By leveraging statistical properties (log-rank test for survival) of patient cohorts defined by binary thresholds, poor-prognosis patients are identified by the sigsquared package via optimization over a cost function reducing type I and II error.