Research Briefs

Berlin, J.A., Santanna, J., Schmid, C.H., and others (2002). "Individual patient- versus group-level data meta-regressions for the investigation of treatment effect modifiers: Ecological bias rears its ugly head." (AHRQ grant HS10064). Statistics in Medicine 21, pp. 371-387.

Most meta-analyses of multiple studies are undertaken with published, group-level data. These investigators explored a real-world example (benefits of anti-lymphocyte antibody induction therapy among renal transplant patients) for which both group-level and individual patient-level data were available and compared the conclusions reached through both methods. They focused on whether there were subgroups of patients in whom therapy might prove particularly beneficial. The endpoint studied was allograft failure within 5 years. The patient-level analysis revealed that treatment was significantly more effective among patients with elevated (20 percent or more) panel reactive antibodies. These patients comprised a small subgroup of patients (15 percent) who benefited from therapy. The group-level analysis failed to detect this interaction. The researchers suggest the use of individual patient data in meta-analyses to avoid the potential for ecological bias introduced by group-level analysis.

Hollingworth, W., Deyo, R.A., Sullivan, S.D., and others. (2002). "The practicality and validity of directly elicited and SF-36 derived health state preferences in patients with low back pain." (AHRQ grant HS09499). Health Economics 11, pp. 71-85.

Many clinical trials now incorporate more than one measure of health-related quality of life. Some of these measures are: disease-specific instruments; generic health profiles (for example, the SF-36); preference-based indexes, for example, the standard gamble (SG) and time trade-off (TTO); and the non-choice-based visual analogue scale (VAS). These authors compared the practicality and validity of SF-36 derived preference scores with directly elicited TTO and VAS scores in a group of patients with low back pain. Choice-based methods (SG and TTO) yielded higher and more uniform estimates of preference than non-choice methods. Directly elicited TTO values were variable and had less power to distinguish among patients with differing severity of low back pain. SF-36 derived preferences demonstrated good practicality and construct validity in this setting. However, different methods will yield disparate estimates of marginal benefit. A standardized algorithm for deriving SF-36 preference scores is needed.

Meenan, R.T., Goodman, M.J, Fishman, P.A., and others. (2002, January). "Issues in pooling administrative data for economic evaluation." (AHRQ National Research Service Award training grant HS00069 and fellowship F32 HS00072). American Journal of Managed Care 8(1), pp. 45-53.

Economic evaluations, which increasingly are being tailored to the perceived needs of health maintenance organizations (HMOs), are steadily gaining acceptance by policymakers responsible for health care resource allocation. This encourages use of HMO administrative data as an efficient source of resource use and cost measures. The best alternative to a nationally representative data set is to pool administrative data from multiple sites within one database. However, pooling administrative data is problematic because HMO data sources reflect differences in systems of care, costs, and coding. These authors describe issues inherent in the pooling of HMO administrative cost data for use in multisite economic evaluations. They describe the attributes of administrative data that are relevant to costing, discuss their implications for multisite economic evaluations, and offer suggestions for researchers working with such data.

Mukamel, D.P., Dick, A., and Spector, W.D. (2001). "Specification issues in measurement of quality of medical care using risk adjusted outcomes." Journal of Economic and Social Measurement 26, pp. 267-281.

Governments and private organizations recently have begun publishing "report cards" that compare quality of hospitals, physicians, and health care plans. These report cards often include quality measures based on risk adjusted health outcomes of the patients treated by each health care provider. Until now, concerns about the accuracy of such measures have focused on their risk adjustment methodology and small sample sizes. These investigators raise a third issue related to the definition of quality measures as either the difference between observed and predicted outcome rates or the ratio between these rates. They present a theoretical analysis of the properties of the two measures. The researchers show that the two risk-adjusted outcome measures of quality may lead to different conclusions about relative quality among providers. They conclude that the choice of measure to be used depends on the underlying relationship between patient risks and quality of care in determining health outcomes.

Reprints (AHRQ Publication No. 02-R053) are available from the AHRQ Publications Clearinghouse.

Mukamel, D.B., and Spector, W.D. (2002). "The competitive nature of the nursing home industry: Price mark ups and demand elasticities." Applied Economics 34, pp. 413-420.

The demand for nursing home care is due primarily to Medicaid and private pay residents, who accounted for over 95 percent of all nursing home residents in New York State in 1991. The Medicaid demand is perfectly price elastic because the Medicaid payment rate is set by State Medicaid programs and is the same for all residents. On the other hand, private pay demand has been assumed in previous studies to be downward slowing, reflecting competitive market structure. Nursing home markets are likely to deviate from a competitive structure because of limitations on nursing home entry imposed by Certificate of Need (CON) regulations and the potential for product differentiation along such attributes as location, religious affiliation, and quality. These authors examine the structure of nursing home markets in New York by calculating price mark ups and residual private pay demand elasticities, an approach that allows estimation of demand elasticities in all markets whether or not CON regulations constrain bed supply. They show that the residual demand elasticity is bound by estimates based on price mark ups above marginal costs and above Medicaid rates.

Reprints (AHRQ Publication No. 02-R052) are available from the AHRQ Publications Clearinghouse.

Normand, S.T., and Zhou, K.H. (2002). "Sample size considerations in observational health care quality studies." (AHRQ grant HS09487). Statistics in Medicine 21, pp. 331-345.

Unlike cluster randomization trials, where clusters often are randomized to interventions to learn about individuals, the target of inference in health care quality studies is the cluster. These authors discuss approaches to sample size determination to compare providers when designing observational studies of health care quality. They focus on process-based measures because this approach has been widely adopted by many regulatory agencies and health plans. They use data from a study designed to develop and test a set of outpatient quality measures across a continuum of care sites, payment systems, and data sources for patients with cardiovascular disease. Drawing from experience gained from this study, they briefly review methods for calculating sample size using marginal models, but the focus is on hierarchical binomial models. The researchers conclude that investigators interested in comparing clusters should use hierarchical models.

Zhan, C., Sangl, J., Meyer, G.S., and Zaslavsky, A.M. (2002). "Consumer assessments of care for children and adults in health plans. How do they compare?" Medical Care 40(2), pp. 145-154.

The Consumer Assessment of Health Plans Study (CAHPS®) surveys include an adult version and a child version for parents or caretakers to rate children's care in health plans. This study examined how adult and child assessments differed in ranking health plans. The researchers used data from 136 commercial health plans participating in the National CAHPS® Benchmarking Database, which included 80,539 adults and 40,003 children. They compared mean assessments for adults and children on four global ratings and five composites and determined respondent characteristics predictive of these assessments. CAHPS® scores for children were significantly higher than those for adults, except for customer service (lower for children) and specialist ratings. There was fair to moderate agreement between adult and child mean scores in ranking health plans. Since adult and child CAHPS® surveys provide similar scores and plan rankings on many aspects of care, consumers who are concerned with quality of care for children may to some degree rely on results of the adult survey.

Reprints (AHRQ Publication No. 02-R047) are available from the AHRQ Publications Clearinghouse.

Return to Contents

Current as of April 2002
AHRQ Publication No. 02-0021