Differences in Generalists' and Cardiologists' Perceptions of Cardiovascular Risk and the Outcomes of Preventive Therapy in Cardiovascular Disease

Methods

Questionnaire

The questionnaire consisted of written case simulations of patients with hypercholesterolemia, isolated systolic hypertension, and chronic stable angina with left main coronary stenosis whose clinical characteristics were representative of the average person in well-publicized clinical trials (Table 6). The hypercholesterolemia case simulation (or scenario) was extrapolated from the Lipid Research Clinics and Helsinki trials [11, 12]; physicians were asked to give unprompted estimates of 5-year risk for myocardial infarction with and without lipid-lowering drug therapy. In the section of the questionnaire dealing with the isolated systolic hypertension scenario, which was derived from the Systolic Hypertension in the Elderly Program trial [13], physicians were asked to give estimates of 5-year risk for stroke with and without antihypertensive drug therapy. In the section that provided the left main coronary stenosis scenario, which was extrapolated from the Veterans Administration Cooperative Study of Surgery for Coronary Arterial Occlusive Disease and the Collaborative Study in Coronary Artery Surgery [14, 15], physicians were asked to estimate 3-year survival with and without coronary artery bypass graft surgery. The “correct” answers were taken from the results of the studies from which each scenario was derived. Respondents were asked not to consult the literature when completing the questionnaire, but simply to give their best estimates.

The physicians were also asked whether they would generally recommend therapy for such a patient (yes or no). They rated their confidence in their estimates and their familiarity with the relevant medical literature on 7-point Likert-type scales (for example, 1 equals not at all confident; 7 equals very confident). Finally, respondents reported their age, sex, practice setting, specialty (cardiology, general internal medicine, or family practice), and the year in which they graduated from medical school.

Sample Selection and Survey Procedure

Sampling was done in two stages. First, an addressing company contracted by the American Medical Association (AMA) drew a self-weighted, nationally representative, systematic sample of 6000 practicing physicians (2417 family physicians, 2901 general internists, and 682 cardiologists) from the AMA masterfile of all licensed physicians. Second, after dividing the first-stage sample into the three specialty groups, we selected three separate simple random samples of 199 family physicians, 200 general internists, and 200 cardiologists to form our mailed sample (Table 1). For the cholesterol scenario, using a two-sample t-test, a standard deviation of 8 percentage points from a pilot study, and a 50% response rate, 200 physicians in each group would have given us more than 90% power (α equals 0.05) to detect a difference of 4 percentage points between the groups' estimates. We mailed the two-page survey three times between September 1993 and February 1994, and a reminder postcard was sent 2 weeks after the second mailing.

Analysis

Self-reported specialty was compared with the specialty category in the AMA masterfile, with a percent agreement of 91%. Because the AMA specialty category was used for demographic comparison with the nonrespondents and overall population, we used the specialty designation from the AMA masterfile. The sex of all nonrespondents was inferred on the basis of first name when possible, before the names were deleted. Sex could not be determined in 2% of instances. Characteristics of the AMA masterfile population were estimated from 1993 year-end data [16].

Absolute risk reduction was calculated as the difference between estimated risk at baseline and after therapy. Relative risk reduction was the absolute risk difference divided by the baseline risk. “Accuracy” was defined as the absolute value of the difference between the physician's estimate and the estimate from the literature, with lower numbers indicating greater accuracy. Because the estimates and accuracy were not normally distributed, we present median estimates with interquartile ranges (the range between the 25th and 75th percentiles). Across specialty groups, we generated differences in medians with confidence intervals, and used the Mann-Whitney U test [17]. Similarly, within each specialty group, we compared the differences in medians between those who would and those who would not recommend treatment.

To compare respondents' and nonrespondents' sex and geographic region, we used the Cochran-Mantel-Haenszel chi-square statistic for contingency tables stratified by specialty [18]. For analyses of categorical variables across specialty groups, we did chi-square tests and calculated confidence intervals. Across specialty groups, we also used analysis of variance to generate differences in means and confidence intervals for comparisons of 1) confidence in estimates and 2) familiarity with the relevant literature [both of which were measured on 7-point Likert-type scales], as well as 3) years since graduation from medical school. The relation between these three variables and accuracy was assessed using the Spearman rank correlation coefficient, with confidence intervals generated using the Fisher Z transformation. For clarity, 95% CIs are given throughout. Because of multiple comparisons, we defined statistical significance at the P ≤ 0.005 level (two-tailed) and noted trends at the P ≤ 0.05 level.

Results

Respondents and Nonrespondents

We received responses from 232 (40%) of 579 eligible physicians. Of the 20 ineligible physicians, 15 did not have a forwarding address and 5 were retired or deceased. Another 5 surveys were illegible, leaving 227 for analysis. Thus, 84 (44%) of 191 eligible family physicians, 77 (40%) of 194 general internists, and 66 (34%) of 194 cardiologists responded (Table 1). The available demographic characteristics of respondents were reasonably similar to those of nonrespondents and of the AMA masterfile population (Table 1).

Estimates and Treatment Recommendations

For both primary prevention scenarios, cardiologists gave lower, more accurate estimates of baseline risk and absolute risk reduction than either general internists or family physicians, although these estimates were inflated and widely distributed for all three specialties (Table 2). For example, 27% of general internists and more than 30% of family physicians estimated that the 5-year absolute risk reduction for myocardial infarction in the cholesterol scenario was greater than 10%, a 10-fold overestimate, compared with only 14% of cardiologists. Similarly, for the systolic hypertension scenario, half of the family physicians and general internists gave two- to threefold overestimates of the 5-year risks for stroke both with and without antihypertensive therapy, whereas most of the cardiologists better approximated the estimate from the literature. Despite differing perceptions of baseline risk and absolute risk reduction, the specialties did not differ significantly as to estimates of relative risk reduction for the primary prevention scenarios.

For the left main coronary stenosis case, cardiologists gave lower estimates of baseline survival and greater estimates of its improvement with surgery Table 2 than general internists and family physicians, but no differences in accuracy (the absolute value of the difference between the physician's estimate and the estimate from the literature) were seen across the specialties. For example, for the absolute improvement in survival after bypass surgery, no differences in median accuracy were noted between cardiologists and either family physicians (point estimate for the difference in medians, 0; 95% CI, − 1 to 5) or general internists (point estimate, 0; CI, − 5 to 3). Nonetheless, cardiologists were twice as likely to recommend surgical revascularization appropriately (Table 3). For the primary prevention scenarios, treatment recommendations did not differ among the specialties.

For all of the scenarios, generalists' decisions to recommend treatment appeared to be associated with greater perceived absolute therapeutic benefit (Table 4). In addition, generalists who would not have recommended primary preventive treatment tended to provide more accurate estimates than did those who would have recommended these therapies. Conversely, for the left main stenosis scenario, physicians who would have recommended bypass surgery provided more accurate estimates.

Confidence in Estimates, Familiarity with the Literature, and Years Since Medical School Graduation

Compared with generalists, cardiologists were more confident in their estimates and perceived themselves as more familiar with the relevant literature for the left main coronary stenosis scenario but not for the primary prevention scenarios (Table 5). However, confidence in their estimates and self-perceived familiarity with the literature were not associated with greater accuracy of estimates for any of the scenarios (data available on request). Time since medical school graduation did not differ between the cardiologists (mean ±SD, 20 ± 11 years) and either the family physicians (difference in means, 0.1 years; CI, − 4 to 4) or the general internists (difference in means, 2 years; CI, − 2 to 6), and was also not correlated with the accuracy of estimates (data available on request).

Discussion

We found that many physicians harbor inflated and widely varying perceptions of cardiovascular risk without treatment and are overly optimistic about the benefit of primary preventive therapy. In particular, although generalists are more likely to be responsible for primary prevention, general internists and family physicians overestimated baseline cardiovascular risk and the benefit of risk-modifying therapy to a greater extent than did cardiologists.

Several possible explanations can be given for the cardiologists' greater accuracy. First, because their knowledge base is narrower than that of generalists, cardiologists may learn and retain more details from the cardiovascular literature. Conversely, in their attempt to maintain a broad knowledge base, generalists may have forgotten the numbers or may never have learned them. They may recall only that a therapy was “beneficial” or “effective.” Over time, these quasi-quantitative expressions may be unconsciously transformed into a wide range of numerical interpretations [19]. Second, research training during fellowship may render cardiologists more facile with quantitative information [20]. Third, because most cardiologists' patients have atherosclerosis and risk factors, cardiologists may perceive these cases as “failures” of prevention, and maintain less optimism about risk factor modification. Finally, generalists may have greater psychological investment in primary prevention because it is one of the defining characteristics of primary care practice. Ego bias may thus have contributed to the generalists' overestimates of both baseline risks and the benefits of therapy for the primary prevention scenarios [21].

However, despite their lower expectations about the efficacy of primary preventive therapy, cardiologists were as inclined as generalists to recommend such therapy. This finding may imply that cardiologists are more disposed to treat to achieve small benefits, which could be related to the observation that specialists provide greater intensity of care than generalists, even adjusting for case-mix [22, 23]. Consistent with the findings of a recent study of specialty care practices [4], cardiologists were also much more likely to recommend coronary bypass surgery appropriately to a patient with left main coronary stenosis.

In addition, we observed a trend in which those generalists who would not have recommended primary preventive therapy gave more accurate estimates of absolute risk reduction than those who would have recommended treatment. This finding implies that some physicians' threshold for what constitutes a worthwhile benefit of therapy may be higher than that shown in clinical trials. A recent study of oncologists' perceptions of the efficacy of adjuvant chemotherapy for breast cancer also suggested that the threshold of benefit at which they would recommend treatment was higher than the amount of benefit shown in previously reported trials [24]. These results may indicate that if some physicians accurately perceived the benefits of primary preventive therapy, they would recommend treatment to individual persons less often. This supposition may apply to patients' acceptance of these therapies as well [5].

Previous studies have shown that some physicians overestimate the probability of certain diseases or outcomes, such as ischemic heart disease, pneumonia, streptococcal pharyngitis, bacteremia, death or rebleeding in patients with previous gastrointestinal bleeding, and complications after invasive diagnostic procedures [25-30]. The literature on cognitive bias provides intriguing explanations for these overestimates [21, 25-33], but further research is needed to determine whether generalists are more subject to these errors than specialists.

Ideally, experience and continuing education should produce more accurate perceptions of risk and benefit. However, the lack of association that we found between years since graduation from medical school and accuracy supports previous observations that experience does not improve these perceptions [21, 30]. Several forces may account for this phenomenon. First, beliefs acquired early in training may be difficult to extinguish, and physicians may process only that information that confirms their opinions [21]. Second, when clinical guidelines or “authoritative” trials recommend therapy, some physicians may feel exempt from learning the quantitative basis for the recommendations. Indeed, one purpose of practice guidelines is to avail busy clinicians of this shortcut. Third, published studies, media coverage, and pharmaceutical advertising tend to emphasize the benefits and downplay the risks of therapy or the existence of moderating views. The common emphasis on relative risk reduction may also magnify perceptions of benefit [34-37]. Fourth, physicians' bias toward action—that is, the idea that doing something is better than doing nothing—may influence estimates of the treatment effects [38, 39]. Still, it is unclear whether the causal association between enthusiasm for therapy and overestimation of its effect is unidirectional or circular. Further study of this association is needed because it has important implications for continuing medical education and the implementation of practice guidelines. In the meantime, however, we believe that the baseline risk and absolute benefit, or number needed to treat [40], should be emphasized when treatment recommendations appear in published studies, guidelines, or educational materials.

Our study has several limitations. First, the 40% response rate, although not unusual for a mail survey of physicians without telephone contact [41, 42], may have introduced response bias. Demographic balance between the respondents and nonrespondents argues against this bias to some degree. Nonetheless, if poorly informed cardiologists were more likely than poorly informed generalists to not respond, our results could be biased in favor of the cardiologists. Second, written scenarios may not reflect real-life practice [32], although they are probably an effective way to elicit physicians' beliefs and attitudes [43], the main focus of this study. Third, the scenarios were sparse in detail because the relevant clinical trials did not provide outcomes for detailed subgroups. Consequently, respondents may have interpreted the questionnaire variably or may have systematically inferred other factors that would increase cardiovascular risk. Fourth, we did not ask about perceptions of the risk for adverse effects from therapy. Physicians' concerns about adverse effects may have decreased the willingness of physicians to treat and may also have influenced estimates of benefit. Fifth, we did not ask about the benefit of therapy for other outcomes, such as all-cause mortality. We only examined outcomes for which therapeutic efficacy was best shown. This might explain why most respondents would have recommended preventive therapies, whereas other studies have suggested that preventive services are underutilized [44]. Finally, although clinicians' widely ranging responses could reflect diverse findings in clinical trials, the results of other relevant studies approximate those of the trials we chose [45, 46].

If physicians' perceptions of risk and benefit influence the counseling that patients receive, then our results raise concerns about the quality of discussions between patients and physicians about primary preventive therapy. Physicians must provide patients with accurate information about the probability of disease and the magnitude of benefit from an intervention in order for the patient to assess the worth of that intervention. The accuracy of the information that the patient receives is particularly crucial for primary preventive therapies because 1) these therapies often have small individual benefits; 2) these therapies may harm asymptomatic persons who may not necessarily develop disease; and 3) patients' decisions to accept therapy are likely to be sensitive to small changes in expected benefit [5, 47-51]. Future investigations should thus determine the extent to which physicians' perceptions of risk and benefit influence clinical decision making.

Presented at the 17th Annual Meeting of the Society of General Internal Medicine, Washington, D.C., 27 April, 1994.

Dr. Brett: Department of Medicine, University of South Carolina School of Medicine, 2 Richland Park, Suite 502, Columbia, SC 29203.

Dr. Mayo-Smith: Ambulatory Care Services, Manchester Veterans Affairs Medical Center, 718 Smyth Road, Manchester, NH 03104.