Modifiable Lifestyle Factors and Risk of Stroke

Supplemental Digital Content is available in the text.

W hile many lifestyle factors have been associated with stroke risk, 1 ascertaining causality and whether their modification will reduce risk is less clear. Demonstrating causality with observational epidemiological studies is challenging due to biases such as confounding and reverse causation. One way to assess causality is with Mendelian randomization (MR), which uses genetic variants as instrumental variables in an approach analogous to a randomized controlled trial, to assess whether risk factors have causal associations with an outcome of interest. 2 Another challenge is that ischemic stroke is caused by various pathologies, which have distinct pathophysiological characteristics. Most epidemiological data examining risk factor modification have not studied the relationships between modifiable risk factors and specific stroke subtypes.
We used an MR approach to investigate the etiological role of multiple modifiable lifestyle factors on stroke and its subtypes.

METHODS
The authors declare that supporting data are available within the article and in the Data Supplement. The included genomewide association studies obtained ethical review board approval and informed consent. Stroke

CLINICAL AND POPULATION SCIENCES
We performed a 2-sample MR analysis using summary statistics from the largest publicly available genome-wide association studies for educational attainment, sleep duration, physical activity, smoking, alcohol and coffee consumption, 4 dietary components, body mass index (BMI), and waist-hip ratio (WHR; Table I in the Data Supplement). For the stroke outcome measures, we obtained summary statistics from the MEGASTROKE Consortium, 3 consisting of 67 162 cases and 454 450 controls (n=60 341 ischemic stroke; 9006 cardioembolic stroke (CES); 6688 large artery stroke (LAS); 11 710 small vessel stroke [SVS]), which we restricted to Europeans since all the lifestyle traits genomewide association studies were conducted in Europeans. We also obtained genome-wide association study summary statistics on intracerebral hemorrhage (ICH), 4 consisting of 1545 cases and 1481 controls. We used inverse-variance weighted meta-analysis 2 to combine the ratio estimates from each genetic variant into a single estimate of the causal effect of the lifestyle traits on each outcome. We also performed MR-based network mediation analysis 5 to examine the extent to which BMI and smoking mediate the protective effects of education on stroke. Further details are provided in the Data Supplement.

RESULTS
We analyzed 12 lifestyle traits for their associations with all stroke, any ischemic stroke (AIS), ICH, and 3 subtypes of ischemic stroke (CES, LAS, and SVS). Figure 1 summarizes the direction and magnitude of the association estimates for each lifestyle trait with each outcome. Figure 2 shows scatterplots of the associations of each genetic variant plotted against their association with the corresponding outcomes for all traits that had significant (false discovery rate, q<0.05) associations. Forest plots of the associations between lifestyle traits and each stroke subtype using the random-effects inverse variance-weighted (IVW) method are shown in Figure 3. Full results are provided in Table II  There were no significant associations of sleep duration, physical activity, dietary components, coffee intake, or alcohol consumption with AIS, ICH, or any stroke subtype.
The proportion of the total effect of education on stroke mediated through BMI and smoking was 1% for AIS and nearly 14% for CES and LAS (Table III in the Data Supplement). However, the indirect effects of education on each stroke subtype were not statistically significant (q<0.05), and for all stroke and SVS, there was inconsistent mediation because the indirect and total effects were in opposing directions, possibly due to insufficient power. Our mediation analysis, therefore, lacked sufficient evidence to state that the effects of education are mediated through either BMI or smoking.

Sensitivity Analyses
Analyses using the MR-Egger, weighted median, simple mode, and weighted mode were similar to the IVW results (Table II in the Data Supplement), with a few exceptionssuch as the effects of education on ICH and smoking on SVS, which had contradictory directions of effect for the MR-Egger method compared with other methods. The discrepancy in the direction of association using the MR-Egger approach may be due to violation of the InSIDE assumption.
We also individually analyzed each smoking trait contributing to the lifetime smoking index. Smoking initiation was significantly associated with LAS, but there were no significant associations of age of initiation of regular smoking, cigarettes per day, or smoking cessation with stroke subtypes (Table IV in the Data Supplement).

DISCUSSION
We found genetically predicted years of education to be inversely associated with AIS, LAS, SVS, and ICH.

SCIENCES
Genetically predicted ever smoking regularly and higher BMI and WHR were associated with AIS and LAS. The associations of education, BMI, and smoking with risk of AIS and its subtypes were independent. Our analysis did not identify any causal associations with CES. A possible explanation is that most cases of CES are due to atrial fibrillation, which has a different risk factor profile to other stroke subtypes.
Our MR results provide genetic evidence for an inverse causal effect of educational attainment on ischemic stroke and ICH, consistent with previous epidemiological data. The association of the lifetime smoking index with risk of AIS, LAS, and SVS is in concordance with observational studies and MR analyses, 6 and our findings for BMI and WHR in a larger sample confirm a previous MR analysis. 7 We did not identify associations with genetically predicted longer sleep duration, increased physical activity, or consumption of fat, protein, carbohydrates, sugar, coffee, or alcohol. The number of genetic instruments was small for physical activity and dietary components (<10 single nucleotide polymorphisms each), which may have introduced weak instrument bias.
Our findings support several potential policy recommendations. Genetic heritability only explains a fraction of total variation in modifiable traits (education: 11%, smoking: 2%, BMI: 6%, and WHR: 3.9%). Therefore, regardless of their genetic risk, individuals can make lifestyle changes to reduce their stroke risk, such as smoking cessation. 1 Our study has strengths and limitations. The large sample sizes provided substantial power, and we investigated a wide range of lifestyle factors. However, we only had access to summary-level data rather than individual-participant data, so we were unable to evaluate at the individual level whether specific genetic variants increase stroke risk. While we captured the impact of a lifelong intervention in each lifestyle factor, we did not determine whether individuals who implement lifestyle changes reduce their stroke risk. Additionally, our analyses involved individuals of European-only ancestry. Overlap in participants between the exposure and outcome data sets was very small so the impact on the results was likely minimal. Finally, the small number