Patients’ Perceptions of the Value of Current Vision: Title and subTitle BreakAssessment of Preference Values Among Patients With Subfoveal ChoroidalNeovascularization—The Submacular Surgery Trials Vision Preference ValueScale: SST Report No. 6 FREE

Choroidal neovascularization (CNV) is the main cause of severe visionloss in patients who have age-related macular degeneration (AMD) and is themost common cause of irreversible central vision loss in older persons livingin the United States.^{1 - 4} Choroidalneovascularization also may cause vision loss in patients who have the ocularhistoplasmosis syndrome or an unknown cause (idiopathic CNV).^{5 - 6} Visionloss due to CNV is an important public health problem.⁷

Previous studies have demonstrated that vision loss can have a profoundeffect on how people view their health-related quality of life.^{8 - 13} However,relatively little work has been done to measure the impact of vision lossdue to CNV on health-related quality of life. One study demonstrated thatpoorer visual acuity in patients with recurrent subfoveal CNV from AMD wasassociated with somewhat poorer scores on the Physical and Mental ComponentSummary scales of the Short Form-36 health survey (SF-36), a generic instrumentfor assessing health-related quality of life.¹² Anotherstudy found that decreased visual acuity in patients with AMD was associatedwith a substantial decrease in the utility or preference values that patientsassigned to their health and vision status,¹⁴ wherepreference values refer to how people feel about the desirability of a healthstate and utility values are preference values measured using methods consistentwith utility theory.¹⁵ Both of these studieshad relatively few participants, thereby limiting the ability to examine clinicaland sociodemographic factors that may influence a patient’s view ofthe impact of vision loss due to CNV on health-related quality of life. Sincequality of life considerations are extremely important in the management ofCNV, clinicians and investigators need a better understanding of how CNV affectshealth-related quality of life, especially as new treatments emerge.

Therefore, to improve understanding and awareness of the impact of subfovealCNV on health-related quality of life, we analyzed data collected from a largepopulation of patients with subfoveal CNV in at least 1 eye to measure thepreference values that patients with subfoveal CNV assigned to their healthand vision status.

We conducted a cross-sectional analysis of the baseline preference valuesand characteristics of participants in the Submacular Surgery Trials (SST),a set of 3 randomized controlled trials among 27 clinical centers designedto evaluate the surgical removal of subfoveal CNV.¹² Patientseligible for the SST had either: (1) subfoveal CNV due to AMD with blood accounting for at least 50% of the total lesion (group B); (2) new (not recurrent) subfoveal CNV due to AMD with bloodaccounting for less than 50% of the lesion (group N); or (3) subfoveal CNVdue to the ocular histoplasmosis syndrome or an unknown cause (group H). Otherprincipal eligibility criteria are summarized in Table 1. For each group, patients were randomized in a 1-to-1 ratiobetween surgery and observation, after completing the consent process andsigning a consent form approved by the institutional review board of eachcenter.

We developed a vision-related preference value scale for the SST, theSubmacular Surgery Trials Vision Preference Value Scale (SST-VPVS), usinga rating-scale technique to measure the preference values that patients assignto their health and vision status (Table 2).The rating-scale technique was used because it has good test-retest reliabilityin most studies and generally has better internal consistency than other scalingtechniques, such as the standard-gamble and time-trade-off techniques.^{8 ,15 - 16} In addition, the rating-scaletechnique can be expected to yield higher response and completion rates thanthe other techniques because it is easier for most people to understand. Ityields reliable and valid results even without an in-person interview or visualaids.¹⁶ These considerations were importantin this study because they allowed us to administer the instrument centrallyby telephone without requiring visual aids that might not be seen by patientswith subfoveal CNV in the second eye after loss of central vision in the firsteye. We did not perform an evaluation of the test-retest reliability of therating scale in our study population, because we did not have any reason toexpect that reliability would be less in this study than in other studiesin which reliability had been established.

Telephone interviewers at the SST Coordinating Center first asked patientsby telephone to rate their current vision using both eyes on a rating scalefrom 0 (completely blind) to 100 (perfect vision). As was done in previouswork,¹⁰ complete blindness and perfect visionwere chosen as the scale anchors because it was judged to be easier for anindividual to rate a visual state on a scale defined in terms of vision thanon a scale defined in other terms, such as from death to perfect health. Theinterviewers then asked the patients to rate what they imagined it would belike to be completely blind in both eyes (with otherwise the same health),using a scale from 0 (death) to 100 (perfect health). Finally, study patientswere asked to rate what it would be like to have perfect vision without anyother change in their health, again using a scale from 0 (death) to 100 (perfecthealth). These 3 questions permitted conversion of the first rating on thevision-only scale to preference values on a scale from death to perfect healthand vision (Figure 1). The responseswere converted to a scale from 0 to 1 to be consistent with the manner inwhich preference values typically are reported.^{8 ,13} Forexample, as shown in Figure 1, if apatient rated current vision as 80 on the vision scale and rated completeblindness (with otherwise same health) as 20 and perfect vision (with otherwisesame health) as 70 on the death to perfect health scale, the final preferencevalue for the current vision and health state would be 0.60 on the final 0to 1 scale from death to perfect health. By asking patients to assign a preferencevalue to perfect vision with their otherwise same current health, we wereable to adjust for the perceived effects of coexisting health problems onratings of their overall health state.

Patients completed interviews about their health-related quality oflife prior to enrollment and random treatment assignment in the SST. Becausemany SST participants would not be able to read a printed questionnaire becauseof vision loss in both eyes, interviews were administered by telephone bytrained interviewers located at the SST Coordinating Center. At follow-upvisits, the SST participants could choose to have an interviewer call themat home or they could call the Coordinating Center from their own homes orfrom the clinic sites where they had their clinical examinations. The interviewersused a computer-generated script and entered responses directly into computerfiles.

The interviews included 3 validated instruments for assessing genericand specific aspects of health-related quality of life in addition to thequestions used to determine the preference value that patients assigned totheir health and vision status: (1) the SF-36¹⁷ ;(2) the 39-item National Eye Institute Visual Function Questionnaire (NEI-VFQ)¹⁸ ; and (3) the Hospital Anxiety and Depression Scale(HADS).¹⁹ In the interviews, the SF-36 immediatelypreceded the HADS and the NEI-VFQ immediately preceded the questions usedto determine the preference value, but a computer program randomly selectedeither the SF-36 or NEI-VFQ to be administered first. After October 9, 1997,all patients enrolled in the SST were asked the preference value questionsin addition to the other health-related quality-of-life questions. On average,it took patients about 26 to 27 minutes to answer all of the questions aboutpreference values and quality of life.

The SST Patient-Centered Outcomes Subcommittee (PCOS) met regularlywith the interviewers to discuss problems encountered during the interviews.Because the interviewers found that some patients could not answer the preferencevalue questions, the PCOS made minor simplifications on 2 occasions to 3 questionsused to determine the SST vision preference value (Table 2). Previous studies have found that up to 30% of the generalpublic has difficulty understanding scaling techniques for measuring preferencevalues.^{15 ,20} The scaling techniquestypically are unfamiliar to people and often require visual aids to help peopleunderstand the questions and scales.

As part of the interview on the health-related quality of life, interviewersasked patients to rate their health in general on a scale of excellent, verygood, good, fair, or poor (SF-36) as well as on a scale from 0 to 10 where0 represented death and 10 was best possible health (NEI-VFQ). The interviewerssimilarly asked patients to rate their eyesight using both eyes as excellent,good, fair, poor, very poor, or completely blind (NEI-VFQ) and on a scalefrom 0 to 10 where 0 represented blind and 10 was best possible vision (NEI-VFQ).

The SST database included data on the sociodemographic and clinicalcharacteristics of enrolled patients,¹² includingage, gender, self-reported race/ethnicity, occupational status, SST group(H, N, or B), specific features of the CNV, treatment assignment (surgeryor observation), right or left eye involvement, clinic site, and best-correctedvisual acuity in each eye.

We summarized the distribution of responses to each of the preferencevalue scale questions by determining means, medians, and interquartile (IQ)ranges. Standard box and whisker plots were created to provide graphic summariesof the distributions. To assess differences in patient characteristics betweenpatients with and without usable preference values, we used the t test, Wilcoxon rank sum test, Kruskal-Wallis analysis of variance,χ² test for homogeneity, or Fisher exact test, as appropriate.To examine the strength of the relation between preference values and patients’self-rated perception of overall health or self-reported perception of vision,we used Spearman rank correlation analysis. To determine whether preferencevalues differed by selected patient characteristics, we used Kruskal-Wallisanalysis of variance and sequential multivariate linear regression. P values <.05 were deemed to indicate any statisticalsignificance; P values <.01 were deemed to indicatemoderate significance; P values <.001 were deemedto indicate high significance. No adjustment to P valueswas made for multiple comparisons. The analyses were performed using SAS softwareversion 8 (SAS, Inc, Cary, NC).

A total of 1015 patients enrolled in the SST between April 1997 andSeptember 2001. Nineteen of these patients (2%) enrolled before the SST-VPVSwas included in the baseline interview. Table3 summarizes the characteristics of the 996 patients who were askedthe preference value questions.

Of the 996 patients questioned, 145 (15%) were unable to answer 1 ormore of the 3 preference value questions. In addition, 59 patients (6%) assigneda value to complete blindness that was equal to or greater than the valuethey assigned to perfect vision, suggesting that they did not understand thequestions. As shown in Table 3, we foundsignificant, albeit small, differences in the demographic and clinical characteristicsof those who did and those who did not give usable answers to the preferencevalue questions, in that those who were not able to answer the questions wereslightly older, more often were retired, had somewhat worse vision in thebetter eye, more often had CNV in the fellow eye, and had, on average, lowerPhysical Component Summary scores and higher Mental Component Summary scoreson the SF-36.

When comparing responses according to the version of the questions thatpatients were asked (Table 2), we founda trend but did not find a significant difference (P>.10)in the percentage unable to answer the questions (17% of 223 for the firstversion, 14% of 647 for the second version, and 12% of 126 for the third version).We also found a trend but did not find a statistically significant trend inthe percentage that rated complete blindness as equal to or better than perfectvision (8% for the first version, 6% for the second version, and 2% for thethird version).

The main analysis was based on the responses of the 792 patients whogave usable answers to the preference questions that permitted calculationof SST-VPVS scores. When they were asked to rate their current vision on ascale (Figure 1) where 0 correspondsto complete blindness and 100 corresponds to perfect vision, they gave a meanrating of 55 (median, 50; IQ range, 50-75). On a scale where 0 correspondsto death and 100 corresponds to perfect health with perfect vision, the patientsgave a mean rating of 35 to indicate how they imagined they would feel ifthey were completely blind in both eyes with their same health otherwise (median,30; IQ range, 10-50). On the same scale, they gave a mean rating of 90 toindicate how they imagined they would feel if they had perfect vision withtheir same health otherwise (median, 95; IQ range, 80-100). We then calculatedthat the mean preference value assigned to current vision and health withCNV was 0.64 (median, 0.68; IQ range, 0.51-0.80) on the scale where 0 correspondsto death and 1 corresponds to perfect health with perfect vision.

The calculated preference values assigned to current vision and healthwith CNV differed between the groups that received different versions of thequestions: (1) mean, 0.61 and IQ range, 0.48 to 0.76 with the first version(168 patients); (2) mean, 0.66 and IQ range, 0.55 to 0.81 with the secondversion (516 patients); and (3) mean, 0.63 and IQ range, 0.47 to 0.78 withthe third version (108 patients) (P<.01 for thedifference across the 3 versions by Kruskal-Wallis test). Because the differencesbetween means and IQ ranges were small despite a P valueless than. 01, when the 3 sets of values were compared, the presentation ofthe rest of the analysis combines the data from all 792 patients who gaveusable answers to the preference value questions. The calculated preferencevalues were similar when we limited the analysis to the 516 patients who gaveusable answers to the second version of the questions (mean, 0.66; median,0.70; IQ range, 0.55-0.81).

As shown in Figure 2, the calculatedpreference values for current vision and health with CNV were moderately wellcorrelated with patients’ ratings of their eyesight as excellent, good,fair, poor, very poor, or completely blind (Spearman correlation coefficient,0.38; P<.001). The preference values for currentvision and health with CNV also were correlated with patients’ ratingsof their eyesight on a 0 to 10 scale (Spearman correlation coefficient, 0.47; P<.001), patients’ ratings of their overall healthon a 0 to 10 scale (Spearman correlation coefficient, 0.36; P<.001), and patients’ ratings of their overall health asexcellent, very good, good, fair, or poor (Spearman correlation coefficient,0.28; P<.001).

As shown in Table 4, the calculatedpreference values for current vision and health differed modestly among the3 trial groups. Group H patients, who had the youngest mean age and the bestvisual acuity scores, had the highest mean preference values. The preferencevalues differed somewhat by employment status and patient age but did notdiffer (P>.05) by gender or race/ethnicity. The preferencevalues differed significantly by visual acuity in the better-seeing eye orworse-seeing eye and differed somewhat by specific features of the CNV lesion.Patients with a visual acuity of 20/200 or worse in both eyes had a mean preferencevalue of 0.53 compared with a mean preference value of at least 0.67 if 1eye had a visual acuity of 20/40 or better. In addition, the preference valueswere lower when the HADS indicated the presence of anxiety or depression orwhen the Mental or Physical Component Summary scores of the SF-36 were lower.

In a multivariate analysis (Figure 3)(Table 5) of the relation of demographiccharacteristics (age, gender, race/ethnicity, and employment status) to thepreference values for current vision and health with CNV, we found that preferencevalues were independently associated only with being employed (P<.001). In a multivariate analysis of the relation of clinicalfeatures of CNV to the preference values (Figure3) (Table 5), we found thatpreference values were independently associated only with visual acuity ofthe better eye (P<.001). When employment statusand visual acuity of the better eye were both considered (Figure 3) (Table 5), onlythe latter was significantly associated with the preference values. When comorbidconditions were considered in the model, visual acuity of the better eye remainedsignificant (P<.001) and only hypertension wasassociated with a decrease in preference values (P = .02).When the HADS classification of depression was added to the model, all 3 variableswere independently associated with the preference values: vision in the bettereye (P<.001), hypertension (P = .049), and HADS depression classification (P<.001), with an overall R² of0.18. When the SF-36 Physical Component and Mental Component Summary scoreswere considered instead of the HADS, the only significant variables in thefinal model were visual acuity of the better eye (P<.001),Physical Component Summary score (P<.001) andMental Component Summary score (P<.001), withan overall R² of 0.20. When the NEI-VFQscore was considered instead of the HADS scores or the SF-36 scores, onlythe NEI-VFQ score was significantly associated with the preference values(P<.001; R² = 0.26).

As one of the largest studies of preference values in patients witheye disease, this investigation has shown convincingly that patients withsubfoveal CNV in one eye, with or without a neovascular lesion in the felloweye, assign low preference values to their health status. A mean preferenceof 0.64 on a scale from 0 (dead) to 1 (perfect health) indicates that subfovealCNV, on average, has a profound impact on how patients feel about their qualityof life. The impact is greatest in those with the most severe loss of vision,but even patients with visual acuity of at least 20/40 in 1 eye had relativelylow preference values. As presented in Table 6 for comparison, previous studies have reported a mean preferencevalue of 0.77 for congestive heart failure,²¹ 0.65for symptomatic human immunodeficiency virus infection or AIDS,¹³ 0.54for chronic renal failure requiring home dialysis,²² 0.50to 0.70 for minor stroke,²³ and 0 to 0.30 formajor stroke.^{13 ,23} The relativelylow mean preference value assigned to subfoveal CNV would suggest that patientswith CNV would be more willing to consider a risky or expensive treatmentthan patients with an ocular condition associated with a higher preferencevalue.

These results are consistent with the pathophysiologic features andnatural history of CNV. It is often a progressive and generally irreversibledisorder that frequently causes severe loss of central vision.^{1 - 3 ,24 - 28} Thus,it is not surprising that patients with central vision loss from CNV in atleast 1 eye report that their vision loss markedly decreased their health-relatedquality of life¹² and the preference valuesthey assigned to their overall health. Moreover, the impact of CNV on health-relatedquality of life and preference values may be worsened by uncertainty aboutthe efficacy of treatment. Chronic impairments typically are associated withlower preference values than temporary impairments, even if patients are ableto adjust to their chronic impairments.^{5 - 6}

The results also are consistent with previous reports on the preferenceor utility values that patients assign to visual impairment (Table 6). Brown et al,¹⁴ in a studyof 80 patients with AMD, reported a mean patient preference value of 0.72using the time-trade-off technique and a mean utility value of 0.81 usingthe standard-gamble technique. The preference or utility values reported byBrown et al may be higher than the values reported in our study because, asempirical work has shown, the time-trade-off and standard-gamble techniquestypically yield higher values than the rating-scale technique we used.¹⁵ We used the rating-scale technique because it iseasier for patients to understand,¹⁵ less dependenton patients’ education, and could be adapted to a telephone interviewrather than require an in-person interview, thereby contributing to the SST’sability to collect such data from a large population of patients with CNV.For this visually impaired population, we also thought it was important touse a scaling technique that did not depend on use of visual aids. In comparison,another relatively small study examined the preference values of 47 patientswith cataract who were scheduled to have cataract surgery. In that study,the mean preference value for preoperative vision was 0.68 using a rating-scaletechnique.¹⁰ Of note, the median Snellen visualacuity in the eye scheduled for surgery was 20/70, better than the worse-eyemedian visual acuity in the current study (20/250). However, in that study,the preference values were not correlated very well with measures of visualacuity, suggesting that measures of visual acuity do not adequately predicthow patients feel about the impact of their visual impairment. Studies alsohave reported that patients would assign an extremely low preference valueto complete blindness, ranging from 0.33 to 0.39.^{10 ,22} Althoughit is important to emphasize that preference values may differ markedly dependingon the scaling technique that is used, the results of our study and previousstudies are consistent in demonstrating that patients with visual impairmenthave a marked decrease in the preference values that they assign to theirhealth status, indicating a profound impact on how patients feel about theirquality of life.

This study was large enough to permit determination of the sociodemographicand clinical characteristics of patients that were associated with a differencein the SST-VPVS scores that patients assigned to their current health withCNV. As expected, visual acuity in the better eye was strongly associatedwith the preference values. Also, the preference values were higher in groupH patients (ocular histoplasmosis syndrome or idiopathic CNV) than in groupsB (CNV with blood) and N (new CNV). However, the data suggest that this mostlikely was due to the lesser effects of CNV on visual acuity at baseline andmore frequent unilateral involvement in group H rather than the younger ageof group H specifically. Group H patients tended to have smaller CNV lesionsand better visual acuity in both the study eye and fellow eye, as comparedwith patients in groups B and N. We did not have enough information on theextent of CNV in the fellow eye to assess whether preference values differedaccording to the severity of CNV in the fellow eye. As presented in Table 4, age was significantly associated withthe preference values. We could not determine whether the preference valuesdiffered by race/ethnicity because so few of the patients were nonwhite.

Older patients with CNV may be particularly vulnerable to the effectsof vision loss on their health-related quality of life because of coexistingdiseases and impairments. This study found that older age by itself was associatedwith a significant decrease in preference values, although in a multivariateanalysis, this association no longer was significant when employment was addedto the multivariate model. The presence of hypertension was associated withdecreased preference values in the multivariate model, likely because hypertensionitself is associated with other complications and the need for medicationsthat may affect quality of life. The SST-VPVS scores for current health withCNV were lower when patients had evidence of anxiety or depression (as measuredby the HADS)¹⁹ or a decrease in mental or physicalfunctioning (as measured by the Mental or Physical Component Summary scoresof the SF-36).¹⁷ In our multivariate analysis,the preference values were independently associated with vision in the bettereye, the Physical Component Summary score, and the Mental Component Summaryscore. The relatively low R² value ofthat multivariate model (0.20) indicates that it is very difficult to predicthow any given patient will feel about the effects of CNV on quality of life.

One limitation of this study is that we used a rating-scale techniqueto measure preference values without comparison with another scaling technique.Some “utility” experts contend that the standard-gamble techniqueis the gold standard for measuring patient utility values,^{8 ,16} butas mentioned previously, use of the standard-gamble technique has been limitedby difficulty in administering it without an in-person interview and accompanyingvisual aids. In populations like ours that include patients with very poorvision, the use of visual aids is not feasible. The time-trade-off techniqueoften is used instead of the standard-gamble technique, but it also is difficultto administer without visual aids. Perhaps more importantly, all 3 scalingtechniques have been shown to have reasonable reliability and validity; approximatelyone third of studies on patient preference assessment have used a rating-scaletechnique.¹³ By using the simpler rating-scaletechnique, we were able to take advantage of a unique opportunity to incorporatean assessment of patient preference values into large multicenter trials.

Approximately 20% of the patients in the SST had difficulty answeringthe rating-scale questions at the baseline interview. This is not surprisingsince previous studies often have found that 10% to 30% of study subjectshave difficulty answering questions that involve quantification of their preferencevalues.^{15 ,20} While Brown et al²⁹ have reported low rates of difficulty in studiesof preference values in ophthalmology patients, it is unclear whether thatwas owing to use of an in-person interviewer or unique aspects of the studypopulations. It is possible that responses were biased by a loss of concentrationrelated to the length of the questions about preference values and qualityof life, but we suspect this would have been a more serious problem if wehad used a more difficult scaling technique. Our modification of the wordingof the questions did not appear to increase understanding of the questionssignificantly, although we saw a slight trend toward a decrease in the percentageof patients unable to answer the questions. The trend suggests that it wouldbe appropriate to use the final version of the questions in future studies.However, we emphasize that estimates of patients’ preference valuesin the SST were similar regardless of the minor differences in wording ofthe questions. An even higher percentage of patients might not have been ableto give appropriate answers to questions based on the standard-gamble techniquebecause that technique is known to be particularly difficult for people tounderstand. It is not clear why patients with slightly better scores on theMental Component Summary of the SF-36 were less likely to give usable answers,but the observed difference in Mental Component Summary scores was so small(median, 55.8 for those with usable answers vs 56.4 for those without usableanswers) that it is unlikely to be clinically significant. Additionally, wenote that patients with severe cognitive impairment would not have been ableto give informed consent to participate in the trials. After excluding patientswho gave answers to the 3 preference questions that were not consistent witheach other, moderate correlation existed between the preference values andmeasures of visual acuity as well as measures of health-related quality oflife. The correlation data indicate that the questions had reasonable constructvalidity.

Of course, our results may not apply to all patients with CNV sincethe study population enrolled in the SST included very few nonwhite patients,consistent with the epidemiological factors.³⁰ However,the study population did include patients with different types of CNV.

We conclude that subfoveal CNV has a profound impact on how patientsfeel about their quality of life. The data provided by the SST-VPVS lend supportto the importance of giving attention to the needs of such patients for visualrehabilitation and mental health services that could minimize the impact ofthe disease on their quality of life while remaining alert to the possibilityof coexisting depression or mental health problems that would further impaira patient’s quality of life. The data also provide further justificationfor pursuit of more effective prevention and treatment options for patientswith subfoveal CNV. Even relatively costly medical or surgical treatmentsfor CNV could prove to be cost-effective from the patient’s point ofview, especially if cost-effectiveness is measured in terms of the recommendedrubric of dollars per quality-adjusted life year gained.³¹

The SST are poised to make a unique contribution to the field by virtueof having included patients’ preference values as an outcome that wasmeasured across time. These data will allow the investigators to analyze changesacross time relative to changes in visual acuity and progression of the ocularcondition. The data also will enable the investigators to analyze the cost-effectivenessof submacular surgery, if it is shown to be efficacious, taking into considerationthe substantial effects of CNV on quality of life. The SST-VPVS also willallow investigators to analyze the cost-effectiveness of other treatmentsfor CNV, again taking into consideration the substantial effects of CNV onquality of life.

Correspondence: Eric B. Bass, MD, MPH, GeneralInternal Medicine, 1830 E Monument St, Room 8068, Baltimore, MD 21287 (ebass@jhmi.edu).

Submitted for Publication: November 13, 2003;final revision received August 16, 2004; accepted August 16, 2004.

Submacular Surgery Trials Patient Centered OutcomesSubcommittee and Writing Committee: Eric B. Bass MD, MPH (Patient CenteredOutcomes Subcommittee and Writing Committee co-chair); Marta J.Marsh MS (WritingCommittee co-chair); Carol M.Mangione MD, MSPH (Patient Centered OutcomesSubcommittee chair); Neil M. Bressler MD; Ashley L. Childs MS; Li Ming Dong PhD; Barbara S. Hawkins PhD; Harris A. Jaffee PhD;and Päivi H. Miskala PhD.

Financial Disclosure: None.

Funding/Support: The Submacular Surgery Trialsare supported by cooperative agreements U10 EY11547, EY11557, and EY11558between the National Eye Institute, National Institutes of Health, Departmentof Health and Human Services; Bethesda, Md, and The Johns Hopkins University,Baltimore, Md. Participating clinical centers are supported by contracts withThe Johns Hopkins University. Participation of Dr Mangione is supported bya contract between The David Geffen School of Medicine, University of California,Los Angeles, and The Johns Hopkins University.

Acknowledgment: We thank the interviewers andpatients for their fine efforts in making this study possible.