Effect of Previous Exposure to Malaria on Blood Pressure in Kilifi, Kenya: A Mendelian Randomization Study

Background Malaria exposure in childhood may contribute to high blood pressure (BP) in adults. We used sickle cell trait (SCT) and α+thalassemia, genetic variants conferring partial protection against malaria, as tools to test this hypothesis. Methods and Results Study sites were Kilifi, Kenya, which has malaria transmission, and Nairobi, Kenya, and Jackson, Mississippi, where there is no malaria transmission. The primary outcome was 24‐hour systolic BP. Prevalent hypertension, diagnosed using European Society of Hypertension thresholds was a secondary outcome. We performed regression analyses adjusting for age, sex, and estimated glomerular filtration rate. We studied 1127 participants in Kilifi, 516 in Nairobi, and 651 in Jackson. SCT frequency was 21% in Kilifi, 16% in Nairobi, and 9% in Jackson. SCT was associated with −2.4 (95% CI, −4.7 to −0.2) mm Hg lower 24‐hour systolic BP in Kilifi but had no effect in Nairobi/Jackson. The effect of SCT in Kilifi was limited to 30‐ to 59‐year‐old participants, among whom it was associated with −6.1 mm Hg (CI, −10.5 to −1.8) lower 24‐hour systolic BP. In pooled analysis allowing interaction by site, the effect of SCT on 24‐hour systolic BP in Kilifi was −3.5 mm Hg (CI, −6.9 to −0.1), increasing to −5.2 mm Hg (CI, −9.5 to −0.9) when replacing estimated glomerular filtration rate with urine albumin to creatinine ratio as a covariate. In Kilifi, the prevalence ratio for hypertension was 0.86 (CI, 0.76–0.98) for SCT and 0.89 (CI, 0.80–0.99) for α+thalassemia. Conclusions Lifelong malaria protection is associated with lower BP in Kilifi. Confirmation of this finding at other sites and elucidating the mechanisms involved may yield new preventive and therapeutic targets.

exposure to malaria as Nairobi and Kilifi have markedly contrasting malaria transmission patterns. Study participants in Nairobi were randomly selected from those who had self-identified as belonging to ethnic groups known to have a high frequency of malaria protective polymorphisms (Luhya, Luo, Teso, Mijikenda) as a result of hailing from parts of Kenya that are known to be endemic for malaria. 3,4 Study participants in Kilifi were predominantly from the Chonyi subtribe of the Mijikenda community. The prevalence of hypertension within the Kilifi Health and Demographic Surveillance System which covers an area of 900km 2 is ~17%. 5 However there are significant differences in the incidence to death due to stroke within the study area. Chasimba where >75% of study participants came from has an incidence of death due to stroke that is three times that of Kilifi township, suggesting that there are local geographical differences in the prevalence of hypertension which is the main risk factor for stroke.
In both Kilifi and Nairobi, trained study staff visited all individuals who had been selected to participate in the study at their homes and requested them to come to the study clinic to undergo study procedures. Those who failed to come to the clinic within 3 months of being requested to do so were considered to have declined our invitation to participate in the study.
At the clinic participants first underwent an interview where they answered questions about their past medical history and their socioeconomic status based on the multidimensional poverty (MDP) index. 6 Weight and height were measured using a validated SECA 874™ weighing machine and a portable stadiometer (Seca 213™), respectively. Body mass index was calculated as the weight in kilograms divided by height in meters squared (kg/m 2 ). We did not classify BMI by age-category in the adolescents that we studied. Mid-upper arm circumference (MUAC) was measured in a standardized manner using TALC™ MUAC tapes. All participants were subsequently fitted with a validated Arteriograph24™(TensioMed Ltd., Budapest, Hungary) device for 24-hour ABPM measurement. 7 The devices were attached on the non-dominant arm and were programmed to take measurements every 20 minutes during daytime hours (0600-2200 hrs) and every 40 minutes at night (2200-0600 hrs). At the end of the 24-hour period, participants returned to the study clinic where the Arteriograph was removed and data downloaded onto computers that would later (within 12 hours) synchronize their data onto an MySQL database hosted on servers located at the KEMRI-Wellcome Trust Research Programme offices in Kilifi, Kenya.
We collected 10ml of blood from participants for full blood count, determination of sickle hemoglobin status and serum electrolytes. After performing automated full blood counts using an ACT 5™ machine, whole blood samples were frozen at -80ºC and then transported to the KEMRI-Wellcome Trust Research Programme laboratories in Kilifi, Kenya for determination of sickle hemoglobin status. DNA was extracted retrospectively from the frozen samples by use of Qiagen™ DNA blood mini-kits (Qiagen, Crawley, United Kingdom) and typed for sickle hemoglobin and  + thalassemia using polymerase chain reaction. Glycosylated hemoglobin levels were determined using the Biorad™ D-10 machine (Bio-rad Laboratories Inc, Hercules, California).
Serum and urine samples collected from participants were frozen at -80ºC within 4 hours of collection and later transported to the laboratories in Kilifi for analysis. We determined urea and creatinine levels in these samples using ion electrophoresis and the jaffe method, respectively. 8 Creatinine measurements were performed using Isotope dilution mass spectrometry traceable methods. In addition, we determined albumin levels in the urine samples by immunoturbidometry using a Quantex™ microalbumin kit. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation in adults and the Schwartz equation in those aged ≤16 years. 9, 10 b) Study Procedures in Jackson, Mississippi, USA The Jackson Heart Study (JHS) 11 is a population-based prospective cohort study, which was designed to evaluate cardiovascular disease risk among blacks. The JHS enrolled 5306 noninstitutionalized blacks, aged ≥21 years, between 2000 and 2004.
The participants were recruited from the Atherosclerosis Risk in the Community site in Jackson, MS, and a representative sample of urban and rural Jackson, MS, metropolitan tricounty (Hinds, Madison, and Rankin counties) residents, volunteers, randomly selected individuals, and secondary family members. 12 The current analysis was restricted to JHS participants who completed ABPM soon after the baseline study visit (visit 1).
During in-home interviews, trained African American interviewers administered standardized questionnaires to collect self-reported information on sociodemographics (e.g. age, sex, education, marital status and socioeconomic status), previously diagnosed co-morbid conditions and selected health-related behaviors (e.g. current smoking). Weight and height were measured during a clinic visit.
At the clinic visit, blood samples were collected for full blood count, genetic studies and determination of serum sodium, potassium and creatinine concentrations. 24hour urine samples were collected for determination of creatinine and albumin concentrations. Full blood counts were performed using the Coulter GenS machine (BeckmanCoulter, Hialeah, Florida, USA). DNA was extracted from whole blood samples using Puregene reagents (Gentra System, Minneapolis, USA) and genetic studies were performed as previously described. 12

b) Sample size estimation
The sample size calculation for Kilifi was based on a two-sample t-test comparing mean 24-hour systolic blood pressure in those with and without the sickle cell trait (SCT). The following assumptions were made: -That the prevalence of SCT would be ≥15% 17 -That the standard deviation of 24-hour systolic BP would be ≤15 mm Hg 5,18 Based on these assumptions we calculated that, for Kilifi, we would need a minimum of 1115 participants with complete data in order to detect a statistically significant 4 mm Hg difference in 24-hour systolic BP with at least 80% statistical power.
For participants in Nairobi/Jackson we assumed that the combined SCT prevalence for the two sites would be ≥ 10%. 19 Other assumptions were similar to those for Kilifi.
Based on these assumptions we calculated that for Nairobi/Jackson, we would need minimum of 1270 participants with complete data in order to detect a statistically significant 4 mm Hg difference in 24-hour systolic BP with at least 80% statistical power.
We assumed that with these numbers, we would achieve enough power for the primary outcome measure, a linear regression to determine the effect of SCT on 24hour BP measures, while adjusting for age, sex and estimated glomerular filtration rate (eGFR). The literature suggests that the major consideration for sample size calculations in linear regression models is to ensure that there are at least 2-50 individuals per variable in the model 20 , a requirement that would almost certainly be achieved if most of the assumptions stated above held true.

c) Quality control criteria for ABPM data
There are 2 internationally recognized quality control criteria used for ABPM data, which are based on completeness of observations. The International Database of Ambulatory blood pressure in relation to Cardiovascular Outcomes (IDACO) study 21 defined ABPM data as acceptable if they include ≥ 10 daytime and ≥ 5 nighttime readings, where daytime is defined as 1000-2000 hrs and nighttime as 0000-0600 hrs. 21 The guidelines from the European Society of Hypertension (ESH) are more stringent; they require ≥20 daytime and ≥7 nighttime readings where daytime is defined as 0900 to 2100 hrs and nighttime as 0100 to 0600 hours. 22 It is important to note that these criteria were arbitrarily set by experts and were not based on outcome studies. As the ESH criteria are more stringent they are likely to lead to a greater loss of data and subsequent loss of power and precision. However, in order to reduce measurement bias and obtain as accurate an effect size as possible, an a priori decision was made to restrict our primary analysis to data that met the ESH criteria.

d) Primary and secondary outcome measures
The primary outcome measure was estimated using a linear regression model to determine the effect of SCT on 24-hour systolic blood pressure, after adjusting for age, sex and estimated glomerular filtration rate (eGFR). Blood pressures were obtained by ambulatory blood pressure monitoring using the Arteriograph24™ device. 7 Numerous studies have shown that the more accurate measurements resulting from repeated inflations and more standardized procedures in ABPM make it a much better predictor of cardiovascular events than other BP measurement methods. 22 The justification for adjusting for age, sex and eGFR is given in the section below on confounders and model building.
Secondary outcome measures were defined as follows: a) effect of  + thalassaemia on 24-hour, daytime and nighttime systolic blood pressures, after adjusting for age, sex and estimated glomerular filtration rate b) effect of SCT on 24-hour, daytime and nighttime diastolic blood pressures, after adjusting for age, sex and estimated glomerular filtration rate c) effect of  + thalassaemia on 24-hour, daytime and nighttime diastolic blood pressures, after adjusting for age, sex and estimated glomerular filtration rate d) prevalence ratio for hypertension in those with and without SCT using logbinomial regression, adjusting for age, sex and estimated glomerular filtration rate Hypertension was diagnosed by any one of the following criteria in individuals aged ≥16 years: 22

e) Adjusting for confounders and model building
The theoretical basis for the malaria-high blood pressure hypothesis has been published previously. 25 Briefly, the primary hypothesis is that individuals in Kilifi who were exposed to more malaria disease in childhood (represented by those having haemoglobin AA) would have higher 24-hour systolic blood pressure than those who were exposed to less malaria disease (represented by those having haemoglobin AS [SCT]). The proposed causal diagram drawn purely for purposes of informing the analytical plan can be found in Figure S2.
For purposes of this analysis it is important to note that because malnutrition and stunting are on the causal pathway from malaria to the outcome, adjustment for body mass index (BMI) and other anthropometric indices (e.g. mid upper arm circumference) would be inappropriate.

i) Confounders
The principle of Mendelian randomization holds that because comparisons are based on genetic traits acquired at conception, any relationships between the genetic trait and the outcome are unlikely to be confounded by other exposures as these will be randomly distributed between carriers and non-carriers of the trait. 26 However age, sex, and BMI are known to have a very strong influence on BP and other cardiovascular diseases 27 , and are usually adjusted for as 'fixed covariates' in MR/Genome wide association studies [28][29][30][31] . We have outlined above why it would be inappropriate to adjust for BMI.
Sickle cell trait has been associated with impaired kidney function as measured by decline in estimated glomerular filtration rate (eGFR) and albuminuria. 19 This association is independent of blood pressure elevation. Impaired kidney function is associated with elevations in blood pressure as a result of sodium retention 32 , increased activity of the renin-angiotensin system 33 , increased sympathetic activity 34 , secondary hyperparathyroidism 35 , impaired nitric oxide synthesis 36 and increased prevalence of nocturnal non-dipping BP pattern. 37 It is also possible that kidney disease could arise from hypertension. 38 The direction of the relationship between blood pressure and kidney function, has been the subject of debate. 39 However, evidence from genetic studies suggests that the association between renal function and blood pressure is likely to likely to be explained by decreased renal function giving rise to high blood pressure. In a large (n>200,000) genome wide association study (GWAS), loci that were associated with blood pressure elevation and cardiovascular disease showed no association with kidney disease or kidney function. 29 If SCT compromises renal function and this in turn leads to elevated blood pressure, this would result in a bias toward a null result when using SCT as a proxy for testing the malaria-high blood pressure hypothesis. As can be seen in Figure S3, impaired kidney function (as measured by eGFR) is associated with both the exposure and the outcome, but is not on the causal pathway from malaria to the outcome. Kidney function is therefore considered a confounder and we adjusted for eGFR in all regression analyses. We also examined the effect of using urine albumin to creatinine ratio in place of eGFR in the regression models.
If, however, renal function lies on the causal pathway between malaria and high blood pressure it would be inappropriate to include it within the regression models.
Severe malaria does occasionally present with acute renal failure and repeated episodes of malaria could result in chronic pyelonephritis and elevated BP. However, acute renal failure is a very rare complication of malaria in Kilifi, for example, it occurred in 2 out of 1844 children admitted with malaria. 40 This suggests that renal failure is an unlikely mediator of the potential association between malaria and elevated BP.
We confirmed that each of the a priori specified covariates (age, sex and eGFR) significantly improved the regression models using the likelihood ratio test.

Confounding due to pleiotropy
A special type of confounding can also occur if the genetic trait influences the outcome through a pathway that is independent of the exposure (pleiotropy) 41 as illustrated in Figure S4.
In contrast with renal function, which is a known confounder and can be measured and adjusted for in regression analyses, confounding due to other (often unknown) causes can only be detected by examining the relationship between sickle cell trait and blood pressure in individuals who have not been exposed to malaria. The existence of pleiotropy can invalidate the use of the genetic trait as a marker for the infectious disease exposure. In order to exclude pleiotropy as a potential explanation for the association between SCT and BP, we studied lifelong residents of Nairobi, Kenya and Jackson, Mississippi, two sites where there is no malaria transmission.
In addition, we conducted a pooled analysis incorporating data from the three study sites and conducted a linear regression with the previously specified covariates plus SCT and study site and their interaction. This increased the power to detect any independent effect of SCT on BP while simultaneously checking for differential effect of SCT according to study site.
ii) Effect modifier:  + thalassemia In a related analysis, we ran a linear regression model examining the effect of  + thalassemia on blood pressure with the same covariates used in the main analysis for SCT. Because  + thalassemia confers less protection against malaria than SCT, we expected that the effect estimate in this model would be lower than that of SCT.

f) Testing for cohort effects
Malaria incidence in Kilifi has been changing over time and we considered that this could influence results obtained. Data on the changing levels of transmission go back to 1990 and they show that a significant drop in transmission in Kilifi began around 1999-2000. 43 In addition, because blood pressure rises with age, it is possible that the effects of malaria on outcome measures may be more apparent later in life. While it is not possible to determine the individual contributions of changing malaria exposure and aging to any differences observed in outcome measures, we attempted to display these differences by performing comparisons of the outcomes by sickle trait in 3 age strata. Did not consent n=166 Consented n=1026 Did not consent n=93 Consented n=1148 Did not consent n=4158        Results of analyses based on data meeting IDACO criteria for completeness        Sickle cell trait Malaria eGFR High blood pressure Figure S4. Illustrating confounding due to pleiotropy.

Sickle cell trait Malaria
Effect not related to malaria High blood pressure