Modification of cardiovascular disease risk by health behaviour change following type 2 diabetes diagnosis

Among adults with type 2 diabetes (T2D), unhealthy behaviours are associated with increased risk of cardiovascular disease (CVD) events. To date, little research has considered whether healthy changes in behaviours following T2D diagnosis reduce CVD risk.


| INTRODUCTION
Health behaviours contribute to the development and progression of type 2 diabetes (T2D), including the risk of cardiovascular disease (CVD) complications. [1][2][3] Moderate levels of physical activity, a nutritionally balanced diet and smoking cessation are recommended for the management of T2D. 4 Studies have suggested that healthy changes in these behaviours after T2D diagnosis may reduce CVD risk. [5][6][7][8] However, these studies have not explored whether the association between changes in diet or physical activity and CVD outcomes varies according to baseline levels of these behaviours.
Previously, The Anglo-Danish-Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care (ADDITION)-Cambridge trial found that participants who increased their physical activity following T2D diagnosis had a lower 5-year risk of CVD events compared to participants who did not. 7 Baseline differences in physical activity among participants who increased activity levels, compared to those who did not, may have contributed to the observed association, but this was not explored. To date, little research has investigated whether associations between physical activity levels and CVD are modified by changes in physical activity following T2D diagnosis.
Dietary behaviours, such as higher fibre intake, have been associated with lower risk factors for CVD in people with T2D. 9 Additionally, changes in these behaviours following T2D diagnosis, such as increasing fruit and vegetable intake, have been associated with reduced CVD risk factors. 5 Separately, ADDITION-Cambridge previously found that reducing alcohol consumption in the year following T2D diagnosis lowered 5-year and 10-year incidence of CVD events. 7,8 However, whether the effect of dietary behaviours and alcohol consumption on CVD risk is modified by behaviour changes following T2D diagnosis remains to be established.
T2D diagnosis may be a 'teachable moment' when those diagnosed become aware of the benefits of adopting healthier behaviours. Participants in the ADDITION-Cambridge cohort were newly diagnosed with screen-detected T2D at time of enrolment and were therefore at an early stage of disease progression. Behaviour changes may affect people differently depending on their behaviours at the time of T2D diagnosis. We aimed to identify whether changes in behaviours impacted the association between behaviours prior to T2D diagnosis and long-term CVD risk. Our objectives were to evaluate the associations between health behaviours at diagnosis and the 10-year incidence of CVD, and to assess whether behaviour changes made within the first year following diagnosis modified these associations.

| Population
ADDITION-Cambridge is a population-based screening programme for T2D, followed by a cluster-randomized intervention trial to evaluate the effectiveness of intensive multifactorial treatment compared to routine care in reducing CVD risk in those with screen-detected T2D. 10,11 Screening invitations were sent to 33,539 individuals who met eligibility criteria: aged 40-69 years, at high risk of undiagnosed diabetes and registered at a participating general practice (GP) within Eastern England (n = 49). 10 High risk was determined by a validated risk tool using routinely collected information from participants' medical records. 12 Of those invited, 73.5% attended the first stage of screening. 13 Diabetes was diagnosed according to World Health Organization criteria (1999). 14 This identified a cohort of 867 individuals with screen-detected T2D, all of whom consented to participate in the treatment study and were enrolled between 2002 and 2006. Participants at practices randomized to routine care (n = 23) received diabetes care according to National Health Service guidelines. 10 Participants at practices randomized to intensive treatment (n = 26) received more frequent consultations and educational material packs on healthy behaviours, while physicians were encouraged to follow a protocol that featured earlier use of medication to achieve more intensive treatment targets. 10 Results suggested that the intervention did not significantly improve CVD outcomes. 11 Thus, data Novelty statement: What is already known on this subject? • Diet and physical activity are risk factors for type 2 diabetes (T2D) and cardiovascular disease (CVD). • Little research has assessed whether changes in these behaviours following T2D diagnosis modify CVD risk. What has this study found? • High fat intake was associated with higher 10-year hazard of CVD. However, reducing fat intake in the year after T2D diagnosis attenuated this risk. What are the clinical implications of the study? • Among those with newly diagnosed T2D, reducing fat intake may lower long-term risk of CVD. This suggests caution is warranted regarding the increasing use of low carbohydrate diets to treat CVD, as such diets typically result in increases in fat intake.
from the two arms were pooled for the present study: an observational secondary analysis of the ADDITION-Cambridge cohort. Ethical approval was obtained from relevant Multi-Centre Research Ethics Committees. 10

| Measurements
Smoking, alcohol consumption, physical activity, fibre, total energy and fat intake, were ascertained by self-report, while plasma vitamin C was measured using a fluorometric assay. 10 All behaviours were measured at baseline (T2D diagnosis) and 1 year later. Physical activity was assessed by the validated European Prospective Investigation into Cancer (EPIC)-Norfolk Physical Activity Questionnaire 10 and was defined as moderate/vigorous physical activity (MVPA) in the past year. Fibre, total energy and fat intakes over the past year were estimated by the validated EPIC-Norfolk food frequency questionnaire. 10 Smoking status (never, former, current), alcohol consumption (units/week), MVPA (minutes/week), plasma vitamin C (μmol/l), fibre intake (g/day), total energy intake (Kcal/day) and percentage of total energy intake from fat (%) are targets of healthy behaviour recommendations for those with T2D. Body mass index (BMI) was derived from measures at baseline using a fixed rigid stadiometer and SECA scales. Socio-demographic information (age, sex, age left full-time education, occupational social class) and prescribed medication use were also self-reported at baseline.

| CVD outcomes
CVD outcomes were assessed from date of T2D diagnosis to 31 December 2014 and were defined as a composite of first cardiovascular event during the study period, including cardiovascular mortality, non-fatal myocardial infarction, non-fatal stroke, non-traumatic amputation and revascularization. 10 National registers, hospital and GP records were used to ascertain incidence of non-fatal CVD events, while Office of National Statistics mortality data was used to ascertain incidence of deaths. 7,10 All events were independently adjudicated. Outcomes were ascertained for 99.9% of participants. 13

| Health behaviours and CVD incidence
Continuous values of plasma vitamin C, fibre intake, total energy intake, percentage of total energy from fat and MVPA were categorized in quartiles. Supplemental analyses considered percentage of energy from saturated fat and mono-and poly-unsaturated fats. For alcohol consumption, categories of abstainer (0 units/week), light-moderate (1-14 units/week) and heavy (>14 units/week) were used. Referent groups were quartile one for all health behaviours except total energy intake (which used quartile three), smoking (never) and alcohol consumption (1-14 units/week). Hazard ratios (HRs) for the association between each health behaviour at baseline and 10-year CVD incidence were estimated by Cox proportional hazards regression. Participants were at risk for an incident CVD event from date of T2D diagnosis. Follow-up time was censored at the earlier of first CVD event, non-CVD death, or 31 December 2014. The Schoenfeld residuals test and log-log survival curves indicated no violations of the proportional hazards assumption. All models were adjusted for age (continuous), sex (men/ women), occupational social class (managerial/professional, intermediate, routine/manual), age left full-time education (<16 years, 16-18 years, >18 years), baseline BMI (continuous), baseline anti-hypertensive medication use (yes/no), baseline lipid-lowering medication use (yes/no), treatment group (intervention/routine care) and baseline smoking status where smoking was not the primary exposure (current, never/former). Confounders were selected a priori and were variables previously established to be associated with the exposure and outcome. We did not adjust for glucose-lowering medication use because only four participants reported use at baseline. Robust standard errors accounted for clustering by GP.
We performed a complete case analysis. Individuals were excluded from an analysis if they were missing behaviour or covariate information. Separately, we performed a sensitivity analysis excluding 99 participants who reported having a history of CVD prior to enrolment.

| Behaviour change and CVD incidence
To assess whether behaviour changes made in the year following T2D diagnosis modified the baseline associations, we modelled an interaction term between the behaviour at baseline and behaviour change between baseline and 1 year. Behaviour change was defined as the value at baseline subtracted from the value at 1 year. In these models, the risk period for CVD events began at the 1-year follow-up visit. Modification by change in smoking status was not investigated as too few participants changed their smoking status (n = 21). The testparm command was used to assess the presence of statistically significant interaction simultaneously across the categories of each health behaviour. A Wald test p-value <0.05 suggested that change in behaviour modified the baseline association with CVD. We then assessed if the presence of modification differed depending on baseline exposure category. HRs for the association between a one standard deviation (SD) increase and decrease in each health behaviour with CVD were calculated from the linear combination of estimates generated from the models. These HRs were then compared to the HRs unadjusted for interaction terms to examine whether behaviour change modified the baseline association between the behaviour and CVD risk.

| RESULTS
Of the 867 study participants, median age at diagnosis was 62.7 years and 61% were men (Table 1). Mean baseline BMI was 33 kg/m 2 . There were 126 CVD events during the study period: 31 CVD deaths, 21 myocardial infarctions, 33 strokes, 40 revascularizations and one amputation. Mean follow-up time was 9.6 years from T2D diagnosis. Seven hundred and fifty-eight (87%) participants attended for 1year measurements.
Compared to never smokers, former and current smokers had a higher hazard of CVD, with HRs of 1.61 (95% CI: 1.07, 2.42) and 1.73 (95% CI: 1.04, 2.87) respectively ( Table 2). Those in the highest quartile of fat intake had a higher hazard of CVD compared to participants with the lowest intake [HR: 1.70 (95% CI: 1.02, 2.85)]. Compared to those with the lowest levels of plasma vitamin C, those with the highest levels had a 56% lower hazard of CVD (95% CI: 0.22, 0.87). Those with the highest fibre intake had a lower hazard of CVD compared to those with the lowest intake [HR: 0.60 (95% CI: 0.36, 0.99)]. Results were robust to the inclusion of participants with a CVD history prior to diagnosis (Table S1).
The association between percentage of energy from fat and CVD was modified by changes in fat intake in the year following T2D diagnosis (p-value for interaction term: 0.04). Among those in the highest quartile of fat intake at baseline, a one SD (6.48%) increase in energy from fat was associated with a higher hazard of CVD  (Table  S2). Modification of these associations by 1-year changes in saturated or unsaturated fat intake was not statistically significant. However, increases in saturated fat intake among individuals with high baseline saturated fat intake appeared to correspond with an increase 10-year hazard of CVD. Among individuals who reduced their saturated fat intake, there was no elevated risk of CVD (Table S3).
The Wald test p-value for overall interaction was >0.05 for other behaviours (Table 3). However, the magnitudes of the stratified HRs for plasma vitamin C and MVPA, while not statistically significant, suggest the potential for modification by behaviour change. For plasma vitamin C, a one SD (22.85 μmol/L) decrease showed a higher hazard of CVD among those in the third highest quartile of intake

| DISCUSSION
Among the 867 adults with screen-detected T2D enrolled in ADDITION-Cambridge, current or former smoking and a high percentage of energy from fat were associated with a higher hazard of 10-year CVD incidence, while higher levels of plasma vitamin C and fibre intake were associated with a lower hazard of CVD incidence. The association between percentage of energy from fat and CVD risk was modified by changes in fat intake in the year following T2D diagnosis, where a 6.48% decrease attenuated the association with CVD. We did not detect statistically significant modification of the associations between other behaviours and CVD.
To date, little research has assessed the impact of withinindividual changes in health behaviours following T2D diagnosis on CVD events. We found that, while high fat intake was associated with elevated CVD risk, there was no elevated risk of CVD among those who had decreased their fat intake in the year after T2D diagnosis. Saturated fat showed similar but non-statistically significant associations with CVD, while unsaturated fat intake was not associated with CVD risk. This may suggest that the observed association between fat intake and CVD may be primarily driven by changes in saturated fat as opposed to unsaturated fat intake. However, future research focused on dietary sources of fat intake would be better suited to address this question. This result is supported by a recent meta-analysis which showed that reductions in saturated fat lowered risk of CVD events 15 and one study has reported this among individuals with T2D. 16 These findings are particularly relevant given the recent popularity of low-carbohydrate diets for T2D treatment. The effects of low-carbohydrate diets on CVD risk remain unclear, 17,18 however several studies have found that low-carbohydrate diets, particularly those coupled with high animal fat intake, are associated with higher all-cause mortality rates. [19][20][21] Our research supports that a diet leading to higher fat intake may increase long-term CVD risk, which raises concerns | 5 of 9 ESTLIN ET aL.
regarding low-carbohydrate high-fat approaches. However, further research considering the sources of fat is needed to inform dietary recommendations.
Prior research has shown that increased MVPA following T2D diagnosis lowers CVD risk. 7 We did not identify any impacts of increases in physical activity on the observed associations with CVD, perhaps due to the small average increases in activity (27.76 min/week), and the low power of the effect modification analyses. For this reason, the lack of evidence from this study on the effect of increases in MVPA on CVD should not undermine the established benefits of physical activity on CVD risk. Furthermore, we did not detect an  association between changes in plasma vitamin C and CVD, although this has been demonstrated previously. 5,16 The null association in our study may also be due to the small average changes in plasma vitamin C levels (1.99 μmol/L) as well as the low power of the stratified effect modification analyses. A recent ADDITION-Cambridge analysis found reducing alcohol consumption by >2 units/week was associated with lower 10-year hazard of CVD. 8 However, our results did not find changes in alcohol consumption following T2D diagnosis to modify the association between baseline alcohol consumption and CVD. One key distinction is that in this analysis we did not count abstainers among those who reduced their alcohol intake. The average change among moderate drinkers was insubstantial, which may have limited our ability to detect modification, further impacted by the low power of the stratified analyses. Additionally, changes in fibre intake and total energy intake were not found to modify the baseline associations with CVD. This may be because the average changes in fibre intake and total energy intake, aside from the group with the highest total energy intake at baseline, were small.
A key limitation of this study is the low number of events, which may have impacted our ability to detect associations. Stratified models assessing modification by behaviour change compromise statistical power, limiting our ability to detect effect modification. Results for alcohol consumption, fibre intake and total energy intake suggest an absence of interaction on the multiplicative scale, however, we did not assess interaction on the additive scale. Additionally, results are sensitive to how behaviour change has been defined. A one SD change may constitute quite a large behaviour change (e.g. increasing fibre intake by 11 g/day equates to eating three pears), and the average behaviour changes among the cohort were generally small. Behaviour change was measured 1 year after diagnosis and may not have been maintained long term. It is possible that changes in behaviours may take longer than 10 years to generate differences in CVD risk. Despite the small number of events and small changes in behaviours, it is evident that increases in fat intake may be harmful. We cannot exclude the possibility that changes in other behaviours may be beneficial or harmful in circumstances where larger changes are achieved and where changes are maintained for longer periods of time.
Another limitation is the exposure assessment. Except for plasma vitamin C, behaviours were ascertained by selfreport, which may result in exposure misclassification and would consequently attenuate the observed associations. Social desirability bias may have influenced reporting of health behaviours, 22 including alcohol consumption and smoking. Additionally, weight-dependent bias may have resulted in those with higher BMI under-reporting dietary intakes. Furthermore, we tested for effect modification by six variables, and while we did not correct for multiple testing, T A B L E 2 Hazard ratios for 10-year cardiovascular disease (CVD) incidence associated with each health behaviour at baseline. T A B L E 3 Hazard ratios for 10-year cardiovascular disease (CVD) incidence associated with a one standard deviation (SD) increase and decrease in the health behaviour in the year following diabetes diagnosis. we interpreted results with caution considering the magnitude and precision of the estimates, in light of the large number of tests run. Participants were invited to the screening programme based on a diabetes risk score which selected individuals with higher BMI. It is possible that the cohort may be more (or less) responsive to changes in diet than individuals with lower BMI. Finally, the majority of participants had overweight/obesity at the time of T2D diagnosis and were white. Thus, results from this study may not be generalizable to other groups. It is possible that the observed associations between dietary fat and CVD risk may vary among subgroups with different diabetes characteristics or cardiovascular risk factors, and future research with the appropriate data should explore this question. ADDITION-Cambridge had a high attendance at screening and all eligible individuals identified with T2D consented to enrol. Since T2D was screen detected and behaviour measurements were repeated the year after diagnosis, this study was able to assess behaviour change early in T2D progression. This ability, along with the population-based nature of the study, are particular strengths compared to other research on health behaviours among adults with T2D. Other strengths are the objective assessment of plasma vitamin C, a marker of fruit and vegetable intake, and the independent adjudication of CVD events.
This study emphasizes the importance of healthy vitamin C, fibre and fat intake, and the prevention of smoking in reducing CVD risk among those with T2D. Additionally, these results support recommending those newly diagnosed with T2D to reduce their fat intake, as this may lower their hazard of CVD incidence and mitigate effects of a high-fat diet on long-term CVD risk. This evidence may raise concern regarding low-carbohydrate diets in diabetes treatment, as these diets typically lead to increases in fat intake. Future research incorporating data on the sources of fat intake are needed to inform dietary recommendations. Few studies have assessed long-term health impacts of changes in behaviours following T2D diagnosis and further research is needed to support these results. Future studies using objective behaviour measurements and with longer follow-up will help to better our understanding of the role of behaviour changes following T2D diagnosis in the prevention of CVD.