Impact of common cardio-metabolic risk factors on fatal and non-fatal cardiovascular disease in Latin America and the Caribbean: an individual-level pooled analysis of 31 cohort studies

Background Estimates of the burden of cardio-metabolic risk factors in Latin America and the Caribbean (LAC) rely on relative risks (RRs) from non-LAC countries. Whether these RRs apply to LAC remains unknown. Methods We pooled LAC cohorts. We estimated RRs per unit of exposure to body mass index (BMI), systolic blood pressure (SBP), fasting plasma glucose (FPG), total cholesterol (TC) and non-HDL cholesterol on fatal (31 cohorts, n=168,287) and non-fatal (13 cohorts, n=27,554) cardiovascular diseases, adjusting for regression dilution bias. We used these RRs and national data on mean risk factor levels to estimate the number of cardiovascular deaths attributable to non-optimal levels of each risk factor. Results Our RRs for SBP, FPG and TC were like those observed in cohorts conducted in high-income countries; however, for BMI, our RRs were consistently smaller in people below 75 years of age. Across risk factors, we observed smaller RRs among older ages. Non-optimal SBP was responsible for the largest number of attributable cardiovascular deaths ranging from 38 per 100,000 women and 54 men in Peru, to 261 (Dominica, women) and 282 (Guyana, men). For non-HDL cholesterol, the lowest attributable rate was for women in Peru (21) and men in Guatemala (25), and the largest in men (158) and women (142) from Guyana. Interpretation RRs for BMI from studies conducted in high-income countries may overestimate disease burden metrics in LAC; conversely, RRs for SBP, FPG and TC from LAC cohorts are similar to those estimated from cohorts in high-income countries. Funding Wellcome Trust (214185/Z/18/Z)


Introduction
Cardiovascular diseases are the leading causes of death globally [1] , and the main causes of these deaths are a set of wellknown cardio-metabolic risk factors such as high blood pressure, overweight/obesity, diabetes and dyslipidemias [2][3][4] . Supporting evidence on the impact of cardio-metabolic risk factors on cardiovascular diseases has mostly come from cohort pooling collaborations [5] , including the Asia-Pacific Cohort Studies Collaboration [ 6 , 7 ], the Prospective Studies Collaboration [8][9][10] , and the Emerging Risk Factors Collaboration [11][12][13] . The results of these collaborations have been used to attribute disease burden to risk factors globally, providing inputs for surveillance and monitoring of cardio-metabolic risk factors and diseases.
These collaborations have little representation from Latin America and the Caribbean (LAC) [14] , a vast region with unique characteristics in terms of non-communicable diseases such as diabetes and raised blood pressure [15] , paired with the fastest rate of transition towards a predominance of urban areas in the developing world [ 16 , 17 ]. Therefore, the findings of these non-LAC collaborations, such as the age-specific relative risks used in global burden of disease estimations, may not apply to LAC countries. In fact, evidence suggests that the association between some risk factors and cardio-metabolic outcomes may be stronger in LAC compared with other world regions [ 4 , 18 ], possibly due to different levels of access to health care [ 19 , 20 ], differences in the distribution of cardio-metabolic risk factors [ 15 , 21-23 ], or incidence of noncommunicable diseases [24][25][26] . We identified and pooled prospective cohort studies in LAC to examine the effect of major cardiometabolic risk factors on cardiovascular outcomes, and to estimate age-specific relative risks for this world region.

Methods
Details about the Cohorts Consortium of Latin America and the Caribbean (CC-LAC) have been reported detailed elsewhere [27] . We analysed cohort data pooled and harmonized by the CC-LAC [27] , a network of health researchers and practitioners in LAC. We have harmonised and pooled approximately population-based cohort data on cardio-metabolic risk factors and outcomes, i.e., participants were not recruited based on cardiovascular diseases (CVD) (e.g., cohort of stroke survivors) or risk factor (e.g., cohort of smokers) history only. Cohort studies were identified through a systematic search and networks of researchers in LAC. We identified 78 approximately population-based cohorts (i.e., did not select participants on the basis of having previous disease) and excluded 31 cohorts that recruited only young participants (e.g., birth cohorts), did not measure exposures/outcomes of interests, or could not be accessed by the original investigators [27] . We accessed 33 cohorts from 13 countries (37% of LAC countries). From these 33 cohorts, 5 included participants who attended a specific health centre [28][29][30] or were members of a professional organization such as The Mexican Teachers' Cohort [31] and the Health Workers Cohort Study [32] . The other cohorts sampled individuals from the general population. Individual-level data from each cohort were received by the CC-LAC and were subsequently harmonised and pooled for the present analyses [27] .
From the pooled 33 cohorts, we excluded two that did not ascertain either fatal or non-fatal cardiovascular events; we further excluded 18 cohorts that did not ascertain non-fatal cardiovascular events from analysis of fatal-and non-fatal outcomes. Thus, our estimates for fatal cardiovascular events were informed by 31 cohorts with a mean follow-up of 8.8 years (standard deviation = 3.1), while our estimates for fatal and non-fatal cardiovascular events were informed by 13 cohorts with a mean follow-up of 8.5 years (standard deviation = 5.3).
We estimated incidence rate ratios which we will hereafter refer to as relative risks (RRs) and 95% Confidence Intervals (95% CI) for each selected cardio-metabolic risk factor on fatal and non-fatal cardiovascular diseases. We used these RRs to estimate the proportion and number of deaths attributable to each risk factor.

Cardiovascular outcomes
We analysed two outcomes separately: i) fatal and non-fatal cardiovascular events and ii) fatal cardiovascular events. Non-fatal cardiovascular events were not analysed alone because of the small number of events in many age groups. Cardiovascular events were identified using data from vital registration systems, clinical records or verbal autopsies, and where relevant adjudicated by each cohort (Supplementary Table 1).

Cardio-metabolic risk factors
The risk factors of interest were systolic blood pressure (SBP, in mmHg), body mass index (BMI, in kg/m 2 ), fasting plasma glucose (FPG, in mmol/L), total serum cholesterol (in mmol/L) and nonhigh-density lipoprotein cholesterol (non-HDL, in mmol/L). These variables were collected following standardised protocols in each cohort. We only examined SBP as the relationship between SBP and CVD outcomes is stronger than that of diastolic blood pressure [ 8 , 33 ].

Handling of missing data
BMI was missing in 12% of the pooled observations, while this number for SBP, total cholesterol, FPG and non-HDL cholesterol ranged from 66% to 80% (Supplementary Table 2). We used multiple imputation and fitted the regression models in each of 50 imputed datasets, pooling the estimates following Rubin's rules [34] . In sensitivity analyses, we used a complete-case dataset, and we observed minor differences for non-HDL cholesterol but overall the results were unchanged using multiple imputation (Supplementary Figure 1). Detailed methods on multiple imputation are available in Supplementary Material (pp. [7][8]. The main results herein presented are based on the multiple imputation data.

Adjusting for Regression Dilution Bias (RDB)
As the selected risk factors have a natural variability during follow-up, the estimated associations between baseline one-off measures underestimate the effect of "usual" exposure. This is often referred to as regression dilution bias [ 35 , 36 ]. To adjust for this bias, we used data from 10 of our cohorts with repeated risk factor measurements and used standard analytic methods (MacMahon method) [ 35 , 36 ] to calculate correction factors. The estimated correction factors were: 1.10 for BMI, 1.50 for SBP, 1.53 for FPG, 1.75 for total cholesterol, and 1.85 for non-HDL cholesterol. Further details about the RDB methods are available in the expanded methods (Supplementary Material pp. 6-7).

Survival analysis
For each cardio-metabolic risk factor, we fitted a Poisson linear mixed effects regression model for each age group separately (35-44, 45-54, 55-64, 65-74, 75-84, 85 + years) in which the independent variable was the risk factor, adjusted for sex and age at risk (i.e., at follow-up/event) within each age-group. A cohortspecific random intercept was included as well as the natural logarithm of the follow-up time as an offset. The coefficient of each risk factor from this model represents the log-incidence rate ratio for one-unit increase in the risk factor in each 10-year age group. We applied the RDB correction factor to these coefficients (in the log scale). For further details, refer to the extended methods (Supplementary Material pp. [5][6][7][8][9][10][11]. To better understand any potential differences in RRs between sub-regions in LAC, we computed RDB-adjusted RRs for fatal cardiovascular diseases for Central America & the Caribbean (Costa Rica, Cuba, Dominican Republic, Puerto Rico and Trinidad & Tobago) versus South America (Argentina, Brazil, Chile, Colombia, Peru, Uruguay and Venezuela). We included Mexico in the former group to preserve geographic proximity. To compare our results with a previous analysis of cohort data pooling studies that reported RRs for fatal ischaemic heart disease and stroke subtypes separately [5] , we weighted their estimated RRs by the relative prevalence of these outcomes in LAC.

Quantifying the population-level impact of risk factors
Following a comparative risk assessment approach [37] , we estimated the population attributable fraction (PAF) for each risk factor on cardiovascular deaths in 35 countries of the region comparing the current mortality burden to the one that would have been observed if the mean levels in the population were optimal (Supplementary Material p. 9). The optimal levels in the population were derived from previous analyses of global burden of disease [5] . Current mean levels of BMI, SBP, total cholesterol and non-HDL cholesterol for each country and by five-year age group and sex, were extracted from the NCD-RisC ( http://ncdrisc.org/ ) [ 15 , 22 , 23 ]. This information was not available for FPG, therefore, we did not include non-optimal glucose in these analyses. The number of cardiovascular deaths for the year 2019 was extracted from the estimates provided in the Global Burden of Disease (GBD) Study [38] . We used the RRs herein estimated (10-year age groups), and we interpolated them into 5-year age groups (Supplementary Figure  2 and Supplementary Table 7) [5] . We used the same age-specific RRs for men and women across countries in LAC, as we found no evidence of different RRs by sex. We estimated crude attributable death rates per 10 0,0 0 0 person-years by multiplying PAFs by total CVD deaths and dividing by the adult population of each country, which was also extracted from the GBD Study. Further details about the comparative risk assessment are available in the expanded methods (Supplementary Material pp. [8][9]. Countries for which we made estimates are those in common between the GBD Study and the NCD-RisC (35 countries and territories in LAC).

Risk of bias in each study
We evaluated three sources of bias. First, selection bias due to enrolment of participants. The risk of selection bias in these cohorts is rather small because inclusion in the study is unlikely to be simultaneously related to exposure and outcome. Second, measurement bias. As explained above and in Supplementary Material p. 04, major variables of interest were measured except for BMI which was self-reported in one cohort [ 31 , 39 ]. These variables are commonly measured in cardiovascular cohorts and were measured following standard procedures. Regarding the outcomes, we did not pool cohorts in which this information was not verified using links to vital registration data or adjudication (Flow Diagram on Supplementary Material p. 04). Third, confounding. We adjusted for age, sex and cohort in all analyses as we were interested in comparing the magnitude of our RRs with those of other global pooling studies which used the same set of potential confounders [5] .

Role of the funding source
The funder of the study had no role in study design, data collation, analysis, interpretation, or writing of the report. RMC-L and GD had full access to the data in the study. RMC-L and GD had final responsibility for the decision to submit for publication.

Results
In the group of cohorts analysed for fatal outcomes, women were younger than men (46.1 vs 55.7 years), while in the second set of cohorts the age was more alike (52.7 vs 52.1 years). In both sets of cohorts (i.e., analysis for fatal as well as for fatal and non-fatal outcomes), women had higher BMI than men (27.4 kg/m 2 vs 26.2 kg/m 2 and 28.6 kg/m 2 vs 26.4 kg/m 2 ); mean total cholesterol was also higher in women (5.4 mmol/L vs 5.2 mmol/L and 5.3 mmol/L vs 5.2 mmol/L). The average non-HDL cholesterol was slightly higher among men than women in both sets of cohorts (4.1 mmol/L vs 4.2 mmol/L and 4.1 mmol/L vs 4.2 mmol/L). In both sets of cohorts, mean SBP was higher among men than women (131 mmHg vs 134 mmHg and 128 mmHg vs 133 mmHg; Supplementary Table 3B).
In regional sub-group analyses, we did not observe substantial differences in the magnitude of the RRs for fatal outcomes. In both sub-regions, we could not observe a clear age gradient with larger RRs in younger groups; except for SBP where there was an age gradient from age 45 years ( Figure 2 , Supplementary Table 6).
For fatal and non-fatal outcomes, the 13 cohorts contributed with 27,554 eligible individuals. Almost two thirds were men (64.1%), and the mean age was 52.3 (10.5) Figure 2. Age-specific relative risks for fatal cardiovascular disease associated with usual levels of selected cardio-metabolic risk factors by sub-regions. All models were adjusted by sex and age (within each age group). Age groups based on age at risk (i.e. at outcome). RR: relative risk; 95% CI: 95% confidence interval; BMI: body mass index; SBP: systolic blood pressure. Only results adjusted for regression dilution bias are presented. Results as per multiple imputation. Insufficient observations to reliably compute these risk estimates for fatal plus non-fatal cardiovascular events. RR: relative risk; 95% CI: 95% confidence interval.
The age-specific RRs for fatal and non-fatal CVD for SBP and FPG were remarkably similar to those reported from cohorts mostly conducted in high-income countries ( Figure 3 ) [5] . For TC, our RRs appeared to be smaller for the two youngest age groups, though these differences were statistically insignificant ( Figure 3 ). For BMI, the RRs were consistently smaller in magnitude for participants younger than 75 years old, with the largest difference for those in the age group 55-64 years ( Figure 3 ): 1.24 (95% CI: 1.11-1.38) vs. 1.50 (95% CI: 1.41-1.61).
Non-optimal SBP was responsible for the largest proportion of attributable cardiovascular deaths across countries, with a proportional effect ranging from 30.7% among Cuban women to 58.0% among men from Grenada. The second largest proportion for attributable CVD mortality was due to non-optimal non-HDL cholesterol, which proportional effect varied between 13.9% (Chile, women) and 31.2% (Guyana, women). The proportional effect of non-optimal BMI and total cholesterol were smaller. For BMI the proportional effect ranged from 6.1% in women from Cuba to 19    these numbers were 4.4% (Guatemala, men) and 18.1% (Guyana, women; Supplementary Figure 3). The proportional effect of non-optimal total cholesterol tended to be larger among women than among men, as was observed in 34 countries ( Figure 4 ); the largest absolute difference between women and men was observed in Guatemala (11.6% in women vs 4.4% in men; Supplementary Figure 3). On the other hand, the proportional effect of non-optimal SBP was higher among men than among women in most countries ( Figure 4 ), with the largest difference in Uruguay (47.5% in men vs 35.4% in women; Supplementary Figure 3).
Over half a million deaths (502,913 (95% credible interval = 340,637-653,242) out of a total of 1,094,795 CVD deaths in LAC) were attributable to non-optimal SBP in 2019. The second largest effect was estimated for non-optimal non-HDL cholesterol at 224,118 (95% credible interval = 83,755-388,176) deaths and the lowest for non-optimal BMI at 119,498 (95% credible interval = 61,201-200,824) ( Table 1 ). Brazil, Mexico and Argentina, in that order, had the largest numbers of attributable deaths across the four risk factors for both women and men; Colombia displaced Argentina from the third place regarding non-HDL cholesterol in women (Supplementary Figure 3).
Most of the cardiovascular disease deaths attributable to nonoptimal BMI were premature ( < 70 years of age; Table 1 ), ranging from 72% among women in Southern and Tropical Latin America, to 83% among men in the Caribbean ( Table 1 ). On the other extreme, the 44% of all cardiovascular disease deaths attributable to non-optimal SBP was premature; this ranged from 31% (women in Andean Latin America) to 54% (men in Southern and Tropical Latin America; Table 1 ).
Across the four sub-regions, the crude attributable death rate per 10 0,0 0 0 person-years due to non-optimal risk factors was consistently larger among men for BMI, SBP and non-HDL cholesterol ( Figure 5 ). Consistently across all sub-regions, non-optimal SBP was responsible for the largest number of attributable deaths for both men and women. Similarly, non-optimal non-HDL cholesterol always ranked second ( Figure 5 ).
For men, country-specific attributable death rates to nonoptimal SBP ( Figure 6 A, Supplementary Figure 3), was the smallest in Peru (54 per 10 0,0 0 0 person-years) and Guatemala (67), and the largest in Dominica (249) and Guyana (282). For non-optimal non-HDL cholesterol, we observed the smallest attributable death rates in Guatemala (25) and Peru (27), and the largest in Dominican Republic (107) and Guyana (158). Finally, for BMI, the lowest attributable death rates were observed in Peru (13) and Guatemala (14), and the largest in Saint Lucia (64) and Saint Kitts and Nevis (75).
For women ( Figure 6 B, Supplementary Figure 3), the smallest attributable death rate to non-optimal SBP was estimated in Peru (38 per 10 0,0 0 0 person-years) and Guatemala (64), while Guyana (228) and Dominica (261) had the largest rates. For non-HDL cholesterol, the lowest attributable rates were observed in Peru (21) and Chile (30), while the largest rates were in Haiti (109) and Guyana (142). For non-optimal BMI, the lowest attributable death rates were observed in Peru (10) and Panama (14), whilst the largest rates were in Saint Kitts and Nevis (51) and Guyana (76).

Discussion
Benefiting from a unique database of pooled individual-level data from 31 cohort studies in 13 LAC countries [27] , we estimated age-specific RRs for major cardiovascular disease risk factors. We observed smaller RRs for BMI compared with those used in the global estimates of disease burden based on cohort collaborations originating mostly in high-income countries. We observed an age gradient whereby young people had higher RRs than older indi- BMI: body mass index; SBP: systolic blood pressure; TC: total cholesterol; Non-HDL: non-HDL cholesterol; CI: 95% credible interval. All ages included observations aged ≥20 years, whereas premature refers to ages between 20 and 69 years. The premature-to-all-ages ratio quantifies the ratio of the estimated attributable deaths below age 70 to the corresponding estimate for all ages expressed as a percentage. viduals. Our results suggested that the RRs did not differ between Central America & the Caribbean sub-region compared with South America. The largest attributable CVD deaths across the selected risk factors were due to non-optimal SBP, followed by non-HDL cholesterol. These risk factors had a much larger impact on cardiovascular deaths in the Caribbean and Southern and Tropical subregions.

BMI
The age gradient of the estimated RRs in our analysis is consistent with prior pooled analysis of large cohort studies [ 2 , 5 , 9 , 10 , 13 , 40 ]. The magnitude of age-specific RRs was similar in our analyses compared with prior pooling projects of cohorts in high-income countries and those of the Asia-Pacific region [5] ; however, for BMI, our estimated RRs were smaller for many age groups below the age of 75, particularly for people aged 55-64. The reported RRs for BMI from the Prospective Studies Collaboration, and the Asia Pacific Cohort Studies Collaboration did not account for RDB [ 10 , 41 ]. Adjusting for RDB, would have led to even higher RRs compared with ours. Such similarity may reflect the same underlying biology of these risk factors and lack of major modifications by lifestyle or environmental risk factors that do differ across regions. In fact, where patterns or lengths of exposure matter as it is the case for smoking and alcohol use, RRs of cardiovascular outcomes differ substantially by region [25] . In contrast, the observed differences in RRs for BMI may be explained by the shorter duration of the weight gain in the LAC region compared with highincome countries. That is, high-income populations have been exposed to non-optimal BMI levels longer than most populations in LAC [22] , and are therefore experiencing the larger cumulative harmful effects of BMI on cardiovascular health. Alternatively, the same level of BMI may correspond to a healthier body fat distribution in LAC compared with high-income populations. The current evidence on such a difference in fat distribution at the same level of BMI is mixed [ 42 , 43 ] and further research is needed using larger population-based surveys with measurements of body composition in LAC. Our RRs for non-HDL cholesterol are consistent with a recent analysis of the PURE study, which did not find substantial differences in RRs for non-HDL cholesterol between high-, middleand low-income countries [4] .
The observed differences in RRs for BMI may explain the differences in our estimates of attributable deaths to cardio-metabolic risk factors in LAC versus those reported by the GBD Study, which uses RRs mostly informed by epidemiological studies in highincome countries. For example, we estimated a crude attributable death rate for non-optimal BMI in women in Peru of 10 per 10 0,0 0 0, compared with 18 cardiovascular disease deaths reported by the GBD Study [44] for Guyana we estimated 76 compared to 86 [44] . Notably, the GBD Study risk estimates include other cardiovascular outcomes besides those herein analysed-partly explaining the differences.
Our results show that non-optimal SBP was responsible for the largest number of cardiovascular disease deaths, followed by non-HDL cholesterol, total cholesterol and BMI. This ranking is similar to the one proposed by the GBD Study in 2019, in which SBP ranked first, followed by LDL-Cholesterol, fasting plasma glucose and BMI [45] . This suggests that the ranking based on global risk estimates still apply to LAC, yet the burden attributable to each risk factor may be different. That difference, as herein proposed, may be overestimating the cardiovascular disease mortality attributable to non-optimal BMI in LAC.
Arguably, LAC-based risk estimates -particularly for BMI-provide more valid metrics for countries in LAC to quantify the burden of key cardio-metabolic risk factors. This evidence could allow prioritizing the risk factor(s) with the largest burden, develop policies and interventions to address these priorities, and set up surveillance systems to monitor the progress towards international and local goals. Our results could be taken as parameters upon which goals can be set to reduce cardiovascular burden in LAC and in each country in the region given that metrics to measure the progress and surveillance of cardiovascular diseases were mostly informed by countries outside LAC. Considering the sharp rise in obesity and diabetes in the region [15] , despite our evidence that shows lower RRs compared with high-income countries, overweight/obesity remains one of the highest-ranking risk factors for CVD; obesity control and prevention policies should continue to remain top priorities.
Our work has several strengths. The risk estimates are agespecific and were computed following consistent methods using the largest pooled database of cohorts in LAC. We analysed data from 13 countries including at least one from each sub-region in LAC, a work never conducted before. Analysing individual level data, in contrast to published estimates [ 18 , 46 ], allowed us to examine interactions between different variables. The RRs were adjusted for regression dilution bias using LAC data providing the RR of "usual" exposure to risk factors. We used multiple imputation to handle missing data for risk factors at baseline. Nevertheless, we acknowledge several limitations. Due to data availability, we could not study other risk factors such as LDL-cholesterol. Likewise, some outcomes were not available, preventing us from disentangling, for example, ischaemic from haemorrhagic stroke. We were also unable to examine RRs in all sub-regions (e.g., Andean Latin America and southern Latin America) due to the small numbers of events. We therefore only explored risk estimates from South America with those from Central America and the Caribbean, and even in this case, confidence intervals were wide, particularly in the youngest and oldest age groups. Many cohorts did not collect data on non-fatal events (possibly due to the younger age of participants or complexities and costs of identifying and adjudicating non-fatal events), precluding a separate analysis. The limited number of non-fatal events could have also affected the main results (RRs for both fatal and non-fatal CVD), as these estimates could have been mostly driven by fatal events; however, results for fatal outcomes only showed the same age pattern and the RRs had a similar magnitude as those including both fatal and nonfatal events. Mortality risk may also be confounded by health care access and control of non-communicable diseases, variables that were not included in the regression models. A few variables had a large proportion of missing values across cohorts mostly because a subset of cohorts did not include these measurements in their protocol, as opposed to non-response or missing measurements within each cohort. We used modelled estimates of CVD deaths by country, age and sex from the GBD 2019 Study to calculate the attributable number of deaths which makes our results comparable and consistent across countries [38] . However, the estimated CVD mortality may be biased in countries especially if local data is not incorporated in the GBD analyses and/or if modelling assumptions are not valid for a particular region/sub-region. Also in relation to the GBD Study, we acknowledge that GBD deliver estimates for several years whereas we only used their most recent estimates (2019); we focused on the most recent year because we aimed to provide estimates to inform policies and goal setting, rather than showing time patterns. Cohorts herein analysed for fatal outcomes included more women than men; interpretation of these estimates should be made in light of this profile. We only presented results at the country level. Future work should also study cardiovascular disease burden at the subnational level, ideally in all countries in LAC considering its substantial geographical and socioeconomic diversity. We encourage researchers in LAC to use the risk estimates herein reported to conduct subnational analysis of cardiovascular disease burden. We pooled multiple cohorts which included a random sample of the general population or studied specific groups (e.g., The Mexican Teachers' Cohort). We studied cardio-metabolic risk factors (e.g., blood pressure and total cholesterol) which were collected following objective, standard and comparable methods between cohorts. The risk of selection bias is quite low because the probability of being selected in these studies is unlikely to be simultaneously related to the exposure and outcome. Regarding BMI, except for one cohort we used measured weight and height which reduces measurement error; this is method is consistent with other cohort pooling projects.
In conclusion, using data from the first pooling project of cohort studies in LAC we found that RRs of cardiovascular disease per unit increase in blood pressure, glucose and cholesterol are remarkably similar to previous pooling projects that used data mostly from high-income countries. In contrast, we observed smaller agespecific RRs for BMI. The estimated RRs offer region-specific evidence that can be used to update estimates of attributable burden of disease to better inform regional policies and goals. One of the strategic lines of action in Pan American Health Organization's Plan of Action for the Prevention and Control of Non-communicable Diseases in the Americas 2013-2019, was to strengthen country capacity for surveillance on non-communicable diseases and their risk factors [47] . Our results can help improve the validity of such surveillance effort s by emphasizing the use of local data and evidence in prioritizing and implementing CVD prevention programs.

Contributions
GD, RMCL and ME conceived the study. RMCL curated the data and conducted all analysis with input from GD and ME. All members of the CC-LAC Steering committee contributed to the design of the analysis and interpretation of the results and conclusions. RMCL wrote the first draft of the manuscript and all co-authors contributed to the revisions. GD and RMCL have access to and have verified the underlying data.

Data sharing
Data is currently only available to CC-LAC collaborators. Expressions of interest to access the CC-LAC data are welcomed and will be handled by the CC-LAC steering committee.