Predicting Life Expectancy in Heart Failure

In this issue of JAMA, Allen and colleagues¹ report findings from a provocative study that identifies a disconnection between patient-predicted and model-predicted life expectancy among ambulatory patients with heart failure. In so doing, this study provides challenges to evaluating the need to know prognosis, communicating prognostic information, and anticipating the consequences of such information for a patient with heart failure.

Allen et al used the Seattle Heart Failure Model (SHFM)² to estimate life expectancy in the setting of stable ambulatory chronic heart failure in a small single-center tertiary care patient population and then determined the actuarial-predicted life expectancy in the absence of any known disease. To determine the patient's own assessment of survival, a de novo questionnaire was constructed to determine life traits and perceptions, with the scores acquired using a visual analog scale. Questions to assess depression (using a modified Beck Depression Inventory) and religiosity (using the Duke University Religion Index) were included. Actual survival over 2.8 years of follow-up was determined from the Social Security Death Index. Twenty percent of the intended patient cohort opted out of participation.

The SHFM predicted a median survival of 10 (range, 2.0-25.2) years; the actuarial model predicted a median survival of 20.5 (range, 3.5-52.6) years. There was no correlation between these models. Patients predicted a median survival of 13 (range, 1-54) years, and the majority overestimated their life expectancy based on the SHFM by approximately 40%. The ratio of patient-predicted life expectancy vs model-predicted life expectancy was more than 2:1 for certain patients, suggesting a significant overestimation. A number of factors correlated with an overly optimistic outlook in univariate unadjusted analysis, including age, race, education level, and severity of heart failure. In multivariable analysis, younger age, lack of college education, higher New York Heart Association class, and nonischemic etiology remained significant for reasons that are neither clear nor intuitive. Measures of religiosity did not correlate with accuracy of prediction. Actual survival data demonstrated no relation to the patient's level of optimism and survival. The authors suggest that both patient and clinician issues are at play in the apparent discrepancy between actual prognosis and patient expectations of outcomes in heart failure; they also suggest that certain levels of decision making are not well facilitated by this lack of an informed prognostic perspective. The study indicates a need to seek better ways to convey difficult information and to have a more completely informed patient population.

However, a relevant question is “Why is it important for a given patient to be aware of precise quantitative prognostic information?” There are several treatment decisions for which this is important—if anticipated survival time in heart failure is short (<1 year), referral for heart transplantation or mechanical support needs to be considered. Similarly, referral for hospice or palliative care would be greatly facilitated by an accurate estimation of even shorter-term survival (approximately 6 months or less); conversely, if the expected survival time is at least 1 to 2 years, referral for implantation of a cardioverter-defibrillator would be appropriate.^{3 - 4} However, beyond these specific examples, no other therapeutic interventions for heart failure exist for which precise knowledge of the likelihood of survival matters in the decision-making process. Most evidence-based therapies for heart failure are indicated because of disease severity, which is usually arbitrated as the extent to which active symptoms are present and not because of threatened survival time; thus, precise knowledge of survival time does not factor directly into the usual decision matrix. Another reason precise awareness of survival may be important is embedded in the “time trade-off” construct.⁵ Knowing that survival is limited, patients with advanced disease might opt for comfort measures or an enhanced quality of life, even at the expense of shortened survival.

Presuming that a need to know can be firmly established, a second relevant question is “Are there accurate tools that allow the prediction of death in chronic heart failure?” Several earlier prognostic models have been suggested in the literature, but some were intended to predict the need for heart transplantation and were validated in patient populations referred for consideration of transplantation, not in a general heart failure cohort.^{6 - 7} Moreover, these models were not representative of contemporary treatment approaches to heart failure. The SHFM is a well-validated, more contemporary, and useable tool, but the validation of this model has not included access to clinical trial databases that are racially diverse, include use of β-blockers, and are represented by substantial use of implantable devices.² For example, the SHFM does not account for the survival advantage due to use of isosorbide dinitrate and hydralazine in African American patients with heart failure or the benefit of using an implantable cardioverter-defibrillator in class II/III heart failure.^{8 - 9} Thus, even the SHFM is an imperfect model. Like any survival model, the error rate reflects either an overestimation of survival (perhaps in patients with more advanced disease) or an underestimation (perhaps in those with more moderate disease and implanted devices).

Fortunately, the care of patients with heart failure is improving, not only because of effective evidence-based strategies but also because of greater adherence to quality metrics, as suggested by secular trends demonstrating steadily improved survival.^{10 - 11} The use of any tool derived from retrospective observations as a means to predict future events in heart failure must be cautious. For example, given that data from Allen et al¹ show a discrimination of fit for the SHFM of 0.73 (C statistic for observed death and model-predicted death), there is a potential error rate of 27% that could affect decision making based on an imperfect estimate of prognosis.

A third relevant question is “What are the potential risks of providing a more quantitative prognostic perspective and being in error?” There is value in providing qualitative prognostic information for patients with heart failure, ie, good, guarded, or poor. But in the context of quantitative estimates, should care be withheld from a patient whose life expectancy is deemed too short, only to discover that a sudden death event might have been prevented with implantation of a cardioverter-defibrillator? What are the emotional consequences of a premature referral to hospice? What if a patient elects to withdraw reasonable care based on a perspective that survival time is short, recognizing that there may be a 30% chance that this is a flawed assessment? Should a payer be allowed to deny coverage for an intervention deemed clinically appropriate and desired by the patient, based on an imprecise estimate of survival? Patients with advanced disease rely on several resources to sustain life and maintain both its quality and dignity, among which is the embrace of hope. The implied precision of a quantitative prognostic assessment may be sufficient to remove hope and thus affect the quality of life.

A fourth relevant question is “How is this information best conveyed?” The physician has the responsibility to be conversant with patients and to represent the full spectrum of available information, ie, diagnosis, evaluation, treatment, and prognosis. Physicians are often eager to share the triumphs of medicine but should not shy away from discussing its more somber aspects. If patients do not appreciate the actual prognosis of their disease, did their clinicians convey that information to them and was the dialogue culturally sensitive without any evidence of disparate health care? Allen et al note that recall of a prior dialogue with a clinician was not associated with a patient's own prognostic assessment. If these conversations are held, were they supported by additional educational resources, social service support, and clerical counseling where desired or needed? What are the characteristics of physicians who are able to communicate this information effectively? The available data portray only half of the story—patients overestimate their survival, and certain patients are especially likely to do so. It is concerning that patient naiveté may be operative, given the association of younger age and lesser levels of education. It is important to better understand clinician-related issues prior to introducing new processes that address this critical patient-physician discussion.

Currently, there is insufficient precision in the prognostication of heart failure, and decision making at the end of life is perhaps the most personalized of all decision making in medicine. Although well-intended and carefully constructed tools and awareness of the natural history of disease are helpful, it is the primacy of the patient-physician interface that must prevail. Until these questions are fully addressed, it is best to avoid adopting an imprecise method, instead continuing to embrace the individualized decision-making process guided by physician judgment that incorporates all patient care considerations.