Antidepressant prescriptions and associated factors in men with prostate cancer and their female partners

All persons living in the greater Copenhagen and Aarhus areas aged between 50 and 64 years without a previous history of cancer (n = 160,725) were invited to participate in the Diet, Cancer and Health cohort study [20]. The overall participation rate was 37% among women and 34% among men. Details about the study design, participation, and non-responder analyses have been published previously [20]. Baseline was defined as 1 January 1997. For participants who entered the study between January and May 1997, baseline was set to the date of study entry. Analyses within the group of patients with prostate cancer and the female partners of men with prostate cancer began at the date of prostate cancer diagnosis of the patient. All men and their female partners participating in this cohort were followed until 31 December 2014 (men) or 31 December 2010 (partners).

Study-specific exclusion criteria at baseline were (1) history of major psychiatric disorders, defined as at least one hospital contact for organic or substance-related mental disorders, schizophrenia, schizotypal or delusional disorder, bipolar or unipolar depression (ICD-8: 290 to 295.99, 296.19, 296.39, 303.00–304.99, and ICD-10: F00-F33); (2) history of cancer except non-melanoma skin cancer; and (3) one or more redeemed prescriptions for antidepressant medications (Anatomic Therapeutic Chemical classification system [ATC] code N06A, excluding Bupropion [N06AX12]).

At enrollment, participants completed questionnaires concerning diet, lifestyle-, and health-related issues, including physical exercise (metabolic equivalents [MET] categorized in quartiles), smoking status (current, former, never), and alcohol consumption converted to gram/day (categorized into 0, 1–36, and > 36 g/day for men and 0, 1–24, and > 24 g/day for women, i.e., the recommended limit in Denmark at the time) [20]. A lab technician conducted anthropometrical measurements including height and weight. Body mass index (BMI) was categorized into underweight (< 18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), or obese (≥ 30 kg/m²).

In patients with prostate cancer, hospital records were screened to obtain levels of prostate-specific antigen (PSA) at the time of diagnosis, Gleason score, and treatment information (first-line treatment categorized as active surveillance, curative, palliative, and watchful waiting). Using the 2004 TNM classification, patients were categorized into localized (T1-2 N0,x M0,x and Tx N0 M0) vs. nonlocalized (all other TNM stages). As full information on cancer stage was not available in all hospital files, supplementary information was retrieved from the Danish Cancer Registry.

Unambiguous linkage of data from nationwide registers was secured by personal identification numbers. The National Cancer Registry [21], the National Prescription Registry [22], the National Patient Registry [23], the Danish Psychiatric Central Research Register [24], the Educational Register [25], and the Civil Registration System [26] were used for dates and ICD codes of cancer diagnoses and TNM stage where the information could not be obtained from hospital charts, dates of FRPA, Charlson Comorbidity Index (CCI), highest level of education (short, mandatory school only; medium, senior high school, or vocational education; long, higher education), cohabitation status (opposite-sex person with maximum of 15-year age difference, living in the same household), date of emigration, date of change in personal identification number, and date of death, respectively (see Supplementary Material S1 for more details).

The outcome for all analyses was the first redeemed prescription for antidepressants (FRPA) following exposure (the male partner’s prostate cancer diagnosis). Antidepressant prescriptions have been successfully used as measure of physician-diagnosed depression and anxiety in patients with cancer [27]. In northern European countries, about two thirds of antidepressant prescriptions are issued to treat depression, followed by anxiety (20%), sleep disorders (10%), obsessive-compulsive disorder (< 5%), and neuropathic pain (< 5%) [28‐30]. We used the date of the first rather than repeated prescriptions as it most closely corresponds to the point in time when a mental disorder requiring pharmacological treatment is diagnosed.

In all analyses of the male cohort, observations were censored by the following dates: diagnosis of non-prostate cancer and non-melanoma skin cancer (ICD-10 C-Diagnoses except C4A, C44, C61), hospital contacts for major psychiatric disorders not including depression (ICD-10 F00-F31), emigration, death or 31 December 2014. The same censoring criteria were applied in the partner cohort, with two additional dates: the date at which their partner was diagnosed with a non-prostate cancer and any change in cohabitation (including separation, divorce, or death of the male partner). As there were too few partners at risk after 2010, the end date for partners was set to 31 December 2010.

To estimate the unadjusted hazard ratios (HR) of FRPA for sociodemographic factors, lifestyle, and somatic comorbidity at baseline, we conducted univariate Cox regression analyses with age as the underlying time.

The adjusted HRs of FRPA were estimated in multivariable Cox regression models including education, BMI, MET, smoking status and alcohol consumption at enrollment, and CCI and cohabitation status (except for partners) as time-varying covariates, using age as the underlying time. The proportional hazard assumption was tested both graphically and using Schoenfeld residuals. Because the proportionality assumption was violated for prostate cancer diagnosis (prostate cancer vs. cancer-free) in the male cohort, we analyzed the interaction of age (cutoff 65 years) and having a prostate cancer diagnosis.

The incidence of FRPA after prostate cancer diagnosis in patients and their partners was estimated using cumulative incidence function (CIF) analyses. We statistically compared differences in cumulative incidence using Gray’s test [31].

To identify potential risk factors of FRPA in men with prostate cancer, we performed a multivariable Cox regression analysis using time since diagnosis as the underlying time, including the following independent variables: education, BMI, MET, smoking status and alcohol consumption at enrollment, CCI and cohabitation status as time-varying variables, tumor spread (localized vs. nonlocalized) at the time of diagnosis, and first-line treatment. The proportionality assumption was not violated. As a post hoc analysis, we compared patients in different treatment groups to cancer-free men, adjusting for the same covariates as in the first model.

All statistical analyses were performed in R version 3.5. [32] using the packages prodlim and cmprsk for cumulative incidence functions [33, 34], survival for Cox modeling [35], and survminer to test the proportional hazard assumption [36]. Cases with missing data were deleted list-wise before each analysis. Cohabitation status and education showed relevant amounts of missing data (4.1% and 2.0% missing values, respectively). All other variables had near-complete data with 0.4% or fewer missing values.

All participants provided written informed consent. The study was conducted in accordance with the Declaration of Helsinki and approved by the regional ethical committees on human studies in Copenhagen and Aarhus (File no.: (KF)11–037/01) and by the Danish Data Protection Agency (File no.: 2013-41-4232).

Out of 26,944 men in the Diet, Cancer and Health cohort, 25,126 men (93%) fulfilled the inclusion criteria (Fig. 1). At baseline, 8785 women from the Diet, Cancer and Health cohort who fulfilled the inclusion criteria were cohabiting with men from the cohort. Most men (median age 57 years; interquartile range 54 to 61 years) had medium education, were cohabiting, overweight, current or former smokers, drank moderate amounts of alcohol, and had no severe somatic comorbidities at baseline (Table 1). Female partners (median age 56 years; interquartile range 53 to 60 years) had similar patterns of education and somatic comorbidity, but their health behavior was more favorable than the men’s (Table 1). The median follow-up time was 17.8 years for the male cohort and 13.7 years for the female partner cohort.

Unadjusted Cox regression models showed that short education, living alone, smoking, high levels of alcohol consumption, and somatic comorbidity were associated with an overall higher risk of FRPA in both the overall male cohort and the partner cohort (Table 1). A BMI below or above normal weight was associated with a higher risk of FRPA in the male, but not the female cohort.

Patients with prostate cancer were followed for a median time of 4.3 years after receiving the diagnosis and female partners of patients for 3.1 years after their partner’s diagnosis (Table 2). Men with prostate cancer had a two-fold increased risk of FRPA and a two- to four-fold increased risk after adjusting for sociodemographic, lifestyle-related factors, and comorbidity, compared with cancer-free men (Table 3). Partners of men with prostate cancer had a non-significantly increased risk (Table 3).

After diagnosis with prostate cancer, patients had a significantly higher incidence of FRPA than their female partners (Gray test p = 0.001; Fig. 2).

During the study period, 1828 men were diagnosed with prostate cancer (Fig. 1). At the time of diagnosis, patients with prostate cancer had a median age of 69 years (interquartile range, 66 to 73 years) and 85% were cohabiting with a partner. Patients’ medical characteristics are shown in Table 2. Patients assigned to palliative treatment received androgen deprivation therapy in more than 90% of the cases.

Palliative cancer treatment, curative treatment (compared with active surveillance/watchful waiting), non-localized disease, short education, and living alone were significantly associated with a higher HR of FRPA (Table 4). When active surveillance and watchful waiting were entered as separate categories into the analysis, both treatments showed a non-significant association with a lower risk of FRPA than patients in curative treatment (HR, 0.33; 95% CI, 0.10 to 1.05; and HR, 0.54; 95% CI, 0.29 to 1.03, respectively). An adjusted post hoc model showed that patients receiving active surveillance or watchful waiting had a similar risk as cancer-free men (HR 0.74, 95% CI, 0.48 to 1.14), while patients receiving curative (HR, 1.87; 95% CI, 1.52 to 2.30) or palliative treatment (HR 4.70; 95% CI, 3.90 to 5.66) had significantly higher risks of FRPA.

Sensitivity analyses adjusting for calendar period or date of diagnosis revealed no substantial effects on the HRs observed (data not shown).

Strengths of our study include the prospective, longitudinal design, including detailed and objective data from before the cancer diagnosis until up to 18 years after the diagnosis and following both patients and their partners. The inclusion of near-complete data on lifestyle factors and detailed clinical data as well as the time-varying impact of cohabitation and somatic comorbidity further increase the scope of our analyses and the robustness of our results.

Our results ought to be interpreted in light of the following limitations: participation in the Diet, Cancer and Health cohort was associated with longer education and better overall health [20, 49]. Thus, it is likely that there is less variance in both general and mental health and our analyses may underestimate the actual effects. Lifestyle variables were self-reported, assessed only at enrollment, and may have changed over time [47]. Due to the registry-based assessment of cohabitation status based on sex, age, and address, it cannot be ruled out that some of the women identified as partners may actually be flat-mates or friends living in the same apartment. However, in this age group, such mislabeling is rare. Despite the large cohort, the power of the partner analyses was insufficient to confirm the significance of relatively weak associations. It should be noted that results on treatment refer to first-line treatment and some patients may have received other treatments later on. Antidepressants may be prescribed for issues other than depression and anxiety and a prescription thus does not exactly correspond to a diagnosis of these mental disorders. However, antidepressant prescriptions can be seen as a proxy for a psychological or neurological issue that was deemed as requiring pharmacological treatment by a physician and thus as an important group of intermediate and late effects of cancer and its treatment. As antidepressants are more frequently described for depression than for anxiety disorders, we did not use prior hospital stays due to anxiety as an exclusion criterion, which represents a small methodological asymmetry in the exclusion criteria.