Abstract:
|
Harrell's c index is widely used to measure the prognosis accuracy in predicting univariate survival. However, disease outcomes may show up in multiple endpoints of interest. The endpoints are often correlated and sometimes competing or semi-competing with each other. In particular, we are interested in the case where subjects in an intermediate disease state can experience both disease regression and disease progression. Disease progression terminates the at-risk process for disease regression, but not vice versa. We propose two extensions of Harrell's c index for composite disease outcomes and conduct inference using their properties as U-statistics. Using simulations, we show that our extensions are more efficient at identifying prognostic variables than Harrell's c index for the primary outcome alone. We apply these new measures to patients with impaired glucose resistance (IGR) who may either progress to type II diabetes or regress to normal glucose resistance (NGR).
|