Abstract:
|
Studying repeated measures time-to events with discontinuous risk intervals requires appropriate data structure and modeling assumptions. We created a data set with discontinuous risk intervals and used generalized estimating equations (GEE) Poisson, GEE logit, and Cox proportional hazards counting process (CCP) models. We applied these models to MarketScan health claims data collected during outpatient visits between 2006 and 2013. We examined whether 163,000 children aged of 0 to 20 years with an autism spectrum disorder (ASD) reported more injury events than 200,000 age-matched controls. After adjustment for demographic factors, ASD was significantly associated with an increased risk for injury visits in GEE Poisson and GEE logit models (ps=.009), while marginally for CCP model (p=.07). After further adjustment for previous injury visits, the estimates for all 3 models were not statistically significant indicating the intra-individual dependency of injury visits. Both GEE and CCP models can be applied to the repeated measures time-to events data. In some cases, adjusting for previous health outcomes may be informative and impacts the interpretation of the findings.
|