Strengths and limitations

The strengths of this study include the general population design, the breadth of coverage of confounding factors and the very large sample size. However, several limitations are acknowledged. Cardiometabolic disorders previously diagnosed by a physician were self-reported by the participants. Similarly, mood disorder features were self-reported, rather than assessed using a formal diagnostic interview, although a structured approach was used (Appendix). Psychiatric diagnoses based on formal interviews were not practical within UK Biobank, so we took a pragmatic approach to mood disorder groupings. It is possible that our criteria for mood disorder are less stringent than formal diagnostic criteria. It should be noted, however, that we are reporting features of mood disorders at a population level, rather than formal clinical diagnoses, and we were unable to validate diagnoses because of the lack of availability of definitive diagnostic information. Similarly, because of a lack of definitive cardiometabolic diagnostic information we were unable to validate outcome measures. This lack of validation for both cardiometabolic and psychiatric diagnoses has implications in terms of the likely sensitivity and specificity of our groupings, which are based on self-report. Linkage to routine hospital and general practice health records will be available in the future and will allow some validation of these groups.

There is, however, evidence for the validity of these groupings using internal variables, ^{Reference Nordentoft, Wahlbeck, Hallgren, Westman, Osby and Alinaghizadeh19} for example, the gender distributions of approximately 1:1 for bipolar disorder and approximately 2:1 (women:men) for depression are consistent with a large body of epidemiological research on lifetime rates of mood disorder in men and women. ^{Reference Kessler, McGonagle, Swartz, Blazer and Nelson20,Reference Hwu, Joyce, Karam, Lee, Lellouch and Wickramaratne21} Furthermore, the lifetime prevalence rates for bipolar disorder (1.1%) and recurrent major depressive disorder (21.2%) are consistent with other population-based lifetime estimates. ^{Reference Smith, Nicholl, Cullen, Martin, Ul-Haq and Evans15}

Although we are reasonably confident that members of the control groups did not have significant features of depression or bipolar disorder, it should be noted that a proportion may have fulfilled criteria for other mental illnesses, such as anxiety disorder or, less likely, schizophrenia, both of which have been associated with poor cardiometabolic health. ^{Reference Nordentoft, Wahlbeck, Hallgren, Westman, Osby and Alinaghizadeh19,Reference De Hert, Correll, Bobes, Cetkovich-Bakmas, Cohen and Asai22,Reference Lahti, Tiihonen, Wildgust, Beary, Hodgson and Kajantie23} There are also limitations with our broad definition of current psychotropic medication use that groups together different classes of medication. In a subanalysis of individuals on psychotropic monotherapy, we found that selective serotonin reuptake inhibitors (SSRIs), other antidepressants and sedatives/hypnotics were all associated with greater risk of myocardial infarction relative to psychotropic-free controls (SSRIs OR = 1.82, 95% CI 1.51–2.21; other antidepressants OR = 1.50, 95% CI 1.20–1.87; and sedatives/hypnotics OR = 2.53, 95% CI 1.76–3.66). It is therefore possible that adverse cardiovascular outcomes in depressive disorders are not limited to the use of antipsychotics but may also be a consequence of other classes of psychotropic medications.

A relatively low proportion of the mood disorder features groups reported taking psychotropic medication (bipolar features 32.1%, depressive features 20.6% and controls 3.4%) but, as noted above, the focus in this study was not clinically diagnosed mood disordered groups but rather lifetime features of mood disorder at a population level. Poor adherence with medication may be an issue. According to some estimates, up to 50% of patients prescribed psychotropic medication do not take their medication regularly. ^{Reference Trivedi, Lin and Katon24–Reference Gadkari and McHorney26} Although adherence for physical health medications may be slightly better than this, ^{Reference Cramer and Rosenheck27} it is likely that the factors that cause poor adherence with psychotropic medication may also lead to poor adherence with cardiovascular medications in individuals with mood disorders. This might explain the proportionately worse cardiovascular outcomes in a less adherent mood disordered group compared with a more adherent non-mood disordered group. As a result of the level of information on medication status that was collected at baseline, we were unable to assess the impact of duration of exposure to psychotropic medication on adverse cardiometabolic outcomes.

We included a range of possible confounding variables in the regression models but were unable to control for physical activity. The available data on physical activity were not collected in terms of standardised measures. Unfortunately, it was not possible to create a standardised measure, such as the metabolic equivalent of task (MET). Similarly, we were unable to adjust for the potential confounding effects of diet. When considering appropriate variables to include as confounding variables, it is important to note that there is uncertainty as to the extent that these variables might represent confounders or mediators. For example, obesity may represent an important component of the pathophysiology for depression, bipolar disorder and coronary heart disease. A prospective cohort study is required to address these concerns and is planned as part of future work in this cohort.

Our analyses did not assess the additive effects of sociodemographic and lifestyle factors. Rather, given the large number of confounding factors involved, we took a broad approach, focusing on cardiovascular outcomes after partial and full adjustment for confounders.

Severity of mood disorder may have contributed to the associations we observed. We therefore conducted an additional analysis of the impact of illness severity on cardiovascular disease outcomes. The depressive features group were divided into those who had a history of a probable single episode of major depression, versus those with a history of probable moderate recurrent depression and those with probable severe recurrent depression (Appendix). The bipolar group was also split into those with features of probable bipolar type I illness and those with features of probable bipolar type II illness (Appendix). In general, individuals in the more severe mood disorder groups had higher risk of adverse cardiovascular outcomes. For the bipolar type I group the odds ratio of ‘any cardiovascular illness’ was 1.39 (95% CI 1.23–1.58), for the recurrent severe depression group the OR was 1.33 (95% CI 1.18–1.50) and for the bipolar type II group the OR was 1.21 (95% CI 1.09–1.32).

Findings in the context of previous work

Many studies have identified an association between major depression and cardiovascular disease but relatively few studies have examined this with respect to bipolar disorder. ^{Reference Hare, Toukhsati, Johansson and Jaarsma28} Although some reports in the literature support the association between bipolar disorder and cardiovascular disease, ^{Reference Ösby, Brandt, Correia, Ekbom and Sparén4,Reference Ramsey, Leoutsakos, Mayer, Eaton and Lee29–Reference Weiner, Warren and Fiedorowicz31} these studies have not been able to adjust for the same range of confounding factors as our study.

Relative to controls, women with features of bipolar disorder had a greater risk of ‘any cardiovascular disease’ than men with bipolar features (OR 1.36 v. 1.19). This finding may be particularly noteworthy given that men are known to have increased rates of cardiovascular disease within the general population. In their 2013 population study, Crump and colleagues ^{Reference Crump, Sundquist, Winkleby and Sundquist30} reported increased all-cause mortality for women with bipolar disorder compared with men with bipolar disorder (adjusted hazard ratio of 2.13 compared with 1.74). Furthermore, the same study reported higher hazard ratios of death from ischaemic heart disease for women with bipolar disorder compared with men with bipolar disorder (2.14 compared with 1.73). Our findings add to the evidence that women with bipolar disorder may be disproportionately affected by cardiovascular disease. ^{Reference Fiedorowicz, He and Merikangas11}

In our fully adjusted model for individuals with depressive or bipolar features, men (but not women) had an increased risk of myocardial infarction, and women (but not men) were at higher risk of stroke (Table DS1). There are several possible explanations for this, including studies that have found that men are more likely to receive a diagnosis of myocardial infarction than women ^{Reference Shah, Griffiths, Lee, McAllister, Hunter and Ferry32} and that stroke is more commonly diagnosed in middle-aged women than in men. ^{Reference Towfighi, Saver, Engelhardt and Ovbiagele33} We also found that, compared with both controls and women, men with depressive features had a higher risk of ‘any cardiovascular disease’ and hypertension, supporting previous findings in this area. ^{Reference Fiedorowicz, He and Merikangas11} Furthermore, we found that in the combined male and female results, depressive features were associated with an increased risk of ‘any cardiovascular disease’. Although odds ratios were not as large as previous studies ^{Reference Herbst, Pietrzak, Wagner, White and Petry12} this may underscore the importance of adjusting for a wide variety of confounding factors where possible.

Possible mechanisms

There are several possible underlying mechanisms for the associations we have observed, including shared behavioural factors, inequalities in treatment, shared genetic vulnerabilities and shared pathophysiology. Individuals with mood disorder may have difficulties in accessing preventative medical interventions, as well as higher rates of risky lifestyle factors. ^{Reference Carney and Freedland34} Such factors include smoking, physical inactivity and reduced adherence with cardiovascular medications. ^{Reference Freedland, Carney and Skala35,Reference Whooley, de Jonge, Vittinghoff, Otte, Moos and Carney36} In keeping with other work, smoking was more common in the bipolar features group than the depressive features and control groups. ^{Reference Heffner, Strawn, DelBello, Strakowski and Anthenelli37} Given the negative impact of smoking on cardiovascular health, ^{Reference Ezzati, Henley, Thun and Lopez38} this is clearly relevant to our findings of increased cardiovascular risk in both the bipolar and depressive features groups. Our results also show a significant attenuation in odds ratios when moving from partially adjusted to fully adjusted analyses, underlining the impact of smoking, alcohol, medication and BMI on cardiovascular risk in individuals with depressive and bipolar features.

Although our study adds to knowledge on the possible contribution of lifestyle and demographic factors to the association between cardiometabolic disease and mood disorder, it cannot contribute substantially to understanding pathophysiological mechanisms. Several biological systems may contribute to the comorbidity between mood disorder and cardiometabolic disease, including (but not limited to) abnormalities of the hypothalamic–pituitary–adrenal (HPA) axis, ^{Reference Rosmond and Björntorp39–Reference Watson, Gallagher, Del-Estal, Hearn, Ferrier and Young42} oxidative stress processes, ^{Reference Khanzode, Dakhale, Khanzode, Saoji and Palasodkar43} shared genetic vulnerability, ^{Reference Kendler, Gardner, Fiske and Gatz44–Reference Torkamani, Topol and Schork47} abnormal inflammatory responses, ^{Reference Hansson48,Reference Joynt, Whellan and O'Connor49} endothelial dysfunction ^{Reference Rybakowski, Wykretowicz, Heymann-Szlachcinska and Wysocki50,Reference Steinberg, Chaker, Leaming, Johnson, Brechtel and Baron51} and haemodynamic reactivity. ^{Reference Strike and Steptoe52}

We did, however, find that the use of psychotropic medication was associated with an increased risk of cardiovascular disease in all groups, even those without features of mood disorder. The metabolic effects of psychotropic medications are known to increase cardiometabolic risk by various mechanisms, including weight gain ^{Reference Allison, Newcomer, Dunn, Blumenthal, Fabricatore and Daumit53} and impaired glucose tolerance. ^{Reference Sernyak, Leslie, Alarcon, Losonczy and Rosenheck54} Given that our analyses were conducted on the basis of current use of any psychotropic medication, it is perhaps noteworthy that adverse cardiometabolic effects may not be limited to antipsychotics. There are reports in the literature of other psychotropic medications used in the treatment of mood disorder, such as antidepressants, lithium and valproic acid, having adverse metabolic effects. ^{Reference Baptista, Teneud, Contreras, Alastre, Burguera and de Burguera55–Reference Fava, Judge, Hoog, Nilsson and Koke57}

Future work

There remain several unanswered questions and opportunities for further research. There is a need for longitudinal and mechanistic studies in this area to better understand causal pathways. ^{Reference Faith, Matz and Jorge58} To date, very few studies have tested specific genetic associations between lifestyle-related risk factors for cardiometabolic disease across the broad clinical spectrum of mood disorders. In addition, further mechanistic studies examining epigenetic factors, as well as oxidative stress, endothelial dysfunction and HPA axis abnormalities will be important to better understand comorbidity between mood disorder and cardiometabolic disease.