Study design and participants
In 1996–2002, 146,556 adults aged ≥30 years were recruited from the general population in Cuba into a prospective cohort study. Details of the study design and survey methods have been reported previously [11]. Briefly, the study recruited men and women from five of Cuba’s fourteen provinces. Within each province, medical offices were selected at random using a computer-generated random allocation sequence. Each medical office subsequently invited all of its patients aged 30 years and older to participate in the study. Overall, 74% of eligible individuals participated.
Trained health-care staff (mostly the family doctor, but on occasion local nurses or other trained health care workers) visited each household. After providing written consent, participants provided information on age, sex, ethnicity, education, occupation, marital status, lifestyle factors, current medications, and medical history (the original questionnaire in Spanish, developed solely for this study, and the English translation are available in Additional file 1: p2–3). Blood pressure was measured twice while the participant was seated (once towards the beginning of the interview and once towards the end), using a calibrated manual sphygmomanometer and standard techniques. Following the home visit, participants were invited to attend their physician’s medical office for height and weight measurements. Between 2006 and 2008, baseline measures were repeated using the same procedures in a formal resurvey of 24,345 participants (17% of the initial study population; all residents of Matanzas province) to assess temporal variation in estimated levels of risk factors at baseline, particularly blood pressure and BMI.
Participants were followed up until 1st January 2017 through electronic record linkage to the Cuban Public Health Ministry’s mortality records using national identification numbers. Mortality records are collected for all deaths in Cuba, and include medically certified causes of death. In 1996–2000, all deaths in Cuba were coded using the 9th edition of the International Classification of Disease (ICD-9), while all deaths from 2001 onwards were coded using ICD-10 (Additional file 1: Table S1).
Statistical analyses
The presence of diabetes was defined as a (self-reported) previous medical diagnosis of diabetes, the use of antidiabetic medication (oral hypoglycemic medication or insulin), or both. The mean of the two systolic blood pressure (SBP), or the two diastolic blood pressure (DBP), measurements was used in all blood pressure analyses, and body-mass index (BMI) was calculated as weight in kg divided by the square of height in m.
Participants with missing SBP, DBP, diabetes status, BMI or other key covariates, were excluded from the main analyses, as were those with implausible or outlying values for SBP (<80 or ≥250 mmHg), DBP (<40 or ≥ 150mmHg), or BMI (<15 or ≥40 kg/m2), those with prior cardiovascular disease at baseline (myocardial infarction, angina, or stroke), and those with no follow-up at ages 35–79 years. To further limit any effect of pre-existing disease on blood pressure or BMI at baseline, the first 5 years of follow-up were also excluded.
Cox regression was used to relate mortality rate ratios (RR) at ages 35–79 years to SBP (using the mean of the two SBP measurements), diabetes and BMI, with adjustment for age at risk (5-year groups), sex, highest completed level of formal education (less than primary, primary, lower secondary, high school [or technical training], university), smoking (never, ex-smoker, current smoker of < 20 cigarettes/day, current smoker of 20 cigarettes/day, current smoker of >20 cigarettes/day, other smoker), alcohol (non- drinker, less than weekly drinker, at least weekly drink of < 1 bottle/week of 35 cl rum [or equivalent alcohol] per week, at least weekly drinker of 1- <3 bottles/week, and at least weekly drinker of 3 plus bottles/week), and, where appropriate, BMI (15- <20, 20- <22.5, 22.5- <25, 25- <27.5, 27.5- <30, 30- <40 kg/m2); analyses of DBP are given in Additional file 1. RRs were corrected for regression dilution (ie, categorising people by their baseline SBP or BMI and estimating the long-term average mean SBP or BMI in each category using the correlation between re-survey and baseline measurements [12]), and are therefore described as associations of usual SBP and usual BMI with mortality [3, 4].
In categorical analyses, SBP was categorized as <125, 125- <145, 145- <165, ≥165 mmHg, and BMI was categorized as 15- <20, 20- <22.5, 22.5- <25, 25- <27.5, 27.5- <30, 30- <35, 35- <40 kg/m2. Confidence intervals (CIs) were calculated using the variance of the log risk, which appropriately attributes variance to all groups, including the reference, and so allows CIs to be used to compare risks in any two groups [13]. Linear associations are reported per 20 mmHg higher usual SBP and per 10 kg/m2 higher usual BMI.
Sensitivity analyses of the main prospective associations were conducted by further adjusting the associations for other potential confounders, and by excluding participants with non-vascular chronic diseases at baseline (including chronic obstructive pulmonary disease, liver cirrhosis, chronic kidney disease, peptic ulcer, and cancer) and those taking blood pressure-lowering medication. The analyses of BMI are also reported with and without further adjustment for usual SBP and diabetes (including the effect of such adjustments on the Wald chi-squared statistic), to assess the extent to which these factors mediate the association.
The fraction of cardiovascular deaths in Cuba attributed to SBP, diabetes and BMI were estimated by age group and sex, using the formula Pe (RR – 1)/(Pe [RR-1] + 1) where Pe is the prevalence of the given risk factor in Cuba as estimated by the 2010 Cuban National Non-communicable Disease Risk Factor Survey [14], and RR is the rate ratio in the present study [15]. These fractions were multiplied by the number of vascular deaths in each age group and sex in Cuba in 2015 [16] to give the overall number of deaths attributable to each of these risk factors at ages 35–79 years. All analyses were conducted in SAS (version 9.4), and results were plotted in R (version 3.6.2).