Aging
Navigate
Submit an Article
Research Paper|Volume 12, Issue 8|pp 7561—7575

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

Alberto Zucchelli1,2, Alessandra Marengoni1,3, Debora Rizzuto1,4, Amaia Calderón-Larrañaga1, Maurizio Zucchelli5, Roberto Bernabei6, Graziano Onder7, Laura Fratiglioni1,4, Davide Liborio Vetrano1,6
  • 1Aging Research Center, Department of Neurobiology, Care sciences and Society, Karolinska Institutet, Stockholm 17165, Sweden
  • 2Department of Information Engineering, University of Brescia, Brescia 25123, Italy
  • 3Department of Clinical and Experimental Sciences, University of Brescia, Brescia 25123, Italy
  • 4Stockholm Gerontology Research Center, Aldrecentrum, Stockholm 11346, Sweden
  • 5Kandou Bus SA, Lausanne 1015, Switzerland
  • 6Department of Geriatrics, Fondazione Policlinico “A. Gemelli” IRCCS and Catholic University of Rome, Rome 00168, Italy
  • 7Department of Cardiovascular, Endocrine-metabolic Diseases and Aging, Istituto Superiore di Sanità, Rome 00161, Italy
Received: February 17, 2020Accepted: April 8, 2020Published: April 28, 2020
https://doi.org/10.18632/aging.103118

Copyright © 2020 Zucchelli et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY 3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Abstract

The frailty index (FI) is one of the most widespread tools used to predict poor, health-related outcomes in older persons. The selection of clinical and functional deficits to include in a FI is mostly based on the users’ clinical experience. However, this approach may not be sufficiently accurate to predict health outcomes in particular subgroups of individuals. In this study, we implemented an optimization algorithm, the genetic algorithm, to create a highly performant (FI) based on our prediction goals, rather than on a predetermined clinical selection of deficits, using data from the Swedish National Study on Aging and Care in Kungsholmen (SNAC-K) and 109 potential deficits identified in the dataset. The algorithm was personalized to obtain a FI with high discrimination ability in the prediction of mortality. The resulting FI included 40 deficits and showed areas under the curve consistently higher than 0.80 (range 0.81-0.90) in the prediction of 3-year and 6-year mortality in the whole sample and in sex and age subgroups. This methodology represents a promising opportunity to optimize the exploitation of medical and administrative databases in the construction of clinically relevant frailty indices.