Patient Population and Data Collection
We utilized the Japan Medical Data Center (JMDC) claims database from January 2010 to December 2015. The database includes health insurance claims data from non-governmental employees together with their family members. It also provides information on patient demographics, diagnostic codes, dates and types of procedures, dispensed prescription drugs, medical services available to inpatients and outpatients, as well as direct costs and expenditures. The JMDC database has been used to investigate a wide range of conditions in Japan such as schizophrenia, diabetes, and cardiovascular disease [12, 13]. To protect patient confidentiality, all personally identifiable information was de-identified, and as such no informed consent was deemed necessary.
We included RA patients with at least two ICD 10th diagnoses of RA (M05, M06.0, M06.2-M06.9). Moreover, patients were required to have received at least two prescriptions for RA treatment [DMARDs/Biologic]. Patients less than 18 years of age and affected with Crohn’s disease, ankylosing spondylitis, juvenile arthritis, psoriasis, ulcerative colitis, psoriatic arthritis, and/or Behçet’s disease were excluded from the analysis. Depression as a comorbidity was defined when a patient had at least two ICD 10th diagnoses of depression such as major depressive disorder, dysthymic disorder, depressive disorder NOS, and depressive mood (F03, F32.0, F32.1, F32.2, F32.8, F32.9, F33.1, F33.2, F33.3, F33.9, F34.1, F34.9, F41.2, and F53.0). In addition, patients were required to have received at least two prescriptions for treatment of depression including selective serotonin reuptake inhibitors (SSRI), serotonin noradrenaline reuptake inhibitor (SNRI), tricyclic antidepressant (TCA), monoamine oxidase inhibitors (MAOI).
Statistical Analysis
Since there is a potential imbalance in baseline covariates between patients with and without depression, we performed a propensity score matching to eliminate the impact of confounding parameters. This procedure allows us to estimate resource and costs attributable to depression in patients with RA. Each patient of the reference cohort was matched with up to four comparison cases with the closest propensity score using a greedy algorithm without replacement (i.e., once a match is made, the match is not reconsidered). The maximum tolerated difference between matched subjects in a “non-perfect” matching (i.e., caliper matching) was 0.2 × standard deviation of the logit of the propensity score [14].
The propensity score was assessed using a multivariable logistic regression model. Each potential covariate of interest was first tested in a univariate model and retained for the multivariate logistic regression model if P < 0.10. The potential confounders include age at index date, gender, calendar year of index-date, comorbidity measures (i.e., Charlson comorbidity index within the pre-index period, renal failure, interstitial pneumonia, COPD, peptic ulcer, chronic liver disease, osteoporosis, diabetes), polypharmacy in the pre-index period, and follow-up duration. The Charlson comorbidity index (CCI) is used to measure 19 comorbidities by assigning a weight between 1 and 6 to each one with higher CCI indicative of greater morbidity affecting the patient [15].
The Kaplan–Meier survival plots were used to estimate the proportions of patients with at least one occurrence of health care utilization (HCU), while the difference between the two groups (RA/depression and RA only) was assessed by the log-rank test. An adjusted Cox regression model was used to assess the relative risk of occurrence of the event between RA patients with depression and the matched population. Independent variables were the index depression status at index date (RA/depression and RA only) and patient characteristics.
The average number of occurrences per person-month attributable to depression was calculated among patients with at least one occurrence and reported at 6 and 12 months.
To assess the difference between the two groups (RA/depression and RA only), we also estimated a ratio between number of occurrences of health care utilization in the “RA/depression” group and in the “RA only”, by health care utilization category and among patients with at least one occurrence.
The incidence rate in turn was calculated based on the number of occurrences per patient-month. The average cost per person-month and cost attributable to depression were calculated and reported during different time periods: 6, 12 months, and the study period.
Adjusted generalized linear models with gamma distribution were fitted to compare the healthcare costs between RA patients with depression and the matched-controlled population in order to assess the impact of patients’ clinical and baseline characteristics. All analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC, USA). A P value of < 0.05 was considered statistically significant.