Association Between Medicaid Prescription Drug Limits and Access to Medications and Health Care Use Among Young Adults With Disabilities

Key Points Question Are policies that cap monthly prescriptions in Medicaid associated with access to medication and health care use among young adults with disabilities in Arkansas and Texas? Findings In this cohort study using difference-in-differences analysis of 28 046 young adults with disabilities, including 8214 in states with a 3-drug limit at age 21 years, the 3-drug limit was associated with lower monthly prescriptions for medications used to treat mental health conditions and higher inpatient admissions among all individuals with disabilities in states with the drug cap policy compared with those in states without this policy. Meaning In this study, state drug cap policies in Medicaid were associated with lower access to medications and higher use of inpatient care.


Main Analyses
The regression model for the difference-in-differences analyses was specified as: = + 1 21 + 2 + 3 * 21 + + + + + In the regression model, the variable 21 is a dummy variable that indicates whether the individual was age 21 or older in that month or quarter and is an indicator for whether the individual resides in one of the drug cap states (Arkansas or Texas). The coefficient of interest is 3 which identifies the change in the outcome of interest before versus after age 21 in drug cap states compared with comparison states. All outcomes were calculated in each month or quarter in the 12 calendar months before and after the month of an individual's 21 st birthday.
The month of an individual's twenty-first birthday was defined as a transition period and not included in the analyses. The regression adjusted for individual sex and race/ethnicity (white, black, or other) as well as whether the individual lived in an urban county (Rural Urban Continuum Codes 1-3) prior to turning age 21. Heteroskedastic-robust standard errors were clustered at the individual level. The functional form of the model differed based on the distribution of the outcomes of interest. Regressions were estimated using a zero-inflated negative binomial model for skewed count outcomes with a high proportion of zeros (i.e., prescriptions, total emergency department visits, inpatient length of stay, and spending) while logistic regressions were used for binary outcomes (i.e., any prescription, more than 3 prescriptions, and any inpatient visit).
All analyses were implemented using STATA/MP version 15. Results with p-values <0.05 were considered statistically significant.

Test for Parallel Trends
The difference-in-difference analysis relies on the assumption that the trends in the outcomes of interest would have been parallel in the absence of the policy change at age 21. Although this "parallel trends" assumption cannot be tested directly, we tested whether this assumption is plausible by testing whether the trends in the outcomes were parallel in the pre-policy period.
The regression model for the parallel trends test was specified as: The sample of all individuals with disabilities includes all Medicaid beneficiaries who were eligible for Medicaid due to a disability prior to turning age 21 and were continuously enrolled in fee-for-service Medicaid in the year before and after turning age 21. The treatment group includes all individuals residing in Arkansas and Texas who were eligible for the drug cap policy at age 21. The control group includes all individuals residing in Colorado, Connecticut, Idaho, Missouri, Nebraska, New Hampshire, Nevada, Virginia, Washington, and Wisconsin who were not eligible for a drug cap policy at age 21. The pre-period includes all prescription drug and health care services in the 12 calendar months before an individual turns age 21 and post-period includes all prescription drug and health care services in the 12 calendar months after an individual turns age 21. Means were calculated by first averaging monthly measures in the 12 months before and after age 21 for each individual and then averaging across all individuals. The sample of all individuals with disabilities includes all Medicaid beneficiaries who were eligible for Medicaid due to a disability prior to turning age 21 and were continuously enrolled in fee-for-service Medicaid in the year before and after turning age 21. The serious mental illness subgroup includes all individuals with disabilities who were diagnosed with schizophrenia and psychotic disorders or bipolar disorder at any time prior to turning age 21. The treatment group includes all individuals residing in Arkansas and Texas who were eligible for the drug cap policy at age 21. The control group includes all individuals residing in Colorado, Connecticut, Idaho, Missouri, Nebraska, New Hampshire, Nevada, Virginia, Washington, and Wisconsin who were not eligible for a drug cap policy at age 21. The pre-period includes all prescription drug and health care services in the 12 calendar months before an individual turns age 21 and post-period includes all prescription drug and health care services in the 12 calendar months after an individual turns age 21.  (2007)(2008)(2009)(2010)(2011)(2012). Notes: Regressions adjusted for covariates listed in methods. Difference-in-differences estimates were calculated among all individuals with disabilities in drug cap states (n=8,214) compared with individuals living in comparison states (n=19,832) and the subgroup of individuals with a serious mental illness in drug cap states (n=1,178) and comparison states (n=2,957). Pre-age 21 means were calculated in the drug cap states (Arkansas and Texas) among all individuals in the 12 calendar months prior to turning age 21. Pre-age 21 means were calculated by first averaging monthly measures in the 12 months prior to age 21 for each individual and then averaging across all individuals. Prescription drug outcomes were measured among all individuals on a monthly basis while health care resource use was measured on a quarterly basis before and after the individual turned age 21. Total spending includes all spending on prescription drugs as well as inpatient and emergency department visits. All results are from the coefficient on the interaction between treated indicator variable and post-policy indicator. All results for count outcomes (total prescriptions) are reported as incidence rate ratios (IRR) from the zero-inflated negative binomial models.  (2007)(2008)(2009)(2010)(2011)(2012). Notes: Regressions adjusted for covariates listed in methods. Difference-in-differences estimates were calculated among all individuals with disabilities in drug cap states (n=8,214) compared with individuals living in comparison states (n=19,832) and the subgroup of individuals with a serious mental illness in drug cap states (n=1,178) and comparison states (n=2,957). Pre-age 21 means were calculated in the drug cap states (Arkansas and Texas) among all individuals in the 12 calendar months prior to turning age 21. Pre-age 21 means were calculated by first averaging monthly measures in the 12 months prior to age 21 for each individual and then averaging across all individuals. Prescription drug outcomes were measured in each of the 12 calendar months before and after the individual turned age 21 and prescriptions in the month of the twenty-first birthday were not included. All results are from the coefficient on the interaction between treated indicator variable and post-policy indicator. All results for count outcomes (total prescriptions and spending) are reported as incidence rate ratios (IRR) from the zero-inflated negative binomial models while results for binary outcomes (more than 3 prescriptions) are reported as odds ratios (OR) from the logistic models.  (2007)(2008)(2009)(2010)(2011)(2012). Notes: Regressions adjusted for covariates listed in methods. Estimates were calculated among all individuals with disabilities in drug cap states (n=8,214) compared with individuals living in comparison states (n=19,832) and the subgroup of individuals with a serious mental illness in drug cap states (n=1,178) and comparison states (n=2,957). Prescription drug outcomes were measured among all individuals on a monthly basis while health care resource use was measured on a quarterly basis in the 12 calendar months before the individual turned age 21. Total spending includes all spending on prescription drugs as well as inpatient and emergency department visits. All results are from the coefficient on the interaction between treated indicator variable and continuous measure of months until 21st birthday in the pre-policy period. All results are reported as incidence rate ratios (IRR) from the zero-inflated negative binomial models except odds ratios (OR) are reported for outcomes from the logistic models (any emergency or any inpatient visit).