This study comprised data from two cohorts: AGDS and the Prospective Imaging Study of Aging (PISA). Participants in both groups provided informed consent before participating. These studies, including all questionnaires used, were approved by QIMR Berghofer Medical Research Institute's Human Research Ethics Committee.

Twenty thousand six hundred eighty-nine participants from across Australia were recruited through an open media campaign and targeted mailout. The publicity campaign, from which 86% of participants were recruited, including both conventional and online social media. The campaign appealed for anyone who “had been treated by a doctor, psychiatrist or psychologist for depression” to visit this website—https://www.geneticsofdepression.org.au. For the targeted mailout, invitation letters were sent by the Australian Government Department of Human Services (DHS) to individuals who, according to their records, had received at least four prescriptions for any of the 10 most commonly used antidepressants in the last 4.5 years. DHS did not, at any time, share any personal information with the research team. Potential participants were directed to the above website, which contained information about the study, a registration and consent form, and a comprehensive online questionnaire. The essential inclusion criteria included having been prescribed and taken antidepressants and providing consent to donate a saliva sample for subsequent genotyping. No participant was excluded based on comorbid conditions. The online survey assessed mental health diagnoses, antidepressant response, suicidality, general health and substance use, among several other variables. A detailed baseline description of the cohort has been published elsewhere (75). The full list and details of instruments used for AGDS phenotyping are available at: https://bmjopen.bmj.com/content/bmjopen/10/5/e032580/DC2/embed/inline-supplementary-material-2.pdf.

The Prospective Imaging Study of Aging (PISA) is a longitudinal cohort of Australian adults (76). The population-based sample recruitment pool comprised adult twins, their spouses, and first-degree relatives of twins and spouses who over previous decades, had volunteered for studies on risk factors or biomarkers for physical or psychiatric conditions and had previously been genome-wide genotyped (77, 78). The PISA protocol consisted of online questionnaires, including a history of mental health diagnoses and the same pain questionnaire in AGDS. It was completed by N = 2,469 PISA participants. For that reason, AGDS and PISA data were used in the present study to assess the effect of depression and demographics (e.g., age, sex) on chronic pain. All other analyses described in this manuscript were performed only in the AGDS cohort.

AGDS participants were asked to self-report whether they had ever been diagnosed with depression by a health professional, and similarly for 19 other psychiatric conditions. Individuals were classified as depression cases if they had reported both a depression diagnosis and had been prescribed antidepressants in the past 5 years (N = 17,849). Of these, 92% fulfilled the DSM-5 diagnostic criteria for a lifetime depressive episode based on detailed descriptions of this cohort (75). Importantly, this figure is within the test-retest reliability estimates of depression ascertainment from DSM-5 based instruments (79, 80). Participants were administered a pain severity numerical rating scale (81). Briefly, patients were asked to indicate whether they experienced chronic pain in their daily life and to rank its intensity on a scale from 0 to 10. Only those reporting a pain rating >0 progressed to the remainder of the pain module, which included questions about the duration and location of their primary pain. Following the IASP guidelines, chronic pain was defined as pain persisting or recurring for at least 3 months (7, 8). Cases were classified as having comorbid chronic pain and depression if they fulfilled the criteria for both conditions (N = 6,895), and controls were classified as those who reported depression but no chronic pain (N = 4,475). We performed a complete case analysis. Thus, participants with missing data for chronic pain (i.e., those who did not complete the section; N = 6,463) were excluded from analyses that needed data for both chronic pain and depression.

Suicidality was assessed using the SIDAS instrument (82). Briefly, suicidal ideation over the last month was measured on a 10-point scale: 0 indicated having no suicidal ideation in the past month (never), and 10 denoted persistent suicidal ideation. Participants with a score >0 were classified as positive cases for suicidal ideation. Suicide attempt was measured using a similar 10-point scale in regard to how close a participant had come to making an attempt. Only those with a score of 10 (labeled as “I have made an attempt”) were considered cases for a suicide attempt. Participants also reported their frequency in using a range of substances over the last 3 months. Alcohol consumption frequency was measured as the number of days the participant drank three or more standard drinks. For all other substances, the response options were: “never” (0), “once or twice” (1), “monthly” (2), “weekly” (3), or “daily” (4). These responses were modeled as continuous variables when assessing their correlation with chronic pain.

Participants were asked whether they had ever been prescribed any of the 10 most commonly used antidepressants in Australia for any indication. These are sertraline, escitalopram, venlafaxine, amitriptyline, mirtazapine, desvenlafaxine, citalopram, fluoxetine, duloxetine and paroxetine. Information regarding the reason(s) for prescription was collected using a checklist of 17 possible responses, including depression, chronic pain, and anxiety (among others). Multiple selections were possible. Participants were asked to report on the best aspects of taking antidepressants using the following item: “What were the best aspects of taking the antidepressant(s)? Include any antidepressant you have taken.” Participants were then able to select all that apply out of a list including relief of depressive symptoms, relief of other symptoms, e.g., sleep disturbance, reduction in suicidal ideation, return of normal emotion, improved relationships, returning to normal activities and restored control over mood. Moreover, participants rated the effectiveness of each antidepressant they had taken, using a scale ranging from 0 (e.g., “sertraline works not at all well for me”) to 2 (e.g., “sertraline works very well for me”). Two analyses were performed: (i) first, antidepressant effectiveness was compared between participants who reported taking an antidepressant prescribed for chronic pain against the rest of the participants (i.e., not prescribed for chronic pain); and (ii) we compared antidepressant effectiveness between participants reporting chronic pain and those reporting no chronic pain (regardless of explicit indication).

In this study, we used complete case analysis and thus removed participants who did not have the required data from specific analyses. The relationship between chronic pain and several other variables of interest was assessed using multivariable logistic regression. This approach enabled us to quantify the associations while adjusting for age, sex and all other relevant factors (e.g., the correlation between alcohol and chronic pain while keeping usage of all other substances equal). Fully adjusted odds ratios were calculated from effect sizes on the logit scale, and p-values were estimated using Wald-tests. For all analyses, the presence of chronic pain was modeled as a binary variable, while chronic pain severity was modeled as a quantitative score from zero to 10. The relationship between chronic pain and antidepressant effectiveness was examined using cumulative link logistic regressions to accurately model treatment response, which was coded on an ordinal scale. Furthermore, to assess the effect of chronic pain across all antidepressants, a random effect was included to account for repeated responses from participants. This analysis was performed in R using the ordinal package and the clm and clmm functions, adjusting for the effects of sex and age when antidepressant treatment started. All other statistical analyses were performed and figures generated in python using these modules: statsmodels, scipy, numpy, pandas, matplotlib, and seaborn.

Demographics and chronic pain prevalence for both AGDS (enriched for depression) and PISA (not enriched for depression) cohorts are shown in Table 1, Supplementary Figures 1, 2. Figure 1 shows the prevalence of chronic pain by age, stratified by cohort. A significant cohort effect is evident. Nonetheless, this cohort effect may be attributable (at least in part) to other differences such as age, sex and education rather than depression. Chronic pain is positively associated with age but despite the PISA cohort being older on average (Supplementary Figure 1), the AGDS cohort showed a higher prevalence of chronic pain. After adjusting for all the relevant factors, the cohort effect was found to be partly attributable to depression status (OR = 1.86 [1.37–2.54]) because residual cohort effects were non-significant after accounting for the effect of depression (Cohort_AGDS OR = 1.32 [0.97–1.79]). Furthermore, a higher age (OR = 1.02 [1.02–1.03]), lower educational attainment (OR = 0.89 [0.86–0.91]), and being female (OR = 1.16 [1.07–1.25]) were associated with chronic pain in the pooled PISA and AGDS sample (Supplementary Table 1).

Results presented here are from AGDS where all participants reported depression. Higher pain severity (intensity) was found to be associated with longer durations of pain (Supplementary Figure 3). Increased pain severity was also associated with an increased number of depressive episodes (Supplementary Figure 4). The prevalence of suicidal ideation was higher in the comorbid chronic pain group (OR = 1.49 [1.38–1.61]). Likewise, recent suicide attempt was associated with chronic pain (OR = 1.88 [1.14–3.09]). Within the chronic pain group, recent suicidal thoughts and suicide attempt scores were also positively correlated with chronic pain severity scores (Supplementary Figure 4).

In this subsection, the results presented are from AGDS where all participants reported depression. Out of the 19 mental health conditions examined, social anxiety disorder was found to have the strongest association with chronic pain (p < 0.01), however, this association did not survive multiple-testing correction. Anorexia nervosa was found to be negatively associated with the likelihood of developing chronic pain (p < 0.05). Although both of these results were nominally significant, no association survived correction for multiple testing (Figure 2; Supplementary Table 2). Notably, chronic pain was significantly associated with decreased use of alcohol, increased use of tobacco, and painkiller misuse (e.g., opioids). Nominal associations were observed for other drugs such as cocaine (negative relationship) and opioids (Figure 2; Supplementary Table 3).

In this subsection, the results presented are from AGDS, where all participants reported depression. The three most commonly prescribed antidepressants for the indication of chronic pain—over and above depression—were amitriptyline (N = 606), duloxetine (N = 288) and sertraline (N = 160; Supplementary Figure 5). Overall, the overwhelming majority of participants with chronic pain (i.e., 75–98% across all antidepressants) reported that their antidepressant prescription did not consider chronic pain. Furthermore, participants with chronic pain predominantly reported taking antidepressants for depression (i.e., more than 90% of participants reported their prescription was for depression). The exception was amitriptyline, for which only 60% of participants with chronic pain reported the prescription was for depression (Supplementary Table 4). Compared to participants without chronic pain, those with comorbid chronic pain were less likely to report positive functional benefits from taking antidepressants such as “relief of depressive symptoms,” “return of normal emotions,” and “getting back to normal daily activities” (Figure 3A). A trend was noted—whereby participants with chronic pain were more likely to report a reduction in suicidal symptoms as a positive aspect of antidepressant treatment—but this finding did not survive correction for multiple testing (Supplementary Table 5). Furthermore, in the chronic pain group, the average self-reported benefits from taking antidepressants were significantly lower compared to the group without chronic pain (OR = 0.75 [0.71–0.80]; p < 2 × 10⁻¹⁶). A similar but non-significant finding was observed between the average response of participants prescribed antidepressants for chronic pain vs. those without a chronic pain indication (OR = 0.94 [0.80–1.1]; Supplementary Figure 6). Next, we examined whether these findings held true for each antidepressant under investigation. For most antidepressants, no statistically significant difference in effectiveness was found between participants with chronic pain (or an indication for chronic pain) and participants without chronic pain. Participants with chronic pain who had taken sertraline, escitalopram or venlafaxine, reported significantly lower effectiveness than participants without chronic pain (Figure 3B). At this point, it remained unclear whether the antidepressant was taken before or after the commencement of chronic pain. To address this question, we performed a secondary analysis defining cases as participants who reported taking an antidepressant where the prescription explicitly considered chronic pain. This analysis revealed lower effectiveness for sertraline, escitalopram and citalopram (Figure 3C). The only antidepressants with a positive effect (i.e., greater effectiveness) were duloxetine, venlafaxine and amitriptyline, but only when prescribed for comorbid chronic pain and depression. However, none of these positive associations reached statistical significance (Figure 3B; Supplementary Tables 6, 7).

The current study is the largest to date examining the relationship between depression and chronic pain with a novel pharmacoepidemiological approach that has yielded new insights into the medical treatment of these highly morbid conditions. However, there are also a number of limitations to be acknowledged. Data were primarily drawn from AGDS, which is a multi-aim study investigating the risk factors for depression and treatment response to antidepressants. As the AGDS employed phenotyping across an extensive range of complex traits (75), on balance it was not feasible to also collect detailed information from participants on dosages and (polypharmacy) combinations taken of the prescribed antidepressants (103); the duration and magnitude of benefits (e.g., for cost-effectiveness analyses) (65, 104, 105); drug tolerability and adverse events (65, 106–111); adjunct psychological therapies and multidisciplinary treatment/rehabilitation programs (112); other prescribed pain pharmacotherapies and questionnaires (113–117). Further pharmacoepidemiological studies focusing on chronic pain and a large range of psychiatric comorbidities (51, 52, 118). can directly address and collect data on these specific issues. As the current data were based on self-reported responses, they may also be subject to a degree of participant recall bias (e.g., time-specific details). For example, the substance use and suicidality phenotypes may be subject to non-disclosure effects due to the potential stigma associated with these conditions and the antidepressant response data may include non-specific (placebo/nocebo) effects (59, 119). However, these effects would be present across all antidepressants and thus alone are highly unlikely to explain the observed differences between the chronic pain (cases) and control participants. Randomized interventional studies comparing treatments for participants with comorbid chronic pain and depression are required to validate our results and elucidate the causal direction of associations reported here. Genetic-based methods can also aid in further examination of our findings by performing discovery genome-wide association studies (GWAS) of comorbid chronic pain and depression to determine individuals' polygenic risk irrespective of whether they have developed chronic pain or not. Furthermore, GWAS will enable: (i) the elucidation of causality between chronic pain and treatment response by using methods such as Mendelian randomization; and (ii) replication in antidepressant treatment-resistant depression cohorts with primary care and genotype data (120).