Competitive Bidding in Medicare: Who Benefits From Competition?

The county-level data contain average plan bids and rebates weighted by plan enrollment, as well as the averagecounty CMS Hierarchical Condition Categories (CMSHCC) risk scores of beneficiaries. CMS-HCC risk scores are calculated using beneficiary age, sex, and clinical diagnoses.¹¹ The plan bid is standardized to a beneficiary of 1.0 risk, making all bids comparable. The plan-level data contain planspecific bids, rebates, and average CMS-HCC risk score of plan enrollees. County benchmark rates and actual fee-forservice costs are also standardized to a beneficiary of 1.0 risk. We restricted analysis to local plans, which bid against the same benchmark. We excluded regional plans, which use different benchmarks. We also excluded employer plans and special needs plans. Our baseline analysis included only HMO plans, the plan type with the longest history. In sensitivity analyses, we included local PPO and then PFFS plans, both  of which are less managed. PFFS plans most closely resemble traditional fee-for-service Medicare.

Study Design

Both county-level and plan-level analyses of the relationship between benchmark rates and bids are inherently limited. Because plans are required to submit a single bid covering all counties they serve, relating the county benchmark to the average bid in any given county ignores the effect of benchmarks in other (often adjacent) counties that may affect the plan’s bid. Alternatively, plan-level longitudinal models require plans to be stable over time. Yet insurers frequently consolidate or split their plans from one year to the next. For example, a “silver” plan’s enrollees in one year may be subsumed under a “gold” plan in the next year, retaining the gold plan’s identification number. Alternatively, the 2 plans may merge and assume a new identification number altogether. Subsequently, this consolidated plan may split again. Such plan dynamics complicate longitudinal models with plans as the unit of the observation.

To combat these limitations, we constructed, for each county, an effective “market” that takes into account the benchmarks in other counties that plans observe when making their bids. First, for each county, we identified all other counties served by plans in the target county. Next, we assigned the benchmarks in these other counties to the target county, weighted by the share of each target county plan’s enrollment in these other counties. We used the plan’s average share across the study period so that the weights were stable. This procedure was repeated for other variables. The resulting markets are a stable unit of analysis over time, allowing for longitudinal models.

Statistical Analysis

The primary analyses were conducted at the market level. We used a longitudinal model with indicators for each market and year (ie, fixed effects), weighted by the number of Medicare beneficiaries in a market. Our dependent variable was average plan bids in a market in a year. The independent variable of interest was the benchmark rate. Additional independent variables included actual fee-for-service costs, number of insurers offering Medicare Advantage plans, number of total beneficiaries, and average risk score of feefor- service beneficiaries to control for market-level health status. We also included the average risk score of Medicare Advantage beneficiaries in a market to control for possible non-random enrollment into plans.

Indicators for year controlled for any underlying time trends in bids across all markets. Indicators for market gave our model a longitudinal rather than cross-sectional interpretation.¹² Specifically, they allowed us to identify the effect of benchmark rates on bids through within-market changes in benchmarks, rather than across-market differences that are more prone to bias. Restricting the variation in all variables to within-market changes helped us control for unobserved area-level characteristics that are fixed across time, especially those characteristics that might be correlated with benchmark rates. We effectively compared changes in bids in markets that experienced large fluctuations in their benchmark versus counties that saw small benchmark changes.

Our model took advantage of relatively exogenous changes in benchmarks due to the 2007 and 2009 “rebasing” of benchmark payments, when benchmarks in all counties were updated by law using national or regional floor updates that were largely uncorrelated with changes in underlying costs in any specific county. Moreover, by controlling for market-level average fee-for-service costs, we further reduced the potential for confounding related to market-level changes in the costs of delivering care. Since markets overlap geographically, we clustered standard errors by state.¹³ This gave us more conservative (larger) standard error estimates than we would get by clustering standard errors at the market level. We conducted a number of sensitivity analyses of the longitudinal market model.

Given the effect of benchmarks on bids, it is straightforward to calculate the residual effect on rebates (Figure 1). As a confirmatory analysis of our bidding model, we estimated analogous longitudinal models using average market rebates as the dependent variable. For every $1 increase in benchmarks, we would expect bids and rebates to move in the same direction as long as the bid increased by less than $1. In other words, a $1 increase in benchmarks must be accounted for by either a change in bids, a change in rebates, or a change in the amount returned to the Medicare program.

RESULTS

Characteristics of Markets and Plans

Table 1 displays the characteristics of markets and of plans competing in each market. Markets that experienced large (above median) benchmark increases saw an average annual increase of $16.54 per beneficiary per month from 2006 to 2009, whereas markets that experienced small (below median) benchmark increase had average annual increases of $3.33. Compared with markets with small increases, those experiencing large increases had lower average benchmark rates. Plans

serving markets with large benchmark changes submitted lower average bids over the 4 years. Because of lower benchmarks in those markets, raw average rebates were also lower. Average risk scores at both the market and plan levels were comparable across the 2 groups.

Effect of Benchmarks on Bids