1. Epidemiology and Global Health
  2. Microbiology and Infectious Disease
Download icon

Impact of asymptomatic Plasmodium falciparum infection on the risk of subsequent symptomatic malaria in a longitudinal cohort in Kenya

  1. Kelsey M Sumner
  2. Judith N Mangeni
  3. Andrew A Obala
  4. Elizabeth Freedman
  5. Lucy Abel
  6. Steven R Meshnick
  7. Jessie K Edwards
  8. Brian W Pence
  9. Wendy Prudhomme-O'Meara
  10. Steve M Taylor  Is a corresponding author
  1. Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, United States
  2. Division of Infectious Diseases, School of Medicine, Duke University, United States
  3. School of Public Health, College of Health Sciences, Moi University, Kenya
  4. School of Medicine, College of Health Sciences, Moi University, Kenya
  5. Academic Model Providing Access to Healthcare, Moi Teaching and Referral Hospital, Kenya
  6. Duke Global Health Institute, Duke University, United States
Research Article
  • Cited 0
  • Views 486
  • Annotations
Cite this article as: eLife 2021;10:e68812 doi: 10.7554/eLife.68812

Abstract

Background:

Asymptomatic Plasmodium falciparum infections are common in sub-Saharan Africa, but their effect on subsequent symptomaticity is incompletely understood.

Methods:

In a 29-month cohort of 268 people in Western Kenya, we investigated the association between asymptomatic P. falciparum and subsequent symptomatic malaria with frailty Cox models.

Results:

Compared to being uninfected, asymptomatic infections were associated with an increased 1 month likelihood of symptomatic malaria (adjusted hazard ratio [aHR]: 2.61, 95% CI: 2.05 to 3.33), and this association was modified by sex, with females (aHR: 3.71, 95% CI: 2.62 to 5.24) at higher risk for symptomaticity than males (aHR: 1.76, 95% CI: 1.24 to 2.50). This increased symptomatic malaria risk was observed for asymptomatic infections of all densities and in people of all ages. Long-term risk was attenuated but still present in children under age 5 (29-month aHR: 1.38, 95% CI: 1.05 to 1.81).

Conclusions:

In this high-transmission setting, asymptomatic P. falciparum can be quickly followed by symptoms and may be targeted to reduce the incidence of symptomatic illness.

Funding:

This work was supported by the National Institute of Allergy and Infectious Diseases (R21AI126024 to WPO, R01AI146849 to WPO and SMT).

Introduction

Asymptomatic Plasmodium falciparum infections, defined as the presence of parasites in the absence of symptoms, are common across sub-Saharan Africa. In 2015, a geo-spatial meta-analysis estimated a continent-wide prevalence of asymptomatic P. falciparum in children aged 2 to 10 years of 24% based on microscopy and rapid diagnostic test (RDT) results (Snow et al., 2017). In high-transmission settings, these infections are more common, with point prevalence among adults exceeding 30% in the Democratic Republic of the Congo (Taylor et al., 2011) and Malawi (Topazian et al., 2020). Though by definition lacking acute symptomatology, these infections when persistent can adversely affect the individual (Cottrell et al., 2015; Maketa et al., 2015; Matangila et al., 2014; Sifft et al., 2016) as well as serve as a reservoir for onward parasite transmission (Gouagna et al., 2004; Tadesse et al., 2018).

Although epidemiologically important, the natural history of asymptomatic P. falciparum and its relationship to future symptomatic malaria remains unclear. In prior studies, asymptomatic P. falciparum infections have been observed to both decrease (Buchwald et al., 2019; Males et al., 2008; Portugal et al., 2017; Sondén et al., 2015) and increase (Le Port et al., 2008; Liljander et al., 2011; Njama-Meya et al., 2004; Nsobya et al., 2004) the risk of symptomatic malaria, and several have correlated this heterogeneity to people’s age or site transmission intensity (Henning et al., 2004; Wamae et al., 2019). Inferences have been further complicated in these studies owing to their cross-sectional capture of asymptomatic infections (Henning et al., 2004; Liljander et al., 2011; Males et al., 2008; Nsobya et al., 2004; Sondén et al., 2015; Wamae et al., 2019), short follow-up periods (Le Port et al., 2008; Njama-Meya et al., 2004), or limited age ranges (Le Port et al., 2008; Liljander et al., 2011; Males et al., 2008; Njama-Meya et al., 2004; Nsobya et al., 2004; Wamae et al., 2019), which collectively undermine a clear understanding of the risk of symptomatic malaria following the detection of an asymptomatic infection.

We investigated the natural history of asymptomatic P. falciparum infections in a high-transmission setting using a 29-month longitudinal cohort of people aged 1 to 85 years in Western Kenya. Using monthly active case detection of asymptomatic infections and passive capture of symptomatic events, we evaluated the likelihood of symptomatic malaria following an asymptomatic P. falciparum infection. We hypothesized that infection with asymptomatic parasitemia would be associated with a decrease in future risk of symptomatic malaria compared to uninfected people and that, because age serves as a proxy for prior cumulative exposure, this effect would be most pronounced in older people.

Materials and methods

Study population, sample collection, and sample processing

Request a detailed protocol

From June 2017 to November 2019, we followed a cohort of 268 people aged 1 to 85 years living in 38 households in a rural setting in Webuye, Western Kenya (O'Meara et al., 2020). The cohort was assembled using radial sampling of 12 households per village for three villages with high malaria transmission. The first household in each village was randomly selected. Two households moved during follow-up and were replaced. For each person, asymptomatic P. falciparum infections were detected monthly by active surveillance through collecting questionnaires and dried blood spot (DBS) samples for post hoc molecular parasite detection. Symptomatic P. falciparum infections were detected using passive surveillance by testing people with self-reported symptoms with a malaria RDT (Carestart Malaria HRP2 Pf from Accessbio) and collecting a DBS (AccessBio, 2019). People with positive RDT results were treated with Artemether-Lumefantrine (AL).

DBS were processed to detect P. falciparum infections by extracting genomic DNA (gDNA) from DBS and then tested in duplicate for P. falciparum parasites using a duplex real-time PCR (qPCR) assay targeting the P. falciparum pfr364 motif and human β-tubulin gene (Plowe et al., 1995; Taylor et al., 2019). From each DBS, three individual punches were deposited in a single well of a 96-well deep well plate and extracted with Chelex-100 following Saponin and Proteinase K treatments. gDNA was ultimately preserved in approximately 100 μL of solution. Each gDNA extract was tested in duplicate. Each reaction contained 2 μL of gDNA template in a 12 μL total reaction volume, and templates were tested in 384-well plates on an ABI QuantStudio6 platform. Samples were defined as P. falciparum-positive if: (i) both replicates amplified P. falciparum and both Ct values were < 40 or (ii) one replicate amplified P. falciparum and the Ct value was < 38.

Parasite densities were estimated using standard curves generated from amplifications on each plate from templates of known parasite density. To generate these templates, parasite strain 3D7 was cultivated in vitro using standard conditions and the parasite density was estimated initially by light microscopy of Giemsa-stained slides and then by hemocytometer. For the latter, after averaging estimates of parasite density from five to six chambers, the non-diluted sample was diluted with fresh whole blood to obtain a 2000 p/μL stock solution. This was then serially diluted with whole blood to obtain 1000, 200, 100, 20, 10, 2, 1, 0.2, and 0.1 p/μL samples. Each of these was then prepared as a DBS of 30 μL volume, and from these gDNA was extracted as above for clinical samples. The result was a series of gDNA samples from mocked templates of known concentrations of 3D7 parasite that were processed identically to the clinical samples.

Exposure and outcome ascertainment

Request a detailed protocol

The main exposure was an asymptomatic P. falciparum infection during monthly active case detection assessments, defined as P. falciparum-positive by qPCR in a person lacking symptoms. People who were P. falciparum-negative by qPCR during monthly visits were considered uninfected. Participant follow-up was imputed for the first consecutive missed monthly visit during each follow-up period by carrying forward the previous month’s value as the exposure status of the missed monthly visit (Nguyen et al., 2018). If a person missed two or more consecutive monthly visits, they were considered lost to follow-up and censored at the time of the imputed monthly visit. A sensitivity analysis was conducted for imputation using a dataset without imputation for missed monthly visits. Participants were allowed to enter and leave the study throughout the study period. At the end of the study period, all participants were censored.

The main outcome assessed was days to symptomatic malaria. We defined symptomatic P. falciparum infection as the current presence of at least one symptom consistent with malaria during a sick visit (i.e. fever, aches, vomiting, diarrhea, chills, cough, or congestion) and P. falciparum-positive by both RDT and qPCR. Outcome events occurring within 14 days of receipt of AL for a symptomatic infection were excluded.

Some participants were classified as symptomatically infected at a monthly visit through passive detection of symptoms; this occurred if a study team member conducting a monthly visit was approached by a participant reporting malaria-like symptoms. The study team member would then perform an RDT and record information as for a passively detected sick visit. Symptoms were not routinely elicited during interviews on monthly visits. If the person met the case definition for symptomatic malaria, they were confirmed as having a symptomatic visit on that day and removed from follow-up until 14 days post-receipt of antimalarials or the next monthly follow-up visit. If that person did not meet our case definition for symptomatic malaria, then they were removed from follow-up for that month and re-entered for follow-up in the following month.

Hazard of symptomatic malaria analysis

Request a detailed protocol

Across all participants, we estimated the hazard of subsequent symptomatic malaria when infected with asymptomatic malaria compared to being uninfected at monthly visits. The hazard of symptomatic malaria was calculated for multiple follow-up periods: (i) 1 month, (ii) 3 months, (iii) 6 months, (iv) 12 months, and (v) 29 months (entire study period). For each follow-up period exceeding 1 month, exposure status was ascertained at every monthly follow-up visit and allowed to vary each month using a method proposed by Hernán et al., 2005. This method treats each monthly follow-up visit as a new study entry, recalculating the time to symptomatic malaria or censoring using each monthly follow-up visit date as the origin and attributing the exposure in that month as the exposure status from that month up until the event or censoring (Figure 1). This exposure coding method was chosen due to its ability to capture the exposure at multiple time points with less risk of misclassification or left truncation bias compared to alternative time-varying coding approaches (Supplementary file 1).

Schematic of how asymptomatic exposure status was ascertained for one participant’s follow-up.

The method treated each monthly follow-up visit as a new study entry for the participant (denoted by circles), recalculating the time to symptomatic malaria (denoted by X) using the monthly follow-up visit date as the origin. The exposure status for each monthly follow-up visit became the exposure status for the follow-up period (denoted by the horizontal line). The follow-up period ended if the participant had a symptomatic infection (X) or was censored due to the study ending or becoming lost to follow-up (denoted by squares). Illustrated here is one hypothetical participant’s follow-up during 29 months, during which they contributed follow-up periods after being uninfected (green lines) and after having asymptomatic infections (orange lines), and suffered one episode of symptomatic malaria (denoted by X), before being censored at the close of follow-up (denoted by squares). As a result, each participant contributes multiple entries to each model equal to the number of exposure assessments, and models include a random effect at the level of the participant to account for repeated observations.

Statistical modeling

Request a detailed protocol

We first estimated the time to symptomatic malaria across the full 29 months using Kaplan-Meier curves and the log-rank test. We compared differences in median time to symptomatic malaria across select covariates using the Wilcoxon rank sum test with continuity correction for dichotomous variables or the Kruskal-Wallis test for polytomous variables. The Bonferroni correction was applied to all table p-values to account for repeated measures during the 29 months of follow-up.

In order to account for anticipated confounders of the relationship between asymptomatic infection and symptomatic malaria, we next computed a multivariate frailty Cox proportional hazards model (Equation 1).

(1) h1(t)ih0(t)i=exp(αi+β1Aymptomaticinfectionim+β2Age5to15i+β3Ageover15i+β4Femalei+β5Regularbednetusagei+β6village:Maruti+β7village:sitabicha+ϵi)

The model controlled for the following confounders as determined by a directed acyclic graph (Figure 3—figure supplement 1): age (<5 years, 5 to 15 years, >15 years), sex, and regular bed net usage (averages > 5 nights a week sleeping under a bed net – yes, no). To account for differences in malaria prevalence across the three villages, we also included a covariate in the model to represent each village. We allowed the main exposure to vary each month based on the monthly follow-up visit infection status (m), and included a random intercept at the participant level (αi) to account for potential correlated intra-individual outcomes. A log-normal distribution was used for the random effect. ϵi represented the model’s error term. Additional models incorporated either an additional random effect at the household level or a robust error estimator at the participant level. The proportional hazards assumption was assessed using Kaplan-Meier curves and Schoenfeld residual plots.

We tested for effect measure modification by age and sex by stratifying the multivariate model by age category (<5 years, 5 to 15 years, >15 years) or sex, computing hazard ratios and 95% confidence intervals (CI) of the main exposure, and comparing a Cox proportional hazards model with an interaction term between the potential modifier and main exposure to Equation 1 using the log-likelihood ratio test.

We computed an additional time-to-event model using a subset of events. Because asymptomatic infections could represent incipiently symptomatic (i.e. ‘pre-symptomatic’) infections, we excluded all monthly follow-up visits occurring within 14 days prior to a symptomatic infection, reducing the possibility that pre-symptomatic infections could be misclassified as asymptomatic. The time frame for identifying potentially pre-symptomatic infections was chosen for consistency with previous work studying time to symptomatic malaria (Buchwald et al., 2019). The analysis was conducted using Equation 1 for the 1-, 3-, 6-, 12-, and 29-month follow-up periods. All statistical analyses were performed using R version 4.0.2 (R Development Core Team, 2020) with the packages tidyverse (Wickham et al., 2019), survminer (Kassambara et al., 2020), survival (Therneau and Grambsch, 2000), coxme (Therneau, 2020), lme4 (Bates et al., 2015), and ggalluvial (Brunson, 2020). Code is available on Github: https://github.com/duke-malaria-collaboratory/time_to_symptomatic_malaria, (Sumner, 2021a; copy archived at swh:1:rev:95b7f8268baa6007af84cc7ee0f110f2a3629631Sumner, 2021b). Statistical significance was assessed at an α level of 0.05.

Detectability of asymptomatic infections

Request a detailed protocol

Asymptomatic infections defined as above were further classified as meeting a series of thresholds of parasite densities: any density, >1, >10, >100, >500, and >1000 parasites/μL. These classifications were assigned in a non-mutually exclusive fashion to asymptomatic infections, and then the 1-month likelihoods of symptomatic malaria relative to uninfected people were modeled separately using the Cox proportional hazards model in Equation 1. As an additional analysis, we repeated this process for each parasite density threshold stratified by participant age (<5 years, 5 to 15 years, >15 years).

Sensitivity analyses

Request a detailed protocol

We computed sensitivity analyses to account for potential misclassification of the outcome of symptomatic malaria in the main models over 1- and 29-month intervals. A ‘permissive’ case definition defined a symptomatic infection as one where a participant had at least one symptom consistent with malaria during a sick visit and was P. falciparum-positive by real-time PCR (qPCR). A ‘stringent’ case definition defined a symptomatic infection as one where a participant had a self-reported fever during a sick visit and was P. falciparum-positive by both RDT and qPCR. Additional sensitivity analyses were computed to investigate the separate effects of additional covariates by incorporating into the frailty Cox proportional hazards model in Equation 1, a new term for the covariate of interest. For seasonality, we classified monthly visits that occurred any time from May to October as the high-transmission season and from November to April as the low-transmission season, based on the region’s rainy seasons. For the number of prior infections, we included in the model as a covariate the number of prior infections as a continuous number. For prior antimalarial treatment, we included a variable coded dichotomously as having received study-prescribed antimalarials up until that monthly visit or not; a person was coded as having not received study-prescribed antimalarials up until their first symptomatic infection, but afterward were coded as receiving treatment from that point forward in follow-up.

Ethical review

Request a detailed protocol

The study was approved by institutional review boards of Moi University (2017/36), Duke University (Pro00082000), and the University of North Carolina at Chapel Hill (19-1273). All participants or guardians provided written informed consent, and those over age 8 years provided additional assent.

Results

For 29 months, we followed 268 participants from three villages in Western Kenya. After excluding participants with less than 2 months of follow-up, the analysis dataset consisted of 257 participants with a median of 222 days (interquartile range [IQR]: 89, 427) of follow-up and a median age of 13 years (range: 1, 85) (Figure 3—figure supplement 2). Overall, 5379 person-months at risk were observed with 1842 (34.2%) person-months of asymptomatic malaria exposure; the median total months of asymptomatic exposure for a participant was 9 (IQR: 5, 17). Exposure status frequently changed for participants and remained constant for only 16 (6.2%) people across follow-up; four people were asymptomatically infected for the entirety of follow-up and only 12 people were never infected (Figure 2A). We recorded 266 symptomatic malaria events. Participants had a median of 1 (IQR: 0, 2) symptomatic infection during follow-up. Median time to symptomatic malaria when asymptomatically infected (173, IQR: 49, 399) was shorter than when uninfected (230, IQR: 98, 402), as well as shorter for participants aged 5 to 15 years or living in the village Maruti (Table 1). Comparison of Kaplan-Meier curves over the full 29 months indicated a difference in the time to symptomatic malaria in the first few months post asymptomatic infection but not long term (p-value = 0.100 by log-rank test) (Figure 2B). Results for secondary case definitions for symptomatic malaria were overall similar and are provided in the supplement (Supplementary files 2 and 3).

Asymptomatic malaria exposure classifications and symptomatic malaria outcomes over time.

(A) The proportion of participants who had either an asymptomatic infection (orange) or were uninfected (green) at each monthly visit is indicated by the bars. The ribbons connecting the bars illustrate the proportion of participants who moved exposure status from month to month. Orange ribbons indicate the proportion of participants with asymptomatic infections and green the proportion that were uninfected. (B) A Kaplan-Meier survival curve of symptomatic events across the follow-up period stratified by asymptomatic malaria exposure is displayed. Each exposure-outcome pair depicted in Figure 1 is plotted, and therefore each study participant is registered in the curve multiple times, equal to the number of exposure classifications.

Table 1
Covariate distribution across symptomatic malaria events in 29 months of follow-up.
Total person-months
(N, %)
Person-months ending in symptomatic infections
(N, %)
Median time to symptoms for entire study
(days, IQR)
p-Value comparing time to symptoms
Main exposure<0.001§
No infection3537 (65.8)1580 (65.7)230 (98, 402)
Asymptomatic infection1842 (34.2)826 (34.3)173 (49, 399)
Age0.015
<5 years812 (15.1)329 (13.7)226 (82, 435)
5 to 15 years2279 (42.4)1319 (54.8)199 (70, 379)
>15 years2288 (42.5)758 (31.5)244 (97, 426)
Sex0.779§
Male2360 (43.9)1190 (49.5)229 (86, 420)
Female3019 (56.1)1216 (50.5)202 (76, 384)
Regular bed net usage*1.000§
No1425 (26.5)730 (30.3)210 (82, 386)
Yes3954 (73.5)1676 (69.7)217 (80, 403)
Village<0.001
Kinesamo1854 (34.5)876 (36.4)233 (89, 418)
Maruti1681 (31.3)745 (31.0)174 (64, 350)
Sitabicha1844 (34.3)785 (32.6)231 (90, 421)
  1. Abbreviations: IQR, interquartile range.

    *Regular bed net usage was a person averaging > 5 nights a week sleeping under a bed net.

  2. Total person-months indicates the total number of monthly follow-up visits ending in a symptomatic infection or censoring for full 29 months of follow-up.

    Symptomatic infections were defined using the primary case definition where a participant was P. falciparum-positive by both RDT and qPCR as well as had at least one symptom consistent with malaria during a sick visit.

  3. §Wilcoxon rank sum test with continuity correction and Bonferroni correction for repeated measures.

    Kruskal-Wallis test with Bonferroni correction for repeated measures.

  4. Significant estimates are bolded.

Short-term effect of asymptomatic malaria exposure

In a univariate frailty Cox proportional hazards model, compared to uninfected people, the 1-month crude hazard ratio of symptomatic malaria for participants with asymptomatic infections was 2.69 (95% CI: 2.12 to 3.43). This association was similar in a model adjusted for covariates (adjusted HR [aHR]: 2.61, 95% CI: 2.05 to 3.33) (Table 2, Figure 3A) as well as when using alternative modeling approaches, alternate outcome case definitions, and in sensitivity analyses. This relationship between asymptomatic malaria and subsequent symptomatic illness was not modified by age (p-value = 0.447 by log-likelihood ratio test), because asymptomatic infections were associated with significantly increased likelihoods of subsequent symptomatic malaria in all age categories: <5 years (aHR: 3.77, 95% CI: 2.02 to 7.04), 5 to 15 years (aHR: 2.45, 95% CI: 1.79 to 3.35), and >15 years (aHR: 2.55, 95% CI: 1.57 to 4.15) (Table 3). In contrast, sex did modify this relationship (p-value = 0.006 by log-likelihood ratio test) (Table 3), whereby the risk of symptomatic malaria following asymptomatic infection was lower for males (aHR: 1.76, 95% CI: 1.24 to 2.50) compared to females (aHR: 3.71, 95% CI: 2.62 to 5.24) (Figure 3—figure supplement 3). We observed similar 1-month elevated risks of malaria in asymptomatically infected people when using both the ‘permissive’ (aHR 1.97, 95% CI: 1.63 to 2.40) and the ‘stringent’ (aHR 2.76, 95% CI: 2.11 to 3.62) alternate case definitions for symptomatic malaria (Figure 3—figure supplement 3).

Figure 3 with 4 supplements see all
Adjusted hazard of symptomatic malaria after asymptomatic infections compared to uninfected over various follow-up periods.

(A) Frailty Cox proportional hazards model results comparing having asymptomatic malaria infections versus being uninfected over time and the 1-month hazard of symptomatic malaria. The main model used all eligible participants while the pre-symptomatic model removed monthly follow-up visits that occurred within 14 days prior to a symptomatic malaria infection. Models controlled for covariates participant age, sex, bed net usage, and village. (B) Main model results using primary outcome coding of symptomatic malaria were computed using differing follow-up periods ranging from 1 to 29 months and controlled for covariates participant age, sex, bed net usage, and village.

Table 2
Predicted hazard of symptomatic malaria across follow-up periods.
1-Month aHR (95% CI)3-Month aHR (95% CI)6-Month aHR (95% CI)12-Month aHR (95% CI)29-Month aHR (95% CI)
Main exposure
No infectionRefRefRefRefRef
Asymptomatic infection2.61 (2.05, 3.33)1.64 (1.40, 1.94)1.38 (1.20, 1.58)1.12 (1.00, 1.25)1.11 (1.01, 1.22)
Age
<5 yearsRefRefRefRefRef
5 to 15 years1.37 (0.90, 2.08)1.61 (1.00, 2.61)1.99 (1.07, 3.71)2.37 (0.97, 5.77)2.52 (1.26, 5.01)
>15 years0.56 (0.36, 0.88)0.74 (0.46, 1.21)0.83 (0.44, 1.53)0.88 (0.37, 2.08)0.97 (0.51, 1.84)
Sex
MaleRefRefRefRefRef
Female0.93 (0.70, 1.24)0.84 (0.61, 1.16)0.80 (0.53, 1.20)0.68 (0.38, 1.21)0.63 (0.40, 0.99)
Regular bed net usage*
NoRefRefRefRefRef
Yes1.00 (0.70, 1.43)0.81 (0.55, 1.20)0.70 (0.43, 1.16)0.59 (0.29, 1.21)0.52 (0.30, 0.89)
Village
KinesamoRefRefRefRefRef
Maruti1.08 (0.77, 1.52)1.11 (0.75, 1.64)1.14 (0.69, 1.88)1.13 (0.56, 2.31)1.09 (0.64, 1.85)
Sitabicha0.72 (0.49, 1.05)0.80 (0.53, 1.21)0.76 (0.45, 1.29)0.73 (0.35, 1.51)0.70 (0.40, 1.23)
  1. Abbreviations: CI, confidence interval; aHR, adjusted hazard ratio; Ref, reference.

    *Regular bed net usage was defined as a person averaging > 5 nights a week sleeping under a bed net.

  2. Significant estimates are bolded.

Table 3
Age- and sex-stratified adjusted hazard ratios of time to symptomatic malaria.
ComparisonAge aHR (95% CI)Sex aHR (95% CI)
<5 years5 to 15 years>15 yearsMaleFemale
1-Month main model3.77 (2.02,7.04)2.45 (1.79,3.35)2.55 (1.57,4.15)1.76 (1.24,2.50)3.71 (2.62,5.24)
1-Month pre-symptomatic2.85 (1.19,6.79)1.61 (1.05,2.46)1.90 (0.93,3.86)1.24 (0.75,2.05)2.34 (1.47,3.71)
3-Month main model2.47 (1.59,3.84)1.49 (1.21,1.85)1.69 (1.23,2.32)1.29 (1.01,1.64)2.03 (1.62,2.55)
3-Month pre-symptomatic2.00 (1.20,3.34)1.16 (0.90,1.48)1.35 (0.94,1.93)1.05 (0.79,1.39)1.52 (1.18,1.97)
6-Month main model1.94 (1.34,2.80)1.32 (1.11,1.57)1.31 (1.01,1.70)1.13 (0.93,1.39)1.62 (1.35,1.94)
6-Month pre-symptomatic1.63 (1.08,2.46)1.11 (0.92,1.34)1.10 (0.83,1.46)0.98 (0.78,1.22)1.33 (1.09,1.62)
12-Month main modelNot calculated*Not calculated*Not calculated*1.10 (0.86,1.19)1.21 (1.05,1.41)
12-Month pre-symptomatic1.24 (0.88,1.74)1.00 (0.86,1.17)0.85 (0.68,1.07)0.91 (0.77,1.09)1.04 (0.88,1.22)
29-Month main model1.38 (1.05,1.81)1.16 (1.02,1.32)0.96 (0.81,1.13)1.08 (0.94,1.24)1.14 (1.01,1.30)
29-Month pre-symptomatic1.23 (0.92,1.64)1.06 (0.93,1.21)0.88 (0.74,1.05)Not calculated*Not calculated*
  1. Abbreviations: CI, confidence interval; aHR, adjusted hazard ratio.

    *Not calculated due to data sparsity.

  2. Statistically significant effect measure modification by the log-likelihood ratio test is bolded.

In a subset analysis accounting for potentially pre-symptomatic infections, compared to uninfected people, the risk of symptomatic malaria was increased in those with asymptomatic infections by more than 1.7 times (aHR: 1.77, 95% CI: 1.26 to 2.47) when limited to those with events more than 14 days after exposure ascertainment (Figure 3A). In this subset, we did not observe effect measure modification by participant age or sex (Table 3). The 1-month adjusted risk of symptomatic malaria was not substantially different in models incorporating seasonality (aHR: 2.46, 95% CI: 1.93 to 3.15), the number of prior asymptomatic infections (aHR: 2.60, 95% CI: 2.03 to 3.31), or prior antimalarial treatment (aHR: 2.61, 95% CI: 2.05 to 3.33), nor in a model using the dataset without imputation (aHR: 2.75, 95% CI: 2.05 to 3.66).

Long-term effect of asymptomatic malaria exposure

Next, we assessed the relationship between asymptomatic infection and subsequent symptomatic malaria over longer follow-up periods. Extending the follow-up period led to a diminution in the risk of symptomatic malaria comparing those asymptomatically infected versus uninfected over 3 months (aHR: 1.64, 95% CI: 1.40 to 1.94), 6 months (aHR: 1.38, 95% CI: 1.20 to 1.58), 12 months (aHR: 1.12, 95% CI: 1.00 to 1.25), or 29 months (aHR: 1.11, 95% CI: 1.01 to 1.22) (Table 2, Figure 3B). In the 29-month analysis, this relationship was modified by participant age (p-value < 0.001 by log-likelihood ratio test) with the strongest relationship between asymptomatic infection and future symptomatic malaria in children < 5 years (aHR: 1.38, 95% CI: 1.05 to 1.81), second-strongest in children 5 to 15 years (aHR: 1.16, 95% CI: 1.02 to 1.32), and weakest in adults > 15 years (aHR: 0.96, 95% CI: 0.81 to 1.13) (Table 3, Figure 3—figure supplement 4). Consistent with the 1-month analysis, we observed modification by sex in some models, with females having higher risk for symptomatic disease (Table 3). The limited association between asymptomatic infection and malaria over the 29-month period was also observed when using both the ‘permissive’ (aHR: 1.20, 95% CI: 1.11 to 1.31) and the ‘stringent’ (aHR 1.02, 95% CI: 0.92 to 1.13) alternate case definitions for symptomatic malaria (Figure 3—figure supplement 4).

We assessed for effect modification of the main exposure-outcome relationship by sex and by age in the main and pre-symptomatic models over various periods of follow-up (Table 3). For age, we observed only significant effect modification over the 29-month period, for which the risk of symptomatic malaria following asymptomatic infection was elevated in under-5s (aHR 1.38, 95% CI: 1.05 to 1.81) but not in adults (aHR 0.96, 95% CI: 0.81 to 1.13; p < 0.001 by log-likelihood ratio test). Conversely, for sex, we observed effect modification only at short follow-up period after an asymptomatic infection: over 1 month, the risk of symptomatic infection was lower in males (aHR 1.76, 95% CI: 1.24 to 2.50) than females (aHR 3.71, 95% CI: 2.62 to 5.24; p = 0.006 by log-likelihood ratio test).

Short-term effect of detectability of asymptomatic infections

Owing to the consistently elevated short-term risk of symptomatic malaria in people with asymptomatic infections, we investigated the effect of parasite density in these infections on the risk of subsequent symptomatic malaria within 1 month. Compared to uninfected people, the 1-month hazard of symptomatic malaria was significantly increased by asymptomatic infections of all parasite densities, with the highest risk for those with densities > 1000 parasites/μL (aHR 3.99, 95% CI: 2.41 to 6.62) (Figure 4). This observed increase in the hazard of symptomatic malaria with increasing parasite density was most pronounced among adults >15 years (Figure 4—figure supplement 1); however, children’s likelihood of symptomatic infection did not appear to be influenced by parasite density.

Figure 4 with 1 supplement see all
Association between parasite density of asymptomatic malaria infections and the short-term, 1-month hazard of symptomatic malaria.

Estimates of the 1-month hazard of symptomatic malaria in people with asymptomatic infections compared to being uninfected are presented in separate frailty Cox proportional hazards models that were each restricted to asymptomatic infections meeting parasite density thresholds. Each model compared people with asymptomatic malaria infections meeting the listed density threshold to uninfected people and controlled for covariates participant age, sex, bed net usage, and village.

Discussion

Using a 29-month longitudinal cohort in a high malaria transmission region of Kenya, we investigated the association between asymptomatic P. falciparum infections and the risk of symptomatic malaria. In the short term, compared to uninfected individuals, people of any age with asymptomatic infections were associated with a more than twofold increased hazard of symptomatic malaria within 1 month. This elevated likelihood of symptomatic malaria was associated with asymptomatic infections at all parasite densities. As follow-up time was expanded, the association between asymptomatic infection and the increased risk of subsequent symptomatic malaria remained significant but attenuated. Collectively, our findings that detection of an asymptomatic P. falciparum infection confers an elevated risk of future symptomatic malaria supports the routine treatment of infections even in the absence of symptoms to prevent clinical cases.

Previous studies that detected asymptomatic infections using microscopy also reported an increased short-term hazard of symptomatic illness among children within 9 to 30 days after having an asymptomatic malaria infection (Le Port et al., 2008; Njama-Meya et al., 2004). We built upon these studies by detecting asymptomatic infections using qPCR, a highly sensitive method with a low limit of detection (Taylor et al., 2019), in participants of all ages and similarly found that asymptomatic infections have a high probability of being quickly followed by symptomatic illness. The increased short-term hazard could reflect misclassification of a ‘pre-symptomatic’ infection that progressed to symptoms as an asymptomatic exposure (Njama-Meya et al., 2004). This interpretation is partially supported by the diminished risk observed in a sub-analysis censoring asymptomatic infections that occurred within 14 days prior to a symptomatic event (Figure 3A). The increased hazard of symptomatic malaria could also have been due to the presence of new genotypes in infections (i.e. superinfection), although we previously reported that such newly apparent genotypes were associated with symptoms only in previously uninfected people (Sumner et al., 2021). It is notable that the increased risk of symptomatic malaria following asymptomatic infection was observed in all age groups (Figure 3—figure supplement 2): though children under 5 years were consistently at highest risk, the increased risk in those >15 years was surprising given the presumption that adults develop functional immunity to clinical disease possibly in part from asymptomatic carriage. Our results indicate that asymptomatic infection is associated with an increased short-term risk of malaria irrespective of age.

Elevated risk of symptomatic malaria within 1 month was present for asymptomatic infections of any parasite density. We did observe a dose-response of the risk of symptomatic malaria as a function of parasite density, particularly among adults, but the risk of malaria was increased relative to uninfected people even when they harbored low-density infections. Prior studies observed conflicting relationships between asymptomatic parasite density and subsequent malaria: PCR-positive infections below the limit of detection of microscopy were not associated with subsequent symptomatic malaria in Ugandan children (Nsobya et al., 2004), but were associated with a reduced risk in Malawian children (Buchwald et al., 2019). To our knowledge, our data are the first to analyze a broad range of parasite densities in asymptomatic infections and their association with subsequent symptomatic malaria. Though higher densities were associated with slightly higher risk of symptomatic illness, the elevated risk across all clinically detectable parasitemia does not clearly support risk stratification by detectability for the purposes of preventing clinical disease. Our results indicate that in a high-transmission setting, despite the absence of symptoms, the detection of P. falciparum parasites of any density is significantly associated with an increased risk of malaria in the forthcoming month and suggest that detection modality should not influence a decision to treat.

We observed significant modification of the relationship between asymptomatic infection and symptomatic malaria risk by sex. Specifically, despite an overall lower burden of symptomatic malaria among females that is consistent with prior studies (Houngbedji et al., 2015; Mulu et al., 2013; Newell et al., 2016), we observed that the short-term hazard of symptomatic malaria following an asymptomatic infection was significantly higher among females (aHR 3.71) compared to males (aHR 1.76). To our knowledge, this effect modification by sex has not been previously reported, with prior studies typically including sex as a covariate in models. A recent study (Briggs et al., 2020) highlighted large gaps in knowledge related to sex-based differences in malaria epidemiology, while reporting that Ugandan females of all ages cleared asymptomatic infections at nearly twice the rate of males. That observation, while not directly comparable, is a challenge to reconcile with ours, which suggests that asymptomatic infections in females, compared to males, are more likely to culminate in symptomatic malaria than in natural clearance. Despite similarities in cohort membership, follow-up, and outcome assessment, a key difference may be the far higher transmission intensity in our cohort: the recent application of control measures in Uganda reduced the incidence of malaria in the area by more than 10-fold and the prevalence of PCR-detectable infections more than threefold. Either the recent fall or the current low-transmission intensity may have differentially affected the natural history of infections in that region. Reconciling these sex-based findings and exploring additional impacts relevant to prevention and control will require rigorous assessments of sex as an effect measure modifier in malaria epidemiology.

We observed that the increased risk for symptomatic disease associated with asymptomatic infection weakened as the follow-up length extended from 3 to 29 months, illustrated in both the multi-level models and Kaplan-Meier curves. One possible explanation for this observation could be inherent to the methodology whereby the magnitude of the average hazard ratio decreases as follow-up time increases (Hernán, 2010). Alternatively, it is biologically plausible that the further removed in time an asymptomatic exposure is, the weaker the relationship to disease outcomes, possibly by waning immunity. This is supported by our observation that older children and adults are no longer at increased risk of symptomatic disease by 29 months, although small children still maintain significantly elevated risk even for this extended follow-up period.

We used a novel approach to capture how asymptomatic malaria varied over time. Most previous work used an intention-to-treat approach for asymptomatic infections identified in cross-sectional surveys (Henning et al., 2004; Liljander et al., 2011; Males et al., 2008; Nsobya et al., 2004; Portugal et al., 2017; Sondén et al., 2015; Wamae et al., 2019); however, this method can misclassify person-time if the exposure frequently changes, as happens with asymptomatic infections in high-transmission areas. For previous studies with more frequent asymptomatic sampling, the projects had short follow-up periods (9 to 30 days) (Le Port et al., 2008; Njama-Meya et al., 2004), or coded the exposure as static after an asymptomatic infection occurred (Buchwald et al., 2019). We recorded asymptomatic malaria exposure using a time-varying method proposed by Hernán et al., 2005 that allows participants to change exposure status throughout follow-up, which may capture a more complete view of infection dynamics with lower risk of exposure misclassification. By producing an effect estimate predictive of future risk regardless of prior exposure, this method is less prone to left truncation bias, which can occur with methods that create additive measures of months of exposure. To our knowledge, though this method (Hernán et al., 2005) has been used in studies of cardiovascular or kidney disease (Danaei et al., 2013; Hernán et al., 2008; Secora et al., 2020), it has not before been used to study malaria. Given the frequency of outcome events in high-transmission settings and the complexity of risk factors for them, this approach could be a useful addition to the analytic toolkit to assess time-varying exposures and their association with malaria outcomes.

This study had some limitations that should be considered when weighing the findings. First, asymptomatic infections were only captured at monthly follow-up visits, potentially missing transient asymptomatic infections between visits. By allowing participant exposure to vary over time, we assumed exchangeability between the exposed and unexposed groups. This was mitigated by the observation that approximately 94% of the study population changed exposure status at least once during follow-up. Finally, we estimated parasite densities using molecular methods and only at a single time point, though densities are known to fluctuate during infections. However, this potential bias in density estimations should be random and non-directional, and therefore mitigated by the analysis of over 1600 density measurements in asymptomatic infections.

In conclusion, using a novel exposure coding method and frequent sampling of both children and adults over 29 months, we found that asymptomatic P. falciparum infections were associated with a high likelihood of being shortly followed by symptomatic illness across all ages and parasite densities. These results suggest interventions focus on treating and reducing asymptomatic malaria in high-transmission settings.

Data availability

Data will be shared under the auspices of the Principal Investigators. Investigators and potential collaborators interested in the datasets will be asked to submit a brief concept note and analysis plan. Requests will be vetted by Drs. O'Meara and Taylor and appropriate datasets will be provided through a password protected secure FTPS link. No personal identifying information will be made available to any investigator. Relevant GPS coordinates would only be provided when (1) the planned analysis cannot reasonably be accomplished without them and (2) the release of the coordinates is approved by the Institutional Review Board. A random error in the latitude and longitude of 50-100 meters will be added to each pair of coordinates to protect individual household identities. General de-identified datasets will be prepared that can accommodate the majority of requests. These will be prepared, with documentation, as the data is cleaned for analysis in order to reduce time and resources required to respond to individual requests. Recipients of study data will be asked to sign a data sharing agreement that specifies what the data may be used for (specific analyses), criteria for acknowledging the source of the data, and the conditions for publication. It will also stipulate that the recipient may not share the data with other investigators. Requests for data use must be made directly to the PI and not through third parties.

References

  1. Book
    1. AccessBio
    (2019)
    CareStart Malaria
    Berkshire: APACOR.
  2. Software
    1. R Development Core Team
    (2020) R: A language and environment for statistical computing
    R Foundation for Statistical Computing, Vienna, Austria.

Decision letter

  1. Marcelo U Ferreira
    Reviewing Editor; University of São Paulo, Brazil
  2. Dominique Soldati-Favre
    Senior Editor; University of Geneva, Switzerland
  3. Marcelo U Ferreira
    Reviewer; University of São Paulo, Brazil
  4. Cristian Koepfli
    Reviewer; Biological Sciences, University of Notre Dame

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This study addresses an important topic with major public health implications: whether chronic, asymptomatic parasite carriage is associated with increased or decreased risk of subsequent episodes of clinical malaria in Africa. The authors followed up individuals exposed to intense malaria transmission in western Kenya. They used a novel analytical approach to assess the association between baseline asymptomatic infections and subsequent symptomatic malaria and found that the risk of disease is higher if someone carried asymptomatic Plasmodium falciparum infection at the baseline, compared to uninfected individuals.

Decision letter after peer review:

Thank you for submitting your article "Impact of asymptomatic Plasmodium falciparum infection on the risk of subsequent symptomatic malaria in a longitudinal cohort in Kenya" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Marcelo U Ferreira as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by a Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Cristian Koepfli (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. How was the cohort assembled to support the representativeness of the participants for the general community?

2. Which symptoms were asked about in the self-report, and what was the algorithm used to define a case as symptomatic or asymptomatic, over what time period?

3. Was any participant classified as symptomatically infected at a monthly visit? If so what status were they given at that visit?

4. How many participants were symptomatic, PCR positive but RDT negative, and therefore not treated? How did their clinical course, if any, compare with those who were also RDT positive and therefore treated?

5. Was there any correlation by individuals in the same household?

6. Are you able to control for spatial heterogeneity in risk, that could also help tease out whether superinfection has confounded your results?

7. Please reconsider your suggestion of causality between asymptomatic and symptomatic infections.

8. Were the time periods that occurred after an individual was treated for a symptomatic illness different than before they had an illness? I noted that treatment was not included in the models.

9. Given that some parasite genotyping data are already available for this cohort, could the authors incorporate genotyping data in their analysis to distinguish between persisting infections that became symptomatic vs. superinfection with new strains?

10. The authors ascertain the exposure status at every monthly follow-up visit, thereby treating each of these visits as a new study entry. It remains unclear whether this method is applicable for infectious diseases for which the risk of subsequent infections changes over time. The key question is whether this change is moderate or not over the time period considered. The stratification by age shows substantial changes over five year periods which is not that much more than a 29-month follow up. Please discuss whether these results are robust against the deviations from moderate time dependence.

Reviewer #1:

The "premunition" hypothesis postulares that individuals living in malaria-endemic areas gradually develop a form of clinical immunity characterized by the chronic carriage of low parasite burdens that reduces the risk of superinfection with more virulent strains and protects from full-blown disease (e.g., doi: 10.1590/s0074-02761994000600013). Whether this hypothesis holds in African settings with contrasting levels of transmission intensity remains unknown.

Here, Sumner and colleagues provide evidence that, contrary to the premunition hypothesis, pre-existing asymptomatic infections at the baseline are associated with increased (rather than decreased) short-term risk of clinical disease in a cohort of 268 Kenyans exposed to intense malaria transmission and followed up for 29 months. This finding has major public health implications as it implies that treating asymptomatic infections (that are typically missed by malaria control strategies) may reduce subsequent malaria-related morbidity.

Sumner and colleagues use multivariate frailty Cox proportional hazards models to account for the repeated observations per individual – an interesting approach that has not been previously applied (to my knowledge) to the study of infectious diseases. One possible caveat is that the model does not account/correct for some degree of acquired immunity elicited by a previous infection diagnosed during the follow-up. For example, if the study participant is uninfected at baseline, asymptomatically infected during the first follow-up visit and again uninfected during the second follow-up visit, his/her level of "clinical immunity" (and thus the odds of having clinical disease) may have changed from the baseline (uninfected) to the second visit (currently uninfected, but recently exposed to the parasite). This might be discussed in the manuscript.

The key biological question underlying this study is whether chronically infected but asymptomatic individuals are more likely than their uninfected counterparts to develop clinical disease because pre-existing parasites eventually multiply and parasite density crosses the (individual) "fever threshold", causing malaria-related symptoms, or, alternatively, whether these infected hosts are more likely (e.g., because of increased exposure) to be superinfected with more virulent parasite strains over the next few weeks and therefore develop clinical disease. Surprisingly, this question remains unaddressed in the present manuscript.

However, the same authors have recently reported a detailed molecular analysis of incident P. falciparum infections in this cohort (doi: 10.1093/cid/ciab357). Molecular data suggest that the second hypothesis is correct: incident infections with only novel haplotypes were significantly associated with increased odds of subsequent symptomatic malaria over 14 months of follow-up (doi: 10.1093/cid/ciab357). Integrating these separately reported results in a single publication would render it much more appealing for the broad audience of eLife readers.

In conclusion, Sumner et al. convincingly show that asymptomatic parasite carriage in a high-transmission setting in Kenya is associated with increased subsequent risk of clinical malaria, but the mechanisms underlying this association remain relatively unexplored in this study.

Comments for the authors:

This is a clearly written manuscript that addresses a topic of major public health significance. My only concern refers to the way the interesting results reported here are explored.

The authors use a relatively new approach to survival analysis to show that carriers of asymptomatic malaria infections are more likely than uninfected individuals in the same populations to develop a clinical disease over the next month, although not necessarily over longer periods of follow-up. To this end, they excluded putatively "pre-symptomatic" infections, that were arbitrarily defined as those occurring within 14 days prior to a symptomatic infection. I see no clear justification for this 14-day time window.

I would favor an alternative, more strict approach: excluding infections leading to clinical symptoms only if identical parasite haplotypes persisted since the first parasite detection until the clinical episode. This would provide evidence of "pre-symptomatic" infections. Combining changes in parasitemia over time would make the case even stronger. Quantifying the proportion of subsequent symptomatic infections associated with persisting haplotypes would provide an information of major biological and epidemiological interest and would allow us to test whether the arbitrary time window of 14 days is appropriate to rule out "pre-symptomatic infections". The authors have already genotyped a large proportion of incident infections in cohort participants (doi: 10.1093/cid/ciab357) but do not consider this rich dataset in their analyses.

The exposure status of each study participant (whether asymptomatically infected or not) was ascertained at every monthly follow-up visit and allowed to vary each month. This is an interesting approach that has been previously used in epidemiological studies of non-communicable diseases, but not of infections. The caveat here is that each new infection (either persisting or self-cured) may reduce the subsequent risk of malaria-related disease. Therefore, someone who is uninfected (at baseline), then infected (first follow-up visit) and again uninfected (second follow-up visit) has a similar "exposure status" during the second follow-up visit as someone who remained uninfected since the baseline evaluation, but this (mis)classification ignores acquired immunity gradually developed in response to each infection experienced by cohort participants, which may reduce the subsequent risk of infection and disease.

Not surprisingly, the risk of disease following asymptomatic parasite carriage increases with increasing parasite density (Figure 4), but I wonder whether age-related fever thresholds may be observed in this population. Given that parasite density data are available, I suggest that a further analysis of parasite density levels associated with symptoms across age groups might provide some interesting insights into the biological/immunological mechanisms underlying the main study finding.

Reviewer #2:

1) At least three processes could potentially result in more frequent clinical infection following asymptomatic infection. First, the infection might be presymptomatic. Each clinical infection is preceded by a short period (a few days) of asymptomatic blood-stage infection. Second, chronic asymptomatic infections might increase in density though hitherto unknown processed, and cause clinical illness. Third, asymptomatic infection might be an indicator for higher exposure. Subsequent new infections might then result in clinical disease.

Depending on the scenario, different control interventions are warranted. If asymptomatic infections develop into symptomatic infections, treating these infections will reduce the number of clinical cases. If they are a marker for higher exposure, vector control will be warranted.

The first option was analyzed by a subgroup analysis (asymptomatic infections <14 days before clinical disease excluded from analysis). The results showed a surprisingly large difference between models, indicating that a large number of 'asymptomatic' infections was in fast presymptomatic.

Scenario 2 and 3 could be distinguished by parasite genotyping, as a new infections will most likely be a different clone. Thus, genotyping would add substantially to the interpretation of the data. Given that the authors already have published genotyping results from a similar study (Sumner et al., 2021), this should be straightforward. It might also help to understand better why some previous studies have found similar results as this one, while other studies have found contrasting results.

2) Seasonality was not considered in the analysis, but might be important to understand the observations. For example, the finding that "a difference in the time to symptomatic malaria in the first few months post asymptomatic infection but not long-term" could be confounded by the fact that most transmission occurs in the wet season, and thus that the probability for both asymptomatic and clinical infection is higher in the wet season.

3) Lines 52-54: It would be good to mention whether these estimates are based on microscopy or PCR. It would also be good to introduce the concept of supatent infections in the introduction, and discuss the possibility that even by PCR some infections were missed.

4) Lines 188-189: While I like brevity, this section should be expanded, given its importance for the study. How many participants were never infected? How often was someone infected at a single time point? What was the duration of infections? Did you observe the sequence infected-non infected-infected, pointing to possible undetected infections at certain time points (in particular if the same clone was observed)?

Likewise, given that age was an important variable, please provide more details on the age of clinical patients. A figure showing age trends in asymptomatic and clinical infections would be helpful.

Reviewer #3:

Sumner et al. report on a prospective cohort study of 254 participants aged 1 to 85 years living in three villages in Western Kenya. Participants gave monthly blood samples for testing by polymerase chain reaction (PCR) for parasite DNA and recorded any symptoms of illness during the previous 30 days from June 2017 through November 2019. Patients who reported symptoms of illness at any time were tested for malaria by rapid diagnostic test (RDT) and had blood samples taken for analysis by PCR. The study assessed whether asymptomatic individuals carrying malaria parasites were more likely, over the following time period, to have a symptomatic episode of malaria than people not infected with malaria. Using an analysis that treated each monthly visit as a new entry in the study, the time to symptomatic malaria or the end of the study was calculated and attributed to the status of the participant in that month (uninfected or infected but asymptomatic). Repeated measures on each participant were accounted for in the models using a random-effect measure at the level of participant.

The researchers found that participants with asymptomatic infections developed symptomatic infections more quickly than those uninfected (median time to symptomatic malaria: 173 days [interquartile range, IQR: 49, 399] vs. 230 days [IQR: 98, 402], respectively). There was a statistically significant difference in the hazard ratio comparing symptomatic malaria between asymptomatic infected and uninfected participants for periods from 1- to 6-months, but not at 12 months, after adjusting for covariates, with a decline in the hazard ratio over longer follow-up periods.

These data add to the ongoing discussion about the importance and clinical relevance of asymptomatic malaria infections. There is widespread recognition that a substantial proportion of the human malaria parasite reservoir in malaria-endemic areas is found in people who do are asymptomatic, paucisymptomatic or afebrile. There is a debate, however, as to whether these infections are: a) the result of previous infections on their way to being resolved; b) chronic infections resulting from partial immunity to malaria; or c) new onset patent parasitemia appearing prior to the onset of symptoms. Additionally, the long-term outcomes of these infections, and their potential deleterious effects, have been the subject of debate. Such infections could: a) disappear without intervention; b) remain for long periods of time without causing harm; c) remain for long periods of time causing mild intermittent symptoms; or d) produce symptoms of illness that prompt healthcare seeking and treatment.

The authors of this paper have taken great care in the design and analysis of the complex data gathered during this study, some aspects of which have been reported elsewhere (O'Meara et al. JID 2020 221;1176-1184). However, the authors' conclusion that asymptomatic infections should be treated with antimalarials because the asymptomatic infections themselves increase the risk of future symptomatic malaria is not fully supported by the evidence. The findings in this study could also result from superinfection, that is, if individuals with asymptomatic infections are more likely to be infected again, a new parasite strain could provoke symptoms of illness. If an asymptomatic infection serves as a marker of living in an area with a higher force of infection, then it would follow that such individuals are at greater risk of multiple infections. The development of symptoms associated with a new infection by a different parasite strain would, therefore, not be affected by treatment of a previous asymptomatic infection.

There is already some suggestion in the report that symptomatic infections were associated with a higher force of infection. The study reported that time to symptomatic malaria was shorter for participants living in the village of Maruti. A previous paper reporting on entomologic indices in the study area showed that this village had the highest density of malaria mosquito vectors per household, and effectively the highest entomologic inoculation rate, compared to the other villages (O'Meara et al. JID 2020 221;1176-1184). Additionally, the previous paper reported that 41 individuals had more than 1 symptomatic infection, suggesting heterogeneity in transmission and force of infection.

Additionally, it was surprising to see that age, which is generally a proxy for greater lifetime exposure to malaria and therefore a higher degree of partial immunity, did not modify the association between asymptomatic infection and subsequent symptomatic infection. This finding could also be explained by the symptomatic infections arising from reinfection with new strains of parasites more likely to cause symptoms.

The strengths of the report lie in the use of a novel approach to time-varying exposure status (asymptomatic infection vs. uninfected) that avoided misclassification of person-time; careful statistical analysis of the data, including excluding post-baseline potential pre-symptomatic episodes; and relatively complete follow-up of a modest cohort of participants.

The weaknesses in this paper, however, limit the paper's ability to conclude that asymptomatic infections should be treated to avoid symptomatic illness. In addition to the points mentioned above, the analysis did not control for the timing of episodes. Although not reported in this paper, the previous paper by this group suggests that the proportion of the cohort that had asymptomatic malaria infections at baseline was >50%. These baseline infections appear to be counted as asymptomatic infections although the length of time that these individuals had been infected was unknown. The authors could clarify whether these baseline asymptomatic infections were more likely to become symptomatic than asymptomatic infections that developed later, perhaps suggesting that many of them were presymptomatic.

Comments for the authors:

The paper is very well written and conceived, but I do not agree that the main conclusion is supported by the data.

Comments on the statistical analysis:

1. Line 105-107: missed monthly visits got an exposure status equal to the previous month's value – was any sensitivity analysis done?

2. Line 107-109: why are people considered lost to follow-up and censored at the time of the imputed monthly visit; I would argue imputation is best not done in this situation

3. Line 136-138: How do the authors account for repeated measures using a Bonferroni correction?

4. Line 143-144; Equation 1: What is ϵi in this formula? What are the underlying distributional assumptions for αi and were these assumptions challenged? Please clarify notation used.

5. Individual are clustered in households (38 in total); was this taken into account in the analysis? This would be very relevant for malaria. A nested frailty approach would be useful here.

6. Throughout the text it would help to clarify which HR are referred to: conditional or unconditional (wrt the frailty). This can be briefly mentioned so that repetition is not needed.

https://doi.org/10.7554/eLife.68812.sa1

Author response

Essential revisions:

1. How was the cohort assembled to support the representativeness of the participants for the general community?

The cohort was assembled using radial sampling of 12 households per village for three villages within Webuye, Western Kenya. The first household in each village was randomly selected. The three villages were geographically close and chosen due to their high malaria prevalence in a previous cross-sectional study in the area [Mean P. falciparum prevalence in 2013: Kinesamo (18.4%), Maruti (20.8%), Sitabicha (22.8%), Obala AA PloS One 2015;10(7)]. We have added some text to the methods to describe this:

“The cohort was assembled using radial sampling of 12 households per village for three villages with high malaria transmission. The first household in each village was randomly selected. Two households moved during follow-up and were replaced.”

2. Which symptoms were asked about in the self-report, and what was the algorithm used to define a case as symptomatic or asymptomatic, over what time period?

We have clarified this by adding to the Exposure and outcome ascertainment section of the manuscript: “The main exposure was an asymptomatic P. falciparum infection during monthly active case detection assessments, defined as P. falciparum-positive by qPCR in a person lacking symptoms. People who were P. falciparum-negative by qPCR during monthly visits were considered uninfected. We defined symptomatic P. falciparum infection as the current presence of at least one symptom consistent with malaria during a sick visit (i.e. fever, aches, vomiting, diarrhea, chills, cough or congestion) and P. falciparum-positive by both RDT and qPCR.” Additional details surrounding exposure and outcome ascertainment are provided in the supplement as described in #3 below.

3. Was any participant classified as symptomatically infected at a monthly visit? If so what status were they given at that visit?

Yes, and this information is now described in the supplement under the header Exposure and outcome ascertainment:

“Some participants were classified as symptomatically infected at a monthly visit through passive detection of symptoms; this occurred if a study team member conducting a monthly visit was approached by a participant reporting malaria-like symptoms. […] If that person did not meet our case definition for symptomatic malaria, then they were removed from follow-up for that month and re-entered for follow-up in the following month.”

4. How many participants were symptomatic, PCR positive but RDT negative, and therefore not treated? How did their clinical course, if any, compare with those who were also RDT positive and therefore treated?

In the primary analysis data set, we included 0 participants who were symptomatic, PCR positive, and RDT negative, because they did not meet our case definition for symptomatic malaria. Looking back at the original data set prior to any censoring criteria being applied, there were 6016 observations collected through monthly and sick visits and 183/6016 (3.0%) of them met the criteria of being symptomatic, PCR positive, and RDT negative. Subsetting the data set to only sick visits, 183/983 (18.6%) of sick visits were symptomatic, PCR positive, and RDT negative. Because this number was fairly few across a large number of participants and 29 months of observation, we have limited ability to make inferences about the natural history of malaria in RDT-negative/PCR-positive people; however, we did conduct a secondary analysis where we computed the hazard of symptomatic malaria using a secondary permissive definition for symptomatic malaria (PCR-positive and having at least one malaria-like symptom) to assess if alternative symptomatic malaria case definitions influenced results. Using the secondary permissive case definition for symptomatic malaria, we observed overall similar results to those produced using the primary case definition. We have added this to the main text:

“We observed similar 1-month elevated risks of malaria in asymptomatically-infected people when using both the “permissive” [aHR 1.97, 95% CI 1.63 to 2.40] and the “stringent” [aHR 2.76, 95% CI 2.11 to 3.62] alternate case definitions for symptomatic malaria (Figure 3—figure supplement 3).”

5. Was there any correlation by individuals in the same household?

We chose to only include a random intercept at the individual and not household level due to the little impact correlation at the household level had on the relationship between asymptomatic/uninfected at monthly visits and subsequent symptomatic malaria. We assessed the possibility of using a nested frailty approach with random intercepts at the individual and household levels. To do so, we re-computed the frailty Cox proportional hazards model described in Equation 1 in the main text using a random intercept at the individual level and an additional random intercept at the household level. Results were similar for the 1-month hazard of symptomatic malaria for those asymptomatically-infected versus uninfected when a random intercept at the household level was included [aHR: 2.64, 95% CI: 2.07 to 3.37] versus excluded [aHR: 2.61, 95% CI: 2.05 to 3.33]. Because including a household level random intercept had little impact on model results, we excluded it from analyses to be able to have model convergence to test effect measure modification of the relationship by age and sex.

6. Are you able to control for spatial heterogeneity in risk, that could also help tease out whether superinfection has confounded your results?

We controlled for spatial heterogeneity in risk by including village fixed effects in the models. We do not expect spatial risk to be substantially different within and between villages due to the radial sampling of households, close proximity of villages to each other (all less than 11 km apart), and similar high malaria prevalence across the three villages [Mean P. falciparum prevalence in 2013: Kinesamo (18.4%), Maruti (20.8%), Sitabicha (22.8%), Obala AA PloS One 2015;10(7)].

7. Please reconsider your suggestion of causality between asymptomatic and symptomatic infections.

We have relaxed the language in the main text to focus more on the association between asymptomatic and symptomatic infections instead of a causal relationship. To do so, we changed the casual language. For example, the Abstract now reads “Compared to being uninfected, asymptomatic infections were associated with an increased 1-month likelihood of symptomatic malaria [adjusted Hazard Ratio (HR):2.61, 95%CI:2.05–3.33]…” We made similar changes in the Introduction, Results, and Discussion, each of which are tracked in the accompanying documents.

8. Were the time periods that occurred after an individual was treated for a symptomatic illness different than before they had an illness? I noted that treatment was not included in the models.

To test if prior treatment during the study period influenced results, we re-computed the frailty Cox proportional hazards model in Equation 1 with a covariate for prior treatment. To do so, we reran the frailty Cox proportional hazards model in Equation 1 with an additional covariate representing prior receipt of antimalarial treatment. This variable was coded dichotomously as having received study-prescribed antimalarials up until that monthly visit or not. For example, a person was coded as having not received study-prescribed antimalarials up until their first symptomatic infection, but afterward were coded as receiving treatment from that point forward in follow-up. Compared to primary models, antimalarial treatment had minimal effect on the hazard of symptomatic malaria when asymptomatically infected versus uninfected in the short-term [1-month aHR: 2.61, 95% CI: 2.05 to 3.33] but made results largely null in the long-term [29-month aHR: 1.04, 95% CI: 0.95 to 1.15].

We have added to the methods section Sensitivity analyses:

“For prior antimalarial treatment, we included a variable coded dichotomously as having received study-prescribed antimalarials up until that monthly visit or not; a person was coded as having not received study-prescribed antimalarials up until their first symptomatic infection, but afterward were coded as receiving treatment from that point forward in follow-up,” and the Results the general statement that “This association was similar in a model adjusted for covariates [adjusted HR (aHR): 2.61, 95% CI: 2.05 to 3.33] (Table 2, Figure 3A) as well as when using alternative modeling approaches, alternate outcome case definitions, and in sensitivity analyses.”

9. Given that some parasite genotyping data are already available for this cohort, could the authors incorporate genotyping data in their analysis to distinguish between persisting infections that became symptomatic vs. superinfection with new strains?

Unfortunately, parasite genotype data are unavailable for more than half of the months of observation that we analyze here. Our recent analysis of parasite genotypes used data from the first phase of this cohort [Sumner KM Clin Infect Dis 2021;ciab357], and we supplemented these 14 months of phase 1 with an additional 15 months from phase 2 for this analysis.

Results from that paper [Sumner KM Clin Infect Dis 2021;ciab357] suggested an association between the presence of new haplotypes in incident infections and an increased likelihood of symptomatic malaria; however, this relationship was overall null for persistent infections. We have referenced some of these results into the discussion:

“The increased short-term hazard could reflect misclassification of a “pre-symptomatic” infection that progressed to symptoms as an asymptomatic exposure (Njama-Meya et al., 2004). […] The increased hazard of symptomatic malaria could also have been due to the presence of new genotypes in infections (i.e. superinfection), although we previously reported that such newly-apparent genotypes were associated with symptoms only in previously-uninfected people (Sumner et al., 2021).”

10. The authors ascertain the exposure status at every monthly follow-up visit, thereby treating each of these visits as a new study entry. It remains unclear whether this method is applicable for infectious diseases for which the risk of subsequent infections changes over time. The key question is whether this change is moderate or not over the time period considered. The stratification by age shows substantial changes over five year periods which is not that much more than a 29-month follow up. Please discuss whether these results are robust against the deviations from moderate time dependence.

To account for the gradual acquisition of immunity in cohort participants and subsequently time dependence, we have re-computed the models including a variable for the number of prior asymptomatic infections a person experienced in the study up until each monthly visit. To assess how the number of prior asymptomatic infections a person had during the study period could have influenced results, we re-computed the frailty Cox proportional hazards model in Equation 1 with a variable added for the number of prior asymptomatic infections each person had in the study up until each monthly follow-up visit. The covariate for number of prior infections was included as a continuous number in the model. We observed that the number of prior asymptomatic infections had little impact on the hazard of symptomatic malaria comparing those asymptomatically infected to uninfected in both the short-term [1-month aHR: 2.60, 95% CI: 2.03 to 3.31] and long-term [29-month aHR: 1.29, 95% CI: 1.17 to 1.42] compared to the primary models” Overall, we did not observe a large difference in the short-term or long-term results when accounting for the number of prior infections a person experienced, suggesting that the results were robust to deviations from time dependence.

We have added to the methods section Sensitivity analyses:

“For the number of prior infections, we included in the model as a covariate the number of prior infections as a continuous number,” and to the Results the general statement that “This association was similar in a model adjusted for covariates [adjusted HR (aHR): 2.61, 95% CI: 2.05 to 3.33] (Table 2, Figure 3A) as well as when using alternative modeling approaches, alternate outcome case definitions, and in sensitivity analyses.”

Reviewer #1:

[…] Comments for the authors:

This is a clearly written manuscript that addresses a topic of major public health significance. My only concern refers to the way the interesting results reported here are explored.

The authors use a relatively new approach to survival analysis to show that carriers of asymptomatic malaria infections are more likely than uninfected individuals in the same populations to develop a clinical disease over the next month, although not necessarily over longer periods of follow-up. To this end, they excluded putatively "pre-symptomatic" infections, that were arbitrarily defined as those occurring within 14 days prior to a symptomatic infection. I see no clear justification for this 14-day time window.

I would favor an alternative, more strict approach: excluding infections leading to clinical symptoms only if identical parasite haplotypes persisted since the first parasite detection until the clinical episode. This would provide evidence of "pre-symptomatic" infections. Combining changes in parasitemia over time would make the case even stronger. Quantifying the proportion of subsequent symptomatic infections associated with persisting haplotypes would provide an information of major biological and epidemiological interest and would allow us to test whether the arbitrary time window of 14 days is appropriate to rule out "pre-symptomatic infections". The authors have already genotyped a large proportion of incident infections in cohort participants (doi: 10.1093/cid/ciab357) but do not consider this rich dataset in their analyses.

We have updated the manuscript text to reflect why the 14 day (2 week) time window was chosen for the pre-symptomatic subset analysis:

“The time frame for identifying potentially pre-symptomatic infections was chosen for consistency with previous work studying time to symptomatic malaria (Buchwald et al., 2019).”

Unfortunately parasite genotype data are only available for less than half of the time period of this cohort. Additionally, incorporation of these complex deep-sequenced genotype data would vastly expand the scope and impact of this report, which is focused on the clinically-actionable aspects of the natural history of asymptomatic infections.

The exposure status of each study participant (whether asymptomatically infected or not) was ascertained at every monthly follow-up visit and allowed to vary each month. This is an interesting approach that has been previously used in epidemiological studies of non-communicable diseases, but not of infections. The caveat here is that each new infection (either persisting or self-cured) may reduce the subsequent risk of malaria-related disease. Therefore, someone who is uninfected (at baseline), then infected (first follow-up visit) and again uninfected (second follow-up visit) has a similar "exposure status" during the second follow-up visit as someone who remained uninfected since the baseline evaluation, but this (mis)classification ignores acquired immunity gradually developed in response to each infection experienced by cohort participants, which may reduce the subsequent risk of infection and disease.

Addressed in Essential revisions #10 above.

Not surprisingly, the risk of disease following asymptomatic parasite carriage increases with increasing parasite density (Figure 4), but I wonder whether age-related fever thresholds may be observed in this population. Given that parasite density data are available, I suggest that a further analysis of parasite density levels associated with symptoms across age groups might provide some interesting insights into the biological/immunological mechanisms underlying the main study finding.

As suggested, we did an additional analysis assessing the 1-month hazard of symptomatic malaria comparing people with asymptomatic infections with parasite densities above a series of thresholds to people who were uninfected, stratified by people’s age groups (<5 years, 5-15 years, and >15 years). Among adults >15 years, we still saw a positive association between increasing parasite density in asymptomatic infections and a higher 1-month hazard of symptomatic malaria (see Figure 4—figure supplement 1); however, parasite density had less of an effect among children. We have added this analysis to the main text:

“As an additional analysis, we repeated this process for each parasite density threshold stratified by participant age (<5 years, 5-15 years, >15 years). This observed increase in the hazard of symptomatic malaria with increasing parasite density was most pronounced among adults >15 years (Figure 4—figure supplement 1); however, children’s likelihood of symptomatic infection did not appear to be influenced by parasite density.”

Reviewer #2:

1) Seasonality was not considered in the analysis, but might be important to understand the observations. For example, the finding that "a difference in the time to symptomatic malaria in the first few months post asymptomatic infection but not long-term" could be confounded by the fact that most transmission occurs in the wet season, and thus that the probability for both asymptomatic and clinical infection is higher in the wet season.

To investigate the effect of seasonality on our results, we re-computed the models including seasonality as a covariate. To investigate how seasonality could have affected results, we re-computed the frailty Cox proportional hazards model in Equation 1 while including a variable for seasonality. We classified monthly visits that occurred any time from May to October as the high transmission season and from November to April as the low transmission season, based on the region’s rainy seasons. With seasonality in the model, a high hazard of symptomatic malaria was observed among those that were asymptomatically infected compared to uninfected at monthly visits [1-month aHR: 2.46, 95% CI: 1.93 to 3.15; 29-month aHR: 1.14, 95% CI: 1.04 to 1.26]; these results were similar to the main analysis presented in the manuscript using Equation 1, suggesting that the increased hazard of symptomatic malaria following asymptomatic infection was not solely attributable to seasonality.

We have added to the methods section Sensitivity analyses:

“For seasonality, we classified monthly visits that occurred any time from May to October as the high transmission season and from November to April as the low transmission season, based on the region’s rainy seasons.,” and to the Results the general statement that “This association was similar in a model adjusted for covariates [adjusted HR (aHR): 2.61, 95% CI: 2.05 to 3.33] (Table 2, Figure 3A) as well as when using alternative modeling approaches, alternate outcome case definitions, and in sensitivity analyses.”

2) Lines 52-54: It would be good to mention whether these estimates are based on microscopy or PCR. It would also be good to introduce the concept of supatent infections in the introduction, and discuss the possibility that even by PCR some infections were missed.

We have now added the specific malaria diagnostics used for the estimates:

“In 2015, a geo-spatial meta-analysis estimated a continent-wide prevalence of asymptomatic P. falciparum in children aged 2-10 years old of 24% based on microscopy and rapid diagnostic test results (RDT) (Snow et al., 2017).”

3) Lines 188-189: While I like brevity, this section should be expanded, given its importance for the study. How many participants were never infected? How often was someone infected at a single time point? What was the duration of infections? Did you observe the sequence infected-non infected-infected, pointing to possible undetected infections at certain time points (in particular if the same clone was observed)?

We have added some details to the first paragraph of results describing participant follow-up: “… person-months of asymptomatic malaria exposure; the median total months of asymptomatic exposure for a participant was 9 (IQR: 5, 17). Exposure status frequently changed for participants and remained constant for only 16 (6.2%) people across follow-up; 4 people were asymptomatically infected for the entirety of follow-up and only 12 people were never infected (Figure 2A).”

Likewise, given that age was an important variable, please provide more details on the age of clinical patients. A figure showing age trends in asymptomatic and clinical infections would be helpful.

We have added more detail about the age of the participants to the main text and have added Figure 3—figure supplement 2 illustrating the age distribution: “… and a median age of 13 years (range: 1, 85) (Figure 3—figure supplement 2).”

Reviewer #3:

Comments for the authors:

The paper is very well written and conceived, but I do not agree that the main conclusion is supported by the data.

Comments on the statistical analysis:

1. Line 105-107: missed monthly visits got an exposure status equal to the previous month's value – was any sensitivity analysis done?

Our primary models impute missed monthly visits to have an exposure status equal to the previous month’s value. We chose this approach based on a previous paper that had employed this approach to study asymptomatic malaria over time [Nguyen Lancet Infect Dis 2018;18(5)]. Imputing missed monthly visits added 715 observations to the primary analysis data set, with imputed monthly visits making up approximately 13% of the observations in the data set (715/5379).

We have added a sensitivity analysis to the manuscript where we repeated the 1 and 29-month hazard of symptomatic malaria analyses using the original data set without imputation for missed monthly visits. We found that results produced using the data set without imputation [1-month aHR: 2.75, 95% CI: 2.05 to 3.66; 29-month aHR: 1.40, 95% CI: 1.22 to 1.61] were similar to the results presented in the manuscript where imputation was performed [1-month aHR: 2.61, 95% CI: 2.05 to 3.33; 29-month aHR: 1.11, 95% CI: 1.01 to 1.22]; notably, imputation produced estimates closer to the null of 1, suggesting that bias from imputation was towards the null.

We have added to the Exposure and outcome ascertainment section “A sensitivity analysis was conducted for imputation using a dataset without imputation for missed monthly visits,” and to the Results “This association was similar in a model adjusted for covariates [adjusted HR (aHR): 2.61, 95% CI: 2.05 to 3.33] (Table 2, Figure 3A) as well as when using alternative modeling approaches, alternate outcome case definitions, and in sensitivity analyses.”

2. Line 107-109: why are people considered lost to follow-up and censored at the time of the imputed monthly visit; I would argue imputation is best not done in this situation.

Addressed above.

3. Line 136-138: How do the authors account for repeated measures using a Bonferroni correction?

We applied the Bonferroni correction to all the table p-values for Table 1 by multiplying each p-value by 29, which was the maximum amount of follow-up visits (and repeated measures) each person could have in the study.

4. Line 143-144; Equation 1: What is ϵi in this formula? What are the underlying distributional assumptions for αi and were these assumptions challenged? Please clarify notation used.

We have clarified the definitions for these terms in the main text:

“We allowed the main exposure to vary each month based on the monthly follow-up visit infection status (m), and included a random intercept at the participant level (αi) to account for potential correlated intra-individual outcomes. A log-normal distribution was used for the random effect ϵi. represented the model’s error term.”

5. Individuals are clustered in households (38 in total); was this taken into account in the analysis? This would be very relevant for malaria. A nested frailty approach would be useful here.

Addressed in Essential revisions #5 above.

6. Throughout the text it would help to clarify which HR are referred to: conditional or unconditional (wrt the frailty). This can be briefly mentioned so that repetition is not needed.

Throughout the main text and supplement, we have updated the HR abbreviation to indicate whether the unconditional, crude HR is being reported (cHR) or the conditional, adjusted HR is being reported (aHR). Models that produced an aHR always included a random intercept at the individual level.

https://doi.org/10.7554/eLife.68812.sa2

Article and author information

Author details

  1. Kelsey M Sumner

    1. Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, United States
    2. Division of Infectious Diseases, School of Medicine, Duke University, Durham, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
  2. Judith N Mangeni

    School of Public Health, College of Health Sciences, Moi University, Eldoret, Kenya
    Contribution
    Data curation, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Andrew A Obala

    School of Medicine, College of Health Sciences, Moi University, Eldoret, Kenya
    Contribution
    Conceptualization, Supervision, Investigation, Methodology, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Elizabeth Freedman

    Division of Infectious Diseases, School of Medicine, Duke University, Durham, United States
    Contribution
    Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  5. Lucy Abel

    Academic Model Providing Access to Healthcare, Moi Teaching and Referral Hospital, Eldoret, Kenya
    Contribution
    Supervision, Investigation, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  6. Steven R Meshnick (deceased)

    Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, United States
    Contribution
    Conceptualization, Supervision
    Competing interests
    No competing interests declared
  7. Jessie K Edwards

    Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, United States
    Contribution
    Formal analysis, Supervision, Writing - review and editing
    Competing interests
    No competing interests declared
  8. Brian W Pence

    Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, United States
    Contribution
    Supervision, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  9. Wendy Prudhomme-O'Meara

    1. Division of Infectious Diseases, School of Medicine, Duke University, Durham, United States
    2. School of Public Health, College of Health Sciences, Moi University, Eldoret, Kenya
    3. Duke Global Health Institute, Duke University, Durham, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing - review and editing
    Contributed equally with
    Steve M Taylor
    Competing interests
    No competing interests declared
  10. Steve M Taylor

    1. Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, United States
    2. Division of Infectious Diseases, School of Medicine, Duke University, Durham, United States
    3. Duke Global Health Institute, Duke University, Durham, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Methodology, Project administration, Writing - review and editing
    Contributed equally with
    Wendy Prudhomme-O'Meara
    For correspondence
    steve.taylor@duke.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2783-0990

Funding

National Institute of Allergy and Infectious Diseases (R21AI126024)

  • Wendy Prudhomme-O'Meara

National Institute of Allergy and Infectious Diseases (R01AI146849)

  • Wendy Prudhomme-O'Meara
  • Steve M Taylor

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We are very appreciative of the Webuye study participants for their participation in this study. We also thank the project manager and field technicians in Kenya for their fastidious work: J Kipkoech Kirui, I Khaoya, L Marango, E Mukeli, E Nalianya, J Namae, L Nukewa, E Wamalwa, and A Wekesa. We thank M Emch (of the University of North Carolina at Chapel Hill) for his analysis considerations and A Nantume (of Duke University) for laboratory sample processing. This work was supported by NIAID (R21AI126024 to WPO and R01AI146849 to WPO and SMT).

Ethics

Human subjects: The study was approved by institutional review boards of Moi University (2017/36), Duke University (Pro00082000), and the University of North Carolina at Chapel Hill (19-1273). All participants or guardians provided written informed consent, and those over age 8 provided additional assent.

Senior Editor

  1. Dominique Soldati-Favre, University of Geneva, Switzerland

Reviewing Editor

  1. Marcelo U Ferreira, University of São Paulo, Brazil

Reviewers

  1. Marcelo U Ferreira, University of São Paulo, Brazil
  2. Cristian Koepfli, Biological Sciences, University of Notre Dame

Publication history

  1. Received: March 26, 2021
  2. Accepted: July 20, 2021
  3. Accepted Manuscript published: July 23, 2021 (version 1)
  4. Version of Record published: August 4, 2021 (version 2)

Copyright

© 2021, Sumner et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 486
    Page views
  • 96
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Epidemiology and Global Health
    2. Genetics and Genomics
    Mohd Anisul et al.
    Research Article Updated

    Background:

    The virus SARS-CoV-2 can exploit biological vulnerabilities (e.g. host proteins) in susceptible hosts that predispose to the development of severe COVID-19.

    Methods:

    To identify host proteins that may contribute to the risk of severe COVID-19, we undertook proteome-wide genetic colocalisation tests, and polygenic (pan) and cis-Mendelian randomisation analyses leveraging publicly available protein and COVID-19 datasets.

    Results:

    Our analytic approach identified several known targets (e.g. ABO, OAS1), but also nominated new proteins such as soluble Fas (colocalisation probability >0.9, p=1 × 10-4), implicating Fas-mediated apoptosis as a potential target for COVID-19 risk. The polygenic (pan) and cis-Mendelian randomisation analyses showed consistent associations of genetically predicted ABO protein with several COVID-19 phenotypes. The ABO signal is highly pleiotropic, and a look-up of proteins associated with the ABO signal revealed that the strongest association was with soluble CD209. We demonstrated experimentally that CD209 directly interacts with the spike protein of SARS-CoV-2, suggesting a mechanism that could explain the ABO association with COVID-19.

    Conclusions:

    Our work provides a prioritised list of host targets potentially exploited by SARS-CoV-2 and is a precursor for further research on CD209 and FAS as therapeutically tractable targets for COVID-19.

    Funding:

    MAK, JSc, JH, AB, DO, MC, EMM, MG, ID were funded by Open Targets. J.Z. and T.R.G were funded by the UK Medical Research Council Integrative Epidemiology Unit (MC_UU_00011/4). JSh and GJW were funded by the Wellcome Trust Grant 206194. This research was funded in part by the Wellcome Trust [Grant 206194]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

    1. Epidemiology and Global Health
    Lawrence Lubyayi et al.
    Research Article

    Background:

    Lack of early infection-exposure has been associated with increased allergy-related disease (ARD) susceptibility. In tropical Africa, little is known about which infections contribute to development of ARDs, and at which time.

    Methods:

    We used latent class analysis to characterise the early infection-exposure of participants in a Ugandan birth cohort and assessed ARDs in later childhood.

    Results:

    Of 2345 live births, 2115 children (90%) had data on infections within the first year of life while 1179 (50%) had outcome data at 9 years. We identified two latent classes of children based on first-year infection-exposure. Class 1 (32% membership), characterised by higher probabilities for malaria (80%), diarrhoea (76%), and lower respiratory tract infections (LRTI) (22%), was associated with lower prevalence of wheeze, eczema, rhinitis, and Dermatophagoides skin prick test (SPT) positivity at 9 years. Based on 5-year cumulative infection experience, class 1 (31% membership), characterised by higher probabilities for helminths (92%), malaria (79%), and LRTI (45%), was associated with lower probabilities of SPT positivity at 9 years.

    Conclusions:

    In this Ugandan birth cohort, early childhood infection-exposure, notably to malaria, helminths, LRTI, and diarrhoea, is associated with lower prevalence of atopy and ARDs in later childhood.

    Funding:

    This work was supported by several funding sources. The Entebbe Mother and Baby Study (EMaBS) was supported by the Wellcome Trust, UK, senior fellowships for AME (grant numbers 064693, 079110, 95778) with additional support from the UK Medical Research Council. LL is supported by a PhD fellowship through the DELTAS Africa Initiative SSACAB (grant number 107754). ELW received funding from MRC Grant Reference MR/K012126/1. SAL was supported by the PANDORA-ID-NET Consortium (EDCTP Reg/Grant RIA2016E-1609). HM was supported by the Wellcome’s Institutional Strategic Support Fund (grant number 204928/Z/16/Z).