Setting and study population
In this population-based retrospective cohort study, the study population consisted of people with type 2 diabetes, aged 15 years or older, living in the Local Health Authority (LHA) of Romagna, that has a catchment area of about 1.1 million people.
Data for the present study were extracted from Emilia-Romagna administrative databases, including the Hospital Discharge Records (HDR) database; Mental Health Information System (MHIS); Residential Mental Health care (RMHC); Pharmaceutical databases; Regional mortality register. HDR database contains admissions and discharge dates, the primary and up to five secondary diagnoses and up to six interventions (identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM coding system). MHIS database includes demographic characteristics and the ICD-9-CM diagnoses of all the adults who have at least one contact with the community mental health centers.
The RMHC database includes information on patients, discharged from no-profit or accredited private facilities, notably admission and discharge dates, principal diagnosis, and destination at discharge. The pharmaceutical databases include drugs reimbursed by the health care system and prescribed by the general practitioner or a specialist, or directly delivered by the hospital pharmacies. These databases contain information on the patient’s sex and age, prescriptions (substance name, ATC System code—V.2013, date of prescription filling, and number of packages), and prescribers. The regional mortality register database was used to collect the patient's date of death. These databases were linked through a unique anonymized patient identifier.
Case definition of diabetes
Beneficiaries of the NHS aged 15 years or older and living in the LHA of Romagna were classified as patients with type 2 diabetes if they had at least one hospitalization with a primary or secondary diagnosis of diabetes (ICD-9-CM code 250.xx) and at least one prescription of Glucose-Lowering Medication (GLM) (ATC code A10), or at least three prescriptions of GLM in distinct periods during the follow-up.
To identify incident cases, we excluded all patients with at least one hospitalization or a GLM prescription in the three years before the date of entry into the study cohort. Uncertain cases of type 2 diabetes, such as patients with insulin as initial and unique treatment, and women diagnosed with gestational diabetes were excluded . We further excluded patients with hospitalizations for the outcomes investigated (Supplementary Table 1) and patients with hospitalizations for depression or prescriptions of antidepressants in the three years before the diabetes diagnosis. The entry date into the study cohort was considered as the date of the diabetes diagnosis.
Patients were followed from the diagnosis of diabetes up to their death or October 31, 2020, whichever came first.
Definition of depression
The presence of depression was ascertained using the following criteria: at least 1 prescription of antidepressant drugs (ATC code N06A), or at least 1 hospitalization (sources HDR, RMHD), or at least 1 outpatient service (source MHIS) with ICD-9-CM diagnosis codes for depression (see Supplementary Table 2). The first date of inpatient or outpatient record or antidepressant prescription was considered as the index depression date.
The presence of comorbid conditions in the three years preceding the onset of diabetes was determined for each patient. The comorbid conditions considered were as follows: other mental disorders (psychosis, bipolar disorders, anxiety/OCD, substance disorders), neurological disorders (epilepsy, dementia, Parkinson's disease), hypothyroidism, respiratory illness (COPD, asthma), and cancer (see Supplementary Table 3 for the detailed list of ICD-9-CM/ATC codes).
The primary outcome was the onset of acute diabetes complications (see Supplementary Table 1 for the detailed list of ICD-9-CM codes) in the first three years of follow-up.
Secondary outcomes were long-term diabetes complications and mortality within ten years from the onset of diabetes (see Supplementary Table 1 for the detailed list of ICD-9-CM codes). Complications were retrieved from the HDRs database.
Demographic and clinical characteristics of patients who developed depression (Dep) during 10 years of follow-up and those who did not develop depression (Non-Dep) were summarized using absolute frequencies and percentages, means and standard deviations or medians and interquartile range (IQR), as appropriate.
Univariate logistic regression models were used to identify the predictors of depression. The possible predictors considered were sex, age (categorized as ≤ 35, 36–55, 56–65, 66–75, > 75 years), urbanization level of the municipality of residence, presence of comorbid conditions, and initial diabetes medication in the first month (1 oral GLM, 2 or more oral GLM, insulin, insulin + oral GLM).
Cox proportional-hazard models were used to investigate whether depression was associated with complications and mortality, unadjusted and adjusted for confounders (sex, age group, presence of comorbid conditions, and initial diabetes medication). Patients were considered exposed to depression only if they developed depression before the outcomes or before the end of follow-up.
In the Cox regression models, depression was included as a time-dependent covariate to take into account that its onset could take place at different times during the follow-up. We tested the proportional-hazard assumption underlying these models using Schoenfeld residuals. When confounders did not meet the proportional-hazard assumption, they were used as strata of the baseline hazard. Results are expressed as odds ratios (ORs) (logistic regression) or hazard ratios (HRs) (Cox regression), with 95% confidence intervals (95%CI).
For all tests, significance was set as p < 0.05. Statistical analyses were performed using IBM SPSS version 25.0 and Stata 15.