Chapter  19:  Management of Acute Exacerbations of Chronic Obstructive Pulmonary Disease

Clinical Assessment

The principal findings concerning the clinical assessment of patients with acute exacerbation of COPD were as follows:

• Patients presenting with acute exacerbation of COPD have a relatively high rate of abnormalities (such as infiltrates or pulmonary edema) on chest roentgenography (CXR), particularly when compared with previous series of patients with asthma, where relatively low rates of abnormalities have been found.

• Historical data and clinical signs and symptoms associated with two common comorbid conditions that often complicate the assessment of acute exacerbation of COPD—CHF and pneumonia—are significant but inexact predictors of two specific abnormalities on CXR, namely pulmonary edema and infiltrate, respectively.

• The prevalence of clinically unsuspected deep venous thrombosis (DVT) among patients hospitalized for acute exacerbation of COPD is high in some studies; however, few data are available to help quantify the risk for pulmonary embolus among patients with or without known DVT.

• Among patients presenting with acute exacerbation of COPD, forced expiratory volume in 1 second (FEV₁) during exacerbation is not well correlated with partial pressure of oxygen (PaO₂) without supplemental oxygen, but is correlated with partial pressure of carbon dioxide (PaCO₂) and pH.

• Physician estimates of FEV₁ during acute exacerbation of COPD are generally inaccurate. Peak expiratory flow rate (PEFR) is not sufficiently well correlated with FEV₁ to substitute for it.

• Among patients on theophylline, neither clinical data on theophylline use (history of dosage, timing of last dose, past drug levels) nor other data (history of cigarette use, body weight) are accurate predictors of drug level during acute exacerbation of COPD.

Prognosis

Major findings related to prognosis were:

• While several factors are associated with worsening clinical conditions in patients with acute exacerbation of COPD, no predictive model accurately predicts clinical outcomes, so ongoing clinical monitoring is needed for many patients.

• Among patients presenting with acute exacerbation of COPD and selected for outpatient treatment, cumulative relapse rates were between 11 percent and 17 percent at 48 hours and between 23 percent and 32 percent at 2 weeks. Hospitalization at index visit ranged from 24.2 percent to 71 percent among patients presenting to the emergency department (ED).

• Data from the previous history of individual patients—e.g., previous visit within 7 days, number of exacerbations in the past year, and relapsing on previous visits—were consistently identified as predictive of relapse. Also found to be predictive in several studies was baseline pulmonary function, as measured by FEV₁ or forced vital capacity (FVC). Data describing acute respiratory physiology, such as FEV₁ during exacerbation or arterial blood gases, predicted hospitalization or relapse. Data describing treatments used in the ED and clinical response also were generally predictive of hospitalization or later relapse.

• Among patients hospitalized for acute exacerbation of COPD and cared for in either ward beds or intensive care units, short-term or hospital mortality ranged from 4 percent to 26 percent. Study populations were not described well enough to explain these differences in overall mortality rates.

• The following were all associated with mortality due to acute exacerbation of COPD:

  1. Acute physiology (as measured by arterial blood gases, FEV₁ during exacerbation, and scores from the Acute Physiology and Chronic Health Evaluation).

  2. Comorbid illness and other baseline, preexacerbation health status measures (such as body mass index and functional status).

  3. Cumulative or longitudinal data on the clinical course (e.g., baseline spirometry, number and frequency of previous acute exacerbations, and previous response to treatment of acute exacerbations of COPD).

• Acute respiratory physiology, as measured by blood gases, was predictive of the need for mechanical ventilation, as were baseline measures such as nutritional status.

Antibiotics

The antibiotic drugs studied were tetracycline, doxycycline, chloramphenicol, penicillin plus streptomycin, ampicillin, amoxicillin, and cotrimoxazole. The major findings related to these drugs were:

• Placebo-controlled randomized trials of antibiotic treatment of acute exacerbations of COPD show overall evidence of improvement in pulmonary function.

• The included trials suggest that patients with more evidence of bacterial infection (sputum purulence) and more severe illness (worse PEFR) benefit more from antibiotics; however, this has not been conclusively demonstrated.

Bronchodilators

The bronchodilators studied were:

• The anticholinergic ipratropium bromide and glycopyrrolate.

• The beta₂-agonist albuterol (salbutamol).

• Fenoterol.

• Metaproterenol.

• Salmeterol.

• Terbutaline.

• The methylxanthines aminophylline and doxofylline.

The major findings related to this class of drugs were:

• Inhaled ipratropium and beta₂-agonists were shown in comparative trials to have similar effects. However, neither class has demonstrated conclusive evidence of benefit in placebo or other no-treatment control trials. Most trials were too small to demonstrate a minimally clinically important benefit.

• Ipratropium is generally associated with fewer adverse effects than are the beta₂-agonists, but it needs to be used cautiously in patients with preexisting urinary retention problems. Beta₂-agonists can cause cardiac arrhythmias in those predisposed to the condition. The arrhythmias are usually not life threatening.

• Bronchodilator therapy delivered by nebulizers and metered-dose inhalers (MDI) show equivalent bronchodilation among patients with stable COPD. However, among patients with acute exacerbation of COPD, who may be unable to hold their breath, nebulizers may be necessary.

• Glycopyrrolate may have a synergistic effect in bronchodilation when given with a beta₂-agonist.

• Parenteral aminophylline did not improve FEV₁, hospitalization rates, or relapse in three placebo-controlled trials. Parenteral doxofylline did show a significant improvement in FEV₁ in a placebo-controlled trial. Moreover, methylxanthines have numerous, sometimes life-threatening, adverse effects and drug interactions.

Corticosteroids

Principal findings related to corticosteroid treatment were the following:

• Several randomized controlled trials provided strong evidence that a course of systemic corticosteroids provides benefit in patients hospitalized with acute exacerbation of COPD. The risk of treatment failure was reduced by approximately 10 percent, and FEV₁ showed an improvement averaging about 0.1 liters in the first hours to days of treatment.

• Doses as low as prednisone 30 mg daily and duration as short as 3 days have been shown to be effective in single trials; however, the optimal dose and duration are not clear from available trials. A single well-designed trial found that there were no significant differences in clinical outcomes between a 2-week course of systemic corticosteroids and an 8-week course.

• Inhaled corticosteroids have not been tested adequately in patients with acute exacerbation of COPD.

• Adverse effects were common in patients treated for acute exacerbation of COPD with systemic corticosteroids. The most frequently observed adverse effect was hyperglycemia.

Mucous-Clearing Treatments

Considered under this heading were mucolytic drugs and physical therapy interventions. The principal findings were:

• Available studies show no benefit from any of the mucolytic drugs studied (ambroxol, bromhexine, domiodol, potassium iodide, and S-carboxymethyl cysteine) in improving ventilatory function in acute exacerbation of COPD. Some studies reported subjective improvement in symptoms associated with decreasing sputum viscosity.

• Studies of chest percussion also failed to show any benefit in improving short-term ventilatory function in patients with acute exacerbation of COPD.

NPPV

Major findings related to NPPV are the following:

• NPPV is an effective alternative to mechanical ventilation by endotracheal intubation for some patients with acute respiratory failure secondary to acute exacerbation of COPD.

• The selection of mask interface and/or ventilator mode can be important to patient cooperation and tolerance, and thus to the efficacy of the intervention. Each type of mask and ventilation mode comes with its own set of morbidities. The pressure-support ventilation (PSV) and continuous or bilevel positive airway pressure modes of ventilation appear to be best tolerated and most effective for correcting hypercarbia. Assist control ventilation (ACV) mode with NPPV is generally poorly tolerated unless volume and rate are adjusted to the individual patient.

Key Questions

In our preliminary literature review, we identified several hundred head-to-head comparisons of different antibiotic treatments for acute exacerbation of COPD. Our Advisory Panel of Technical Experts suggested that we limit the studies of antibiotic treatment to placebo-controlled trials. A systematic review of the comparator trials might have been useful to inform selection of particular classes of antibiotics, particularly if it allowed stratification by severity of exacerbation. However, nearly all studies of antibiotic treatment for acute exacerbation of COPD were performed in outpatient populations with moderately severe exacerbations; it is not clear if the findings from these studies are applicable to hospitalized patients with more severe exacerbations, particularly because differences in the respiratory tract microbial flora occur based on severity of disease. Furthermore, this review would provide limited clinical usefulness, as geographic and temporal patterns of antibiotic resistance and the availability of antimicrobial agents may be more important determinants of antibiotic selection at the local level. Finally, we were concerned that this large and complex review could command all the resources that were available to this project, to the exclusion of other questions.

Causal Pathway

Figure 1. Evidence model for management of acute exacerbations (more...)

   Figure 1. Evidence model for management of acute exacerbations of COPD

Figure 1

Search Strategies and MeSH Terms

Drs. Douglas McCrory and Cynthia Brown reviewed several iterations of the literature search strategies and reached a level of refinement that excluded as many nonrelevant articles as possible without jeopardizing the inclusion of relevant articles. The search strategies were critically reviewed by the other clinical investigators in the local Work Group (Drs. David Matchar, Eugene Oddone, and Neil MacIntyre). As an additional quality check, the search strategies were reviewed by the COPD Guideline Collaboration Expert Panel that was convened by our study partners, the American College of Chest Physicians (ACCP) and the American College of Physicians-American Society for Internal Medicine (ACP—ASIM). The Work Group members and the ACCP/ACP—ASIM Guideline Collaboration Expert Panel provided several useful suggestions that were subsequently incorporated. The final search strategies are provided in Appendix A, which also includes the MeSH terms. Briefly, the search strategies combined a concept for “COPD” with a “methods” concept and an “intervention” concept. A subset of the MEDLINE search with high specificity (high yield) was achieved by adding an “acute exacerbation” concept; this strategy was used in the EMBASE and Cochrane Controlled Trials Register (CCTR) databases (MEDLINE [online database], 1999; EMBASE [online database], 1999; CCTR [database on CD—ROM], 1999). Based on recommendations from the Work Group, we eliminated articles with asthma or antibiotics as the focus (see “Exclusion Criteria”).

Bibliographic database searches from the preliminary literature review during the topic assessment and refinement phase of the project also contributed citations to the database.

Databases Searched and Years Included

The most productive literature databases were MEDLINE, CCTR, and EMBASE. MEDLINE was searched from the years 1966 to June 1999 for the topics of clinical assessment, treatment, and NPPV; EMBASE was searched from 1974 to 1999 for all topics; and the CCTR database (Issue 4, 1998, of The Cochrane Library) was searched for the treatment topic. The Cumulative Index to Nursing and Allied Health (CINAHL) and Health Services, Technology, and Research (HealthSTAR) databases also were searched, but they yielded few articles. For the EMBASE searches, Andrew Eisan, a Duke University Medical Center librarian, translated the MEDLINE search strategy for use in EMBASE. The number of articles identified for each of the three major topic areas was: diagnostic = 893, treatment = 1,609, and NPPV = 546.

Screening Criteria

Drs. McCrory, Matchar, and Brown led the effort to develop the screening criteria. These criteria, and those for quality rating of the articles, were discussed at length in several Work Group meetings. The criteria underwent several revisions before their final acceptance by the investigators. The final screening criteria for inclusion and exclusion of articles are described below.

Inclusion Criteria

Articles were selected for inclusion in the systematic review based on the following:

• Patient population. Study subjects were adults who were likely to have COPD based on clinical diagnosis, spirometry, or known or suspected history; subjects must have been experiencing an acute exacerbation of respiratory symptoms. Qualifying respiratory symptoms may have included increased dyspnea, increased quantity or purulence of sputum, or ARF.

• Interventions. Studies of clinical assessment, antibiotics, bronchodilators, corticosteroids, mucous clearing strategies, and NPPV were included. Additional criteria were as follows:

  • Bronchodilators: All dose ranges/timing; all formulations (metered-dose inhaler [MDI], nebulized, oral, and IV.

  • Corticosteroids: All dose ranges; all formulations (oral, inhaled, and IV).

  • Mucolytic/expectorant interventions: Drug treatment/respiratory therapies (e.g., bland aerosols; all dose ranges; all formulations [nebulized, oral, and IV]) and physiotherapy (including chest percussion, pneumatic vests, flutter device, etc.).

• Study designs: We included reports of original research or systematic reviews, but limited ourselves to certain study designs according to the topic and purpose (Sackett, Haynes, and Tugwell, 1985).

  • Therapeutics: Randomized controlled trials (RCTs) and quasirandomized or nonrandomized prospective controlled trials; retrospective historical or concurrent cohort comparisons.

  • Clinical assessment—Diagnosis: Studies describing the results of a test (with or without a reference standard test) in a series of patients suspected of having the condition or in two groups of patients, one known to have the condition and the other known not to have the condition.

  • Clinical assessment—Prognosis: Case series or cohort studies providing longitudinal data.

• Outcomes: For efficacy, we included studies with outcomes that were measured at least 4 hours after the start of the intervention. In addition, at least one of the following outcomes was measured and reported: mortality, hospitalization and LOS, relapse after discharge from outpatient care, relapse after discharge from inpatient care, health-related quality of life (HRQL)/QOL, symptom severity or duration, decreased need for intubation, improved breathing mechanics, improved ventilation (in arterial partial pressure of carbon dioxide [PaCO₂]), decreased need for supplemental oxygen, decreased ICU admissions, improved or lack of deterioration of mental status, ability for information at initial workup to predict any of the aforementioned, or adverse reactions and side effects of any intervention mentioned.

Exclusion Criteria

Articles were excluded from review if they were based on population types, interventions, study designs, and other criteria described below.

• Patient population—Excluded were studies of patients with: chronic MV needs, tracheostomies, asthma, bronchiolitis obliterans with organizing pneumonia, bronchiolitis obliterans, bronchiectasis, cystic fibrosis, and immunocompromised status (had known lung cancer, HIV/AIDS, or tuberculosis; or were on chemotherapy or radiation therapy for any cause).

The decision to exclude studies of patients with asthma was recommended by the Advisory Panel. There was broad agreement that the mechanisms underlying COPD and asthma are different and result in important differences in response to certain treatments. We decided to include patients with COPD who have coincident asthma, and patients with acute exacerbations of COPD who show evidence of airway reactivity. The following were excluded:

• Study designs—Nonsystematic reviews (traditional narrative reviews).

• Interventions—Studies relating to the use of sputum culture.

• Other—Studies published in non-English languages.

Quality Rating Criteria

The criteria used to rate the quality of the articles included: one for assessing external validity (applied to all included studies) and two for assessing internal validity (one for treatment articles and another for articles on prognosis).

Screening of Titles and Abstracts

Table 4. Interrater agreement in citation screening (more...)

Table 4. Interrater agreement in citation screening

Clinician Pair	Kappa Statistic	95% Confidence Interval
Reviewers 1 and 2	0.466	0.352 to 0.579
Reviewers 1 and 3	0.398	0.280 to 0.517
Reviewers 2 and 3	0.397	0.275 to 0.520
Reviewers 2 and 4	0.393	0.272 to 0.513
Reviewers 2 and 5	0.393	0.272 to 0.513
Reviewers 1 and 4	0.375	0.247 to 0.503
Reviewers 3 and 4	0.363	0.241 to 0.485
Reviewers 3 and 5	0.304	0.201 to 0.407
Reviewers 1 and 5	0.188	0.107 to 0.268
Reviewers 4 and 5	0.159	0.070 to 0.248

The clinical investigators reviewed the record of each article identified through the literature search and made a judgment, based on the previously described inclusion and exclusion criteria, about whether to include it for further review using the full text of the article. The first set of search results (n = 216 records) was reviewed by all five clinicians, and the strength of their agreement to include/exclude each article was tested by a kappa statistic (Table 4). Clinician pairs are ordered by agreement.

Agreement was significantly better than chance for all clinician pairs; however, agreement was only fair for most pairs. To ensure high sensitivity of the screening process despite lack of good to excellent agreement, we used several strategies. First, the search results were stratified into high-yield and low-yield sets (see search strategies). The high-yield set was relatively small (n = 216 MEDLINE articles) and was screened by the five clinicians, with a net inclusion rate of 48 percent. The low-yield stratum was much larger (n = 1,767 MEDLINE articles). Each citation was screened by two clinicians, with a net inclusion rate of 8.6 percent. The search strategies employed in the CCTR and EMBASE databases were of the more specific high-yield variety. Second, we used the combination of a pulmonary medicine clinician and a methodologist clinician as screeners. We believed that the differing perspectives of the reviewers would improve the sensitivity of the selection process.

We took several steps to preserve the integrity of the selection process. First, the reviewers were made aware of the disagreements; next, they reviewed a sample of the citations about which they disagreed so that they could identify the reasons for disagreement. Second, because there was evidence of poor agreement, we required that at least two reviewers screen each citation; citations that were included by EITHER reviewer were kept. This ensured that all relevant citations were retained, even though different raters may have been using slightly different thresholds.

The numbers of articles that were included after the “title and abstract” screening were: clinical assessment = 162, treatment = 185, and NPPV = 166. The inclusion rates for the three topics were 18.1, 11.5, and 30.4 percent, respectively.

Data Abstraction and Development of Evidence Tables

Those articles that passed the full-text screening were grouped according to topic and were carefully read in their entirety. At this stage, final determination of inclusion and exclusion criteria was made. Data were abstracted onto a computerized form that had been specially designed with input from all investigators. As data abstraction on each article was completed, the form paired with the article was reviewed by another investigator (clinician) for accuracy and completeness. (An example of a completed data abstraction form is shown in Appendix.)

Statistical Analysis Methods

For most topics, we did not perform any statistical analysis. We reported analyses performed by the authors of original reports. In summarizing data on groups of studies, we used standard statistical methods for rates and proportions or group means. In a few instances, when several comparable studies provided estimates of effectiveness for a given treatment comparison, we tested for homogeneity and, if reasonably similar, combined the individual study estimates using a fixed effects model. Meta-analysis was performed using FAST*PRO software (FAST*PRO, 1992).

Summary of Search Results

Table 5. Literature selection process by topic

Table 5. Literature selection process by topic

Step 1	Step 2	Step 3	Step 4
Number of Articles Identified by Literature Searches1	Number of Articles Included After Title & Abstract Screening (% Included)	Number of Articles Included After Full-Text Screening (% Included, Step 2 → Step 3)	Final Number of Articles Included (% Included, Step 3 → Step 4)	Percent of Articles Included, Step 1 → Step 4
Clinical Assessment 892/839	160 (17.9%)	101 (63.1%)	35 (34.6%)	3.9%
Treatment 1,610/1,588	182 (11.3%)	88 (48.4%)	40 (45.4%)	2.5%
NPPV 546/501	165 (30.2%)	63 (38.1%)	10 (15.9%)	1.8%
Total 2,833/2,793	4882 (17.2%)	2472 (50.6%)	85 (34.4%)	3.0%

1

The number before the diagonal (/) is the number of articles identified by the targeted search plus those referred for screening from targeted searches on other topics. The number listed after the diagonal is the number of articles identified by the targeted search only.

2

Totals are slightly lower than sum of each topic because of overlap between topics.

NPPV = noninvasive positive pressure ventilation.

Table 6. Literature identification by source database (more...)

Table 6. Literature identification by source database

	MEDLINE1	Cochrane Controlled Trials Register2	EMBASE3	Other	Total
Clinical Assessment Articles	664	39	372	2	893
Excluding MEDLINE		6	224	2
Excluding CCTR			220	2
Excluding EMBASE				2
Treatment Articles	1,258	345	293	8	1,610
Excluding MEDLINE		236	144	8
Excluding CCTR			108	8
Excluding EMBASE				8
NPPV Articles	499	26	124	1	546
Excluding MEDLINE		11	42	1
Excluding CCTR			35	1
Excluding EMBASE				1
Total	2,227	345	640	11	2,833
Excluding MEDLINE		236	395	11
Excluding CCTR			359	11
Excluding EMBASE				11

1

MEDLINE [database online]. Bethesda (MD): National Library of Medicine; 1985-. Updated weekly. Available from: National Library of Medicine; BRS Information Technologies, McLean VA; DIALOG Information Services, Inc., Palo Alto, CA; Ovid Technologies, Inc., New York, NY.

2

Cochrane Controlled Trials Register [database on CDROM]. In: The Cochrane Library. London: BMJ Publishing Group; 1998 Issue 4. Updated quarterly. Available from: BMJ Publishing Group, London, United Kingdom.

3

EMBASE [database online]. New York (NY): Elsevier Science Secondary Publishing Division; 1974-1999. Updated weekly. Available from: Elsevier Science Secondary Publishing Division, New York, NY; DIALOG Information Services, Inc., Palo Alto, CA.

NPPV = noninvasive positive pressure ventilation

Internal Quality Control Mechanisms

Quality control procedures were integrated into each step of the literature review. The search strategies had several checks on their comprehensiveness: (1) the strategies were checked and finalized by a medical librarian at the Duke University Medical Center Library who specializes in evidence-based literature reviews; (2) the articles cited in reference lists of reviewed articles were compared with those in the ProCite literature database and, when appropriate, articles were copied, reviewed, and added to the ProCite database; and (3) the Advisory Panel of Technical Experts was asked to submit articles and other documents that met the specified inclusion criteria as a check on the effectiveness of the search strategies.

With regard to the content of the evidence tables, several quality control procedures were included: (1) screening of all articles by at least two clinicians; (2) abstracting of information into the evidence report by two clinician-trained abstractors; and (3) overreading of all abstracted information by at least one clinician. The clinicians responsible for screening articles, training the abstractors, and overreading the abstracted information were paired so that each pair included a pulmonologist and an internist. Kappa statistics were calculated to determine the strength of agreement between clinician-reviewers in the article screening stage and to identify areas where agreement was low. In such cases, the reviewers identified discrepancies and agreed on an interpretation of the selection criteria. To maximize the sensitivity of the screening process, we included articles selected by either reviewer, recognizing that the data abstraction process would permit reexamination of inclusion/exclusion status.

External Quality Control Mechanisms

Two external oversight and review panels were established—the Advisory Panel of Technical Experts and the Peer Review Panel.

The Advisory Panel of Technical Experts, which was initiated early in the project's 12-month timeframe, reviewed progress on the evidence report at four key stages of its development: (1) the identification of the key literature search questions; (2) the first drafts of the evidence tables; (3) the need for any supplemental analyses; and (4) the draft of the evidence report. Draft documents were discussed as a group in two conference calls, in individual telephone calls, and in written communications. The nine-member Panel consisted of clinical and methodological experts in relevant specialty areas, including pulmonology, critical care medicine, respiratory therapy, internal medicine, and epidemiology. Panel members also were chosen to assure representation of three other important constituencies: the Department of Veterans Affairs, managed care organizations, and consumers. Consumers were represented by the American Lung Association's North Carolina office.

The primary function of the Peer Review Panel was to review and comment on the complete draft of the evidence report. The 27-member Panel, which included AHRQ staff and the Advisory Panel of Technical Experts, consisted of clinical and methodological experts in relevant specialty areas, including pulmonology, critical care, respiratory therapy, internal medicine, and epidemiology. Other constituencies represented were consumers, the Department of Veterans Affairs, managed care organizations, and the Cochrane Airways Group. The report was modified based on the Panel's comments, with close attention to relevant studies not included in the report, misinterpretation of findings, and other issues deserving revision within the constraints of the methodology, timeframe, and budget. The format of the report was designed by AHRQ.

Differential Diagnosis

Severity of Illness as Gauged by Spirometry

Spirometric assessment is the hallmark of the diagnosis of COPD; however, few data are available on the usefulness of spirometric assessment in acute exacerbations. We sought studies describing the accuracy, repeatability, and associations with other markers of severity or clinical outcome of spirometric measures taken in patients with acute exacerbation of COPD. We identified three such studies (Emerman, Connors, Lukens, et al., 1989a; Emerman and Cydulka, 1996; Emerman, Lukens, and Effron, 1994). Each of these studies was set in the same urban medical center ED and used essentially identical inclusion criteria: age over 50 years, a clinical diagnosis of COPD, and a recent increase in respiratory symptoms leading to presentation to the ED. Although the study dates were not specified, the publication dates ranged from 1989 to 1996. External validity scores in the three studies ranged from 1/4 to 3/4; all were graded Level 3 on the internal validity scale.

Emerman, Connors, Lukens, et al. (1989a) (external validity, 3/4; internal validity, Level 3) described the correlations between FEV₁ and measured pH, PaCO_2, and PaO₂ on initial ABG. The study population had a baseline stable FEV₁ of 1.37 L (57 percent predicted) and FEV₁ at presentation of 0.71 L (25.7 percent predicted). Initial ABGs showed an average pH of 7.43, PaCO₂ of 39.7, and PaO₂ of 63.5 millimeters of mercury (mm Hg). Only 4.3 percent (3/70) of patients were acidemic (pH < 7.36); 20 percent were hypercapnic (PaCO₂ > 45 mm Hg); and 34 percent were hypoxic (PaO₂ < 60 mm Hg). The correlation between FEV₁ at the time of presentation and PaO₂ on initial ABG was not statistically significant. FEV₁ showed a weak, but statistically significant, correlation with both PaCO₂ (correlation coefficient [r] = -0.47; probability [p] < 0.001) and pH (r = 0.36; p < 0.01). The authors noted that these results were different from those seen in asthmatics presenting to the emergency room, where spirometry and ABGs are ordinarily highly correlated. Clinically, this poor correlation between FEV₁ and ABGs means that the spirometric measure cannot obviate the need for ABG measurement in patients presenting with acute exacerbation of COPD.

Emerman, Lukens, and Effron (1994) (external validity, 2/4; internal validity, Level 3) described the accuracy of physicians' estimations of FEV₁ in 90 patients presenting with acute exacerbation of COPD. The patient sample was similar to that of the previously described study, with FEV₁ at admission of 30.9 percent predicted. Physicians were asked to estimate FEV₁, both pre- and posttreatment. In estimating posttreatment FEV₁, physicians were allowed to see the previously measured spirometric values. The correlation between observed and predicted pretreatment FEV₁ was relatively weak but statistically significant (r = 0.34; p < 0.001). Physician estimates were roughly accurate in 38 percent of patients, high in 48 percent, and low in 14 percent, indicating that physicians tended to overestimate FEV₁. Posttreatment estimates were both better correlated (r = 0.57; p < 0.0001) and better calibrated with posttreatment measures. Physician estimates of posttreatment FEV₁ were roughly accurate in 46 percent of patients, high in 43 percent, and low in 11 percent. Attending physicians' predictions were significantly more accurate than those of residents (p < 0.001).

Emerman and Cydulka (1996) (external validity, 1/4; internal validity, Level 3) compared two spirometric measures, PEFR and FEV₁, in a population of 199 patients presenting with acute exacerbation of COPD. Although PEFR can be measured quickly using an inexpensive, handheld peak flow meter, this study measured both FEV₁ and PEFR simultaneously using a spirometer. Each patient gave three attempts, with the highest simultaneously obtained measures being used. Measures were taken before treatment, 1 hour after treatment, and before discharge from the ED. A total of 556 measurements from 199 patients were used to calculate a correlation coefficient. The mean FEV₁ at admission was 0.85 L (30.2 percent predicted) and PEFR was 113.1 L/minute (min) (26.3 percent predicted). The correlation between FEV₁ and PEFR for all measurements was high (r = 0.84; p < 0.001); however, the authors pointed out that for a minority of patients, the absolute difference between percent predicted values based on FEV₁ and PEFR was greater than 10 percent. The authors concluded that PEFR is not sufficiently well correlated to substitute for FEV₁. Although they were not used in this study, handheld peak flow meters (e.g., mini-Wright device) would likely show poorer correlation with FEV₁. However, in patients experienced in performing PEFR measurement, the accuracy might be better; when used for serial monitoring, data on changes in PEFR over time may be more clinically useful than single FEV₁ measures.

Outpatient and ED Studies

Table 7. Outpatient studies of predictors

Table 7. Outpatient studies of predictors

Study	Setting	Outcome	Analysis	FEV1	Significant Predictors
Murata, Gorby, Chick, et al., 1989	ED	Relapse (ED)	Uni	46.3% pred (baseline)	Nighttime presentation (vs. daytime)
					Weekend presentation (vs. weekday)
Murata, Gorby, Kapsner, et al., 1992a	ED	Relapse (ED)	Multi	40.4% pred (baseline)	Relapse on previous visits
					Use of aminophylline
					Greater no. of doses of nebulized bronchodilators
					Previous visit within 7 days
					No use of glucocorticoids at discharge
Emerman, Effron, and Lukens, 1991	ED	Relapse (ED)	Uni	0.81 L (acute)	Lower pretreatment FEV1
					Older age
					Female sex
					Greater number of bronchodilator treatments in ED
Ball, Harris, Lowson, et al., 1995	Office-based outpatient practice	Relapse (office)	Multi	N/S	Presence of comorbid cardiopulmonary disease
					More than 4 previous exacerbations in past year
		Hospitalization	Multi	N/S	Presence of comorbid cardiopulmonary disease
					Greater duration of COPD
					History of smoking (ever or current)
Murata, Gorby, Chick, et al., 1991	ED	Relapse (ED)	Multi (univariable predictors listed)	43.8% pred (baseline)	By patient:
				0.74 L (acute)	Greater bronchodilator response on baseline FEV1
					Greater number of bronchodilator treatments in ED
					More frequent use of parenteral adrenergic drugs
					By visit:
					Shorter duration of dyspnea
					Lower entry FEV1
					Lower discharge FEV1
					Greater number of bronchodilator treatments in ED
					More frequent use of parenteral adrenergic drugs
					Less frequent use of oral prednisone on discharge
Murata, Gorby, Kapsner, et al., 1992b	ED	Relapse (hospital)	Multi	40.2% pred (baseline)	High admission rate on previous visits
					High relapse rate for previous visits
					Low proportion of past discharges in which conservative therapy was used
					Low postbronchodilator FEV1 at baseline
					High postbronchodilator FEV1/FVC ratio
Parshall, 1999	ED	Hospitalization	Multi	N/S	Use of intravenous fluids
Fedullo, Swinburne, and McGuire-Dunn, 1986	ED	Hospitalization	Uni	N/S	Greater body temperature
					Lower arterial pH
					Higher arterial PaCO2
					Lower arterial PaO2

COPD = chronic obstructive pulmonary disease; ED = emergency department; FEV₁ = forced expiratory volume in 1 second; FVC = forced vital capacity; Multi = multivariable; N/S = not specified; PaCO₂ = partial pressure of carbon dioxide, arterial; PaO₂ = partial pressure of oxygen, arterial; pH = hydrogen ion concentration; pred = predictive; Uni = univariable

The same group of investigators reported another study in a group of patients assembled from the ED log book at the same site as the previous study (Murata, Gorby, Chick, et al., 1991) (external validity, 4/4; internal validity, Level 3). Three hundred fifty-two consecutive patients were assembled. Detailed clinical data were gathered from a medical record review. The investigators used both univariable and multivariable analyses to identify, from the clinical history and laboratory values, those factors that predicted relapse, which was defined as return to the ED within 14 days of discharge. They also reported data on hospitalization; the hospitalization rate in patients with at least one ED discharge was 24 percent (280/1,157 visits). Although this figure is lower than that reported in the previous study, it underestimates the actual hospitalization rate because 84 patients who were admitted on every ED visit during the study period were excluded. For visits resulting in ED discharge, the relapse rate was 32 percent (281/877 visits).

The investigators analyzed the data on predictors of relapse in several ways. Taking each visit as a unique observation, they identified the following predictors of relapse (univariable analysis): shorter duration of dyspnea (p = 0.002); lower entry FEV₁ (p = 0.027); lower discharge FEV₁ (p = 0.04); greater number of nebulized bronchodilator treatments (p = 0.009); more frequent use of parenteral adrenergic drugs (p = 0.006); and less frequent use of oral prednisone on discharge (p = 0.016). Taking each patient as a unique observation, the investigators identified the following predictors of relapse: more improvement in FEV₁ with bronchodilators on baseline PFTs (p = 0.047); more bronchodilator treatments in ED (p = 0.001); and more frequent use of parenteral adrenergic drugs (p = 0.006). Finally, the investigators used multivariable linear regression to predict the number of relapse visits per patient. Bronchodilator “response index” (defined as change in FEV₁ with bronchodilators divided by the difference between prebronchodilator FEV₁ and predicted FEV₁ from spirometry at baseline), number of nebulizer treatments, use of parenteral adrenergic drugs, and oral prednisone on discharge were independent predictors. Overall, the model was highly significant (p < 0.001), but it explained little of the variance in relapse number (multiple R² = 0.202).

Emerman, Effron, and Lukens (1991) (external validity, 2/4; internal validity, Level 3) described predictors of relapse in patients presenting to an urban ED with acute exacerbation of COPD. Of a total 83 patients in the study, 37 were hospitalized at the index ED visit (45 percent). Of the 46 patients discharged from the ED, various clinical data available at presentation or during ED management were used to predict return to ED or hospital admission within 48 hours of discharge from the ED. The investigators found that the following data were associated with relapse: lower pretreatment FEV₁, older age, female sex, and greater number of aerosol treatments in the ED. No multivariable analyses were performed. The investigators described the sensitivity and specificity of several proposed admission criteria based on spirometric measures and ABG values. For example, if all patients with posttreatment FEV₁ < 40 percent of predicted normal had been admitted, this would have identified 43 of the 45 patients who were admitted at either the index ED visit or during the next 48 hours (sensitivity = 0.96). However, this criterion would have resulted in admitting 22 of the 38 patients who did not require admission (specificity = 0.58). Other proposed criteria were: posttreatment FEV₁ < 70 percent of baseline FEV1 (sensitivity 0.83, specificity 0.56); pretreatment FEV₁ < 30 percent of predicted normal (sensitivity 0.76, specificity 0.59); less than 10 percent increase in FEV₁ after treatment (sensitivity 0.44, specificity 0.42); PaO₂ < 60 mm Hg (sensitivity 0.46, specificity 0.85); and PaCO₂ > 45 mm Hg (sensitivity 0.23, specificity 0.92). The threshold value of 40 percent predicted normal was apparently selected as the optimal cutoff for the FEV₁ clinical prediction rule based on a receiver operating characteristic curve (ROC) derived from these same data. Thus, the performance of this rule in another population is likely to be worse. The rationale for selecting thresholds for other variables is not described, but it also may be based on these data and, thus, overestimate the performance in another population.

Two 1992 publications described the development and validation of clinical prediction rules for relapse within 48 hours of ED discharge and for hospitalization within 2 weeks of ED visits. The study populations used for these two studies overlapped substantially. The populations were assembled from patients with a history of COPD presenting with a chief complaint of increase dyspnea to a Department of Veterans Affairs hospital ED. Murata, Gorby, Kapsner, et al. (1992a) (external validity, 2/4; internal validity, Level 3) identified factors that predicted relapse in 705 episodes of acute exacerbation of COPD in which the patient was discharged from the ED. The relapse rate in this population was 11.6 percent (82/705 visits). Several factors were found to increase the risk of relapse: high relapse rate for previous visits; use of aminophylline; use of antibiotics at discharge; number of doses of nebulized bronchodilators; and previous visit within 7 days. Two factors decreased the risk of relapse: long-term oxygen therapy and use of prednisone at discharge. A multivariable logistic regression model developed in this population was tested in a separate group of 159 patients with 476 ED discharges accrued in 1988-1990. The model was used to classify visits as high or low risk, based on model predicted values. All of the factors that had been found in the training set to increase the risk of relapse were validated; however, the factors that decreased relapse risk were not predictive in the validation set. The relapse rate was 18.4 percent in those visits predicted as high risk by the model, and 6.1 percent in those visits predicted as low risk. The sensitivity was 0.57 and specificity was 0.72.

In a companion paper, the same authors described the development of a model to predict relapse, which was defined as hospital admission within 14 days of an ED visit for COPD exacerbation (Murata, Gorby, Kapsner, et al., 1992b) (external validity, 2/4; internal validity, Level 3). In this analysis, similar methods were used to develop a predictive model in a training set of 693 episodes of acute exacerbation of COPD discharged from the ED. In keeping with the authors' goal of prehospital screening, the predictors that were considered included only data about baseline status and outcomes of previous visits. The hospitalization rate was 210/693 visits (30.3 percent). In the training set, a multivariable analysis showed the following factors to be significant independent predictors of hospitalization: higher admission rate for previous visits; higher relapse rate for previous visits; lower proportion of past discharges in which conservative therapy was used; lower postbronchodilator FEV₁ (baseline) percent of predicted; and lower postbronchodilator FEV₁/vital capacity (VC) ratio (baseline) percent of predicted.

In the validation set, the hospitalization rate was 76/269 visits (28.3 percent). Of the 269 visits, 171 (63.6 percent) were classified as high risk, and 98 (36.4 percent) were classified as low risk. Hospitalization was needed for 68/171 (39.8 percent) patients who were predicted to be at high risk and 8/98 (8.2 percent) patients who were predicted to be low risk. This corresponds to a relative risk of 4.9 (95 percent CI, 2.5 to 9.8), a sensitivity of 89.5 percent, and a specificity of 46.6 percent. Positive predictive value was 39.8 percent, and negative predictive value was only 8.2 percent.

Ball, Harris, Lowson, et al. (1995) (external validity, 1/4; internal validity, Level 1) studied 471 patients presenting to any of several office-based general medical practices in the United Kingdom with features of acute exacerbation, clinical diagnoses of COPD, and age > 30 years. Various predictors were tested for association with relapse, which was defined as return to the general practitioner with a chest problem or hospitalization within 28 days. Of 423 patients for whom followup data were available, 56 (13.2 percent) relapsed, 9 of whom were hospitalized. Although age and sex were not associated with relapse, the presence of comorbid cardiopulmonary disease and a history of more than four previous exacerbations in the past year were significant independent predictors of relapse. Patients were classified according to the Winnipeg criteria, and the number and identity of the clinical features (increased amount of sputum, increased purulence of sputum, or increased dyspnea) were not associated with relapse.

Two studies described predictors of hospital admission in patients with COPD who were managed in an ED (Fedullo, Swinburne, and McGuire-Dunn, 1986; Parshall, 1999). Parshall (1999) (external validity, 0/4; internal validity, Level 1) described a group of 239 patients from the 1992 National Hospital Ambulatory Care Survey (NHAMCS), which was a probability sampling of every 20th emergency visit from 437 EDs in participating nonfederal hospitals. Information on chief complaint, visit urgency, type(s) of medication used in the ED, and use of IV fluids was assessed to identify possible predictors of visit disposition (hospital admission versus discharge). More than half of the patients with COPD were hospitalized (54.8 percent, l31/239 patients). A chief complaint of dyspnea, urgent versus nonurgent visit, and use of IV fluids were associated with admission. Age and use of bronchodilators had no predictive value. Use of IV fluids was the best predictor, and when considered in a multivariable analysis, it was the only significant independent predictor of admission.

Fedullo, Swinburne, and McGuire-Dunn (1986) (external validity, 1/4; internal validity, Level 3) identified all patients presenting to an ED with a complaint of dyspnea and they provided separate results for those patients with COPD. Seventy-one percent of the 24 COPD patients were admitted, and the investigators compared several variables between those patients admitted and those discharged from the ED. Those admitted had higher body temperature, lower pH and PaO₂ values, and higher PaCO₂ values than those discharged from the ED. No data on the clinical course following ED discharge were available in this retrospective study. Age, duration of breathlessness, heart rate, respiratory rate, and presence of peripheral edema or wheezing were no different between those admitted and those discharged; however, the small sample size limited the power of this study to identify differences. In particular, the pulse rate, respiratory rate, and leukocyte count were higher in those patients admitted, but these differences were not statistically significant.

In summary, the risk of relapse in patients presenting with acute exacerbations who were selected for outpatient treatment was between 11 and 17 percent at 48 hours, and between 23 and 32 percent at 2 weeks. In all patients presenting to the ED, hospitalization at index visit varied from 24.2 to 71 percent. Mean age, spirometric values, and other descriptive data were reasonably similar in the studies, suggesting similar populations. Data from the previous history of individual patients were consistently identified as having value for predicting relapse; for example, previous visit within 7 days, number of exacerbations in the past year, and relapsing on previous visits. Also predictive in several studies was baseline pulmonary function as measured by FEV₁ or FVC. Data describing acute respiratory physiology such as FEV₁ during exacerbation or ABGs predicted hospitalization or relapse. Data describing treatments used in the ED and clinical response in the ED also were generally predictive of hospital admission or later relapse. However, despite demonstrating statistically significant associations with outcome, most prediction rules were either not sufficiently accurate to be clinically useful or performed poorly on attempts at validation.

Inpatient Studies

Table 8. Inpatient studies of predictors

Table 8. Inpatient studies of predictors

Study	Setting	Analysis	FEV1	Significant Predictors11,2
Mortality
Connors, Dawson, Thomas, et al., 1996	ICU	Multi	N/S	Higher acute physiology score
				Lower BMI
				Older age
				Worse functional status 2 wks prior to admission
				Lower PaO2/FiO2 ratio
				Absence of comorbid CHF
				Lower serum albumin
				Absence of comorbid cor pulmonale
Seneff, Wagner, Wagner, et al., 1995	ICU	Multi	N/S	Higher nonrespiratory APS score
				Greater number of pre-ICU hospital days
Burk and George, 1973	Hospital ward/ICU	Uni	“Severe obstructive ventilatory impairment”	Use of mechanical ventilation (vs. conservative care)
				General medical ward care (vs. ICU care)
				CHF as etiology of ARF (vs. respiratory infection)
Warren, Flenley, Millar, et al., 1980	Hospital ward	Uni	N/S	Older age
				Highest level of arterial PaCO2 during controlled oxygen therapy
				Lowest pH < 7.26 (p < 0.025)
Jeffrey, Warren, and Flenley, 1992	Hospital	Uni	N/S	Measured at admission:
				High blood urea concentration
				Low systolic BP
				Low arterial pH
				Measured throughout hospital stay:
				Lowest pH < 7.26
				Lowest pH < 7
Heuser, Case, and Ettinger, 1992	ICU	Multi	N/S	Older age
				Presence of comorbid COPD
				Age-comorbid COPD interaction
				Primary diagnosis pneumonia (note asthma bronchitis was reference case)
				MedisGroups Admitting severity group 3 or 4
Portier, Defouilloy, and Muir, 1992	ICU	Multi	N/S	Presence of cachexia
				Low serum sodium
				Required mechanical ventilation in first 24 hrs
				Not COPD as underlying chronic respiratory insufficiency*
				Previous confinement to home
				Presence of edema*
Fuso, Incalzi, Pistelli, et al., 1995	Hospital	Multi	N/S	Older age
				PA-aO2 > 41 mm Hg
				Presence of atrial fibrillation
				Presence of ventricular arrhythmias
Dardes, Campo, Chiappini, et al., 1986	Hospital	Uni	N/S	None identified
Moran, Green, Homan, et al., 1998	ICU	Multi	0.70 L (baseline)	None identified
Need for mechanical ventilation
Vitacca, Clini, Porta, et al., 1996	Hospital	Multi	35.6% pred (admission)	Worse nutritional status
				Lower FVC (% predicted)
Bone, Pierce, and Johnson, 1978	ICU	Multi	0.67 L (baseline)	Worse blood gas values (PaO2, PaCO2, pH) on admission
				Reduction in pH after initial oxygen therapy
Moran, Green, Homan, et al., 1998	ICU	Multi	0.70 L (baseline)	None identified
Length of stay
Loukides and Polyzogopoulos, 1996	Hospital	Uni	1.2 L (admission)	Comorbid diabetes mellitus
Moran, Green, Homan, et al., 1998	ICU	Multi	0.70 L (baseline)	Greater time on mechanical ventilation
Mushlin, Black, Connolly, et al., 1991	Hospital	Multi	N/S	PaCO2 level > 44 mm Hg in ED
				Symptoms present for more than 1 day
				Received antibiotics on day of admission

1

Presence of predictor or noted direction is associated with an increased risk in the outcome indicated.

2

Asterisk indicates variable in multivariable analysis was not significant in univariable analysis. APS = acute physiology score; ARF = acute respiratory failure; BMI = body mass index; BP = blood pressure; CHF = congestive heart failure; ED = emergency department; FEV₁ = forced expiratory volume in 1 second; FiO₂ = fraction of inspired oxygen; FVC = forced vital capacity; Hg = mercury; ICU = intensive care unit; L = liter; mm = millimeter; Multi = multivariable; N/S = not specified; PaCO₂ = partial pressure of carbon dioxide, arterial; PaO₂ = partial pressure of oxygen, arterial; PAO₂ = partial pressure of oxygen in the alveoli; PaO₂ -PAO₂ = alveolar-arterial difference in partial pressure of oxygen; pH = hydrogen ion concentration; pred = predictive; Uni = univariable

Seneff, Wagner, Wagner, et al. (1995) (external validity, 1/4; internal validity, Level 1) enrolled 362 patients who were admitted to ICUs with respiratory failure due to COPD. Excluded were patients with pneumonia, pulmonary edema, or PE. The inhospital mortality rate of 23.8 percent was predicted by the number of pre-ICU hospital days and the “nonrespiratory” component of the acute physiology score, which included the Glasgow Coma Score, heart rate, mean blood pressure, temperature, hematocrit, WBC count, creatinine, urine output, serum urea nitrogen, sodium, albumin, and glucose. Although predictive on univariable analyses, the following factors were not significant independent predictors on multivariable analysis: age, activities of daily living, “respiratory” acute physiology score, or use of MV on ICU day 1. The “respiratory” acute physiology score comprised respiratory rate, pH, PaCO₂, and PaO₂ or alveolar arterial difference in partial pressure of oxygen (PA-aO₂).

A separate analysis of predictors of 180-day mortality was conducted. Significant predictors included both components of the acute physiology score, older age, and a higher number of pre-ICU hospital days. Activities of daily living was a significant predictor on univariable analysis, but did not demonstrate independent predictive ability in the multivariable analysis. The 180-day mortality rate was not described.

The earliest report examining predictors of mortality in patients hospitalized with acute exacerbation of COPD described 74 patients at a Department of Veterans Affairs hospital (Burk and George, 1973) (external validity, 1/4; internal validity, Level 3). Study participants had ARF, which was defined as clinical deterioration plus hypercapnia or hypoxia. The hospital mortality rate was 26 percent. Predictors of inhospital mortality on univariable analysis were use of MV and presence of CHF (versus respiratory infection). ICU care was associated with lower hospital mortality than general medical ward care (19 percent versus 32 percent); however, because this study spanned the interval during which the ICU was introduced, allocation to the ICU was based, in part, on the time period of enrollment and availability of the unit.

Warren, Flenley, Millar, et al. (1980) (external validity, 2/4; internal validity, Level 3) reported predictors of inhospital mortality in 135 patients on 157 consecutive admissions for ARF at an acute care hospital in the United Kingdom. The mortality rate was 18 percent (24/135 patients). Predictors of inhospital mortality on univariable analysis included older age, the highest level of PaCO₂ during oxygen therapy, and whether the lowest pH measured was less than 7.26.

In a subsequent study from the same institution (Jeffrey, Warren, and Flenley, 1992) (external validity, 2/4; internal validity, Level 3), these predictors were tested in a new group of 95 patients with COPD who had been hospitalized for acute hypercapnic respiratory failure. Although PaCO₂ was not a significant discriminator between those who died and those who survived in this group, pH was. With a threshold of 7.28, the lowest pH during hospital course performed as the best predictor of mortality. Contrary to expectation, the lowest pH was not recorded immediately before death in most cases (13/17 or 76 percent). Additional analyses identified several predictors measurable at admission that were predictive of inhospital mortality: high blood urea nitrogen, low systolic blood pressure, and low arterial pH. Two studies of somewhat more heterogeneous populations of patients admitted to the ICU with respiratory failure evaluated predictors of inhospital mortality; both included substantial numbers of patients with acute exacerbation of COPD (Heuser, Case, and Ettinger, 1992; Portier, Defouilloy, and Muir, 1992). Heuser, Case, and Ettinger (1992) (external validity, 0/4; internal validity, Level 2) studied 3,050 patients older than age 50 who were admitted to U.S. ICUs with lower respiratory tract diseases, 813 of whom had asthma or bronchitis. This retrospective study used clinical and laboratory data obtained at admission to calculate the MedisGroup's admitting severity group (ASG) score. For patients with obstructive lung disease, inhospital mortality was 6.7 percent (51/762). Patients with pneumonia had significantly higher mortality, as did those with worse ASG scores and older patients. The diagnosis of COPD was associated with a higher mortality rate, but this relationship varied with age. Younger patients with COPD had a lower mortality rate than those without COPD, but patients older than 76 years with COPD were more likely to die than those without COPD.

Portier, Defouilloy, and Muir (1992) (external validity, 1/4; internal validity, Level 1) included 322 patients who were admitted with ARF to one of nine medical ICUs in France. Patients with a variety of causes of chronic respiratory insufficiency were included; 45 percent had COPD. The ICU mortality rate for all patients was 10.9 percent (35/322 patients). The mortality rate was lower for those with COPD as the underlying cause of chronic respiratory insufficiency than for patients with other diagnoses. Other independent predictors of mortality were cachexia, low serum sodium, MV during the first 24 hours, confinement to home before admission, and presence of peripheral edema.

Fuso, Incalzi, Pistelli, et al. (1995) (external validity, 3/4; internal validity, Level 2) performed a large retrospective cohort study in Italy that described predictors of inhospital mortality in 590 consecutive patients who were hospitalized with acute exacerbation of COPD. Significant independent predictors included older age, PA-aO₂ gradient more than 41 mm Hg, presence of atrial fibrillation, and presence of ventricular arrhythmias. Variables that were significant in the univariable analysis, but not in the multivariable analysis, included use of digitalis, history of myocardial infarction, and the need for MV.

Two additional small studies failed to find any predictors of inhospital mortality (Dardes, Campo, Chiappini, et al., 1986; Moran, Green, Homan, et al., 1998) (external validity, 2/4 and 3/4, respectively; internal validity, both Level 3).

In summary, mortality in patients who were hospitalized for acute exacerbations and cared for in either general or ICU beds varied, ranging from 4 percent to 26 percent; study populations were not described well enough to explain this difference in overall mortality rates. Variation in mortality rates may be affected by local differences in ICU admission standards or other factors. Despite the wide variation in mortality rates, there is some agreement between studies about the prognostic factors that explain variation in mortality rate in patients. While potential predictors varied in studies, and the definitions and thresholds used for similar data varied, several trends can be observed. First, measures of acute physiology were well correlated with mortality (e.g., ABGs, Acute Physiology and Chronic Health Evaluation [APACHE] scores). Second, comorbid illness and other baseline preexacerbation health status measures (e.g., BMI, functional status) were associated with mortality. Few of the included studies reported specific data on baseline pulmonary function. Third, cumulative or longitudinal data on the clinical course also were important in describing mortality.

Need for mechanical ventilation. Three studies described predictors of the need for MV in patients hospitalized with an acute exacerbation of COPD (Bone, Pierce, and Johnson, 1978; Moran, Green, Homan et al., 1998; Vitacca, Clini, Porta, et al., 1996). All three studies were based on relatively small samples. Vitacca, et al. (1996) (external validity, 3/4; internal validity, Level 3), used discriminant analysis to identify predictors of the need for MV in a population of 39 patients in an Italian acute care hospital; 14 of these patients required MV. Those requiring ventilation had significant differences in 11 physiological parameters, most of which were included in the APACHE II score, which also was predictive. The multivariable analysis suggested that a worse nutritional prognostic index and FVC were the best independent predictors, discriminating between those patients who required MV and those who did not with a sensitivity of 0.69 and specificity of 0.79. However, given the small number of patients on which the modelling is based (only 14 required MV), the results are unreliable. The identified predictors were not tested in a separate group of patients.

An earlier prospective study (Bone, Pierce, and Johnson, 1978) (external validity, 2/4; internal validity, Level 3) identified predictors of the need for MV in a population of similar size: 50 patients with acute exacerbation of COPD, 13 of whom required ventilation. Blood gas values on admission and the change in pH after initial oxygen therapy were better predictors of the need for MV than were baseline spirometric values or changes in PaO₂ or PaCO₂ in response to oxygen therapy. A clinical prediction rule based on pH and PaO₂ had a sensitivity of 0.77 and specificity of 0.92 for predicting which patients would require ventilation. This prediction rule was tested in another group of patients who were assembled later at another institution. In the validation set, the prediction rule showed a slightly higher sensitivity of 0.81, but a lower specificity of 0.84.

A recent study of 75 patients with acute exacerbation of COPD who were admitted to the ICU failed to identify any factors that predicted the need for MV (Moran, Green, Homan, et al., 1998) (external validity, 3/4; internal validity, Level 3). Forty-three of 100 admissions were associated with ventilation. The relatively small size of the study may have prevented the identification of any premorbid or admission factors as predictive. Furthermore, because all patients were admitted to the ICU with symptoms of respiratory failure, the differences that may have led to MV may have been apparent only after admission as the clinical course was monitored.

In summary, several individual factors have been shown to be associated with the need for MV resulting from ARF in acute exacerbations of COPD. Acute respiratory physiology, as measured by blood gases, was predictive of the need for MV, along with baseline measures such as nutritional status. In one study, worsening acidosis after initiation of supplemental oxygen therapy added additional prognostic value to the blood gas values alone. When taken together in a multivariable model (Bone, Pierce, and Johnson, 1978), several factors have been shown statistically to discriminate between patients needing MV and those who did not. While the model was shown valid, it was not highly accurate and probably not sufficiently accurate for clinical decisionmaking regarding invasive ventilation.

Length of stay. LOS was reported in several cohort studies, and associated factors were described in three papers (Loukides and Polyzogopoulos, 1996; Moran, Green, Homan, et al., 1998; Mushlin, Black, Connolly et al., 1991). One study from Greece focused on differences in LOS in patients hospitalized with acute exacerbation of COPD with and without diabetes mellitus (Loukides and Polyzogopoulos, 1996) (external validity, 2/4; internal validity, Level 3). While the overall average LOS was 8.9 days, patients without comorbid diabetes mellitus averaged 8.53 days versus 10.76 days for patients with diabetes. Furthermore, diabetics who were insulin dependent had an average LOS of 15.63 days (significantly longer than patients with COPD alone) versus 7.69 days for noninsulin-using diabetics (not significantly different from patients with COPD alone).

An Australian study of 75 patients with acute exacerbation of COPD who were admitted to a respiratory ICU assessed predictors of LOS in the ICU and hospital (Moran, Green, Homan, et al., 1998) (external validity, 3/4; internal validity, Level 3). The average ICU LOS for the group was 5 days (range 1 to 40 days), and the average hospital LOS was 17 days (range 4 to 97 days). The ICU LOS was associated with the use of MV, greater time on ventilation, and the presence of tracheostomy; time on MV was the best independent predictor of LOS. Hospital LOS was associated with admission status (whether first ICU admission, repeat ICU admission, or repeat ICU admission within a single hospitalization) and tracheostomy status; however, admission status did not provide significant improvement to the discrimination of tracheostomy status alone in a multivariable model.

The third study described predictors of LOS in patients with acute exacerbation of chronic lung disease, half of whom had acute exacerbation of COPD (Mushlin, Black, Connolly, et al., 1991) (external validity, 0/4; internal validity, Level 3). (Forty-five percent had bronchospasm as the reason for admission.) The 83 consecutive patients in two hospitals had an average LOS of 8.7 days (range 1 to 57 days); 10 percent (8/83) required admission to the ICU (average ICU LOS 7.4 days). The investigators did not specify all of the predictive factors that they considered. Significant independent predictors of a longer LOS included hypercapnia on presentation, duration of symptoms of more than 1 day, and receipt of antibiotics on the day of admission.

In summary, few data exist to accurately predict LOS for patients who were hospitalized with acute exacerbation of COPD.

Antibiotics

Efficacy

The meta-analysis—published relatively recently in a high-profile, general medical journal—summarizes nearly all of the extant randomized trials comparing an antibiotic with placebo in patients with COPD and symptoms of exacerbation. The authors required studies to consider at least a 5-day duration of followup and to report data on continuous outcome measures. All antibiotic agents were considered together in the analysis. No subgroup analyses were planned a priori, but several post hoc analyses were performed based on level of care (inpatient or outpatient) and on a selection of particular outcome measures from individual reports. Adjustments were made for trials that used the number of exacerbations instead of the number of patients as the unit of analysis.

Nine trials met the inclusion criteria for the meta-analysis (Anthonisen, Manfreda, Warren, et al., 1987; Berry, Fry, Hindley, et al., 1960; Elmes, Fletcher, and Dutton, 1957; Elmes, King, Langlands, et al., 1965; Fear and Edwards, 1962; Jørgensen, Coolidge, Pedersen, et al., 1992; Nicotra, Rivera, and Awe, 1982; Petersen, Esmann, Høncke, et al., 1967; Pines, Raafat, Greenfield, et al., 1972). These trials used a variety of main outcome measures: PEFR, duration of exacerbation, PaO₂, symptom score, and overall score by physician. Of the nine studies, three found statistically significant effects individually (Anthonisen, Manfreda, Warren, et al., 1987; Berry, Fry, Hindley, et al., 1960; Pines, Raafat, Greenfield, et al., 1972);, three suggested a trend favoring antibiotics (Elmes, Fletcher, and Dutton, 1957; Elmes, King, Langlands, et al., 1965; Fear and Edwards, 1962); and three failed to show any difference from placebo (Jørgensen, Coolidge, Pedersen, et al., 1992; Nicotra, Rivera, and Awe, 1982; Petersen, Esmann, Høncke, et al., 1967).

To combine results from these disparate outcome measures, the authors calculated effect sizes (or standardized mean differences), a unitless measure of efficacy. Effect sizes that were based on these outcomes were statistically homogeneous and were combined to yield an overall estimate of 0.22 (95 percent CI, 0.1 to 0.34), which indicated a small but statistically significant effect favoring antibiotics over placebo.

Because effect sizes are difficult to interpret clinically, the authors analyzed a subset of trials that reported PEFR as the most frequently reported outcome measure—reported in six of the nine trials. Two of the six trials showed a trend (Elmes, King, Langlands, et al., 1965) or significant improvement (Anthonisen, Manfreda, Warren, et al., 1987) in PEFR favoring antibiotics. The trials were statistically homogeneous. A combined estimate of the difference in mean PEFR between antibiotic- and placebo-treated patients was 10.75 L/minute (min) (95 percent CI, 4.96 to 16.54).

The authors also reported an analysis that was stratified by level of care (outpatient versus inpatient). The summary effect size for outpatient studies was 0.17 (95 percent CI, 0.03 to 0.30) and 0.38 (95 percent CI, 0.13 to 0.62) for hospitalized patients.

The authors concluded that antibiotics yielded a small but statistically significant improvement compared with placebo that may be clinically significant, especially in patients with low baseline flow rates. However, they also acknowledged that these results should be interpreted with caution because the trials had important differences in selection of patients, outcomes measured, and specific antibiotic agents utilized.

In the meta-analysis, it was not possible to investigate a relationship between antibiotic efficacy and severity of illness, sputum purulence, or bacterial cultures—with the exception of an analysis stratified by level of care. However, several of the trials analyzed the efficacy of antibiotics according to subgroups that were defined either by evidence of bacterial infection or severity of illness (Anthonisen, Manfreda, Warren, et al., 1987; Berry, Fry, Hindley, et al., 1960; Elmes, King, Langlands, et al., 1965). One trial found that a priori criteria that were proposed to select patients with signs of infection (Winnipeg criteria) showed a relationship of better outcomes with antibiotic versus placebo treatment (Anthonisen, Manfreda, Warren, et al., 1987) (external validity, 5/5; internal validity, 4/5). Patients with type-1 exacerbations (who met all three criteria: increases in amount of sputum, purulence of sputum, and dyspnea) benefited the most, with resolution of symptoms in 63 percent of the antibiotic-treated exacerbations and 43 percent of the placebo-treated exacerbations. Patients with type-3 exacerbations (who met one of the above three criteria) did not show any benefit, with 74 percent of exacerbations resolving on antibiotics and 70 percent resolving on placebo. Those with type-2 exacerbations (who met two of the above three criteria) showed an intermediate (and not statistically significant) benefit, with 70 percent resolving on antibiotics and 60 percent resolving on placebo.

One trial that compared oxytetracycline with placebo in 53 patients with acute exacerbations assessed the severity of exacerbation at presentation (Berry, Fry, Hindley, et al., 1960) (external validity, 2/5; internal validity, 3/5). The 4-point severity scale was subjective, with investigators rating the severity as “baseline,” “mild,” “moderate,” or “severe.” The analyses for efficacy of antibiotic therapy—assessed on days 2, 7, and 14—were stratified by severity. For patients presenting with mild attacks (n = 24), there were no significant differences in severity of symptoms between antibiotic- and placebo-treated patients at any time point. For patients presenting with moderate or severe attacks, antibiotic-treated patients had significantly less severe symptoms on days 2 and 7; however, the differences were not statistically significant at day 14.

Another trial matched patients based on severity of illness, which was defined as two or more of the following: fever higher than 37.5 °C, pulmonary consolidation, or purulent sputum (Elmes, King, Langlands, et al., 1965) (external validity, 2/5; internal validity, 5/5). The trial was terminated based on a statistical stopping rule when a significant effect of antibiotic treatment was found; however, this analysis was based on an assessment that was not blinded to bacteriologic results and, thus, may have been biased. A later independent, blinded assessment failed to find a significant difference between antibiotic- and placebo-treated patients. Patients with greater severity of illness had more relapses in the hospital than did less severely ill patients in the placebo group; however, this relationship was not found in the antibiotic group. It should be noted that the criterion for “severely ill” (met by 19 of the 74 patients in the trial) would have resulted in exclusion from other studies of acute exacerbation of COPD because of either fever or pulmonary consolidation.

Table 9. Characteristics of RCTs of antibiotics in (more...)

Table 9. Characteristics of RCTs of antibiotics in acute exacerbations of COPD

Study	Number of Patients	Mean PEFR at entry (L/min)	Patients with Purulent Sputum (%)	Level of Care	Gluco-corticoid Use	Results1
Jørgensen, Coolidge, Pedersen, et al., 1992	268	2952	33%3	Opt	Prohibited	- Overall clinical assessment
						- Symptoms (MD)
						- PEFR
Sachs, Koëter, Groenier, et al., 1995	71	233	27%	Opt	Prescribed	- Symptoms (pt)
						- PEFR
Petersen, Esmann, Høncke, et al., 1967	19	2144	74%	Inpt	N/S	- PEFR
Anthonisen, Manfreda, Warren, et al., 1987	173	190	60%	Opt	Permitted (42% all)	+ Overall clinical assessment
						+ PEFR
Nicotra, Rivera, and Awe, 1982	40	160	N/S	Inpt	Permitted (75% abx; 65% pbo)	- Symptoms (pt)
						- Symptoms (MD)
						- FEV1, PEFR, FVC
Pines, Raafat, Greenfield, et al., 1972	259	146	100% 5	Inpt	N/S	+ Overall clinical assessment
						+ Symptoms (MD)
						- PEFR
Pines, Raafat, Plucinski, et al., 1968	30	88	100%	Inpt	N/S	+ Overall clinical assessment
Elmes, King, Langlands, et al., 1965	58	79	78%	Inpt	Prohibited	- Overall clinical assessment
						- PEFR
						- Length of stay
Berry, Fry, Hindley, et al., 1960	53	N/S	60%	Opt	N/S	+ Symptoms (MD), pts with moderate to severe exacerbations
						- Symptoms (MD), pts with mild exacerbations
Elmes, Fletcher, and Dutton, 1957	59	N/S	N/S	Opt	N/S	- Duration of symptoms
						- Work days lost
Fear and Edwards, 1962	62	N/S	N/S	Opt	N/S	- Symptoms (MD)
						- Duration of symptoms

1

Results: “+” indicates “statistically significant difference between antibiotic- and placebo-treated patients”; “-” indicates “no significant difference between antibiotic- and placebo-treated patients.”

2

Estimated from figure.

3

Sputum color: “yellow” versus “none, clear, or white.”

4

Weighted average of men's median PEFR and women's median PEFR in control group (n = 10); value for active-treatment group could not be similarly estimated.

5

Moderately purulent or purulent.

Abx = antibiotics; COPD = chronic obstructive pulmonary disease; FVC = forced vital capacity; FEV₁ = forced expiratory volume in 1 second; inpt = inpatient; L = liter; MD = physician-assessed; min = minute; N/S = not specified; opt = outpatient; pbo = placebo; PEFR = peak expiratory flow rate; pt = patient-assessed; RCT = randomized controlled trial

The different study populations appear to have clinically important differences in severity of illness. Table 9 describes characteristics of the study populations in terms of PEFR, sputum purulence, and level of care. One publication (Pines, Raafat, Plucinski, et al., 1968) (external validity, 2/5; internal validity, 5/5) described a pilot trial comparing penicillin and streptomycin with placebo in 30 patients hospitalized with acute exacerbation of chronic bronchitis. All patients had purulent sputum, and half were characterized as severely ill. The study population had a mean age of 67.5 years, a mean PaCO₂ of 70 mm Hg, a mean PEFR of 88 L/min, a mean leukocyte count of 12 × 10⁹/L, and a mean fever of 37.3° C. Patients were examined daily by a clinical assessor who was blinded to treatment assignment, and their clinical status was graded based on “obvious changes in the degree of well-being, colour, and dyspnoea,” but not based on the state of the sputum. In the antibiotic-treated group, 10 patients improved, 3 remained unchanged, and 2 deteriorated (1 of whom died); in the placebo group, 3 patients improved, 3 remained unchanged, and 9 deteriorated (3 of whom died). The difference in deterioration was statistically significant (p < 0.05). The trial was halted early because of the high number of deteriorations. The investigators concluded that it was unethical to include a placebo group in the subsequent trial. Results were not stratified according to severity of illness.

Another trial, which was not included in the meta-analysis, considered 71 outpatients with COPD and increasing dyspnea (despite increased inhaled bronchodilator therapy) (Sachs, Koëter, Groenier, et al., 1995) (external validity, 4/5; internal validity, 4/5). These patients were randomly allocated to receive co-trimoxazole, amoxicillin, or placebo; all patients received prednisolone (35 milligrams [mg] on the first day, reduced by 5 mg daily). Symptoms and PEFR were assessed during the ensuing 2 weeks. The patients had a mean age of 51.7 years and a mean PEFR of 233 L/min. Fifty of the 71 patients produced sputum, and sputum was purulent in 19 patients (27 percent). No differences were observed in recovery rate or changes in symptom score, PEFR, temperature, or sputum. The authors suggest that the antiinflammatory effect of the prednisolone cointervention preempted a clinically important effect from antibiotic treatment. They noted that the study population's relatively high PEFR and the relatively low proportion of patients with purulent sputum (unlike previous studies that showed a benefit to antibiotic treatment) also might explain this finding.

Studies varied with regard to whether corticosteroid treatment was permitted (see Table 9). Only one study provided any comment on the association of corticosteroid treatment with antibiotic efficacy (Anthonisen, Manfreda, Warren, et al., 1987) (external validity, 5/5; internal validity, 4/5). In this study, corticosteroid therapy was permitted, but not prescribed or assigned in any way by the study protocol. Corticosteroids were used in approximately 42 percent of the 362 exacerbations. The success rate was reported to be better for antibiotic- than placebo-treated patients in patients who were on corticosteroids and those who were not (though no test of significance is reported).

Bronchodilating Drugs

Worsening airflow obstruction is characteristic in the COPD patient with an acute exacerbation. To some extent, the severity of the exacerbation is based on the degree of constriction or obstruction present. Along with airway inflammation, contraction of smooth muscle within the bronchial tree is responsible for this phenomenon. Bronchodilators primarily relax smooth muscle and also may reduce inflammation. The three most commonly used classes of agents—anticholinergics, beta₂-agonists, and methylxanthines—are discussed here.

Anticholinergics reduce airflow obstruction by blocking the effects of acetylcholine, a potent bronchoconstrictor, and by reducing the production of secretions. The two anticholinergics discussed here are ipratropium bromide and glycopyrrolate. Beta₂-agonists bind to beta₂-receptors in bronchial smooth muscle. Once bound, their action results in bronchodilation, decreased production of secretions, and vasoconstriction. The beta₂-agonists studied in the included trials were albuterol (salbutamol), fenoterol, metaproterenol, salmeterol, and terbutaline. Theophylline and aminophylline are the best known of the methylxanthines. Researchers believe that these agents reduce airflow obstruction, at least partially, by blocking the function of intracellular phosphodiesterase, allowing the smooth muscle-relaxing effects of cyclic adenosine monophosphate to persist.

Studies addressing the efficacy of bronchodilators in acute exacerbation of COPD fall into four categories: (1) head-to-head comparisons of two or more drugs or drug combinations, (2) addition of a drug to a standard treatment regimen, (3) dose comparisons involving a single drug, and (4) comparisons of different drug delivery systems (MDI versus nebulizer).

Efficacy

The SCCOPE trial, a large multicenter study conducted in U.S. Department of Veterans Affairs hospitals, was designed to compare treatment failure (death, intubation, readmission, or intensification of drug treatment) rates between patients who did and did not receive systemic corticosteroids (Niewoehner, Erbland, Deupree, et al., 1999) (external validity, 2/5; internal validity, 4/5). Patients who were admitted for an acute exacerbation of COPD and who had not used systemic corticosteroids in the previous 30 days were assigned to receive 3 days of IV methylprednisolone or placebo. The IV steroids were followed by oral prednisone in a tapering dose over 12 days or 8 weeks. Allocation to the 2- or 8-week course was random. There were no important differences in any efficacy outcomes or adverse effects between the two groups; for most analyses reported, the two groups were combined. For the combined corticosteroid group, the risk of treatment failure was reduced by 10 percent, and FEV₁ showed an improvement averaging about 0.1 L in the first 3 days of treatment. No differences were observed in length of hospitalization or mortality. The improvement in FEV₁ observed in this trial is remarkably similar to the magnitude of benefit reported in several previous small trials, thus reinforcing the generalizability of this finding to different settings, populations, and dosages.

The ATS suggests that a difference of 13 percent in FEV₁, developing over a short period, is clinically important (American Thoracic Society, 1991). For the population in the SCCOPE study, which had an FEV₁ at admission of 767 mL, the 100 mL difference that was observed represents precisely a 13 percent difference. Thus, the magnitude of improvement in FEV₁ that is attributable to systemic corticosteroid observed in the SCCOPE trial appears to be clinically important. Furthermore, the SCCOPE trial provides a link between improvement in FEV₁ and improvement in clinical outcomes—length of hospital stay and treatment failure rate.

Table 10. Dosage comparison of corticosteroid trials (more...)

Table 10. Dosage comparison of corticosteroid trials

Study	Corticosteroid Agent	Dose (mg)	Route	Schedule	Equiv. 1st-day Dosage of Methyl-prednisolone (mg)
Bullard, Liaw, Tsai, et al., 1996	Hydrocortisone	100	IV	Once	20
Davies, Angus, and Calverley, 1999	Prednisolone	30	PO	Daily for 14 days	37.5
Thompson, Nielson, Carvalho, et al., 1996	Prednisone	60	PO	Daily for 3 days, then taper	75
Emerman, Connors, Lukens, et al., 1989b	Methyl-prednisolone	100	IV	Once	100
Albert, Martin, and Lewis, 1980	Methyl-prednisolone	35 (based on 0.5 mg/kg)	IV	Ev. 6 hours for 3 days	140
Niewoehner, Erbland, Deupree, et al., 1999	Methyl-prednisolone	125	IV	Ev. 6 hours for 3 days, followed by PO prednisone taper	500

Equiv. = equivalent; Ev. = every; IV = intravenous; kg = kilogram; mg = milligram; PO = orally (per os)

Several trials have examined the time course of improvement in FEV₁ during treatment with systemic corticosteroids. The two trials that considered short-term outcomes of ED treatment failed to find significant differences in FEV₁ between corticosteroid- and placebo-treated patients (Bullard, Liaw, Tsai, et al., 1996; Emerman, Connors, Lukens, et al., 1989b) (external validity, 1/5 and 3/5, respectively; internal validity, 4/5 and 4/5, respectively). However, trials that measured FEV₁ changes over a longer period of time did show significant differences. One trial measured FEV₁ improvement at only one time point (72 hours) and found a statistically significant improvement in patients treated with methylprednisolone (Albert, Martin, and Lewis, 1980) (external validity, 4/5; internal validity, 5/5). Other trials measured FEV₁ at multiple time points over longer time frames and found that most of the improvement occurs in the first 3–5 days of corticosteroid treatment (Davies, Angus, and Calverley, 1999; Niewoehner, Erbland, Deupree, et al., 1999; Thompson, Nielson, Carvalho, et al., 1996). Thompson, Nielson, Carvalho, et al. (1996) (external validity, 5/5; internal validity, 3/5) measured FEV₁ at days 3 and 10; while there was a trend toward better FEV₁ improvement in the steroid-treated group at day 3, this effect was significant at day 10, and the mean slope was significantly better. Davies, Angus, and Calverley (1999) (external validity, 4/5; internal validity, 5/5), measuring FEV₁ daily, found that by day 5, patients in the steroid-treated group had increased postbronchodilator FEV₁ to 92 percent of discharge values compared with 85 percent in the placebo-treated group (p < 0.04). In the SCCOPE trial, the difference in FEV₁ between corticosteroid- and placebo-treated patients was highest after the first day of treatment, remained statistically significant after the second and third days, and was no longer significant at 2 weeks.

Adverse Effects

The most common adverse effect that was associated with systemic corticosteroids for acute exacerbation of COPD was hyperglycemia, which was reported in each of the three trials that provided data on adverse effects (Albert, Martin, and Lewis, 1980; Davies, Angus, and Calverley, 1999; Niewoehner, Erbland, Deupree, et al., 1999). Transient glycosuria developed in 6 of 28 patients who were treated with 30 mg of prednisolone, but it did not develop in the 22 patients who were on placebo (Davies, Angus, and Calverley, 1999); the clinical importance of the glycosuria was not described. After 3 days of methylprednisolone at 140 mg/d, mean blood glucose was significantly higher in the methylprednisolone group (164 ± 42 mg/deciliter [dL]) than in the placebo group (139 ± 29 mg/dL) (p < 0.05), with the highest glucose level being 265 mg/dL in a corticosteroid-treated patient (Albert, Martin, and Lewis, 1980). In the SCCOPE trial, hyperglycemia that required treatment occurred significantly more often in the combined corticosteroid groups (15 percent) than in the placebo group (4 percent) (p = 0.002) (Niewoehner, Erbland, Deupree, et al., 1999). Two-thirds of the episodes of hyperglycemia that required treatment occurred in patients who were known to have diabetes mellitus. Nearly all episodes occurred in the first 30 days; whether hyperglycemia was more frequent or severe in the 8-week or 2-week course was not described.

The other trials considered above provided only limited data about short-term adverse effects that were associated with corticosteroids.

Summary

Several RCTs provide good evidence for a benefit from a short course of systemic corticosteroids in patients with acute exacerbation of COPD who require hospitalization. The SCCOPE trial included a randomized comparison between a 2- and 8-week course of systemic corticosteroids. Based on the finding that these courses were not importantly different in clinical outcome, the investigators concluded that the shorter course, which reduced adverse effects, is preferred. The optimal dose and duration of treatment remain uncertain, however, because small studies suggest that even lower doses (Davies, Angus, and Calverley, 1999) and even shorter courses of treatment (Albert, Martin, and Lewis, 1980) also may be effective.

Mucolytic Drugs

Efficacy. We identified six controlled trials involving six different mucolytic drugs. Comparisons tested included domiodol versus control (Finiguerra, Conti, Figura, et al., 1982) (external validity, 1/5; internal validity, 1/5), bromhexine versus placebo (Langlands, 1970) (external validity, 2/5; internal validity, 5/5), ambroxol versus control (Peralta, Poderoso, Corazza, et al., 1987) (external validity, 2/5; internal validity, 3/5), S-carboxymethylcysteine versus bromhexine (Aylward, 1973) (external validity, 3/5; internal validity, 4/5), erdostein versus ambroxol (Fumagalli, Balzarotti, Banfi, et al., 1988) (external validity, 2/5; internal validity, 3/5), and potassium iodide versus chloramphenicol, physiotherapy, and placebo (Petersen, Esmann, Høncke, et al., 1967) (external validity, 2/5; internal validity, 1/5). None of these trials reported statistically significant differences in mean FEV₁ between treatments. Although we did not abstract data on sputum volume or viscosity, we did record data on subjective symptom scores on difficulty with expectoration. Of the five trials measuring such data, two reported a statistically significant difference (p < 0.01) favoring the mucolytic drug over control (Finiguerra, Conti, Figura, et al., 1982; Peralta, Poderoso, Corazza, et al., 1987).

Adverse effects. Limited data on adverse effects were available from the controlled trials described above. The effects reported were primarily gastrointestinal intolerance that was associated with bromhexine.

Limitations. Most trials involving mucolytic drugs focused on populations with stable chronic bronchitis and were excluded. The remaining trials often recorded data on outcomes that we perceived as either irrelevant or difficult to interpret (e.g., sputum production or sputum viscosity). Seven trials were excluded because they had only physiological outcomes (e.g., sputum rheology or viscosity). We recorded data on the ventilatory parameters and symptoms of ease or difficulty of expectoration when these were available; however, one trial that included patients with acute exacerbation of COPD and measured suitable clinical outcomes could not be included because it did not report results for patients with acute exacerbation of COPD separately from the results for patients with acute bronchitis—who formed the majority of the study population (Jager, 1989).

Summary. Available studies show no benefit from any mucolytic drugs studied (ambroxol, bromhexine, domiodol, potassium iodide, and S-carboxymethyl cysteine) in improving ventilatory function in acute exacerbations; some studies report subjective improvement in symptoms associated with decreasing sputum viscosity.

Physical Therapy

Efficacy. Three RCTs of chest physiotherapy were included (Newton and Bevans, 1978; Petersen, Esmann, Høncke, et al., 1967; Wollmer, Ursing, Midgren, et al., 1985) (external validity, 3/5, 2/5, and 2/5, respectively; internal validity, 3/5, 1/5, and 1/5, respectively). A fourth study in a group of patients with acute exacerbation of COPD did not report suitable outcome data (only blood gases, temperature, and sputum production) (Anthonisen, Riis, and Søgaard Andersen, 1964). Three other controlled trials of various physical therapy modalities were conducted in patients who were not in acute exacerbation (Maloney, Fernandez, and Hudgel, 1981; van Hengstum, Festen, Beurskens, et al., 1990; van Hengstum, Festen, Beurskens, et al., 1991) or who were in postexacerbation (Kirsten, Taube, Lehnigk, et al., 1998).

None of the included trials reported any benefit over control for ventilatory function (FEV₁ or FVC). One trial described a significantly lower FEV₁ in patients who received chest percussion therapy compared with control (Wollmer, Ursing, Midgren, et al., 1985). A similar transient decrease in FEV₁ following chest percussion was previously described in an uncontrolled study (Campbell, O'Connell, and Wilson, 1975).

Adverse effects. Other than the data on short-term decrease in FEV₁ immediately following chest physiotherapy, no other information on adverse effects was provided.

Summary. Available studies of chest physiotherapy fail to show any improvement in short-term ventilatory function for patients with acute exacerbation of COPD.

Data from RCTs

Patient populations were similar in the four trials with no treatment controls (Barbé, Togores, Rubí, et al., 1996; Bott, Carroll, Conway, et al., 1993; Brochard, Mancebo, Wysocki, et al., 1995; Kramer, Meyer, Meharg, et al., 1995). Each trial included patients with severe COPD—as defined by the ATS (American Thoracic Society, 1995)—who had respiratory failure. Excluded were patients with causes for respiratory failure other than exacerbation of COPD, such as CHF, pneumothorax, and PE. Each study also excluded patients who needed immediate intubation due to hemodynamic instability or central nervous system depression. Respiratory failure was primarily hypercapneic in origin, with varying levels of hypoxia.

The studies employed a variety of techniques for NPPV (including the use of pressure-, time-, or volume-cycled ventilation) and equipment (including face and nasal masks, or nasal pillows). Either investigators or a respiratory therapist monitored how well these devices fit their patients. Cointerventions typically included bronchodilators, corticosteroids, supplemental oxygen, and antibiotics; however, they were standardized for all subjects in only one study (Barbé, Togores, Rubí, et al., 1996). Measured outcomes we considered were: improvements in blood oxygen and carbon dioxide levels (PaO₂, PaCO₂), spirometry (FEV₁, FVC), admissions to the ICU, need for subsequent intubation, hospital LOS, and mortality. Any adverse effects reported in the trials also were recorded.

The rate of intubation was compared in four of the RCTs (Barbé, Togores, Rubí, et al., 1996; Bott, Carroll, Conway, et al., 1993; Brochard, Mancebo, Wysocki, et al., 1995; Kramer, Meyer, Meharg, et al., 1995). A significant difference in the need for intubation was found in two trials, with a reduced need for intubation in the NPPV groups (Brochard, Mancebo, Wysocki, et al., 1995; Kramer, Meyer, Meharg, et al., 1995) (external validity, 2/5 in both trials; internal validity, 3/5 and 2/5, respectively). Mortality was examined in these two trials and in three other trials that compared NPPV with control (Angus, Ahmed, Fenwick, et al., 1996; Barbé, Togores, Rubí, et al., 1996; Bott, Carroll, Conway, et al., 1993) (external validity, 3/5, 3/5, and 2/5, respectively; internal validity, 1/5, 2/5, and 2/5, respectively). The mortality data were analyzed in a recent systematic review on this topic (Keenan, Kernerman, Cook, et al., 1997). This analysis included a meta-analysis of the subset of studies in which NPPV was used for ARF due to acute exacerbation of COPD exclusively. The findings of this meta-analysis were that NPPV significantly reduced both the odds ratio (OR) for mortality (OR 0.22; 95 percent CI, 0.09 to 0.54) and the OR for the need for intubation (OR 0.12; 95 percent CI, 0.05 to 0.29) in comparison with control.

Since this meta-analysis was performed, new data have become available from Barbé, Togores, Rubí, et al. (1996) (external validity, 3/5; internal validity, 2/5). Also, although the abstract by Ahmed, Fenwick, Angus, et al. (1992) comparing the use of NPPV with doxapram, a respiratory stimulant, was included in the meta-analysis, data from these and from additional patients were reported in the final paper of this trial (Angus, Ahmed, Fenwick, et al., 1996) (external validity, 3/5; internal validity, 1/5). We repeated the analysis, incorporating the new data from these two trials. For mortality, an overall test for homogeneity of the OR was not significant (p = 0.765). Using a fixed-effect OR that was calculated using Peto's method, the OR for mortality associated with NPPV versus control was 0.25 (95 percent CI, 0.11 to 0.55), which is similar to the findings reported in the original analysis (Keenan, Kernerman, Cook, et al., 1997). The new trials (Angus, Ahmed, Fenwick, et al., 1996; Barbé, Togores, Rubí, et al., 1996) had small populations (n = 17 and n = 24, respectively) compared with two of the largest trials (n = 60 and n = 85, respectively) (Bott, Carroll, Conway, et al., 1993; Brochard, Mancebo, Wysocki, et al., 1995) that were included by Keenan, Kernerman, Cook, et al. (1997). We did not analyze data regarding the need for intubation, but we expect that this summary estimate would not change substantially with the inclusion of the new data.

Data from Retrospective Concurrent or Historical Cohort Comparisons

Prospective case series with historical control groups also have been described (Brochard, Isabey, Piquet, et al., 1990; Corbetta, Ballerin, Putinati, et al., 1997; Foglio, Vitacca, Quadri, et al., 1992; Hilbert, Gruson, Gbikpi-Benissan, et al., 1997; Servera, Peréz, Marín, et al., 1995). The controls did not receive NPPV and were well matched to the cases in terms of demographics and clinical status at presentation. Results from these trials were similar to those from the RCTs: the groups receiving NPPV had significant improvements in PaCO₂ and pH, fewer intubations, and a lower mortality rate when compared with the historical controls. Also decreased were hospital LOS and use of invasive ventilation when compared with the controls that were eventually intubated. The exception to these conclusions was the study by Foglio, Vitacca, Quadri, et al. (1992). This research group found no increased effectiveness of NPPV over more conventional treatment with bronchodilators and corticosteroids. In addition, there were large numbers of adverse effects associated with the use of the NPPV.

Other Designs/Questions

Additional questions that were addressed in the literature included comparisons between NPPV and invasive ventilation, optimal NPPV delivery methods, and predictors of successful application of NPPV. One retrospective historical cohort study compared the prognosis of patients with ARF secondary to acute exacerbation of COPD who received NPPV with the prognosis of patients who were intubated for MV (Vitacca, Clini, Rubini, et al., 1996). This study looked at mortality rates in hospitals at 3 and 12 months, and at hospital days during the year post-discharge. They found a significantly increased mortality rate for patients who had been intubated (63 percent versus 30 percent in the NPPV group, p < 0.005). In both groups, patients with a diagnosis of pneumonia had a worse outcome. There also were more ICU admissions, longer hospital stays, and longer ventilation times for patients undergoing invasive ventilation. One author found similar results in his review of assist control, pressure-targeted ventilation using BiPAP^® (Respironics, Pittsburgh, PA) and traditional MV patients (Confalonieri, Parigi, Scartabellati, et al., 1996).

Predictors of Success

A retrospective analysis of 59 episodes of acute exacerbation of COPD was undertaken by several investigators in an attempt to identify parameters that could predict a successful outcome of NPPV (Ambrosino, Foglio, Rubini, et al., 1995). Identifying these parameters could potentially prevent delays in those patients who will eventually require intubation. The investigators looked at anthropometric and demographic characteristics, nutritional status, spirometry, blood gases, and causes of acute exacerbation of COPD. Factors that predicted success included higher pH, lower PaCO₂, and higher FVC (p < 0.05). Poor outcomes were associated with a diagnosis of pneumonia, poor nutritional status, and decreased compliance with the apparatus.

Mask Interfaces

A recurring theme throughout the reports is the need for a mask that fits reasonably well and is comfortable. One investigator reviewed a series of 25 studies involving the use of NPPV for ARF and found that the incidence of mask intolerance ranged from 3 to 33 percent (Meduri, 1996). The review did not uncover a clear association between mask type and facial skin necrosis. Brochard, Mancebo, Wysocki, et al. (1995) were the only investigators to use facemasks in their study of NPPV using an RCT design. They also had the highest complication rate (20.9 percent), including four deaths in the NPPV arm. However, mask comfort/tolerance was not described in this study. One study had three episodes of claustrophobia that complicated the tolerability of a nasal mask (Barbé, Togores, Rubí, et al., 1996). Authors of another study (Kramer, Meyer, Meharg, et al., 1995) provided great detail about masks and tolerance. The subjects were administered BiPAP® (Respironics, Pittsburgh, PA) through a nasal mask. When they were first placed on the ventilation, “coaching” was necessary to help with anxiety and synchronization of respiration. The inability to synchronize led to the eventual intubation of one person. Two patients (18 percent) did not tolerate the nasal mask.

Adverse Effects

The adverse effects of NPPV can be compared with those of the invasive ventilation techniques or conservative management methods used in the RCTs discussed below. In addition, there are adverse effects that were unique to the equipment and techniques involved in administering NPPV therapy to patients. Traditional invasive ventilation has been associated with complications such as sinusitis, pneumonia, and barotrauma, but NIV has not. However, both techniques have been associated with gastric distention. At present, no study has been done to compare the adverse effects that are associated with these two ventilation techniques.

In the five RCTs, we looked for comparisons of adverse effects that occurred in the use of NPPV and conservative treatment. Three studies (Angus, Ahmed, Fenwick, et al., 1996; Barbé, Togores, Rubí, et al., 1996; Bott, Carroll, Conway, et al., 1993) did not provide data on adverse effects other than mortality and the need for invasive intubation. These have been discussed previously in this report. The other two studies (Brochard, Mancebo, Wysocki, et al., 1995; Kramer, Meyer, Meharg, et al., 1995) provided limited comparison data. Nosocomial pneumonia, tension pneumothorax, and a steroid-induced myopathy were seen in control-group patients requiring intubation in one study (Kramer, Meyer, Meharg, et al., 1995). The NPPV group did not experience these complications as a result of the treatment. The other study (Brochard, Mancebo, Wysocki, et al., 1995) showed a three-fold increase in the incidence of pneumonia for the conservative treatment group compared with the NPPV-treated group.

Claustrophobia is a common problem and is a cause of much anxiety for patients with COPD. Wearing a facemask could aggravate that problem; however, nasal masks also were associated with claustrophobia in three episodes (Boye and Gaustad, 1991).

One investigator compiled a table of adverse effects that were associated with the use of NPPV based on data from 25 controlled and uncontrolled trials (Meduri, 1996). No incidence of barotrauma was noted in any of the studies reviewed. The most common adverse effects reported in the trials are facial skin abrasions/necrosis (2.1 to 20.6 percent), pneumonia (0.6 to 5.5 percent), retention of secretions (0 to 16.7 percent), conjunctivitis (0 to 14.2 percent), and gastric distention (1.8 to 13.3 percent). Many studies did not report any adverse effects.

Table 11. Studies of characteristics and adverse effects (more...)

Table 11. Studies of characteristics and adverse effects of noninvasive positive pressure ventilation¹

							Adverse Effects
Study	No. of Pts	Mask	VMODE1	VMODE2	Success (No.)	%	INTOL	FN	PN	SEC	CONJ	GD
Meduri, Conoscenti, Menashe, et al. (1989)	10	0	C	P	7	70	1	1	0	0	0	0
Brochard, Isabey, Piquet, et al. (1990)	13	0	I	--	12	92	--	--	--	--	--	--
Elliott, Steven, Phillips, et al. (1990)	6	1	A	--	5	83	1	0	0	1	0	0
Meduri, Abou-Shala, Fox, et al. (1991)	18	1	C	P	13	72	2	1	1	0	0	0
Marino (1991)	13	01	A	--	9	69	4	--	--	--	--	--
Chevrolet, Jolliet, Abajo, et al. (1991)	6	--	A	--	3	50	0	--	--	--	--	--
Pennock, Kaplan, Carlin, et al. (1991)	29	1	B	--	22	76	3	0	0	0	0	0
Hodson, Madden, Steven, et al. (1991)	6	1	A	--	5	83	--	--	--	--	--	--
Benhamou, Girault, Faure, et al. (1992)	30	1	A	--	18	60	7	2	0	5	0	0
Wysocki, Tric, Wolff, et al. (1993)	17	0	P	--	8	47	1	1	0	0	0	0
Vitacca, Rubini, Foglio, et al. (1993)	29	0	P	A	24	83	0	6	0	0	0	1
Fernandez, Blanch, Valles, et al. (1993)	12	0	C	P	9	75	0	0	0	2	0	0
Bott, Carroll, Conway, et al. (1993)	30	1	A	--	26	86	--	--	--	--	--	--
Pennock, Crawshaw, and Kaplan (1994)	110	1	B	--	84	76	--	--	--	--	--	--
Meduri, Fox, Abou-Shala, et al. (1994)	11	0	C	P	7	64	1	1	0	0	0	0
Soo Hoo, Santiago, and Williams (1994)	12	1	A	--	5	42	--	--	--	--	--	--
Tognet, Mercatello, Polo, et al. (1994)	15	01	P	--	6	40	5	0	0	0	0	2
Wysocki, Tric, Wolff, et al. (1995)	21	0	C	P	12	62	--	--	--	--	--	--
Confalonieri, Aiolfi, Parigi, et al. (1995)	28	1	B	--	18	64	2	2	0	0	4	0
Kramer, Meyer, Meharg, et al. (1995)	16	1	B	--	27	93	2	2	0	0	0	0
Sacchetti, Harris, Paston, et al. (1995)	22	1	B	--	20	91	--	--	--	--	--	--
Pollack, Fleisch, and Dowsey (1995)	50	0	B	--	43	7	2	--	--	--	--	--
Brochard, Mancebo, Wysocki, et al. (1995)	43	1	I	--	32	74	--	1	1	0	0	0
Ambrosino, Foglio, Rubini, et al. (1995)	59	0	P	A	46	78	--	--	--	--	--	--
Meduri, Turner, Abou-Shala, et al. (1996)	158	0	C	P	112	65	6	20	1	0	0	3

1

Adapted with publisher's permission from: Meduri GU. Noninvasive positive-pressure ventilation in patients with acute respiratory failure. Clin Chest Med 1996;17(3):513–53.

A = assist-control ventilation; B = BiPAP (bilevel positive airway pressure); C = CPAP (continuous positive airway pressure); CONJ = conjunctivitis; FN = facial necrosis; GD = gastric distention; I = IPAP (inspiratory positive airway pressure); INTOL = intolerance; P = pressure-controlled ventilation; MASK = type of mask used; 0 = facial; 1 = nasal; PN = pneumonia; SEC = retention of secretions; SUCCESS = no. of pts successful on NPPV; VMODE1 = ventilator mode 1; VMODE2 = ventilator mode 2.

Limitations

Limitations of the trials were few, but some were substantial. First, NPPV trials cannot be blinded, because the introduced bias would be impossible to eliminate; however, the randomized studies did prove to be a more valid test of effectiveness than the many nonrandomized studies. The cointerventions were not standardized in most instances, and researchers did not control this aspect of the studies. The criteria to determine (1) the need for intubation and (2) the subsequent decision to intubate were not clearly defined by the authors. All of the RCTs used PaO₂ and/or PaCO₂ as their criteria for inclusion. Brochard, Mancebo, Wysocki, et al. (1995) had a more seriously ill population, which may explain a high intubation rate that was not seen in the other studies.

Summary

In summary, NPPV is an effective means of avoiding MV for patients with ARF. Optimal use must include considerations of both patient condition and type of apparatus. Patients must be conscious and able to tolerate using the apparatus. Those with rapidly deteriorating respiratory status (pH < 7.3) should be considered for immediate intubation and MV. The selection of a nasal or face mask should be based on patient comfort. PSV and CPAP or IPAP/EPAP are the ventilation modes that are best tolerated by patients and are most effective for correcting blood gas abnormalities. NPPV also is a labor-intensive ventilation method. Respiratory care or nursing personnel must follow the patient closely to ensure that his/her breathing is synchronized with the ventilator and to help minimize air leaks and adverse effects. Ventilators must be monitored for malfunctions, and patients must be monitored for the effects of NPPV so that intubation, if needed, is not delayed. Such monitoring of patients and devices usually is performed in a special unit such as an ICU or respiratory care unit.

Outpatient Studies

Risk of relapse in patients presenting with acute exacerbation of COPD and selected for outpatient treatment was between 11 and 17 percent at 48 hours and between 23 and 32 percent at 2 weeks. In all patients presenting to the ED, hospitalization at index visit varied from 24.2 to 71 percent. Mean age, spirometric values, and other descriptive data were reasonably similar in studies, suggesting similar populations. Data from the previous history of individual patients were consistently identified as having value for predicting relapse, (e.g., previous visit within 7 days, number of exacerbations in the past year, and relapsing on previous visits). Also predictive in several studies was baseline pulmonary function, as measured by FEV₁ or FVC. Data describing acute respiratory physiology such as FEV₁ during exacerbation or ABGs also predicted hospitalization or relapse. Data describing treatments used in the ED and clinical response in the ED also were generally predictive of hospital admission or later relapse. However, despite demonstrating statistically significant associations with outcome, prediction rules were either not sufficiently accurate to be clinically useful or performed poorly on attempts at validation.

Inpatient Studies

Studies describing prognosis in patients hospitalized for acute exacerbation focused on predicting mortality, the need for MV, and LOS; however, there were few data on LOS.