The National Committee for Quality Assurance (NCQA) is a “not-for-profit organization dedicated to improving health care quality.” One of their major initiatives is the Health Plan Employer Data and Information Set (HEDIS) which aims to evaluate the quality of care offered by various health plans.
In a 2001 NBER working paper (“Learning…“), researchers Michael Chernew, Gautam Gowrisankaran and Dennis Scanlon look at employee data from General Motors and see whether or not the HEDIS health plan report cards affects plan choice. GM is a large national employer so authors are able to look at data for about 70,000 employees in many different markets over two years. Between years one and two, health plan report cards information was disseminated to non-union employees. GM employees use flex dollars to pay from health insurance plans categorized into 4 types: fee-for-service basic (FFSB), fee-for-service enhanced (FFSE), HMOs and PPOs.
The authors use a Bayesian learning model to evaluate the impact of the introduction of the health plan report cards to the employees. A person’s utility is assumed to depend on perceived health plan quality, price, other plan attributes and other unobserved factors. Using a nested logit model, the authors estimate the conditional distribution of quality at time 0 (i.e.: the prior) and the conditional distribution of quality at time 1 (i.e.: the posterior). The prior is a function of plan reputation and experience while the posterior is a function of the prior and the signal (i.e.: the HEDIS report card value). The authors also include plan-market fixed affects as well as plan type-time interactions.
The authors note: “Since we include a fixed effect for the prior quality of each plan in each market, endogeneity would occur only if particular ratings or changes in prices are correlated with changes in unobservable plan characteristics that might change market shares even in the absence of the changes in price or ratings.” The authors do not believe this to be the case, but it is very possible that health plans will shift resources from procedures which are included in the HEDIS quality measure and away from medical care whose quality level will not be captured in the HEDIS report card.
The authors find that $100 increase in the price of a health plan will cause a 2.7% decrease in market share. Plans receiving a superior or average rating increased market share, but plans with below average or no data had reduced market share. The ‘no data’ category had the largest decrease in market share. The estimated value of the HEDIS information is about $20 for all people but $488 ex-post for the people who switched. The average value is small because only 12.4% of people switched plans in the sample after the HEDIS introduction and only about 3.9% switched plans due to the quality ratings. The authors conclude that “plan priors are more important than either ratings or prices,” which is likely due to 1) plan switching costs employees incur and 2) patient loyalty to their doctor.