This study was conducted using data from the IBM MarketScan Commercial Claims and Encounters Database, the IBM MarketScan Medicare and Supplemental and Coordination of Benefits Database, and the MarketScan Health and Productivity Management Database. The Commercial database contains longitudinal medical and drug information, including paid amounts, for several million individuals (including spouses and dependents) across multiple employer-sponsored private health insurance plans. The Medicare supplemental database contains claims data for retirees with Medicare supplemental insurance through employers and includes approximately 3 million individuals annually. The MarketScan Health and Productivity database includes data on workplace absence, short-term disability, and worker’s compensation for a subset of enrollees in the Commercial database. MarketScan databases are Health Insurance Portability and Accountability Act (HIPAA) compliant. This study was conducted using anonymized patient data from MarketScan and as such review board approval and patient consent were not necessary.
Study Design/Search Strategy
This was a retrospective cohort analysis of HCRU, costs, and workplace productivity comparing a cohort of individuals with depression and OAB (the case cohort) to a propensity score matched cohort with depression but no OAB (the control cohort). Cohorts were constructed from a prevalent population of patients with depression (identified on the basis of medical and prescription claims for depression, Appendix 1 in the electronic supplementary material, ESM). The first diagnosis of OAB (or date of first prescription for an OAB medication) served as index for the case cohort (“OAB index date”). For patients with depression and without OAB, a proxy index date was constructed using the Harvey et al.  method selected randomly from within an April 1, 2012 to December 31, 2015 identification period. The 6-month period prior to the index date was the pre-index (baseline) period and the 12-month period following the index date was the post-index period. The full study period was October 1, 2011 to December 31, 2016.
Individuals in both cohorts were required to be at least 18 years of age with continuous insurance enrollment (in the pre-index through the post-index periods) and have a diagnosis of depression during an October 1, 2011 to December 31, 2015 identification period. Depression was identified on the basis of the presence of a depression diagnosis code on one inpatient claim or two outpatient claims AND one or more prescription claims for an antidepressant medication with first depression identification date (diagnosis or treatment) occurring prior to the index date. For the case cohort, an antidepressant medication prescription claim had to occur within 1 month prior or 1 month following the OAB index date; for the control cohort, within 1 month prior or 1 month following the proxy index date. Patients with OAB were identified using criteria similar to those of previously published studies of patients with OAB [9, 10]. Included were patients with the presence of at least one inpatient or two outpatient claims for an OAB diagnosis (International Classification of Diseases [ICD]-9 or ICD-10; Appendix 1 in ESM) or a prescription claim for a medication indicated almost exclusively to treat OAB (darifenacin, fesoterodine, oxybutynin, solifenacin, tolterodine, trospium, mirabegron; patients who may have been treated for other indications of these medications (i.e., neurogenic bladder) were excluded). A complete list of the ICD-9 and ICD-10 diagnosis codes and prescription drug codes associated with depression and OAB, respectively, is provided in Appendix 1 in ESM.
Patients were excluded if they had a diagnosis or procedure code indicative of pregnancy, malignant neoplasms (cancer), renal impairment, hepatic insufficiency, or organ transplantation during the pre- or post-index observation period or a diagnosis of neurogenic bladder, or physical trauma during pre-index. A diagnosis for bipolar depression, or psychosis, schizophrenia, other psychosis-related disorders, paranoid states, other mood disorders, drug-induced depression, Alzheimer’s disease, Parkinson’s disease, or dementia during the pre- or post-index observation period was also cause for exclusion, as was a long-term care or hospice stay in the pre-index period.
Depression-related HCRU measures included inpatient admissions (acute and non-acute), total bed days per patient, outpatient visits (excluding physician office and emergency room [ER] visits), physician office visits, and ER visits. Claims were considered depression-related if there was a primary diagnosis for MDD in the medical claim (Appendix 1 in ESM). Depression-related pharmacy utilization was defined as the total number of unique medications indicated for treatment of depression (Appendix 1 in ESM). Healthcare costs were captured as per patient costs over the 12-month post-index period and reported costs include depression-related total costs, total medical costs, and total pharmacy costs. Total depression-related medical costs included costs incurred during inpatient admissions (acute and non-acute), outpatient (excluding physician office and ER visits) physician, and ER visits. Depression-related pharmacy costs were calculated using outpatient pharmacy claims. For workplace productivity, absentee measures presented were total time absent from work as well as the proportion with each absence type (sick, leave, disability, recreational). For short-term disability, measures presented were total case days and total payments among all patients and for those using short-term disability.
Covariates and Other Measures
Demographic and enrollment characteristics of the patient population, including age, sex, geographic region, and health plan, were described during the pre-index period. Clinical characteristics described during this period included the Charlson Comorbidity Index, based on the Quan enhanced ICD-9 and ICD-10 set  and the number of Elixhauser comorbidities . Baseline depression treatment characteristics described included types of treatments (selective serotonin reuptake inhibitor [SSRI], serotonin–norepinephrine reuptake Inhibitor [SNRI], atypical antidepressant, atypical antipsychotic, lithium, other antidepressants), cumulative days’ supply of any antidepressant drug, and total number of unique treatments for depression observed per patient. The full list of propensity score matching variables is indicated in Table 1.
For HCRU analysis, propensity score matching was used to balance the case and control cohorts on a range of important baseline characteristics without the level of attrition that may result from an exact matching process on a large set of variables. Patients with depression and OAB (cases) and patients with depression without OAB (controls) were propensity score matched in a 1:1 ratio. Prior to matching, imbalance existed (standardized difference ≥ 0.1) between cohorts in 29 variables. The optimal propensity score model was selected on the basis of the model with the most balance observed between cohorts. Standardized differences were reported for all baseline covariates included in the propensity score model, both before and after matching.
Independent variables in the final selected propensity score model included patients’ age, sex, baseline clinical characteristics, health resource utilization, and other baseline characteristics that were important confounders with outcomes (see Table 1 for propensity score model variables). The analyses included comparisons of depression-related HCRU measures, costs, and productivity measures between the two cohorts. For the matched analyses, statistical tests accounted for matching and included Wilcoxon signed rank test for non-normally distributed continuous variables, paired t test for normally distributed continuous variables, and McNemar chi-square for categorical variables, unless otherwise noted.
Regression models (generalized linear models with gamma distribution accounting for matched pairs) were used to elicit the multiplicative impact of OAB status on total depression-related healthcare costs. These costs were assessed on the basis of the statistical significance of the ratio of costs for patients with OAB versus patients without OAB.
Within the subset with workplace absenteeism data, multivariate regression was conducted to address covariate imbalances between cohorts while for the short-term disability analysis no further covariate adjustment was conducted since cases and controls were observed to be balanced on baseline covariates. Negative binomial models were used to model number of absentee hours and logistic regression models were used to model binary outcomes (absent from work (yes/no); absence type: sick (yes/no), leave (yes/no), disability (yes/no), recreational (yes/no); short-term disability utilized (yes/no)). All data analyses for this study were conducted using SAS version 9.4 (SAS Institute, Cary, NC). The a priori alpha level for all inferential analyses was 0.05 and all statistical tests were two-tailed. Data were evaluated for violations of assumptions underlying the associated statistical tests as appropriate.