|Year : 2021 | Volume
| Issue : 1 | Page : 15-22
Examining the prevalence of hypertension by urban–rural stratification: A Cross-sectional study of nepal demographic and health survey
Md Salauddin Khan1, Sabira Naznin1, Henry Ratul Halder2, Umama Khan3, Md Murad Hossain4, Tanjim Siddiquee1
1 Statistics Discipline, Khulna University, Khulna, Bangladesh
2 Statistics Discipline, Khulna University, Khulna, Bangladesh; Department of Statistics, Jahangirnagar University, Savar, Dhaka, Bangladesh
3 Biotechnology and Genetic Engineering Discipline, Khulna University, Khulna, Bangladesh
4 Department of Statistics, Bangabandhu Sheikh Mujibur Rahman Science and Technology University, Gopalganj, Bangladesh
|Date of Submission||30-Aug-2020|
|Date of Decision||28-Oct-2020|
|Date of Acceptance||27-Nov-2020|
|Date of Web Publication||9-Feb-2021|
Henry Ratul Halder
Department of Statistics Discipline, Khulna University, Khulna - 9208
Source of Support: None, Conflict of Interest: None
Introduction: Nepal has one of the highest prevalences of hypertension in South Asia, which also causes other cardiovascular diseases. However, no studies investigated the prevalence and risk factors of hypertension by urban-rural stratification. Methods: We used a machine learning, Boruta algorithm to select risk factors and a tenfold random forest classifier to evaluate their performance. Finally, multivariate logistic regression estimated crude and adjusted odds ratios with 95% confidence intervals for knowledge generation. Results: The study included 7825 participants (urban: 4939; rural: 2886), where rural participants were slightly older (median: 37 years; interquartile range: 26–53) and females were more hypertensive (urban: n = 606, 34.5%; rural: n = 308, 31.2%). The prevalence of hypertension was 35.6% in urban and 34.1% in rural regions. The odds of hypertension increased in rural regions for advancing age, provinces (province 4 and 5), and ecological zones (hill and terai). Overweight and obese participants were more likely to have hypertension in both regions. Conclusion: The study recommends the rigorous improvement of public health programs in rural regions of province 4 and 5, concentrating on Dalit and Janajati older males from hill and terai ecological zones. Overweight and obese people from both regions also need special focus. Finally, policymakers and government officials have to tailor campaigns differently for robust implementation of the essential health-care package and multisectoral action plans to prevent and control hypertension.
Keywords: Hypertension, machine learning, multivariate logistic regression, Nepal, urban-rural stratification
|How to cite this article:|
Khan MS, Naznin S, Halder HR, Khan U, Hossain MM, Siddiquee T. Examining the prevalence of hypertension by urban–rural stratification: A Cross-sectional study of nepal demographic and health survey. Asian J Soc Health Behav 2021;4:15-22
|How to cite this URL:|
Khan MS, Naznin S, Halder HR, Khan U, Hossain MM, Siddiquee T. Examining the prevalence of hypertension by urban–rural stratification: A Cross-sectional study of nepal demographic and health survey. Asian J Soc Health Behav [serial online] 2021 [cited 2022 Dec 8];4:15-22. Available from: http://www.healthandbehavior.com/text.asp?2021/4/1/15/308816
| Introduction|| |
A substantial percentage of disabilities and deaths occur globally due to cardiovascular diseases. Hypertension is one of the predominant risk factors for these diseases, disabilities, and deaths. Hypertension is also an alarming issue in numerous low- and middle-income countries (LMICs). A goal was set to reduce the prevalence of hypertension by one-fourth between 2013 and 2025. Yet, the estimated global prevalence was found significantly higher in 2015. The prevalence of this disease can be controlled for the achievement of the targeted goal by emphasizing the LMICs.
Behavioral risk factors (such as unhealthy eating habits, sedentary lifestyle, excessive alcohol, and tobacco consumption) contribute to high blood pressure (BP) and its related complications. Various sociodemographic characteristics are also responsible for this. Research is scarce on the associated risk factors for hypertension in many LMICs, including Nepal. Furthermore, the country is undergoing an epidemiological transition, which increases the burden of noncommunicable diseases (NCDs). The estimated prevalence of hypertension in Nepal was between 21% and 34%. Besides, research indicated that urban–rural stratification is a key factor for this higher prevalence rate., The cause behind this is the diverse behavioral patterns in urban and rural regions. For instance, urban residents are more inactive and consume calorie-dense food that may lead to a higher body mass index (BMI) and increases the risk of hypertension. The Nepal Demographic and Health Survey (NDHS) 2016 also reported inequalities between urban and rural regions for other health indicators. Prior studies only explored the socioeconomic and demographic characteristics of hypertension in Nepal., However, previous studies have not implemented a stratified setting to examine the prevalence and risk factors for hypertension. This approach will help policymakers and government officials to strengthen the policies and alleviate hypertension in specific places. It is also important to understand the variation in the distribution of risk factors that could contribute to the overall prevalence of this disease.
Therefore, the study aimed to stratify the prevalence of hypertension by urban–rural place of residence among adults (aged ≥35 years) in Nepal. We used the Boruta algorithm for the extraction of significant risk factors and performed multivariate logistic regression by each stratification for knowledge generation using the nationally representative survey of NDHS 2016. The final aim of the study is to provide policy recommendations and a multisectoral action plan to prevent and control hypertension.
| Methods|| |
We implemented a secondary investigation on a cross-sectional dataset of NDHS 2016. The survey estimated the prevalence of hypertension and diabetes among adults (aged ≥35 years), including other health indicators. The NDHS 2016 was designed to cover both urban and rural regions for the country's seven provinces including all ecological zones. The survey successfully collected data from 11,040 households with a 99% response rate. The biomarker questionnaire was used to record anthropometry, BP measurement, and hemoglobin test. The details of survey design, data collection method, questionnaires, sample size calculation, and results are published in the NDHS 2016 report.
Participants' BP was recorded using UA-767F/FAC (A&D Medical) BP monitor. The enumerators measured BP three times at an interval of five or more minutes. The average of the second and third measurements was used to report the final BP level. For BP measurements, we used the cutoff points of the National Institutes of Health guideline. Briefly, individuals were categorized as hypertensive if their systolic blood pressure (SBP) was ≥140 mmHg and/or if their diastolic blood pressure (DBP) was ≥90 mmHg. Individuals with SBP of 120–139 mmHg and/or DBP of 80–89 mmHg were categorized as prehypertensive. The normal category was formulated with SBP <120 mmHg and/or DBP <80 mmHg.
Risk factors selection
We performed the Boruta algorithm to extract risk factors for hypertension from the entire NDHS 2016 biomarker dataset. This is a machine learning algorithm around the random forest classifier (RF) to select risk factors based on the importance score. Nine risk factors were included in the study for further investigation. The selected risk factors were age (18–24, 25–34, 35–44, 45–54, and above 54), ecological zone (mountain, hill, and terai), province (province 1, province 2, province 3, province 4, province 5, province 6, and province 7), wealth index (poor, middle, and rich), alcohol consumption (no, yes), sex (male, female), marital status (single, married, and others), education level (no formal education, primary, and secondary or above), and BMI (normal, underweight, and overweight or obese). We recoded continuous variables (age and BMI) and transformed them into categorical variables for adjustment purposes. The categories for BMI (normal 18.5–25 kg/m2, underweight <18.5 kg/m2, and overweight or obese >25 kg/m2) were calculated using the World Health Organization classification system.
All the statistical analyses were conducted using R version 4.0.2 (Bell Laboratories, New Jersey, USA). The entire sample of NDHS 2016 was utilized in the study to identify the influencing risk factors using the Boruta algorithm. In the following step, we evaluated the performance of selected risk factors in each stratification using a tenfold RF classifier (training data = 70% and testing data = 30%) with various metrics (accuracy, sensitivity, and specificity). This classification technique is relatively quick and can be performed without tuning the hyperparameters. In the next stage, multivariate logistic regression with stratification was performed to obtain both the adjusted odds ratio (AOR) and the crude odds ratio (COR) with 95% confidence intervals (CIs). Besides, the variance inflation factor tested the multicollinearity among risk factors before adjusting for confounders. [Figure 1] briefly describes the complete analytical process of the study.
The New ERA conducted the NDHS 2016 under the aegis of the Ministry of Health of Nepal. Survey funding was supported by the United States Agency for International Development (USAID), while the Inner City Fund (ICF) International provided technical assistance through the Demographic and Health Survey (DHS) Program. The NDHS dataset was available from https://dhsprogram.com/data. Ethical approval is not required since the NDHS 2016 is a secondary dataset. However, Macro International obtained informed consent from individuals who participated in the DHS surveys.
| Results|| |
Performance evaluation of selected risk factors
[Table 1] exhibits the performance of selected risk factors for hypertension using a tenfold RF classifier (training data = 70%, and testing data = 30%), without tuning hyperparameters. The accuracy of the classifier was 91.84% and 92.05% for urban and rural regions, respectively. Moreover, sensitivity and specificity also reported a high percentage for accurate predictions. As >70% accuracy, sensitivity, and specificity scores indicate a better predictive model, we concluded that the Boruta algorithm retrieved the most informative risk factors from the dataset.
|Table 1: Performance evaluation of selected risk factors for hypertension using random forest classifier|
Click here to view
Background information of the participants
From [Table 2], 7825 participants (urban: 4939; rural: 2886) were included in the study. The median SBP and DBP in the two regions were almost similar. Around 35% of the participants had hypertension in both regions. Overall, participants from rural regions were slightly older (median: 37 years; interquartile range [IQR]: 26–53) than their urban counterparts (median: 35 years; IQR: 25–50). In urban regions, half of the participants were from terai (n = 2503; 50.7%) ecological zones. Province 1, province 2, province 5, and province 7 had more hypertensive participants in urban regions, whereas the result contrasted for province 3, province 4, and province 6. The number of alcoholic hypertensive participants was higher in rural areas (n = 18; 1.8%) in comparison to urban areas (n = 10; 0.6%). For overall participants, almost half of the rich people (n = 2290; 46.4%) were from urban regions and the gender distribution was nearly equal. However, there was a high frequency of hypertensive females in both urban (n = 606; 34.5%) and rural (n = 308; 31.2%) areas. The distribution of hypertension was higher in urban regions for people with single (n = 97; 5.5%) and others (n = 298; 16.9%) marital status. In contrast, married (n = 784; 79.4%) rural residents had a high frequency of hypertension. The total number of educated participants was more in urban areas. Finally, overweight or obese hypertensive participants were remarkably high in urban regions (n = 677; 38.5%) than their rural (n = 251; 25.5%) counterparts.
|Table 2: Characteristics of all study variables for hypertensive participants stratified by urban-rural stratification (n=7825)|
Click here to view
Prevalence of hypertension by urban–rural stratification
[Table 3] presents the prevalence of hypertension with 95% CI for the selected risk factors stratified by urban–rural place of residence. According to most of the characteristics, urban participants had a higher prevalence of hypertension than their rural counterparts. Overall, older participants were more likely to have hypertension. People aged above 54 in urban regions (55.11%; 95% CI: 51.83–58.40) had the highest prevalence of hypertension. Furthermore, 38.74% (95% CI: 36.73–40.76) and 37.75% (95% CI: 35.20–40.31) participants from hilly ecological zones, and 48.10% (95% CI: 44.35–51.84) and 42.48% (95% CI: 37.50–47.46) participants from province 4 were more prevalent to hypertension in both regions. Prevalence of hypertension was the highest for urban rich (36.68%; 95% CI: 34.71–38.66) and rural alcoholic (60.00%; 95% CI: 42.47–77.53) participants. Females were more hypertensive than males, with 36.55% (95% CI: 34.23–8.87) in urban and 34.61% (95% CI: 31.48–37.73) in rural regions. The majority of people with others marital status were more prevalent in hypertension in both urban (56.33%; 95% CI: 52.11–60.56) and rural (46.47%; 95% CI: 41.17–51.77) regions. For education level, urban participants had a higher prevalence of hypertension for all categories than their rural counterparts. More than half of the overweight and obese people in both regions (urban: 56.66%, 95% CI: 52.04–61.27; rural: 53.86%, 95% CI: 51.10–56.61) had a higher prevalence of hypertension.
|Table 3: Prevalence of hypertension stratified by urban-rural place of residence (n=7825)|
Click here to view
Results of multivariate logistic regression model
[Table 4] reports the estimation of COR and AOR for associated risk factors of hypertension by urban–rural stratification. The oldest age group (above 54 years) was more likely to have hypertension in both urban (AOR: 7.64; 95% CI: 5.65–10.34) and rural (AOR: 3.99; 95% CI: 5.24–11.14) regions, reference to the youngest age group (18–24 years). Urban hill (AOR: 1.18; 95% CI: 0.82–1.69) and rural terai (AOR: 1.51; 95% CI: 1.02–2.21) had higher odds of hypertension when compared to mountain ecological zones. People from province 4 (AOR: 1.86; 95% CI: 1.44–2.42) and province 5 (AOR: 1.88; 95% CI: 1.48–2.39) had significantly higher odds of hypertension in urban regions. However, participants from province 7 were less likely to have hypertension in urban (AOR: 0.85; 95% CI: 0.66–1.10) and rural (AOR: 0.81; 95% CI: 0.58–1.13) regions. Urban female (AOR: 1.03; 95% CI: 0.89–1.19) participants having a middle wealth index (AOR: 1.04; 95% CI: 0.86–1.26) with others marital status (AOR: 1.07; 95% CI: 0.75–1.54) were more likely to have hypertension. Furthermore, rural alcoholic (AOR: 1.66; 95% CI: 0.73–3.79) participants with secondary or above education level (AOR: 1.05; 95% CI: 0.79–1.39) had a higher odds of hypertension. Finally, the overweight or obese group was more likely to have hypertension in urban (AOR: 2.43; 95% CI: 2.08–2.85) and rural (AOR: 2.74; 95% CI: 2.16–3.47) region.
|Table 4: Estimation of crude and adjusted odds ratio for risk factors associated with hypertension stratified by urban-rural place of residence (n = 7825)|
Click here to view
| Discussion|| |
The study investigated a nationally representative survey of Nepal to explore the prevalence of hypertension and its associated risk factors by urban-rural stratification. Participant's age and ecological zones exhibited a positive significant relationship with hypertension only in rural regions, while overweight or obese BMI portrayed analogous results in both regions.
In this study, advancing age is a significant risk factor for hypertension only in rural regions. Preceding literature corresponds with this finding in the absence of stratification., A plausible reason is the age structural conversion of blood vessels, which flattens the vascular lumen and increases the risk of hypertension in the elderly. The fertility rate is declining and life expectancy is improving, which will inundate the number of geriatric population shortly in Nepal. Consequently, researchers are predicting the rapid growth of hypertension. By 2020, Nepal has a target to reduce the prevalence of hypertension and this target cannot be achieved without a high awareness among the rural geriatric population. Prior studies also reported a remarkably low awareness level for hypertension in the country., This increased level of hypertension among geriatric population can significantly be reduced by visiting healthcare institutions for frequent health checkup, and by intaking antihypertensive medication.,
Unlike previous findings, this study presents new insights concerning provinces and ecological zones. Hill and terai ecological zones, province 3 and 5 highlighted significant differences only in rural regions. However, this result contrasts from a study without stratification.
Besides, a significant association was visible for urban participants from province 4 and 5, which aligns with previous studies aggregating region of residence., Studies found, regional variation and ethnicity play a role in these differences., Furthermore, Dalit and Janajati ethnic groups are the majority in these provinces (province 4 and 5) and ecological zones (hill and terai). Physical inactiveness and consumption of alcohol, tobacco, and salty foods are the reasons for a higher likelihood of hypertension in these provinces and ecological zones.,
Without implementing a stratified setting, previous studies concluded that males are more prevalent in hypertension than their female counterparts.,, However, this study found that males only from rural regions are more prevalent in hypertension. This groundbreaking information will help Nepal to tackle hypertension effectively. Sex hormones, cultural, behavioral, and genetic differences are liable for this., Comprehensive health education and screening programs can reduce the burden of hypertension among rural male participants. The effect of BMI had significant positive odds in both regions and the result is also parallel with prior research. The Nepalese Government has to implement evidence-based policies (such as restriction on the advertisement and revenue enhancement on unhealthy foods and capitalize the consumption on healthy foods) at a large scale for prevention and control of overweight or obesity and hypertension.
This study also found a difference in education level and wealth index, which is a new addition to the knowledge of hypertension. Poor participants with no formal education had a higher chance of hypertension in both regions. However, previous studies have found a positive relationship between higher education level and hypertension., Illiterate individuals usually have a lower socioeconomic status, which results in lesser awareness for prevention and control of hypertension using antihypertensive medication. Hence, the study suggests a rise in public health awareness, especially for poor and illiterate Nepalese people. Besides, the essential health-care package (ESP) of Nepal does not provide service for NCDs (such as diabetes and hypertension) and the majority of people do not have adequate access to it. Drastic improvement is required, which includes the NCDs related services (such as providing antihypertensive drugs, an essential medication for NCDs at low or no cost).
Strengths and limitations
This is the first epidemiological study to investigate the prevalence of hypertension in Nepal stratified by urban–rural place of residence. The major strength of this study is that the NDHS 2016 survey covered both urban and rural regions, including all the seven provinces and three ecological zones. Therefore, the findings of this study is reflecting the entire Nepalese population. The highly skilled enumerators also provided an appropriate measurement for the participants' BP. Finally, the study selected risk factors from the entire NDHS database using a machine learning algorithm and evaluated their performance. This double validation helped to rigorously measure the AOR and COR from multivariate logistic regression model for robust knowledge generation. Despite these strengths, the study has some limitations. First, the absence of other risk factors in the NDHS 2016 dataset (such as the family history of hypertension, concomitant diabetes, physical activity, and salt intake) that could provide a more robust assessment. Second, the longitudinal measurement of BP with sphygmomanometers is recommended by the standard guidelines, but the BP measurement in the NDHS 2016 was recorded with an automated machine on a single day. Third, there was a possibility of measurement and classification errors caused by the enumerators. Finally, caution should be exercised while interpreting the findings (particularly for the central development region) given that the survey was conducted a year after the huge 7.8 magnitude earthquake that hit the country in 2015, leaving a widespread impact on the Nepalese population.
| Conclusion|| |
To conclude, this evidence-based study enhances the knowledge of hypertension. Overweight or obese participants were more hypertensive in both regions; hence, they need to adopt a healthier lifestyle to control their BMI. Besides, based on the distribution of important risk factors in both regions, policymakers and government officials need to tailor their campaigns differently. This study also revealed some groundbreaking insights, which will help the multisectoral action plan and the ESP tremendously for the prevention and control of hypertension. Both of the programs have to focus on hypertension management, especially for rural Dalit and Janajati older male residents from province 4 and 5, and hill and terai ecological zones to attain the targeted level of hypertension.
The authors would like to thank the MEASURE DHS (Monitoring and Evaluation to Assess and Use Result, Demographic and Health Survey), ICF International (Rockville, Maryland, USA) for allowing us to use NDHS 2016 dataset for analysis.
Conflicts of interest
There are no conflicts of interest.
| References|| |
Hay SI, Jayaraman SP, Truelsen T, Sorensen RJ, Millear A, Giussani G, et al
. GBD 2015 disease and injury incidence and prevalence collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990-2015: A systematic analysis for the global burden of disease stud. Lancet 2017;389:E1.
Forouzanfar MH, Afshin A, Alexander LT, Biryukov S, Brauer M, Cercy K, et al
. Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks, 1990-2015: A systematic analysis for the Global Burden of Disease Study 2015. Lancet 2016;388:1659-724.
NCD Risk Factor Collaboration (NCD-RisC). Worldwide trends in blood pressure from 1975 to 2015: A pooled analysis of 1479 population-based measurement studies with 19·1 million participants. Lancet 2017;389:37-55.
Nepal Ministry of Health & Population. Nepal Demographic And health Survey. Nepal Ministry of Health & Population; 2016.
Hasan M, Sutradhar I, Akter T, Das GR, Joshi H, Haider MR, et al
. Prevalence and determinants of hypertension among adult population in Nepal: Data from Nepal demographic and health survey 2016. Li Y, editor. PLoS One 2018;13:e0198028.
Kibria GM Al, Swasey K, Sharmeen A, Sakib MN, Burrowes V. Prevalence and associated factors of pre-hypertension and hypertension in Nepal: Analysis of the Nepal Demographic and Health Survey 2016. Heal Sci Rep 2018;1:e83.
Neuman M, Kawachi I, Gortmaker S, Subramanian SV. Urban-rural differences in BMI in low- and middle-income countries: The role of socioeconomic status. Am J Clin Nutr 2013;97:428-36.
Roth Z, Brown MM. The seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure: The JNC report. Vol. 7, Evidence-Based Ophthalmology. 2006. Available from: https://www.ncbi.nlm.nih.gov/pubmed/20821851
. [Last accessed on 2020 Aug 07].
Kursa MB, Rudnicki WR. Feature selection with the boruta package. Stat Softw 2010;36:1-13.
Neupane D, Shrestha A, Mishra SR, Bloch J, Christensen B, McLachlan CS, et al
. Awareness, prevalence, treatment, and control of hypertension in Western Nepal. Am J Hypertens 2017;30:907-13.
Pinto E. Blood pressure and ageing. Postgrad Med J 2007;83:109-14.
Government of Nepal National Planning Commission. Demographic Changes of Nepal: Trends and Policy Implications. Kathmandu; 2017. Available from: https://www.unicef.org/nepal
. [Last accessed on 2020 Aug 14].
World Health Organization. Multisectoral Action Plan for the Prevention and Control of Non Communicable Diseases (2014-2020). World Health Organization; 2014.
Devkota S, Dhungana RR, Pandey AR, Bista B, Panthi S, Thakur KK, et al
. Barriers to treatment and control of hypertension among hypertensive participants: A community-based cross-sectional mixed method study in municipalities of Kathmandu, Nepal. Front Cardiovasc Med 2016;3:1.
Karmacharya BM, Koju RP, LoGerfo JP, Chan KC, Mokdad AH, Shrestha A, et al
. Awareness, treatment and control of hypertension in Nepal: Findings from the Dhulikhel Heart Study. Heart Asia 2017;9:1-8.
Mehata S, Shrestha N, Mehta R, Vaidya A, Rawal LB, Bhattarai N, et al
. Prevalence, awareness, treatment and control of hypertension in Nepal: Data from nationally representative population-based cross-sectional study. Hypertens 2018;36:1680-8.
Harshfield E, Chowdhury R, Harhay MN, Bergquist H, Harhay MO. Association of hypertension and hyperglycaemia with socioeconomic contexts in resource-poor settings: The Bangladesh Demographic and Health Survey. Int J Epidemiol 2015;44:1625-36.
Meshram II, Vishnu Vardhana Rao M, Sudershan Rao V, Laxmaiah A, Polasa K. Regional variation in the prevalence of overweight/obesity, hypertension and diabetes and their correlates among the adult rural population in India. Br J Nutr 2016;115:1265-72.
Thapa N, Aryal KK, Puri R, Shrestha S, Shrestha S, Thapa P, et al
. Alcohol consumption practices among married women of reproductive age in Nepal: A population based household survey. Portolés M, editor. PLoS One 2016;11:e0152535.
Khanal MK, Dhungana RR, Bhandari P, Gurung Y, Paudel KN. Prevalence, associated factors, awareness, treatment, and control of hypertension: Findings from a cross sectional study conducted as a part of a community based intervention trial in Surkhet, Mid-western region of Nepal. Kiechl S, editor. PLoS One 2017;12:e0185806.
Hossain FB, Adhikary G, Chowdhury AB, Shawon MS. Association between body mass index (BMI) and hypertension in south Asian population: Evidence from nationally-representative surveys. Clin Hypertens 2019;25:1-9.
Chang YP, Liu X, Kim JD, Ikeda MA, Layton MR, Weder AB, et al
. Multiple genes for essential-hypertension susceptibility on chromosome 1q. Am J Hum Genet 2007;80:253-64.
Mills KT, Bundy JD, Kelly TN, Reed JE, Kearney PM, Reynolds K, et al
. Global disparities of hypertension prevalence and control. Circulation 2016;134:441-50.
Afshin A, Forouzanfar MH, Reitsma MB, Sur P, Estep K, Lee A, et al
. Health effects of overweight and obesity in 195 countries over 25 years. N Engl J Med 2017;377:13-27.
Busingye D, Arabshahi S, Subasinghe AK, Evans RG, Riddell MA, Thrift AG. Do the socioeconomic and hypertension gradients in rural populations of low and middle-income countries differ by geographical region? A systematic review and meta-analysis. Int J Epidemiol 2014;43:1563-77.
[Table 1], [Table 2], [Table 3], [Table 4]