Duke 2008 | ASHE | Equity and Efficiency in Health and Healthcare

Background: The expected benefit of early detection of asymptomatic, early stage cancer tumors declines with the risk of death from competing causes. It makes no sense to subject patients to invasive tests and procedures if they are likely to die of an unrelated cause (e.g. heart disease) before a screen-detected cancer tumor will become symptomatic. Though the principle that patients with limited life expectancy do not benefit from early detection is widely recognized, it is difficult to implement in clinical practice. Some patients may have trouble understanding how the gains from screening are related to competing mortality risks. Guidelines do not provide clear rules about when clinicians and patients should discontinue screening. Fee-for-service reimbursement, "pay-for-performance" schemes, and the malpractice system provide incentives for over-screening. These factors suggest that screening behavior may be insufficiently responsive to variation in patient life expectancy.

Methods: There are a number of studies that use cross-sectional variation to examine the relationship between indicators of life expectancy and prostate cancer screening. However, these may be biased by omitted variable bias because many men never undergo screening in the first place. We use a different approach. Using Medicare claims data, we identified men age 65+ who received a prostate specific antigen (PSA) test in 1997 ("the index PSA") and were not previously diagnosed with prostate cancer (N = 32,039). We then examined the impact of negative health shocks -- heart attacks, strokes, hospitalization for any cause, and hospitalization for a "serious" cause -- on continuation of screening. Specifically, we used a Cox model to estimate time to receipt of a 2nd PSA test as a function of negative health shocks, controlling for age, race, and socioeconomic status. Observations for men who were alive at the end of the follow-up period (December 31, 2004) without having received a 2nd PSA were right-censored.

Results: 28,297 (88%) of the sample had a 2nd PSA test. Among the remaining 3,742 men, 2,447 died without receiving a 2nd PSA and 1,279 were alive at the end of 2004. The odds ratio associated with having a heart attack was 0.65 (SE: 0.05, p<0.01), implying that men who experienced a heart attack were less likely to receive a 2nd PSA test. Results were similar for other negative health shocks.

Conclusions: Encouragingly, we find that some patients and physicians react to new information about patient life expectancy; they do not blindly persist in past screening behavior. By limiting our sample to men previously screened for prostate cancer, we avoid some of the omitted variables problems that plague earlier studies. However, we are unable to determine whether screening decisions are optimal. In future work, we will apply these methods to mammography.