Individual health indices via register-based health records and machine learning

Quantifying individual latent health has potential value in healthcare policy. This paper constructs an individual-level health index using detailed population-wide registers for individuals at advanced ages. In the methodology, we use traditional and machine learning techniques. We provide importance rankings of health variables and evaluate their ability to form an accurate stratification of health states. The high dimensionality of our data allows us to derive a small subset of health measures that enables similar analysis in settings where individual health registers may not be available. Finally, we use our indices to forecast individual-level mortality.