We analysed the prescription trajectory of all cohort participants from the date of depression onset until December 2020 to ascertain their TRD status, and matched patients with TRD 1:4 to the non-TRD based on their year of first depression diagnosis, and then age and sex using nearest-neighbour matching. The follow-up started from the prescription date of the third regimen (index date for patients with TRD) until any censoring events. The same index date was applied to the four non-TRD matches. Patients whose deaths or onset of autoimmune diseases occurred earlier than the assigned index dates were further excluded (Supplementary Fig. 1 ). We reported the incidence of autoimmune diseases per 10,000 person-years using Poisson distribution. Next, we estimated the hazard ratios (HRs) of any development of autoimmune diseases associated with TRD status, using a multivariable Cox regression model. Censoring events were development of autoimmune diseases (outcome), death and end of study period, whichever came first. We performed a Schoenfeld residual-based test, which showed no violation of proportional hazard assumption. Covariates included any history of physical disorders (obesity, type II diabetes, hypertension, cardiovascular diseases and tumours) and history of psychiatric conditions (attention-deficit hyperactivity disorder, autism, psychosis or schizophrenia, epilepsy, anxiety disorder, personality disorder, substance use disorder, dementia, bipolar disorder, obsessive compulsive disorder and eating disorder) on or before index date. Detailed ICD-9-CM codes for the covariates are reported in the Supplementary Table 1 .
Free full text: Click here
