Chapter  6:  Prevention and Management of Urinary Tract Infections in Paralyzed Persons: Evidence Report/Technology Assessment Number 6

Objectives

The objective was to analyze the evidence on aspects of the prevention and management of urinary tract infections in paralyzed persons. The two most common conditions affected are spinal cord injury (SCI) and multiple sclerosis (MS). Both conditions have a predilection for onset in young adulthood. Eighty percent of persons with SCI experience a urinary tract infection (UTI) by their 16th year post-injury, and diseases of the urinary system are the 5th most common cause of death. Over 70 percent of persons with MS develop bladder dysfunction over the course of their disease.

Specific questions addressed were (1) what combination of signs, symptoms, and laboratory findings are associated with risks to this population, (2) what are risk factors for recurrent UTIs, and (3) what are the risks and benefits of antibiotic prophylaxis.

Search Strategy

An expert and consumer panel was convened to focus the literature review. A research librarian performed a search of MEDLINE (1966-January 1998) and EMBASE (1974-January 1998) databases, using the terms urinary tract, urinary tract infections, bacteriuria, paraplegia, quadriplegia, spinal cord injuries, multiple sclerosis, neurogenic bladder, and neuropathic bladder. CINAHL (1982-July 1998) was also searched. Some articles were identified by panel members and by review of reference lists.

Selection Criteria

All titles were reviewed, then abstracts of non-rejected titles, where available. Full-length articles were reviewed for accepted abstracts and for titles with no abstract. Selection criteria included human studies of adults and adolescents with neurogenic bladder due to spinal cord dysfunction and relevant to a key question, and non-acute SCI patients. Excluded were case reports, reviews, editorials, and letters, and studies published before 1979 on risk factors for recurrent UTI. For prophylaxis of UTI, only randomized controlled trials were included, as were studies of acute SCI.

Data Collection and Analysis

As articles were reviewed they were designated as addressing one of the key questions. Project investigators reviewed full-length articles and excluded those having insufficient data or not otherwise addressing a question. Data from remaining articles were extracted into evidence tables. Quality of controlled trials and of cohort studies was formally assessed. A formal meta-analysis was undertaken on prophylaxis of UTI. A draft evidence report was critiqued by 22 experts and consumers.

Main Results

Study samples in most of the published literature were patients with SCI. Bacteriuria is a common occurrence; pyuria with bacteriuria may be associated with symptomatic infections, but these findings are also common in asymptomatic patients. The occurrence of febrile episodes in prior years is associated with upper urinary tract complications or abnormalities; bladder calculi are associated with prior cultures of certain bacterial species and of multiple organisms. Other evidence regarding the significance of signs, symptoms, and laboratory findings is sparse or inconclusive due to study design limitations. Indwelling catheterization is associated with more frequent infections than bladder management methods not involving a catheter. The literature does not support firm conclusions regarding most other risk factors. Antibiotic prophylaxis reduces bacteriuria but is not associated with a reduced number of symptomatic infections in the populations studied and results in two-fold increases in the occurrence of antibiotic-resistant bacteria.

Conclusions

Febrile episodes are associated with the later occurrence of upper tract complications. Intermittent catheterization is associated with a lower risk of urinary tract infections. The regular use of antibiotic prophylaxis for most patients with spinal cord dysfunction cannot be supported. Future research should focus in the areas of (1) prospective cohort studies to assess the short-term and long-term significance of signs, symptoms, and laboratory findings (level of bacteriuria and type of organism, pyuria, others); (2) large, multicenter prospective studies of risk factors - potentially modifiable risk factors, in particular - for urinary tract infection; and (3) randomized controlled trials in the subgroup of patients who have frequent, recurrent urinary tract infections that limit their functioning. Studies should include both SCI and MS patients, where feasible, and state-of-the-art methods for maximizing the quality of the study designs should be employed.

This document is in the public domain and may be used and reprinted without permission, except for those copyrighted materials noted for which further reproduction is prohibited without the specific permission of copyright holders.

Suggested Citation

Vickrey BG, Shekelle P, Morton S, et al. Prevention and Management of Urinary Tract Infections in Paralyzed Persons. Evidence Report/Technology Assessment No. 6. (Prepared by Southern California Evidence-Based Practice Center/RAND under Contract No. 290-97-0001.) AHCPR Publication No. 99-E008. Rockville, MD: Agency for Health Care Policy and Research. February 1999.

Overview

The objective was to analyze the evidence on selected aspects of the prevention and management of urinary tract infections in paralyzed persons. The two populations most commonly affected are persons having spinal cord injury (SCI) and people with multiple sclerosis (MS). Both of these conditions have a predilection for onset in young adulthood. Eighty percent of persons with SCI experience a urinary tract infection (UTI) by their 16th year post-injury, and diseases of the urinary system are overall the 5th most common primary or secondary cause of death in this population. Between 70 to 90 percent of persons with MS develop bladder dysfunction over the course of their disease, placing them at increased risk for UTIs. Urinary complications are responsible for a large proportion of hospitalization-related episodes in these patient populations. UTI is the most frequent secondary medical complication reported by the federally designated Model Spinal Cord Injury Systems during acute care and rehabilitation, and UTI was the primary or secondary diagnosis for nearly one-third of hospitalizations of MS patients over the age of 65, according to 1989 Medicare data.

Reporting the Evidence

The specific questions addressed in this report are (1) what combination of signs, symptoms, and laboratory findings are associated with risks to persons with paralysis due to neurogenic bladder, (2) what are risk factors for recurrent UTIs, and (3) what are the risks and benefits of antibiotic prophylaxis.

The literature review for the first key question was broad and included studies of both short-term and long-term risks as related to episodes of various combinations of signs, symptoms, and laboratory findings, for example, the presence of fever, the level of bacteriuria, the type or organism, the presence of varying levels of pyuria, or some combination. For the literature search on risk factors for UTI, types of risk factors examined were socioeconomic status, insurance status, behavioral factors, personal hygiene, sex, and domicile, as well as intermediate risk factors of bladder management method (or drainage), time since injury, and level of functioning (or injury). Regarding prophylaxis, the efficacy of any oral antibiotic therapy and the efficacy of specific oral antibiotics were examined. All analyses were further stratified by acute versus non-acute SCI patients and by asymptomatic and symptomatic UTIs.

Study populations included adults and adolescents (13 years and older). In studies that had patient samples with spinal cord injury, the review focused on non-acute patients (defined as more than 90 days out from their injury) for all key questions, with the additional inclusion of studies of acute SCI patients for the analysis of antibiotic prophylaxis.

Methodology

A 13-member panel of experts, consumers, and a managed care organization representative was convened to focus the literature review on a set of key questions and to develop potential causal pathways for each question. Subsequently, a research librarian performed searches of MEDLINE (1966-January 1998) and EMBASE (1974-January 1998) databases, using the terms urinary tract, urinary tract infections, bacteriuria, paraplegia, quadriplegia, spinal cord injuries, multiple sclerosis, neurogenic bladder, and neuropathic bladder; case reports and animal studies were excluded. CINAHL (1982-July 1998) was also searched. Foreign language articles were not excluded from any searches. Some additional articles were identified by panel members and by review of citations of articles obtained from searches.

All titles were reviewed by two physicians, then abstracts of non-rejected titles, where available. Full-length articles were reviewed for accepted abstracts and for titles with no abstract. Twelve translators assisted in the screening and evaluation of articles in 14 different foreign languages.

Selection criteria included human studies of adults and adolescents with neurogenic bladder due to spinal cord dysfunction and relevant to a key question and inclusion of a potentially relevant outcome measure, such as bacteriuria or UTI. For the first two key questions, studies of acute SCI patients (i.e., limited to within the first 90 days following injury) were excluded. For prophylaxis of UTI, only randomized controlled trials were included; both acute and non-acute SCI study samples were included for this key question. Rejection criteria for all key questions were case reports, reviews, editorials, and letters; studies published before 1979 on risk factors for recurrent UTI were also excluded, because bladder management methods and their associated risks changed greatly with the introduction of intermittent catheterization at that time. As articles were reviewed, they were designated as addressing one of the key questions. Project investigators reviewed full-length articles and excluded those having insufficient data or not otherwise addressing a question. Data from remaining articles were extracted into evidence tables, and results summarized. Quality of controlled trials and of cohort studies was formally assessed.

A meta-analysis was conducted for the key question on benefits and harms of long-term use of antibiotic prophylaxis for UTI in people with neurogenic bladder due to spinal cord dysfunction. Steps included obtaining any additional information needed from authors of studies, identification of the outcomes and subgroups for analyses, formal assessment of evidence for publication bias, selection of an appropriate statistical pooling method, assessment and incorporation of heterogeneity, combination of data across studies, and execution of sensitivity analyses.

A draft evidence report was circulated for critique by the 13-member panel previously convened and by 7 additional content experts, methodologists, and a managed care organization representative. The meta-analysis was additionally reviewed by two outside experts in meta-analysis.

Findings

• Study samples in most of the published literature were patients with SCI.

• Bacteriuria is a common occurrence; pyuria with bacteriuria may be associated with symptomatic infections, but these findings are also relatively common in asymptomatic patients.

• There is convergent data from several large cohort and case-control studies that the occurrence of febrile episodes in prior years is associated with a higher occurrence of upper urinary tract complications or abnormalities at long-term followup.

• The presence of certain bacteria or of multiple organisms early after spinal cord injury is associated with an approximately 3@@@frac12@@@-fold increased odds for developing bladder calculi at 2 years, but the presence of other signs and symptoms and treatment status were not included in the single study of this issue that was identified.

• Other evidence regarding the significance of signs, symptoms, and laboratory findings either is sparse or is inconclusive due to study design limitations.

• Indwelling catheterization is associated with more frequent infections than that involving intermittent catheterization, which in turn is associated with more frequent infections than methods not involving a catheter. (However, severity of disease affects choice of method, particularly the alternatives involving use of a catheter versus no catheter.)

• The literature does not support firm conclusions regarding most other risk factors.

• Antibiotic prophylaxis significantly reduces bacteriuria among acute spinal cord injury patients (p <0.05), and there is a trend for reduction in bacteriuria among non-acute spinal cord patients (p = 0.06). However, antibiotic prophylaxis is not associated with a reduced number of symptomatic infections in the populations studied.

• Antibiotic prophylaxis results in two-fold increases in the occurrence of antibiotic-resistant bacteria.

Future Research

Future research should focus in the areas of (1) prospective cohort studies to assess the short-term and long-term significance of signs, symptoms, and laboratory findings (level of bacteriuria and type of organism, pyuria, others); (2) large, multi-center, prospective cohort or randomized trial studies of risk factors for UTIs, particularly targeting potentially modifiable risk factors like behavioral factors and catheterization techniques; and (3) randomized controlled trials in the subgroup of patients who have frequent, recurrent urinary tract infections that limit their daily functioning and well-being. These studies should include both SCI and MS patients, where feasible, and should enroll a sufficient number of patients for adequate statistical power to detect meaningful clinical differences. In addition to traditional clinical measures, these studies should also measure quality-of-life outcomes and costs. State-of-the-art methods for maximizing the quality of the study designs and the rigor with which they are executed should be employed.

Purpose

This evidence report details the methodology, results, and conclusions of a literature search on the prevention and management of urinary tract infections (UTIs) in paralyzed persons. The information is intended to assist health care providers and organizations with the development of clinical practice guidelines on this topic.

Scope of Work

Diagnosis, treatment, and prevention of recurrent UTIs were all initially considered as part of the scope of this project. However, our technical experts agreed that prevention of recurrent infections and subsequent lower and upper tract complications is the clinician's most important responsibility and the area in which the biggest gap exists between best and usual care. Therefore, the literature review focused specifically on this topic.

Prevalence

Urinary tract dysfunction is often a secondary complication of spinal cord injury¹ (SCI) [defined as any damage to the spinal cord that results in paralysis and/or loss of sensation in various parts of the body (National Foundation for Brain Research, 1992)], multiple sclerosis,² and other disorders that cause spinal cord dysfunction. Although renal failure and other related urinary tract complications have traditionally been the leading causes of death after spinal cord injury, the death rate has declined in recent decades due to increased attention and new methods of managing the neurogenic bladder (DeVivo, Black, and Stover, 1993). In this 1993 study, DeVivo, Black, and Stover found diseases of the urinary system to rank 13 out of 19 primary causes of death in spinal cord injured patients. However, as a secondary cause of death, diseases of the urinary system were the most common cause of death in these patients. Overall, as a primary or secondary cause of death, diseases of the urinary system were the fifth most common cause of death. For this reason, urinary tract infection continues to pose a major threat to the health of persons with paralysis due to spinal cord damage. For example, UTI is the most frequent secondary medical complication reported by the federally designated Model Spinal Cord Injury Systems during acute care and rehabilitation. Eighty percent of individuals in this population reported a UTI (74 percent of those with incomplete paraplegia and 87 percent of those with complete tetraplegia). UTI is also the most common complication after discharge: 57 percent of individuals experience a UTI in the first post-injury year. Eighty percent of the individuals will have experienced a UTI at some point by their 16th year post-injury.

Disturbances in bladder function are very common in patients with multiple sclerosis (MS) (Edlich, Westwater, Lombardi et al., 1990). It is estimated that 70 percent to 90 percent of MS patients develop bladder function abnormalities in the course of their disease (Rabey, Moriel, Farkas, et al., 1979; Blaivas, Bhimani, and Labib, 1979). In turn, this urinary dysfunction often places MS patients at risk for urinary tract infections.

Fleming and Blake (1994) used the 1989 Quality of Care Medicare Provider Analysis and Review (MEDPAR) file to study the prevalence of comorbidities in hospitalized patients with MS aged 65 years and older. They found that urinary tract infection was the most common primary or secondary discharge diagnosis for these patients, with a prevalence rate of nearly 31 diagnoses per 100 discharges.

Anecdotally, it appears that upper urinary tract involvement for people with MS is less common than for those with spinal cord injury. However, in one study, a relatively larger percentage (over 1 in 5) of those MS patients having urinary symptoms had clinically significant upper urinary tract abnormalities (Sliwa, Bell, Mason et al., 1996).

Burden of Illness Due to Recurrent UTI in Paralyzed Persons

There are substantial costs associated with SCI and its secondary medical complications. DeVivo, Whiteneck, and Charles (1995), in analyzing the cost data of the federally designated Model Spinal Cord Injury Systems program, arrived at total first-year post-injury charges of $198,335 (1992 dollars). Average charges incurred annually after paralysis amounted to $24,154, much of this being related to rehospitalization costs for treatment of secondary medical complications. Although average cost data are not available for specific diagnostic and therapeutic interventions (such as treatment of UTIs), DeVivo, Whiteneck, and Charles (1995), in their predictive model of first-year charges following spinal cord injury, project $12,503 for each rehospitalization and $13,926 for each complication requiring rehospitalization. They estimate the aggregate costs of new spinal cord injuries to be $7.2 billion annually ($3.1 billion in direct costs and $4.1 billion in indirect costs). In a more recent study, DeVivo (1997) estimates the total costs of spinal cord injury in the United States has increased to $7.74 billion annually.

Multiple sclerosis also results in substantial costs. Based on the results of a 1994 survey of 606 persons with multiple sclerosis, Whetten-Goldstein, Sloan, Goldstein et al. (1998), at the Duke University Center for Health Policy Research and Education, reported that the annual individual cost of multiple sclerosis was more than $34,000 per person (1994 dollars) and that the total annual lifetime cost per case was $2.2 million. They further estimated that the annual national cost of multiple sclerosis was $6.8 billion.

Meeting with Consortium for Spinal Cord Medicine

The Project Task Order Manager (Barbara G. Vickrey, MD) and Evidence-Based Practice Center (EPC) Director (Paul G. Shekelle, MD) met with the Steering Committee of the Consortium for Spinal Cord Medicine Clinical Practice Guidelines on October 24, 1997, to:

• 1

  Present the scope of work and key steps proposed for the EPC.

• 2

  Request the nominations of technical experts and affected persons to serve on an advisory panel and provide peer review for project documents.

• 3

  Identify key articles, including those on magnitude of the problem, cost, and evidence.

The Consortium for Spinal Cord Medicine is supported by the Paralyzed Veterans of America, who nominated this topic. The Steering Committee includes representatives from the consortium. The consortium member organizations are:

• American Academy of Orthopedic Surgery

• American Academy of Physical Medicine & Rehabilitation

• American Association of Neurological Surgeons

• American Association of Spinal Cord Injury Nurses

• American Association of Spinal Cord Injury Psychologists and Social Workers

• American Congress of Rehabilitation Medicine

• American Occupational Therapy Association

• American Paraplegia Society

• American Physical Therapy Association

• American Psychological Association

• American Spinal Injury Association

• Association of Academic Physiatrists

• Association of Rehabilitation Nurses

• Congress of Neurological Surgeons

• Department of Veterans Affairs

• Insurance Rehabilitation Study Group

• Paralyzed Veterans of America

Panel Meeting

Initial Meeting

Table 1. List of Technical Experts

Table 1. List of Technical Experts

Technical Expert	Area of Expertise	Affiliation/Location
Carol Bennett, MD	Urology	UCLA Dept. of Urology Los Angeles, CA
Michael Burns, M Ed	Consumer	VA Medical Center, San Diego San Diego, CA
Rabih Darouiche, MD	Infectious diseases	VA Medical Center, Houston Houston, TX
Jeffrey Davis, MD	Medical Director, managed care organization	Santa Clara Family Health Plan San Jose, CA
Bruce Dobkin, MD	Neurorehabilitation (Neurology)	UCLA Dept. of Neurology Los Angeles, CA
Michael Dunn, PhD	Clinical psychologist, consumer	VA Palo Alto Health Care System Palo Alto, CA
Angela Joseph, MSN	Urologic nursing	VA San Diego Health Care System San Diego, CA
Todd Linsenmeyer, MD	Urology/physical medicine and rehabilitation	Kessler Institute for Rehabilitation West Orange, NJ
Frederick Maynard, MD	Physical medicine and rehabilitation	Metrohealth Center for Rehabilitation Cleveland, OH
Inder Perkash, MD	Urology/physical medicine and rehabilitation	VA Palo Alto Health Care System Palo Alto, CA
T. Peter Seland, MD	Neurology (multiple sclerosis)	Private Practice; Kelowna General Hospital Multiple Sclerosis Clinic British Columbia, Canada
Ken Waites, MD	Microbiology	University of Alabama Birmingham, AL
Mary Nancy Young, RN, MS	Spinal cord injury nursing	Sun Lakes, AZ (formerly at Rancho Los Amigos)

Project staff scheduled an initial meeting for 13 technical experts, consumers, and a managed care organization representative, at RAND (Santa Monica, California) on December 4-5, 1997. (See Table 1 for the final list of 13 panelists, including name, expertise or perspective, and geographic location.)

Purpose of the Meeting

The purpose of the meeting was to familiarize the technical experts with the project and to utilize their expertise to define key parameters for the project literature review and analysis. They addressed the following topics at the meeting:

• 1

  Definition of the clinical target population,

• 2

  Determination of the clinical outcomes of interest,

• 3

  Selection of the key research questions,

• 4

  Development of causal pathways for each of the selected key questions, to guide the review of the evidence, and

• 5

  Identification of potential keyword search terms and evidence outside the usual databases.

Causal pathways or evidence models were developed for key questions that were broad and could be conceptualized as having linkages between different sets of scientific evidence, for example, questions where there is a surrogate or intermediate outcome that may be related to a clinically relevant health outcome (Mulrow, Langhorne, and Grimshaw, 1997). These pathways are the framework for assessing the strength and amount of evidence linking different steps within a pathway.

Table 2. Summary Data on Potential Target Conditions (more...)

Table 2. Summary Data on Potential Target Conditions¹

		Population
Summary data	Spinal Cord Injury	Multiple Sclerosis	Stroke
Prevalence of condition (US)	~200,000 525-1125/million	200,000-500,000 580-1730/million	~3,000,000
Male/female ratio	4:1 M:F <20% F	1:2-2.5 M:F ~60-70% F	<1:1 M:F 56% F
Age of onset	mean age=29 yrs	median=29 yrs	Median=74yrs
Estimated cost of illness (National Foundation for Brain Research, 1992)	$16-36 billion annually	$3-8 billion annually	$18 billion annually ($30 billion in elsewhere)
Rates of occurrence of/ care for:
· UTI	about 20% annual incidence 46% of inpatient stays	most common discharge diagnosis in elderly MS patients; 9/100 hospital discharges
· Other lower tract	34-35% of inpatient stays
· Upper tract	15% of inpatient stays
· Urinary symptoms		up to 90% over course of disease
· Incontinence			60% 1st week 40%@4 weeks 30%@12 weeks
Patterns of urinary dysfunction over course of illness	acute and chronic	variable over course of disease	Early after stroke
# of articles identified in MEDLINE/ EMBASE search, all languages, 1985- present, "UTI+condition" clinical trial only	95	22	12
Estimated # of 'relevant' articles (Note: this is '85-'97 only)	~75	~12	~3

1

Handout to facilitate discussion among technical experts; source includes NFBR, 1992

To inform their discussion, the technical experts were provided with a handout (Table 2)containing summary data on spinal cord injury, multiple sclerosis, and stroke (although stroke was subsequently excluded by the expert panel from the scope of the evidence report). Prior to the panel meeting, these three conditions were deemed most likely to cause impairment that would lead to urinary tract dysfunction and to serve as the primary study populations for the report.

Subject Definition

The technical experts agreed that management of urinary tract dysfunction is best directed by physiology rather than by specific cause. Thus, it would be most useful to consider UTIs in terms of the bladder physiology and urodynamic abnormalities, rather than in terms of the specific causes of the urinary tract problems (e.g., spinal cord injury, multiple sclerosis, etc.).

Because the type of bladder dysfunction caused by stroke is dissimilar to that due to spinal cord dysfunction, the panel recommended that the focus of the evidence report be on conditions resulting in spinal cord dysfunction and that the report exclude literature on stroke and disorders of the brain.

Identification of the Main Key Questions

The panel also agreed that prevention of recurrent infections is the clinician's most important responsibility and the area in which the biggest gap exists between best and usual care.

Table 3. Original List of Potential Key Questions¹

Table 3. Original List of Potential Key Questions¹

1 What is significant bacteriuria as applied to people with paralysis, where significant means has potential risks or harms without treatment? How is a urinary tract infection in people with paralysis defined? 2 What are risk factors (structural, physiological, behavioral, demographic) for urinary tract infection in people with paralysis? 3 How are these risk factors further influenced by methods of urinary drainage? 4 What diagnostic studies should be done when people with paralysis present with symptoms of urinary tract infections to confirm this diagnosis and make treatment and management decisions? 5 What are the indications for hospitalization for urinary tract infection? For referral to specialists? 6 What are the appropriate indications for antibiotic use for urinary tract infection in people with paralysis? Are indications altered by duration of injury, treatment, setting, or method of urinary drainage? 7 When antibiotics are used, what factors influence choice of drugs, dosage, duration, and route? 8 Are there indications for long-term use of prophylactic antimicrobial agents or for other prophylactic agents, like ascorbic acid? 9 What are indications for followup testing in people with paralysis who have had urinary tract infections? 10 What are the preventive, diagnostic, and management issues for other lower tract complications, i.e., orchitis, epididymitis, urethral fistula, bladder calculi? 11 What are the preventive, diagnostic, and management issues for upper tract complications, i.e., renal calculi, hydronephrosis, renal and perirenal infection, renal insufficiency and failure?

1

Developed by project staff and onsite technical experts

The panelists and project staff then reviewed and discussed the 11 proposed key questions originally developed by project staff and the onsite experts prior to the meeting (Table 3).

Table 4. Revised Key Questions¹

Table 4. Revised Key Questions¹

1. What combinations of symptoms, signs, and urinary laboratory findings are associated with risks to the persons with neurogenic bladder due to spinal cord dysfunction? Potential factors to consider are drainage method, sex,2 and how the sample was obtained.

2. What are the risk factors for recurrent UTI in persons with neurogenic bladder due to spinal cord dysfunction? Kinds of risk factors include method of drainage, behavioral, sex, bladder physiology, and others.

4a. What diagnostic studies should be performed to detect UTI in persons with neurogenic bladder due to spinal cord dysfunction?

4b. What diagnostic studies should be performed to localize UTI and/or detect the presence of UTI with complications in persons with neurogenic bladder due to spinal cord dysfunction?

5a. What are the indications for hospitalization for UTI?

5b. What are the indications for referral to specialists?

7. What factors influence the choice of antibiotics, dose, duration, and route?

8. What are the benefits, harms, and costs of the long-term use of prophylactic agents? (dependent upon drainage)

9. What type of followup is needed for persons who have neurogenic bladder due to spinal cord dysfunction? Kinds of followup include diagnostic testing, patient education, and referral to specialists.

10. What are the benefits, risks, and costs associated with different drainage methods?

1

Revised after discussion with the expert panel.

2

Although initial questions and pathways developed by the panel used the term "gender," an AHCPR editor specified that the term "sex" should be used when the meaning is "male and female." Thus, "sex" replaces "gender" throughout the report.

As seen in Table 4 (Revised Key Questions), by discussion and consensus, the meeting participants agreed to do the following:

• 1

  Rewrite Question 3 to reflect the benefits, risks, and costs associated with drainage methods.

• 2

  Renumber Question 3 and combine it with Question 10.

• 3

  Fold Question 11 into Question 10.

• 4

  Divide Questions 4 and 5 into two questions each.

• 5

  Collapse Questions 6 and 7 into one question.

The initial development of the original set of key questions and their subsequent revision were based on the panel members interpretation of the evidence known to them and their clinical judgment.

Ranking of Conditions

Table 5. Technical Experts Ranking of Key Questions (more...)

Table 5. Technical Experts Ranking of Key Questions¹

Round 1	(Note that P13 ratings had to be approximated as he/she gave two 5's and three 4's.)
Question	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	Sum
1	0	5	5	1	3	5	5	5	5	3	5	5	4	51
2	1	0	3	2	5	3	1	4	0	1	4	2	4	30
4a	0	0	0	0	0	0	0	0	0	0	3	0	0	3
4b	0	0	1	0	0	2	0	3	4	0	0	3	0	13
5a	0	0	0	0	0	0	0	2	0	0	0	0	0	2
5b	0	1	0	0	0	0	0	0	0	0	0	0	0	1
7	3	0	0	0	0	0	0	0	2	0	0	0	0	5
8	2	4	2	3	2	1	4	0	1	5	2	0	2.3	28.3
9	5	3	4	5	4	4	3	1	3	2	0	4	2.3	40.3
10	4	2	0	4	1	0	2	0	0	4	1	1	2.3	21.3
Total														194.9

1

Each technical expert ranked 5 questions, providing weights from 5 (most relevant) to 1 (least relevant).

Using the approach of a modified Delphi procedure, each technical expert then anonymously ranked the top five key questions that he or she believed the evidence report should address. Rankings were based on the importance and the perceived availability of data to answer the question. As seen in Table 5, Questions 1, 2, 8, 9, and 10 received the highest rankings.

Table 6. Causal Pathways for Top-Ranked Key Questions (more...)

Table 6. Causal Pathways for Top-Ranked Key Questions

	Key Question #1
	Signs, Symptoms, and Laboratory Findings
Sex	-	What level of bacteriuria (data from able-bodied used as standard)
Drainage
	-	Indwelling (F>M)
	-	Intermittent catheter (F=M)
	-	Reflex voiding (M>F or M=F), with and without external collector
	-	Catheter-free (F<M)
	-	Urinary diversion (no data on sex available)
Sample collection method
	-	Indwelling (pre-existing vs. new)
	-	All others (catheterized vs. non-catheterized)
Signs and symptoms (The Prevention and Management of Urinary Tract Infections Among People With Spinal Cord Injuries, NIDRR Consensus Statement, 1992, p. 196)
	-	Leukocytes in the urine
	-	Discomfort or pain over the kidney or bladder, or during urination
	-	Onset of urinary incontinence
	-	Fever
	-	Increased spasticity
	-	Autonomic hyperreflexia
	-	Cloudy urine with increased odor
	-	Malaise, lethargy, or sense of unease
	-	Change in urinary frequency or hesitancy
	-	Symptoms of an exacerbation or relapse in people with multiple sclerosis
Additional keywords for search:
	-	quality of life
	-	spastic bladder
	-	sphincterotomy
	-	hospitalization
	-	urodynamics
	-	bladder function
	-	bladder dysfunction
	-	dyssynergia of detrusor sphincter
	-	bladder catheterization
	-	neurostimulation of bladder
	-	pyelonephritis
	-	autonomic dysreflexia
	-	detrusor sphincter dyssynergia

Key Question #2: Risk Factors for Recurrent UTI

Key Question #8: Prophylactic Agents

Key Question #9: Type of Followup for Persons with Neurogenic Bladder

Key Question #10 Benefits and Disadvantages of Drainage Methods
Method	Benefits	Disadvantages
Indwelling catheter	- convenience	- increased stones
	- continuous drainage	- increased infection
	- prevention of skin breakdown	- increased upper tract changes
	- control over fluid status	- incontinence
	- decreased number of caregivers	- lack of flexibility for other options (i.e., intermittent catheterization)
	- increased independence	- need for anticholinergics
	- increased insurance-paid nurse support (regional)	- bladder cancer
		- autonomic dysreflexia
		- sexual problems
		- urethral complications
		- latex reactions
		- epididymitis
		- severity of infection
		- decreased aesthetics
Suprapubic tube	-sexual function	- need operation
	- decreased urethral complications	- increased physician involvement (may imply increased costs)
	- ? effect on dysreflexia	- more incontinence
	- easy insertion	- latex reaction
	- convenience	- increased stones
	- continuous drainage	- increased infection
	- prevention of skin breakdown	- increased upper tract changes
	- control over fluid status	- incontinence
	- decreased number of caregivers	- lack of flexibility for other options (i.e., intermittent catheterization)
	- increased independence	- need for anticholinergics
	- increased insurance-paid nurse support (regional)	- bladder cancer
		- autonomic dysreflexia
		- sexual problems
		- latex reactions
		- epididymitis
		- severity of infection
		- increased upper tract changes
Intermittent catheterization	- appliance free	- often require anticholinergics
	- increased self image	- urethral complications
	- sexual function	- concern about fluid intake (compulsive behavior)
	- stable bladder function	- inconvenient
	- decreased incidence of urinary complications	- requires adequate hand function
	- more flexibility for change to alternatives	- more concern about severity of infection
	- less dysreflexia
	- nonsurgical
Reflex voiding with external collection	- convenience	- need for treatment for sphincter (med or surg): If done:
	- no fluid concern	- social embarrassment
	- less need for attendant care	- only used in males
	- colonization	- skin breakdown If not done:
	- no need for hand function	- urethral trauma or damage
		- increased vesicoureteral reflux (VUR)
		- increased dysreflexia
		- requires more urological followup
		- incontinence because of trouble with the condom collection device
		- sexual problems for the true reflex voiding individual
Non-catheter (Valsalva crede)	- no collection device	- inguinal hernias
	- more independence	- hemorrhoids
	- better self image	- rectal prolapse
	- less colonization	- incontinence
	- sexual function	- high pressure voiding
	- no surgery	- increased VUR
	- less urethral complications	- inadequate drainage
Urinary diversion (incontinent)	- simplicity of surgery and collecting devices	- reflux
	- convenience	- stones
		- skin problems
		- major surgery with complications
		- electrolyte imbalance
		- altered self image
		- need more medical attention
		- pyelonephritis
Urinary diversion(continent)	- continence	- major surgery
	- self image	- stones
	- ? decreased infection	- still have increased risk of upper urinary tract complications
	- more accessible catheterization	- rupture
	- low pressure system
Newer methods to investigate:
- Electrostimulation methods, also called sacral stimulation (used in Europe and under research protocols in certain centers in the US)
- Magnetic stimulation
- Urethral plug

A second round of votes among the five key questions confirmed the order of the first round of balloting. Question 1 was viewed as the most important, followed by 9, 2, 8, and 10. Table 6 contains the causal pathways or tables for the five key questions determined by the technical experts to be most promising. These causal pathways were developed by the experts as a group during the meeting, guided by senior project staff (Drs. Shekelle and Vickrey). A subset of these five key questions would be the basis for the subsequent literature search.

Revision of Key Questions and Causal Pathways

Table 7. Technical Experts' Comments on Report of Panel (more...)

Table 7. Technical Experts' Comments on Report of Panel Meeting

Comments on Issues Raised During Meeting:

Item 6: Add: Prevent recurrence of symptomatic UTIs.

Item 7: Add: Stroke patients do not have dyssynergia and thus do not have obstructive voiding like spinal cord injury patients.

Item 8: Categorization of bladder dysfunction: Subgrouping by physiology: It is preferable to use the terminology specified by the International Continence Society (ICS): Detrusor hyperreflexia Detrusor hyperreflexia with external sphincter dyssynergia and/or internal sphincter dyssynergia Detrusor areflexia Detrusor hyperactivity with impaired bladder contractility Change "emptying pressure is also an issue" to "filling and emptying pressures are also issues." Change first sentence as follows: Categorization of bladder dysfunction can be looked at as either a failure to store urine, a failure to empty, or a failure to store and to empty (e.g., a person who has overflow incontinence); voiding pressure is also an issue (high vs. low pressure).

Comments/Questions Regarding (Original List of) Key Questions:

1 Complete the following sentence as follows: Urease-producing pathogens need more aggressive treatment to prevent stone disease 2 Under the question, "Can we present information (based on evidence) for patient to make an educated choice?" Change question to: Can we present information (based on evidence) for a person with spinal cord injury to make an educated choice? Add "Bladder stimulation" as a drainage method. Replace "Condom catheterization" and "Reflex voiding" with "Reflex voiding with or without a condom collector." Add to "Reflex voiding": Reflex voiding using a condom catheter Change sentence as follows: Method and compliance while using clean technique for catheterization will influence risk in intermittent catheterization; a practical and economical cleaning method needs to be found. Note that a clean and economical cleaning method also needs to be found with condom catheter-cleaning of bag and daily changing of the condom catheter. Add "volitional voiding" to list of drainage methods.

Key Question #1: What combinations of symptoms, signs, and urinary laboratory findings are associated with risks to the persons with neurogenic bladder due to spinal cord dysfunction?

Revision of Key Question: Additional keywords for search: Change "catheterization" to "bladder catheterization." Add: Pyelonephritis, Autonomic dysreflexia, Detrusor sphincter dyssynergia.

Figures/Tables Representing Key Questions: Drainage: Indwelling is higher among women. On a percentage basis, the frequency of use of intermittent catheterization is about equal among men and women who use intermittent catheterization (two technical experts made this comment). Reflex voiding is almost never used by women; should be M>F or M=F; also, change "external catheter" to "external collector." Catheter free should be changed to F<M. Urinary diversion: "No data on gender" should be noted.

Key Question #2: What are the risk factors for recurrent UTI in persons with neurogenic bladder due to spinal cord dysfunction?

Revision of Key Questions: Change order of listed risk factors to: method of drainage, bladder physiology, gender, behavioral, and others.

Figures/Tables Representing Key Questions:

Delete "Psychosocial" and add "Personal Hygiene" in middle box, left margin.

Comments on Key Questions, continued:

Key Question #4b: What diagnostic studies should be performed to localize UTI and/or detect the presence of UTI with complications in persons with neurogenic bladder due to spinal cord dysfunction?

Revision of Key Questions: It might be worth defining to the reader whether "UTI with complications" are "complicated UTIs or UTI in which secondary complications occur.

Key Question #8: What are the benefits, harms, and costs of the long-term use of prophylactic agents?

Figures/Tables Representing Key Questions: Asymptomatic colonization of the bladder may require adherence to uroepithelial cells; therefore, would delete "adherence" from the "Tissue Invasion" box. As a corollary, both "Prophylactic Agents" and "Anti-adherence Prophylaxis" may act on preventing "Asymptomatic Colonization" and "Tissue Invasion." Juice is an anti-adherence prophylaxis. The term "bacterial interference" includes "other microbes."

Key Question #9: What type of followup is needed for persons who have neurogenic bladder due to spinal cord dysfunction?

Figures/Tables Representing Key Questions: Add "CT scan" under Potential Tests.

Key Question #10: What are the benefits, risks, and costs associated with different drainage methods?

Indwelling Foley Catheter: Delete the word "Foley" Harms: Add: "severity of infection"

Suprapubic (SP) Tube: Benefits: The benefits listed for "Indwelling Foley" also apply to "SP tube." Harms: With the exception of "urethral complications," all the harms listed for "Indwelling Foley" also apply to "SP tube." With "increased physician involvement," note that increased visits and tests imply increased costs. Add "Increased upper tract changes." Add: "severity of infection."

Intermittent Catheterization: Harms: Change "Requires better functional status to Requires adequate hand function. Change "may require anticholinergics" to "often require anticholinergics." Why is "more incontinence" listed?

Reflex Voiding with External Collection: Benefits: Colonization is an issue that needs to be addressed rather than bacteria. Delete "Need for hand function." (Another technical expert made a similar comment: Add 'No' before "Need for hand function.") Harms: Incontinence is also a harm because of trouble with the condom collection device. Add "Sexual problems" for the true reflex voiding individual. Separate harms into whether or not treatment done for sphincter: If done: social embarrassment, only used in males, skin breakdown. If not done: urethral configuration, increased VUR, increased dysreflexia, requires more urological followup. What is meant by "urethral configuration"? Perhaps it means urethral trauma or damage?

Non-catheter (Valsalva crede): Harms: It is not inconvenient Add: High pressure voiding Add: Increased VUR Add: Inadequate drainage

Urinary Diversion (Incontinent): Harms: Add: Pyelonephritis

Urinary Diversion (Continent): Benefits: Skeptical that "less upper tract complications" is backed up with data. Harms: Is incontinence a real harm in patients with "Continent urinary diversion" vs. "Incontinent urinary diversion"? Is "less complications of surgery" considered harmful? Clarify that continent is also part of urinary diversion.

Several weeks following the initial meeting, project staff circulated the summary report of the panel meeting and the figures/tables for the five key questions selected to the technical experts for comments, corrections, and other input. Their comments were collated and divided into several categories:

• Comments on issues that were raised during the meeting

• Comments and questions regarding the original list of key questions

• Comments on the five key questions selected at the meeting for indepth literature review (Table 7)

Comments took the form of additions, corrections, and wording clarification. Project staff incorporated the technical experts' comments into the meeting summary and then sent them the revised summary with a letter indicating that their comments and changes were included in the literature review strategy.

A preliminary search on Key Question 10 indicated that synthesizing the volume of available literature would be beyond the size and scope of this project. Project staff continued with literature searches on the four remaining key questions.

Literature Search

Panel Input

During the initial meeting, the panel members were invited to provide the staff research librarian with information on non-standard data sources and to suggest literature search strategies and key words relevant to the five selected key questions. Some of the terms suggested included the following:

• Quality of life

• Spastic bladder

• Sphincterotomy

• Hospitalization

• Urodynamics

• Bladder function

• Bladder dysfunction

• Dyssynergia or detrusor sphincter

• Catheterization

• Neurostimulation of bladder

Table 8. Literature Searches and Review of Titles and (more...)

Table 8. Literature Searches and Review of Titles and Abstracts

I. SEARCH DESCRIPTIONS AND STRATEGY	Yield
	MEDLINE	EMBASE 1	Total
II. REVIEW OF TITLES AND ABSTRACTS	Yield
Preliminary search for spinal cord injury (SCI), multiple sclerosis (MS), & cerebrovascular disease (CV); randomized controlled trials wereidentified and articles photocopied	796 79	480 50	1,276 129
Key Question #1: Symptoms, signs, and laboratory findings associated with the risk of urinary tract infection (UTI) in paralyzed people
(UTI OR SCI OR MS) PLUS (symptoms OR laboratory tests)	110	68	178
Broaden search to include quadriplegia, paraplegia, neurogenic bladder, neuropathic bladder
(UTI OR SCI OR MS) PLUS (paraplegia OR neurogenic bladder OR neuropathic bladder)	1,757	822	2,579
UTI alone	9,611	8,710	18,321
Add bacteriuria to search
(UTI OR SCI OR MS) PLUS (paraplegia OR neurogenic bladder OR neuropathic bladder) OR bacteriuria	1,881	898	2,779
FINAL SEARCH:
Add urinary tract (UT) alone to search; exclude case reports and animal studies; run separately for 1964-1984 and 1985-present
(UTI OR UT OR bacteriuria) AND (SCI OR MS) AND (paraplegia OR neurogenic bladder OR neuropathic bladder); search separately for 1966 (1974 for EMBASE) to 1984 and 1985 to present
1966/74-1984	2,313	446	2,759
1985-present	1,546	693	2,239
All years	3,859	1,139	4,998
MEDLINE	EMBASE	Total
IIA. Review of Titles
1. Reject:	2,787	650	3,437
2. Obtain full citation with abstract:
1966/74- 1984	622	204	826
1985- present	450	285	735
All years	1,072	489	1,561
IIB.Review List (IIA2) with MEDLINE/EMBASE Abstracts (where available)
1. Reject:
Rejected on basis of abstract:
Not relevant to a key question	155	85	240
Study population less than 13 years of age	70	49	119
Case report/editorial/letter/non-human population	3	8	11
Surrogate outcome	4	3	7
Too acute (<90 days from spinal cord injury) AND not key question 8	7	3	10
Nonrandomized controlled trial AND (Key Question 8)	3	3	6
Other reason	1	0	1
Subtotal	243	151	394
No abstract but rejected on basis of re-review title and on remote year of publication (MEDLINE 1966-74)	190	0	190
Overview/review articles and not relevant	84	40	124
Total	517	191	708
2. Obtain articles for screening:
Article pulled based on review of abstracts and match to Key Questions:
Key Question 1	31	7	38
Key Question 2/Case series	57	29	86
Key Question 2/Not case series	49	12	61
Key Question 2/Cannot tell if case series	32	9	41
Key Question 8	18	2	20
Key Question 9	156	54	210
No abstract available: Article pulled on basis of title	241	187	428
Total	584	300	8842
(Review articles: Obtained for search of references	12	10	22)

1

Tallies for EMBASE searches exclude articles that were also listed in the comparable MEDLINE search.

2

Note: 884 is greater than the number of articles actually obtained because some articles were classified as potentially relevant to two key questions.

Based on discussions at the technical experts' meeting, the research librarian performed MEDLINE and EMBASE searches for Key Question 1 (the symptoms, signs, and laboratory findings related to the risk of UTI in paralyzed people). "Urinary tract infection" was combined with "spinal cord injury" and "multiple sclerosis," and with selected symptoms (e.g., fever, malaise) and laboratory tests (e.g., leukocyte counts, microbiology techniques). This search yielded 178 journal articles (110, MEDLINE; 68 EMBASE, non-duplicated) (Table 8.)

Based on a review of search results, "spinal cord injury" was broadened to include quadriplegia, paraplegia, neurogenic bladder, and neuropathic bladder (searched as a free- text term because it did not appear as a subject heading in either MEDLINE or EMBASE; yield=2,579: 1757 MEDLINE and 822 EMBASE). (A separate search using the term "UTI" alone yielded a total of 18,321 articles.) Subsequently, the term "bacteriuria" was added to the search strategy (yield=2,779; 1,881 MEDLINE and 898 EMBASE).

Final Search Strategy

Table 9. Literature Search Strategy

Table 9. Literature Search Strategy

SEARCH #1: Urinary tract (exploded in MEDLINE) or Urinary tract infections (exploded in MEDLINE and EMBASE) or Bacteriuria and Paraplegia (exploded in MEDLINE) or Quadriplegia or Spinal cord injuries (exploded in MEDLINE and EMBASE) or Multiple sclerosis

SEARCH #2: Neurogenic bladder or Neuropathic bladder (free-text term)

Searches 1 and 2 were "or'ed" together to avoid duplicate records. Other search qualifiers included the following: - Human only - No case reports Foreign language materials were not excluded from the search. The results of all of the searches were run through a DIALOG duplicate detection function to remove many of the records that were duplicated between MEDLINE and EMBASE. The totals for the searches were as follows: - MEDLINE-3,859 - EMBASE-1,139

Based on results of these literature searches, the research librarian was provided with the final strategy for all Key Questions (Tables 8 and 9). It was decided to: add the term "urinary tract" as well as "urinary tract infection," and exclude case reports and animal studies. The revised search was run for two time periods: 1966-84 and 1985-present. The yield for these final searches was a total of 4,998 articles (3,859, MEDLINE; 1,139, EMBASE).

Literature Assessment

Literature Tracking

Table 10. Instructions for Coding Articles in EndNote (more...)

Table 10. Instructions for Coding Articles in EndNote

The research librarian will convert all literature searches into EndNote format and send them by e-mail to the Task Order Coordinator and UCLA research assistant.

1 The first step is to code each article as to source, main search topic, date of search, and RCT status PRIOR to incorporating it into the Master Search file. These codes should be inserted in the 'Notes' field of EndNote. The codes are: Source: SO1 EMBASE SO2 MEDLINE SO3-SO7 Provided by technical experts S08 Article identified by reference checking Topic-Date (date is date of literature search): TO1-date Spinal cord injury TO2-date Multiple sclerosis TO3-date Cerebrovascular disorder Randomized Controlled Trial Status: RCT1 Assumed to be RCT RCT1 Not assumed to be RCT Status of Article Retrieval: As the articles are photocopied, code them as follows: RECD Article photocopied ORD Article not yet photocopied Blank Article not requested to be photocopied 2 After search-specific coding has been completed, incorporate the sub-files into the Master file. 3 Articles received from a source other than a literature search should be hand-entered into the Master File: 1 Check to see if they are already in the master file; if so, do not enter again; however, still enter the additional source, even if two sources now appear for the article 2 Code the date as instructed 3 Code each article RECD

Thus, when all the spinal cord injury articles identified in the EMBASE database are searched, one will "find" (under "References") SO1 in Notes AND TO1 in Notes. When the dates are specified, one can identify exactly when the search was run. ("SO1" will pull up all articles coded as SO1 and SO1-date)

We can add other codes later, such as who reviewed an article, the result of the review, if the article was really an RCT, etc. These codes will be entered to the "master."

The next steps will be: 1 Create full lists of articles for the researchers to review, including abstracts; they will give you the specification (e.g., articles with topic of multiple sclerosis identified) 2 Code the articles they want photocopied in the Notes field 3 Sort the articles they want photocopied by journal name 4 Code the photocopy status (i.e., if the article was photocopied and who it went to)

The capabilities will be to: 1 Identify articles by the source, date search sent, search topic, RCT pre- and post-review, reviewer, score, etc. 2 Create a list for photocopying articles, sorted by journal 3 Track if an article has been requested 4 Track if an article has been photocopied 5 See if a newly requested article has already been received 6 Code the reviewer 7 Code the score 8 Create the project bibliography

All reviewed articles were coded into EndNote, following the instructions shown in Table 10. Articles were coded by source (EMBASE, MEDLINE, technical expert, back search, bibliography), topic (spinal cord injury, multiple sclerosis), and RCT status. The status of article retrieval was also coded using the EndNote software.

Abstract Reviews

Table 11. Key to Abstract Coding

Table 11. Key to Abstract Coding

Accept
	A1= Key Question #1
	A2 = Key Question #2 if A2, then also code as either:
	A8 = Key Question #8	S1 = case series
	A9 = Key Question #9	S2 = anything else
		S3 = cannot tell
	Note: non-RCTs studies included for all key questions except #8; if it is unclear whether the study is an RCT or not, include (accept) the article.
	Note: An article may have more than one Key Question
Reject
	R1 = not relevant to a key question
	R2 = study population < 13 years
	R3 = case report/editorial/letter/non-human
	R4 = surrogate outcome
	R5 = too acute (< 90 days solely)
	R6 = non-RCT for #8
	R7 = other reason
	Note: Select one code above; apply the smallest number that denotes a rejection criterion for a particular article
Overview/ Review Article
	V1 = not relevant
	V2 = relevant - pull and review

The selected abstracts were reviewed to identify which full articles to pull for further review. Project staff developed abstract coding criteria for this purpose and then trained the physician reviewers in their use. Strict screening criteria and training were necessary to ensure the reliable (i.e., reproducible) review of abstracts. As seen in Table 11, articles were accepted if they potentially addressed at least one of the four highest-priority key questions (i.e., Key Questions 1, 2, 8, or 9). Only randomized controlled trials (RCTs) were accepted for Key Question #8 (Harms, Benefits, and Costs of the Long-term Use of Prophylactic Agents); non-RCTs were accepted for Key Questions 1, 2, and 9. If there was doubt as to whether the article represented an RCT, its RCT status was deferred until review of the full article.

At the abstract review stage, project staff rejected articles for one of the seven following reasons:

• 1

  Non-relevance to a key question

• 2

  Study population less than 13 years of age

• 3

  Case report, editorial, or non-human population

• 4

  Report of a surrogate outcome

• 5

  Study population only included subjects within 90 days of a spinal cord injury (unless article addressed Key Question 8 on prophylaxis for UTI)

• 6

  Study was not an RCT and it only addressed Key Question 8 on prophylaxis for UTI

• 7

  Miscellaneous (reason had to be provided)

Abstractors were instructed to use the lowest number rejection code (e.g., if an article was both not relevant to a key question and was a case report, the Reject Code would be R1). Overview and review articles deemed relevant to the evidence report were coded to indicate they should be pulled and their citation lists reviewed.

After the first 76 abstracts were reviewed independently by each reviewer, a meeting was held with project leaders to review agreement. Of the first 76 abstracts, 11 actually had no abstract in the MEDLINE or EMBASE database and 7 were review articles; for 47 abstracts there was agreement between reviewers, and for 11 articles there was disagreement. Reasons for disagreement were discussed, and the reviewers proceeded with their independent assessments. Subsequently, an abstract was only rejected if rejected by both reviewers.

A total of 708 citations were rejected at the abstract stage. Of these, 394 (56 percent) were rejected because a review of their abstracts indicated that the citation met one of the seven rejection criteria listed above. In addition, another 190 articles that had no abstract (and were published between 1966 and 1974) and 124 review articles were rejected on the basis of their titles.

The articles for the remaining 831 citations were pulled and photocopied for full screening. (Table 8 indicates that 884 articles were relevant to a key question. This number is larger than 831 because 53 articles were classified as potentially relevant to two key questions.) Of these, 428 were pulled on the basis of title only because no abstract was available. An additional 22 overview articles were pulled so that their references could be reviewed.

Article Screening

Table 12. Screening Form

Table 12. Screening Form

Each article was screened independently by at least two persons using a form designed to estimate study quality (Table 12). The form recorded the relevant outcome(s) measured in the study, study design, sample size, and key question(s) addressed. The results were entered into an Excel spreadsheet for analysis. The spreadsheet was programmed to generate a list of articles about which the two reviewers disagreed to accept or reject them. Dr. Vickrey or Dr. Shekelle reviewed each article for which there was a discrepancy and made a final decision as to its status.

Table 13. Screening of Foreign Language Articles Requiring (more...)

Table 13. Screening of Foreign Language Articles Requiring Translator Assistance

Language	# of Articles	Disposition			Key Question Addressed 1 (for those accepted)
		No Translator Available	Reject	Accept	1	2	8	9
French	29		27	2				2
German	27		23	4		2		3
Japanese	18		14	4		4		3
Italian	14		11	3		2		1
Russian	11		11
Norwegian	6		5	1	1			1
Spanish	5		2	3		1		2
Portuguese	3		2	1		1
Dutch	3		3
Swedish	2		2
Serbo-Croatian	2	2
Danish	2		2
Hebrew	2		2
Polish	2		1	1		1		1
Chinese	1		1
Finnish	1	1
Total	128	3	106	19	1	11	0	13

1

The total # of articles addressing key questions (1+11+13=25) is greater than the total number accepted (19) because an article could address more than one key question.

If possible, foreign language articles with an English abstract were screened. However, some of these articles could not be classified without additional review by someone fluent in that language. Most of these were in French, German, Italian, and Japanese (Table 13).

Articles were rejected at this phase for one of the five following reasons:

• 1

  Study population was primarily less than 13 years of age or included subjects without neurogenic bladder due to spinal cord dysfunction.

• 2

  No relevant outcome measures.

• 3

  The publication was a case report, letter, review, or the article reported a non-human population.

• 4

  The study did not address a key question.

• 5

  The study addressed Key Questions 2 or 9 but the article was published in 1978 or earlier.

Rejection criterion #5 was added after review of the first 60 articles and discussion with the onsite technical experts (Drs. Bennett and Dobkin at UCLA) indicated that the introduction and dissemination of intermittent catheterization in the early- to mid-1970s greatly changed the management of neurogenic bladder and risks for UTI. Thus, it was agreed that studies performed prior to 1978 were not relevant to current risks for recurrent UTI or to the issue of management to prevent long-term complications of neurogenic bladder.

Of the 853 full-length articles selected for screening, 509 (60 percent) were rejected. All but 3 out of the 128 foreign language articles requiring translation assistance were either screened by a methodologist fluent in that language, or reviewed by a member of the project team in a face-to-face meeting with someone fluent in that language. These 3 articles represent 0.4 percent of the total number of 853 articles identified for review and screening.

Table 14. Distribution by Journal Title of Articles (more...)

Table 14. Distribution by Journal Title of Articles That Could Not Be Obtained

Journal Name	Frequency
Aktuelle Neurologie	1
Aktuelle Urologie	2
Ann. Readapt. Med. Phys.	1
Annales de Medecine Physique	1
Cah. Kinesither	1
Chang Gung Medical Journal	1
Chemotherapy (Tokyo)	1
Chirurgia (Milan)	1
Drugs	1
Drugs of the Future	1
Fizik Tedavi Rehabilitasyon Dergisi	2
Fortschritte der Medizin	1
Hokkaido Journal of Medical Science	1
Infirmiere Canadienne	1
International Urology and Nephrology (Magy. Urol.)	1
Japanese Journal of Urology	1
Journal d-Urologie et Nephrologie	1
Journal d Echographie et de Medecine par Ultrasons	1
Journal of Rheumatology and Medical Rehabilitation	1
Journal of the American Paraplegia Society	1
Journal of the Norwegian Medical Association	1
Khirurgiia (Sofia)	3
Kurortologiia I Fizioterapiia	1
Medula Espinal	1
Nishinihon Journal of Urology	2
Revue Medicale de Toulouse	1
The Canadian Journal of Nursing Research	1
The Journal of Neurological & Orthopaedic Medicine & Surgery	1
Urologe Ausgabe A	2
Total # of articles	35

Only 35 articles (4 percent of total) could not be secured through the UCLA Interlibrary Loan service. On review of the 29 journals in which these articles were published, none appear to be widely circulated (Table 14). In addition, none of these journals was on a list of relevant journals having to do with urinary tract complications in spinal cord dysfunction provided by one of the panelists early in the project. This list identified periodicals reviewed for "Publications of Interest" in the Journal of Rehabilitation Research and Development (1997;34(4):515-6).

Table 15. Results of Screening Full-Length Articles (more...)

Table 15. Results of Screening Full-Length Articles

SCREENING OF FULL ARTICLES		Total
1 Total # of Articles Accepted for all Four Key Questions		306 1
Risks of Signs, Symptoms, and Laboratory Findings
Randomized controlled trial	2
Nonrandomized controlled trial	3
Prospective cohort	6
Retrospective cohort	5
Case control	0
Case series	12
Subtotal	28
Risk Factors for UTI
Randomized controlled trial	5
Nonrandomized controlled trial	10
Prospective cohort	24
Retrospective cohort	20
Case control	2
Case series	59
Unsure	1
Subtotal	121 2
Long-term followup
Randomized controlled trial	6
Nonrandomized controlled trial	5
Prospective cohort	21
Retrospective cohort	33
Case control	6
Case series	137
Unsure	4
Subtotal	212
Prophylaxis of UTI	28 3
2 Rejected			509 4
Invalid study population	183
No relevant outcome measure	40
Case report/editorial/review/letter	212
No key question addressed	56
Publication date prior to 1978 & addresses Key Question 2 or 9	83
3 No Translation Available			3
4 Unable to Obtain through InterLibrary Loan			35
	Total		853

1

The total number of articles accepted by key question (28+121+212+28=389) is greater than 306 because an article could deal with more than one key question

2

A total of 124 articles were screened as addressing this key question, because two articles were added from a subsequent search of the CINAHL database, and one article was identified by panelists and was not in the literature searches

3

A total of 34 articles were reviewed for this key question, because 6 articles were obtained from reference lists of other articles

4

The total number of reasons for rejecting an article (183+40+212+56+83=574) is greater than 509 because articles could be rejected for more than one reason

Of the 306 articles definitely accepted for further study, the breakdown by key question is provided in Table 15.

Case series study designs (63 articles) were subsequently excluded from further consideration for Key Question 2 because risk factors could not adequately be determined from a case series.

Process for Review and Comment on Draft Evidence Report

Table 16. Peer Reviewers

Table 16. Peer Reviewers

Name	Area of Expertise	Affiliation/Geographic Location
Diana Cardenas, MD	Physical medicine and rehabilitation	University of Washington Seattle, WA
John Montgomerie, MBChB, FRACP	Infectious diseases	Los Angeles, CA (formerly of Rancho Los Amigos Medical Center)
Marie Namey, RN, MSN	Nurse specialist in multiple sclerosis	The Cleveland Clinic Foundation Cleveland, OH
Victor J. DeFino, MD	Managed care medical director	Lifeguard HMO San Jose, CA
W.J.J. Assendelft, MD	Systematic review expert	Vrije University Amsterdam, The Netherlands
Martin Roland, MD	Systematic review/ methodology expert	University of Manchester Manchester, UK
Eric Hurwitz, PhD	Epidemiology	University of California, Los Angeles Los Angeles, CA
Meta-analysis reviewers: Vic Hasselblad, PhD		Duke University Durham, North Carolina
Allan R. Sampson, PhD		University of Pittsburgh Pittsburgh, Pennsylvania

Table 17. Instructions for Reviewing Draft Evidence (more...)

Table 17. Instructions for Reviewing Draft Evidence Report

Enclosed is a draft evidence report on the prevention and management of urinary tract infections in persons with neurogenic bladder. You may make your comments either directly on the draft evidence report, or on a separate sheet of paper. If you choose to record your comments on a separate piece of paper, please use the page and paragraph number to identify to which part of the report your comments pertain.

We ask that you consider the following questions while you read this report. We realized that some of the questions may not pertain to your area of expertise. Please feel free to comment only on the those that you feel most suited to answer.

Overall evaluation:

Is it clear what we did? You may agree or disagree with our methods, findings, or conclusions, but you should be able to understand what it is we did in order to produce this report.

Methodology:

Are the methods we used appropriate: For identifying the key questions of interest from the panel of technical experts? For searching and review the identified literature? For synthesizing the literature?

Evidence:

Did we miss any crucial pieces of information in our literature search? Does the evidence support the conclusions?

Utility:

Would you find this information to be useful if you had to develop clinical practice guidelines or medical review criteria for urinary tract infection in persons with neurogenic bladder?

In late June 1998 a draft of the evidence report was circulated to all of the technical experts who had met in December 1997. In addition, seven peer reviewers were identified. These included three technical experts who were nominated to the expert panel but were unable to attend the meeting, one managed care organization medical director, and three methodologists (Table 16). Information and questions for all reviewers to consider in reviewing the draft were provided (Table 17). Tabulation of reviewer comments and responses to questions and how they were addressed by the project team is provided in Appendix I of this report.

The section of the evidence report on the key question of signs, symptoms, and laboratory findings associated with risks to persons with spinal cord dysfunction was circulated to the panel and to peer reviewers in a later mailing. Comments and the center's responses to suggestions are in Appendix II.

In addition, two experts in meta-analysis (Vic Hasselblad, PhD, and Allan R. Sampson, PhD) reviewed and provided written comments on that section of the report.(Table 16; prophylaxis of UTI)

Defining a Urinary Tract Infection: Signs, Symptoms, and Laboratory Findings Associated With Risks to Persons With Spinal Cord Dysfunction

Development of Causal Pathways

Table 18. Causal Pathway for Signs, Symptoms, and Laboratory (more...)

Table 18. Causal Pathway for Signs, Symptoms, and Laboratory Findings Associated With Risks to Persons With Spinal Cord Dysfunction

1

Bacteriuria refers to both the amount or extent of bacteriuria and the organism or type of organism.

On re-review of the key question and the associated literature designated as addressing that key question, the project team met again with the onsite technical experts and discussed the desirability of developing a causal pathway for this key question, because it was broad and the existing literature addressed different subcomponents of the broader question. A pathway was developed and reviewed by those experts, then subsequently revised based on their comments (Table 18). This pathway served as the framework for the subsequent assessment of the literature and the development of evidence tables addressing this question:

"What combinations of symptoms, signs, and urinary laboratory findings are associated with risks to the person with neurogenic bladder due to spinal cord dysfunction?
Potential factors to consider are drainage method, sex, and how the sample was obtained."

Of note, some aspects of the pathway overlap with the key question on the impact of prophylactic agents, which is presented later in the report as a meta-analysis.

Table 19. Influence of Sample Collection Method and (more...)

Table 19. Influence of Sample Collection Method and Source on Measurement of Bacteriuria

The method of attainment of the sample for laboratory analysis affected the measurement of bacteriuria (with or without signs, symptoms, and other laboratory findings) and, potentially, subsequent decisionmaking. A separate causal pathway was developed for this component of the key question that concerned the relationships between method of obtaining the sample and measurement of the extent of bacteriuria (Table 19).

Initial and Subsequent Literature Screens

There were 28 studies that passed our initial screening as being potentially relevant to components of either the primary causal pathway for this key question or the sample collection method pathway.

Table 20. 18 Studies Further Excluded from Key Question (more...)

Table 20. 18 Studies Further Excluded from Key Question on Signs, Symptoms, and Laboratory Findings Associated with Risks to Persons with Spinal Cord Dysfunction

Author		Year	Reason for Exclusion
Anderson, Hatami-Tehrani		1979		Compares different techniques for cultivating bacteria
Darouiche, Preibe, Clarridge		1997		Compares limited vs. full microbiological studies to manage symptomatic polymicrobial UTI's
Donovan, Hull, Rossi		1996		Assesses whether recolonization is due to same or different strains (by plasmid analysis)
Fabrizio, Chancellor, Rivas et al.		1996		Describes utility of renal scintigraphy for differentiating fever due to pyelonephritis from that due to other causes
Florez Alia, Soria Fernandez de Cordoba, Diaz Pienna		1978		Does not address risk (relationship between lithiasis and current bacterial strain)
Giroux, Perkash		1985		Addresses a different issue (localization of infection)
Hashitani, Kimoto, Iwatsubo et al.		1992		Describes diagnostic utility of leukocyte esterase test for bacteriuria
Katz, Greenstein, Midha et al.		1994		Addresses a different issue
Kuhlemeier, Lloyd, Stover		1983		Addresses a different issue (localization of infection to kidney vs. bladder)
Lewis, Corrion, Lockhart et al.		1984		Addresses a different issue
Lindan		1981		Addresses a different issue (localization of infection to upper tract)
Menon, Tan		1992		Does not measure symptoms or signs
Montgomerie, McCary, Bennett et al.		1997		Compares urethral + urine culture in women
Newman, Price, Ederer		1980		Addresses a different issue (localization of infection)
Oshima, Masu		1988		Does not measure symptoms or signs
Thorley, Barbin, Reinarz		1978		Addresses a different issue (prevalence of antibody-coated bacteria in different subgroups)
Tuel, Meythaler, Cross et al.		1990		Addresses a different issue (cost-effectiveness of screening for infection by nursing staff)
Tysnes, Krokeide, Bjerke et al.		1996		Does not measure symptoms or signs

Of these 28 articles, 8 studies addressed parts of the main causal pathway, primarily the relationships between symptoms, pyuria, and bacteriuria, and two studies addressed the impact of how the sample was obtained on measurement of bacteriuria. Data from these 10 studies were abstracted and entered into Evidence Tables. The other 18 of these 28 studies were excluded for various reasons (Table 20).

In addition, all titles of studies initially classified during the full-length article review as addressing appropriate long-term followup of patients with neurogenic bladder due to spinal cord dysfunction (Key Question 9) and accepted for full-length article review were re-reviewed by one project investigator to identify articles addressing risks of signs, symptoms, and laboratory findings in this population associated with lower and upper-tract morbidities. Of the 35 articles identified from this title review, the full-length articles were then reviewed, and 7 of those 35 studies were incorporated into Evidence Tables for this key question on risks associated with signs, symptoms, and laboratory findings. These seven studies were included because they measured and analyzed the relationships between selected symptoms or laboratory findings, and outcomes of lower or upper tract morbidity.

Thus, a total of 15 articles (8 from the original list of articles designated for review for this key question and 7 from articles originally designated for review for the key question on long-term followup) are analyzed with respect to the key question on significance of signs, symptoms, and laboratory findings. Another two articles report relationships between sample collection method and source, and measurement of bacteriuria.

Risk Factors for Urinary Tract Infection

Conceptual Model

Table 21. Causal Pathway for Risk Factors for Recurrent (more...)

Table 21. Causal Pathway for Risk Factors for Recurrent UTI

Table 21 presents the conceptual model for the key question on risk factors for UTI and forms the basis for the organization of this synthesis.

Initial Literature Screen

Table 22. Studies Excluded from Risk Factors for Recurrent (more...)

Table 22. Studies Excluded from Risk Factors for Recurrent UTI

Author	Year	Reason for Exclusion
Awad, Gajewski, Sogbein et al.	1984	Cross-sectional study design; outcome is urinary symptoms; high likelihood of temporal ambiguity between risk factor and outcome
Barkin, Dolfin, Herschorn et al.	1983	Risk factor studied is urinary drainage; data and comparisons presented do not allow the effect of different drainage techniques to be isolated
Barnes, Timoney, Moulas et al.	1992	Cross-sectional study design; high likelihood of temporal ambiguity between risk factor and outcome
Blaivas, Holland, Giesser et al.	1984a	No data or statistical test presented
Chai, Chung, Belville et al.	1995	Risk factor studied is urinary drainage; study design is case series with comparison to published data
Dmochowski, Ganabathi, Leach	1995	Outcome was a composite measure; data for urinary tract infections alone not presented
Gallien, Lerbot-Le Borgne, Niclolas et al.	1995	Cross-sectional study; high likelihood of temporal ambiguity between risk factor and outcome
Giacobini, Cruciani, Fagiola et al.	1982	Case series
Goepel, Stoher, Burgdorfer et al.	1996	No data or statistical test presented
Gotoh, Yoshikawa, Otani et al.	1990	Case series
Greenstein, Rucker, Katz	1992	No comparison group; outcome is upper urinary tract changes
Hachen	1980	Case series
Hackler, Hall, Zampieri	1989	Cross-sectional study design; high likelihood of temporal ambiguity between risk factor and outcome
Hjeltnes, Jansen	1990	Cross-sectional study design; high likelihood of temporal ambiguity between risk factor and outcome
Jackson, DeVivo	1992	Outcomes are effective renal plasma flow and urologic complications
Jensen	1981	Pre/post study design for test of efficacy
Kim, Bird, Priebe et al.	1997	Cross-sectional study design; high likelihood of temporal ambiguity between risk factor and outcome
Koyanagi, Togashi, Mora et al.	1990	Study of drug therapy on urinary flow; no UTI outcomes reported
Kuhlemeier, Lloyd, Stover	1985	Outcome is effective renal plasma flow
Kuhn, Rist, Zacch	1991	No comparison group
Kumazawa, Yamashita, Kitadi et al.	1992	Study of drug therapy on urinary flow; no UTI outcomes reported
Kumazawa, Kimoto, Yamashita et al.	1997	Study of drug therapy on urinary flow; no UTI outcomes reported
Kwias, Aniszcenko, Stryla	1987	No tests were conducted to relate risk factors that were measured to UTI outcome
Lavrinenko	1984	No UTI outcomes reported
Liguori, Cardenas, Ullrich	1997	Cross-sectional study design; high likelihood of temporal ambiguity between risk factor and outcome
Llyod, Zervos, Mahayai et al.	1998	Outcome is enterococcal UTI; risk factors reported are for enterococcal UTI vs. Other UTI; no risk factors for UTI vs. no UTI
Madersbacher, Stohrer, Richter et al.	1995	Study of drug therapy on urinary flow; no UTI outcomes reported
Maynard, Diokho	1982	No comparison group
Maynard, Glass	1987	Risk factor studied is urinary drainage; study design is retrospective cohort
Newman, Price	1985	Outcomes are glomerular filtration rate and urinary morphologic changes
Oshima, Masu	1988	Cross-sectional study; high likelihood of temporal ambiguity between risk factor and outcome
Pansadoro, Pulone	1982	Study of drug therapy on urinary flow; no UTI outcomes reported
Perkash, Giroux	1993	No comparison group
Perrouin-Verbe, Labat, Richard et al.	1995	No data or statistical tests presented
Philp, Thomas	1980	Pre/post study design for test of efficacy
Polito, Carazeni, Villanova et al.	1990	No data or statistical tests presented
Porru, Campus, Garau et al.	1997	Cross-sectional study; no UTI outcomes reported
Ruutu	1985	No data or statistical tests presented
Sekar, Wallace, Waites	1997	Outcome is effective renal plasma flow
Shingleton, Bodner	1993	Outcomes are upper tract complications
Singh, Thomas	1997	Definition of UTI appears to be different among groups
Song, Fam, Lee et al.	1984	Definition of outcome is inadequate to determine study design; UTI outcome appears to be cross-sectional
Stotts	1986	Outcome is hospitalizations for urinary tract complications; data for UTI alone not presented
Van Kerrebroeck, Van der Aa Bosch et al.	1997	Case series
Waller, Jonsson, Norlen et al.	1995	Case series
Wyndaele, Males	1990	Study population is primarily acute patients

There were 68 studies that passed our initial screening as being potentially relevant and not having a case series study design. Of these, 46 studies were excluded for various reasons, which are presented in Table 22. The remaining 22 studies had data abtracted and entered into Evidence Tables 6, 7, 8, 9, 10, 11, 12.

Prophylaxis of Urinary Tract Infection (Supplemental Analysis)

Review of the key questions indicated that prophylaxis of UTI was most likely to have evidence that would be appropriate for meta-analysis, both because of the nature of the question and because of the availability of enough studies of adequate quality.

Methodology for Supplemental Analysis

Literature Search

The literature search identified 28 studies as potentially relevant to this Key Question 8, and a reference list search identified an additional 6 studies. All 34 studies were retrieved and categorized by study design, patient population, treatments, and outcomes. The studies had to meet all of the four following criteria in order to be included in the meta-analysis:

• 1

  They had to be controlled clinical trials.

• 2

  They had to study patients with neurogenic bladder.

• 3

  The treatment had to be given to prevent recurrent urinary tract infection.

• 4

  The outcome measured had to be either bacteriuria or urinary tract infection.

Table 23. Studies Excluded from Meta-Analysis

Table 23. Studies Excluded from Meta-Analysis

Author		Year	Reason for Exclusion
Fried, Goetz, Potts-Nulty et al.		1996		Prophylaxis for urinary procedures
Hetey, Kleinberg, Parker et al.		1980		Outcome measure is urine pH, not UTI 1
Sapico, Lindquist, Montgomerie et al.		1980		Aminoglycoside used as treatment for infection, not prophylaxis
Griffith, Khonsari, Skurnick et al.		1988		Outcome is urinary stone growth, not UTI
Merritt, Erickson, Opitz		1982		Study design appears to be cohort
Lamid		1983		Outcome measure is urine pH, not UTI
Stamm, Counts, McKevitt et al.		1982		Patient population is women with recurrent UTI but without spinal cord dysfunction
Pedersen, Horbov, Biering-Sorenson et al.		1990		Case series
Brumfitt, Hamilton-Miller, Gargan et al.		1983		Patient population is women with recurrent UTI but without spinal cord dysfunction
Britt, Garibaldi, Miller et al.		1977		Patient population is women undergoing gynecologic procedure
Thorsteinsson, Keys		1983		Study reported in abstract only, with insufficient data to characterize it
Vainrub, Musher		1977		Case series
Dubo, Mallory, Ramsey et al.		1982		Study reported in abstract only, with insufficient data to characterize it; attempt to contact first author unsuccessful (unable to locate author)
Rogers		1991		Case series
Foote, Bennett, Cowles et al.		1994		Letter; no data

1

UTI = urinary tract infection

A project investigator reviewed and categorized all studies. In total, 19 met our initial inclusion criteria. Table 23 details the reasons for excluding the remaining 15 studies from the meta-analysis.

Project staff and onsite technical experts (Drs. Bennett and Dobkin from UCLA) also examined the impact of heterogeneity on the analysis, because several of the clinical trials differed in the nature of their patient population: some studied acute (<90 days after spinal cord injury) rather than non-acute (chronic) patients. The onsite technical experts considered these two patient populations to be sufficiently different that statistical combination of results across the non-acute and acute populations would not be valid. Nevertheless, the onsite technical experts believed that the data on acute patients would help inform decisions about prophylaxis for non-acute (or chronic) patients. Therefore, they requested that we perform separate syntheses for acute and chronic patients.

Evidence Tables 13 and 14 list relevant characteristics of studies of acute and non-acute patients, respectively, including the population, specific treatments, the sizes of the study groups, the outcomes measured, and the results. The studies listed in Evidence Tables 13 and 14 were also graded by one project investigator according to Jadad quality criteria and the adequacy of concealment of allocation (Moher, Pham, Jones et al., 1998).

Analyses

Based on the meeting with the technical experts, we performed the following two supplemental analyses:

• 1

  Efficacy of any oral antibiotic therapy, stratified by acute versus chronic patients; and

• 2

  Efficacy of specific oral antibiotics, stratified by acute versus chronic patients.

In addition, our discussion revealed the comparability of various dosages of specific drugs, enabling us to define a priori specific treatment (drug) subgroups to analyze.

Analytical Subgroups

Table 24. Analysis Subgroups

Table 24. Analysis Subgroups

In summary and as shown in Table 24, we used a hierarchical stratification to define our subgroups for analysis. We analyzed the findings for acute and non-acute patients separately, and then within each patient group, we analyzed asymptomatic and symptomatic results separately. Within each of these four strata, we analyzed overall drug versus no drug (control) comparisons, and then specific drug versus no drug (control) comparisons for three drugs (Nitrofurantoin, Methenamine, Trimethoprim) and for a group of bladder instillation methods.

Obtaining Needed Data From Principal Investigators

In the process of constructing the evidence tables, we identified five articles that contained insufficient information to complete the tables. We attempted to contact the principal investigators of all of these five studies to clarify issues of study design and nature of the patient population, or to obtain additional data.

We were able to obtain the needed missing information for two of the studies:

• In their 1994 study, Reid, Dafoe, Delaney et al. did not specify whether the study population was acute or non-acute spinal cord injury patients, nor whether the study design was randomized. After contacting the primary author, Dr. Reid, we learned that the study sample was acute and that the study was randomized.

• In their 1987 study, Mohler, Cowen, and Flanigan did not specify whether their randomized study design was double-blind or not. A conversation with the primary author, Dr. Mohler, revealed that the study design was, in fact, double-blind.

We were unable to secure the missing information for the three other studies:

• In their 1985 study, Kuhlemeier, Stover, and Lloyd reported the results per trial. Our meta-analysis required the data to be per patient, so we attempted to contact the primary author to obtain this information. We were subsequently informed by one of the co-authors, Dr. Stover, that the original data for this study are no longer available.

• In their 1991 study, Banovac, Wade, Gonzalez et al. did not specify whether their study design was randomized or not. Dr. Banovac has not returned several contact phone calls that were placed to his office

• In their 1994 study, Biering-Sorensen, Hoiby, Nordenbo et al. did not specify their patient population as inpatient or outpatient in their cross-over outcome data. Dr. Biering-Sorensen lives in Denmark, and the project team was unable to reach him despite contacting the institution at which he was affiliated at the time the study was published.

Studies and Comparisons

In total, 11 studies on acute patients and 8 studies on non-acute patients met our inclusion criteria (see above for criteria). Each study had two or more patient groups for which results were available. For example, in Evidence Table 13, Maizels and Schaeffer had three patient groups, each receiving different treatments. In Evidence Table 14, the Kuhlemeier, Stover, and Lloyd (1985) study had five treatment groups receiving different drugs and one control group, which received nothing. For each study, we defined one or more "comparisons" between two patient groups.

Table 25. Definitions of Comparisons for Acute Patients (more...)

Table 25. Definitions of Comparisons for Acute Patients

Asymptomatic Comparisons
Author	Comparison	Treatment Group Received	Control Group Received
Stover	1	methenamine hippurate (1 gm) BID	ascorbic acid (1 gm) QID
Anderson	2	neomycin + polymycin B bladder irrigant after each catheterization	nothing
Anderson	3	nitrofurantoin (100 mg) QD	nothing
Anderson	4	nitrofurantoin + neomycin + polymyxin B	nothing
Banovac	5	methenamine hippurate (1 gm) twice daily	nothing
Maizels	6	saline instillation into drainage bag	conventional closed drainage
Maizels	7	hydrogen peroxide instillation into drainage bag	saline instillation into drainage bag
Maizels	8	hydrogen peroxide instillation into drainage bag	conventional closed drainage
Pearman	9	kanamycin + colistin bladder irrigation	trisdine bladder irrigation
Krebs	10	methenamine mandelate (2 gm) 4 times a day + hemiacidrin bladder instillation at each catheterization	nothing
Lindan	11	nitrofurantoin (50 mg) QD	nothing
Schaeffer	12	silver oxide-coated drainage system	conventional closed drainage
Maynard	13	trimethoprim (80 mg) + sulfamethoxazole (400 mg) daily	nothing
Reid	14	trimethoprim (160 mg) + sulfamethoxazole (800 mg) daily	nothing
Symptomatic Comparisons
Author	Comparison	Treatment Group Received	Control Group Received
Gribble	15	trimethoprim (40 mg) + sulfamethoxazole (200 mg) daily	placebo
Krebs	16	methenamine madelate (2 gm) 4 times a day + hemiacidrin bladder instillation at each catheterization	nothing
Maynard	17	trimethoprim (80 mg) + sulfamethoxazole (400 mg)	nothing

BID = twice a day; QD = once a day; QID = four times a day; mg = miligram

Table 26. Definitions of Comparisons for Non-Acute (more...)

Table 26. Definitions of Comparisons for Non-Acute Patients

Asymptomatic Comparisons
Author	Comparison	Treatment Group Received	Control Group Received
Sandock	18	trimethoprim (80 mg) & sulfamethoxazole (400 mg)	placebo
Schlager	19	nitrofurantoin (25 mg) for children < 25 kg or (50 mg) QD	placebo
Hachen	20	immunotherapy, 6 mg of extract daily	placebo
Kuhlemeier	21	nitrofurantoin (50 mg) TID	nothing
Kuhlemeier	22	ascorbic acid (1 gm) QID	nothing
Kuhlemeier	23	nalidixic acid (500 mg) QID	nothing
Kuhlemeier	24	methenamine hippurate (1 gm) BID	nothing
Kuhlemeier	25	trimethoprim (80 mg) & sulfamethoxazole (400 mg) BID	nothing
Kuhlemeier	26	weighted average of treatment effects across comparisons 21, 23, 24, and 25	nothing
Kevorkian	27	methenamine mandeltate + ammonium chloride(each 1 gm) QID	placebo
Duffy	28	nitrofurantoin (100 mg) daily	placebo
Mohler	29	trimethoprim (160 mg) & sulfamethoxazole (800 mg)	placebo
Mohler	30	trimethoprim (160 mg) & sulfamethoxazole (800 mg)	placebo
Sandock	31	trimethoprim (80 mg) & sulfamethoxazole (400 mg) daily	nothing
Schlager	32	nitrofurantoin (25 mg) for children < 25 kg or (50 mg) QD	placebo
Biering- Sorensen	33	ciprofloxacin (100 mg) nightly	placebo

BID = twice a day; QD = once a day; QID = four times a day; mg = miligram

These comparisons are shown in Tables 25 and 26. Within each outcome strata (asymptomatic and symptomatic), the comparisons are ordered by outcome definition (as shown in Evidence Tables 13 and 14) from the least restrictive to the most restrictive, and then alphabetically for those studies which have the same outcome definition.

Generally a comparison is defined between each treatment in a study and the control group which received a placebo or nothing. For those studies that did not have a standard control group, a specific patient group was designated the control group based on clinical judgment. For example in Table 25, the control group for Pearman, Bailey, and Harper (1988) was "trisdine bladder irrigation."

We conducted our different subgroup analyses as described in Table 24 on the comparison level, that is the unit of observation in the meta-analysis described in the following is a comparison, not a study. In the specific drug and drainage method subgroup analyses, no study had more than one comparison available for each analysis. However, for the "Any Drug Versus No Drug" analysis, some studies had more than one comparison available. We did not want a single study to enter a subgroup analysis more than once, or otherwise that study's control group patients would be double-counted, i.e., allowed to enter the analysis for every comparison that study contributed. For all but one study that had more than one comparison available for an analysis, we chose the most clinically relevant comparison. For one study (Kuhlemeier, Stover, and Lloyd, 1985), we produced a summary statistic as described below that combined results across all relevant treatment groups.

Extracting Data From Cross-Over Studies

Four studies, all on non-acute patients (Biering-Sorensen, Hoiby, Nordenbo et al., 1994; Duffy and Smith, 1982; Hachen, 1990; Schalger, Anderson, Trudell et al., 1998), employed cross-over designs as noted in Evidence Table 14. As discussed above in the section on obtaining data from principal investigators, we would like to conduct analysis on the results from the study period prior to cross-over only. This strategy of using only the "phase one" data from cross-over studies is ideal as it reduces bias. For example, for the group of patients who receive treatment and then placebo, this strategy does not allow the possible lingering effects of treatment to contaminate the observed effects from the subsequent assessment of the placebo effect. As noted in Evidence Table 14, for only one study (Hachen, 1990) were we able to obtain data stratified by study phase, and consequently only for this study could we include data from the study period prior to the cross-over. For the three remaining cross-over studies, we used the complete data, which aggregated results across both phases of the studies.

Choice of Outcome Measure

The data varied widely across studies. To perform a meaningful analysis, we had to reconcile discrepancies in the following issues:

• Outcome definition: Studies varied in their definitions of both asymptomatic and symptomatic UTIs. For example, some studies required a colonization count of 10⁴ colonies per ml, while others required 10⁵ colonies per ml.

• Reported statistics: Some studies reported the number of infections among a certain number of cultures, others reported the number of patients who had at least one infection over a certain time period, while still others reported statistics about the time to first infection. To summarize the results across all studies, we needed to choose a common outcome measure for conversion of all studies.

• Measurement differences: The length of the study's observation period varied, as well as the frequency with which observations were made. Symptomatic UTIs were rare as compared with asymptomatic infections, and occurrence of symptoms tended to motivate observations. For example, a patient develops a relevant symptom, such as a fever, goes to see the doctor, and a symptomatic UTI is diagnosed. Asymptomatic UTIs, in contrast, do not manifest themselves to the patient or physician. As a result, investigators generally take periodic cultures (e.g., weekly) to determine whether an asymptomatic infection has occurred. Measurements taken in this manner may miss an asymptomatic culture that resolves on its own. Sometimes studies focused on asymptomatic infections alone but mentioned that they recorded a few symptomatic infections. We treated such studies as asymptomatic studies, as symptomatic infections were rare, and we could not distinguish which infections were of which type.

• Treatment of infections: If a patient experienced a symptomatic UTI, they were administered a clinically appropriate antibiotic for UTI treatment, in addition to or in place of the antibiotic prophylaxis that was being tested. Studies varied as to whether they treated asymptomatic infections and in how clearly they described their treatment protocols for when a patient experienced an asymptomatic or symptomatic UTI. In addition, we could often not determine whether the patients who were treated for a UTI during the course of the study were removed from the population at risk, i.e., from measurement and analysis for the treatment period.

Given this disparity across studies, we decided upon the weekly infection rate as our common outcome measure. For example, if the weekly infection rate is 0.5, then half an infection occurs per week on average, or one infection occurs every two weeks on average. We assumed that infections occur as a Poisson process and that the weekly rate is the mean parameter of this process. As a result, we implicitly assumed that the time between infections is distributed exponentially. From a clinical perspective, the main Poisson assumption is debatable. The assumption implies that infections occur randomly, and that the time between infections is "memoryless." In other words, it assumes that the likelihood of incurring an infection in the next week is the same, whether or not the patient just had one.

Given the lack of natural history data that might allow us to make our infection model more clinically accurate, in our judgment the Poisson model was a good first approximation to the underlying process. The Poisson is the simplest count model and is known to be robust. Without additional information, we cannot apply a more complex model such as a Negative Binomial, which would account for both the within-person and between-person variation. For example, based on our clinical understanding of the physical process, we hypothesized that the between person variation in rates may occur because certain patients have experienced many past infections and as a result have a high rate of infection in the future while patients who have experienced few past infections have a low rate of infection in the future. Thus the choice of the Poisson model probably underestimates the true variance. However, our use of a random effects model to construct a pooled estimate, and our choice of the number of patients as the sample size in the standard error calculations, both of which are discussed in the following, will add additional variation to our estimates.

To arrive at our common treatment effectiveness measure or "effect size," we estimated the weekly infection rates for the treatment and control groups in a particular comparison and then took the difference between them (treatment minus control). Therefore, we converted the study's reported statistics into three measures for each comparison (specific details on the calculations are found below):

• 1

  The estimated weekly infection rate for the control group;

• 2

  The estimated weekly infection rate for the treatment group; and

• 3

  The estimated difference between these weekly infection rates.

If treatment is effective, we expect that the difference between the weekly infection rates will be negative, indicating that treatment is associated with a decreased weekly infection rate of UTIs. In addition, we estimated the standard errors of the group rates and the difference and constructed 95 percent confidence intervals for each estimate.

Calculating the outcome measure

We faced the challenge of converting each study's reported statistics into a weekly infection rate. The studies generally reported a Poisson infection rate for symptomatic UTIs. Although they occasionally reported daily or monthly rather than weekly infection rates, we were able to use the standard facts about the Poisson distribution to produce an estimated weekly rate and its standard error.

In computing all standard errors, we had to determine the appropriate sample size for each study. For example, if a study had 10 patients, and weekly cultures were collected for 10 weeks, producing 100 culture results, what is the appropriate sample size: 10 or 100? Given the correlation between cultures taken on the same patient, choosing 100 as the sample size is extreme. However, we clearly have more data available than 10 would suggest. The correct approach would be to estimate the design effect due to the clustering of cultures within-patient. This adjustment would be appropriate if we were conducting patient-level analysis within a single study. Unfortunately, we do not have data available to perform this adjustment. Thus, regardless of the type of outcome presented in an individual study, the Poisson rate as described above or the measures described below, we used the patient sample size for our standard error calculations. This choice is clearly conservative, as it will overestimate the standard error. However, we concluded that this decision was appropriate for several reasons. First, we did not want to allow a study with many measurements to overwhelm another study with the same number of patients but fewer measurements in the combined calculations. Second, our understanding of the clinical process we were modeling led us to believe that our Poisson model was not incorporating enough variance between patients to begin with. Third, although our random effects approach explained below will attempt to incorporate between-study variation, we acknowledge that this additional source of uncertainty might also be underrepresented in our approach.

Our choice of a common outcome measure and the way in which we constructed the standard errors may cause the individual study results we report to differ in terms of statistical significance from what is reported in the individual studies themselves. This may be due to differences based on the measure choice, which is similar to the way in which an odds ratio conclusion may be different from a risk difference one. In addition, our conservative standard error estimate may also reduce our reported significance, especially for studies that had few patients but many culture observations. Although these choices affect our results, to make comparisons across studies, we had to choose a clinically relevant statistic that could be obtained from all studies. We also had to determine the relative degree of uncertainty that studies had in relation to each other.

For asymptomatic infections, studies generally reported the number of infected cultures observed over a particular time period at particular intervals. For example, a study might report that patients were observed for 8 weeks, that a culture was taken weekly on each patient, and then give the total number of infected cultures. In our calculations, we used the ratio of the number of infected cultures to the total number of cultures as an estimated Binomial mean; that is, we determined the number of weekly cultures or trials that were infected. The estimated Binomial mean is thus an estimate of the probability that at least one infection occurred in a week. Using the Binomial rather than the Poisson in studies with this type of observation protocol takes into account the possibility that infections may have occurred during the week but were not recorded by the culture taken at the end of the week, as they had resolved on their own. To estimate the Poisson rate from the Binomial data, we solve the following equation:

where Y is the number of infections in a week, which is distributed Poisson (), with the Poisson parameter or the mean weekly rate;
x is the number of infected cultures; and
n is the total number of cultures.

This equation produces an estimate of the weekly mean rate . We determined the standard error for the estimated weekly mean rate by using the Delta Method (Mood, Graybill, and Boes, 1974).

Some studies reported the median or mean time until infection, either asymptomatic or symptomatic. If the median was reported, we first estimated the mean time until infection using the relationship between the median and mean for an Exponential random variable, namely that the mean is equal to the median divided by the natural logarithm of two (Johnson and Kotz, 1970). We then used the fact that the time between Poisson events is distributed Exponentially to estimate the mean weekly rate of the related Poisson distribution, i.e., the mean of the Poisson is the inverse of the mean of the Exponential.

Our estimate of the standard error of the Poisson rate when the median time until infection was reported required several steps. First we calculated the variance of the median time until infection using variances of quantiles results that rely on the quantile ordinate value as discussed in Kendall and Stuart (1977). If the mean time until infection was reported, we estimated its variance as the variance of an exponential divided by the sample size. In both the median and mean situations, we then used the Delta Method for an inverse function to determine the variance of the Poisson mean.

After calculating the weekly rates, we arrived at an effect size by taking the difference between the treatment and control weekly infection rates. We calculated the standard error of the effect size by taking the square root of the sum of the variances of the two weekly rates. For the individual studies, we constructed 95 percent confidence intervals for the treatment and control weekly rates, respectively, based on the Inverse Incomplete Gamma distribution. In the calculation of these intervals, we used the patient sample size and the estimated rate to calculate an expected number of infections, and then calculated an exact confidence interval in the statistical package "Intercooled Stata 5.0 for Windows 95" (1985-1996, Stata Corporation, College Station, TX). If the expected number of infections was a fraction, we rounded down to obtain the lower confidence limit and rounded up to obtain the upper confidence limit. All other confidence intervals were constructed using the standard Normal approach, that is, the bounds were plus/minus 1.96 times the standard error.

Statistical Pooling Method

We used the DerSimonian and Laird random effects model to pool effect sizes across studies (DerSimonian and Laird 1986). This method produces a summary measure that is a weighted mean. It weights each study's measure by the inverse of the sum of the within-study variance and the between-study variance. This approach allows both sampling variation and between-study heterogeneity to affect the pooled estimate. We present the pooled estimate and its 95 percent confidence interval both numerically and graphically.

In addition to the pooled estimate, we report the p-value for the chi-squared test of heterogeneity, which tests the null hypothesis that the individual study results are homogeneous (Laird and Mosteller 1990). Even if the test does not reject the null hypothesis, heterogeneity between studies may still exist, as the test is known to lack power to detect heterogeneity. Thus, we protect against spurious conclusions resulting from combining clinically heterogeneous patients or treatments in two ways regardless of the outcome of the chi-squared test of heterogeneity. First, we always fit a random effects model, which may incorporate some between-study variance even if the chi-squared test does not reject. Second, we perform subgroup analyses and sensitivity analyses to assess the impact of possible heterogeneity.

Data were prepared for analysis in the software packages "Microsoft Excel for Windows 95, Version 7.0a" (1985-1996,Microsoft Corporation) and "SAS System for Windows, Release 6.12" (1989-1996, SAS Institute Inc., Cary, NC). All pooling and graphical analyses were conducted in the beta-test version of the software package "MetaGraphs" (1998, Belmont Research, Inc. 84 Sherman Street, Cambridge, MA 02140) on a Windows 95 platform.

We considered conducting a meta-regression strategy in which we would model the relationship between the study effect sizes and covariates of the studies; for example, the baseline (control) rate of infection or the quality score. Such an analysis might well help us understand more thoroughly the reasons for heterogeneity across the studies, beyond just acknowledging and incorporating such heterogeneity in our pooled estimates via the random effects model and the examination of it via the stratified analysis we undertook. However, given the strong a priori hypothesis of our clinical experts that the four strata of acute or non-acute patients by asymptomatic or symptomatic infections results were so clinically dissimilar, we judged it inappropriate to combine results across these four strata in a main effects model. To fit a model with interactions of every covariate with these four strata would be impossible, given the small number of comparisons (observations) in the strata. Thus, we did not conduct a meta-regression.

Other Study Heterogeneity Issues

Although the random effects model attempts to model between-study heterogeneity, additional sources of heterogeneity that can be identified clinically should be considered when assessing the pooled results. Five dimensions of clinical heterogeneity should be taken into consideration. First, the number of cultures taken (if taken more frequently that the protocol states) will affect the incidence of asymptomatic bacteriuria. Second, the level of 'significant bacteriuria' used for diagnosis is usually clearly stated in these studies. However, less than significant levels may not mean the absence of infection. In some studies this is confirmed by repeat culture while in others it may not be. Third, the administration of antibiotics for various reasons (an upper respiratory infection or pneumonia, prophylaxis for surgery, or a skin infection, etc.) is very common in spinal cord injury patients. It would be desirable to understand whether antibiotics were administered for reasons other than UTI in some cases. Fourth, it is important to know the manner in which the microbiology lab cultured the urine (especially if the patient had received antibiotics), as the approach used could conceivably vary across studies. Finally, the effect of antibiotics may vary across patient subgroups depending upon level of injury, drainage method, and other patient characteristics.

Results for Supplemental Analysis

Individual Study Results

Table 27. Individual Comparison Results for Studies (more...)

Table 27. Individual Comparison Results for Studies on Acute Patients

	Comparison	Treatment Sample Size	Control Sample Size	Treatment Weekly Infection Rate	95% CI	Control Weekly Infection Rate	95% CI	Difference in Weekly Infection Rates (treatment-control)	95% CI
Asymptomatic
Stover	1	49	51	0.17	(0.07, 0.35)	0.69	(0.48, 0.98)	-0.52 1	(-0.93, -0.11)
Anderson	2	17	16	0.43	(0.17, 0.93)	0.47	(0.18, 0.99)	-0.04	(-0.35, 0.27)
Anderson	3	15	16	0.15	(0.02, 0.58)	0.47	(0.18, 0.99)	-0.31 1	(-0.55, -0.07)
Anderson	4	16	16	0.19	(0.04, 0.64)	0.47	(0.18, 0.99)	-0.28 1	(-0.52, -0.03)
Banovac	5	34	22	0.27	(0.12, 0.54)	0.86	(0.49, 1.35)	-0.59 1	(-1.11, -0.07)
Maizels	6	10	12	0.97	(0.41, 1.84)	1.08	(0.52, 1.85)	-0.11	(-1.89, 1.68)
Maizels	7	9	10	0.69	(0.24, 1.60)	0.97	(0.41, 1.84)	-0.28	(-1.84, 1.29)
Maizels	8	9	12	0.69	(0.24, 1.60)	1.08	(0.52, 1.85)	-0.39	(-1.97, 1.20)
Pearman	9	7	8	0.08	(0.00, 0.80)	0.09	(0.00, 0.70)	0.00	(-1.20, 1.20)
Krebs	10	20	20	0.42	(0.17, 0.85)	1.18	(0.73, 1.79)	-0.76 1	(-1.49, -0.03)
Lindan	11	31	29	0.02	(0.00, 0.18)	0.19	(0.06, 0.45)	-0.17	(-0.34, 0.01)
Schaeffer	12	41	33	0.13	(0.04, 0.32)	0.61	(0.37, 0.97)	-0.47 1	(-0.91, -0.03)
Maynard	13	23	27	0.13	(0.03, 0.45)	0.47	(0.23, 0.82)	-0.34	(-1.14, 0.47)
Reid	14	7	7	0.78	(0.23, 1.87)	1.14	(0.40, 2.25)	-0.36	(-1.71, 0.98)
Symptomatic
Gribble	15	66	60	0.01	(0.00, 0.08)	0.04	(0.01, 0.15)	-0.03	(-0.09, 0.02)
Krebs	16	20	20	0.02	(0.00, 0.28)	0.10	(0.01, 0.44)	-0.08	(-0.24, 0.07)
Maynard	17	23	27	0.03	(0.00, 0.24)	0.06	(0.00, 0.27)	-0.04	(-0.15, 0.08)
Asymptomatic
Sandock	18	20	23	0.28	(0.08, 0.65)	0.28	(0.10, 0.63)	0.00	(-0.34, 0.34)
Schlager	19	15	15	1.08	(0.61, 1.81)	1.35	(0.81, 2.14)	-0.28	(-1.38, 0.83)
Hachen	20	33	34	0.19	(0.07, 0.44)	0.32	(0.14, 0.59)	-0.14	(-0.30, 0.03)
Kuhlemeier	21	23	30	0.49	(0.24, 0.91)	0.49	(0.26, 0.82)	0.00	(-0.55, 0.55)
Kuhlemeier	22	19	30	0.69	(0.36, 1.24)	0.49	(0.26, 0.82)	0.21	(-0.53, 0.95)
Kuhlemeier	23	14	30	0.49	(0.16, 1.03)	0.49	(0.26, 0.82)	0.00	(-0.64, 0.64)
Kuhlemeier	24	23	30	0.49	(0.24, 0.91)	0.49	(0.26, 0.82)	0.00	(-0.55, 0.55)
Kuhlemeier	25	19	30	0.32	(0.12, 0.76)	0.49	(0.26, 0.82)	-0.16	(-0.63, 0.31)
Kuhlemeier	26	79	30	0.45	(0.31, 0.63)	0.49	(0.26, 0.82)	-0.04	(-0.43, 0.35)
Kevorkian	27	17	22	0.25	(0.06, 0.69)	0.66	(0.35, 1.12)	-0.41 1	(-0.80, -0.02)
Duffy	28	31	31	0.03	(0.00, 0.23)	0.09	(0.01, 0.28)	-0.06	(-0.13, 0.01)
Mohler	29	21	25	0.22	(0.05, 0.56)	0.18	(0.04, 0.47)	0.03	(-0.17, 0.24)
Symptomatic
Mohler	30	21	25	0.08	(0.00, 0.34)	0.13	(0.02, 0.41)	-0.05	(-0.24, 0.13)
Sandock	31	20	23	0.04	(0.00, 0.28)	0.04	(0.00, 0.24)	0.01	(-0.11, 0.13)
Schlager	32	15	15	0.03	(0.00, 0.37)	0.06	(0.00, 0.37)	-0.03	(-0.18, 0.12)
Biering-Sorenson	33	21	21	0.01	(0.00, 0.27)	0.12	(0.01, 0.42)	-0.11	(-0.26, 0.05)

1

p <0.05, the 95% confidence interval does not contain 0. Rates, differences in rates, and confidence limits have been rounded to two decimal places.

Table 27 reports on statistics for each comparison made within a study. This table, which shows all the studies results converted to a common outcome measure, should be examined in conjunction with Evidence Tables 13 and 14. These two earlier tables show the results of the individual studies using the original presentation of the data. As discussed previously, the conversion to a common outcome measure is a necessary step in pooling across studies that allows a common statistical perspective on the effectiveness of the treatment but may mask important individual study results. Thus, both the individual study results and the pooled results should be considered in drawing conclusions.

For example, Anderson and Hatami-Tehrani (1979) report on three treatment groups. Thus, this study has three comparisons listed in Table 27 (#2, 3, 4). We have ordered the comparisons by outcome definition from the least restrictive to the most restrictive, and then alphabetically within the four strata (acute/non-acute patients by asymptomatic/symptomatic infections).

One particular study to note is Kuhlemeier. This cross-over study reported on five treatments (comparisons #21-24). For the main drug versus no drug analysis, we produced a weighted treatment infection rate by combining the treatment infection rates across the four drug comparisons for this study using weights equal to the sample size of each treatment group. We excluded the fifth treatment of ascorbic acid (comparison #22: See Table 26) from this summary statistic. The resulting combined comparison is listed as comparison #25.

Table 27 presents the treatment and control weekly infection rates for each comparison. Understanding the clinical relevance of a weekly infection rate is important. For example, if the control weekly infection rate is 0.69 (as it is for asymptomatic infections among acute patients in the control group studied by Stover, 95 percent confidence interval (0.48, 0.98)), this means that on average such a patient has 0.69 infections per week. This is equivalent to one infection every 1.5 (1/0.69) weeks on average.

The control infection rate is a gauge of the underlying propensity for infections in the study populations, and these rates range quite widely:

• 0.08 to 1.14 infections per week for acute asymptomatic patients;

• 0.04 to 0.10 infections per week for acute symptomatic patients;

• 0.09 to 1.35 infections per week for non-acute asymptomatic patients; and

• 0.04 to 0.13 infections per week for non-acute symptomatic patients.

Our clinical experts suggested that these varying underlying rates can be considered a priori evidence of population heterogeneity. This possible source of heterogeneity is discussed in more detail when the pooled analysis is presented.

Table 27 also reports the effect size for each study, that is, the difference between the treatment and control weekly infection rates for that particular study. For example, in comparison 1 from Stover, the difference between the rates is -0.52 with a 95 percent confidence interval of (-0.93, -0.11). The negative value suggests that a decrease in infection rate is associated with use of prophylaxis antibiotics. The absolute value of the decrease is 0.52, which can be interpreted as follows: an individual patient receiving the treatment would experience on average one less infection every 2 weeks than a patient in the control group.

Finally, Table 27 also reports the 95 percent confidence interval for each rate and denotes statistically significant (p<0.05) effect sizes (as compared to zero) with an asterisk.

Any Drug Versus No Drug Results

Table 28. Funnel Plot

Table 28. Funnel Plot

"AA" is an acute patients asymptomatic infections comparison; "AS" is an acute patients symptomatic comparison; "NA" is a non-acute patients asymptomatic infections comparison; and "NS" is a non-acute patients symptomatic infections comparison. Only comparisons in the Any Drug Versus No Drug analysis are plotted (Table 29).

Six comparisons (#1, 3, 5, 11, 13, 14) on acute patients asymptomatic infections, two comparisons (#15, 17) on acute patients symptomatic infections, six comparisons (#18, 19, 26, 27, 28, 29) on non-acute patients, asymptomatic infections, and four comparisons (#30, 31, 32, 33) are included in the Any Drug Versus No Drug analysis (see Tables 25 and 26 for comparison definitions). Table 28 shows a "funnel plot" for these comparisons (Light and Pillemer 1984). Each comparison is represented by a plotting symbol with "AA" for acute patients asymptomatic infections comparisons; "AS" for acute patients symptomatic comparisons; "NA" for non-acute patients asymptomatic infections comparisons; and "NS" for non-acute patients symptomatic infections comparisons. The purpose of a funnel display is to examine whether publication bias may have occurred, which would be indicated by a lack of studies with nonsignificant results. Table 28 does not indicate a paucity of studies with a difference in weekly infection rates (the effect size) of zero. Therefore, this funnel plot does not support a significant bias due to lack of publication of non-significant results. It does demonstrate the larger effect size for the acute asymptomatic comparisons, indicated by the number of "AA" points to the left.

Table 29. Pooled Baseline Risk For Any Drug Versus (more...)

Table 29. Pooled Baseline Risk For Any Drug Versus No Drug Analyses (continued)

Subgroup	Comments	Control Group	Summary Control Weekly Infection Rates	95% CI	Chi-square Test for Heterogeneity p-value
Acute Asymptomatic	All	1, 3, 5, 11, 13, 14	0.52	(0.27, 0.77)	0.02 1
Acute Asymptomatic #2	without 11	1, 3, 5, 13, 14	0.58	(0.40, 0.75)	0.46
Acute Asymptomatic	104/ml or 103/ml	1, 3, 5	0.59	(0.38, 0.81)	0.29
Acute Asymptomatic	105/ml	11	0.19	(0.06, 0.45)	NA
Acute Asymptomatic	other	13, 14	0.67	(0.07, 1.27)	0.31
Acute Symptomatic	All	14, 16	0.04	(0.00, 0.09)	0.67
Non-acute Asymptomatic	All	18, 19, 26, 27, 28, 29	0.33	(0.14, 0.51)	0.002 1
Non-acute Asymptomatic	104/ml or 103/ml	18, 19	0.73	(0.00, 1.77)	0.02 1
Non-acute Asymptomatic	105/ml or 104/ml twice	26, 28, 28, 29	0.28	(0.09, 0.47)	0.002 1
Non-acute Symptomatic	All	30, 31, 32, 33	0.07	(0.01, 0.12)	0.59

1

p<0.05.

NS is not significant at p<0.05; NA is not applicable. Rates, differences in rates, and confidence limits have been rounded to two decimal places.

Table 29 presents the "Any Drug Versus No Drug" pooled analyses. For example, the pooled difference in rates for the Acute Asymptomatic stratum (the first line of the table) is -0.27 (p<0.05), with a 95 percent confidence interval ranging from -0.40 to -0.15. This value indicates that one patient would require 3.7 (1/0.27) weeks of prophylactic antibiotics to prevent one asymptomatic UTI. Obviously, the possible side effects that might occur over this extended treatment period would have to be weighed against the negative effects of one UTI.

Table 30. Any Drug Versus No Durg Pooled Analysis Acute (more...)

Table 30. Any Drug Versus No Durg Pooled Analysis Acute Asymptomatic

Individual Comparison Effect Size, Horizontal Line is 95% Confidence Interval

Pooled Effect Size, Symbol Width is 95% Confidence Interval

To continue with this example, the chi-squared test of heterogeneity for the Acute Asymptomatic stratum does not reject (p=0.50). We used six comparisons to produce this pooled result. We show the result shown graphically in Table 30. This graph is a shrinkage plot in which the individual comparison and pooled rate differences are shown, along with standard errors. In the plot, the studies are ordered from smallest (top) to largest (bottom) in terms of total number of patients. The rectangle for each comparison is that comparison's individual difference in weekly infection rates, treatment minus control, i.e. that comparison's effect size (see Table 27). The horizontal line through the rectangle indicates the 95 percent confidence interval for the individual comparison. The vertical line at zero breaks the effect size range into those effect sizes that indicate that treatment is favored, i.e., less than zero, and those that indicate the control condition is favored, i.e., greater than zero. The diamond at the bottom of the graph labeled "Overall" shows the location of the pooled or "overall" estimate. The width of the diamond demonstrates the 95 percent confidence interval for the pooled estimate. As shown, this confidence interval is to the left of zero and does not cross zero, which indicates a significant decrease in infection rate.

Table 31. Any Drug Versus No Drug Pooled Analysis Acute (more...)

Table 31. Any Drug Versus No Drug Pooled Analysis Acute Symptomatic

Individual Comparison Effect Size, Horizontal Line is 95% Confidence Interval

Pooled Effect Size, Symbol Width is 95% Confidence Interval

Table 32. Any Drug Versus No Drug Pooled Analysis Non-Acute (more...)

Table 32. Any Drug Versus No Drug Pooled Analysis Non-Acute ASymptomatic

Individual Comparison Effect Size, Horizontal Line is 95% Confidence Interval

Pooled Effect Size, Symbol Width is 95% Confidence Interval

Table 33. Any Drug Versus No Drug Pooled Analysis Non-Acute (more...)

Table 33. Any Drug Versus No Drug Pooled Analysis Non-Acute Symptomatic

Individual Comparison Effect Size, Horizontal Line is 95% Confidence Interval

Pooled Effect Size, Symbol Width is 95% Confidence Interval

Table 29 also shows pooled results for each of the other three major strata: Acute Symptomatic, Non-acute Asymptomatic, and Non-acute Symptomatic. Tables 31, 32, and 33 depict those results graphically. None of the other strata besides Acute Asymptomatic exhibited significant differences in weekly infection rates, although for non-acute patients the reduction in asymptomatic infections approached statistical significance (p = 0.06).

Further stratification is also done as appropriate. When possible, the comparisons are stratified into those that come from studies that had a Jadad quality score of three or more (3+) and those that had a Jadad score of two or less (2-) (see Evidence Tables 13 and 14 for Jadad scores). This cutpoint of three has been found in previous literature to be a good breakpoint in terms of quality (Moher, Pham, Jones et al., 1998). We also had hoped to stratify the studies by whether or not they had appropriate concealment of allocation (see Evidence Tables 13 and 14). However, only two studies, Gribble and Puterman (1993) and Schlager, Anderson, Trudell et al. (1998), had appropriate concealment so such stratification was impossible.

The second page of Table 29 presents the results of a sensitivity analysis to determine if population heterogeneity exists, as suggested by the different underlying control rates of infection described above. Among the six Acute Asymptomatic comparisons, the chi-squared test of heterogeneity rejects (p=0.02), indicating that heterogeneity, which is unexplained by sampling variation, may exist in the control rates of infection among these studies. Further subgroup analyses indicated that comparison #11 from Lindan was markedly different from the rest of the comparisons. As shown in Table 29, Lindan (listed as Acute Asymptomatic, study outcome definition of 10⁵ colonies per ml) has an underlying infection rate of 0.19 (95 percent confidence interval of [0.06, 0.45]). The other Acute Asymptomatic comparisons pooled according to outcome definition have pooled rates of 0.59 (studies with outcome definitions of 10⁴ colonies per ml; 95 percent confidence interval of [0.38, 0.81]) and 0.67 (for studies with "other" outcome definitions; 95 percent confidence interval of [0.07, 1.27]). The Lindan study's small weekly infection rates in both the treatment and control groups produces a small standard error as compared to the other studies (Table 27) because the standard error is a function of both the sample size and the rates. The Lindan result may be explained by the difference in definition (this study required 10⁵ colonies per ml) or may be due to other differences, such as population, protocol, setting, etc. When we pooled the remaining Acute Asymptomatic comparisons (Acute Asymptomatic #2, second page of Table 29), their pooled control infection rate was 0.58 (95 percent confidence interval of [0.40, 0.75]), and the chi-squared test for heterogeneity did not reject (p=0.46).

Table 34. Effect Size Versus Baseline Risk

Table 34. Effect Size Versus Baseline Risk

"AA" is an acute patients asymptomatic infections comparison; "AS" is an acute patients symptomatic comparison; "NA" is a non-acute patients asymptomatic infections comparison; and "NS" is a non-acute patients symptomatic infections comparison. Only comparisons in the Any Drug Versus No Drug analysis are plotted (Table 26).

Table 34 shows a scatterplot of the "baseline risk" or weekly infection rate in the control group versus effect sizes for the Any Drug Versus No Drug comparisons. The same plotting symbols are used as in the previous funnel plot in Table 28. The Acute Asymptomatic comparisons have higher control weekly infection rates and more negative, favoring treatment, effect sizes. The Lindan comparison #11 can be seen as the most extreme "AA" point to the left at a baseline risk (horizontal point) of 0.19, and an effect size (vertical distance) or -0.17. This comparison looks more similar to the "NA" or Non-acute Asymptomatic comparisons than to the "AA" or Acute Asymptomatic comparisons.

Removal of the Lindan study produces the "Acute Asymptomatic #2" pooled result of -0.39 (p<0.05, 95 percent confidence interval of (-0.58, -0.21) in the first page of Table 29. The underlying infection rates in the control group of the non-acute asymptomatic stratum were heterogeneous, as shown on the second page of Table 29.

Specific Drugs and Bladder Instillations

Table 35. Pooled Effect Sizes for Specific Drugs and (more...)

Table 35. Pooled Effect Sizes for Specific Drugs and Instillation Methods

Subgroup	Comments	Comparisons	Summary Difference in Weekly Infection Rates (treatment - control)	Number of Weeks Needed To Prophylaxis To Prevent One Infection	95% CI	Chi-square Test for Heterogeneity p-value
Nitrofurantoin
Acute Asymptomatic		3, 11	-0.22 1	4.5 1	(-0.36, 0.08)	0.34
Acute Symptomatic	no studies
Non-acute Asymptomatic		19, 21, 28	-0.06	NS	(-0.13, 0.01)	0.91
Non-acute Symptomatic		32	-0.03	NS	(-0.18, 0.12)	NA
Methenamine
Acute Asymptomatic		1, 5	-0.55 1	1.8 1	(-0.87, -0.22)	0.84
Acute Symptomatic	no studies
Non-acute Asymptomatic		24, 27	-0.25	NS	(-0.65, 0.14)	0.23
Non-acute Symptomatic	no studies
Trimethoprim and Sulfamethoxazole
Acute Asymptomatic		13, 14	-0.34	NS	(-1.03, 0.35)	0.97
Acute Symptomatic		15, 17	-0.03	NS	(-0.08, 0.02)	0.95
Non-acute Asymptomatic		18, 25, 29	0.00	NS	(-0.16, 0.17)	0.76
Non-acute Symptomatic		30, 31	-0.01	NS	(-0.11, 0.09)	0.59
Instillations
Acute Asymptomatic		2, 7, 12	-0.21	NS	(-0.52, 0.10)	0.28
Acute Symptomatic		no studies
Non-acute Asymptomatic		no studies
Non-acute Symptomatic		no studies

1

p<0.05, the 95% confidence interval does not cross zero. NS is not significant at p<0.05; NA is not applicable. Rates, differences in rates, and confidence limits have been rounded to two decimal places. The statistical tests in this table should be interpreted with caution as we did not adjust the significance levels for multiple comparisons.

Table 35 is analogous to the first page of Table 29. It depicts analyses of the specific drugs and bladder instillations employed. We note before discussing the results that the statistical tests should be interpreted with caution as we did not adjust the significance levels for multiple comparisons. The nitrofurantoin and methenamine studies appear to be the driving force behind the acute asymptomatic results shown in Table 29. In particular, pooling comparisons 3 and 11 (nitrofurantoin) produced a difference in weekly infection rates of -0.22 (p<0.05, 95 percent confidence interval of (-0.36, -0.08)). Pooling comparisons 1 and 5 (methenamine) produce an effect size of -0.55 (p<0.05, 95 percent confidence interval of (-0.87, -0.22)). No other significant results were observed in the stratified analyses shown in Table 29.

Defining a Urinary Tract Infection: Signs, Symptoms, and Laboratory Findings Associated with Risks

Bacteriuria is a common occurrence in these study populations; pyuria with bacteriuria appears to be associated with symptomatic infections, but these laboratory abnormalities are also relatively common in asymptomatic patients. The overall impact of treatment including both short- and long-term benefits and harms is not well-addressed in the literature. For example, although one study concluded that antibiotic treatment of persons having both bacteriuria and pyuria but no symptoms delayed the time to onset of symptomatic UTI by over 2 months, neither the potential adverse effects of treatment nor long-term risks and benefits were addressed. The National Institute of Disability and Rehabilitation Research conference of 1992 concluded, primarily based on clinical consensus, that fever was the sole symptom that routinely warrants concern and consideration for antibiotic treatment in the setting of bacteriuria. In the literature review for this evidence report, several case-control and cohort studies provide convergent data in support of this commonly accepted clinical axiom, based on long-term outcomes. For example, in one study, the occurrence of febrile episodes in prior years was associated with upper urinary tract complications or abnormalities at follow-up several years later. With regards to type or number of bacteria and long-term outcomes, one large retrospective cohort study showed an approximately 3@@@frac12@@@-fold increased odds for developing bladder calculi at 2 years if urine cultures at the time of discharge from an initial admission following spinal cord injury yielded a Proteus species or "other or multiple" organisms.

Other evidence regarding the significance of signs, symptoms, and laboratory findings is sparse or is inconclusive due to study design limitations.

Risk Factors for Urinary Tract Infection

Generally accepted risk factors for urinary tract infection (NIDRR Consensus Statement, 1992) include:

• Over-distention of the bladder

• Vesicoureteral reflux

• High pressure voiding

• Large post-void residuals

• Presence of stones in the urinary tract

• Outlet obstruction

With the exception of bladder residual volume, we could identify no literature specific to the neurogenic bladder that related these factors (such as overdistention of the bladder) to the risk of urinary tract infection.

We found only weak or inconclusive evidence that any of the non-modifiable risk factors in our conceptual framework (sex, level of injury, time since injury, etc.) were associated with the risk of urinary tract infection. Depending on the risk factor, there were either not enough studies, not enough studies of sufficient quality, or large enough samples to draw stronger conclusions. Given that the lifetime prevalence of urinary tract infection among persons with SCI and MS is so high, for all intents and purposes, every person with a neurogenic bladder should be considered at high risk for a urinary tract infection. Thus, this shifts the focus from immutable to mutable risk factors. (However, we note that non-modifiable risk factors can be helpful in increasing the ability to identify high-risk patients. This could be useful in defining homogeneous groups for future randomized controlled trials, for example.)

By far, the greatest amount of evidence that we were able to identify concerned the effect of bladder drainage method on the risk of infection. There was the relatively consistent finding that indwelling catheterization is associated with more frequent infections than intermittent catheterization, which in turn is associated with more frequent infections than drainage methods that do not rely on a catheter. While the evidence supporting this conclusion falls far short of proof, it seems prudent to make every effort to help persons with neurogenic bladder rely on the least invasive method of bladder drainage compatible with the patient's level of function, until better evidence becomes available.

There is conflicting evidence from three controlled trials regarding the benefits of sterile or "no touch" techniques for intermittent catheterization versus conventional clean intermittent catheterization. There exists scant evidence concerning other mutable risk factors, such as various behavioral factors, personal hygiene, insurance status, etc. These are areas for valuable future research (see later section of report).

Prophylaxis of Urinary Tract Infection

Due to the high frequency of urinary tract infection or bacteriuria in persons with neurogenic bladder, assessment of the impact of prophylactic antibiotic treatment is an important issue. Our literature search identified 19 controlled clinical trials relevant to this issue. Our meta-analysis of these trials showed that antibiotic prophylaxis is significantly associated with a reduced amount of bacteriuria among patients with neurogenic bladder in the acute phase of illness (p=0.05) and approaches a statistically significant association for non-acute patients (p=0.06). This conclusion is tempered by the fact that there were relatively few trials, and none of them were double-blinded.

However, there is no evidence to support the claim that such prophylaxis has been associated with a reduced number of symptomatic infections in the general populations that have been studied. In addition, there is good evidence that the use of oral antibiotic prophylaxis results in a two-fold increase in the proportion of antibiotic-resistant bacteria cultured from patients. Therefore, the regular use of antibiotic prophylaxis for most patients cannot be supported. It is possible, though, that antibiotic prophylaxis will reduce symptomatic infections among certain subgroups. In particular, some clinicians believe that prophylactic therapy helps to reduce the number of infections among persons who suffer from infections whose frequency and severity interfere with functioning and well-being in a chronic fashion. This is an important patient population to study in future randomized, clinical trials.

Defining a Urinary Tract Infection: Signs, Symptoms, and Laboratory Findings Associated With Risks

Only two studies of the impact of different urine sampling sources and methods on the measurement of bacteriuria in people with spinal cord dysfunction were identified in this review. Given the importance of understanding whether or not different thresholds of bacteriuria have different clinical risks depending on the sample source and method, future studies of this issue in different subgroups is also warranted.

Risk Factors for Urinary Tract Infection

The most important issue for future research on this topic to address is the identification of the modifiable risk factors for recurrent, clinically significant urinary tract infections. Although most persons with neurogenic bladder will experience urinary tract infections from time to time, the population of persons with frequent, symptomatic, recurrent infections are most in need of improved methods of identification and prevention. An initial study along these lines should adhere to the following parameters:

• It should use a prospective, cohort design and use appropriate methods to ensure minimal loss to followup. Given the changes in health care resulting in shorter lengths of stay for spinal cord injury and the need for long-term followup to assess certain risk factors, study designs will need to include plans for outpatient followup, if recruitment is in the inpatient setting.

• It should enroll a sufficient number of persons with the various hypothesized risk factors (bladder drainage method, behavioral factors, knowledge, sex, socioeconomic status, insurance status, etc.) to allow for meaningful comparisons across strata.

• It should use multivariate methods in the analysis to adjust, as suitable, for baseline differences among populations.

• It should measure the health costs and impacts on life associated with symptomatic infections, and it should have a followup time sufficient to measure these outcomes.

After identification of risk factors strongly associated with UTIs, the next step would be to test the effect of risk factor modification on rate of urinary tract infection in a randomized clinical trial. Through this type of research program, it should be possible to identify not only the risk factors for recurrent symptomatic infections, but also methods for reducing these infections through risk factor modification.

Prophylaxis of Urinary Tract Infection

The most important question for future research on this topic is to determine whether or not prophylactic antibiotic therapy reduces the frequency of recurrent, disabling urinary tract infections. Because of the problems with internal validity that are inherent in studies of comparative efficacy which use observational study designs, this future research should consist of randomized clinical trials. Some elements of such trials have been shown to be important at reducing the risk of bias; these should receive careful attention. In particular, such trials should do the following:

• Use a method of randomization that helps to ensure concealment of allocation

• Use double-blind methods to avoid expectation and detection bias

• Pay careful attention to withdrawals and dropouts and attempt to keep these to a minimum.

These randomized, clinical trials should also adhere to the following parameters:

• 1

  They should define the study population carefully and explicitly. Specifically, the population should consist of persons with neurogenic bladder who suffer from recurrent, disabling urinary tract infections.

• 2

  They should define the precise number of infections and degree of disablement or impairment in functioning and well-being associated with recurrent infections, as well as measure factors at baseline that could predispose to different infection rates, such as presence of vesicoureteral reflux and bladder and renal calculi.

• 3

  They should enroll a sufficient number of patients, so that the trial will have the statistical power to detect meaningful clinical differences.

• 4

  They should measure the outcomes both in terms of the number of symptomatic urinary tract infections, and in terms of impacts on health-related quality of life and on health care utilization and costs.

• 5

  They should measure the potential harms from treatment, particularly the development of antibiotic-resistant organisms.

To give an example of the size of such a study, if we were to define the following:

• The number of recurrent, disabling infections as four per year

• The clinically meaningful benefit worth detecting as an halving of this rate

• The followup time as 1 year

This estimate is a minimum number as it is dependent on convening a group of expert clinicians to define a relatively homogeneous patient group whom they believe, based on clinical experience, has the greatest likelihood of benefiting from prophylaxis, for example, a study of males using clean intermittent catheterization and having bladder calculi. If expert consensus cannot agree or concludes that one or more other key factors is critical to address the issue of prophylaxis and that a single homogeneous sample is not identifiable, then the sample size estimate would need to be adjusted upwards to assure that other potential covarying factors can be assessed with adequate statistical power.

For example, for low prevalence factors such as sex (male/female) in spinal cord injury, an adequate subgroup sample size would be necessary to assess this potentially key risk factor and its importance. Attaining such a sample size might require a stratified sampling approach.

For purposes of both feasibility (in terms of patient enrollment) and generalizability (in terms of the results), a proposed trial of this nature would probably need to be multi-center. While such a trial would no doubt be expensive to conduct properly, by generating a definitive result, it would produce health benefits many times greater than its initial cost.

Faxes and emails

1a. "It may be useful to spell out more carefully what you think the meanings of the existing literature are and what additional studies need to be performed. This was done more extensively in the previous sections that we got several weeks ago."

EPC response: text added

1b. "You have done an exceptionally thorough job overall in amassing this information and providing an assessment of the state of knowledge in this area."

EPC response: no action needed

EPC response: done

3a. "Including both spinal cord injury patients [SCI] and MS [multiple sclerosis] patients in a single spinal cord category makes this question more difficult to respond to. The literature for SCI patients included cohorts of mostly male patients the MS literature (which is minimal) includes mostly female patients. I don't think that the literature data can cross over and be generalized to the other population. I wonder if the risk factors for SCI and MS are the same?"

Note: This comment was from a peer reviewer not a panelist, who thus did not attend the panel meeting in December.

EPC response: The panelists discussed this extensively in December and felt that the final common denominator was physiology; thus, the literature for both SCI and MS could and should be reviewed together. In the end, there was very little literature on this topic for MS relative to the literature for SCI.

3b. "It is also unclear to me if this document supports treatment for infections that are asymptomatic. Did you intend to make a more definitive statement?"

EPC response: A more definitive statement like a guideline is the responsibility of those who may choose to use this evidence report to develop a guideline. The evidence we reviewed suggests no overall benefit for prophylaxis; there is insufficient literature to address the issue of which if any patients who are asymptomatic warrant treatment.

3c. "The preference for male/female terminology is "gender" instead of "sex" and should be changed in the text and pathway for consistency. "

EPC response: AHCPR editor specifically requested that we change from use of "gender" to "sex"

4. "I have reviewed the material you FAXED recently and have no useful comments to make about your extensive and valiant efforts. I am somewhat disappointed that the review was done without a definition of 'urinary tract infection' or 'risks' but having been involved in previous efforts to do that I realize that it is not realistic."

EPC response: no action possible

5. "No comment. It appears to present the evidence, but I will have to defer to the experts (microbiologists)."

EPC response: no action indicated

6. "Looks ok to me."

EPC response: no action indicated

Phone calls and messages

7.No substantial comments: did a really good job. Concerned that studies used were based on patients with spinal cord problems and having catheters - are there any studies with patients who did not have spinal cord injury but also had some form of urinary drainage to compare with these studies of people with spinal cord injury?

Note: This comment was from a peer reviewer not a panelist, who thus did not attend the panel meeting in December.

EPC response: We did not search for literature in populations without spinal cord dysfunction, at the specific direction of the panel. The panelists felt that this literature could not be extrapolated to people with a neurogenic bladder due to spinal cord dysfunction.

8. Suggested indicating more explicitly that causal pathways are not "proven" causal pathways but rather models based on expert opinion and knowledge of the literature, and that these models are subject to evaluation based on the strength of the evidence in the literature being analyzed.

EPC response: text revised where causal pathways are first described

9a.From practical experience, the patient's report of their symptoms is usually accurate.

EPC response: no action needed

9b.In the future studies section, consider noting the issue of shorter lengths of stay (2-3 months in the past, now about 4 weeks), which affects the designs of these studies.

EPC response: section revised to indicate that designs must include plans for outpatient follow-up; cannot rely solely on inpatient studies