Predicting 10-Year Mortality for Older Adults

To The Editor: Preventive interventions, such as cancer screening, expose patients to immediate risks with delayed benefits, suggesting that risks outweigh benefits in patients with limited life expectancy. Recent guidelines recommend considering patients' life expectancy when deciding whether to pursue preventive interventions with long lag times to benefit (≥ 7 years) such as colorectal cancer screening and intensive glycemic control for diabetes.^1,2 However, most mortality indices have focused on short-term risk (≤ 5 years).^3,4 We examined whether our previously developed 4-year mortality index⁵ accurately predicted 10-year mortality.

Like our previous analysis, this analysis uses the 1998 wave of the Health and Retirement Study (HRS), a nationally representative cohort of community-dwelling US adults older than 50 years. The HRS cohort was divided geographically into development (East, Central, and West; n = 11 701) and validation (South; n = 8009) cohorts. Self-report data were collected primarily through telephone interviews (response rate 81%).

The primary predictor was a 12-item mortality index (ages 60-64 years: 1 point, ages 65-69 years: 2 points, ages 70-74 years: 3 points, ages 75-79 years: 4 points, ages 80-84 years: 5 points, ages ≥ 85 years: 7 points; male sex: 2 points; current tobacco use: 2 points; body mass index <25: 1 point; diabetes: 1 point; nonskin cancers: 2 points; chronic lung disease: 2 points; heart failure: 2 points; difficulty bathing: 2 points; difficulty managing finances: 2 points; difficulty walking several blocks: 2 points; and difficulty pushing/pulling large objects: 1 point). Our outcome was death through 2008 (10-year mortality), confirmed with the National Death Index.

A risk score was calculated for each participant by summing the points for each risk factor present. We calculated the 10-year mortality rates across point scores. Kaplan-Meier methods were used to display the validation cohort survival experience, and logistic regression with bootstrapping was used to determine the C statistic, 95% confidence intervals, and 2-sided P values. A P value of less than .05 was considered statistically significant. Cox proportional hazards analyses yielded similar results.

The committee on human research of the University of California, San Francisco, approved this study with a waiver for informed consent. The statistical software used was STATA version 12.0 (StataCorp).

Baseline characteristics of the cohort were described in detail previously.⁵ Briefly, in the validation cohort, the mean (SD) age of participants was 67 (10) years; 56% (n = 4516) were women, 11% (n = 826) reported a history of cancer, 16% (n = 1246) reported diabetes mellitus, 18% (n = 1414) reported difficulty in at least 1 activity of daily living, and 32% (n = 2527) died during the 10 years of follow-up. The development cohort had similar characteristics.⁵

In the development cohort, 10-year mortality rates ranged from 2.5% (95% CI, 1.1%-3.9%; n = 12/486) for participants with 0 points to 96% (95% CI: 94%-98%; n = 298/310) for participants with 14 or more points. In the validation cohort, 10-year mortality rates ranged from 2.3% (95% CI, 0.7%-3.8%; n = 8/354) to 93% (95% CI, 90%-96%; n = 239/257) (Table). The C statistic for the index was 0.838 (95% CI, 0.830-0.846) in the development cohort and 0.834 (95% CI, 0.824-0.843) in the validation cohort. There was no evidence of poor calibration (validation cohort, Hosmer-Lemeshow P = .38).

The Kaplan-Meier survival curves showed that the differences in survival by point score seen at 4 years were magnified at 10 years (Figure).

We validated a mortality index that accurately stratified older adults into groups at varying risk for 10-year mortality. Extending the index from 4 to 10 years did not diminish the model discrimination (validation cohort, C statistics 0.817 vs 0.834; P = .35), suggesting that the risk factors important for 4-year mortality prediction are also important for 10-year mortality prediction. The model compares favorably with other mortality indexes that predict mortality beyond 7 years.³

One limitation of the index is that it was developed and validated in a single, large, national study and should be validated on a separate population to assess generalizability.

Patients identified by this index as having a high risk of 10-year mortality may be more likely to be harmed by preventive interventions with long lag times to benefit, whereas patients identified as having a low risk of 10-year mortality may be good candidates for such interventions.

Author Contributions: Dr Lee had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Cruz, Covinsky, Lee.

Acquisition of data: Stijacic-Cenzer, Lee.

Analysis and interpretation of data: Cruz, Covinsky, Widera, Stijacic-Cenzer, Lee.

Drafting of the manuscript: Cruz, Stijacic-Cenzer, Lee.

Critical revision of the manuscript for important intellectual content: Covinsky, Widera, Lee.

Statistical analysis: Stijacic-Cenzer, Lee.

Obtained funding: Covinsky, Lee.

Administrative, technical, or material support: Cruz, Covinsky, Lee.

Study supervision: Widera, Lee.

Funding/Support: Dr Lee's effort was supported by the American Federation for Aging Research and the National Institute on Aging through the Beeson Career Development Award K23AG040779. Dr Covinsky's effort was supported by grant K24AG029812 from the National Institute on Aging.

Role of the Sponsor: The funders had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.

Previous Presentation: This work was presented as an oral abstract at the 2011 Society for General Internal Medicine National Meeting, Orlando, Florida.