Health Status Predicts Long-Term Outcome in Outpatients With Coronary Disease
Background— Although patient-reported health status measures have been used as end points in clinical trials, they are rarely used in other settings. Demonstrating that they independently predict mortality and hospitalizations among outpatients with coronary disease could emphasize their clinical value.
Methods and Results— This study evaluated the prognostic utility of the Seattle Angina Questionnaire (SAQ), a disease-specific health status measure for patients with coronary artery disease. Patients were enrolled in a prospective cohort study from 6 Veterans Affairs General Internal Medicine Clinics. All patients reporting coronary artery disease who completed a SAQ and had 1 year of follow-up were analyzed (n=5558). SAQ predictor variables were the physical limitation, angina stability, angina frequency, and quality-of-life scores. The primary outcome was 1-year all-cause mortality, and a secondary outcome was hospitalization for acute coronary syndrome (ACS). Lower SAQ scores were associated with increased risks of mortality and ACS admissions. Prognostic models controlling for demographic and clinical characteristics demonstrated significant independent mortality risk with lower SAQ physical limitation scores; odds ratios for mild, moderate, and severe limitation were 1.5, 2.0, and 4.0 versus minimal limitation (P<0.001). Odds ratios for mild, moderate, and severe angina frequency were 0.8, 1.2, and 1.6 (P=0.078). The odds ratios for ACS admission among those with mild, moderate, and severe angina frequency were 1.4, 2.0, and 2.2, respectively (P=0.016).
Conclusions— SAQ scores are independently associated with 1-year mortality and ACS among outpatients with coronary disease and may serve a valuable role in the risk stratification of such patients.
Received January 8, 2002; revision received April 19, 2002; accepted April 22, 2002.
Health status measures quantify patients’ perceptions of how their disease affects them: their function, their symptoms, and their quality of life. Although increasingly used as end points in clinical trials, the utility of these measures in routine clinical practice has yet to be demonstrated. Although the physician’s history for patients with coronary artery disease (CAD) usually addresses their angina, physical limitations, and quality of life, the collection, scoring, and interpretation of formal health status questionnaires is considered too great a burden for most clinical settings. Yet, if there were additional advantages to using health status instruments, they might be more readily accepted by practitioners, both as end points in clinical trials and as tools in patient care.
See p 5
One potential use of formal health status assessment could be as a means of identifying patients at high risk for mortality or acute coronary syndromes (ACS). This would be consistent with one of the major objectives of modern cardiovascular practice, stratifying patients according to their risk for adverse events so that treatment can be tailored to individual patients. For example, coronary revascularization most clearly benefits those with the highest underlying mortality,1 and identifying high-risk patients is a cornerstone of current cardiovascular care. Linking health status scores to subsequent clinical outcomes can potentially provide enough value to clinicians that such measures might become a more routine component of clinical practice.
This study’s objective was to evaluate, in a large cohort of outpatients with CAD, the association between patients’ reported functional limitations, frequency of angina, and quality of life, as measured by the Seattle Angina Questionnaire (SAQ), with subsequent mortality and hospitalization after controlling for traditional risk variables.
These analyses use data from the Ambulatory Care Quality Improvement Project, a randomized trial of continuous quality improvement conducted from January 1997 to December 1999 in 6 Veterans Affairs Medical Centers. All patients enrolled in the General Internal Medicine Clinics at the start of the study were mailed a questionnaire packet including sociodemographic information, a medical history questionnaire, and a Short Form-36 survey. They were then followed up for 2 years. In addition, new patients entering the clinics during the study were enrolled, sent questionnaires, and followed-up until December 1999. On receipt of the medical history questionnaire, those reporting angina, a history of CAD, or a previous coronary event or revascularization procedure were identified as having CAD and were sent the SAQ. Because they were identified through their enrollment in an outpatient clinic, these patients represent outpatients with chronic CAD. The sensitivity and specificity of these self-reported selection criteria have previously been reported as 97% and 93%, respectively.2 SAQ data and 1-year follow-up were available on 5558 patients (83% of those reporting CAD). This project was approved by the Human Subjects Committee of the University of Washington and the institutional review boards at each study site.
Health Status Data
The SAQ is a self-administered, disease-specific measure for patients with CAD that has previously been demonstrated to be valid, reproducible, and sensitive to clinical change.3,4⇓ The SAQ quantifies patients’ physical limitations caused by angina, the frequency of and recent changes in their symptoms, their satisfaction with treatment, and the degree to which they perceive their disease to affect their quality of life. Each scale is transformed to a score of 0 to 100, where higher scores indicate better function (eg, less physical limitation, less angina, and better quality of life).
To simplify interpretation, the physical limitation and angina frequency scores were classified as minimal (scores 75–100), mild (50–74), moderate (25–49), and severe (0–24). Severe angina reflects having angina several times per day, moderate angina indicates having symptoms several times per week to every day, mild angina frequency occurs weekly, and minimal angina occurs less than once a week or not at all. Angina stability scores were classified as much better (76–100), slightly better (51–75), unchanged (50), slightly worse (25–49), and much worse (0–24), and SAQ quality-of-life scores were classified as excellent (75–100), good (50–74), fair (25–49), and poor (0–24).
All demographic, clinical, and health status data were obtained through self-administered questionnaires. Specific variables included demographic characteristics (sex, race), medical comorbidities (previous hospitalization, cancer, chronic heartburn, depression, drug abuse, kidney problems, liver disease/yellow jaundice/hepatitis, posttraumatic stress disorder, seizures or convulsions, peptic ulcer disease, and thyroid disease), cardiac risk factors (diabetes, hypertension, and smoking history), cardiovascular disease severity (stroke, congestive heart failure, previous hospitalization for ACS, previous coronary revascularization procedure, and previous myocardial infarction), and the SF-36.5
The primary end point was all-cause mortality. Survival status was determined through a query of local Veterans Affairs medical systems and the national VA Beneficiary Identification and Record Locator System, a validated source of survival status.6 Secondary analyses examined hospital admission for an ACS. Hospital admissions were identified from the discharge records of each participating medical center. Primary or secondary ICD-9 codes of 410.X and 411.1 were used to identify ACS admissions. Because patients may present with ACS admissions to non-VA hospitals, sensitivity analyses were conducted examining the influence of distance from patients’ residence to the VA hospital (the assumption being that the farther patients lived from a VA hospital, the more likely they were to be admitted to a non-VA facility). These analyses demonstrated a negative correlation between distance and ACS admission at the VA hospital (odds ratio, 0.90 per 10 miles, P=0.0003). When distance was included in the multivariable models, however, no significant changes in the relative risks associated with SAQ scores were detected.
Survival curves for all-cause mortality were estimated by Kaplan-Meier analysis, stratified by baseline SAQ score groups, and were compared by use of log-rank tests. Predictive models were developed for 1-year mortality and ACS admission by logistic regression. For each end point, 3 models were constructed. The first included only SAQ domain scores, and the second included only demographics and clinical variables. In each case, models were developed by incorporating predictors with significant univariate associations with outcome (probability value ≤0.1) and using backward elimination until all remaining factors were significant at the 0.1 level. Age was entered as a continuous predictor, and the remaining independent variables were categorical. A final model was developed by combining significant variables from the first 2 models to estimate the incremental prognostic value of the SAQ above demographic and clinical characteristics alone. For all models, P values and estimated odds ratios are presented.
Model calibration was assessed by comparing predicted and observed outcomes by deciles of predicted risk. Goodness-of-fit was determined by the Hosmer-Lemeshow test. Discriminatory power was evaluated by use of c statistics and compared by Mann-Whitney tests.7 Model discrimination and calibration were validated on 200 bootstrap resamples of the original data set.8 Supporting analyses were conducted by use of Cox proportional hazards models. All analyses were conducted with SAS version 6.12 and S-Plus 2000 for Windows.
Handling of Missing Data
Baseline demographic and clinical variables were missing in very few cases (<0.3%) and were imputed with the mean value (continuous variables) or by random sampling from the complete data (categorical variables). Nineteen percent of patients had missing SAQ scores because of incomplete questionnaires. Among the patients with incomplete SAQ instruments, the 1-year mortality was higher (7.5% versus 5.3%, P=0.005) and the ACS rate lower (2.3% versus 3.6%, P=0.04) than for those with complete data. Primary analyses were conducted on those with complete SAQ data. Supporting analyses to evaluate the impact of missing data were conducted with multiple imputation techniques9 conditioning on the available SAQ and SF-36 scores. Imputation was performed by use of S-Plus functions developed by Frank Harrell at the University of Virginia.10 No substantial differences from the original models were detected, so only complete-data analyses are presented.
Baseline characteristics and univariate odds ratios for 1-year mortality and ACS admission are listed in Table 1. The mean SF-36 physical component score was 30.8±10.5, indicating that these patients were almost 2 SD below the mean of the US population in terms of general physical health status.11 Over the first year of observation, there were 238 deaths (5.3%) and 154 ACS admissions (3.6%).
Univariate Predictors of Outcome
Table 1 describes the relative odds of death at 1 year by demographic characteristics, comorbidities, and SAQ scores. The most powerful predictors were age (OR per 10 years, 1.6; 95% CI, 1.4 to 1.9), presence of congestive heart failure (OR, 2.0; 95% CI, 1.8 to 3.2), and health status. Figure 1 diagrams the relationship of each SAQ domain with mortality. The strongest health status predictor was SAQ physical limitation; patients with severe limitations had a relative odds of dying of 6.2 (95% CI, 3.8 to 10.5) versus those with minimal limitation. SAQ physical limitation, angina frequency, and quality-of-life domains all had a significant trend toward lower mortality with higher patient function, poorer symptoms, and better quality of life (ie, higher ranges of score). The angina stability scale examines recent changes in symptoms; those patients reporting a significant deterioration in their angina over the preceding month had a 1-year mortality rate of 11.4% compared with 4.9% for the rest of the population. Health status was also predictive of time to death throughout the 2-year follow-up period. This is demonstrated graphically in the Kaplan-Meier curves for the SAQ physical limitation domain (Figure 2). The separation of the curves, especially for those with the worst function, begins almost immediately and continues throughout the period of observation.
Table 1 also describes the odds of admission for ACS at 1 year. The strongest predictor of ACS admission was previous hospitalization for an ACS (OR, 10.3; 95% CI, 6.6 to 15.7). The next most powerful predictors were SAQ angina frequency, physical limitation, and quality-of-life domains. For example, those reporting severe angina were 3.1 (95% CI, 1.7 to 5.3) times more likely to be admitted to the hospital for an ACS than those reporting minimal angina.
Multivariable Prognostic Models
Of the 4 SAQ domains examined, physical limitation and angina frequency constituted most of the predictive power for both 1-year mortality and ACS admission. The independent variables that remained in the models are described in Tables 2 and 3⇓. In the SAQ-only models, odds ratios were 6.0 (95% CI, 3.6 to 10.6) for patients with severe physical limitations, 2.6 (95% CI, 1.6 to 4.4) for those with moderate limitations, and 1.7 (95% CI, 1.0 to 1.3) for those with mild limitations versus patients experiencing minimal limitations. Angina frequency (independent of physical limitation) was marginally associated with increased mortality (P=0.068) at 1 year but was significantly associated with mortality at 2 years (P=0.033). The angina frequency scale was strongly predictive of hospitalization for ACS; the odds ratio for patients with severe angina was 2.2 (95% CI, 1.2 to 4.1) compared with the least symptomatic group. Patients reporting severe physical limitation resulting from angina were also more likely to be admitted for an ACS than those with minimal limitation (OR, 2.0; 95% CI, 1.2 to 3.1). Other significant predictors in the clinical model for 1-year mortality included age, congestive heart failure, cancer, previous hospitalization, diabetes, and stroke. Previous hospitalization and previous PTCA or CABG were the strongest clinical predictors of admission for ACS.
Comparing the clinical and combined clinical-SAQ models, the SAQ provided significant independent prognostic value in the presence of clinical predictors (likelihood ratio test P values <0.001 for both mortality and ACS models). Estimated odds ratios remained consistent between the individual and combined models, implying minimal confounding between the SAQ and clinical variables, although the P value for physical limitation in predicting ACS admission did drop from 0.03 to 0.10.
Inclusion of SAQ domains significantly increased the model c-statistics for both the mortality (0.69 to 0.72, P=0.004) and ACS (0.69 to 0.73, P=0.003) models. By use of Wilcoxon signed-rank tests, the predicted probability of death increased significantly from the clinical to the combined model for patients who died (P<0.001) and decreased significantly for those who survived (P<0.001). Similar results were observed for the ACS models (P<0.001 for those not admitted and P=0.005 for those admitted). Calibration was good for all models, with Hosmer-Lemeshow tests showing no significant differences between observed and predicted probabilities by decile of risk.
Supporting analyses used Cox proportional hazards models on time to death and time to ACS admission. These analyses yielded very similar results, corroborating a significant incremental prognostic contribution of the SAQ.
In outpatients with chronic ischemic heart disease, the patients’ health status (their symptoms, physical function, and quality of life) was a strong predictor of subsequent mortality and ACS admission. Only age was a more powerful predictor of mortality, and only a history of previous hospitalizations (particularly ACS admissions) more strongly predicted ACS than certain domains of the SAQ. For most SAQ domains, the relationship of outcome to SAQ score followed a predictable relationship: higher scores (less impairment or fewer symptoms) were associated with significantly lower event rates. These models had discriminatory power (c statistics) that was comparable to that of prognostic models of survival after bypass surgery12 or angioplasty13 and had excellent calibration in predicting the observed event rates among the population. Whereas previous investigators have demonstrated a relationship between generic health status and survival in elderly patients14 and between generic measures and perioperative mortality during bypass surgery,15 this is the first study to demonstrate a similar relationship among outpatients with CAD.
The independent prognostic ability of health status may have been expected on clinical grounds. For example, it is obvious to a clinician that a physically active patient with infrequent angina and a positive outlook on life will survive longer than a patient who is unable to walk because of severe and frequent symptoms, even if the 2 patients have identical coronary anatomy and left ventricular function. Yet, such clinical judgment is not necessarily reproducible, whereas the self-administered SAQ can capture such data in a reproducible and valid manner.4 Furthermore, this study augments the work of Califf and colleagues,16 who described the incremental prognostic value of angina over angiographic data to predict outcome.
This investigation was conducted in a geographically diverse population of the United States and included a large number of patients. Its principal limitation is that it was conducted among patients within the Veterans Affairs Health System and should be repeated in a more diverse sample of patients, including women. In addition, there was a potential selection bias in that not all patients returned completed health status surveys. Nevertheless, there were few observed differences between responders and nonresponders, and when the results of this study are generalized to future populations, a similar selection bias with respect to incomplete surveys is likely to be encountered. A third potential limitation is our reliance on administrative data for ACS admissions. Because some patients may have been admitted outside of the VA system and not properly identified in our analysis, our prediction model may be biased. Nevertheless, we did not detect a change in the SAQ-ACS admission relationship when distance from the VA was included in the model. Furthermore, recent work by Eisenstein et al17 reveals a 2-year ACS event rate that is approximately half as large as their mortality rate (≈6.7% mortality and ≈3% myocardial infarction), a ratio of events similar to that which we report (8.8% and 4.9%). Finally, demonstrating a relationship between cross-sectional health status and survival does not necessarily indicate that changes in patients’ health status will be associated with changes in survival. Longitudinal analyses are needed to explore this, but the expectation is that this relationship would not be present because interventions that greatly improve the control of angina, such as percutaneous coronary interventions, have never been shown to increase survival.
In summary, these data support at least 3 uses of the SAQ. The SAQ is an appropriate end point for clinical trials not only because it quantifies highly relevant goals of therapy (the alleviation of symptoms and the improvement of both functioning and quality of life) but also because it is highly correlated with survival—the ultimate “hard” outcome in patients with coronary disease. Second, the SAQ can enhance quality assessment efforts by increasing the validity of risk-adjustment models of mortality. And finally, the SAQ may prove useful in the management of populations of patients with coronary disease. It is essential to have a mechanism of identifying patients for whom additional interventions, such as revascularization or more aggressive medical therapy, are indicated. Although all patients should receive optimal risk-factor modification, including control of their blood pressure and hyperlipidemia and smoking cessation, some may warrant even more aggressive interventions. The SAQ may prove useful in identifying such patients (ie, those with a very high 1-year event rate). Future work is needed to better understand the potential role of health status assessment in quality assessment and patient care.
The research reported here was supported by the Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development Service, grants SDR 96-002 and IIR 99-376 with Institutional Review Board approval. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs.
Dr Spertus developed and owns the copyright for the Seattle Angina Questionnaire.
- ↵Fisher SG, Weber L, Goldberg J, et al. Mortality ascertainment in the veteran population: alternatives to the National Death Index. Am J Epidemiol. 1995; 141: 242–250.
- ↵Little R, Rubin D. Statistical Analysis With Missing Data. New York, NY: Wiley; 1987.
- ↵Harrell F. Hmisc S-PLUS function library. In: lib.stat.cmu.edu; 1999.
- ↵Jones RH, Hannan EL, Hammermeister KE, et al. Identification of preoperative variables needed for risk adjustment of short-term mortality after coronary artery bypass graft surgery. The Working Group Panel on the Cooperative CABG Database Project. J Am Coll Cardiol. 1996; 28: 1478–1487.
- ↵Wolinsky FD, Johnson RJ. Perceived health status and mortality among older men and women. J Gerontol. 1992; 47: S304–S312.
- ↵Rumsfeld JS, MaWhinney S, McCarthy M Jr, et al. Health-related quality of life as a predictor of mortality following coronary artery bypass graft surgery. Participants of the Department of Veterans Affairs Cooperative Study Group on Processes, Structures, and Outcomes of Care in Cardiac Surgery. JAMA. 1999; 281: 1298–1303.