Research Paper Volume 12, Issue 8 pp 7561—7575

Using a genetic algorithm to derive a highly predictive and context-specific frailty index

Figure 1. (A) Phases of the genetic algorithm: 1) an initial population of FIs is created; 2) the fitness (AUC) of each FI is tested; 3) the fittest FIs have higher chances to be selected for recombination; 4) two crossing-over points are randomly found for each parent FI: children FIs are created by combining different parts of parents FI; 5) a low probability of random mutations of a deficit is introduced; 6) children FIs replace the least fit FI; (B) Output of the genetic algorithm: iteration by iteration, the AUC of the best FI and average AUC of the population of FIs increases until convergence. The number of deficits included can vary iteration by iteration; (C) Distribution of the ga-FI in the whole population (histogram) and density functions in different subsamples. Abbreviations: FI = Frailty Index, AUC = Area under the Curve, CO = Crossing Over point; ga-FI = best genetic algorithm-derived FI.