How Does Drug Coverage Vary by Insurance Type? Analysis of Drug Formularies in the United States | Page 3
Published Online: April 21, 2014
Stephane A. Regnier, PhD, MBA
Table 4 shows the impact of introducing additional variables (see online Appendix 3 for additional details). As expected, the sums of the coefficients for P-TKIs and for the interaction between the Medicare/commercial Medicaid plans and P-TKIs were negative and positive respectively (P <.001), which means that Medicare (respective to commercial Medicaid) plans covered fewer (respective to more) P-TKIs than HMGs in a low tier. In other words, Medicare plans had a substantially lower proportion of P-TKIs than on-patent HMGs in tier 1 or 2. Commercial Medicaid plans were less restrictive for P-TKIs than for HMGs. The average number of competitor plans had a significant negative effect for P-TKIs (P = .05). The more competition faced by plans, the less likely they were to cover P-TKIs in the low copay tiers. However, the number of competitor plans did not impact coverage of on-patent ARBs differently from coverage of HMGs. Overall, the number of competitors was not significant across therapeutic areas (F test P = .15). The other variables remained significant, with similar coefficient values. The interaction variables added some information to the core model (Akaike’s information criterion of 0.99 vs 0.96, respectively).
Finally, a variable measuring the number of insured lives by plan was introduced to assess whether the size of a plan had an impact on on-patent drug coverage. The hypothesis was that larger plans had lower general marketing expenses (as percentage of sales) due to economies of scale, and could afford to cover more on-patent drugs in tiers 1 and 2. The variable measuring insured lives was significant and positive, meaning that plans with more enrollees covered, on average, more on-patent drugs in tiers 1 and 2. All variables remained significant after adding this variable and analyzing a subset of the original data set. One variable (commercial Medicaid) experienced a significant switch in sign. The regression did not include municipal plans or discount prescription programs since only 1 of 14 discount prescription programs and 79 of 235 municipal plans had information on lives covered. Since the data set with lives data is only a subset of the universe, care must be taken when comparing the results, in particular Akaike’s information criterion, of previous regressions.
Logit Model Specification Tests
The logistic model with interaction terms passed the Pregibon link test but not the Hosmer–Lemeshow test (F test P = .46 and P = .01, respectively). The model with interaction terms and lives covered variable passed both tests (F-test P = .85 and P = .23, respectively) and its goodness of fit was not rejected.
Finally, a bootstrap estimation (using 1000 bootstrap replications at 95% CI) was performed for the GLM, with interaction variables. The coefficients from the bootstrap estimates were found to be similar to those of the GLM. The only difference was that the commercial Medicaid variable became significant. Based on these results, the coefficients’ estimates and significance levels from the GLMs were not driven by outliers.
Tobit and Negative Binomial Models
The sign and significance of coefficients were similar for the Tobit and negative binomial models (Table 5). The sign and significance of coefficients were also similar to those of the logit model (Table 5).
To the author’s knowledge, this article is the first report of an attempt to analyze drug tier data in this way. Across all models and therapeutic areas, Medicare plans and discount prescription programs had consistently fewer on-patent drugs in the lowest copay tiers than did commercial plans. These results were expected. The revenue per enrollee of Medicare PDP and MA plans is capped, since these plan providers have to offer standard plan designs (or designs at least actuarially equivalent to the standard plan designs). Therefore, these plans may try to contain costs to increase profits. Discount prescription programs are basic drug insurance programs that mainly cover generics. Conversely, PBMs, employer, and union plans had higher on-patent drug coverage than commercial plans. Therefore, employees had access to more on-patent drugs in the low copay tiers when companies outsourced pharmacy benefits to a PBM and when companies created their own formulary. State Medicaid covered more on-patent drugs in a low-copay tier than any other types of plans. This result reflects the fact that 46 out of 49 Medicaid plans only had 2 tiers. In addition, only 1 Medicaid plan (Mi Salud, Puerto Rico Medicaid) did not reimburse all on-patent drugs. One explanation could be that 2 tiers may be enough to convince patients to use drugs in tier 1 since a marginal copay difference may dissuade patients on low incomes from using a drug. Alternatively, Medicaid plans may have high bargaining power, and/or pharmaceutical companies may be willing to offer substantial rebates to grant access to this segment of the population. Medicaid administrators may also limit cost-sharing for on-patent drugs so that cost is not a barrier to compliance with treatment regimens.
The significance of coefficients for commercial Medicaid and municipal plans varied by therapeutic area and/ or methods. Commercial Medicaid plans covered, on average, more P-TKIs (all within patent) than did commercial plans as a whole. This was not the case for HMGs (Table 5). One explanation may be that state Medicaid agencies, when negotiating with commercial Medicaid companies, include clauses regarding maximum allowable cost sharing for classes with generics.
Across all plans, the percentage of on-patent drugs covered in a low tier was substantially higher for P-TKIs than for HMGs and ARBs, despite P-TKIs being more expensive. One explanation might be that the high unmet medical need associated with CML justifies an affordable choice (ie, the clinical data outweigh the economics). On-patent ARBs were slightly less well covered than on-patent HMGs, despite a lower degree of generic competition. Plan providers might consider that other classes of antihypertensives (eg, calcium channel blockers or beta-blockers) are acceptable alternatives to ARBs, and thus do not feel compelled to place numerous drugs in tiers 1 or 2.
Plan competition had a negative impact on P-TKI coverage, but no significant impact on on-patent ARB or HMG coverage. In other words, when on-patent drugs were more scarce and expensive (such as P-TKIs), plans in competitive states limited the number of on-patent drugs with a low copay, possibly to reduce costs. The key findings were robust: the coefficients’ signs and significance levels remained similar for the core variables when: (1) additional covariates were added; (2) alternative models were applied; and (3) subsets of data were analyzed. Additionally, the most comprehensive logit model passed the goodness-of-fit tests.
The analysis presented in this paper could have some public health implications. Newhouse argued that the optimal cost sharing indicated by the RAND health insurance experiment (HIE) (25% coinsurance) may be suboptimal for medications for chronic diseases, especially medications whose benefits only become apparent in the long term.26,27 For instance, Goldman et al found that compliance with cholesterol-lowering therapy was associated with a reduction in the annual rate of hospitalization. However, the study authors also found that higher levels of cost sharing were associated with a reduction in compliance.28 As such, Medicare plan designs may be suboptimal because they rely heavily on cost-sharing (ie, tier 4) for expensive chronic therapies such as P-TKIs.
There were some limitations in the data. First, coverage information was not complete for all plans. For example, of 1768 plans in the database, 1579 plans had coverage information for all on-patent HMGs; 1673 for all ARBs; and 1631 for all P-TKIs. Only plans that had coverage information for all on-patent drugs within a therapeutic area were included in the analysis, which could, therefore, suffer from censoring issues. Of the plans that did not have information for all HMGs, 39% were commercial plans and 29% were employer plans. In the whole Fingertip Formulary database, commercial and employer plans represented 22% and 11% of the plans, respectively. There was a high overlap across therapeutic areas between plans without coverage information for all products. For instance, 92% of the plans that did not have coverage information for all ARBs did not have information for all HMGs.
Second, clear conclusions for municipal plans could not be derived because they are heterogeneous, as described in the Background and Objectives sections.
Third, only 11 union plans were included, which made it difficult to draw robust conclusions. If prescription data were available, a Herfindahl index based on the market share of plans may be a better measure of competition than a plan count.
Finally, the analysis was a snapshot of the situation in 2011. As more generics launch in each therapeutic area and as more new drugs become available, the results of this study may not be applicable.
PDF is available on the last page.