Strategies for Treating Clinically Localized Prostate Cancer

Expectant Management or "Watchful Waiting"

The strategy of expectant management includes several approaches that range from reserving palliative interventions for patients who develop symptomatic local progression or metastatic disease to withholding potentially curative therapy until signs of disease progression are seen [3]. Expectant management is currently the most favored strategy outside of the United States [4]. Data on the long-term outcomes of this strategy are limited ([5-12]; Johansson JE. Natural history in early primary untreated prostate cancer [Presented paper]. 85th Annual Meeting of the American Urological Association; 1993; San Antonio, Texas), and almost no data are available on the distant prognosis of tumors that are discovered by PSA measurement.

In a structured review of the literature [13], disease-specific mortality and metastatic rates that were seen with expectant management did not differ significantly from those seen in published, uncontrolled surgical series. The rates associated with expectant management were actually lower than those reported for radiotherapy. However, a larger proportion of patients who were treated with radiotherapy had poorly differentiated tumors, reflecting important differences in patient selection. If disparities in the extent of pathologic staging are taken into account, patients who are treated by nonsurgical methods almost certainly have more advanced disease than can be appreciated on clinical grounds. Such analyses highlight the hazards of nonrandomized treatment comparisons [13-15].

Adolfsson and colleagues [16] identified seven studies of men who had palpable, clinically localized cancer that had been treated since 1980 and who had received expectant management. The 10-year disease-specific survival rate (which was measured in only two of the studies) was lower for expectant management (84%) than for radical prostatectomy (93%) but was lower still for radiotherapy (74%).

The prospective study by Johansson and colleagues of 223 men (mean age, 72 years) in Sweden is one of few published studies that is population based [11]. Persons who had clinically localized disease were enrolled between 1977 and 1984. After a mean follow-up of 12.5 years, mortality related to prostate cancer was 10%, whereas 66% of the group died from other causes. Ten-year metastasis-free survival corrected for deaths from other causes was 83%. This study has been criticized for enrolling too many older men and men who had incidental tumors diagnosed at transurethral resection of the prostate [17]. However, similar results were reported for a subset of men who were potential candidates for radical prostatectomy [11, 18].

More recent 15-year population-based data from Connecticut support the findings of earlier analyses of expectant management [19]. A patient-level meta-analysis of 762 men (mean age, 70 years) who were enrolled in six studies of expectant management and showed the following 10-year prostate cancer-specific survival rates by histologic grade: 87% for a well-differentiated tumor, 87% for a moderately differentiated tumor, and 34% for a poorly differentiated tumor [20].

In contrast, a more recent Scandinavian series [21] suggested poorer outcomes with expectant management. However, this study used a retrospective design, unlike the prospective cohort approach of Johansson and colleagues [11] and the retrospective cohort design of Albertsen and coworkers [19]. In addition, cause of death was not determined using blinded reviewers in the study by Aus [21] as it was in the study by Albertsen and coworkers [19]. Nonetheless, for 108 men who had clinically localized tumors that were well-differentiated or moderately differentiated, a 10-year crude survival rate of 78% was reported; 40% of all deaths in this group were caused by prostate cancer.

Men who receive expectant management are more likely than men who receive radical surgery to have local progression of prostate cancer. However, the clinical significance of local progression and its effect on quality of life have not been well studied [22]. Of the patients who were followed by Johansson (Johansson JE. Natural history in early primary untreated prostate cancer [Presented paper]. 85th Annual Meeting of the American Urological Association; 1993; San Antonio, Texas), 22% had cancer that progressed to clinical stage T3 on digital rectal examination after 10 years. However, only six cases were associated with local problems that were considered substantial.

Radiation Therapy

The effectiveness of radiation therapy has not been established in controlled trials [13]. A randomized trial of 97 men who had stage A2 or B cancer that was treated with radiation or surgery had a significantly longer time to recurrence after surgery but showed no difference in mortality rate [23, 24]. This analysis was from a less desirable "treatment-given" perspective rather than an intention-to-treat perspective. Few published observational studies of radiation therapy have stratified outcome by grade and stage [13]. Recent cohort studies of patients receiving radiation therapy for clinically localized cancer indicate that at 10 years, overall survival rates are no different for these patients than for age-matched controls [25-28].

In a recent structured review [13], radiation therapy was associated with higher median rates of metastasis and cancer-specific mortality than was surgery or expectant management. However, such comparative analyses are confounded by selection and staging bias because patients treated with radiation therapy tend to be more likely to have poorly differentiated tumors and occult nodal involvement.

Potential complications of radiation therapy include death (0.2% to 0.5%) and various gastrointestinal and genitourinary illnesses, including incontinence and sexual dysfunction, that may become chronic [13].

Radical Prostatectomy

Substantial improvements in the technique of radical retropubic prostatectomy have been reported [24]. Studies suggest that men who are found to have pathologically organ-confined disease after radical prostatectomy have a low risk for recurrence and essentially normal life expectancy. However, a recent analysis of data from national Medicare claims [29] indicates that approximately 25% of men who are reported to have pathologically localized disease at the time of radical prostatectomy require additional treatment within 5 years. Radical prostatectomy seems to reduce the probability of developing future problems with local disease even in patients who are found to have extracapsular disease during surgery [30].

The benefit that can be attributed to radical prostatectomy remains uncertain. Excellent prognoses in radical prostatectomy series done at single institutions [31] may be an effect of better case selection [32]. Outcomes from surgical series reported in 1981 to 1993 do not differ statistically from outcomes reported for expectant management [13]; however, differences in patient characteristics make such comparisons hazardous.

One randomized trial [33] directly compared expectant management with radical prostatectomy. No difference in survival was seen after 15 years of follow-up. However, because this study had only 111 patients, its power to detect clinically important differences in outcome was limited [34].

The risks of radical prostatectomy include perioperative death, early cardiovascular complications, and chronic impairments of urinary and sexual function [3, 13]. Risk may be greater for patients who are treated in community settings. In a survey of a national probability sample of 1070 Medicare-eligible men who had had radical prostatectomy between 1988 and 1990, Fowler and associates [35] found that 30% were wearing pads or other devices to manage urinary incontinence 2 to 4 years after surgery. Sixty-one percent reported having no or only partial erections since surgery (> 90% said that they had been potent before surgery). However, Medicare claims from one state documented only a 0.7% risk for death 30 days after radical prostatectomy [36]. Litwin and colleagues [37] recently documented that patients who have had radical prostatectomy are more likely to develop impairment in urinary and sexual function than are patients who have received radiation therapy or expectant management. On the other hand, Fowler and coworkers [38] found that almost all of the men in their survey would have radical prostatectomy again, despite the complications.

Methods

Top Methods Results Author & Article Info References

Estimating the Benefits and Harms of Early Detection

Decision models of prostate cancer screening or treatment have yielded conflicting results [39-47]. The absence of definitive data for many important variables, particularly the prognosis of clinically localized cancer that is not treated and the effectiveness of aggressive treatment for such cancer, can result in models that support a wide array of screening policies.

Using available data that have already been reviewed, we estimated the risks and maximum benefits of one-time testing using digital rectal examination and PSA measurement for men who are older than 50 years of age. We believe that the available data are insufficient to model the current American Cancer Society recommendation of an annual digital rectal examination and PSA measurement for men older than 50 years of age. As our analysis shows, available data do not allow a definitive assessment of the cost-effectiveness of one-time screening, let alone serial screening. Nevertheless, our model provides important insights about the implications of an early detection strategy.

The treatment model used here to project the risks, benefits, and costs of early detection was previously published by our group [40]. However, we have responded to criticism [41] by using recent outcome data from the meta-analysis by Chodak and associates [20] that examined outcomes of cancer that were achieved with expectant management. Furthermore, instead of making specific adjustments for quality of life, we report separately estimated incidences of nonfatal complications for alternative strategies.

To estimate the health outcomes of one-time screening for each of three age groups (50 to 59 years of age, 60 to 69 years of age, and 70 to 79 years of age), the model tracks a hypothetical cohort of 100 000 men using a Markov process with 6-month cycles. In the early detection strategy, men receive digital rectal examination and PSA measurement (a PSA level >4.0 ng/mL was considered abnormal) and are followed according to a set of assumptions about subsequent interventions and the probabilities of various outcomes. If men who harbor prostate cancer are not screened, we assume that their cancer will progress and may be fatal, as reflected by the cancer-specific mortality rates presented by Chodak and associates [20].

In our primary analysis, future health benefits and harms were not discounted.

Model Assumptions

Table 1 summarizes the assumptions and probability estimates that were derived from the literature and were used for the model's base case. These assumptions are discussed further in other published versions of the model [40, 55]. To estimate the maximum benefits of the early detection strategy, the model incorporates several important assumptions that favor early detection. However, we also did sensitivity analyses that included the current ranges of uncertainty that are associated with the critical variables of treatment effectiveness and the natural history of cancer and might be detected earlier with digital rectal examination and PSA measurement.

For our baseline analysis, we assumed that radical prostatectomy would be the initial treatment recommended to men who received a diagnosis of clinically localized disease. This assumption is supported by the recent dramatic increase in the use of this procedure [53]. Because preoperative staging cannot determine the volume, true grade, and pathologic extent of detected tumors, we assumed that all men who were found to have cancer would be offered aggressive treatment. Compliance rates with biopsy and treatment are based on data from Richie and colleagues [48].

One area of great uncertainty involves the effectiveness of therapy for clinically localized prostate cancer. We initially assumed that all men whose cancer was found, during surgery, to be confined to the prostate capsule would be cured by radical prostatectomy. We also assumed that all men who had overt extracapsular disease would not be cured and would have the same grade-dependent, disease-specific survival rates as men who did not receive treatment (Table 1). Although an unknown but significant portion of men whose cancer has established capsular penetration but negative surgical margins may be cured, these cases are balanced by the observation that 25% of men whose cancer has only focal penetration have progression within 8 years of radical prostatectomy [56]. Moreover, the recent linkage of the National Cancer Institute's SEER (Surveillance, Epidemiology, and End Results Program) data with Medicare claims suggests that approximately 25% of men whose cancer appears to be pathologically localized at the time of radical surgery receive additional treatment within 5 years [29]. Given the proportion of men who have extracapsular disease for each tumor grade, the assumption of a 100% cure for men who have intracapsular disease reduces mortality related to prostate cancer after radical prostatectomy by 97% for well-differentiated tumors, 70% for moderately differentiated tumors, and 56% for poorly differentiated tumors.

The prior probabilities of prostate tumors of different pathologic categories (tumors <0.5 mL, intracapsular tumors >0.5 mL, and extracapsular tumors >0.5 mL) for men in each age group are determined on the basis of autopsy reports [1]. The distribution of cancer by grade within each pathologic category was derived from a series of men who had clinical stage Tlc cancer and subsequently had surgical staging [49].

The probability that combined digital rectal examination and PSA measurement leads to a recommendation for biopsy and the number of cases of cancer in each pathologic category that would be detected are derived from a recent study of 6630 volunteers [48, 57]. Although these estimates are the best available, they are biased in favor of early detection because volunteers for the screening study may have had a higher prevalence of cancer.

Complications after biopsy and radical prostatectomy were derived from the recent literature [1, 13]. The model includes only the two most common treatment-related long-term complications: incontinence and impotence. Other, less common complications, such as rectal injury, are not considered [13].

On the basis of the meta-analysis by Chodak and associates [20], we assume that cancer-specific mortality rates for men who have clinically localized cancer are determined by the grade rather than by the extent of the tumor. Thus, an untreated tumor that is moderately differentiated is assigned the same prognosis whether it is intracapsular or extracapsular. However, we assume that tumors smaller than 0.5 mL in volume do not have metastatic potential; cancer in this category is considered truly "incidental." The assumption that the prognosis of cancer of a certain grade is the same whether the tumor is intracapsular or extracapsular particularly favors early detection because intracapsular cancer is assumed to be curable with aggressive treatment. To the extent that future cancer-specific death is more likely to be generated by extracapsular rather than by intracapsular cancer, our model overestimates the beneficial effect of early detection on mortality rate.

Assumptions about the management and outcome of men who have distant metastases have been reported elsewhere [40, 55]. The rates of metastatic disease were adjusted to reflect newer cancer-specific mortality rates that were reported by Chodak and associates [20]. Use of the raw metastatic rates reported by Chodak and associates would have resulted in cancer-specific mortality rates that were higher than those actually seen in that analysis, particularly for men with moderately differentiated cancer. We also assumed that no additional treatment would be provided after radical prostatectomy unless patients developed overt metastatic disease. Receipt of additional therapy after radical prostatectomy was reported by too many Medicare beneficiaries to allow us to attribute this therapy entirely to the development of metastatic disease [29, 35]. As a result, our model underestimates the real costs associated with early detection.

Results

Top Methods Results Author & Article Info References

Estimates of the risks and maximum benefits of screening 100 000 men in each of the three age strata using both digital rectal examination and PSA measurement are shown in Table 2, Table 3, and Table 4. In the absence of an early detection program, the projected base-case probability that a 65-year-old man will die of prostate cancer is 4.1%.

The predicted distribution of cancer that would be detected early and treated for men 60 to 69 years of age (the group most commonly represented in studies of radical prostatectomy) is as follows: tumors smaller than 0.5 mL, 11%; intracapsular tumors larger than 0.5 mL, 65%; and extracapsular tumors larger than 0.5 mL, 24%. Moreover, only 7% of all tumors are predicted to be smaller than 0.5 mL and well differentiated or truly insignificant (that is, without metastatic potential). This proportion of "insignificant" cancer is actually lower than that reported in some recent series of stage Tlc cancer that was treated surgically [58, 59].

Early detection would prompt many biopsies of the prostate (10 752 to 27 046 per 100 000 men, depending on age); cause few surgical deaths (8 to 23 per 100 000 men); and render many men newly impotent (657 to 1594 per 100 000 men), incontinent (131 to 317 per 100 000 men), or both (196 to 476 per 100 000 men). The attendant number of life-years during which patients would have such complications is substantial. Over time, the model predicts that 407 men 50 to 59 years of age, 653 men 60 to 69 years of age, and 427 men 70 to 79 years of age who would develop metastatic prostate cancer in the absence of early detection methods would die of something other than prostate cancer if screening were implemented. The net benefit (undiscounted) of early detection would be 4606, 4612, and 1618 life-years saved for the three age groups, respectively, or 17, 17, and 6 days per man screened, respectively.

If surgery for detected cases were completely ineffective, these cohorts would lose about 200 life-years from surgical death and have more than 20 000 life-years with incontinence, impotence, or both. The predicted net benefits, therefore, are sensitive to assumptions about the effectiveness of treatment.

Most physicians already do digital rectal examination in men who are older than 50 years of age to detect prostate and colorectal cancer. What is the marginal value of adding PSA measurement to the digital rectal examination? In the study by Catalona and coworkers [57], the results of digital rectal examinations suggested prostate cancer in 16% of men who were 60 to 69 years of age; the examination had a positive predictive value of 22% and a detection rate of 3.5%. Adding PSA measurement would have increased the detection rate from 3.5% to 6% if there had been full compliance with biopsy. Because the ratio of intracapsular to extracapsular disease was roughly equal between the cancer detected by digital rectal examination and the cancer detected by combination screening, approximately 60% of the risks and maximum benefits presented in Table 3 would be seen with digital rectal examination alone.

Estimated Cost-Effectiveness of Early Detection

Estimates of the costs of an early detection program are important to policy decisions and should incorporate the costs of the initial screening test or tests, all subsequent tests, biopsies, staging evaluations, early treatment, and therapy for treatment complications. To the extent that early detection and treatment are effective, savings accrue from avoiding later costs that are associated with cancer progression. These late savings are diminished by discounting because they occur far in the future.

To estimate the cost of an early detection program with digital rectal examination and PSA measurement among our hypothetical cohorts of 100 000 men in three age groups, we used cost estimates from the 1992 Medicare fee schedule, which was initially developed for an analysis of a policy of Medicare coverage for prostate cancer screening [60]. The cost estimates used in this analysis are shown in Table 5. Future cost and health benefits in this analysis are discounted at 5% per year.

We also estimated the probability that men in the cohorts would require future therapy for cancer progression, benign prostatic hyperplasia, and treatment complications, and we included the cost of such treatment in our analysis (5, 35; Johansson JE. Natural history in early primary untreated prostate cancer [Presented paper]. 85th Annual Meeting of the American Urological Association; 1993; San Antonio, Texas).

Table 6 shows the estimated discounted cost-effectiveness of screening compared with no screening in the three age groups of 100 000 men each with assumptions that favor screening, as discussed previously. The discounted maximal average number of days of life saved per person screened is 11 days for patients 50 to 59 years of age, 7 days for patients 60 to 69 years of age, and 3 days for patients 70 to 79 years of age. Estimates of the discounted average cost per person screened are $216 at 50 to 59 years of age, $387 at 60 to 69 years of age, and $532 at 70 to 79 years of age. The costs of the initial test, which are primarily for PSA measurement, are a small component of the overall average cost when compared with the major costs that accrue for follow-up after suspicious test results and for treatment of patients proven to have cancer. If these optimistic estimates of treatment benefit are correct, then the cost per year of life saved as a result of prevalence screening with digital rectal examination and PSA measurement would be $12 491 at 50 to 59 years of age, $18 769 at 60 to 69 years of age, and $65 909 at 70 to 79 years of age. Such estimates largely fall within the range of derived cost-effectiveness ratios for many commonly accepted medical practices [62], including cancer screening [55]. However, many of these other practices have empirical estimates of effectiveness that have been determined in well-designed controlled trials, whereas our estimates of effectiveness are hypothetical.

The natural history of the cancer that is currently being detected by digital rectal examination and PSA measurement has not been defined, and the effectiveness of aggressive treatment has not been established in controlled trials. Table 7 presents a two-way sensitivity analysis of the effect of varying the cancer-specific mortality rate (using lower rates derived from our original treatment analysis [40] and higher rates taken from the more recent meta-analysis by Chodak and associates [20]) and the effectiveness of treatment (assuming that 50%, 75%, or 100% of intracapsular tumors are cured by radical prostatectomy). Any relaxation of the favorable assumptions that we used in the base-case scenario increases the ratio of cost to benefit, an effect that increases with age.

The results of the base case of this analysis are more favorable than the results recently published by Krahn and coworkers [39] in terms of the days of life gained per person screened (1.1 undiscounted days at age 60) and cost-effectiveness ($127 000 per life-year saved, without quality adjustments). The models have different structures and are difficult to compare. However, we believe that our systematic pattern of making assumptions that favor early detection account for these differences, which include a higher underlying prevalence of cancer, higher cancer-specific mortality rates for men who do not receive treatment, greater treatment efficacy, and lower cost estimates for resource inputs.

Conclusions

The aggregate morbidity and mortality that can be attributed to prostate cancer are certainly sufficient to justify a search for effective and efficient strategies for early detection [3]. However, without randomized trials to quantify the risks and benefits of screening for prostate cancer or proof that treating clinically localized cancer reduces disease-specific mortality rates, the available observational data do not warrant the conclusion that early detection of prostate cancer using the tests now available unequivocally does more good than harm. However, our analysis of the cost-effectiveness of screening suggests that given a combination of favorable assumptions, early detection efforts that use digital rectal examination and PSA measurement might well be cost-effective, at least for men in their 50s and 60s, when compared with previously published estimates for other screening methods. Therefore, early detection cannot be dismissed out of hand as definitely ineffective or cost-inefficient.

As emphasized in part I of our study [1], the best combination of tests for early detection and the most cost-effective frequency of their use are not yet known. None of the currently available methods, alone or in combination, works optimally as a screening tool.

The critical question that now faces clinicians, patients, policymakers, and insurers is whether early detection efforts should be supported or discouraged pending completion of ongoing trials of screening and treatment. Although this debate is fueled by various interpretations of the scientific data that are already available, much of it stems from a clash of perspectives. Put simply, should we encourage widespread screening until we can prove that it does not work, or should we avoid it until we can prove that it does work? One end of the spectrum is advocated by those who argue that it is important to marshal all available resources to reduce mortality related to prostate cancer until controlled trials prove (if they ever prove) that such efforts are ineffective or cause net harm [63, 64]. At the other extreme, more cautious persons and groups emphasize the need to avoid the risks of early detection unless a net benefit of screening can be definitively shown in clinical trials [65-68]. We tend to support the latter perspective.

Pending the results of key clinical trials and given the availability of digital rectal examination and PSA measurement for the early detection of prostate cancer, decisions must be made by individual physicians and patients about the possible benefits and harms of early detection and treatment of this disease. In this context, our analysis identifies the salient issues that primary care physicians must consider when deciding whether to recommend an early detection strategy for a given patient. Our cost-effectiveness model suggests that one-time screening with digital rectal examination and PSA measurement may be defensible in younger men, given favorable but reasonable assumptions about existing nonexperimental data. Conversely, the model makes a reasonably strong case that early detection efforts for men 70 years of age and older are only marginally beneficial, even with very favorable assumptions. However, even for men in their 50s, the maximum average health benefit averaged across all patients is no more than a few additional weeks of life expectancy. For men identified with clinically localized cancer, surgery offers a potential gain of approximately 3 years of life at 55 years of age and 1.5 years at 65 years of age.

Our model quantifies the extent to which all men screened would be exposed to various discomforts and risks, ranging from the chance that biopsy of the prostate would be required to the probabilities of incontinence, impotence, and death. For some patients and physicians, the relatively well-defined harms of screening and treatment will appear to overwhelm even the maximum benefits presented, especially because risks are faced immediately whereas potential benefits are usually delayed for years. Others will discount the counterproductive aspects of such complications and accept the associated hazards, even for a benefit of uncertain magnitude [38, 69]. Men who place the highest premium on the possibility of avoiding advanced prostate cancer should probably choose screening. However, men who are averse to treatment-related risks or those who prefer to pursue only therapy that has a proven benefit should probably reject screening.

Our model and those of others [39, 41] highlight the importance of 1) refining the estimated grade-specific metastatic rate of prostate cancer that is currently being detected with digital rectal examination and PSA measurement and 2) identifying the facilities and surgeons that provide superior treatment with the fewest complications. The disparities between the reported outcomes of patients treated at single institutions and those of persons from broad-based populations, such as those identified through Medicare claims, also indicate that patients may face different risks and benefits depending on where and by whom they are treated.

Primary care physicians should be aware of the uncertainties in the key variables that influence early detection decisions and the tradeoff of potential benefits for known risks before they and their patients make a screening decision. We believe that the lack of proof of net benefit from early detection with digital rectal examination and PSA measurement and the potential for serious attendant harm mandate a higher level of informed consent than exists for most "simple" diagnostic tests. Primary care physicians should provide a balanced presentation of this issue so that patients can make a decision with which they are comfortable. Patients should understand the long-term ramifications of screening, including the relatively high probability that further evaluation (including biopsy) will be required and the potentially difficult decision about therapy that is associated with considerable morbidity rates and uncertain benefit if cancer is discovered. Screening should not be recommended for men who are unwilling to consider aggressive treatment if a tumor of potential clinical significance is found or who are not candidates for such therapy.

Clinicians do digital rectal examinations for reasons other than prostate cancer screening. These reasons range from the evaluation of nonspecific perirectal symptoms to screening for colorectal cancer. Our conclusion about the limited usefulness of this examination for reducing prostate cancer morbidity and mortality rates does not preclude any benefit that may be derived from using this test for other indications. However, the net value of the digital rectal examination for these other purposes, particularly for colorectal cancer screening [70], is also poorly substantiated. Nevertheless, digital rectal examination has traditionally been part of the comprehensive physical examination and may be an expected part of preventive health care for many older men. As a result, we suspect that many physicians will focus on PSA measurement rather than digital rectal examination in discussions with patients about screening for prostate cancer. However, discovery of abnormalities of the prostate in the course of such evaluations often triggers a cascade of additional evaluations, even if the physician and the patient were not specifically interested in screening for prostate cancer.

Although this "informed consent" approach to office-based screening for prostate cancer is ethically responsible [71], it also poses considerable logistical difficulties. Busy primary care clinicians must handle many medical problems and preventive concerns, each of which competes for time during a patient visit [72]. Various strategies, ranging from simple handouts to videotapes, may be used to improve the efficiency of the discussions about early detection of prostate cancer [73, 74].

We did not formally model the screening decision for men who are at a higher lifetime risk for prostate cancer on the basis of race or family history. We believe that such men should be informed by their physicians that they face a higher risk for developing clinically apparent prostate cancer [1]. However, they should also be informed that whether screening reduces morbidity or mortality rates has not been proven in controlled studies and that men at higher risk for prostate cancer face the same risk from treatment as other men.

Available evidence does not justify the common but arbitrary policy of annual digital rectal examination and PSA measurement for men who are older than 50 years of age. Insufficient data are available to formally model serial screening or to define an optimal screening interval. Physicians need to negotiate, on the basis of scant evidence, the need for follow-up testing with patients who decide to have initial screening.

Drs. Barry and Mulley: Medical Practices Evaluation Center, Massachusetts General Hospital, 50 Staniford Street, Boston, MA 02114.

Dr. Fleming: Health Outcomes Associates, 602 Northeast 134th Street, Vancouver, WA 98685.

Dr. Fahs: Department of Community Medicine, Box 1043, Mount Sinai School of Medicine, One Gustave Levy Place, New York, NY 10029.