A comparison of different modelling techniques in predicting mortality with
the Tilburg Frailty Indicator (TFI) (Preprint)
BACKGROUND Background Modern modelling techniques may potentially provide more accurate predictions of dichotomous outcomes than classical techniques. OBJECTIVE Objective We aimed to study the predictive performance of eight modelling techniques to predict mortality by frailty. METHODS Methods We performed a longitudinal study with a 7-year follow-up. The sample consisted of 479 Dutch community-dwelling people >=75 years. Frailty was assessed with the Tilburg Frailty Indicator (TFI), a self-report questionnaire. This questionnaire consisted of eight physical, four psychological and three social frailty components. The municipality of Roosendaal (a city in the Netherlands) provided the mortality dates. We compared modelling techniques such as support vector machine, neural net, random forest, least absolute shrinkage and selection operator and classical techniques such as logistic regression, two 1Bayesian networks and recursive partitioning. The area under the ROC-curve (AUC) indicated the performance of the models. The models were validated using bootstrapping. RESULTS Results We found that the neural net model had the best validated performance (AUC=0.812) followed by the support vector machine model (AUC=0.705). The other models had validated AUCs <0.700. The recursive partitioning model had the lowest validated AUC (0.605). The neural net model had the highest optimism (0.156). The predictor variable ’difficulty in walking’ was important for all models. CONCLUSIONS Conclusions Because of the high optimism of the NN model, we prefer the SVM model for predicting mortality in community-dwelling older people with the TFI with added to it ’gender’ and ’age’. External validation is a necessary step before applying the prediction models in a new setting.