Aetna Behavioral Health provided claims data from their national database to examine persistence with AUD medications and healthcare costs incurred. Study patients were drawn from a “real world” population, meaning that patients received AUD pharmacotherapy based on their regular providers’ clinical judgment. AUD pharmacotherapies were prescribed to older and sicker patients than those receiving psychosocial therapy alone. XR-NTX patients had more comorbid diagnoses, higher pre-treatment spending and utilization of outpatient behavioral health services, inpatient services, and emergency services than any other group.
Patients taking XR-NTX persisted with treatment longer than patients receiving oral pharmacotherapies, controlling for demographics and pre-treatment physical health, mental health, and drug abuse comorbidities. Oral and XR-NTX patients persisted with treatment significantly longer than acamprosate and disulfiram patients, suggesting a naltrexone drug effect. XR-NTX patients persisted significantly longer than oral NTX patients, indicating a naltrexone delivery mode effect as well. However, persistence needs to be interpreted carefully, since no direct information on alcohol-related health outcomes was available.
Although direct health outcomes were not available in this study, cost and utilization outcomes were. XR-NTX patients decreased non-pharmacy healthcare spending and utilization of inpatient and emergency services relative to oral pharmacotherapies and psychosocial therapy only. One potential explanation for the decrease in spending and utilization is that patients were avoiding necessary healthcare. Arguing against avoidance is the observation that utilization of outpatient behavioral health services increased in all groups. Further evidence to suggest that patients were not simply avoiding healthcare in the post-treatment period lies in the statistic that over 90% of each medication group had nonzero healthcare spending in both pre- and post-treatment periods (data not shown). The relatively healthy psychosocial therapy only group had lower proportions of patients with nonzero healthcare spending in the pre-treatment (87%) and post-treatment (74%) periods.
This study demonstrated persistence, utilization, and spending patterns similar to those reported in other analyses. 5,7–10,14,17,18,22
Lower levels of persistence in the current study may reflect a more stringent definition of persistence, differences in the health of the study populations, or unique features of the Aetna Behavioral Health system. Spending outcomes differed from a recent study10
of oral NTX spending and utilization because we did not include pharmacy costs in the analyses, but overall patterns were similar. Notably, our cost and utilization results replicated findings that suggest XR-NTX is associated with decreased healthcare costs and utilization compared with oral medications when treating alcohol dependence.18,22
The consistency of results across different patient populations and different analytic approaches lends robustness to these findings. We have also demonstrated an advantage of XR-NTX over oral medications in terms of persistence with treatment. Prior work has found that persistence with XR-NTX is associated with improved drinking outcomes,17
and nonpersistence with oral NTX has been linked with increased healthcare utilization.9
The persistence benefit with XR-NTX provides a plausible explanation for the observed cost and utilization advantages, but the relationship between persistence, health outcomes, and costs/utilization demands further study.
There are several strengths in this study’s design and analytic approach. The data come from Aetna’s nationwide claims and utilization database, which provides a geographically diverse sample of patients and accurate cost and utilization information. All patients included in the study were continuously enrolled with Aetna Behavioral Health for at least 6 months before and after treatment initiation, so there is no loss to follow-up. Survival analysis provided temporal data on medication persistence throughout the follow-up period. The difference-in-differences analytic approach minimized the effects of unmeasured confounding factors by controlling for time-independent baseline differences between groups. Available confounders were explicitly accounted for in multivariate regression models.
In addition to its strengths, the study was subject to several limitations. First, treatment was not randomly allocated and the study was limited in its capacity to control for underlying differences between study groups. Considerable differences in demographics, pre-treatment comorbid diagnoses, and pre-treatment utilization patterns between study groups suggest that unmeasured factors could have influenced treatment allocation and confounded outcome measurements. Important variables that were not available included AUD severity and psychometrics such as motivation to change drinking behavior. High pre-treatment utilization suggests that XR-NTX was prescribed to patients with more severe AUDs. Severity is likely to be a negative confounder that biases the results towards the null, but the overall confounding effects of unmeasured factors and non-random treatment allocation is unknown. At the same time, the number of patients receiving each medication reflected current utilization patterns and the number in each group is relatively large. The large number of patients increases confidence in the stability of the results.
The application of selection criteria to study groups with considerable underlying differences could have introduced selection bias. A lower proportion of patients who received XR-NTX (13%) were excluded than oral medication or psychosocial therapy only patients (54% to 64%). Excluded patients are likely to represent individuals who lost their jobs and lacked continuous enrollment, individuals who were prescribed 2 or more AUD medications, and individuals with single claims over $25,000. These exclusions should preferentially omit individuals with relatively poor outcomes, thereby biasing our results towards the null.
The difference-in-differences method is a repeated measures design that renders the analysis susceptible to regression to the mean. We cannot exclude the possibility that regression to the mean is responsible for the observed spending and utilization patterns over time. The analysis is also susceptible to types I and II statistical errors. However, the consistency of the results across several outcomes increases confidence that the observed differences are not statistical artifacts. It is unlikely that such consistency would be observed if the results were attributable to chance.
Continuously enrolled patients with AUDs who were prescribed XR-NTX persisted with treatment longer and experienced larger decreases in non-pharmacy healthcare spending and utilization than those who received oral medications or psychosocial therapy only. Although this study was not able to account for the cost of medication, XR-NTX demonstrated favorable persistence and utilization patterns among a cohort of patients in clinical practice. Future research on this topic is necessary to clarify the associations between persistence, healthcare spending and utilization, and direct health outcomes. Cost-effectiveness analyses of XR-NTX and oral medications would provide critical information for practicing clinicians and insurance providers.