A conundrum in cancer quality measurement: Performance on long-term survival reflects performance a long time ago.

Abstract

e18222 Background: Long term outcomes, such as five-year survival following cancer treatment, are widely accepted as important metrics of cancer care quality. But because they take years to accrue, they reflect the quality of care delivered in the past. The objective of this study was to delineate to what extent measures of long term outcomes related to cancer care could be reliably used to predict care quality in future periods. Methods: Taking advantage of FFS Medicare data from related studies, we assessed survival of cancer patients at one and four years by hospital for four separate index years 2006, 2011-2013. We assembled pairs of years across which we could compare risk adjusted outcomes separated by one year (2011 to 2012; 2012 to 2013), two years (2011 to 2013), and five through seven years comparing 2006 to 2011, 2012, or 2013 respectively. The outcome was risk adjusted mortality rates at the hospital level determined by dividing the observed number of deaths by an expected number, adjusting for age, and median income level of the zip code of residence. For each hospital and for every pair of analytic years we separately compared the correlation of one and four-year risk adjusted mortality using both Pearson and Spearman correlation statistics. A mixed effects model was fit to determine if size and year gap between measurements could explain possible differences in the correlation of hospital performance, while controlling for hospitals as a random effect. Results: There were 1,640 hospitals and more than 350,000 patients in each analysis year. Focusing on the correlation of the four year survival outcomes, the highest volume hospitals have significantly higher correlations compared to the other hospital categories (P < 0.001). The greater the elapsed time between measurement periods the lower the correlation. But in the mixed model analysis neither time elapsed between measurement periods (p = 0.51) nor the interaction between time elapsed and hospital volume (p = 0.95) were significant. Analyses of one-year outcomes produced similar findings. Conclusions: Performance on four-year survival outcomes of cancer patients was reasonably correlated for large volume hospitals, but less so for hospitals with lower volumes.