Inferring the Extent of Health Plan Provider Choice among the Privately Insured

A common feature of many health plans today in the US is that only a subset of all providers of a given service are included in a network. In these “narrow network” plans, consumers either have no coverage for “out-of-network” providers, or in “tiered network” plans are held accountable for additional cost sharing when going out-of-network. Restricted provider choice has important implications for patient sorting across plans, access to care and consumer behavior, all of which play a crucial role in containing costs and improving quality.

While narrow network health plans have recently gained market shares in the US, little is understood about their impacts, due partly to a lack of reliable measures of breadth of provider networks. In the Health Insurance Marketplace, regulatory assessment of plan’s network adequacy has largely relied on insurer-reported provider directories which can be outdated or difficult to quantify. The goal of this paper is to infer health plan level provider choice using claims level data.

I use the Truven’s MarketScan claims database, focusing on 400 million outpatient services records of privately insured in the US between 2008 and 2013. The “modeling sample” consists of 31,382 plan-metropolitan statistical area (MSA)-year aggregated observations spanning 445 plans, 439 MSAs, and 6 plan types. Studied plan types include Comprehensive, Exclusive Provider Organization (EPO), Health Maintenance Organization (HMO), Non-Capitated Point-of-Service (POS), Preferred Provider Organization (PPO), and Consumer-Driven/High-Deductible Health Plans (CDHP/HDHP).

The first part of the paper uses distinct provider counts per claim as a metric for provider networks and shows initial evidence of differences in network breadth across plan types. Simple sample means suggest that compared to PPOs, EPOs on average have fewer provider choices, whereas Comprehensive plans are more often observed with more choices. This is consistent with the subsequent regression analyses controlling for additional covariates (MSA fixed effects and average plan level risk scores), in which Comprehensive plans predict the highest level of provider choice, followed by POS, HMO, PPO, CDHP/HDHP, and EPO.  

In the second part of the paper (in progress), I develop a maximum likelihood model to infer the breadth of provider choice at the plan level. The model starts with the assumption that all providers in a plan serving an area within the same in-network versus out-of-network status are equally likely to be chosen by patients (later relaxed), and that health plans choose a constant fraction, λ, of providers in each MSA to cover. The modeling strategy builds on the hypothesis that narrow panels will result in fewer providers seen, and counts of patient visits will be more evenly distributed across providers than the market area average. Preliminary results suggest that provider concentrations (measured by Herfindahl-Hirschman Index at the plan-MSA-year level) are higher in plan types with relatively more provider choices, and that the proportion of providers covered, λ, varies across plans and by plan type as expected.

A methodology for creating an index of provider network breadth would benefit many other studies trying to understand the consequences of narrow panels on diverse outcomes.