Response to Letter to the Editor for “Self-reported snoring and incident cardiovascular disease events: results from the Jackson Heart Study.”

We have read the letter to the editor on our original manuscript, “Self-reported snoring and incident cardiovascular disease events: results from the Jackson Heart Study.” We thank Gao and colleagues for their insight into our work, and critique of the research we conducted. We appreciate the opportunity to engage in a dialogue relevant to the topic of snoring and risk of cardiovascular disease (CVD). The authors outline several points as to why the conclusion reached using our data may be flawed (i.e. the finding that there was no significant association between self-reported snoring and incident CVD in African Americans). We would like to clarify our understanding of the available data in light of some of the points made by authors of the letter.

First, the authors of the letter mention that metabolic syndrome and its components might mediate, and be causal in the association between obstructive sleep apnea (OSA) and CVD, which may offset the significance level. Additionally, the authors believe that a mediation analysis might have been used to address this mediation effect.

The criteria for the diagnosis of metabolic syndrome include waist circumference (obesity), hypertension (HTN), impaired fasting glucose, and dyslipidemia. Several components of the metabolic syndrome may mediate the association between OSA/snoring and CVD, including obesity, HTN, and impaired fasting glucose. Therefore, in addition to the fully adjusted model, we also reported an intermediate model, that adjusted for all covariates, excluding HTN and DM, as these may be in the causal/intermediary pathway for CVD. Despite this, we did not find a significant association between self-reported snoring and incident CVD events. Moreover, of the components of the metabolic syndrome, the only one that current evidence supports as causally related to OSA is hypertension. The remainder are not definitely proven to be on the causal pathway between OSA and CVD. Therefore, although mediation analysis would help explore the causal mechanisms that link OSA and CVD, this was not the purpose of our manuscript. We agree that future independent studies should conduct mediation analyses using these variables to further investigate the relationship between OSA and CVD, and assess the relative magnitude of different pathways and mechanisms by which this exposure may affect the outcome.

Second, the authors mention that additional snoring related variables (i.e. age, sex, smoking status, body mass index (BMI)) included in the adjusted model, may introduce a “multicollinearity” problem, which may not provide the valid result.

We adjusted for several variables in our fully and partially adjusted models, and agree that this could potentially introduce a multicollinearity problem. However, we chose to adjust for these variables as confounders in the relationship between snoring and CVD, given that the chosen variables are known risk factors for CVD, based on multiple prior studies. Therefore, our results could have been biased towards the null due to this, however the validity of our results could also be questioned had we excluded these variables from our multivariate analyses.

Third, the authors point out that effect modification might also contribute to the null snoring-stroke association in the unadjusted models, given the significant differences between snorers and non-snorers in terms of sex, HTN, DM, smoking, and family history of heart disease found in table 1. Additionally, they report that data presented on demographics by incident CVD might have strengthened the overall analysis, with multi-variable models to evaluate the snoring-CVD association by effect modifiers.

We agree that Table 1 in our manuscript shows statistically significant differences in sex, education status, diagnoses of HTN, and history of smoking, which were adjusted for in our analysis of CVD events amongst snorers and non-snorers. However, there were only a total of 48 composite CVD events in the snorer group and 260 composite CVD events in the non-snorer group. Our study was not powered or designed to evaluate the association between snoring and stroke-specific CVD event stratified by the covariates listed above.

Fourth, the authors of the letter questioned whether the baseline demographic data used as covariates were based on previous studies and/or references.

Our study adjusted for several demographic covariates, including (but not limited to): age, sex, education, income, marital status, smoking status, BMI, hypertension, diabetes, etc. Each of these covariates were in fact based on prior studies. Age, sex, BMI, hypertension, diabetes and smoking are known confounders for CVD, and were chosen based on prior published association of cardiovascular events with all-cause mortality or sleep disorders.[1] Additionally, prior studies have shown that individuals who are unmarried, without bedpartners, unemployed, or have a lower education status have greater difficulties in responses, and may not accurately report snoring, which may under-estimate the total number of snorers in the group, therefore biasing the results toward the null due to a smaller sample size in the exposure group.[2]

Lastly, the authors argued that our definition of habitual snoring lacked validation, and did not capture the complexities of snoring. Moreover, they mentioned that definitions for CVD outcome, time to CVD event, and included/excluded criteria are debatable.

In regard to snoring, we agree that the definition of habitual snoring used in our study has some inherent weaknesses, outlined in the manuscript itself. Our study had a very low prevalence of unsure snoring responses (n = 2 reported “do not know” for their snoring status). It is possible that the way the question was phrased (“You are told that you snore loudly and bother others”) resulted in misclassification due to an obligatory response, despite being unsure of their snoring status. Therefore, it is possible that the JHS interviewer technique may have led to a lack of unsure responses, and an overall under-estimation of habitual snorers. However, we must recognize, that much of the literature dedicated to the assessment of habitual snoring faces a similar problem — the lack of an agreed-upon, gold-standard definition. The determination of the presence of habitual snoring should be standardized and quantified to allow for uniformity in comparing study findings. Snoring sound intensity or frequency should be further investigated in order to obtain objective snoring data and definitively answer the question of whether habitual snoring is independently associated with atherosclerosis and CVD.

Regarding the definitions for the primary outcomes, these were clearly defined in the methods section of the paper. The primary outcome was an incident CVD event defined as hospitalization for a stroke, myocardial infarction, coronary revascularization procedure, or death from CHD. Time to CVD event was defined as person- years of follow-up, starting at the baseline examination and ending with the earliest CVD event date detected at censoring date (12/31/2012), date of death, or lost-to-follow-up. Lastly, in response to the authors comment regarding significance of p-values, we report that a p-value of < .05 was considered to be statistically significant in our study.

Once again, we appreciate the opportunity to engage in this rebuttal with Gao and colleagues, and thank the authors for their input. As outlined in the manuscript, we acknowledge several limitations in our study, including small sample size and cardiovascular event rate, use of self-reported rather than objective snoring measures, and a lower prevalence of snoring reported in this population, which maybe secondary to the JHS interview technique, resulting in a biased “unsure” response. Our manuscript should prompt future investigations of the snoring-CVD association among African Americans, which will be better served with the inclusion of more objective data on both snoring and sleep apnea, in a larger sample size and a longer follow-up period to determine CVD events.