What Happens After a Patient Safety Event? Medical Expenditures and Outcomes in Medicare

Introduction

Background

Several studies have estimated the cost of injuries related to medical interventions.^2–4 All of these studies rely on professional judgment to identify preventable adverse medical events and to estimate the cost of this event by summing the cost of a hospital day, an outpatient visit, and a physician office visit attributed to the event. Yet, because of the time and cost involved in reviewing medical records, studies based on the review of hospital medical records involve a relatively small number of cases.

In particular, Johnson and colleagues² estimated the cost of medical injuries from the New York Medical Malpractice study, which is based on interviews with 794 individuals who had suffered medical adverse events in 1984; Thomas and colleagues³ estimated the cost of medical errors that occurred in 459 cases in Utah and Colorado hospitals in 1992; and Bates and colleagues⁴ estimated the cost of adverse drug events using 247 adverse drug events in 207 hospitalized patients in 2 hospitals in Boston over a 7-month time period in 1993.

However, two other studies^{5, 6} have used large administrative databases to analyze patient safety events. The study by Kalish and colleagues used algorithms to screen 372,680 California hospital discharge abstracts in an effort to identify cases where a medical intervention had resulted in an unintended consequence and where this consequence was determined to have been potentially preventable. Zhan and Miller used the AHRQ Patient Safety Indicators to identify patient safety events among 7.45 million discharge abstracts nationwide. Both of these studies used hospital charge data to estimate the cost of the unintended consequence. However, these studies omitted the cost of services that were not billed by the hospital (e.g., physician services for anesthesia and surgery, and charges for laboratory and imaging services provided by outside facilities) and omitted all subsequent outpatient costs and drug costs.

Methods

Our primary source of data was the 2000 MarketScan Medicare Supplement and Coordination of Benefits Database. This database was created by the Medstat Group, Inc., and contains claims data for inpatient care, outpatient care, and prescription drugs for employees, dependants, and retirees over age 65 years in employer-sponsored retiree benefit plans for 41 large employers in all 50 States. The data includes all employer and Medicare coordination of benefits for these enrollees with both employer coverage and Medicare coverage.

We included all medical claims incurred within 90 days after the surgery admission date. We chose a 90-day period since Brennan et al. reported that 50 percent to 70 percent of patients with adverse events recovered within 90 days.⁷ The unit of observation was any major surgery admission (identified as the “index” surgery) that occurred between July 1, 1999, and October 1, 2000, that did not follow another major surgery admission within the previous 90 days for that patient.

We had a total of 22,477 observations. The surgeries occurred at 1,725 hospitals nationwide. Hospital characteristics were obtained from the American Hospital Association's Annual Survey 1999-2000. County characteristics were obtained from the Area Resource File (Bureau of Health Professions, Health Resources and Services Administration).

Statistical methods

Our main patient safety variable was modeled as a binary variable that codes a major surgery hospitalization as “1” if at least 1 of the 14 patient safety events occurred during that hospitalization, and codes the surgery hospitalization as “0” if none of the 14 patient safety events occurred. With this model, we then examined (1) expenditures, and (2) patient outcomes associated with potentially preventable adverse medical events. While there may be variation in expenditures and outcomes between each of the 14 PSIs, we did not have a large enough sample to examine each indicator individually. Thus, we pooled all 14 indicators and looked at their average expenditures and outcomes compared to surgeries without any patient safety event.

Expenditures

Linear regression analysis at the discharge level was used to examine the relationship between the occurrence of a potentially preventable adverse medical event during a surgery and the natural logarithm of the total medical claims cost of treating that patient for 90 days after the admission. The natural logarithm of the expenditures was used instead of the absolute level of expenditures because the distribution of the absolute level of expenditures was skewed. We ran five separate regressions, one on each category of costs: (1) total 90-day expenditures, (2) 90-day inpatient hospital expenditures, (3) 90-day inpatient physician expenditures, (4) 90-day outpatient expenditures, and (5) 90-day drug expenditures. Patients who die in the hospital have high inpatient expenditures but no outpatient or drug expenditures. Since the death rate is much higher for patients experiencing potentially preventable adverse medical events, we chose to conduct the 90-day outpatient and drug regressions with the sample restricted to those who did not die during the 90 days.

Table 1

Sample means of covariants for patient (more...)

Table 1

Sample means of covariants for patient safety events ^a

Variables	PSI	No PSI
Emergency admission	0.051	0.082
Female	0.499	0.539
Age	74.819	76.266
Union	0.253	0.286
Fee-for-service plan	0.748	0.726
Preferred provider organization	0.231	0.250
Point-of-service plan	0.006	0.006
Capitated plan	0.015	0.018
Area characteristics (county)
HMO penetration	0.247	0.273
Media income	$37,609	$38,683
Hospital characteristics
Small bedsize hospital	0.076	0.088
Medium bedsize hospital	0.249	0.249
Large bedsize hospital	0.675	0.663
Teaching hospital	0.508	0.512
Public hospital	0.079	0.062
Not-for-profit hospital	0.857	0.895
For-profit hospital	0.064	0.043
Urban hospital	0.868	0.911
Patient chronic conditions
Congestive heart failure	0.038	0.171
Arrhythmias	0.077	0.217
Valvular disease	0.027	0.091
Pulmonary circulation disease	0.003	0.008
Peripheral vascular disease	0.026	0.062
Hypertension	0.151	0.179
Paralysis	0.005	0.016
Other neurological disorders	0.019	0.061
Chronic pulmonary disease	0.064	0.132
Diabetes	0.052	0.076
Diabetes with chronic complication	0.009	0.133
Hypothyroidism	0.012	0.013
Renal failure	0.020	0.057
Liver disease	0.003	0.003
Peptic ulcer disease × bleeding	0.003	0.008
Aids	0.001	0.001
Lymphoma	0.004	0.007
Metastatic cancer	0.019	0.026
Solid tumor w/out metastasis	0.068	0.085
Rheumatoid arthritis coolagen vas	0.007	0.007
Coagulopathy	0.007	0.020
Obesity	0.001	0.001
Chronic blood loss anemia	0.004	0.009
Deficiency anemias	0.023	0.043
Alcohol abuse	0.002	0.001
Drug abuse	0.001	0.001
Psychoses	0.004	0.012
Depression	0.002	0.006
Sample size	21,574	903

a

PSI refers to a patient safety event occurring.

Since we wanted to estimate the portion of the 90-day expenditures that were due to the potentially preventable adverse medical event and not due to patient, hospital, health plan, or market characteristics, we included the covariates from Table 1 in each of the five expenditure regressions. We included the following to control for potential confounding effects of patient severity: age, age squared, gender, an indicator for emergency room admission, and indicators for 30 chronic conditions developed by Elixhauser et al.¹¹ and updated by the Stanford EPC in the AHRQ PSI software.⁹

Next, to control for demand-side factors that may influence the patient's degree of use, we controlled for the type of the health plan, union status, and the median household income for the patient's county. Since we pooled data from 2 years, we also included indicators for each year to control for any time trend.

To control for market characteristics, we included the 1998 county health maintenance organization (HMO) penetration rate. Hospital characteristics include teaching hospitals, rural hospitals, hospital ownership, and hospital bed size. The hospital bed size (small, medium, large) is defined in the Healthcare Cost and Utilization Project (HCUP) National Inpatient Sample documentation (see http://www.hcup-us.ahrq.gov/).

Finally, not shown in Table 1, we controlled for the type of surgery by including indicator variables for 221 Diagnosis-Related Groups (DRGs). Since the five expenditure regressions used the log of expenditures, we did not report the estimated coefficients. Instead, we used the regression results to simulate the non-logged expenditures that were due to potentially preventable adverse medical events. Since the disturbances were heteroscedastic, we chose to use the log-link Generalized Linear Method (GLM) of regression with a gamma distribution family in Stata 7.0. In all our GLM regressions, the robust standard errors were estimated using the Huber/White sandwich estimator.

Next, we used a simulation method that is similar to that found in Escarce et al.¹² Using each of the five GLM expenditure regressions, we first assumed everyone in the sample had a potentially preventable adverse medical event and predicted their expenditures. We then assumed no one in the sample had a potentially preventable adverse medical event and predicted their expenditures. The difference in the predictions would be the predicted expenditures due to patient safety events. The standard errors of the predicted difference in expenditures due to a potentially preventable adverse medical event were computed using the Delta Method.^{12, 13}

Finally, since long-term care was rarely used by patients in the sample, we predicted long-term care use in two stages: first, the probability of the use of long-term care following a potentially preventable adverse medical event using a logistic regression; second, using a log-link gamma GLM regression on those patients with positive long-term care expenditures during the 90 days to predict long-term care expenditures due to potentially preventable adverse medical events.

Outcomes

Using a logistic multivariate regression at the discharge level, we estimated the probability of dying within 90 days after a surgery in which a potentially preventable adverse medical event occurred. Similarly, we estimated the probability of being readmitted within 90 days after a hospitalization in which a potentially preventable adverse medical event occurred. While the first (index) admission had to be a major surgery, the readmission did not have to be a surgery. Finally, using logistic analysis, we estimated the probability of using long-term care within 90 days after a hospitalization in which a potentially preventable adverse medical event occurred. In all logistic regressions, the robust standard errors were estimated using the Huber/White sandwich estimator and were corrected for clustering at the hospital.

Table 4

Predicted outcomes for following a patient safety (more...)

Table 4

Predicted outcomes for following a patient safety event, adjusted for covariates ^a

	90 Day Death Rate	Readmission Rate for Survivors	Long Term Care Use
PSI	6.15%	10.85%	9.21%
No PSI	1.62%	8.37%	5.61%
Difference	4.53%**	2.48%*	3.60%**
	(1.41)	(1.11)	(1.10)
Percentage Difference	279.63%	29.63%	64.17%

a

These are predicted outcomes in the 90 days following the admission date of a major surgery in 1999 and 2000 for Medicare enrollees with secondary employer coverage, adjusting for the covariates in Table 1 and for MDCs. PSI refers to a patient safety event occurring. Standard errors are in parentheses.

**

Significantly different from zero at the 99% level.

*

Significantly different from zero at the 95% level.

All three outcome logistic regressions used the same covariates found in Table 1, and also controlled for 23 Major Diagnostic Categories (MDCs). All three outcome logistic regressions were then used to predict the probability of each of the three outcomes under a patient safety event. The predicted probabilities of each outcome are reported in Table 4.

Results

Table 2

The financial burden of a patient safety event, (more...)

Table 2

The financial burden of a patient safety event, adjusted for covariates ^a

	Observations	Total Payment
No PSI	21,574	$17,319
PSI	903 (PSI Rate: 4.02%)	$24,317

a

These are payments for medical care in the 90 days following a hospital admission in 1999 and 2000 for Medicare enrollees with secondary employer coverage, adjusting for the covariates in Table 1 and for DRGs. PSI refers to a patient safety event occurring.

Table 3

Predicted medical expenditure for patient safety (more...)

Table 3

Predicted medical expenditure for patient safety events, adjusted for covariates ^a

	Inpatient Hospital Payments	Inpatient Physician Payments	Outpatient Payments	Drug Payments	Total Payments	Portion Due To Long Term Care
PSI	$19,455	$2,104	$2,746	$551	$24,317	$631
No PSI	$12,775	$1,744	$2,458	$570	$17,319	$486
Payment Difference	$6,680** (1,367)	$360** (119)	$288* (134)	-$19 (24)	$6,998** (1,269)	$145** (68)
Percentage Difference	52.3%	20.6%	10.5%	-3.3%	40.4%	29.8%

a

These are predicted payments for medical care in the 90 days following a major surgery admission in 1999 and 2000 for Medicare enrollees with secondary employer coverage, adjusting for the covariates in Table 1 and for DRGs. PSI refers to a patient safety event occurring. Outpatient and drug payment means are conditional upon not dying. However, total payment is not conditional upon not dying, so the sum of the subpayments across each row is greater than the total payment in the final column. Standard errors are in parentheses.

**

Significantly different from zero at the 99% level.

*

Significantly different from zero at the 95% level.

Discussion

The results on the costs of medical errors presented in the Institute of Medicine's report, To Err Is Human, were based on the small sample, State studies of New York^{2, 7, 14} and Colorado-Utah.^{3, 15, 16} These samples were too small to derive cost estimates of patient safety for Medicare patients. To address this problem, in this paper we examined a national sample of elderly Medicare patients with secondary employer coverage.

First, our results provide insight into the composition of medical expenditures 90 days after a potentially preventable adverse medical event. The bulk of the extra expenditures over the 90 days, due to potentially preventable adverse medical events, are inpatient hospital payments. About 5 percent of the 90-day expenditures due to adverse events are physician inpatient payments. On average, about 4 percent of the 90-day expenditures due to adverse events are outpatient payments. Even though all the enrollees had drug coverage from the employer, patient safety events did not result in any extra outpatient drug payments.

This is in contrast to the Utah-Colorado study, where malpractice claims adjustors proposed that 46 percent of the extra lifetime medical costs due to patient safety problems would be outpatient costs.³ This large outpatient composition of expenditures in the Utah-Colorado study is due to the fact that their malpractice claims adjustors attributed 37 percent of the lifetime expenditures on patient safety care to nursing home care. In contrast, we find that only 2.1 percent of the 90-days expenditures due to potentially preventable adverse medical events are attributable to long-term care.

Next, our results provide insight into the magnitude of the 90-day medical expenditures attributable to potentially preventable adverse medical events. We find that patient safety events are responsible for $6,998 in 90-day medical expenditures spent by patients experiencing adverse events. The strength of our study is that we used actual insurance claims (transacted payments) to calculate the costs of a potentially preventable adverse medical event. No other expenditure study uses payments. For example, the studies by Kalish et al.⁵ and Zhan and Miller⁶ use hospital charges instead of payment claims, overestimating what was actually paid. Kalish et al. found that major surgeries with complications (a broader category than patient safety event) in 1988 had inpatient charges of $16,023 ($22,188 in 2000 dollars) attributable to the complication.

A second study, the Harvard Medical Practice Study, interviewed 794 patients in 51 hospitals in New York who had been injured during hospitalization in 1984 to ascertain their outpatient medical care use between the hospitalization in 1984 and July 1988.² They roughly estimated that the lifetime medical costs for these injuries (attributable to injuries and not the illness) were $18,305 per person injured in 1989 dollars ($23,367 in 2000 dollars). However, in that study, much of the costs were based on what the patient or family recalled 5 years after the potentially preventable adverse medical event, with the authors assigning prices to the recalled utilization.

A third study by Thomas et al.³ examined only 265 preventable hospital adverse events that occurred in 1992 in 28 hospitals in Utah and Colorado. They then had 10 malpractice insurance claims adjusters provide an expert opinion on what the lifetime costs of these injuries would be. They found that the lifetime medical costs for these injuries were $17,976 per person injured in 1996 dollars ($20,036 in 2000 dollars). However, in that study, costs were based on claims adjustors' expert opinion of what utilization probably would have occurred, given the medical chart of the medical injury.

Recall that these potentially preventable adverse medical events are based only on 14 measurable PSIs. Thus, there may have been many more preventable safety events (as well as close calls) that occurred but that were not included in our analyses, such as medication errors. In fact, we did not consider drug-related errors, diagnostic errors, and errors in choice of therapy, all of which accounted for 12 percent of surgical errors in the Colorado-Utah study.¹⁶ Thus, our expenditure results are an underestimate of all the expenditures attributable to all preventable adverse events.

Nevertheless, these 14 potentially preventable adverse medical events were deemed to be highly preventable by 11 clinical panels, and yet they resulted in excess deaths, readmissions, and expenses for elderly surgery patients with both Medicare and employer coverage. Moreover, our employer data comes from some of the Fortune 500 corporations. Of all Medicare patients in the country, these would be the ones suspected of having top-notch health insurance coverage, with the best choice of doctors and hospitals. Yet, among surgeries alone, these 14 potentially preventable adverse medical events caused 1.1 percent of their readmissions, 9.4 percent of their deaths, and 1.6 percent of their 90-days medical expenditures.

Acknowledgments

This research was funded by the Agency for Healthcare Research and Quality. The views herein do not necessarily reflect the views or policies of AHRQ, nor the U.S. Department of Health and Human Services. We thank our programmers Linda Andrews of Social and Scientific Systems, Inc., and Yafu Zhao of Coda Research, Inc.

Author affiliation

Both authors are affiliated with the Agency for Healthcare Research and Quality, Center for Delivery, Organization and Markets.

Address correspondence to: William E. Encinosa, Ph.D.; Center for Delivery, Organization and Markets; AHRQ; 540 Gaither Road; Rockville, MD 20850; phone: 301-427-1437; fax 301-427-1430; e-mail wencinos@ahrq.gov.

References

1.

Kohn LT, Corrigan JM, Donaldson MS, editors. To err is human: building a safer health system. A report of the Committee on Quality of Health Care in America, Institute of Medicine. Washington, DC: National Academy Press; 2000.

2.

Johnson W G, Brennan T A, Newhouse J P. et al. The economic consequeneces of medical injuries: implications for a no-fault insurance plan. JAMA. 1992 May 13;267(18): 2487–92. [PubMed]

3.

Thomas E, Studdert D M, Newhouse J P. et al. Costs of medical injuries. Inquiry. 1999; 36(3): 255–64. [PubMed]

4.

Bates D W, Spell N, Cullen D J. et al. The costs of adverse drug events in hospitalized patients. JAMA. 1997; 277(4): 307–11. [PubMed]

5.

Kalish R L, Daley J, Duncan C C. et al. Costs of potential complications of care for major surgery patients. American Journal of Medical Quality, Spring. 1995; 10(1): 48–54.

6.

Zhan C, Miller M. Excess length of stay, charges, and mortality attributable to medical injuries during hospitalization. JAMA. 2003; 290(14): 1868–74. [PubMed]

7.

Brennan T, Leape L, Laird N. et al. Incidence of adverse events and negligence in hospitalized patients: results from the Harvard Medical Practice Study I. New England Journal of Medicine. 1991; 321: 480–4. [PubMed]

8.

Weingart S, Iezzoni L. Looking for medical injuries where the light is bright. JAMA. 2003; 290(14): 1917–19. [PubMed]

9.

McDonald K, Romano P, Geppert J, et al. Measures of patient safety based on hospital administrative data—the patient safety indicators. Technical Review 5. AHRQ Publication No. 02-0038. Rockville, MD: Agency for Healthcare Research and Quality; August 2002.

10.

Romano P S, Geppert J J, Davies S. et al. A national profile of patient safety in U.S. hospitals. Health Aff. 2003 Mar–Apr;22(2): 154–66.

11.

Elixhauser A, Steiner C, Harris D. et al. Comorbidity measures for use with administrative data. Med Care. 1998 Jan;36(1): 8–27. [PubMed]

12.

Escarce J, Kapur K, Joyce G F, Van Vorst K A. Medical care expenditures under gatekeeper and point-of-service arrangements. Health Services Research. 2001; 36(6): 1037–57. [PubMed] [Free Full text in PMC]

13.

Manning W. The logged dependent variable, heteroscedasticity, and the retransformation problem. The Journal of Health Economics. 1998; 17: 283–95.

14.

Leape L. et al. The nature of adverse events in hospitalized patients: results from the Harvard Medical Practice Study II. New England Journal of Medicine. 1991; 324(6): 377–84. [PubMed]

15.

Thomas E J, Studdert D M, Burstin H R. et al. Incidence and types of adverse events and negligent care in Utah and Colorado. Med Care. 2000 Mar;38(3): 261–71. [PubMed]

16.

Gawande A A, Thomas E J, Zinner M J. et al. The incidence and nature of surgical adverse events in Colorado and Utah in 1992. Surgery. 1999; 126(1): 66–75. [PubMed]