Associations between multimorbidity and depression among breast cancer survivors within the UK Biobank cohort: a cross-sectional study

UK Biobank contains data from over 500,000 volunteers, recruited in the period between 2006 and 2010 from across England, Scotland, and Wales. The volunteers were aged between 40 and 70 at the time of recruitment, and provided data from touchscreen questionnaires, physical measurements, genotyping and longitudinal follow-up, with further data continuing to be added [17]. The data are linked to cancer registries, allowing identification of individuals previously diagnosed with breast cancer prior to baseline recruitment.

Included in this study were women with a primary diagnosis of a malignant neoplasm of the breast (International Classification of Diseases 10th Revision (ICD-10) code C50 and ICD-9 code 174). 9364 women matching this criterion were identified.

UK Biobank received ethical approval from the NHS National Research Ethics Service North West (16/NW/0274) and was performed in accordance with the Declaration of Helsinki. Informed consent was obtained for all UK Biobank participants. The analyses in this study were performed under UK Biobank application number 41686.

Current depression was the primary outcome variable, which was assessed using the validated two-item Patient Health Questionnaire (PHQ-2) at baseline [18]. The PHQ-2 operates by asking participants about two key depressive symptoms (low mood and anhedonia) via two simple questions – in the 2 weeks prior how often have they “felt down, depressed, or hopeless” (data field 2050), and how often they “had little interest or pleasure in doing things” (data field 2060). Responses available to participants were “not at all” (assigned a score of 0), “several days” (assigned a score of 1), “more than half the days” (2), and “nearly every day” (3). Participants answering “do not know” or “prefer not to answer” were coded as missing. The combined total provides a score from 0 to 6, with participants scoring at least three categorized as having probable current depression and those scoring less than three categorized as no current depression [18]. We chose to use a measure of current depression to limit the possibility of reverse causation as compared with using a measure of lifetime depression (reported by 7.2% of the sample). We selected the cut-off value of ≥3 as a measure of more severe depression. Using this approach, 5.3% of the sample were classed as currently depressed, compared to 16.6% using a cut-off value of ≥2 [19]. The proportion of people classed as currently depressed using the higher cut-off threshold is also more similar to the measure of self-reported lifetime depression.

Multimorbidity was measured by a count of the number of self-reported doctor diagnosed lifetime chronic health conditions recorded during the baseline nurse-led interview (data fields 20001 and 20002) and informed by previous literature on multimorbidity [5, 20, 21]. The physical and mental health conditions reported by participants were classified into 43 groups following a previously published approach [20‐22]. We excluded lifetime depression, cancer and prostate disorder (as our sample was exclusively women) from the count of co-morbid conditions and multimorbidity was coded as 0, 1, 2, 3, 4 or 5+. Please refer to Additional file 1: Table S1 for full details of the chronic health conditions considered.

We included a number of potential confounding variables: age (5-year bands), education level (degree, non-degree level), socioeconomic deprivation (measured using the Townsend Index [23] and divided into quartiles), ethnicity (white, non-white), marital/cohabitation status (living with spouse/partner, not living with spouse/partner), whether the participant had been previously diagnosed with cancer more than once (yes, no), and time since the primary breast cancer diagnosis (less than 1 year, 1–3 years, 3–5 years, 5+ years).

Descriptive statistics were used to summarise the sample and the most commonly reported chronic health conditions. Logistic regression models were calculated to analyse the association between multimorbidity and the odds of depression. Model 1 included the number of chronic conditions and patient age as predictors. Model 2 additionally included ethnicity, education level, socioeconomic deprivation, marital/cohabitation status, more than one cancer diagnosis and number of years since the primary breast cancer diagnosis as predictors. These models were used to calculate odds ratios with 95% confidence intervals and predicted probabilities (using Stata’s margins command with the atmeans option) to determine the effect sizes. A significance level of p < 0.05 was used in this study. Missing data were excluded (n = 926, 9.9%) from the analysis, with most missing data being found for PHQ-2 (N = 675). A small amount of missing data was also observed for multimorbidity (N = 66), ethnicity (N = 34), education level (N = 168), marital/cohabitation status (N = 51), and socioeconomic deprivation (N = 12). After the exclusion of missing data 90.1% of participants were included in the final analysis sample (N = 8438). Analyses were completed in R version 4.0.2 and Stata/MP 16.0.

Data from UK Biobank is not without flaws – one being evidence of a “healthy volunteer” selection bias. The average UK Biobank participant is healthier than the average member of the UK population. Additionally, participants in UK Biobank were in the age range of 40–70 at the time of recruitment, meaning the ages outwith this bracket are not represented. However, assessment of exposure-disease relationships, such as in the primary aim of this study, may remain generalisable without requiring participants to be reflective of the larger population [30].

Use of UK Biobank data enabled a large sample size of 8438 participants with breast cancer to be obtained and the inclusion of a range of potential confounding factors, the central strength of this study. A key limitation of this study is reliance on self-reporting of the exposure and outcome variables, which can be subjective and therefore less reliable than objective measures such as a formal diagnosis from medical records. However, as previously mentioned, the PHQ-2 is a valid tool for measuring depression and it enabled assessment of people with depression who may not have presented to health services [18]. We used the higher cut-off threshold of ≥3 to identify more severe depression and there are some potential limitations with this approach, including lower sensitivity, but higher specificity [19]. It is possible that using other tools to measure current depression would have produced different results and further work using other outcome measures is needed [31]. Our measure of multimorbidity was also based on a count of self-reported lifetime chronic conditions which may be an overestimate of those conditions currently experienced. However, it may be the case that using a measure of current multimorbidity would be more strongly associated with current depression. Further research using different definitions of multimorbidity is needed, including using administrative data from different sources (e.g. primary and secondary care).