Measuring Restrictiveness of Medicare Advantage Networks : A Claims-Based Approach

As is common in the private insurance market, insurers offering Medicare Advantage (MA) plans typically restrict the network of providers from which patients can seek care. Existing work evaluating the breadth of these provider networks relies on provider directories and typically measures breadth as the fraction of providers in-network within some geography. However, well-documented errors prevalent in provider directories and data identifying provider locations can lead to errors with this approach. We use Medicare prescription drug event data from 2011 to 2017 to identify interactions between MA beneficiaries and providers, creating a predictive model that estimates counterfactual utilization to measure the degree to which networks restrict the number of providers seen by patients. We detail the construction, tuning, and estimation of this model for one high-prescribing specialty (primary care). A random forest model outperformed other choices. Our estimates indicate that, as theory predicts, preferred provider organizations (PPOs) are less restrictive than health maintenance organizations (HMOs), that issuers with higher market share impose more restrictive networks, and that increased provider supply is associated with less restrictive networks. These results help validate our methodological approach, providing justification for application more generally.We also provide guidance on how our approach can be replicated for other markets and/or specialties