A comparison of administrative versus clinical data: coronary artery bypass surgery as an example. Ischemic Heart Disease Patient Outcomes Research Team.

Health services researchers rely heavily on administrative data bases, but incomplete or incorrect coding may bias risk models based on administrative data. The best method for validating administrative data is to collect detailed information about the same cases from independent sources, but this approach may be too costly or technically difficult. We used data on coronary artery bypass surgery from four sites (Duke University; Minneapolis--St Paul; California; and Manitoba) to demonstrate an alternative approach for assessing diagnostic coding and to explore the implications of miscoding. The first two sites have clinical data; the second two have administrative data. The prevalences of 14 comorbidities and the associated risk ratios for short-term mortality were compared across data sets. Some comorbidities could not be precisely mapped to ICD-9-CM. Chronic or asymptomatic conditions such as mitral insufficiency, cardiomegaly, previous myocardial infarction, tobacco use, and hyperlipidemia were far less prevalent in administrative data than in clinical data. The prevalence of diabetes, unstable angina, and congestive heart failure were similar in administrative and clinical data. Estimates of relative risk derived from clinical data equalled or surpassed those derived from administrative data for all conditions. Hospitals should be encouraged to improve reporting of coexisting conditions on discharge abstracts and claims. In the meantime, researchers using administrative data should assess the vulnerability of their risk models to bias caused by selective underreporting.

Modifying managed competition to address cost and quality.

California's health care marketplace is undergoing rapid transformation--from one characterized by aggressive but unregulated competition to one that approximates managed competition. The Pacific Business Group on Health purchases health insurance for large employers; the California Public Employees Retirement System acts as a purchasing pool for both state workers and smaller public employers; and The Health Insurance Plan of California is a pool for small employers. Each focuses on price-conscious competition among health maintenance organizations. These large purchasing cooperatives also are addressing biased selection among the health plans they offer. Initially focused on simple demographic adjustments, risk adjustment now also entails adjustment for high-risk cases, such as acquired immunodeficiency syndrome (AIDS) and liver transplants. These purchasers also devote more detailed attention to the quality and process of care.

Does managed care lead to better or worse quality of care?

We analyzed evidence on managed care plan (mostly health maintenance organization, or HMO) performance from thirty-seven recently published peer-reviewed studies. Quality-of-care evidence from fifteen studies showed an equal number of significantly better and worse HMO results, compared with non-HMO plans. However, in several instances, Medicare HMO enrollees with chronic conditions showed worse quality of care. Evidence comparing hospital and physician resource use showed no clear pattern, whereas evidence on enrollee satisfaction varied by measure and enrollee type. Although recent research provides useful findings, interpreting and generalizing from these relatively few studies is difficult. Fears that HMOs uniformly lead to worse quality of care are not supported by the evidence, although all quality data were collected prior to the recent round of cost cutting that started in 1992. Hopes that HMOs would improve overall quality also are not supported, in part because of slow clinical practice change, lack of risk-adjusted capitation rates, and inadequate quality measurement and reporting.

The impact of hospital market structure on patient volume, average length of stay, and the cost of care.

A variety of recent proposals rely heavily on market forces as a means of controlling hospital cost inflation. Sceptics argue, however, that increased competition might lead to cost-increasing acquisitions of specialized clinical services and other forms of non-price competition as means of attracting physicians and patients. Using data from hospitals in 1972 we analyzed the impact of market structure on average hospital costs, measured in terms of both cost per patient and cost per patient day. Under the retrospective reimbursement system in place at the time, hospitals in more competitive environments exhibited significantly higher costs of production than did those in less competitive environments.

The sensitivity of conditional choice models for hospital care to estimation technique.

It is plausible that distance, quality, and hospital charges all influence which hospital patients (and their referring physicians) choose. Several researchers have estimated conditional choice models that explicitly incorporate the existence of competing hospitals. To be useful for hospital administrators, health planners and insurers, however, estimates must be made for specific types of patients and include entire market areas. Data sets meeting these requirements have many combinations of hospitals and locations with zero patients. This raises computational difficulties with the linear estimation techniques used previously. In this paper, we use data on patients undergoing cardiac catheterization in several market areas to assess alternative estimation techniques. First, we estimate the conditional choice model with the two techniques used previously to transform the non-linear choice model. These involve using as a reference (1) a single hospital, or (2) the geometric mean of all the hospitals in the market. When there are many zeros, these techniques require extensive adjustments to the data which may lead to biased estimators. We then compare these results with maximum likelihood estimates. The latter results are substantively and significantly different from those using traditional techniques. More importantly, the linear estimates are much more sensitive to the proportion of zeros. We thus conclude that maximum likelihood estimates are preferable when there are many zeros.

Willingness to pay for poison control centers.

We used the willingness-to-pay (WTP) method to value the benefits of poison control centers when direct access was blocked, comparing WTP among: (1) blocked callers (n = 396), (2) callers after access was restored (n = 418), and (3) the general population (n = 119). Mean monthly WTP was $6.70 (blocked callers), $6.11 (non-blocked callers), and $2.55 (general population). Blocked and non-blocked callers had a significantly higher WTP than general population respondents (p < 0.001). We conclude that the WTP method measured benefits that are difficult to quantify; however, WTP surveys need to be carefully conducted to minimize bias. We discuss how this approach could be useful for other health care services.

Correlation of travel time on roads versus straight line distance.

Studies of hospital demand and choice of hospital have used straight line distance from a patient's home to hospitals as a measure of geographic access, but there is the potential for bias if straight line distance does not accurately reflect travel time. Travel times for unimpeded travel between major intersections in upstate New York were compared with distances between these points. The correlation between distance and travel time was 0.987 for all observations and 0.826 for distances less than 15 miles. These very high correlations indicate that straight line distance is a reasonable proxy for travel time in most hospital demand or choice models, especially those with large numbers of hospitals. The authors' outlier analyses show some exceptions, however, so this relationship may not hold for studies focusing on specific hospitals, very small numbers of hospitals, or studies in dense urban areas with high congestion and reliance on surface streets.

Assessing the relationship between quality of care and the characteristics of health care organizations.

In the past two decades, relationships among health plans, medical groups, and providers have grown more complex and the number of clinical management strategies has increased. In this context, determining the independent effect of a particular organizational strategy on quality of care has become more difficult. The authors review some of the issues a researcher must address when studying the relationship between organizational characteristics and quality of care. They offer criteria for selecting a research question, list organizational characteristics that may influence quality, and suggest sampling and study design techniques to reduce confounding. Since this type of research often requires a health care organization as collaborator, the authors discuss strategies for developing research partnerships and collecting data from the partner organization. Finally, they offer suggestions for translating research into policy.

Health services research as a scientific process: the metamorphosis of an empirical research project from grant proposal to final report.

The process of health services research is rarely examined; attention is usually focused on results and policy implications. Large and small decisions made during the execution of a study, however, can have major impacts on its outcomes. This article describes a project that underwent major changes because of problems discovered in the basic data and threats to the valid interpretation of econometric results uncovered by qualitative case studies. Although the combination of difficulties encountered in this project may be unusual, it is likely that many similar problems and opportunities occur in other empirical studies.

Calculating the probability of rare events: why settle for an approximation?

OBJECTIVE: Health services researchers often need to compute the probability of observing a certain number of events when only a few such events are expected. Our objective is to show that the standard approaches (Poisson, binomial, and normal approximations) are inappropriate in such instances, and to suggest an alternative. DATA SOURCES: Patients undergoing cholecystectomy (34,234) in 465 California hospitals in 1983 are used to demonstrate the biases arising from various methods of calculating the probability of observing a given number of deaths in each hospital. Similar data from other procedures and diagnoses with lower and higher mortality rates are also used for illustration. STUDY DESIGN: The computational methods to derive probabilities using the Poisson, normal, simulation, and exact probabilities are discussed. Using a previously developed risk factor model, the probability of observing the actual number of deaths (or more) is calculated given the expectation of death for each patient in each hospital. Results for the four methods are compared, showing the types of random and systematic errors in the Poisson, normal, and simulation approaches. DATA COLLECTION: Routinely collected hospital discharge abstract data were provided by the California Office of Statewide Planning and Development. PRINCIPAL FINDINGS: The Poisson and normal approximations are often biased substantially in calculating upper-tail p-values, especially when the expected number of adverse outcomes is less than five. Simulations allow unbiased calculations, and the degree of random error can be made arbitrarily small given enough trials. Exact calculations using a simple recursive algorithm can be done very efficiently on either a mainframe or personal computer. For example, the whole set of cholecystectomy patients can be assessed in less than 90 seconds on a Macintosh. CONCLUSIONS: Calculating the probability of observing a small number of events using standard approaches may result in substantial errors. The availability of a simple and inexpensive method of calculating these probabilities exactly can avoid these errors.

Regionalization of services within a multihospital health maintenance organization.

Among the many factors that may explain lower costs for enrollees in Health Maintenance Organizations (HMOs) is the possibility that the HMO provides inpatient services more efficiently. While direct cost comparisons are in appropriate, it is reasonable to examine whether the Kaiser program in the San Francisco Bay Area regionalizes services among its ten hospitals. The presence of each of 43 facilities/services reported is examined in a regression model that includes type of hospital, size, a size-type interaction, and the distance to the nearest competing facility. When the generally smaller size of the Kaiser hospitals was controlled for, Kaiser hospitals had fewer technologically based services and concentrated these services in larger hospitals. Kaiser had more outpatient-oriented services. Among non-Kaiser hospitals, some specialized facilities were competitively distributed.

Measuring hospital mortality rates: are 30-day data enough? Ischemic Heart Disease Patient Outcomes Research Team.

OBJECTIVE: We compare 30-day and 180-day postadmission hospital mortality rates for all Medicare patients and those in three categories of cardiac care: coronary artery bypass graft surgery, acute myocardial infarction, and congestive heart failure. DATA SOURCES/COLLECTION: Health Care Financing Administration (HCFA) hospital mortality data for FY 1989. STUDY DESIGN: Using hospital level public use files of actual and predicted mortality at 30 and 180 days, we constructed residual mortality measures for each hospital. We ranked hospitals and used receiver operating characteristic (ROC) curves to compare 0-30, 31-180, and 0-180-day postadmission mortality. PRINCIPAL FINDINGS: For the admissions we studied, we found a broad range of hospital performance when we ranked hospitals using the 30-day data; some hospitals had much lower than predicted 30-day mortality rates, while others had much higher than predicted mortality rates. Data from the time period 31-180 days postadmission yield results that corroborate the 0-30 day postadmission data. Moreover, we found evidence that hospital performance on one condition is related to performance on the other conditions, but that the correlation is much weaker in the 31-180-day interval than in the 0-30-day period. Using ROC curves, we found that the 30-day data discriminated the top and bottom fifths of the 180-day data extremely well, especially for AMI outcomes. CONCLUSIONS: Using data on cumulative hospital mortality from 180 days postadmission does not yield a different perspective from using data from 30 days postadmission for the conditions we studied.

Appropriate measures of hospital market areas.

As public and private policymakers turn to market-oriented strategies to control hospital prices, it is necessary to understand the conceptual underpinnings of hospital market area measurement. This article provides a framework for evaluating which definitions of hospital market areas are suitable for various types of analyses. Hospital market areas can be defined from two perspectives: an individual hospital perspective and that of the overall market. From each perspective, empirical definitions can be based on geopolitical boundaries, distance between hospitals, and patient-origin data. In this article, market areas are compared based on various descriptions using data on California hospitals and patient discharge abstracts.

The volume-outcome relationship: practice-makes-perfect or selective-referral patterns?

Various studies have demonstrated that hospitals with larger numbers of patients with a specific diagnosis or procedure have lower mortality rates. In some instances, these results have been interpreted to mean that physicians and hospital personnel with more of these patients develop greater skills and that this results in better outcomes--the "practice-makes-perfect" hypothesis. An alternative explanation is that physicians and hospitals with better outcomes attract more patients--the "selective-referral pattern" hypothesis. Using data for 17 categories of patients from a sample of over 900 hospitals, we examine the patterns of selected variables with respect to hospital volume. To explore the plausibility of each hypothesis, a simultaneous-equation model is also used to test the relative importance of the two explanations for each diagnosis or procedure. The results suggest that both explanations are valid, and that the relative importance of the practice or referral explanation varies by diagnosis or procedure, in ways consistent with clinical aspects of the various patient categories.

Choice of hospital for delivery: a comparison of high-risk and low-risk women.

OBJECTIVE: This article tests whether or not the factors that affect hospital choice differ for selected subgroups of the population. DATA SOURCES: 1985 California Office of Statewide Health Planning and Development (OSHPD) discharge abstracts and hospital financial data were used. STUDY DESIGN: Models for hospital choice were estimated using McFadden's conditional logit model. Separate models were estimated for high-risk and low-risk patients, and for high-risk and low-risk women covered either by private insurance or by California Medicaid. The model included independent variables to control for quality, price, ownership, and distance to the hospital. DATA EXTRACTION: Data covered all maternal deliveries in the San Francisco Bay Area in 1985 (N = 61,436). ICD-9 codes were used to classify patients as high-risk or low-risk. The expected payment code on the discharge abstract was used to identify insurance status. PRINCIPAL FINDINGS: The results strongly reject the hypothesis that high-risk and low-risk women have the same choice process. Hospital quality tended to be more important for high-risk than low-risk women. These results also reject the hypothesis that factors influencing choice of hospital are the same for women covered by private insurance as for those covered by Medicaid. Further, high-risk women covered by Medicaid were less likely than high-risk women covered by private insurance to deliver in hospitals with newborn intensive care units. CONCLUSIONS: The results show that the choice factors vary across several broadly defined subgroups of patients with a specific condition. Thus, estimates aggregating all patients may be misleading. Specifically, such estimates will understate actual patient response to quality of care indicators, since patient sensitivity to quality of care varies with the patients' risk status.

Duplicate health insurance coverage: determinants of variation across states.

Although it is recognized that many people have duplicate private health insurance coverage, either through separate purchase or as health benefits in multi-earner families, there has been little analysis of the factors determining duplicate coverage rates. A new data source, the Survey of Income and Education, offers a comparison with the only previous source of state level data, the estimates from the Health Insurance Association of America. The R2 between the two sets is only .3 and certain problems can be traced to the methodology underlying the HIAA figures. Using figures for gross and net coverage, the ratio of total policies to people with private coverage ranges from .94 in Utah to 1.53 in Illinois. Measures of industry distribution, per capita income and employment explain a large portion of the variance, but it appears that these factors operate in opposite directions for group and non-group policies. Similar sociodemographic variables also explain net coverage. These findings have substantial implications for research and the structuring of employee health benefits.

Health maintenance organization environments in the 1980s and beyond.

Throughout the past decade, health maintenance organizations (HMOs) were buffeted by dramatic regulatory and competitive changes. In this article, literature of the 1980s is reviewed to update our knowledge on the HMO industry and to suggest future research. The influence of intensified competition on these organizations and the determinants of market entry, expansion, and exit are examined. These organizations are now beginning to require copayments and deductibles and to offer point-of-service choice, while indemnity plans are developing sophisticated utilization management techniques. Given these significant structural changes, past distinctions among HMO, preferred provider organization and fee-for-service medicine must be replaced with a distinction between degree of provider choice and level of benefits.

Border crossing for hospital care and its implications for the use of statewide data.

A major concern of researchers using state data sets for population-based analyses and market share studies in the health care sector is the potential bias caused by 'border crossing'--patients receiving care out of state. By using the Health Care Financing Administration (HCFA) discharge abstract files for 1987 and 1988, we found that 'border crossing' is not a serious problem for the two large states we examined. Only 4.4% of New York patients and 2.15% of California patients received care out of state. At the county and zip code level, 'border crossing' is more frequent but tends to be concentrated in areas adjacent to other states. Even excluding all zips with more than 10% of patients crossing the 'border' results in a small loss of patients (2.2% for New York and 1.0% for California).

Estimating hospital admission patterns using Medicare data.

It is often necessary in health services research and strategic planning to simultaneously describe the geographic pattern of admissions to multiple hospitals. Obtaining the data necessary to accomplish this can often be problematic. In some states discharge abstract data on all admissions to all hospitals in the state are compiled and maintained by a government agency, but in 23 states these data are not available. Furthermore, problems arise when a substantial fraction of admissions cross state borders, such that data from more than one state is required for description of 'patient flows'. Individual hospitals typically maintain data on the geographic source of their own admissions, but are not likely to have access to such data regarding other hospitals in their area. Patient flow data on Medicare admissions are available for all states and are readily accessible, but heretofore it has not been known how closely the admission patterns of Medicare patients approximate those of other types of patients. We examine the accuracy of using data on Medicare admissions to estimate, at the hospital level, the admission patterns of other types of patients. Using zip code-to-hospital patient flow data for all non-federal hospitals in California, we calculated the correlation between Medicare admission patterns and those of three other groups of patients (other adults, pediatrics and obstetrics) for each hospital. For the majority of hospitals, Medicare data predict the admissions of other adults quite well, and the admissions of pediatric and obstetric admissions moderately well.(ABSTRACT TRUNCATED AT 250 WORDS)

Should surgery be regionalized?

The authors suggest that new surgical procedures be carried out initially in selected institutions and that complex procedures for which it has been or can be demonstrated that mortality is inversely related to the volume of experience also be regionalized. Regionalization in the latter instance can have a small overall impact on surgical practice but a large impact on the adverse consequences of high risk operations that are performed only occasionally.