Impact of major depression on cardiovascular outcomes for individuals with hypertension: prospective survival analysis in UK Biobank.

OBJECTIVES
To assess whether a history of major depressive disorder (MDD) in middle-aged individuals with hypertension influences first-onset cardiovascular disease outcomes.


DESIGN
Prospective cohort survival analysis using Cox proportional hazards regression with a median follow-up of 63 months (702 902 person-years). Four mutually exclusive groups were compared: hypertension only (n=56 035), MDD only (n=15 098), comorbid hypertension plus MDD (n=12 929) and an unaffected (no hypertension, no MDD) comparison group (n=50 798).


SETTING
UK Biobank.


PARTICIPANTS
UK Biobank participants without cardiovascular disease aged 39-70 who completed psychiatric questions relating International Classification of Diseases-10 Revision (ICD-10) diagnostic criteria on a touchscreen questionnaire at baseline interview in 2006-2010 (n=134 860).


PRIMARY AND SECONDARY OUTCOME MEASURES
First-onset adverse cardiovascular outcomes leading to hospital admission or death (ICD-10 codes I20-I259, I60-69 and G45-G46), adjusted in a stepwise manner for sociodemographic, health and lifestyle features. Secondary analyses were performed looking specifically at stroke outcomes (ICD-10 codes I60-69 and G45-G46) and in gender-separated models.


RESULTS
Relative to controls, adjusted HRs for adverse cardiovascular outcomes were increased for the hypertension only group (HR 1.36, 95% CI 1.22 to 1.52) and were higher still for the comorbid hypertension plus MDD group (HR 1.66, 95% CI 1.45 to 1.9). HRs for the comorbid hypertension plus MDD group were significantly raised compared with hypertension alone (HR 1.22, 95% CI 1.1 to 1.35). Interaction measured using relative excess risk due to interaction (RERI) and likelihood ratios (LRs) were identified at baseline (RERI 0.563, 95% CI 0.189 to 0.938; LR p=0.0116) but not maintained during the follow-up.


LIMITATIONS
Possible selection bias in UK Biobank and inability to assess for levels of medication adherence.


CONCLUSIONS
Comorbid hypertension and MDD conferred greater hazard than hypertension alone for adverse cardiovascular outcomes, although evidence of interaction between hypertension and MDD was inconsistent over time. Future cardiovascular risk prediction tools may benefit from the inclusion of questions about prior history of depressive disorders.


STRENGTHS AND LIMITATIONS 53
• There were methodological advantages over similar studies including a very large sample 54 size, adjustment for a more comprehensive range of confounders and inclusion of non-fatal 55 adverse cardiovascular events from hospital admission data, along with death registry data. 56 • Definition of prior MDD history was based on ICD-10 diagnostic criteria as opposed to a 57 score on a questionnaire and our composite definition of hypertension incorporated past 58 history, current medication and objective blood pressure measurements. 59 • Sample was adjusted for a broad range of baseline factors such as smoking status, BMI, 60 psychotropic medication use and diabetes status amongst others, we were unable to see 61 how these factors changed over the course of follow-up, or assess adherence to medication. 62 • Although trained nurses interviewed participants to obtain medical information for group 63 assignment, as well as medication information, the self-reported nature may limit the 64 accuracy of information. 65 • UK Biobank does have some issues with selection bias, as such those with more severe 66 depression may be less likely to attend an assessment centre By 2030 major depressive disorder (MDD) and cardiovascular disease (CVD) will be the two leading 70 causes of disability worldwide 1 . It is established that individuals with MDD are at increased risk of 71 developing CVD and that they experience worse long-term outcomes, with higher mortality 2 . To 72 date, studies investigating the association between MDD and CVD have tended to focus only on 73 ischaemic heart disease, with the possibility of a specific association between MDD and 74 hypertension relatively under-investigated 3 . 75 Hypertension is extremely common (affecting 1 billion people worldwide) 4 and is responsible for 76 50% of all stroke and 50% of all ischaemic heart disease cases 5 . It is commonly comorbid with MDD, 77 particularly in older age-groups 6 7 . Furthermore, a biological link between hypertension and MDD is 78 supported by genome-wide association studies which have found that variants in calcium-channel 79 genes, important in blood pressure control and hypertension 8 , also act to increase risk for mood 80 disorders such as MDD 9 10 and bipolar disorder 11 12 . 81 Here we make use of prospective data from the UK Biobank cohort 13

Study design 89
This was population cohort study using data from UK Biobank. Four mutually exclusive groups 90 (hypertension only, MDD only, hypertension plus MDD, and a comparison group) were compared for 91 adverse CVD outcomes, as well as stroke outcomes 92 plus self-report of currently taking antihypertensive medication. This composite classification was 117 used to ensure that undiagnosed hypertensive participants were not omitted from analyses and is in 118 line with similar epidemiological studies 15-17 . According to these criteria, n=68,964 participants 119 (51.1% of the sample) had hypertension for the adverse cardiovascular outcomes analysis and 120 n=73,671 participants (52% of the sample) had hypertension in the stroke outcome analysis. 121 A past history of MDD was defined according to the criteria for mood disorders used in several 122 previous studies with UK Biobank data 18 19 . Participants were classified as MDD if they reported at 123 least one episode which comprised depression and/or anhedonia, lasting at least two weeks, plus 124 had consulted with a general practitioner or psychiatrist for mental ill-health (n=28,027 adverse 125 cardiovascular outcomes; n =29,528 stroke outcomes) 18 . 126

Confounding variables 150
Information on potential confounding factors was available for age, sex, socioeconomic status 151 (Townsend score) 21 , self-reported ethnicity, age of leaving full-time education, history of diabetes, 152 body mass index (BMI), systolic blood pressure, history of hypercholesterolemia, alcohol use, 153 smoking history, sedentary behaviour (number of hours each day spent sitting at a computer, 154 television or driving), physical activity levels 22 and psychotropic medication use. Specific details on 155 these variables are provided in supplementary digital content. 156 Analyses 157 Baseline characteristics were compared between groups using Chi-squared tests for categorical 158 variables and Kruskal Wallis for continuous variables. Confounding variables were assessed for 159 differences in adverse cardiovascular outcomes using log rank sums. For the four groups of interest 160 (comparator group, hypertension only group, MDD only group and hypertension plus MDD group) 161 we assessed associations with adverse cardiovascular outcomes using Cox proportional hazard 162 regression and the Efron method for ties 23 . Models were applied in a staged process. Our findings 163 are reported as unadjusted (model one), partially adjusted (model two) and fully adjusted (model 164 The assumption of proportionality of hazard was assessed for the four groups and each study 169 covariate using Schoenfeld residuals 24 . 170 We also assessed for evidence of an interaction between hypertension and MDD. The relative 171 excess risk due to interaction (RERI) 25 was calculated to assess for additivity in the risk at each month 172 where the proportionality assumption for the variables of interest was not met. All analyses were 173 performed with Stata statistical software, version 12 26 with the exception of RERI which was 174 calculated using the Microsoft Excel method of Andersson and colleagues, which allows for 175 comparison of adjusted outcomes 27 . 176 Psychotropic medication use was included as a confounding variable because of reports that these 177 medications may increase risk of mortality 28 but we also conducted a sensitivity analysis which 178 excluded the relatively small proportion of participants who were taking psychotropic medication. 179 Sub-group analyses looking separately at hazard rates in male and female groups only was also 180 carried out to assess for any gender specific differences a priori.  hazards assumption as tested by Schoenfeld residuals. Table 3 presents unadjusted and multivariate-212 adjusted HRs for adverse cardiovascular outcomes across the groups. In the fully adjusted model, 213 relative to the comparator group, the HR for adverse cardiovascular outcomes was significantly 214 raised for the hypertension only group (HR=1. 36, 95%CI 1.22-1.52) and was higher still for the 215 comorbid hypertension plus MDD group (HR=1.66, 95%CI 1.46-1.9) (sensitivity analysis HR=1.43, 216 95%CI 1.27-1.62; HR=1.72, 95%CI 1.49-1.999 respectively). Table 4 presents unadjusted and 217 multivariate-adjusted HRs for adverse cardiovascular outcomes across the groups with the 218 hypertension only group as comparator. In the fully adjusted model, relative to the hypertension 219 group, the HR for adverse cardiovascular outcomes was significantly raised for the comorbid 220 Within the sub-analysis, the model containing only the males showed a significant increase in hazard 225 ratio for hypertension (male HR 1.29, 95% CI 1.13-1.47) (table 5 of the supplementary digital 226 content) and comorbid MDD and hypertension (male HR 1.47, 95%CI 1.24-1.74). However, the 227 difference between comorbid disease and hypertension only was not statistically significant (male 228 HR 1.14, 95%CI 0.995-1.3). The female only sub-analysis showed an increase in hazard ratio for 229 hypertension (female HR 1.64, 95%CI 1.33-2.02) and a greater increase in comorbid MDD and 230 hypertension (female HR 2.18, 95%CI 1.82-2.92). The difference between comorbid disease and 231 hypertension only was also statistically significant (female HR 1.33, 95%CI 1.14-1.56). Sensitivity 232 analysis supported these findings. 233  Table 11 in the supplementary digital content  237 shows the full results for this analysis. 238

Stroke Outcomes 239
None of the independent variables for stroke outcome failed the proportionality assumption. Table 5  240 presents unadjusted and multivariate-adjusted HRs for stroke outcomes across the groups. In the 241 fully adjusted model, relative to the comparator group, the HR for stroke was insignificantly raised 242 for the hypertension only group (HR=1.21, 95%CI 0.97-1.51) and the depression only group 243 (HR=1.20, 95%CI 0.89-1.63) but significantly raised for the comorbid hypertension plus MDD group 244 (HR=1.37, 95%CI 1.04-1.79). In the hypertension comparator group, no group was significantly 245 different from the hypertension only group (table 6). Similar trends were shown in the gender 246 subset analysis but mainly not reaching significance. (Tables 7-8

in supplementary digital content) 247
An adjusted survival plot is shown in figure 3. Again, all results were supported by sensitivity 248 analysis. 249

DISCUSSION 250
In this large population cohort of middle-aged adults without CVD (adjusted for a broad range of 251 confounders), individuals with co-morbid hypertension and MDD were at increased risk of an 252 adverse cardiovascular event over time when compared to those with hypertension alone, 253 depression alone and neither condition. There was some evidence of an additive effect between 254 hypertension and MDD at baseline, but not throughout follow-up or within subgroup analyses. 255 Differences between co-morbid hypertension and depression were more marked in females. For 256 stroke outcomes, comorbid depression and hypertension was the only group that showed 257 significantly increased hazard ratios. 258 These observations are broadly consistent with our results but our study has a number of 284 methodological advantages, including a very large sample size, adjustment of analyses for a more 285 comprehensive range of confounders, and a focus on new-onset non-fatal and fatal adverse 286 cardiovascular events. We also used a definition of prior MDD history which was based on 287 diagnostic criteria within ICD-10 (rather than a threshold score on a depressive symptoms or general 288 wellbeing scale) and our composite definition of hypertension incorporated past history, baseline 289 medication and blood pressure measurements. 290 Limitations 291 However, some limitations are acknowledged. Recruitment criteria for UK Biobank may lead to 292 selection bias for this form of study. Specifically, age-restrictions may lead to underrepresentation of 293 early-onset hypertension and those with more severe forms of MDD may be less inclined to attend 294 for assessment. We also acknowledge limitations with our classifications of MDD and hypertension, 295 which were primarily self-report rather than formal diagnostic assessments. Although we have 296 excluded prior cardiovascular events where possible, the MDD plus hypertension sub-type may 297 capture older individuals with a degree of vascular depression, which has an established association 298 with raised blood pressure 33 . In addition, although we adjust for a host of risk factors at baseline 299 such as smoking status, BMI and psychotropic medication, we are limited by the lack of follow-up 300 data which could show change and modification of said risk factors over time. Similarly we were 301 unable to assess for medication adherence and transitions from one investigatory group to another. 302 Such modifications could explain the non-proportional nature of the depression group, which may in 303 itself be a predictor of poor medication adherence 34 . Although adherence to medication was not 304 formally assessed, the number and duration of antihypertensive medications used in the 305 hypertension plus MDD group was the same as for the hypertension only group (supplementary 306 digital content, table 12). As such, worse outcomes in the MDD plus hypertension group are 307 not explained by less intensive antihypertensive treatment at baseline. The amelioration of the 308 increased sympathetic activity which may be further increased in comorbid states 51 . Furthermore, 335 MDD is also commonly associated with unhealthy lifestyle factors such as smoking, sedentary 336 behaviour and poor diet and increased weight 19 and cardiovascular side effects of medications have 337 been commonly reported. Given the finding that comorbid disease and hypertension lead to 338 increased cardiovascular events, it would be useful to assess how treatment of these conditions 339 influences outcome. 340 Potential menopausal effects are tempting explanations for the variation with time in the female 341 only and overall analysis, especially given the age range of the cohort. It is widely accepted that 342 oestrogen has a protective impact on the heart which may be lost at menopause 53 . Furthermore, 343 increases in blood pressure are also noted at menopause 54 and depression may have a second 344 incidence peak around this time too. 55 56 Due to the MDD only group generally being younger and 345 having more females than the other groups, it may capture more of the menopausal change in 346 cardiovascular hazard during follow up than other groups leading to disproportionate hazards. 347 However, such findings are found in the female only analysis where age is similar between the 348 comparator group and the MDD only group. 349 The disproportional hazards in the MDD only group leads to a trend of lowered hazard ratios at the 350 start of follow up and a significantly increased hazard at the end of follow up. This lowered trend 351 initially leads to a significant RERI finding which is not maintained. Of note, MDD correlates highly 352 with neuroticism which has been shown to be inconsistent in regards to whether it is a risk factor or 353 protective, including in UK Biobank. 57 It is thought that conscientiousness and poor self-reported 354 health interact with neuroticism for better outcomes, it may be that premenopausal or 355 perimenopausal states may influence survival positively in women 58 . Of further investigatory interest 356 may be to assess if those with depression are more likely to develop hypertension during 357 Overall, our findings may have important implications for routine clinical practice, particularly within 361 primary care settings. Although evidence of an additive interaction is inconsistent, we found that 362 comorbid hypertension and depression conferred greater hazard than hypertension alone for 363 adverse cardiovascular outcomes. This significant finding remained after adjustment for factors such 364 as BMI, smoking status and diabetes and was robust to sensitivity analysis excluding those on 365 psychotropic medication. One possible implication is that clinicians should be more aware of the 366 negative long-term impact on CVD outcomes caused by a history of MDD in the context of 367 hypertension, particularly patients with no previous history of CVD. Although this work awaits 368 replication and testing in other cohorts and settings, it may be that future iterations of CVD risk 369 prediction tools, such as ASSIGN 59 , would benefit from the addition of a question on whether 370 individuals have a past history of MDD, so that they can be offered more intensive support to 371 prevent CVD 60 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y 27  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y 30     1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47 o n l y  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  429.2 were also excluded. hospital records are not available for the entire lifetime of study individuals, potentially missing some early cardiovascular events, as such those with selfdeclared prior cardiovascular disease at baseline were also excluded.

Blood Pressure
Blood pressure was measured in a sitting position partway through the interview and at the end of the interview using a digital blood pressure monitor (Omron HEM-7015IT.). Full protocol is available online https://biobank.ctsu.ox.ac.uk/crystal/docs/Bloodpressure.pdf

Physical activity
Physical activity was based on self-report, utilising the short form International Physical Activity Questionnaire (IPAQ). Participants reported the frequency and duration of moderate and vigorous activity along with walking undertaken in a typical week 3 . Data were analysed in accordance with the IPAQ scoring protocol 4 and total physical activity was computed as the sum of walking, moderate and vigorous activity, measured as metabolic equivalents (MET-hours/week). Physical activity was used in analyses as a continuous variable. Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Sedentary behaviour
Sedentary behaviour duration was derived from the sum of self-reported time spent driving, using computer and watching television. Those stating that they had performed "less than an hour" of sedentary activities were coded as 0.5hrs to allow use of a continuous variable. Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Socio-demographic and other covariates
Self-report on taking antihypertensive medication was taken from a question specific to cardiovascular medications, where antihypertensive medication was an option to respond.
Area-based socioeconomic status was derived from postcode of residence, utilising the census-derived Townsend deprivation index scored on housing, employment, social class and car availability where a negative score represents greater affluence 5 6 . Age was calculated from dates of birth and baseline assessment date. Smoking status was categorised into never, former and current smoking based on self-report, those who wished not to answer were coded as missing. Drink frequency was categorised into daily, three or four times a week, once or twice a week, one to three times a month, special occasions only, and never based on self-report. Those who wished not to answer were coded as missing. Medical history of diabetes and high cholesterol was collected from the selfcompleted, baseline assessment questionnaire of medical conditions. Ethnicity was categorised as Caucasian, black/mixed and Asian/mixed based on self-report. Other ethnicities coded as missing due to small numbers. Age at completing full-time education was categorised as (<16, 16, >16). Height and body weight were measured by trained nurses during the initial assessment centre visit. Body mass index (BMI) was calculated as (weight/height 2 ) and the WHO criteria 7 to classify BMI into: underweight <18. 5

Statistical analysis:
A best-fit multivariable regression spline model (stata command "mvrs") was used to find the best model to adjust for non-linear covariates. For the adverse cardiovascular outcomes, A single knot was fitted for age at age 50 and two knots were fitted for total physical activity at 1.65 and 8.062 metabolic equivalent hours. In the male subgroup analysis two knots were fitted for total physical activity at 1.7 and 8.507 metabolic equivalent hours, in the female subgroup two knots were fitted for total physical activity at 1.57 and 3.75 and two knots were fitted at systolic blood pressure 121.5 and 147.5. No bends were noted in any models for the stroke outcomes.

Model selection and covariate adjustment
Two continuous variables, age and total physical activity, expressed non-linearity within the main analysis and male subgroup analysis for cardiovascular outcomes and as such regression splines were used with two and three knots respectively. Two knots were included within the female subgroup analysis for physical activity. For stroke outcome there were no bends in the main or sexspecific models. Further detail on this is provided in the supplementary digital content. Within the main analysis for cardiovascular outcomes, the groups of depression only, Asian/Asian British ethnicity and BMI<18.5 covariates failed the proportionality assumption and as such, were incorporated into the model as a time varying coefficients. Within the sex specific models depression only failed the PH test within the female only analysis and ethnicity and BMI failed within the male only analysis. For the stroke outcomes gender and BMI class failed the proportionality assumption within the main analysis, with no failures within the sex-specific analysis. Analysis was repeated with the hypertension only as the comparator group to assess for any significant difference between the co-morbid group and the hypertension only group.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.
In your methods section, say that you used the STROBE cohort reporting guidelines, and cite them as: interaction between hypertension and MDD on survival 5 6 . MDD is well known to worsen post-81 cardiovascular event survival 6 7 . The risk to first onset cardiovascular is not known. Within this study 82 we look specifically at first onset events, irrespective of whether they lead to death or not.

83
Hypertension is extremely common (affecting 1 billion people worldwide) 8

193
Psychotropic medication use was included as a confounding variable because of reports that they 194 may increase risk of mortality 42 but we also conducted a sensitivity analysis which excluded None of the independent variables for stroke outcome failed the proportionality assumption.

283
There are distinct differences between our current paper and the previous publication. Follow-up 284 data within UK-Biobank has been released to allow meaningful prospective studies be conducted.
285 Thus, the current paper has the benefits of using hospital records and death certification for 286 outcomes, rather than self-reported data. We are also able to make inferences about the direction 287 of effect regarding MDD and CVD and assess the influence of hypertension and MDD over time, both  9 359 and 10). The findings from our study in this context suggest MDDs role as a risk factor for 360 cardiovascular disease and its relationship with blood pressure may be much more complex than 361 initially thought, in particular within female populations.

Blood Pressure
Blood pressure was measured in a sitting position partway through the interview and at the end of the interview using a digital blood pressure monitor (Omron HEM-7015IT.). Full protocol is available online https://biobank.ctsu.ox.ac.uk/crystal/docs/Bloodpressure.pdf

Depression definition
The criteria for lifetime MDD were created via the the following questions via touchscreen questionnaire were: "Looking back over your life, have you ever had a time when you were feeling depressed or down for at least a whole week?" (depression); "Have you ever had a period of time lasting at least two days when you were so irritable that you found yourself shouting at people or starting fights or arguments?" (irritability); "How many weeks was the longest period when you were feeling depressed or down?" (duration); "Have you ever seen a general practitioner (GP) for nerves, anxiety, tension or depression?" (consulted GP); "Have you ever seen a psychiatrist for nerves, anxiety, tension or depression?" (consulted psychiatrist). Participants were classified as having a history of MDD if they reported at least one episode which comprised of depression and/or irritability, with a duration of at least two weeks, plus had consulted with either a general practitioner or psychiatrist for mental ill-health.

Physical activity
Physical activity was based on self-report, utilising the short form International Physical Activity Questionnaire (IPAQ). Participants reported the frequency and duration of

Sedentary behaviour
Sedentary behaviour duration was derived from the sum of self-reported time spent driving, using computer and watching television. Those stating that they had performed "less than an hour" of sedentary activities were coded as 0.5hrs to allow use of a continuous variable.
Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Socio-demographic and other covariates
Self-report on taking antihypertensive medication was taken from a question specific to cardiovascular medications, where antihypertensive medication was an option to respond.
Area-based socioeconomic status was derived from postcode of residence, utilising the census-derived Townsend deprivation index scored on housing, employment, social class and car availability where a negative score represents greater affluence 5 6 . Age was calculated from dates of birth and baseline assessment date. Smoking status was categorised into never, former and current smoking based on self-report, those who wished not to answer were coded as missing. Drink frequency was categorised into daily, three or four times a week, once or twice a week, one to three times a month, special occasions only, and never based on self-report. Those who wished not to answer were coded as

Statistical analysis:
A best-fit multivariable regression spline model (stata command "mvrs") was used to find the best model to adjust for non-linear covariates. For the adverse cardiovascular outcomes, A single knot was fitted for age at age 50 and two knots were fitted for total physical activity at 1.65 and 8.062 metabolic equivalent hours. In the male subgroup analysis two knots were fitted for total physical activity at 1.7 and 8.507 metabolic equivalent hours, in the female subgroup two knots were fitted for total physical activity at 1.57 and 3.75 and two knots were fitted at systolic blood pressure 121.5 and 147.5. No bends were noted in any models for the stroke outcomes.

Model selection and covariate adjustment
All variables were tested against outcome measures (cardiovascular outcomes and stroke outcomes) using univariate analysis to assess appropriateness for inclusion in the final model. All covariates were significantly associated with the outcomes. and were Two continuous variables, age and total physical activity, expressed non-linearity within the main analysis and male subgroup analysis for cardiovascular outcomes and as such regression splines were used with two and three knots respectively. Two knots were included within the female subgroup analysis for physical activity. For stroke outcome there were no bends in the main or sex-specific models.
Within the main analysis for cardiovascular outcomes, the groups of depression only, Asian/Asian British ethnicity and BMI<18.5 covariates failed the proportionality assumption and as such, were incorporated into the model as a time varying coefficients. Within the sex specific models depression only failed the PH test within the female only analysis and ethnicity and BMI failed within the male only analysis. For the stroke outcomes gender and BMI class failed the proportionality assumption within the main analysis, with no failures within the sex-specific analysis. Analysis was repeated with the hypertension only as the comparator group to assess for any significant difference between the co-morbid group and the hypertension only group.

Time varying covariates
Due to the finding of MDD failing the proportionality assumption in the cardiovascular outcome in the primary analysis a series of further analyses have been performed to find when the assumption was not met. A log (-log) plot (fig 3) showed the proportionality assumption was broken at 22.5 months in the fully adjusted model in the primary analysis. As such, separate models were

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.
In your methods section, say that you used the STROBE cohort reporting guidelines, and cite them as: von

95
Here we make use of prospective data from the UK Biobank cohort 23 to test the hypothesis that a 96 lifetime history of MDD in individuals with hypertension impacts adversely on first-episode 97 cardiovascular events. We also assess whether MDD exacerbates the effects of hypertension as a 98 risk factor for CVD. 120 These exclusions were based on self-report (individuals who listed schizophrenia or BD from a list of 121 pre-existing medical conditions), or criteria for BD as per Smith et al, 24 or where they responded 122 "don't know" or "prefer not to answer" to questions or data was missing that would limit our ability None of the independent variables for stroke outcome failed the proportionality assumption.  Tables 7-8). An adjusted survival plot is shown in figure 2.

261
Again, all results were supported by sensitivity analysis excluding those on psychotropic medication.  to allow meaningful prospective studies be conducted. Thus, the current paper has the benefits of 288 using hospital records and death certification for outcomes, rather than self-reported data. We are 289 also able to make inferences about the direction of effect regarding MDD and CVD and assess the

Blood Pressure
Blood pressure was measured in a sitting position partway through the interview and at the end of the interview using a digital blood pressure monitor (Omron HEM-7015IT.). Full protocol is available online https://biobank.ctsu.ox.ac.uk/crystal/docs/Bloodpressure.pdf

Depression definition
The criteria for lifetime MDD were created via the the following questions via touchscreen

Physical activity
Physical activity was based on self-report, utilising the short form International Physical Activity Questionnaire (IPAQ). Participants reported the frequency and duration of  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   3 moderate and vigorous activity along with walking undertaken in a typical week 3 . Data were analysed in accordance with the IPAQ scoring protocol 4 and total physical activity was computed as the sum of walking, moderate and vigorous activity, measured as metabolic equivalents (MET-hours/week). Physical activity was used in analyses as a continuous variable. Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Sedentary behaviour
Sedentary behaviour duration was derived from the sum of self-reported time spent driving, using computer and watching television. Those stating that they had performed "less than an hour" of sedentary activities were coded as 0.5hrs to allow use of a continuous variable.
Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Socio-demographic and other covariates
Self-report on taking antihypertensive medication was taken from a question specific to cardiovascular medications, where antihypertensive medication was an option to respond.
Area-based socioeconomic status was derived from postcode of residence, utilising the census-derived Townsend deprivation index scored on housing, employment, social class and car availability where a negative score represents greater affluence 5 6 . Age was calculated from dates of birth and baseline assessment date. Smoking status was categorised into never, former and current smoking based on self-report, those who wished not to answer were coded as missing. Drink frequency was categorised into daily, three or four times a week, once or twice a week, one to three times a month, special occasions only, and never based on self-report. Those who wished not to answer were coded as  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Statistical analysis:
A best-fit multivariable regression spline model (stata command "mvrs") was used to find the best model to adjust for non-linear covariates. For the adverse cardiovascular outcomes, A single knot was fitted for age at age 50 and two knots were fitted for total physical activity at 1.65 and 8.062 metabolic equivalent hours. In the male subgroup analysis two knots were fitted for total physical activity at 1.7 and 8.507 metabolic equivalent hours, in the female subgroup two knots were fitted for total physical activity at 1.57 and 3.75 and two knots  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   5 were fitted at systolic blood pressure 121.5 and 147.5. No bends were noted in any models for the stroke outcomes.

Model selection and covariate adjustment
All variables were tested against outcome measures (cardiovascular outcomes and stroke outcomes) using univariate analysis to assess appropriateness for inclusion in the final model. All covariates were significantly associated with the outcomes. and were Two continuous variables, age and total physical activity, expressed non-linearity within the main analysis and male subgroup analysis for cardiovascular outcomes and as such regression splines were used with two and three knots respectively. Two knots were included within the female subgroup analysis for physical activity. For stroke outcome there were no bends in the main or sex-specific models.
Within the main analysis for cardiovascular outcomes, the groups of depression only, Asian/Asian British ethnicity and BMI<18.5 covariates failed the proportionality assumption and as such, were incorporated into the model as a time varying coefficients. Within the sex specific models depression only failed the PH test within the female only analysis and ethnicity and BMI failed within the male only analysis. For the stroke outcomes gender and BMI class failed the proportionality assumption within the main analysis, with no failures within the sex-specific analysis. Analysis was repeated with the hypertension only as the comparator group to assess for any significant difference between the co-morbid group and the hypertension only group.
Based on the STROBE cohort guidelines.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.

95
Here we make use of prospective data from the UK Biobank cohort 23 to test the hypothesis that a 96 lifetime history of MDD in individuals with hypertension impacts adversely on first-episode 97 cardiovascular events. We also assess whether MDD exacerbates the effects of hypertension as a 98 risk factor for CVD. 120 These exclusions were based on self-report (individuals who listed schizophrenia or BD from a list of 121 pre-existing medical conditions), or criteria for BD as per Smith et al,24 or where they responded 122 "don't know" or "prefer not to answer" to questions or data was missing that would limit our ability None of the independent variables for stroke outcome failed the proportionality assumption.  Tables 7-8). An adjusted survival plot is shown in figure 2.

Sedentary behaviour
Sedentary behaviour duration was derived from the sum of self-reported time spent driving, using computer and watching television. Those stating that they had performed "less than an hour" of sedentary activities were coded as 0.5hrs to allow use of a continuous variable. Participants who reported greater than 24 hours a day doing all activity were classified as missing.

Socio-demographic and other covariates
Self-report on taking antihypertensive medication was taken from a question specific to cardiovascular medications, where antihypertensive medication was an option to respond. Areabased socioeconomic status was derived from postcode of residence, utilising the census-derived Townsend deprivation index scored on housing, employment, social class and car availability where a negative score represents greater affluence 5 6 . Age was calculated from dates of birth and baseline assessment date. Smoking status was categorised into never, former and current smoking based on self-report, those who wished not to answer were coded as missing. Drink frequency was categorised into daily, three or four times a week, once or twice a week, one to three times a month, special occasions only, and never based on self-report. Those who wished not to answer were coded as missing. Medical history of diabetes and high cholesterol was collected from the self-completed, baseline assessment questionnaire of medical conditions. Ethnicity was categorised as Caucasian, black/mixed and Asian/mixed based on self-report. Other ethnicities coded as missing due to small numbers. Age at completing full-time education was categorised as (<16, 16, >16). Height and body weight were measured by trained nurses during the initial assessment centre visit. Body mass index (BMI) was calculated as (weight/height 2 ) and the WHO criteria 7

Statistical analysis:
A best-fit multivariable regression spline model (stata command "mvrs") was used to find the best model to adjust for non-linear covariates. For the adverse cardiovascular outcomes, A single knot was fitted for age at age 50 and two knots were fitted for total physical activity at 1.65 and 8.062 metabolic equivalent hours. In the male subgroup analysis two knots were fitted for total physical activity at 1.7 and 8.507 metabolic equivalent hours, in the female subgroup two knots were fitted for total physical activity at 1.57 and 3.75 and two knots were fitted at systolic blood pressure 121.5 and 147.5. No bends were noted in any models for the stroke outcomes.

Model selection and covariate adjustment
All variables were tested against outcome measures (cardiovascular outcomes and stroke outcomes) using univariate analysis to assess appropriateness for inclusion in the final model. All covariates were significantly associated with the outcomes. and were Two continuous variables, age and total physical activity, expressed non-linearity within the main analysis and male subgroup analysis for cardiovascular outcomes and as such regression splines were used with two and three knots respectively. Two knots were included within the female subgroup analysis for physical activity. For stroke outcome there were no bends in the main or sex-specific models.
Within the main analysis for cardiovascular outcomes, the groups of depression only, Asian/Asian British ethnicity and BMI<18.5 covariates failed the proportionality assumption and as such, were incorporated into the model as a time varying coefficients. Within the sex specific models depression within the main analysis, with no failures within the sex-specific analysis. Analysis was repeated with the hypertension only as the comparator group to assess for any significant difference between the co-morbid group and the hypertension only group.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.