We performed a nationwide register-based cohort study including all patients diagnosed with incident invasive bladder cancer in Denmark from 2005 through 2015. The Danish population, approximately 5.6 million inhabitants, has access to tax-supported healthcare, with free access to hospital-based and primary medical care provided by the Danish National Health Service [15]. All inhabitants in Denmark have a unique Civil Personal Register (CPR) number, which is recorded along with administrative and medical information in registries and databases and allows for person-specific linkage of information [16].
Data sources
From the Danish National Patient Registry (DNPR), we retrieved diagnosis and procedure codes and associated dates. The DNPR has tracked all somatic hospitalizations in Denmark since 1977 and outpatient and emergency room visits to hospitals since 1995. Recorded data include CPR numbers, dates of admission and discharge, and up to 20 diagnoses, classified according to the International Classification of Diseases, tenth revision (ICD-10) since 1994 [17]. We ascertained information on tumor pathology from The Danish National Pathology Registry, which contains descriptions of pathological specimens and has a coverage of almost 100% [18]. The registry was established in 1997, and data from earlier years have been added [18]. We defined exposure status using data from the Danish National Database of Reimbursed Prescriptions [19]. This registry encompasses information on all redeemed prescriptions from 2004 onwards, including date of redemption and Anatomical Therapeutic Chemical (ATC) code. Finally, the Danish Civil Registration System provided data on marital and vital status. This database contains nearly complete, demographic data and is updated daily [16].
Study population
We identified patients with incident, histologically verified invasive bladder cancer (thus not including patients with carcinoma in situ or pTa tumors) based on ICD-10 diagnosis codes (C67) and bladder cancer pathology (Additional file 1: Table S1). To increase the probability of truly incident cases we excluded patients fulfilling these criteria between 1995 and 2004.
For the surgical outcome analyses, we identified patients with bladder cancer and a recorded cystectomy excluding patients who received intended curative radiation therapy before cystectomy (Additional file 1: Table S1).
Exposure, outcome measures, and covariates
We defined individuals as exposed (‘antidepressant users’) if they filled two or more antidepressant prescriptions (ATC code N06A) on separate occasions in the year preceding bladder cancer diagnosis. Using this definition we sought to reduce the number of non-adherent patients in the exposure group. Individuals filling one prescription or less are termed ‘non-users’. Patients who filled an antidepressant prescription after bladder cancer diagnosis, but not in the year before, were treated as non-users.
Stage at diagnosis was defined as muscle-invasive (pT2+) or non-muscle-invasive (pT1), and stage at cystectomy as organ confined (pT0-T2 and pN0) or non-organ confined (pT3-T4 or pN+).
We compared the cumulative incidence of cystectomy within three months following a diagnosis of muscle-invasive disease. We chose three months as a clinically relevant end of follow-up based on prior work [20]. Following surgery, we assessed outcomes expected to reflect a complicated postsurgical course: length of hospital stay, acute readmission to a somatic hospital within 30 days after discharge from the primary admission, 90-day rate of postoperative procedures, and all-cause mortality at one and three years. We defined postoperative procedures as any invasive procedure within 90 days after cystectomy (any NOMESCO Classification of Surgical Procedures code).
Based on existing literature we included as potential confounders: age, sex, marital status at diagnosis, Charlson Comorbidity Index (CCI) score [21] (excluding bladder cancer), and alcohol-related disorders. We identified comorbidities from DNPR within 10 years before cancer diagnosis (Additional file 2: Table S2) [22].
Statistical methods
Patients’ characteristics at bladder cancer diagnosis were summarized according to exposure groups. Continuous variables were either categorized or described by their median value and interquartile range. For categorical variables, we computed proportions of individuals at each categorical level by exposure group.
We examined the association between antidepressant use and stage at diagnosis (muscle-invasive (pT2+) or non-muscle-invasive (pT1)) or surgery (organ confined (pT0-T2 and pN0) or non-organ confined (pT3-T4 or pN+)) using logistic regression, adjusting for age at diagnosis (included as an unrestricted spline with four knots in the analysis of stage at diagnosis and squared in stage at surgery), sex, CCI score (0, 1–2, 3+), alcohol-related disorders (yes/no), and marital status at diagnosis (married/not married). The analysis of stage at surgery excluded patients receiving neoadjuvant chemotherapy, which was introduced nationwide in Denmark in 2013 (Additional file 1: Table S1).
Due to missing data for tumor stage at diagnosis and at surgery, two sensitivity analyses were performed to increase the proportion of individuals with stage data: we restricted the analysis of stage at diagnosis to patients diagnosed during 2011–2015, and we repeated the analysis of stage at surgery including the latest stage recorded before surgery, if stage at surgery was missing.
For the analysis of time to surgery, we followed patients not receiving neoadjuvant chemotherapy from the date on which muscle invasiveness was first detected until cystectomy, death, emigration, end of the three months of follow-up, or 11 April 2016, whichever came first. Follow-up for postoperative procedures began at date of surgery while follow-up for readmission began at discharge. For both events, follow-up ended at event of interest, death, emigration, end of the 30-day or 90-day follow-up, or 11 April 2016, whichever came first.
We computed the cumulative incidence of cystectomy and used Cox proportional hazards regression to investigate the association between antidepressant use and cystectomy. Here, death and curative-intended radiation therapy were treated as competing risks.
We used Cox proportional hazards regression to investigate associations between antidepressant use and readmission and postoperative procedures adjusting as described for the logistic regression of stage at surgery. Death was considered competing risk. Length of hospital stay was compared between groups using a multiple linear regression model including covariates as in the logistic regression of stage at diagnosis.
For the mortality analysis, we followed patients from date of surgery until death, emigration, end of the one-year or three-year follow-up or 19 April 2016, whichever came first. We used the Kaplan-Meier estimator and compared mortality using Cox regression adjusting as described for the logistic regression of stage at surgery.
Additionally, we investigated whether neoadjuvant chemotherapy, stage at surgery, and type of surgery (open or laparoscopic and robot-assisted, (Additional file 1: Table S1)) mediated surgical outcome differences between exposure groups by addition of each of these variables individually to the regression models described above.
For all Cox regressions, the assumption of proportional hazards was assessed by log-minus-log plots.
All analyses were performed using Stata version 14 (StataCorp LP, College Station, Texas).