Study design

Information from the Peruvian Demographic Health Survey (DHS or ENDES in Spanish) was used. The survey is annually conducted by the National Institute of Statistics and Informatics (INEI in Spanish), and collects information on variables including poverty, fertility, violence, and health. For this analysis, data from the DHS 2014 to 2020 were pooled and analyzed.

Study population and selection criteria

The Peruvian DHS utilized a bietapic sampling approach. In rural areas, the primary sampling units comprise clusters of 500 to 2,000 subjects, whereas the secondary sampling units are households within each of these clusters. However, in urban areas, the approach is a little different: the primary sampling units are blocks or groups of blocks with more than 2,000 subjects within an average of 140 households, whilst secondary sampling units comprise households as in rural areas [16].

The sampling process and techniques used are described in the technical documents of the DHS [17]. The Peruvian DHS is conducted every year; however, the sample frame used for participants’ selection is updated every three years (i.e., 2012–2014, 2015–2017, 2018–2020). The method used for sampling is known as the method of the cube, which allow obtaining samples with estimates of totals approximately equal to the characteristics of the population of the survey that replicates the population structure within the selected sample considering the groups of age, sex and other balance variables.

The present analysis included subjects aged between 40 and 74 years as prediction equations have been validated in this age range [18], with complete information to estimate the 10-year cardiovascular risk (i.e., age, sex, current smoking status, and body mass index), as well as urbanization and altitude information.

Outcome

The outcome of interest was the 10-year predicted absolute CV risk. Since the Peruvian DHS did not collect blood samples for fasting glucose or total cholesterol, we decided to utilize the non-laboratory version of the World Health Organization (WHO) 10-year risk of cardiovascular disease (CVD) [19]. Such risk score used data from the Emerging Risk Factors Collaboration and recalibrates the predicted risk scores to reflect the expected 10-year risk in contemporary population in 21 global regions, including sex, age (in years), smoking status and body mass index (kg/m²). CV risk was estimated utilizing the whocvdrisk command in STATA. For analysis purposes, the CV risk score was described according to standard thresholds (<5%, between 5% to <10%, between 10% to <20%, and ≥20%), and then categorized using a cutoff of 10% as usually individuals with such CV risk level or above should receive an evaluation using laboratory-based charts [18].

Procedures

All the questionnaires and tools were applied by trained staff using tablets (LENOVO B600F or B600H) using a special software developed by INEI [20]. In addition, GPS data collection was also available with the software used.

Anthropometrical procedures (height, weight and blood pressure) were collected by trained personnel [23]. Height was measured using a standard stadiometer, whereas weight was measured utilizing an electronic scale (SECA model 872, 874 and 878). Blood pressure levels were collected using a digital monitor (OMRON, model HEM-713). Two types of cuffs were utilized depending on the arm circumference of participants. Blood pressure measurements were assessed twice with the participant sitting and their right arm resting on a flat surface at the heart level. The first measurement was taken after a resting period of at least five minutes, and the second measurement was done two minutes after the first measurement [23].

Statistical analysis

Statistical analyses were conducted using STATA 16 for Windows (StataCorp, College Station, TX, USA). All the analyses took into account the multistage complex design of the sample, and subpopulation (subpop command) analyses were also pursued [24].

Initially, a description of the study population, including a comparison between those included and not included in the analyses, was conducted. Means and standard deviation (SD) and absolute and relative frequencies were utilized to describe numerical and categorical variables, respectively. In addition, the prevalence of variables of interest were reported as well as their respective 95% confidence intervals (95% CI).

Then, the description of the study population by the exposures (i.e., urbanization and altitude) and the outcome (i.e., absolute CV risk) was also conducted. Comparisons were carried out utilizing the Chi-squared test corrected for the survey design with the second-order correction of Rao and Scott [25].

To assess the associations of interest, crude and adjusted Poisson regression models were built. Multivariable models were adjusted by age, sex, education, wealth index, and survey year. In addition, urbanization and altitude were evaluated in the same regression model to examine if both variables were independently associated with CV risk. Finally, urbanization was evaluated as potential effect modifier of the association between altitude and CVD risk using the testparm command in STATA. Prevalence ratios and 95% CI were reported. A p-value <0.05 was considered statistically significant.

Characteristics of the study population

Data of 86,206 participants fulfilled the inclusion criteria and were pooled using information from ENDES 2014 to 2020. Of them, 143 records were excluded because pregnancy and 5,654 records because of lack of data of interest (i.e., there was not information about altitude, urbanization or variables associated with estimation of the CV risk). Comparison of study characteristics between individuals included and excluded in the analyses are available in Online Supplement E-Table 1. As a result, the information of 80,409 subjects, mean age 54.3 (SD: 8.8), 42,640 (54.4%) females, and 18,098 (29.1%) with 12+ years of education, were further analyzed.

Urbanization and altitude

A total of 30,722 (25.4%; 95% CI: 24.7% – 26.2%) individuals were from rural areas. Age (p < 0.001), sex (p < 0.001), education (p < 0.001), wealth index (p < 0.001), alcohol use (p < 0.001), hypertension (p < 0.001), and survey year (p < 0.001) were associated with urbanization in bivariable analyses (Table 1).

Regarding altitude, 60.6% (95% CI: 59.4% – 61.8%) of subjects were living at <500 m.a.s.l., whereas only 9.3% (95% CI: 8.7% – 10.0%) were living at ≥3500 m.a.s.l. All the variables, except survey year, were associated with altitude in bivariable model (Table 2).

Absolute cardiovascular risk

The average 10-year predicted absolute CV risk was 4.5% (SD: 3.1), with values ranging from 0.5% to 37.4%. In addition, 56,333 (66.5%; 95% CI: 66.0% – 67.1%) had a cardiovascular risk <5%; 18,885 (25.7%; 95% CI: 25.1% – 26.2%) had a risk between 5% and <10%; 5,012 (7.5%; 95% CI: 7.2% – 7.9%) had a risk between 10% and <20%; and 179 (0.3%; 95% CI: 0.2% – 0.4%) had a risk ≥20%. Age (p < 0.001), sex (p < 0.001), education (p < 0.001), wealth index (p < 0.001), current smoking (p < 0.001), and hypertension (p < 0.001) were factors associated with an absolute CV risk ≥10% in bivariable analysis (Table 3).

Urbanization, altitude and absolute cardiovascular risk

After controlling for age, sex, education, wealth index, and survey year, higher altitude was associated with lower absolute cardiovascular risk (Table 4). Thus, the proportion of subjects with CV risk ≥10% was lower among those living between 2,500 – 3,490 m.a.s.l. (PR = 0.81; 95% CI: 0.74 – 0.88) compared to those living <500 m.a.s.l., and such proportion was much lower among those living at ≥3,500 m.a.s.l (PR = 0.66; 95% CI: 0.59 – 0.73). Additionally, in multivariable model, the proportion of subjects with CV risk ≥10% living in rural area was lower (PR = 0.82; 95% CI: 0.75 – 0.89) compared to those living in urban areas. When urbanization and altitude were included in the same regression model, both variables were independently associated with CV risk (Table 4).

Urbanization was an effect modifier of the association between altitude and CV risk (p < 0.001). Thus, the protective effect of altitude on CV risk was seen ≥2500 m.a.s.l among those living in urban areas, whereas such association was observable only among those living ≥3500 m.a.s.l. in rural settings (See Table 5).

Main results

Our study reports evidence that both, urbanization, specifically rurality, and altitude, mainly settings at ≥2,500 m.a.s.l., had a lower proportion of individuals having a CV risk ≥10% in multivariable models. Moreover, the association of urbanization with CV risk was independent of the effect of altitude, and vice versa. In addition to that, the association between altitude and CV risk is seen at ≥2500 m.a.s.l. in urban settings and ≥3500 m.a.s.l. in rural ones, highlighting that altitude may be the main factor associated with lower CV risk.

Implications of the findings

While 50% of the world’s population lives below 150 m.a.s.l, only 6% de the world’s population lives over 1,500 m.a.s.l. [26, 27], and most of these people are from low- and middle-income countries (LMIC). Thus, disentangling the potential effect of urbanization and altitude on CV risk seems to be relevant, especially for resource-constrained settings where the burden of CV is high and risk-based prevention much needed. As many of the rural areas are at high altitude, our findings help to better understand the epidemiology of absolute CV risk in these populations. Thus, our results highlight that rurality and high altitude (≥2,500 m.a.s.l.) are negatively and independently associated with high CV risk; however, the potential effect of high-altitude on CV risk is seen at lower altitude among urban dwellers.

Contrary to the usual paradigm that altitude and rurality are associated with a low prevalence and incidence of cardiovascular risk factors (i.e., smoking, obesity, hypertension, and type 2 diabetes), current literature suggest that the distribution of these risk factors in rural and high-altitude settings could have changed. Thus, a recent analysis reported that ≥55% of the global rise in BMI in the last three decades, and ≥80% in some low- and middle-income regions, was attributed to increase of BMI in rural areas [28]. This has been also confirmed in Peruvian less urbanized settings [29]. Whereas cross-sectional studies have reported low prevalence of risk factors in rural settings, a longitudinal analysis in Peru reported a higher risk of developing hypertension in rural areas relative to rural-urban migrant or urban groups [30]. In addition to that, a dramatic change in lifestyle, whereby unhealthy lifestyles are more common, has taken place in Peruvian rural and high-altitude populations. Although urbanization is usually associated to increased consumption of fat, sugar and salt, the results of a study comparing diet patterns in urban and rural settings in a high-altitude Peruvian region reported mixed findings with more added salt and oil consumption than urban areas [31]. Despite of this, according to our findings, subjects with CV risk ≥10% are still less common in rural and high-altitude settings in Peru. A potential explanation of these findings may be the duration of the exposure to these cardiometabolic risk factors; in other words, these rural and high-altitude populations may have been exposed to these risk factors not long enough to have experienced the deleterious effects on cardiovascular health.

Beyond traditional risk factors, altitude may also have a direct effect on cardiovascular physiology. Thus, with chronic altitude exposure, there is a decrease in maximum heart rate, with a decrease in heart rate change during exercise [32]. Similarly, altitude exposure is associated with lower levels of systolic blood pressure [33]. In addition, ultraviolet radiation from sunlight, which tend to increase with altitude, may have a substantial impact on CVD risk [34]. This effect may be mediated by vitamin D levels and by lower blood pressure [35]. Thus, there is a need to understand mechanisms behind the association between altitude and CVD risk.

Public health relevance

For this manuscript, we used one of the six modules of the HEARTS technical package developed by the World Health Organization (WHO) [18]. The module is focused on risk-based CVD management and includes information on a total risk approach to the assessment of CVD, including subregion-specific risk charts [19].

Despite the limitation of risk scores we used, our findings highlight the need to target individuals at higher risk of CVD, especially those located in urban areas and living <2500 m.a.s.l., but without excluding those from rural or high-altitude settings. Targeting individuals at high CV risk has been recognized as a beneficial strategy in terms of number of CV events avoided [36, 37]. Providing CVD risk score may reduce total cholesterol, systolic blood pressure, and multivariable CVD risk [37]. We decided to use an absolute CV risk approach as a subject with multiple risk factors, even if each factor is only slightly or moderately elevated, may be at a higher risk of CVD than someone with an elevated level of a single risk factor.

Finally, whereas a risk stratification approach, as the used by the WHO HEARTS risk score, is mainly suitable to settings with limited resources, our results suggest the need of ensuring laboratory infrastructure to confirm results and conduct follow-up of individuals, appropriate drug therapy availability for primary prevention of CVD, and counseling strategies in different settings, especially in rural areas.

Strength and limitations

This study benefits from pooled data derived from consecutive nationally representative surveys in Peru. Our study expands to disentangle relationships between urbanization, altitude and predicted CV risk. However, our analysis has some limitations that merit discussion. First, the cross-sectional nature of the data only permits to evaluate association and not causality. Second, across the years, the sampling frame of the Peruvian DHS has changed; thus, the sampling frame of the last three years was not similar to that of the previous years. However, we have used the year-specific sampling weights, stratum, and sampling units. Because of the sampling frame, although there is a possibility to have some bias because of sampling, this is relatively negligible. Moreover, although there is a small probability to select the same participant in different surveys (for example in 2014 and 2020), this probability is very reduced as different sampling frames are utilized. Third, as blood samples are not usually taken in these surveys (i.e., lipid profile and fasting glucose are not available), we only utilized the non-laboratory chart of the WHO absolute CV risk score. However, a moderate agreement between WHO CVD risk predictions using laboratory and non-laboratory algorithms has been described in population samples [18]. Finally, some residual confounding may arise as an exhaustive list of confounders was not used because of data availability.