This quasi-experimental study examines oral albuterol dispensing and Medicaid PDLs between 2011 and 2018.
Data source for albuterol dispensing
We obtained data on albuterol prescription dispensing to Medicaid enrollees in all states between 2011 and 2018 from the State Drug Utilization Database, which is maintained by the Centers for Medicare & Medicaid Services . This database reports the quarterly number of prescriptions dispensed to Medicaid enrollees by national drug code (NDC). We identified products containing albuterol using a list of NDCs obtained from IBM Micromedex RED BOOK (see Table E1 in Online Supplement for full list) . Our list included all albuterol NDC codes that were active during the study period.
In the State Drug Utilization Database, data are suppressed for any quarterly count less than 11. We imputed suppressed counts in the same manner as prior studies using the same dataset . This method leveraged the discrepancy between the sum of state-level totals and the national total. Imputed counts were weighted inverse to the number of quarters during the year that the count was suppressed, so that counts less frequently suppressed received greater weight (see Methods E1 in Online Supplement for description of the full methodology). When suppressed cells were dropped instead of imputed, results from difference-in-differences analyses were unchanged (see Table E2 in Online Supplement, which contains the results without imputed counts).
Data source for PDLs
For the analysis of oral albuterol inclusion on 2018 PDLs, data on PDLs were obtained from Medicaid program websites. If PDLs were unavailable on websites, published minutes and notes for the state’s Pharmacy and Therapeutics Committee were reviewed for indications of changes. If neither was available, the state’s Medicaid office was contacted to clarify coverage or obtain prior PDL versions. When more than one list was published during 2018, all versions were reviewed to evaluate whether changes in coverage occurred. An albuterol formulation was considered preferred if it was included on the PDL; otherwise, it was considered non-preferred.
The analysis was limited to PDLs for FFS plans. Four states (Hawaii, New Jersey, New Mexico, and South Dakota) did not use PDLs for their FFS plans in 2018  and were excluded, leaving 46 states and the District of Columbia (henceforth referred to as “states”). Of the 47 states, 11 had FFS only (no managed care organization (MCO) plans) in 2018, 13 had both FFS and MCO plans and used the same PDL for both (uniform PDL), and 23 had both FFS and MCO plans but used different PDLs . For these 23 states, the PDLs for MCO plans were not examined.
For each of the 47 states using PDLs in 2018, we determined whether at least one formulation of oral albuterol was considered preferred on the state’s 2018 FFS PDL. Additionally, we calculated total reimbursement for oral albuterol prescriptions across all state Medicaid programs in 2018. We conducted these analyses to determine the degree to which Medicaid programs continue to include low-value oral albuterol on PDLs and to estimate the direct amount of wasteful spending on this drug by Medicaid programs.
Selection of states for difference-in-differences analysis
Nineteen states did not consider any form of oral albuterol as preferred on their FFS PDL in 2018. For these states, PDLs from 2011 to 2018 were reviewed to determine whether any oral albuterol formulations had changed from preferred to non-preferred status. Changes were identified in six states (2012: Idaho; 2016: Connecticut, Delaware; 2017: Michigan; 2018: Kentucky, New Jersey, North Carolina). Three states were not considered, because PDL removal occurred too early or late during the study period to have sufficient pre-intervention or post-intervention data. Two states (Connecticut and Delaware) were not considered because dispensing totals of oral albuterol in the quarter before PDL removal were 0.1–0.3%, levels that were too low to measure impact. In contrast, dispensing totals before PDL removal were high enough to measure impact in Arkansas, which changed oral albuterol syrup from preferred to non-preferred in March 2014, midway through the study period.
To identify potential control states for Arkansas, the percentage of dispensed albuterol prescriptions that were for oral albuterol syrup in 2011 was calculated for each state. This percentage was used because it is unaffected by the number of Medicaid enrollees, in contrast to raw prescription counts. States with a similar percentage of dispensed albuterol prescriptions that were for oral albuterol syrup to Arkansas in 2011 were considered. Iowa was chosen as the control state for several reasons. First, the trends for this percentage during the pre-intervention period between 2011 and 2013 were parallel to Arkansas. Second, Iowa did not experience any changes in albuterol coverage on its PDL between 2011 and 2018. Third, while Iowa transitioned from being fee-for-service only to a combination of fee-for-service and managed care organization plans in April 2016, the state utilized a uniform PDL that applied to both types of plans throughout the study period. Finally, Iowa expanded Medicaid under the Affordable Care Act in January 2014, as did Arkansas.
To examine the association between removal of oral albuterol syrup from Arkansas’ PDL in 2014 and dispensing of this drug, a difference-in-differences analysis was performed using data between 2011 and 2018. The pre-intervention period consisted of 12 quarters between January 2011 and December 2013, and the post-intervention period included 19 quarters between April 2014 and December 2018. The quarter containing March 2014 was excluded. A second difference-in-differences model including only data through 2015 was also estimated, as Iowa started to use MCOs in 2016, while Arkansas had only FFS plans from 2011 to 2018.
The dependent variable was the percentage of albuterol prescriptions dispensed that were for oral albuterol syrup. The model included indicators for quarter to adjust for seasonal patterns in dispensing. Robust standard errors were used.
To test the parallel trends assumption, a linear regression model was fit using only data from 2011 to 2013. Terms included time as a continuous variable, an indicator for Arkansas, and their interaction.