Penalty-based spatial smoothing and outlier detection for childhood obesity surveillance from electronic health records

Childhood obesity is associated with increased morbidity and mortality in adulthood, leading to substantial healthcare cost. There is an urgent need to promote early prevention and develop an accompanying surveillance system. In this paper, we make use of electronic health records (EHRs) and construct a penalized multi-level generalized linear model. The model provides regular trend and outlier information simultaneously, both of which may be useful to raise public awareness and facilitate targeted intervention. Our strategy is to decompose the regional contribution in the model into smooth and sparse signals, where the characteristics of the signals are encouraged by the combination of fusion and sparse penalties imposed on the likelihood function. In addition, we introduce a weighting scheme to account for the missingness and potential non-representativeness arising from the EHRs data. We propose a novel alternating minimization algorithm, which is computationally efficient, easy to implement, and guarantees convergence. Simulation shows that the proposed method has a superior performance compared with traditional counterparts. Finally, we apply our method to the University of Wisconsin Population Health Information Exchange database.