How well do physicians predict the life expectancy of older adult outpatients? A systematic review and discussion of the implications for cancer screening guidelines

Background: Shortened life expectancy may limit older adults’ ability to benefit from certain medical interventions, such as preventive services. Appropriate consideration of a patient’s life expectancy may help physicians weigh potential benefits against potential harms for that individual. Some recent cancer screening guidelines have recommended this individualized approach to cancer screening decisions to avoid overscreening or underscreening. However, it is not clear how competent physicians are at predicting life expectancy, and there is little consensus about how these predictions should guide their recommendations for or against cancer screening. Purpose: This systematic review attempts to characterize physicians’ accuracy and reliability at predicting the life expectancy of older adult outpatients. A discussion further considers what level of evidence is needed to justify the replacement of an age-based cancer screening guideline to a life expectancy guideline for stopping cancer screening. Data Sources: MEDLINE, EMBASE, PsycINFO, reference lists Study Selection: Studies that assess physicians’ estimates of life expectancy for older adults who are not terminally ill are included. Data Extraction: Data relating to physicians’ accuracy and reliability, as well as physician attributes associated with predictive performance, are extracted. Study quality is assessed according to criteria adapted from the Centre for Review and Dissemination’s quality assessment criteria for diagnostic tests. Data Synthesis: Six small, cross-sectional studies utilizing patient vignettes and comparing physicians’ predictions to some reference standard are included. As a group, physicians have a tendency to underestimate life expectancy by 1–2 years and 10-year mortality risk by 10-15%, with poor inter-rater reliability. Among studies that reported LE predictions in years (n = 3), the prediction error (±SD) ranged from -1.63 (± 3.70) years to +0.34 (± 3.29) years. Limitations: Quality ratings for included studies were generally fair or poor. Major limitations include the use of patient vignettes in place of real patient encounters, reference standards for life expectancy of unclear quality, and small, nonrandom samples of physicians. Conclusions: As a group, physicians are moderately accurate in their predictive ability for the life expectancy of older adults but reliability among individuals is poor. Larger studies of physicians’ predictions for real patients are needed, possibly incorporating life expectancy prediction rules to assist physicians. Six criteria for evaluating whether there is enough evidence to support a life expectancy-based cancer screening guideline are proposed.Master of Public Healt