While we found little variation between states in the proportion of beneficiaries who received appropriate medications (in every state at least 80 percent of beneficiaries with schizophrenia or bipolar disorder had at least one claim for an evidence-based medication), there was wide variation in the proportion of beneficiaries with high medication continuity (Figure 3). The proportion of beneficiaries with high medication continuity ranged from more than 80 percent in North Dakota to less than 50 percent in the District of Columbia and Mississippi.
Underlying differences in the demographic characteristics of state Medicaid populations explains some of the state variation in medication continuity. For example, many of the states with the lowest rates of medication continuity had a higher-than-average proportion of African American beneficiaries, who tend to have worse rates of medication continuity. To examine the extent to which state variation in medication continuity reflected differences in state demographic characteristics, we standardized the proportion of beneficiaries with high medication continuity in each state to account for age, gender, and race/ethnicity. The adjusted rate of medication continuity used the observed rates of medication continuity for each demographic group in the state, but applied the rates to a population with the same demographic composition as the entire study population in all 22 states. Thus, the adjusted rate reflects the expected rate for a state if the beneficiaries in that state had more closely resembled the entire study population. Using this technique, we found that while the proportion of beneficiaries who were female or older varied between states that variation had little to no effect on the proportion of beneficiaries in a state with high medication continuity. After adjusting the medication continuity rates to account for state differences in race and ethnicity, some states such as Alabama, Maryland, and Mississippi had slightly better rates of medication continuity whereas some states, including Oklahoma and Nevada, had worse rates of medication continuity. Other states such as the District of Columbia, Louisiana, and New Hampshire saw an improvement in continuity rates of between 5 and 10 percentage points once the measure was standardized to account for differences in race and ethnicity. However, states with below average rates of medication continuity remained below average while those with above average rates remained above average even after adjusting for demographic characteristics. Thus, differences in medication continuity in Figure 3 do not only reflect differences in state demographic characteristics.
FIGURE 3. Gap Between Proportion of Medicaid Beneficiaries with Continuous Use and Any Use of Appropriate Medications in 2007, by State
SOURCE: MAX data, calendar year 2007.
States differed in the proportion of beneficiaries with schizophrenia who received depot antipsychotics in 2007. At the low end, only 2 percent of beneficiaries in the District of Columbia received depot antipsychotics in 2007, whereas 20 percent received depot antipsychotics in Alabama. However, the variation in depot medication use did not appear related to medication continuity within a state. Due to a small sample size, we did not examine state variation in depot antipsychotics for bipolar disorder.
Medication continuity was associated with the region of the country, with states in the South Census region having the lowest proportion of beneficiaries with high medication continuity. Figure 4 illustrates that states in the South consistently scored below the median on our measure of medication continuity (states in the bottom left quadrant of the figure have the worst medication continuity, while those in the upper right quadrant have the best).
FIGURE 4. Medication Continuity Among Medicaid Beneficiaries with Schizophrenia or Bipolar Disorder in 2007, by Census Region
SOURCE: MAX data, calendar year 2007.
NOTE: Dotted lines indicate median value among all 22 states in the study.
This finding could be attributable to systematic differences in the funding or orientation of state mental health systems in different regions of the country; states in the South Census region had a lower proportion of total state mental health funding dedicated to providing services in the community (Table 5). It also might be related to other unmeasured differences between states that vary regionally.
Median per capita SMHA Spending on All Services ($)
Median Percentage ofFunding Spent on Community-Based Services
Median per capitaSMHA Spending on Community-Based Services ($)
|SOURCE: NRI Revenue and Expenditure Reports, FY 2007.|
Given that a beneficiary may live in a state with several different medication policies, we further examined the relationship of these variables with medication continuity using random effects regression analyses to model the odds that an individual had high medication continuity as a function of several beneficiary and state characteristics (the full regression models are presented in Table A.10 of the technical appendix).9 These exploratory regression analyses were conducted to identify factors that may be associated with medication continuity while controlling for other factors. However, this study was not able to measure every beneficiary, state, or Medicaid program characteristic that could influence medication continuity. Therefore, the regression findings should be interpreted cautiously since other unmeasured state or beneficiary characteristics may be associated with medication continuity.
For beneficiaries with schizophrenia, the regression results suggested that a two-dollar copayment for preferred or generic medications (Odds Ratio [OR]): 0.45; 95 percent Confidence Interval [CI]: 0.27-0.76) or a $3 copayment (OR: 0.73; 95 percent CI: 0.57-0.94) versus no copayment were inversely associated with high medication continuity. In addition, a $1 copayment for branded medications (OR: 0.62 95 percent CI: 0.41-0.95) compared with no copayment was inversely associated with medication continuity whereas $2-$3 copayments were not. It may be that states exempt certain branded medications used for schizophrenia from these higher copayment requirements, but we cannot determine that using claims data alone. These findings suggest that a beneficiary with schizophrenia living in a state with no medication copayment requirement for generic medications who had a 70 percent probability of high medication continuity given his age, race, and co-occurring conditions, would have only a 63 percent probability of high medication continuity in a state that was otherwise similar with respect to the Medicaid and mental health system characteristics that we controlled for in the model but that required a copayments of $3. It is possible that beneficiaries could acquire medications outside of Medicaid-funded programs if faced with the challenge of paying a copayment. Nonetheless, these findings suggest that living in a state with a higher copayment for generic drugs was associated with a higher likelihood of medication interruptions.
The regression results also suggested that enrollment in an HMO was inversely associated with medication continuity (OR: 0.72; 95 percent CI: 0.67-0.75), but these findings should be interpreted with caution because few HMOs were included in the study. The regression confirmed the descriptive findings in that beneficiaries who were African American (OR: 0.45; 95 percent CI: 0.44-0.47), Hispanic (OR: 0.69; 95 percent CI: 0.66-0.73), or another or unidentified race and ethnicity (OR: 0.87; 95 percent CI: 0.83-0.91) had significantly lower odds of medication continuity compared with nonHispanic Caucasian beneficiaries. The presence of substance abuse disorder or cardiovascular disease was inversely associated with medication continuity, whereas the presence of diabetes was positively associated with medication continuity. Also consistent with the bivariate findings, a 1 year increase in age was associated with slightly better medication continuity. Controlling for other variables in the model, medication continuity was slightly worse among beneficiaries who lived in a county that was designated as a mental health provider shortage area (OR: 0.94; 95 percent CI: 0.92-0.98). Medication continuity was slightly better among beneficiaries who had at least one claim for a psychosocial service (OR: 1.07; 95 percent CI: 1.04-1.11), Finally, controlling for other variables in the model, medication continuity was better among beneficiaries who lived in states in which a higher percentage of state mental health funding was dedicated to community-based services rather than institutional care (OR: 1.06; 95 percent CI: 1.01-1.11).
We conducted similar multivariate analysis to model the odds of high medication continuity among beneficiaries with bipolar disorder.10 The regression results suggested that the number of prior authorization policies for different classes of mental health drugs (antidepressants, antipsychotics, and anticonvulsants) was inversely associated with high medication continuity such that the odds of medication continuity slightly decreased with each additional prior authorization requirement (OR: 0.90; 95 percent CI: 0.82-1.003), although this was only marginally statistically significant (p=0.058). Copayment amounts for generic medications were not statistically associated with medication continuity whereas, similar to our finding for beneficiaries with schizophrenia, a $1 copayment requirement for branded medications was inversely associated with medication continuity (OR: 0.52; 95 percent CI: 0.31-0.92). Copayment amounts may have a weaker relationship with medication continuity for bipolar disorder versus schizophrenia because of differences in the populations. Individuals with bipolar disorder may be able to maintain their functional status, employment, and relationships for longer periods of time than those with schizophrenia--which may provide them with more resources to pay for their medications. The medications used for bipolar disorder are also different from the medications for schizophrenia; states may exempt some medications for bipolar disorder from copayments but we do not have that detailed information.
As was the case with schizophrenia, medication continuity was inversely associated with enrollment in an HMO (OR: 0.79; 95 percent CI: 0.73-0.86) or the presence of a comorbid substance abuse diagnosis (OR: 0.49; 95 percent CI: 0.46-0.53) or cardiovascular disease (OR: 0.86; 95 percent CI: 0.77-0.96) while being positively associated with diabetes (OR: 1.37; 95 percent CI: 1.28-1.46). It may be that beneficiaries with diabetes have more regular contact with the health care system and thus more opportunities to receive their medications. Conversely, those beneficiaries with high medication continuity but poor medication monitoring may be more prone to developing diabetes. We cannot determine the direction of the relationship between the presence of comorbid conditions and medication continuity using a single year of cross-sectional claims data. Finally, similar to our findings for beneficiaries with schizophrenia, those beneficiaries with bipolar disorder who had at least one claim for a psychosocial service were slightly more likely to have high medication continuity (OR: 1.13; 95 percent CI: 1.07-1.20)
Both regression models for schizophrenia and bipolar disorder originally controlled for whether the state Medicaid program had monthly medication supply or refill limits. These variables were not statistically associated with medication continuity and were therefore excluded to develop the most parsimonious models. Their exclusion did not change the statistical significance, direction, or magnitude of the other coefficients. However, the lack of a statistical relationship should not be interpreted to imply that these Medicaid program characteristics are unimportant; we may have lacked enough variation to detect a relationship or these variables could have been measured imperfectly.
Given that beneficiaries in California accounted for 35 percent of the entire study population and 6o percent of beneficiaries in the sample who were enrolled in HMOs, we conducted sensitivity analyses to repeat these regression analyses excluding California. The direction, magnitude, and statistical significance of the coefficients did not substantially change when beneficiaries from California were excluded. HMO enrollment, in particular, was still inversely associated with high medication continuity, suggesting that the large number of beneficiaries from California in the study were not solely driving the findings. Nonetheless, the study included relatively few beneficiaries enrolled in HMOs from a few states such that the findings must be interpreted with caution. Finally, as previously mentioned, the study did not have reliable measures of every beneficiary, mental health system, or Medicaid program characteristic that could influence medication continuity. There may be unmeasured systematic differences between states that could confound the results. These exploratory findings point to certain Medicaid policies and features of mental health systems that may facilitate or impede medication continuity. Further research is needed to replicate these findings and understand the dynamic and complex way in which beneficiary, mental health system, and Medicaid program characteristics interact at the state level to potentially influence medication continuity as well as the receipt of other EBPs.
We considered including a variable in the model to serve as a proxy for the relationship between the state Medicaid and mental health agency to examine whether the strength of the relationship between the two agencies was associated with the delivery of EBPs. Unfortunately, the only variable that we could find from 2007 (the year of our claims data) to serve as a weak proxy for this relationship was from a survey conducted by NRI and available online through the NRI State Mental Health Agency Profiling System, which asked SMHA directors to report whether the state Medicaid and mental health agency “combined or coordinated funding.” This variable did not indicate whether the combining or coordination of funding was directed toward EBPs and we did not have any further information from the survey to contextualize or describe the specific nature of how or in what manner funding was combined or coordinated in each state. Thus, this variable could have different interpretations or meanings for each state. In addition, this variable was missing for the District of Columbia and two states (Mississippi and Missouri) in our study, which would require that these states to be dropped from the analysis if this variable was included in the regressions. Omitting these states would result in a sizeable reduction of our sample--9.7 percent of beneficiaries with schizophrenia and 13.6 percent of beneficiaries with bipolar disorder. Finally, this variable lacked variability among the remaining states. The seven states that did not coordinate or combine funding included Alaska, California, Idaho, Illinois, South Dakota, West Virginia, and Wyoming. With the exception of California and Illinois, these states have a small number of beneficiaries with schizophrenia or bipolar disorder, and therefore, in the context of these regression analyses, this variable may be functioning as a proxy for California and Illinois rather than serving as a true measure of whether states combine or coordinate funding. When we attempted to include this variable in the regressions, we found that it produced unstable estimates that changed in direction and statistical significance. We also did not see a clear relationship between medication continuity and whether the Medicaid and mental health agencies combined or coordinated funding in bivariate analyses. Given these limitations, we did not include this variable in our final regression models. There may be opportunities in future research to develop or include better measures of the degree to which Medicaid and mental health agency collaboration influences the delivery of EBPs.
FIGURE 5. Proportion of Medication Users in Each State Receiving Recommended Medication Monitoring or Health Screening in 2007
SOURCE: MAX data, calendar year 2007.
NOTES: Each diamond in the figure represents the proportion of beneficiaries in a state who received medication monitoring.