Setting and data sources
Montefiore Medical Center is an academic medical center and integrated health care delivery system in the Bronx, NY. In addition to 4 hospitals (over 80,000 total inpatient admissions annually) and 4 emergency departments (over 300,000 total visits annually), Montefiore provides primary and specialty care through a network of outpatient clinics with over 3 million visits annually. We conducted a retrospective cohort study using all de-identified patient, emergency department visit, outpatient visit, inpatient hospitalization, problem list, and prescription data from the Montefiore electronic health record between January 1, 2010 and December 31, 2016. The Montefiore Medical Center/Albert Einstein College of Medicine Institutional Review Board approved this study.
Study population
We included all patients: (1) 18 years and older, (2) with an AUD diagnosis, defined as an International Classification of Diseases, 9th Edition, Clinical Modification (ICD-9-CM), code or free-text problem list entry indicating an AUD (Additional file 1: Appendix 1) entered between January 1, 2010 and December 31, 2014. Given evidence that AUD remains under-diagnosed among patients with regular contact with the health care system [4], we included problem list entries “alcohol consumption binge drinking” and “alcohol consumption heavy” as evidence of AUD diagnosis, while also tracking how many entered our cohort with either of these problem list entries. A diagnosis code indicating AUD in remission did not qualify for cohort entry. We excluded patients if they were prescribed AUD pharmacotherapy within the 6 months prior to the AUD diagnosis. Patients entered the cohort at the time of AUD diagnosis and we extracted data on each patient for 2 years after this index date.
Main exposure: level of primary care engagement
Our main exposure was level of primary care engagement within 2 years of AUD diagnosis. Similar to previous research [19], we coded primary care engagement as a three-level categorical variable: no primary care, limited primary care, or engaged with primary care. We defined “no primary care” as no visit to a primary care facility (i.e., internal or family medicine primary care clinic). We defined “primary care engagement” as one visit to a primary care facility (among 21 primary care facilities within the Montefiore Medical Center network) within 1 year of AUD diagnosis and at least one additional visit at the same facility between 90 and 365 days after the first visit. We defined “limited primary care” as one or more visits to a primary care facility but no additional visits to the same primary care facility between 90 and 365 days after the first primary care visit.
Main outcome: prescription of AUD pharmacotherapy
For our primary outcome, we defined AUD pharmacotherapy prescribing as at least one instance of a provider prescription for acamprosate, naltrexone, topiramate, or disulfiram within 2 years of AUD diagnosis. This approach includes all medications approved by the Food and Drug Administration plus topiramate, which we included given the evidence for efficacy within a 2014 meta-analysis [6]. We examined the percentage of patients who had any AUD pharmacotherapy prescription ordered, and for each individual medication.
Demographic and clinical characteristics
For eligible patients, we extracted demographic and clinical characteristics, and health care utilization data. For demographic characteristics, we extracted: age, sex (male, female), race/ethnicity (Hispanic/Latino of any race, White non-Hispanic, Black non-Hispanic, and other/unknown), and insurance status (private, public, self-pay/unknown). For clinical characteristics, we extracted: year of cohort entry (i.e., year of AUD diagnosis), Charlson comorbidity index based on ICD-9-CM codes 1 year prior to AUD diagnosis [20, 21], and presence of a psychiatric comorbidity within 1 year prior to AUD diagnosis (yes, no; based on an ICD-9-CM code or problem list entry indicating depression, anxiety, schizophrenia, or bipolar disorder, Additional file 1: Appendix 2). For health care utilization, we extracted the number of: emergency department visits, inpatient hospitalizations, outpatient psychiatric visits, and outpatient substance use disorder treatment visits within 2 years of AUD diagnosis.
Statistical analysis
First, by level of primary care engagement (no primary care, limited primary care, and engaged with primary care), we compared demographic and clinical characteristics, and health care utilization. For categorical variables we used Pearson Chi square test, and for continuous variables we used Kruskal–Wallis tests.
Next, to assess the association between level of primary care engagement (main exposure) and AUD pharmacotherapy prescribing (main outcome), we used a multivariable logistic regression model. We included age, sex, race/ethnicity, year of cohort entry, insurance status, Charlson score, psychiatric comorbidity, emergency department visits, inpatient hospitalizations, and outpatient psychiatric and substance use specialty visits in the model because all were associated with the primary outcome during bivariate testing with a P value < 0.2. We did not find evidence of collinearity between variables in our final model. We present logistic regression results as predictive margins and adjusted differences based on recycled prediction with all analyses completed in Stata 13 (StataCorp, College Station, TX).
Sensitivity analysis
We conducted two sensitivity analyses to determine the robustness of our findings. To account for alternative indications for prescribing topiramate, we removed patients with epilepsy and migraine headache syndromes (based on ICD9-CM codes or problem list entries, Additional file 1: Appendix 3) noted within 1 year prior to cohort entry. To account for a more narrow definition of AUD, we repeated our analysis while excluding patients with either “alcohol consumption binge drinking” and “alcohol consumption heavy” problem list entries.