Cover Image

PAPERBACK
$66.00



View/Hide Left Panel
Click for next page ( 81


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 80
13 Administrative Data in Effectiveness Studies: The Prostatectomy Assessment Elliott S. Fisher* and John E. Wennberg Comprehensive, population-based administrative health care data bases provide an increasingly accessible and important source of data for studies of the effectiveness of health care (1~. To illustrate their potential uses, their strengths, and their limitations, we describe the role that administrative data have played in the ongoing assessment of treatments for benign pros- tatic hyperplasia, one of the more common conditions affecting elderly men. OVERVIEW OF THE PROSTATECTOMY ASSESSMENT Analyses of administrative health care data bases have long documented marked variations in population-based rates of prostatectomy (2,3~. To understand the causes of these variations, a multidisciplinary team com- posed of practicing urologists from Maine and researchers from academic medical centers in the United States, Canada, and Europe was assembled. The assessment team, funded under the Patient Outcome Assessment Research Program of the National Center for Health Services Research, undertook a comprehensive program of evaluation, the early findings of which are de- scribed in a series of recent publications (4-8~. These findings are briefly summarized in Table 1 to provide a context for the description of the analyses based on administrative data. The first goal of the assessment process was to identify possible explana- tions for the observed variations in utilization rates. This entailed both a *The paper was presented by Dr. Fisher, but it represents the ongoing research of many investigators in the Prostatectomy Patient Outcomes Research Team of which Dr. Wennberg is the principal investigator. The work is now part of the Patient Outcome Research Team program of the Agency for Health Care Policy and Research. 80

OCR for page 80
USE OF LARGE DATA BASES TABLE 1 Aims, Methods, and Data Sources for Assessment of Treatments for Benign Prostatic Hyperplasia 81 Aim Describe patterns of use of treatments arid characterize the theories of efficacy advanced by their proponents Identify, define, and develop (where necessary) measures for the full spectrum of relevant outcomes Establish the best estimates for probabilities of the relevant outcomes of alternative treatments Assess the efficacy of alternative treatment theories Integrate results, identify questions for further research Method and Data Source Geographic variation studies using insurance claims and other large data bases Structured literature review and focus groups with practicing physicians Literature review and semi-structured interviews with patients, physicians Identification or development of valid and reliable outcome and case-mix measures Claims-based cohort studies; linkage of claims and other data bases Prospective cohort studies (Maine Interview Study) Decision analysis, meta-analysis Observational studies, randomized trials where appropriate Publication of results and impart . . . . . . 1nc trigs to practicing p ~yslclans Development of interactive video for Shared Medical Decision-making Procedure review of the scientific literature and discussions with practicing urologists in Maine. Two conflicting theories concerning the indications for prostatectomy were identified. Many physicians believed that prostatectomy should be performed early in the course of symptomatic prostatism on the theory that if the operation is delayed, the patient will be at higher risk when the surgery becomes unavoidable. Because overall life expectancy would be reduced by delay, those who held to this preventive theory believed that watchful waiting was not a reasonable option. In contrast, urologists who believed the quality of life theory argued that prostatectomy is not inevitable. For patients without evidence of actual or impending renal dysfunction, the primary indication for the procedure should be improvements in functional status and quality of life. According to this theory, watchful waiting is a reasonable option. To evaluate these competing theories, the assessment team identified all relevant outcomes through discussions with patients and physicians. A

OCR for page 80
82 EFFECTIVENESS AND OUTCOMES IN HEALTH CARE review of the medical literature demonstrated serious gaps in existing knowledge about these outcomes. Claims-based analyses made possible reliable mea- sures of the likelihood of mortality in the postoperative period and of reoperation (41. The probabilities for other outcomes-such as incontinence, impo- tence, and postoperative symptom relief and improvement in functional sta- tus required the development of new measurement instruments and the implementation of a prospective interview study of patients undergoing prostatectomy in Maine (6~. The findings of the literature review, the claims-based analyses, and the interview study provided sufficient data to assess the efficacy of watchful waiting versus transurethral prostatectomy (TURP) through decision analy- sis (5~. The decision analysis demonstrated that for most patients the deci- sion to undergo prostatectomy results in a slight decrease in life expect- ancy. These findings confirmed the opinion of those physicians who believed that the operation was justified primarily for its value in reducing symptoms. However, the assessment also demonstrated (a) that improvements in symp- toms were only available to those willing to accept the risks of the surgery, and (b) that patients with identical symptoms differed greatly in their attitudes toward those symptoms and, presumably, toward the risks of surgery. The assessment thus revealed that variations in utilization rates induced by practice style were primarily a function of differences in providers' attitudes toward the preventive theory and of difficulty in integrating pa- tients' preferences into the decision to undergo prostatectomy. To help address these difficulties, the assessment team developed a computer-assisted, inter- active video presentation that provides a comprehensive description of the risks and benefits of the alternatives and is tailored to the individual patient viewing the presentation. This Shared Medical Decision-making Procedure (SMDP) has been implemented in several participating centers, with both surgical and watchful waiting patients being followed up to provide further refinements in the probability estimates for outcomes. The assessment steps described above required the application of mul- tiple research methodologies. In the remainder of this chapter, we describe the role that administrative data bases played both in the assessment of TURP versus watchful waiting and in addressing a specific question that emerged from early analyses. OVERVIEW OF METHODS Because we are describing the results of a series of studies conducted over many years, it is impractical to present in detail the methods used in each of the analyses. The general approach followed, which was similar in all analyses, will be reviewed briefly. The reader is referred to the primary publications for additional detail (3,4,8,9~.

OCR for page 80
USE OF LARGE DATA BASES 83 All of the studies relied on administrative or health insurance data bases as primary sources of data. The Health Care Financing Administration (HCFA) maintains comprehensive files on inpatient, outpatient, and skilled nursing home care for virtually the entire U.S. population over the age of 65 (10,11~. Similar files have long been maintained by the Manitoba Health Services Commission, in the Oxfordshire Region of England, and in Den- mark (8~. Three features of these files are essential to the analysis. First, the eligible population can be precisely defined, the date of death can be ascer- tained independent of health care utilization, and patients can be located for long-term follow-up studies. Second, administrative procedures in each system ensure that virtually all hospital utilization is documented. Third, unique personal identifiers allow utilization files to be linked to each other, to the population files, and to other sources of data. The methods used to define cases for inclusion in the study population and to define relevant variables were similar in all claims-based analyses reported here. They thus represent a generalizable approach to the use of administrative data bases for cohort studies. CASE IDENTIFICATION AND VARIABLES All patients were initially identified on the basis of computerized hospi- tal discharge abstracts or physician claims documenting a prostatectomy during the various study periods encompassed by the assessments. Where both physician and hospital claims were available (HCFA and Manitoba), potential cases were identified, consistency checks carried out, validity of claims determined, and appropriate exclusions applied. For each case, the first prostatectomy during the study period was defined as the index opera- tion. Based on the claims data, three classes of variables were defined. Outcomes The population file was searched to determine whether and when patients might have died. Reoperation was defined based on the presence of subse- quent claims for prostatectomy. Other possible complications were defined based on combinations of diagnoses and procedures coded on inpatient hos- pital records and on physician claims for both inpatient and outpatient services. Patient Covariables Diagnoses recorded on the index hospitalization claim and on physician and hospital claims preceding the index prostatectomy were used to mea- sure comorbidity.

OCR for page 80
84 Treatment Variables EFFECTIVENESS ED OUTCOMES IN HEALTH CARE The specific codes recorded on hospital and physician claims were used to define the type of prostatectomy received by the patient (open versus transurethral). USES OF ADMINISTRATIVE DATA IN PROSTATECTOMY ASSESSMENT VARIATIONS IN UTILIZATION RATES First, and perhaps most important, studies of small-area variations in prostatectomy rates provided the initial stimulus for the research project and were critical to engaging the interest of practicing urologists in the assessment. Early studies documented age-adjusted population-based utili- zation rates for prostatectomy that varied by a factor of four across small areas of New England (3~. Other studies documented variations across large geographic regions (12) and between and within countries with differ- ent health care financing and organizational structures (131. Discussed extensively elsewhere (14,15), small-area analyses have highlighted the clinical uncertainty surrounding many decisions in medicine and underlined the need for comprehensive assessments of the risks, benefits, and alternatives to specific treatments. POPULATION-BASED ESTIMATES OF ADVERSE OUTCOME RATES As mentioned above, urologists in Maine disagreed in their understand- ing of the risks and benefits of prostatectomy. Some of this disagreement could be attributed to gaps and flaws in the existing medical literature. Physicians usually rely on reports from clinical trials and case series to estimate the risks of adverse outcomes following specific surgical or medi- cal interventions. Unfortunately, these sources suffer from several limita- tions. One problem with case series is reporting bias: only where the results are better than previously reported is there a strong incentive to publish. Consequently, published rates of adverse outcomes may underesti- mate the risk in most clinical settings. Clinical trials usually report findings on highly selected populations and therefore may be difficult to generalize. Moreover, sample sizes are usually limited, and follow-up and choice of outcomes for study vary among case series. Consequently, confidence in- tervals (CIs) are likely to be wide, rare events may not be documented at all, and results are difficult to pool. Claims data can overcome these limita- tions.

OCR for page 80
USE OF LARGE DATA BASES 85 In the mid-1970s, a review of the literature on prostatectomy stated that mortality rates following TURP were under 1 percent and that patients rarely required reoperation (16~. Wennberg, Roos, and colleagues, using claims data from Maine and Manitoba for 1974 through 1976, found that over 3 percent of patients died within 90 days of surgery and that the overall rate of reoperation following TURP was 20.2 percent at eight years (4~. While these findings demonstrate the difficulty of relying on small, highly selected samples to estimate the likelihood of various outcomes, the data are by now quite old. What are the current risks of prostatectomy? We have used Medicare data for New England to examine mortality and morbidity following prostatectomy in the 1980s (Tables 2 and 3~. Because of the large sample sizes, mortality rates for prostatectomy can now be precisely estimated: 30-day mortality ranges from 0.3 percent for patients between the ages of 65 and 69 to 2.6 percent for patients age 80 and over. Studies of morbidity are more difficult when relying on claims data alone, because few of the diagnostic and procedure codes used on hospital discharge abstracts or physician claims specify that a given complication or procedure is the direct consequence of a prior prostatectomy. Consequently, using methods similar to those described by Roos et al. (17), we asked physicians to group codes into those that were possibly complications of the procedure (that is, outcomes occurring with increased frequency following any opera- tion) and those that were probably complications (because they are more directly related to prostatectomy). More than 10 percent of patients had a probable complication, while 16 percent had a possible complication. In all, almost one-quarter of patients had significant adverse outcomes in the 90 days following prostatectomy. TABLE 2 Mortality Rates Following Prostatectomy Among Medicare Enrollees Who Were New England Resident Patients Without Indication of Prostate or Bladder Cancer, 1984-1986 Patient Dead Patient Dead Within 30 Days Within 90 days Cases of Surgery of Surgery Age Group (No.) (Percent) (Percent) 65-69 6,428 0.3 1.2 70-74 6,946 0.8 2.3 75-79 5,740 1.2 3.2 80 and over 5,652 2.6 6.8 Total 24,766 1.2 3.3 NOTE: Based on Medicare Part A and Part B claims and Medicare Enrollment (HISKEW) files.

OCR for page 80
86 EFFECTIVENESS AND OUTCOMES IN HEALTH CARE TABLE 3 Morbidity Rates Within 90 Days of Transurethral Prostatectomy Among Patients Without Indication of Prostate or Bladder Cancer Who Were New England Resident Medicare Enrollees, 1984-1986 Possible Complications Percent Probable Complications Percent Myocardial infarction 1.0 Bladder infection 2.0 Pulmonary embolus 0.3 Kidney infection 0.1 Respiratory infection 3.0 Prostate infection 0.2 Wound infection 0.3 Other urinary infection 0.3 Congestive heart failure 1.5 One or more urinary infection 2.3 Phlebitis 0.2 Stricture~treatment 3.7 Deep venous thrombosis 0.3 Retention treatment 1.8 Arterial embolus 0.4 Other invasive testing 4.5 Bleeding 7.5 Second prostatectomy 0.6 Miscellaneous 3.3 One or more invasive procedures 8.7 One or more of above 16.0 One or more of above 10.3 One or more possible or probable complications, 23.4 percent NOTE: Based on Medicare Part A and Part B claims files. Although these data demonstrate that a variety of adverse events may be detected through the claims data, several limitations must be acknowledged. First, the completeness and accuracy of the coding in claims data bases has been questioned (18,19~. However, if the accuracy of the data could be confirmed and if administrative safeguards were enacted to ensure their complete and accurate documentation, then claims-based measures could be used to monitor the outcomes of care for patients undergoing prostatectomy. Second, the scope of the data is limited. Many outcomes critical to the prostatectomy assessment, such as disease-specific functional status and quality of life, could not be ascertained from the claims data. The next section provides examples of how these specific limitations of the claims data can be overcome. COMPARISONS OF TRANSURETHRAL AND OPEN PROSTATECTOMY The initial claims-based analyses of prostatectomy outcomes in Maine and Manitoba also compared the long-term results of TURP with those of open prostatectomy. Both operations have the same purpose-to relieve urinary obstruction. The open procedure is usually performed through an incision in the abdominal wall, whereas the transurethral procedure is per- formed through the urethra. Because of its less invasive nature, TURP was believed by urologists to be both safer and more effective than the open

OCR for page 80
USE OF LARGE DATA BASES 87 operation. Although a randomized clinical trial has never been conducted, TURP has gradually replaced open prostatectomy to the point where, in the 1980s, only about 5 percent of prostate operations in our data base were open. The claims data provided an opportunity to compare the long-term out- comes of the two procedures. Controlling for both patient and hospital characteristics, our study showed that patients undergoing TURP were twice as likely to require reoperation within eight years and appeared to face a significantly elevated long-term risk of death, compared to patients receiv- ing the open procedure (4~. These findings raised potentially important questions about both the safety and the efficacy of TURP compared with open prostatectomy. To evaluate further the association between the type of operation re- ceived by patients and their long-term outcomes, several additional studies were conducted. The first study sought to determine whether the increased risk associated with TURP would be found across different time periods and in different countries. Retrospective cohorts were assembled; these cohorts consisted of all patients aged 55 through 85 (except those with bladder or prostate cancer) who underwent prostatectomy between 1977 and 1985 in Denmark, between 1972 and 1985 in Manitoba, and between 1963 and 1977 in the Oxfordshire region of England (8~. The risk of reoperation was consistently higher among patients who received a TURP, ranging from a relative risk of 2.7 at eight years in Denmark to 6.7 at eight years in Oxford. Also, the risk of death following TURP was consistently higher at five and eight years, the relative risk of TURP to open being 1.2 to 1.3 at eight years. There remained the possibility that physicians were selecting only rela- tively healthy patients for the open procedure and that increased severity of illness among TURP patients might explain the excess mortality observed. Data from a teaching hospital in Manitoba were reviewed to investigate this possibility. All patients who underwent prostatectomy at the hospital be- tween July 1974 and December 1983 were identified through the claims data. Those with bladder or prostate cancer were excluded. All claims records before and after prostatectomy were identified and used to define patient covariables, including age, the presence of cancer diagnoses, prior hospitalizations with high-risk diagnoses, and nursing home residence. A clinical data base collected by anesthesiologists for a study of all surgical patients at this hospital was identified, and key clinical variables were ex- tracted and linked to the prostatectomy records. The linked variables included the American Society of Anesthesiologists' risk score and medication use. Among all cases the adjusted relative risk of death within five years was 1.45 (95 percent CI, 1.15, 1.84) (see Table 4~. Similarly, after excluding all cases with evidence of significant comorbidity, the relative risk remained

OCR for page 80
88 EFFECTIVENESS AND OUTCOMES IN HEALTH CARE TABLE 4 Relative Risks of Death for Patients Receiving Transurethral (TURP) and Open Prostatectomy, Operated On at Manitoba University Hospital, 1974-1983, by Selected Demographic and Clinical Characteristics All Patients Healthiest Patientsa Characteristics (N = 1650) (N = 557) TURP vs. Open prostatectomy 1.45 (1.15, 1.84)b 1.60 (0.93, 2.77) Age Groups 85+ vs. 55-69 3.75 (2.75, 5.09) 5.92 (2.44, 14.40) 80-84 vs. 55-69 2.77 (2.07, 3.72) 5.22 (2.57, 10.60) 75-79 vs. 55-69 2.35 (1.78, 3.10) 3.54 (1.85, 6.79) 70-74 vs. 55-69 1.48 (1.12, 1.96 1.60 (0.79, 3.24) Cancer diagnosis prior to surgery 3.93 (2.92, 5.28) NAC Hospitalized with high-risk diagnoses prior to surgery Within 6 months 1.46 (1.14, 1.87) NA Within 7-12 months 1.54 (1.13, 2.10) NA Nursing home resident 1.17 (0.76, 1.80) NA ASA Score 3+ 1.91 (1.57, 2.36) NA On digitalis 1.40 (1.10, 1.78) NA High-risk diagnosis 1.42 (1.15, 1.76) NA Prostatic hyperplasia only diagnosis 0.54 (0.38, 0.77) - .0.41 (0.22, 0.74) NOTE: Cox regression results based on linked claims and anesthesia data bases. aHealthiest defined as not resident in nursing home, had no current or previous diagnosis of cardiovascular disease, had no diagnosis of cancer, took no medica- tions preoperatively, had no other high-risk diagnosis, and had a physical status score of 1 or 2 (healthy or mild disease). bg5 percent confidence intervals in parentheses. CNot applicable. SOURCE: Roos et al. (8) . elevated at 1.60 (95 percent CI, 0.93, 2.77), although the confidence limits increased because of the smaller sample size. There remained a concern that the elevated risk might reflect subtle char- acteristics of patients known to their physicians and recorded in the medical record but not in either the claims data or the anesthesiologists' study. To address this concern, the medical records of a sample of TURP and open patients were abstracted to obtain a broad range of clinical data from patients'

OCR for page 80
USE OF LARGE DATA BASES 89 histories, physical examinations, and laboratory findings at the time of sur- gery. The medical record data were used to determine an index of comorbidity and a measure of functional health, both of which have been previously demonstrated to predict long-term survival (20,21~. Two Cox regression models were developed. In one, we used the indices of comorbidity and functional health to control for differences in illness levels. In the other we allowed all variables significantly associated with long-term survival into the model. Using these models, the relative risk of death within five years of operation was elevated for patients undergoing TURP compared to open prostatectomy, and it was similar in magnitude to the relative risk obtained from the claims data alone (Table 5~. These analyses suggest several conclusions. First, they confirm our ini- tial observation of increased mortality and reoperation rates among TURP patients in the original small sample from Maine and Manitoba. Second, measures of case mix that were obtained retrospectively did not explain the findings. However, it is important to note that patients may appear similar based upon retrospective review of their charts, but that the measures ob- tained retrospectively may not identify significant prognostic differences. For example, physicians may record characteristics of patients differently, based on their own assumptions about the relative safety of TURP compared to open prostatectomy. Nevertheless, because of the large numbers of patients undergoing TURP and the potential public health importance of the observed increased mortality following TURP, the evidence we found should not be TABLE 5 Relative Risk of Death for Patients Receiving Transurethral (TURP) versus Open Prostatectomy, Operated On at Manitoba University Hospital, 1974-1983 Variable Adjusted Relative Risk (95% confidence interval) TURP vs. Open Age 70-74 vs. under 70 Age over 75 vs. under 70 Comorbidity index > 2 Decreased functional statusa 1.59 (1.06, 2.37) 1.69 (1.05, 2.64) 2.23 (1.38, 3.58) 2.52 (1.74, 4.08) 2.66 (1.74, 4.08) NOTE: Cox regression results based on linked claims and chart review data, N = 485. Decreased functional status defined as a Karnofsly score < 70. SOURCE: Malenka et al. (9).

OCR for page 80
9o EFFECTIVENESS AND OUTCOMES IN HEALTH CARE ignored. We are pleased that the American Urological Association has joined with our assessment team to undertake the prospective clinical trials needed to resolve the issue. IMPROVING THE USEFULNESS OF ADMINISTRATIVE DATA BASES Administrative data have played an important role in stimulating the current interest in studying the effectiveness of medical care and offer an important resource for assessments of current treatment patterns. To make use of their full potential, we should build on their strengths and make the investment necessary to overcome their limitations. STRENGTHS As recognition of the importance of further evaluation of medical prac- tice has grown, so has advocacy of the Medicare claims files and similar data bases as sources of data for technology assessment. The assessment of prostatectomy exploited four major strengths that Medicare data offer for outcomes research. First, the enrollment file provides not only the popula- tion counts required for epidemiological studies, but also a means to effi- ciently ascertain death, eligibility status, and change of residence for long- term follow-up studies. Second, universal coverage offers the opportunity to study populations that are free from selection bias and are of sufficient size to document rare outcomes. Virtually all health care utilization by the covered population is identified in these files. Third, individual identification numbers allow records to be linked across time and providers. Such linkage is essential to longitudinal studies of health care outcomes and utilization. Finally, the individual identification numbers provide a mechanism to link Medicare data to other sources of data. Potential sources of supplemental data include those reported here, existing clinical data bases, and medical records. It is also feasible to obtain names and addresses so that individuals could be surveyed to ascer- tain outcomes not recorded in either the claims themselves or the medical records, such as functional status and quality of life. LIMITATIONS As with any source of data, limitations in Medicare data must be ac- knowledged and, when possible, overcome. Treatments and diagnoses in the claims files are recorded in nonresearch settings using International Classification of Diseases (ICD-9-C~ codes (hospitals) and Common Procedural

OCR for page 80
USE OF LARGE DATA BASES 91 Terminology (CPT-4) codes (physicians). The precision of the codes them- selves and the accuracy with which they are recorded limit the kind of studies that may be successfully undertaken. Major surgical procedures have been found to be accurately coded, and the precision of these codes allows reasonable cohorts to be defined. In contrast, fine distinctions among different subgroups of patients with medical conditions are poorly documented within the existing coding conventions; it would be difficult, for example, to define a cohort of patients with unstable angina. Similarly, the records do not document either the timing of the onset of medical conditions within a hospitalization or the affected side (left vs. right) for procedures or conditions that may affect either side of the body. Codes for many new technologies and treatments are rarely introduced in a timely fashion. Specific codes for coronary angioplasty were introduced several years after the widespread adoption of the technique in practice. Finally, the scope of data recorded is limited, and utilization rather than the incidence of a medical event is recorded. Some patients with adverse out- comes may not bother or be able to afford to see their physicians. Certain events (mortality, reoperation) can be accurately measured, but other variables (clinical risk factors, functional status, quality of life) cannot be ascertained directly from the claims data. SUGGESTIONS These limitations suggest several steps we could take to enhance the value of administrative data bases for health care research and outcomes assessment. First, we should improve the completeness and accuracy of the coding used in claims data bases. Establishing codes for new technologies as soon as they become eligible for reimbursement would markedly enhance assessment efforts. Documentation and publication of the accuracy of cod- ing in administrative data bases by the agency responsible for collecting the data would enhance the utility of the data bases to all users. Second, because the scope of the data is limited, additional data will be required for many analyses. We should be cautious in our strategies for supplementing data, however. There is a tension between the desire to collect all possibly relevant data on each patient and the needs of a given assessment. For example, the specific variables required to study angioplasty or prostatectomy are not likely to be included in even the most comprehen- sive data set. Consequently, we should determine efficient, flexible means of supplementing the data base. These might include not only facilitating access to medical records to supplement claims data, but also developing strategies for routine posttreatment interviews to determine functional sta- tus and quality of life.

OCR for page 80
92 EFFECTIVENESS AND OUTCOMES IN HEALTH CARE REFERENCES 1. Roper, W.L., Winkenwerder, W., Hackbarth, G.M., et al. Effectiveness in Health Care: An Initiative to Evaluate and Improve Medical Practice. New England Journal of Medicine 319:1197-1202, 1988. 2. Wennberg, J.E. and Gittelsohn, A.M. Small-Area Variations in Health Care Delivery. Science 183:1102-1108, 1973. 3. Wennberg, J.E. arid Gittelsohn, A.M. Variations in Medical Care Among Small Areas. Scientific American 246: 120-134, 1982. 4. Wennberg, J.E, Roos, N.P., Sola, L., et al. Use of Claims Data Systems to Evaluate Health Care Outcomes: Mortality and Reoperation Following Prostatectomy. Journal of the American Medical Association 257:933-936, 1987. 5. Barry, M.J., Mulley, A.G., Fowler, F.J., et al. Watchful Waiting vs. Immedi- ate Transurethral Resection for Symptomatic Prostatism: The Importance of Patients' Preferences. Journal of the American Medical Association 259:3010-3017, 1988. 6. Fowler, P.J., Wennberg, J.E., Timothy, R.P., et al. Symptom Status and Quality of Life Following Prostatectomy. Journal of the American Medical Asso- ciation 259:3018-3022, 1988. 7. Wennberg, J.E., Mulley, A.G., Hanley, D., et al. An Evaluation of Prostatectomy for Benign Urinary Tract Obstruction: Geographic Variations and the Assessment of Medical Care Outcomes. Journal of the American Medical Association 259:3027-3030, 1988. 8. Roos, N.P., Wennberg, J.E., Malenka, D.J., et al. Mortality and Reoperation After Open and Transured~ral Resection of the Prostate for Benign Prostatic Hyperplasia. New England Journal of Medicine 320:1120-1124, 1989. 9. Malenka, D.J., Roos, N.P., Fisher, E.S., et al. Further Study of the Increased Mortality Following Transurethral Prostatectomy. Urology. In press. 10. Lave, J., Dobson, A., and Walton, C. The Potential Use of Health Care Financing Administration Data Sets for Health Care Services Research. Health Care Financing Review 5 :93-98, 1983. 11. Hatten, J. Medicare's Common Denominator: The Covered Population. Health Care Financing Review 2:53-63, 1980. 12. Chassin, M.R., Brook, R.H., Park, R.E., et al. Variations in the Use of Medical and Surgical Practices by the Medicare Population. New England Journal of Medicine 314:285-290, 1986. 13. McPherson, K., Wennberg, J.E., Hovind, O.B., et al. Small-Area Variation in the Use of Common Surgical Procedures: An International Comparison of New England, England and Norway. New England Journal of Medicine 307:1310-1314, 1982. 14. Wennberg, J.E. Dealing with Medical Practice Variations: A Proposal for Action. Health Affairs 3 :6-33, 1984. 15. Paul-Shaheen, P., Clark, J.D., and Williams, D. Small Area Analysis, A Review and Analysis of the North American Literature. Journal of Health Politics, Policy and Law 12:741 -809, 1987. 16. Grayhack, J.T. and Sadlowski, R.W. Results of Surgical Treatment of Benign ProstaticHyperplasia Pp.125-134InBenignProstaticlIyperplasia. Grayhack,

OCR for page 80
USE OF LARGE DATA BASES 93 J.T., Wilson, J.D. and Scherbenske, M.J., eds. DHEW Publication No. (NIH) 76- 1113. Washington, DC: Government Printing Office, Washington, D.C., 1976. 17. Roos, L.L., Cageorge, S.M., Austen, E., et al. Using Computers to Identify Complications After Surgery. American Journal of Public [health 75: 1288-1295, 1985. 18. Greenfield, S., Aronow, H.U., Elashoff, R.M., et al. Flaws in Mortality Data: The Hazards of Ignoring Comorbid Disease. Journal of the American Medical Association 260:2253-2255, 1988. 19. Jencks, S.F., Williams, D.K., and Kay, T.L. Assessing Hospital-Associated Deaths from Discharge Data: The Role of Length of Stay and Comorbidities. Journal of the American Medical Association 260:2240-2246, 1988. 20. Charlson, M.E., Pompei, P., Ales, K.L., et al. A New Method for Classify- ing Prognostic Comorbidity in Longitudinal Studies: Development and Validation. Journal of Chronic Diseases 40:373-383, 1987. 21. Stanley, K.E. Prognostic Factors for Survival in Patients with Inoperable Lung Cancer. Journal of the National Cancer Institute 65:25-33, 1980.