Supplements Economic Considerations for the Treatment of Opioid and Alcohol Dependence
Alcohol Dependence Treatments: Comprehensive Healthcare Costs, Utilization Outcomes, and Pharmacotherapy Persistence
Objectives: To determine the healthcare costs associated with treatment of alcohol dependence with medications versus no medication and across the 4 medications approved by the US Food and Drug Administration (FDA).
Study Design: Retrospective claims database analysis.
Methods: Eligible adults with alcohol dependence were identified from a large US health plan and the IMS PharMetrics Integrated Database. Data included all medical and pharmacy claims at all available healthcare sites. Propensity score–based matching and inverse probability weighting were applied to baseline demographic, clinical, and healthcare utilization variables for 20,752 patients, half of whom used an FDA-approved medication for alcohol dependence. A similar comparison was performed among 15,502 patients treated with an FDA-approved medication: oral acamprosate calcium (n = 8958), oral disulfiram (n = 3492), oral naltrexone (NTX) hydrochloride (n = 2391), or extended-release injectable naltrexone (XR-NTX; n = 661). Analyses calculated 6-month treatment persistence, utilization, and paid claims for: alcoholism medications, detoxification and rehabilitation, alcohol-related and nonrelated inpatient admissions, outpatient services, and total costs.
Results: Medication was associated with fewer admissions of all types. Despite higher costs for medications, total healthcare costs, including inpatient, outpatient, and pharmacy costs, were 30% lower for patients who received a medication for their alcohol dependence. XR-NTX was associated with greater refill persistence and fewer hospitalizations for any reason and lower hospital costs than any of the oral medications. Despite higher costs for XR-NTX itself, total healthcare costs were not significantly different from oral NTX or disulfiram, and were 34% lower than with acamprosate.
Conclusion: In this largest cost study to date of alcohol pharmacotherapy, patients who received medication had lower healthcare utilization and total costs than patients who did not. XR-NTX showed an advantage over oral medications in treatment persistence and healthcare utilization, at comparable or lower total cost.
(Am J Manag Care. 2011;17:S222-S234)
The dominant mode of treatment of alcohol dependence is psychosocial treatment or counseling, and several models have shown evidence for effectiveness.6 Although 4 medications have been approved by the US Food and Drug Administration (FDA) for the treatment of alcohol dependence, there is little adoption of these agents.7,8 Survey results published in 2007 reported that pharmacotherapies for substance-use disorders (SUDs) were offered in less than 25% of public and private specialty treatment programs7 and a 2007 study reported that SUD medications comprised less than 1% of all SUD treatment costs.8 Nevertheless, the National Institute on Alcohol Abuse and Alcoholism has issued recommendations stating that medications are “helpful to patients in reducing drinking, reducing relapse to heavy drinking, achieving and maintaining abstinence, or a combination of these effects” and clinicians should “consider adding medication whenever [they] are treating someone with active alcohol dependence.”6
There are multiple reasons why medication-assisted treatment (MAT) for alcohol dependence is not widely used, including long-standing traditions rooted in the mutual help movement, but adoption of MAT is also predicated on concerns about poor patient adherence to medication, modest efficacy, and poor costeffectiveness.9-11 Retrospective insurance database studies of oral medications have reported that 50% of patients fail to obtain their first refill,12,13 and refill rates are worse for alcoholism medications than for statins and psychiatric medications.14 Clinical trials have found that medication adherence is crucial to efficacy.15
Medication adherence in substance-dependence treatment has been a priority concern of the National Institutes of Health for over 3 decades.16 In 2006, the FDA approved the first extended-release formulation for the treatment of alcohol dependence, extended-release naltrexone (XR-NTX), which was designed to address the challenge of adherence through a once-monthly injection.17 Of the 4 agents FDA-approved for the treatment of alcohol dependence studied in a retrospective claims analysis of commercial insureds, XR-NTX was associated with reduced estimated charges and utilization of inpatient detoxification days and alcoholism-related inpatient days, compared with all 3 oral agents (ie, oral naltrexone, disulfiram, and acamprosate calcium).18 Given the importance of alcohol dependence treatment for public health and healthcare cost containment, the present study was designed to extend current knowledge of real-world effectiveness with alcohol dependence treatments, including treatment with no medication, any approved medication, and among the approved medications, treatment with each specific agent. This study sought to examine a larger cohort of insured patients treated with XR-NTX than previously studied, and to determine a comprehensive range of healthcare utilization and actual expended healthcare costs for each treatment category.
Data Sources and Study Population
This was a retrospective database analysis conducted using commercial enrollees from a large US health plan affiliated with i3 Innovus and the PharMetrics Integrated Database from 2005 to 2009. These databases included medical and pharmacy claims from all available healthcare sites (inpatient, hospital outpatient, emergency department [ED], physician’s office, and surgery center) for virtually all types of provided services, including specialty, preventive office-based treatments, and retail and mail order pharmacy claims.
For the comparison of the “no medication” group to the “any medication” group, patients were required have at least 1 claim for alcohol dependence (Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, code 303.xx) during the pre- or post-index period, have an alcohol use disorder diagnosis pre-index, and have at least 6 months of continuous enrollment pre-index and 6 months post-index. The earliest pharmacy claim for alcohol medication was set as the index date for the any medication group. The index date was defined as the first medical claim for a nonpharmacologic treatment such as a detoxification facility claim, a substance abuse treatment facility claim, or a substance abuse counseling claim. Patients in the nonpharmacologic substance group had no prescription fills for alcoholism medication while patients in the any medication group had at least 1 fill for any of the 4 alcoholism medications. Patients with liver failure during the pre-index period were excluded. Furthermore, patients were excluded if they had claims for pharmacological treatment in the month prior to the index date (with the exception of the XR-NTX group, because this group was occasionally required to demonstrate prior oral medication failure). These inclusion/exclusion criteria led to a final sample of 20,670 patients in the no medication group and 15,502 patients in the any medication group. Figure 1 presents the sample sizes after applying the inclusion/exclusion criteria.
Similar criteria were required for patients in the comparison of the 4 alcoholism medications. Patients treated with XR-NTX were identified on the basis of an outpatient drug claim using the National Drug Code (NDC) or medical claims with the Healthcare Common Procedure Coding System code. The other medications, such as oral naltrexone, disulfiram, or acamprosate were identified using outpatient drug claims based on NDCs. The final sample of 661 patients in the XR-NTX group, 2391 patients in the oral NTX group, 8958 patients in the disulfiram group, and 3492 patients in the acamprosate group, was identified after applying the inclusion/exclusion criteria.
We derived demographic and clinical characteristics of the study populations at baseline. In particular, age, sex, and geographic location were measured at the index date. Deyo-Charlson comorbidity score,19 Elixhauser score,20 and the number of distinct psychiatric diagnoses and medications were calculated during the pre-index period. The Deyo-Charlson comorbidity score is an International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) code adaption of the Charlson index, which assigns a range of weights, from 1 to 6 according to disease severity, for 19 conditions. The Elixhauser score is also a claims-based comorbidity index which sums a patient’s comorbid conditions from among 30 ICD-9-CM comorbidity flags, differentiating secondary diagnoses from comorbidities by using diagnosis-related groups.
For socioeconomic status (SES), we constructed a summary measure for each US Zone Improvement Plan (ZIP) code using data on income, education, and occupation from the 2000 US Census and then linked this information to the patient’s ZIP code of residence in the analytic files.21 Factor analysis was used to identify 6 census variables that could be meaningfully combined into a summary socioeconomic status score. These variables included 3 measures of wealth/income (median household income, median value of housing units, and proportion of households with interest, dividend, or rental income), 2 measures of education (proportion of adult residents completing high school and college), and 1 measure of occupation/employment (proportion of employed residents with management, professional, and related occupations).22
Healthcare utilization and costs were calculated during both the pre-index and post-index periods. In terms of inpatient utilization, the number of detoxification facility days, and the number of detoxification and/or rehabilitation (admissions with an ICD-9-CM procedure for detoxification or rehabilitation), alcohol (admission with a principal diagnosis), and non-related inpatient admissions were measured. ED visits, alcohol-related physician visits, alcohol and substance abuse psychosocial provider visits, and non-alcohol-related outpatient visits were calculated. Utilization measures were presented per 1000 patients. Associated costs related to these measures and total costs were also calculated.
In addition to healthcare utilization and costs, we evaluated adherence by analyzing medication possession ratio and days of persistence with index medication refills post-index date.
Baseline characteristics were compared between the patient cohorts, and descriptive statistics were calculated as percentages and standard deviations. Differences between the cohorts were analyzed using the t-test, Mann-Whitney U test, and chi-square test, and standardized differences were calculated. It has been demonstrated that standardized differences 10% and higher between the baseline variables are significant, and need to be adjusted to compare the outcome measures among the groups.23,24
Propensity-score matching was applied to compare the risk-adjusted outcomes between the no medication group and the any medication group. Propensity-score matching is a technique that aims at adjusting for selection bias in nonexperimental, nonrandomized, and retrospective studies like the present one.25 By using propensity-score matching, each patient in the any medication group was “mirrored” by a patient with similar predefined characteristics in the no medication group. The following characteristics were used to match: age, sex, region, comorbid scores, SES, baseline healthcare utilization, and costs. Logistic regression was used to estimate propensity scores. Several interaction variables were constructed, but they were not determined to be significant. Estimation power of the logistic regression was determined by C statistics. Following the guidelines set forth by Baser, it was determined that one-to-one matching created the best balance among the groups.26
Following Imbens and Lechner, we applied propensity-score matching that accounts for multilevel treatments when comparing the 4 alcoholism medication groups.27,28 Several applications of this method are presented in the medical literature.29-31 The first step uses multinomial logistic regression to estimate conditional probabilities of being in the particular treatment group. The second and final step estimates conditional expectation of outcome given the treatment level. Adjusted Wald tests were performed to test for the difference in weighted characteristics across the treatment cohorts.