Research Activities, October 2009
Alexander, C. G. and Stafford, R. S. (2009, June). "Does comparative effectiveness have a comparative edge?" (AHRQ grant HS15659). Journal of the American Medical Association 301(23), pp. 2488-2490.
In this commentary, the authors discuss the factors needed to permit comparative effectiveness research to substantially affect the practice of medicine. They argue there is a need for comparative effectiveness research to generate data more rapidly (before a clinical practice is widely adopted), link the evidence to strategies proven to modify practice, broaden the agenda beyond drugs and devices (to different treatment strategies and systems of care), alter the regulatory environment (particularly the Food and Drug Administration's [FDA's] drug and device approval process), and consider the cost implications of practice alternatives. Barriers to creating and implementing comparative effectiveness findings include lack of comparison of medical management with surgical procedures, the FDA's historic focus on evaluating efficacy of new drugs against placebo rather than standard therapy in making approval decisions, and the impact of direct-to-consumer advertising of new, typically expensive drugs. The primary problem that may lead to failure of the competitive effectiveness initiative is not the absence of comparative knowledge, but of ways to put this knowledge to work, the authors conclude.
Clancy, C. M. (2009, May/June). "Re-engineered hospital discharge process lowers re-admissions, reduces costs." Patient Safety & Quality Healthcare 6(3), pp. 8-9. Reprints (AHRQ Publication No. 09-R069) are available from the AHRQ Publications Clearinghouse.
The author begins by noting that shortened hospital stays in recent years require better planning for and explanation to the patient of posthospital care. She reports that a recent study found that 36 percent of patients discharged from the geriatric or medicine service at a large teaching hospital failed to get the after-discharge workups they needed. In addition, slightly more than half of the discharge summaries failed to document the workups recommended in the patients' hospital charts. Clinicians are seeking ways to close these information gaps. One such effort, Project RED (for Re-engineered Discharge) at Boston Medical Center in Massachusetts, has transformed how patients are discharged from the hospital and seen for followup care. The author summarizes the simple principles at the heart of Project RED, and the 11 steps that the project uses to put these principles into action. A randomized study of standard vs. re-engineered discharge found that patients who participated in the RED program had 30 percent fewer emergency department visits and readmissions than control patients, and were almost three times more likely to leave with followup appointments with their primary care physicians.
Clancy, C. M. and Hughes, R. G. (2009, April). "The stress of the care environment." American Operating Room Nurses Journal 89(4), pp. 751-753. Reprints (AHRQ Publication No. 09-R066) are available from the AHRQ Publications Clearinghouse.
In this commentary, the authors note some of the sources of stress that have led to nurses abandoning their profession for other careers. They point out that such loss of experienced nurses weakens the health care system's ability to provide quality care. Drawing on Patient Safety and Quality: An Evidence-Based Handbook for Nurses, published by the Agency for Healthcare Research and Quality, the authors highlight some key stressors that affect nurses. These include verbal bullying, restrictions on autonomy by physicians (who often see themselves as leaders of the health care team), and exclusion of nurses from the clinical decisionmaking process. Other stressors affecting nurses include the tradition-bound and highly stratified nature of most hospitals, the divergence between nurses' training and their actual work environment, and the demands of the job.
Dhruva, S. S., Phurrough, S. E., Salive, M. E., and others (2009, June). "CMS's landmark decision on CT colonoscopy-Examining the relevant data." New England Journal of Medicine 360(26), pp. 2699-2701. Reprints (AHRQ Publication No. 09-R075) are available from the AHRQ Publications Clearinghouse.
The authors of this commentary discuss the recent decision by the Centers for Medicare & Medicaid Services (CMS) to deny coverage of computed tomographic (CT) colonoscopy for cancer screening. CMS chose to deny coverage because the evidence of effectiveness of this technique was inadequate, especially for the Medicare age group. They note that the clinical trials that showed benefit of screening with this noninvasive technique had a mean age significantly lower than that of Medicare beneficiaries. What's more, there were no studies either evaluating this technology in the elderly or analyzing subgroups of participants older than 65 years of age. The authors reviewed the evidence used to make decisions on other recent payment decisions by CMS and found similar problems regarding age groups, analysis by sex, and representation of racial or ethnic minority groups. They note that groups with a financial stake in specific technologies-specifically, radiologists and equipment manufacturers involved with cardiac CT-have been able to influence CMS to back down on previous denials of coverage by lobbying Congress, often leaving a geographic patchwork of payment. The CT colonoscopy decision was different, in part, because Congress had authorized CMS to consider costs as well as clinical effectiveness, the authors note. They hope that CMS holds to its colonoscopy decision and expands its evidence-based approach.
Elliott, M. N., Morrison, P. A., Fremont, A., and others (2009). "Using the Census Bureau's surname list to improve estimates of race/ethnicity and associated disparities." (AHRQ Contract No. 282-00-0005). Health Services Outcomes Research Methods 9, pp. 69-83.
What's in a name? According to this paper's authors, using a list of relatively common surnames from the 2000 Census can help improve estimates of race or ethnicity in patient data, and therefore give a better estimate of associated disparities in care. The 151,671 surnames listed by at least 100 individuals represent 89.8 percent of all individuals enumerated in the 2000 Census, the authors say. Using the self-reported racial and ethnic affiliations for each surname, together with geographical information, researchers can calculate a set of likelihoods for someone with a specific surname being white, black, Hispanic, etc. For example, someone with the surname Washington is 90 percent likely to be black, while individuals named Yoder or Novak have a 97 chance of being non-Hispanic whites, the authors note. By using Bayesian analysis to adjust the national racial probabilities for a given surname for the person's Census block of residence, the authors get a set of updated probabilities of the individual belonging to a particular race or ethnic group. The authors suggest that use of first name listings may improve prediction estimates, particularly for blacks and some Asian subgroups.
Fongwa, M., Cunningham, W., Weech-Maldonado, R. and others (2008, November). "Reports and ratings of care: Black and white Medicare enrollees." (AHRQ grants HS09204 and HS16980). Journal of Health Care for the Poor and Underserved 19(4), pp. 1136-1147.
Consistent with the Institute of Medicine's observation that racial and ethnic minorities are less likely than whites to receive needed services, a new study finds that blacks insured by Medicare have far worse experiences than their white counterparts with getting care quickly, getting needed care, office staff helpfulness, and health plan customer service. They also rate their specialist care and health plans more negatively. However, elderly blacks reported better provider communication than their white counterparts and rated their personal doctors and nurses more positively. The study used the 2002 Consumer Assessment of Healthcare Providers and Systems survey of 101,189 white and 8,791 black Medicare enrollees randomly sampled from 321 health plans.
Fowler, F. J., Epstein, A., Weingart, S. N., and others (2008, October). "Adverse events during hospitalization: Results of a patient survey." (AHRQ grant HS11928). Joint Commission Journal on Quality and Patient Safety 34(10), pp. 583-590.
After interviewing nearly 2,600 patients discharged from 16 diverse Massachusetts hospitals over a 5-month period in 2003, researchers found that one in four patients experienced a medical error or safety problem. These adverse events rates were substantially higher than the rates published in hospital records. Reactions to drugs newly prescribed at the hospital accounted for 40 percent of the events, while another 34 percent of the adverse events were related to surgery. Ongoing hospital care also presented problems, such as falls (14 percent of all adverse events), rashes caused by bandages (15 percent), and drops in blood pressure (11 percent). Physician reviewers on the research team determined that 31 percent of the reported adverse events were likely preventable.
Goldmakher, G. V., Camargo, E. C. S., Furie, K. L., and others (2009, January). "Hyperdense basilar artery sign on unenhanced CT predicts thrombus and outcome in acute posterior circulation stroke." (AHRQ grant HS11392). Stroke 40(1), pp. 134-139.
Blockage of the basilar artery along the brainstem at the back of the brain is typically a devastating stroke event that can lead to disability and death. Early diagnosis is critical so that clot-busting therapy or other supportive measures can begin. A new study suggests that community hospitals can use unenhanced computerized tomography (CT) to diagnose these dangerous blood clots. The researchers examined unenhanced CT scans obtained within 24 hours of symptom onset in 95 patients with suspected posterior circulation stroke. Three neuroimagers rated the presence of hyperdense basilar artery (HDBA) sign, a strong predictor of a blood clot in the basilar artery, on unenhanced CT scans. HDBA sign had 71 percent sensitivity, 98 percent specificity, 94 percent accuracy, 83 percent positive predictive value, and 95 percent negative predictive values for basilar artery occlusion. The HDBA sign increased fivefold the likelihood of poor long-term outcome.
Hammill, B. G., Hernandez, A. F., Peterson, E. D., and others (2009, June). "Linking inpatient clinical registry data to Medicare claims data using indirect identifiers." (AHRQ grant HS10548). American Heart Journal 157(6), pp. 995-1000.
The authors of this paper demonstrate that it is possible to create a high-quality link between inpatient clinical registry data and Medicare claims data, even if there are no direct patient identifiers (patient name or Social Security number) common to the two data sources. The absence of such direct identifiers in disease- and procedure-specific registries has made it difficult, if not impossible, to follow the registry patients as outpatients. Among all Medicare claims, almost all records were unique when unique dates of birth were considered with any combination of admission or discharge date, regardless of patient sex, the authors say. They validated the links between Medicare claims data and clinical registry data by comparing Medicare hospital names for the claim with the hospital names in the registry. This process was limited to patients age 65 years or older who were not treated in a Veterans Affairs hospital. Once registry patients have been identified in Medicare, researchers can track postdischarge outcomes such as mortality, readmission, and subsequent inpatient procedures, the authors note. This would permit researchers to investigate questions of long-term safety and efficacy of inpatient treatments.
Hersh, A. O., von Scheven, E., Yazdany, J., and others (2009, January). "Differences in long-term disease activity and treatment of adult patients with childhood- and adult-onset systemic lupus erythematosus." (AHRQ grant HS13893). Arthritis & Rheumatism 61(1), pp. 13-20.
Individuals living with childhood-onset systemic lupus erythematosus (cSLE) have worse outcomes and take more steroids than those who developed the disease as adults (adult-onset or aSLE), according to a new study based on self-report data of 90 adults with cSLE and 795 with aSLE from the University of California Lupus Outcomes Study. Those with aSLE were more likely to have active disease on all measures of disease activity, such as flareups in the past 3 months (49.9 vs. 23.7 percent), and were more likely to develop pulmonary disease. Those with cSLE were more likely to be on prolonged steroid therapy and to develop SLE-related kidney disease. Rates of heart attack were similar between the two groups; however, the average age of first heart attack was 32 years in the cSLE group and 48 in the aSLE group. These findings may help develop tailored treatments based on age at SLE onset and underscore the need for clinical support as patients transition from pediatric to adult rheumatology care.
Hill, S. C. and Miller, G. E. (2009). "Health expenditure estimation and functional form: Applications of the generalized gamma and extended estimating equations model." Health Economics (available at http://www3.interscience.wiley.com). Published online May 11, 2009. Reprints (AHRQ Publication No. 09-R067) are available from the AHRQ Publications Clearinghouse.
The authors note that regression models of health care expenditures are used in a wide variety of economic analyses, including risk adjustment and evaluations of both programs and treatments. Specifically, generalized gamma models (GGMs) and extended estimating equations models (EEEs) have proven useful in expenditure estimation in situations with a variety of data problems. The authors use data from the AHRQ Medical Expenditure Panel Survey to compare the bias, predictive accuracy, and marginal effects of these models and four other estimation models in more common use. The study shows that the EEE model is a robust estimator for health expenditures, minimizing bias and prediction error. However, it requires large samples to obtain precise estimates of coefficients and marginal effects, the authors report.
Jackson, S. L, Taplin, S. H., Sickles, E. A., and others (2009, June). "Variability of interpretive accuracy among diagnostic mammography facilities." (AHRQ grant HS10591). Journal of the National Cancer Institute 101(11), pp. 814-827.
To try to identify the factors that lead to differences in results among radiology facilities that do diagnostic mammography, the authors conducted a study of facilities associated with three mammography registries belonging to the national Breast Cancer Surveillance Consortium. They analyzed rates of false-positive mammograms, sensitivity, and likelihood of cancer among women referred for breast biopsies at 32 radiography facilities in the Pacific Northwest, Colorado, and New Hampshire. The analyses included 28,100 diagnostic mammograms and data on 118 radiologists who interpreted them. The authors found that, after adjustment for patient and radiologist characteristics, the facilities differed significantly only in the percentage of false-positive mammograms. Facilities with higher concern about malpractice had higher false-positive rates, the authors found.
Kirby, J. B. (2008, September). "Poor people, poor places and access to health care in the United States." Social Forces 87(1), pp. 325-355. Reprints (AHRQ Publication No. 09-R047) are available from the AHRQ Publications Clearinghouse.
This study used data from the 2000 Medical Expenditure Panel Survey and U.S. Census data to identify the prevalence of poverty in communities and its relationship to health care access. The researcher found that individuals living below 200 percent of the Federal poverty line were not adversely affected by community-level poverty when it came to accessing health care. Instead, those individuals living above the 200 percent poverty line were the most affected by poverty in their community in terms of health care access. In fact, this group completely accounted for the negative relationship found between community-level poverty and access to health care. There was no evidence that higher income residents living in poor communities can use their extra resources to overcome the disadvantage of living in these environments.
Korthuis, P. T., Saha, S., Fleishman, J. A., and others (2008, December). "Impact of patient race on patient experiences of access and communication in HIV care." (AHRQ grant HS13903 and contract 290-01-0012). Journal of General Internal Medicine 23(12), pp. 2046-2052. Reprints (AHRQ Publication No. 09-R032) are available from the AHRQ Publications Clearinghouse.
This study suggests that HIV care sites that provide services to a greater proportion of blacks and Hispanics than whites may be more difficult to access for all patients. Researchers surveyed 915 HIV-infected adults receiving care at 14 U.S. HIV clinics. On average, blacks and Hispanics spent more time traveling to the care site than whites (36 and 37 vs. 29 minutes). Further, travel time to the HIV care site lengthened as the proportion of black and Hispanic patients increased at a given site. Waiting times at care sites were longer for Hispanics and blacks than whites (36 and 31 vs. 27 minutes) and increased with the proportion of Hispanic (but not black) patients at each site. Improving access will provide more patient-centered care for blacks and Hispanics who need treatment for HIV. Despite travel and wait times, patients gave high ratings for the quality of communications with their providers.
Lautenbach, E., Babson, A., Santana, E., and others (2009, April). "Assessment of the use of urine samples to detect colonization with fluoroquinolone-susceptible and fluoroquinolone-resistant Escherichia coli." (AHRQ grant HS10399). Infection Control and Hospital Epidemiology 30(4), pp. 396-397.
The authors examined whether urine samples could give additional information about the presence of fluoroquinolone-susceptible or -resistant strains of Escherichia coli (E. coli) bacteria beyond that obtained from rectal swabbing. The study, carried out in two long-term care facilities associated with the University of Pennsylvania, required growing the bacteria from swabs or urine on culture dishes in the presence or absence of the antibiotic fluoroquinolone. E. coli was identified in 55 percent of 56 perirectal swabs, and in 11 percent of 56 urine samples. In all six perirectal swab samples paired with urine samples that produced bacterial strains, the same strain was found in the urine sample. However, in some cases distinct bacterial strains were found in urine samples, but were not found in subsequent perirectal samples. While the use of urine samples may give a more complete measure of total exposure of the patients to colonizing bacterial strains, not all of the distinct E. coli strains were found to colonize the gut. The value of adding urine collection to the study will depend on the aims of the study, the authors conclude.
Levtzion-Korach, O., Alcalai, H., Orav, E. J., and others (2009, March). "Evaluations of the contributions of an electronic web-based reporting system: Enabling action." (AHRQ grant HS11046). Journal of Patient Satisfaction 5(1), pp. 9-15.
Incident reporting represents a key tool in safety improvement, and electronic voluntary reporting systems are increasingly being implemented. Since most studies of such systems have focused on the act of reporting, the researchers decided to study not only the rate and content of such reports, but also the actions and followup that result from the reports. They analyzed 14,179 reports submitted over a 31-month period through a Web-based system operating at a large urban hospital. Most reports involved laboratory results (30.4 percent), followed by medications (17.2 percent), and falls (10.9 percent). The majority of reports (70 percent) were reviewed by an average of four people within 72 hours. First manager review was completed within a median of 22 hours. The followup actions taken included change of equipment, change of policy, further education, and change of staffing levels. The researchers conclude that the Web-based reporting system effectively captured incidents, actions, and followup.
Messina, C. R., Lane, D. S., and Colson, R. C. (2009, July/August). "Colorectal cancer screening among users of county health centers and users of private physician practices." (Cooperative agreement between AHRQ and the National Cancer Institute). Public Health Reports 124, pp. 568-578.
Compliance with colorectal cancer (CRC) screening recommendations remains suboptimal. The researchers examined the use of CRC screening methods among 500 county health center registrants and 570 private physician patients aged 52 to 75 years in Suffolk County, New York. The screening methods were the fecal occult blood test (FOBT), sigmoidoscopy, and colonoscopy. Among county health center registrants, FOBT was more frequent; among private physician patients, sigmoidoscopy and colonoscopy were more common. County health center registrants less frequently cited no physician recommendation as a barrier to FOBT, but more frequently cited no recommendation as a barrier to sigmoidoscopy and colonoscopy compared with private physician patients. Among private physician patients, there was an association between perceived barriers to screening and lower odds of any screening. Lack of health insurance or inadequate health insurance remain barriers to equitable use of endoscopy among both groups of patients.
Resnik, L., and Plow, M. A. (2009, May). "Measuring participation as defined by the International Classification of Functioning, Disability, and Health: An evaluation of existing measures." (AHRQ grant T32 HS00011). Archives of Physical Medicine Rehabilitation 90, pp. 856-866.
The researchers compared and evaluated measures designed to assess participation, disability, and handicap by using the International Classification of Functioning, Disability, and Health (ICF). The World Health Organization's ICF model is the most recent and comprehensive model of functioning and disability. Content analysis was conducted by classifying participation-related items of each measure into one of the nine activities and participation chapters of the ICF taxonomy. Forty generic and condition-specific self-report measures that met inclusion criteria were evaluated. Five measures that contained participation items linked to all nine chapters were the Community Living Skills Scale, the Assessment of Life Habits, the Mayo-Portland Adaptability Inventory, the Participation Measure for Post-Acute Care, and the Psychosocial Adjustment to Illness Scale. These five measures were compared with respect to their breadth and coverage. Each of the measures differed in specifics of coverage and the approach to assessing participation. Clinicians and researchers can use these findings to select the most comprehensive participation outcome measure for their population.
Sedrakyn, A. (2009, June). "CABG versus PCI for multivessel coronary artery disease." The Lancet 373, pp. 2199-2200. Reprints (AHRQ Publication no. 09-R077) are available from the AHRQ Publications Clearinghouse.
In a letter to the editor, the author criticizes some of the presentation and reporting of data in a study (Hlatky and colleagues, April 4, p. 1190) pooling data from trials of coronary artery bypass grafting (CABG) vs. percutaneous coronary interventions. He criticizes a table depicting 5-year event rates for showing only the overall hazard ratio of 0.91, instead of including relative risk estimates and p values for 5-year followup. Also unclear is whether the final statistical model included the two interaction terms that were significant (age and diabetes). In addition, using a very high standard for reporting subgroup effects ignores the evidence favoring the lower mortality in CABG patients with stable symptoms. Finally, even if only diabetes and older people are the subgroups that will have lower mortality with CABG, in the real world they represent more than 55 percent of people undergoing revascularization.
Shin, J.K. (2009). "Inpatient stays of Asian patients with psychiatric diagnoses in New York City." (AHRQ grant HS00149). Issues in Mental Health Nursing 30, pp. 112-121.
Asian Americans may resist using Western mental health services for many reasons including stigma, fear of institutionalization, and lack of faith in the benefits of psychotherapy; however, a number of studies have suggested that those who do enter the mental health system are among the most severely disturbed patients. Using New York City data from 1995 to 1999, the researcher sought to determine whether Asians, after adjusting for other covariates in large data sets, have longer length of stay (LOS). The author found that schizophrenic disorders were the most frequent diagnoses among Asians, with a rate twice as great among Asians as among non-Asians. In the study period, Asians stayed between 2.59 and 1.67 additional days longer than non-Asians. Also, Asians, when compared to non-Asian groups, were underrepresented in inpatient services. They were only about 1 percent among total hospitalized patients with psychiatric disorders, while the population of Asians in New York City was 10 percent during this study.
Sterkel, R., Banister, C., Bruns, J. and others (2009, June). "Collaboration with pediatric call centers for patient recruitment." (AHRQ grant HS15378). Archives of Pediatric Adolescent Medicine 163(6), pp. 588-589.
Recruitment of study subjects who are representative of community members is difficult. The researchers collaborated with a pediatric call center to recruit subjects for a study to evaluate whether a telephone coaching program for parents of young children with persistent asthma would reduce asthma morbidity. They found 114 pediatricians who used the call center and gave their permission to contact families from their practice who had received asthma care from the call center. Initially, call center nurses attempted recruitment, but parents were rarely at home during hours when the nurses could call, so study staff were allowed to contact families directly. Although multiple calls were sometimes necessary, there were few problems in the recruitment process. The collaboration with the call center enabled the researchers to reach their recruitment target of 360 children within 13 months. The use of the call center allows access to a broad sample of potentially eligible subjects who are likely representative of the general pediatric population.
Wakefield, D. S., Brokel, J., Ward, M. M., and others (2009). "An exploratory study measuring verbal order content and context." (AHRQ grant HS15196). Quality and Safety in Health Care 18, pp. 169-173.
The use of verbal orders (VO), while essential in some health care settings, has been identified as a potential contributor to poor quality and less safe care. Both the content and the context of VOs potentially contribute to errors. The authors first identify variables related to VO content and context, and then provide detailed analyses from two exploratory studies conducted in a community hospital. Content variables include complexity, urgency, use of abbreviations, high-alert medications, and drugs with sound-alike names. Context variables include the type of care setting, the time of day, the communication process (face-to-face vs. telephone), personnel characteristics, and environmental factors such as background noise and staffing levels. The authors present analyses that focus on variations in types and timing of verbal orders hospitalwide as well as for specific inpatient units, changes in the use of VOs following implementation of a computerized provider order entry system, and the presence of sound-alike and high-alert medications in VOs.
Wideroff, L., Phillips, K. A., Randhawa, G., and others (2009). "A health services research agenda for cellular, molecular, and genomic technologies in cancer care." Public Health Genomics 233-244. Reprints (AHRQ Publication No. 09-R071) are available from the AHRQ Publications Clearinghouse.
Extensive resources have been invested to develop cellular, molecular, and genomic (CMG) technologies with clinical applications that span the continuum of cancer care. In December 2006, the National Cancer Institute (NCI) sponsored the first workshop of researchers to focus on the delivery of CMG interventions in cancer care. The authors report on the outcomes of this workshop. The workshop identified a comprehensive research agenda to better understand the delivery of care for emerging CMG interventions. This agenda incorporates health and safety endpoints, utilization patterns, patient and provider preferences, quality of care and access, disparities, economics and decision modeling, trends in cancer outcomes, and health-related quality of life among target populations. The workshop discussion included such issues as the scope of CMG technologies in cancer care, gaps in efficacy and effectiveness research, how to address these gaps, and challenges in addressing gaps.
Winter, L., Moss, M.S., and Hoffman, C. (2009). "Affective forecasting and advance care planning: Anticipating quality of life in future health statuses." (AHRQ grant HS13785). Journal of Health Psychology 14(3), pp. 447-456.
Affective forecasting, the process by which people anticipate their emotional responses to future events, has attracted considerable research attention in recent years. Decades of research using various methods document that quality-of-life valuations from patients with a particular illness tend to be higher than those without. Hence, sicker individuals express wishes for longer life in future poor-health statuses than do those who are well. To understand the decision process that underlies these choices, the authors previously proposed Prospect Theory, a powerful decision model that explains why closer prospects are easier to distinguish than more distant prospects. For this study, they interviewed 230 elderly persons by asking the Years of Desired Life questions which were linked to 9 separate health scenarios. They found that the interaction between current health status and health scenario supported the relative acceptability of poor-health prospects to sicker individuals, confirming the initial hypothesis.
Zuvekas, S. H., and Olin, G. L. (2009, Spring). "Accuracy of Medicare expenditures in the Medical Expenditure Panel Survey." Inquiry 46, pp. 92-108. Reprints (AHRQ Publication No. 09-R068) are available from the AHRQ Publications Clearinghouse.
In detailed comparisons of the Medical Expenditure Panel Survey (MEPS) with the National Health Expenditure Accounts (NHEA), an earlier study had found an estimated gap in expenditures of 14 percent between MEPS and NHEA. The authors examined underreporting and underrepresentation of high expenditure cases in the MEPS that had been identified as the primary drivers of this gap. They found that underreporting in the MEPS affected all population groups, but there were important differences in the extent of underreporting. Underreporting was relatively greater for nonwhites compared with whites in the analytic matched sample. There was also significant geographic variation in the reporting of Medicare expenditures in the analytic matched sample. Overall, there was a 19 percent gap between the MEPS estimates of Medicare expenditures in the matched analytic sample and a comparably defined sample of beneficiaries in the Medicare Current Beneficiary Survey. About half of the gap was due to underreporting and the other half appeared due to underrepresentation of high-expenditure cases in the matched analytic sample.