Discordance Between Physician and Patient Self-Rated Health and All-Cause Mortality

INTRODUCTION

Community-based cohorts and clinic-based studies show a strong association between participants' responses to a single question about general self-rated health (GSRH) and subsequent need for health expenditures and health services.1-3 Patient responses are also strongly associated with subsequent mortality even when adjusted for an array of covariates.4,5 GSRH has also been demonstrated to improve the predictive power of claims-based and clinical models.3

GSRH has not been widely used for predicting patient outcomes because practitioners have assumed that it adds little value over what an informed physician assessment and other relevant objective data can provide. Although one might believe that GSRH and physician assessment of health would be highly correlated, few studies have addressed this association. We hypothesized that GSRH may add value to physician assessments of health because many of the other important influences on health and outcome of which the patient is inherently conscious (socioeconomic status, emotional health, etc) are not routinely captured by either the health system or the physician. Therefore, we assessed the concordance between patient GSRH and physician assessment of health after adjustment for an array of objective data, as well as the risk for mortality associated with GSRH independent of physician-rated health in a nationally representative sample of US adults.

RESULTS

Overall, 52.2% of participants reported Excellent/Very Good GSRH, and 32.9% and 14.9% of participants reported Good and Fair/Poor GSRH, respectively (Table 1). Health was assessed by the physicians as being Excellent/Very Good, Good, and Fair/Poor for 70.9%, 22.2%, and 6.9% of participants, respectively. Overall participant GSRH and physician-assessed health were concordant for 53.8% of participants (42.1% Excellent/Very Good, 8.7% Good, and 3.0% Fair/Poor). The weighted kappa statistic for agreement between participant GSRH and physician-assessed health was 0.20.

Among participants for whom their physician-assessed health was Excellent/Very Good, 59.4% reported Excellent/Very Good GSRH, while 30.7% and 10.0% reported Good and Fair/Poor GSRH (Figure). For participants whose physician-assessed health was Good, 39.4%, 39.0%, and 21.6% reported Excellent/Very Good, Good, and Fair/Poor GSRH, respectively. The distribution of participant GSRH was 19.8%, 36.7%, and 43.5% (Excellent/Very Good, Good, and Fair/Poor, respectively) among those whose health was assessed as being Fair/Poor by their physicians.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE.

Distribution of participant general self-rated health (GSRH) by level of physician-assessed health.

Within each physician-assessed health grouping, participants with worse GSRH were older, more likely to be non-Hispanic blacks and Mexican-Americans, have less than a high school education, earn an annual household income <$20,000, and be physically inactive (Table 2). Worse GSRH was also associated with higher systolic blood pressure and diabetes.

Over a median of 13 years of follow-up (maximum follow-up: 18 years), 3,460 deaths occurred. The most common cause of death was cardiovascular disease (n  =  1,512; 44%). Worse GSRH was associated with higher mortality rates before and after adjustment for an array of health conditions and physician-assessed health (Table 3). After multivariable adjustment, those with Good and Fair/Poor GSRH and physician-assessed health had 1.35 (95% CI: 1.18-1.55) and 1.74 (95% CI: 1.47-2.03) higher risk for mortality, respectively, compared to their peers whose GSRH was Excellent/Very Good. Patterns of increased hazard ratios for mortality associated with worse GSRH were present across subgroups defined by age, race/ethnicity, and sex as shown in Table 4. Within each physician-assessed health category, worse participant GSRH was associated with a significantly higher multivariable adjusted hazard ratio for mortality (Table 5). Results were consistent for the outcome of cardiovascular mortality (Table 6).

Following multivariable adjustment for age, race, sex, education, income, current and former cigarette smoking, BMI, systolic blood pressure, use of antihypertensive medications, total and HDL cholesterol, diabetes, estimated glomerular filtration rate, microalbuminuria, and elevated CRP, participants who reported better GSRH than the physician assessment of their health experienced lower mortality (hazard ratio  =  0.76, 95% CI: 0.66-0.87). Also, participants reporting worse health than the physician assessment of their health experienced higher mortality (hazard ratio  =  1.45, 95% CI 1.24-1.70).

DISCUSSION

In this analysis of a nationally representative sample of US adults, participant GSRH and physician perceptions of general health were not highly correlated. Participants with worse self-rated health had a significantly higher risk for mortality even after adjusting for physician perception of their health and an array of objective health data. This relationship was present in the overall population and was markedly consistent across age, race/ethnicity, and sex subgroups. Also, participants with self-assessed health worse than that rated by the study physicians experienced significantly higher mortality. These data suggest that GSRH provides important patient-centered information in addition to that readily available to clinicians.

The low level of concordance between GSRH and physician assessments of health observed in the current study is consistent with prior literature. Furthermore, physician-assessed health and GSRH may not always be concordant, especially if physicians do not consider nonclinical psychosocial information such as quality of life, income, and education.10,11 In a group of older patients, Finnish investigators reported significant discordance between individuals' self reports of health and their physicians' assessment.10 Also, in a study of 456 patients attending a primary care clinic, discordance between physician assessment of an individual's health and self-rated health was present in 32% of the patients.11 In a separate study of participants recruited in a primary care setting, overall agreement between patients' reports of their health and objective measures was only 54%.12 The current results are consistent with these previous findings; overall agreement and the agreement beyond that expected by chance (ie, kappa statistic) were low.

Reasons for the discordance between self-rated and physician-assessed health are not entirely clear. Patients with depression, advancing age, and lower socioeconomic status are more likely to have discordance with their physician assessments.10,13 Because patients capture a broad array of dimensions when assessing their own health that may not be incorporated into physician-assessed health, the rates of discordance are not surprising.14,15 Patients inherently know their past medical history, current symptoms (such as masked depression), social circumstances, and family longevity. All can contribute to mortality and can be easily missed by a physician assessment. Even when stratified by physician-assessed health, characteristics of participants in the current study varied widely by GSRH.

Prior studies have reported worse GSRH to be associated with increased mortality even after controlling for objective measurements of health, such as claims-based measures of chronic diseases.3,16-18 In one study, participants who rated their health as being better than expected based on the presence of chronic diseases had improved health outcomes. In contrast, those who reported worse health than expected based on the presence of chronic diseases (termed realists or pessimists) had higher rates of adverse outcomes.12 Although studies have documented that GSRH provides additional information above and beyond objective health information, few data are available on the additional information provided through self-rated versus physician-assessed health.

The added value of GSRH to physician assessments plus clinical data is expected given the broad array of dimensions captured in patients' general health assessments that are potentially not known to the clinician and not routinely part of what is collected in administrative and clinical databases. GSRH is a summative measure accounting for a variety of personal and relevant social characteristics. When considering their own health, individuals are known to consider their overall physical, mental, and social situation.19,20 GSRH is also highly correlated with a number of biomarkers that are associated with morbidity and mortality.21,22

Creating a more patient-centered health system will require systematizing patient perspectives on their health and healthcare. A single question assessing a patient's GSRH performs well in predicting future health states and needs. Because GSRH information takes only seconds to collect, it may be valuable to add to the routine data collected from patients to aid in predictive modeling. Collecting the answer to a single GSRH question could easily be added to routine intake information at primary care appointments for use by administrators. The information could be easily merged with patient age, sex, and race/ethnicity and could be useful for predictive modeling by payers and health system managers. GSRH responses add important information to administrative data for the prediction of future healthcare expenditures.3 Also, adding GSRH to the Framingham Risk Score improves prediction of future coronary heart disease incidence.23 In addition to the administrative value of adding GSRH to prediction models, the information captured through the collection of GSRH answers may prove valuable in guiding the patient into care management programs at the point of care. If GSRH information is collected in an automated fashion in an electronic health record, patients with poor or declining GSRH can be further evaluated for care management, even in the absence of physician referral. Given the discordance seen between patient and physician assessments of health and the significant impact patient GSRH has on outcomes when patients report worse health than their physicians assess, the use of GSRH could dramatically improve the identification of high-risk patients and earlier entry into beneficial care programs.

The findings from the present study should be considered in the context of potential limitations. GSRH is a subjective measure that can vary in response to significant life events whether physical or mental. However, it is a generally reliable measure and at the population level, sufficiently stable.24,25 Although the physician assessments were performed in a rigorous manner consistent with the NHANES protocol, the study physicians were handicapped by a lack of continuity relationship with participants. This limits the inclusion of a longitudinal perspective on participants' health when providing an assessment. Despite these limitations, the current study has many strengths. Most notably, NHANES III data were collected using a rigorous study protocol with extensive quality control procedures. The extensive data collection in NHANES III allowed adjustment for many objective measures of health (eg, the presence of hypertension, diabetes, albuminuria, inflammation, etc). The strong association between GSRH and all-cause mortality after controlling for these factors is noteworthy. Also, the results are representative of the US noninstitutionalized civilian population.

CONCLUSION

In this large nationally representative sample of US adults, GSRH and physician assessments of health maintained low agreement. Furthermore, regardless of physician rating of a participant's health, self-assessed health via GSRH maintained a strong, graded association with mortality. These results suggest that important prognostic information is contained in an individual's GSRH assessment beyond other objective data and physician impression of health. Routinely collecting GSRH may enhance the focus on patient-centered care.