Subjective, Objective, and Observed Long-Term Survival: A Longitudinal Cohort Study

To the Editor

Many professional guidelines recommend using life expectancy when considering diagnostic or treatment interventions in which the time to benefit may exceed patients’ survival.¹ When subjected to such tests and treatments, these patients are put at risk for up-front harms with little chance of reaping benefits.²

Patients’ perceptions of prognosis are important. Clinicians who follow recommended guidelines may urge patients to change health routines to which they have become accustomed. Patients who underestimate their survival may choose to forego interventions that are likely to help them, while those who overestimate may choose to undergo interventions that are more likely to cause harm. However, little is known about how well older adults estimate their survival, the aim of this analysis.

Methods

We drew a sample of 64-, 69-, 74-, 79-, 84-, and 89- year-old participants in the 2000 wave of the Health and Retirement Study (HRS), a nationally representative, longitudinal, prospective cohort study of adults in the United States.³ Institutional review board approval for HRS was received from the University of California, San Francisco, Committee on Human Research. Informed consent was obtained by HRS at the time of the interview. Our variable of interest was participants’ subjective estimate of long-term survival, assessed by the question, “What is the percent chance you will live to be x or more?” where x was 75, 80, 85, 90, 95, or 100 years old. Our sampling strategy ensured that all participants predicted survival across the same period. We calculated an objective estimate of life expectancy using the Lee life expectancy calculator.⁴ Observed survival was determined using mortality through 2010 (confirmed using the National Death Index). We compared participants’ subjective estimates to observed survival (using the C-statistic) and the objective estimates (using a best-fit analysis). Participants were classified as underestimating or overestimating their survival if their estimate was more than 25 percentage points less or greater than the calculated prediction, respectively.

Results

A total of 2018 respondents of the specified ages were interviewed directly in the 2000 wave of the HRS, and our final sample included 1722 participants who had complete responses necessary for the calculator (56%female, 88%white). We combined the 84- and 89-year-olds because of the small number of respondents in each group. Overall, discrimination was moderate for participants’ subjective estimates of survival compared with observed survival (C statistic = 0.62; P < .001). The 64- and 69-year-olds were moderately able to estimate their survival (C statistic = 0.62 and 0.58, respectively; P < .001 for both), but older participants fared no better than chance (Table). Substantial dissimilarity was seen between participants’ subjective and the objective estimates of survival (Figure). Overall, 54.7% of participants had estimates similar to the objective calculation; however, 32.7%underestimated and 11.5%overestimated. Underestimation was relatively similar across age groups, but overestimation increased significantly with age (Table) (P < .001).

Table 1.

Subjective, Objective, and Observed Survival^a

	Characteristics	Overall (N=1722)	Age 64 (n=544)	Age 69 (n=447)	Age 74 (n=322)	Age 79 (n=265)	Age 84/89 (n=144)
Survival	Survived follow-up %	62.1	83.7	71.2	56.9	36.7	19.7
	Subjective Estimate of survival (self-report) Mean % (SD)	58.2 (31.0)	68.8 (26.4)	62.0 (28.7)	56.9 (30.6)	43.3 (32.2)	41.6 (33.7)
	Objective Estimate of Survival (Lee life-expectancy calculator) Mean % (SD)	66.9 (23.2)	80.4 (16.6)	74.1 (17.4)	63.5 (20.5)	51.2 (21.6)	36.4 (20.6)
Discrimination compared to observed survival	Discrimination of Subjective Estimate c-statistic (95% CI)b	0.62 (0.59–0.65)	0.62 (0.54–0.69)	0.59 (0.52–0.65)	0.52 (0.44–0.58)	0.56 (0.48–0.63)	0.56 (0.45–0.68)
	Discrimination of Objective Estimate c-statistic (95% CI)b	0.79 (0.76–0.81)	0.76 (0.70–0.81)	0.67 (0.61–0.73)	0.70 (0.63–0.76)	0.71 (0.65–0.78)	0.80 (0.71–0.89)
Similarity (subjective vs. objective)c	Underestimated %	32.7	30.9	36.2	32.3	36.8	20.4
	Similarly Estimated %	54.7	63.4	56.3	52.8	45.5	49.4
	Overestimated %	11.5	4.7	7.5	14.8	17.0	29.2

^aReported values incorporate survey weights to account for the complex survey design.

^bC statistic is significant for 95%confidence intervals that do not cross 0.5.

^cCategorized as underestimated if the participant’s subjective estimate of survival was at least 25 percentage points less than the objective estimate (subjective – objective estimate of survival ≤ −0.25); Categorized as similarly estimated if objective estimate of survival was within ±25 percentage points of objective estimate; and categorized as overestimated if subjective estimate of survival was at least 25 percentage points greater than the calculated estimate (subjective – objective estimate o survival ≥ 0.25).

Figure 1. Subjective vs. Objective Estimates of Long-Term Survival.

This figure illustrates the dissimilarity between participants’ subjective estimate of long-term survival and the calculated objective estimate. A best-fit line was statistically significant but accounted for very little variability (R²=0.12, p<0.001). The blue line divides the region in the lower right, which represents those participants underestimated survival, from the upper left region, which represents those who overestimated. Underestimation was more common than overestimation.

Discussion

Our findings have important implications for clinicians. First, approximately half the time, participants accurately estimated survival compared with an objective clinical estimate, and when in error, they were more likely to underestimate than overestimate. Consequently, disclosing prognosis is not necessarily bad news. Second, because of the substantial dissimilarity between the subjective and objective estimates, patients may be hesitant to follow guideline recommendations to change long-standing regimens. Third, our findings make a case for using prognostic calculators, which were more accurate than participants’ subjective estimates and have been shown to be superior to clinician estimates.⁵ Fourth, participants in the HRS were willing to make an estimate of their survival and report it. Previous research suggests that older adults are generally willing to discuss long-term prognosis.⁶ Clinicians should solicit patients’ individual perception of prognosis and use this information as a starting point for further discussion, particularly among older patients who may be more prone to errors. Then, in conjunction with prognostic tools, they can begin to bridge the gap between subjective and objective estimates of survival.