Epidemiology and Global Health

Evidence from a natural experiment that malaria parasitemia is pathogenic in retinopathy-negative cerebral malaria

University of Pennsylvania, United States
Michigan State University, United States
Blantyre Malaria Project, Malawi
University of Liverpool, United Kingdom
Royal Liverpool University Hospital, United Kingdom
University of California, San Francisco, United States
Malawi-Liverpool-Wellcome Trust Clinical Research Programme, Malawi

Short Report Jun 7, 2017

Cited 14
Views 599
Annotations

Cite this article as: eLife 2017;6:e23699 doi: 10.7554/eLife.23699

Article
Figures and data
Abstract
eLife digest
Introduction
Results
Discussion
Materials and methods
Appendix 1
Data availability
References
Decision letter
Author response
Article and author information
Metrics

Abstract

Cerebral malaria (CM) can be classified as retinopathy-positive or retinopathy-negative, based on the presence or absence of characteristic retinal features. While malaria parasites are considered central to the pathogenesis of retinopathy-positive CM, their contribution to retinopathy-negative CM is largely unknown. One theory is that malaria parasites are innocent bystanders in retinopathy-negative CM and the etiology of the coma is entirely non-malarial. Because hospitals in malaria-endemic areas often lack diagnostic facilities to identify non-malarial causes of coma, it has not been possible to evaluate the contribution of malaria infection to retinopathy-negative CM. To overcome this barrier, we studied a natural experiment involving genetically inherited traits, and find evidence that malaria parasitemia does contribute to the pathogenesis of retinopathy-negative CM. A lower bound for the fraction of retinopathy-negative CM that would be prevented if malaria parasitemia were to be eliminated is estimated to be 0.93 (95% confidence interval: 0.68, 1).

https://doi.org/10.7554/eLife.23699.001

eLife digest

Malaria is a life-threatening disease caused by a parasite that is transferred between people by infected mosquitoes. Most infected individuals suffer flu-like symptoms, but in rare cases malaria can affect the brain, resulting in brain damage, coma or death.

The World Health Organization defines a person as suffering from cerebral malaria if the person is in a coma, has malaria parasites in his or her blood, and has no known alternative cause of the coma. Patients suffering from cerebral malaria are categorized based on whether they have damage to the back of the eyes known as retinopathy. It had previously been found that children who died of “retinopathy-positive” cerebral malaria (i.e. those who had retinopathy) had malaria parasites stuck in small vessels in their brains, which likely caused the coma. By contrast, children who died of “retinopathy-negative” cerebral malaria lacked this parasitic condition, and often also had other infections that can cause a coma, such as meningitis or sepsis.

Because hospitals in many of the areas most affected by malaria often lack the ability to identify what – other than malaria – caused a coma, it was not clear whether malaria parasites influence how retinopathy-negative cerebral malaria develops.

People with certain genetic variants – such as those that underlie sickle cell disease – are protected against the symptoms of malaria infections, and so these variants should also protect against cerebral malaria cases caused by the parasites. Small et al. therefore looked through data that had been collected over several years from people who had been admitted to a hospital in Malawi for cerebral malaria. This revealed that the genetically inherited sickle cell trait is highly protective against retinopathy-negative (as well as retinopathy-positive) cerebral malaria. Therefore, malaria parasites do play a role in a substantial proportion of cases of retinopathy-negative cerebral malaria.

Although Small et al. provide evidence that malaria parasites play a role in retinopathy-negative cerebral malaria, they may not be the only cause of the coma. In the future, the absence of retinopathy could be used as a sign to look for additional factors that contribute to the coma. Currently, all cerebral malaria patients are treated in the same way. Understanding how malaria parasites interact with other illnesses to produce a coma could lead to the development of targeted treatment plans for retinopathy-negative patients.

https://doi.org/10.7554/eLife.23699.002

Introduction

Cerebral malaria (CM) is responsible for a substantial proportion of the approximately 500,000 annual malaria deaths and 2,000,000 severe malaria cases (WHO, 2014, WHO, 2015). CM is defined by the World Health Organization (WHO) as unarousable coma with circulating malaria (Plasmodium) parasitemia and no known non-malaria causal explanation (WHO, 2000). Based on the presence or absence of malaria-specific retinal changes, CM can be classified as retinopathy-positive (Ret+) or retinopathy-negative (Ret-) (Lewallen et al., 1999; Beare et al., 2006). Ret- CM is a common and devastating condition – 40% of CM cases in our cohort were Ret- and of these, 12% died and 10% developed neurological problems (Table 1). Autopsy data show that children dying of Ret+ CM have a high degree of sequestration of parasitized red blood cells in cerebral vasculature (Taylor et al., 2004), considered the pathological hallmark of CM. The pathogenesis of Ret- CM has been a puzzle, in particular the role of malaria parasitemia (Postels and Birbeck, 2011). In the only autopsy study among children dying with CM, that we are aware of ( Taylor et al. (2004) with follow-up results in Milner et al. (2015) and Barrera et al. (2015)), among children for whom retinopathy was assessed, 41 of 42 children dying with Ret+ CM had substantial cerebral sequestration of parasitized red blood cells in the cerebral microvasculature (defined as ≥23% of cerebral capillaries had sequestration) and mostly lacked other identified potential causes of death besides the malaria parasitemia, whereas all 15 children dying with Ret- CM lacked substantial cerebral sequestration (<23% of cerebral capillaries had sequestration) and mostly had non-malarial etiologies of death (see Appendix 1 for causes of death); these numbers update those in Taylor et al. (2004) to include patients enrolled after 2004. Incidental malaria parasitemia is common in people living in areas of high malaria transmission. Therefore, it is possible that at least some children with Ret- CM have a non-malarial etiology of coma and an incidental (asymptomatic) malaria parasitemia. The pathogenesis of Ret-CM is therefore unclear, and the role of malaria parasitemia in the etiology of the acute illness is unknown (Postels and Birbeck, 2011). Our aim of the research presented here was to assess the contribution of acute malaria infection in the pathophysiology of Ret- CM.

Table 1

Characteristics of study participants at admission, Means ± SD for continuous variables. The proportions of missing data are shown in Appendix 1. There are 3704 community controls, but their characteristics are not shown because only their genotypes and not their clinical characteristics were collected. Bold denotes p-value less than 0.05.

https://doi.org/10.7554/eLife.23699.003

	Retinopathy Positive CM	Retinopathy Negative CM	Non-Malaria Hospital Controls	p-value, Ret + vs. Ret -	p-value, Ret + vs. Controls	p-value, Ret – vs. Controls
Number of participants	438	288	204
Female	50%	52%	43%	0.54	0.11	0.05
Age (months)	40 ± 26	44 ± 30	46 ± 30	0.10	0.05	0.53
Mid-upper arm circumference (cm)	14.9 ± 1.6	15.0 ± 1.7	14.8 ± 1.8	0.72	0.53	0.39
Weight (kg)	12 ± 4	13 ± 5	13 ± 6	0.37	0.29	0.74
Height (cm)	90 ± 16	91 ± 17	91 ± 20	0.30	0.40	1.00
Temperature (°C)	38.6 ± 1.2	38.4 ± 1.4	37.7 ± 1.5	0.03	<0.001	<0.001
Febrile (Temperature≥ 37.5°C)	81%	77%	56%	0.23	<0.001	<0.001
Pulse rate – beats/minute	152 ± 26	148 ± 24	139 ± 28	0.06	<0.001	<0.001
Respiratory rate – breaths/minute	47 ± 15	45 ± 13	45 ± 15	0.12	0.17	0.93
Liver size – cm below costal margin	2.0 ± 1.9	1.5 ± 1.9	1.1 ± 1.7	<0.001	<0.001	0.04
Spleen size – cm below costal margin	1.7 ± 2.1	1.6 ± 2.1	0.9 ± 1.6	0.56	<0.001	<0.001
Deep breathing	33%	25%	30%	0.03	0.59	0.18
Blantyre Coma Score: 0 1 2 3 4 5	14% 35% 49% 1% 0% 0%	19% 38% 43% 0% 0% 0%	28% 40% 23% 3% 0% 5%	0.01	0.08	0.90
CSF opening pressure – mm of water	176 ± 75	152 ± 82	176 ± 99	0.001	0.96	0.07
Hematocrit -- %	19.8 ± 6.9	28.2 ± 7.5	28.1 ± 9.6	<0.001	<0.001	0.86
Platelets	81,220± 67,219	161,600± 124,747	248,400± 162,287	<0.001	<0.001	0.86
Malaria parasitemia – parasites/mm³	230,500± 321,924	180,500± 280,676	3,619± 28,917	0.03	<0.001	<0.001
White blood cells	13,040± 9163	13,020± 8923	13,930± 9544	0.97	0.29	0.31
Lactate – mmol/liter	8.6 ± 5.0	7.3 ± 4.4	5.5 ± 3.9	0.05	<0.001	0.007
Blood glucose – mmol/liter	6.1 ± 3.9	6.8 ± 4.4	7.6 ± 5.3	0.03	<0.001	0.05
CSF white cell count – % ≥ 5	16%	20%	24%	0.31	0.06	0.37
Blood culture positive for pathogen	4%	2%	14%	0.51	<0.001	<0.001
HIV positive	18%	17%	15%	0.91	0.59	0.66
Outcomes
Discharge outcome: Full recovery Neurological Sequalae Died	69% 10% 21%	78% 10% 12%	57% 15% 28%	0.003	0.01	<0.001

Figure 1 depicts three possible pathways to clinically-defined (WHO-defined) CM (Postels and Birbeck, 2011). One pathway is to Ret+ CM for which there is evidence that malaria parasites play a primary role. As noted above, at autopsy, Ret+ CM is associated with the sequestration of parasitized red cells in the cerebral microvasculature (Taylor et al., 2004). Compared to children with Ret- CM, those who are Ret+ have increased concentrations of P. falciparum HRP2, a parasite-produced protein reflecting total body parasite burden (Seydel et al., 2012). Ocular funduscopic findings in Ret+ CM mirror the microvascular pathology observed on fluorescein angiography (MacCormick et al., 2015) and are correlated with the severity of sequestration in both the retina and the brain at autopsy (Barrera et al., 2015). Two pathways to Ret- CM, pathway (a) and pathway (b) are depicted in Figure 1. As noted above, in patients dying with Ret- CM, the cerebral microvasculature does not have substantial sequestered parasitized erythrocytes (<23% of cerebral capillaries have sequestration), plasma concentrations of HRP2 are decreased, and a variety of non-malarial causes of death have been identified (Taylor et al., 2004). For Ret- CM, one potential pathway is asymptomatic parasitemia and another illness that is sufficient, in and of itself, to produce coma (pathway (a) in Figure 1). Another potential pathway is parasitemia leading to uncomplicated malaria illness (e.g., fever) combined with a second insult (innate or acquired), resulting in coma (pathway (b) in Figure 1); the two hits (symptomatic malaria+ innate or acquired second factor) result in the clinical syndrome of Ret- CM. A key unanswered question about the pathogenesis of Ret- CM is, are malaria parasites incidental to coma (only pathway (a) exists) or do they play a role in the pathogenesis of Ret- CM (pathway (b) exists) (Bearden, 2012; Postels and Birbeck, 2011)?

Figure 1

Download asset Open asset

Potential pathways to clinically-defined cerebral malaria and genetic bottle necks.

There are three potential pathogenetic routes to WHO-defined cerebral malaria (CM). The first, shown in red, is the classical pathway: a malaria infection evolves into retinopathy-positive (Ret+) CM. The second and third possibilities produce retinopathy-negative (Ret-) CM. In (a) the coma is entirely the result of another etiology and the malaria parasitemia is incidental. In (b), the coma is a product of the interaction between the malaria parasitemia and an additional cause (or causes) of coma. Sickle cell trait is underrepresented in patients with Ret+ and Ret- cerebral malaria (CM) because of the bottleneck at the transition between 'malaria infection' (asymptomatic malaria) and 'malaria disease' (uncomplicated malaria). Blood group O is underrepresented in patients with Ret+ CM, but not in those with Ret- CM. Taken together, the results for sickle cell trait and blood group O suggest that some Ret- CM cases occur through pathway (b) (because sickle cell trait is underrepresented in Ret- CM) and that malaria parasites contribute to the pathogenesis of these cases, and that sickle cell trait reduces the pathogenetic potential of malaria infection for Ret- CM but do not provide evidence that blood group O reduces the pathogenetic potential of malaria infection for Ret- CM.

https://doi.org/10.7554/eLife.23699.004

Whether malaria parasites play a role in the pathogenesis of Ret- CM could in principle be tested by a randomized experiment. For example, Smith (2007) considered a hypothetical blood-stage malaria vaccine that reduces parasite density by 50%; the vaccine would reduce malaria illness but not the incidence of parasitemia. If such a vaccine existed, then a way to test whether malaria parasites are pathogenetic in Ret- CM would be to randomize a large number of children to either (i) the blood-stage vaccine or (ii) placebo. If malaria parasites are never pathogenetic in Ret- CM, then we would expect no difference in Ret- CM because the blood stage vaccine would fail to prevent the cause of the development of the Ret- CM whereas if malaria parasites are sometimes pathogenetic in Ret- CM in a way that requires the development of uncomplicated malaria illness, then the blood stage vaccine would prevent the development of Ret- CM in some cases. Such an experiment is not currently feasible because no blood-stage vaccine has reached a Phase III trial (Miura, 2016) and even if an effective blood-stage vaccine was developed, the experiment would require a huge sample size to have power to detect a change in Ret- CM rates.

Though a randomized experiment with a blood-stage malaria vaccine that would have power to detect a difference in Ret- CM rates is not currently feasible, nature provides traits that protect against malaria illness in a random way through genetic inheritance. The general approach of using genetic variation to construct natural experiments is called Mendelian randomization (Smith and Ebrahim, 2003). The sickle cell trait (HbAS) – inheritance of one abnormal allele of the betaglobin gene – protects against symptomatic malaria (Modiano et al., 2001; Taylor et al., 2012; Williams et al., 2005; Willcox et al., 1983). Thus, a person who inherits one abnormal allele of the betaglobin gene has an antimalarial biochemical protection provided by nature. A second inherited trait which affects susceptibility to malaria illness is blood group O (BGO), which protects against CM compared to groups A, B or AB (Cserti and Dzik, 2007; Malaria Genomic Epidemiology Network, 2014). Analogous to the randomized trial described above, possession vs. lack of a malaria protective trait assigns children to an arm in which some malaria illness is prevented vs. not prevented. For possession vs. lack of a trait to be fully analogous to the blood stage vaccine randomized trial described above, possession of the trait cannot affect malaria parasitemia incidence just like assignment to the blood-stage vaccine arm in the randomized trial does not affect malaria parasitemia incidence. If the trait affects malaria parasitemia incidence, then it could decrease the Ret- CM rate not because it decreases coma but because the WHO definition of Ret- CM requires malaria parasitemia; the natural experiment induced by the trait would then be biased for assessing the effect of the trait on coma in the same way that a randomized trial is biased if the treatment could affect the measurement of the outcome (WHO-defined Ret- CM) without affecting the true outcome of interest (coma without malarial retinopathy). For both BGO and HbAS, it is plausible that the traits do not protect against malaria parasitemia incidence as systematic reviews have not found consistent evidence for protection (Uneke, 2007; Taylor et al., 2012); we will assume no protection for our main analysis but do sensitivity analyses that allow for protection. Under the assumptions that a trait does not affect malaria parasitemia incidence and the trait is randomly assigned, then if the trait decreases the probability of developing both Ret+ and Ret- CM, this suggests that malaria parasites contribute to the pathogenesis of both conditions. If the trait decreases the probability of developing Ret+ CM but not Ret- CM, this would suggest that malaria parasites are pathogenetic for Ret+ CM but are either not pathogenetic for Ret-CM or the trait affects an aspect of disease not causal to the development of Ret- CM.

Results

Using data gathered from 1996 to 2007 in a study of CM pathogenesis in Blantyre, Malawi (Taylor et al., 2004; Seydel et al., 2015) as well as the MalariaGEN consortium (Malaria Genomic Epidemiology Network, 2008), we compared children with CM to two types of controls – (1) community controls; (2) hospital controls, children who were admitted to the Paediatric Research Ward with a known non-malarial cause of illness – meningitis, non-malarial anemia or other non-malaria illness. Table 1 shows admission characteristics of the hospitalized participants. The Blantyre coma score is statistically significantly higher in Ret+ CM patients than Ret- CM patients, but we do not regard the difference as clinically significant. In general, the patients with Ret+ CM were more severely ill than those with Ret- CM (higher lactate, more deep breathing and a higher chance of death). The malaria illness is more severe in Ret+ CM than Ret- CM patients (lower platelet count and more anemia). The higher opening CSF pressures in Ret+ CM patients compared to Ret- CM patients suggests a higher proportion of Ret+CM patients have increased brain volume. Comparing CM cases to non-malaria controls, as expected, laboratory abnormalities associated with malaria infection (e.g. low hematocrit and platelet count) were more frequent in the CM cases.

For each trait t (t=HBAS or blood group O (BGO)), we test the null hypothesis ( $H_{0}^{t}$ ) that the trait frequency is the same in controls and true Ret- CM cases vs. the alternative hypothesis ( $H_{a}^{t}$ ) that the trait frequency is higher in controls than true Ret- CM cases. Here, true Ret- CM refers to Ret- CM measured without error; in the actual data, retinopathy status may be measured with error and this measurement error is taken into account in the inferences (Materials and methods). Under a model in which the trait does not affect other potential contributors to Ret- CM besides malaria and does not affect which CM cases are admitted to the Paedeatric Research Ward (as compared to dying before reaching the Ward or recovering before being referred to the Ward), the null hypothesis $H_{0}^{t}$ implies that malaria parasitemia is an incidental finding in children with Ret- CM and/or the trait affects an aspect of disease not causal to development of Ret- CM, while the alternative hypothesis $H_{a}^{t}$ implies that malaria parasitemia is necessary for some Ret- CM cases and the trait reduces the pathogenetic potential of malaria infection for Ret- CM. Note that if either $H_{0}^{H b A S}$ or $H_{0}^{B G O}$ is false, this implies that malaria parasitemia plays a pathogenetic role in at least some Ret- CM cases.

We calculated HbAS and BGO proportions in study subjects and made inferences about odds ratios (Table 2) and tested $H_{0}^{H b A S}$ and $H_{0}^{B G O}$ . For both HbAS and BGO, non-malaria hospitalized controls did not differ from community controls (HbAS p-value=0.86; BGO p-value=0.83); therefore, subsequent analyses combined the control groups. Controls had a higher proportion of HbAS than true Ret- CM patients (odds ratio: 14.33, 95% CI: 3.21, 257.24) and true Ret+ CM patients (odds ratio: 1223.22, 95% CI: 9.87, $\infty$ ). For BGO, the controls were comparable to true Ret- CM patients (odds ratio: 1.03, 95% CI: 0.83, 1.29) but higher than true Ret+ CM patients (odds ratio: 1.23, 95% CI: 1.01, 1.50). There is strong evidence to reject $H_{0}^{H b A S}$ (p-value<0.0001) but not $H_{0}^{B G O}$ (p-value=0.79); these results are insensitive to different plausible assumptions about the false discovery rate and false omission rate for malarial retinopathy (Appendix 1). Taken together, these tests suggest that malaria parasitemia is pathogenic for a proportion of Ret- CM cases. Sickle cell trait protects against Ret- CM, but blood group O does not.

Table 2

The top panel displays sickle cell trait (HbAS) proportions in retinopathy-positive (Ret+) cerebral malaria (CM), retinopathy-negative (Ret-) CM and control groups. The bottom panel displays ABO blood group gene proportions in Ret+ CM, Ret- CM and control groups. The last two rows of each panel display the odds ratios comparing controls to true Ret+ and true Ret- CM groups, which account for the fact that there is measurement error in observed retinopathy status (false discovery rate = 0.07 and false omission rate = 0.05).

https://doi.org/10.7554/eLife.23699.005

	Ret+ CM	Ret- CM	Non-malaria hospital controls	Community controls
Sample size	438	287	192	3657
HbAS^*	0	1	8	175
HbAA	437	286	184	3482
Proportion of HbAS	0	.003	.042	.048

			Odds ratio (95% CI)
Non-malaria hospital controls vs. community controls			0.87 (0.36, 1.78)
Controls vs. true Ret- CM			14.33 (3.21, 257.24)
Controls vs. true Ret+ CM			1223.22 (9.87, $\infty$ )

	Ret+ CM	Ret- CM	Non-malaria hospital controls	Community controls
Sample size	433	286	199	3543
Blood Group O	175	135	96	1739
Blood Group A, B or AB	258	151	103	1804
Proportion of Blood Group O	.404	.472	.482	.491

			Odds ratio (95% CI)
Non-malaria hospital controls vs. community controls			0.97 (0.72, 1.30)
Controls vs. true Ret- CM			1.03 (0.83, 1.29)
Controls vs. true Ret+ CM			1.23 (1.01, 1.50)

* HbAS (sickle cell trait) means that that the person has one normal and one abnormal copy of the hemoglobin beta gene. HbAA means the person has two normal copies of the hemoglobin beta gene.

In Materials and methods, we formulate a sufficient-component cause model (Rothman, 1976) based on Figure 1 and describe how to make inferences about the fraction of Ret- CM cases that are due to pathway (b) in Figure 1, i.e., the malaria parasitemia attributable fraction of Ret- CM (the fraction of Ret- CM cases that would be prevented if malaria parasitemia were eliminated [Benichou et al., 1998]). The fraction itself cannot be estimated without strong biological assumptions (Greenland and Robins, 1988), but a lower bound can be estimated under plausible assumptions. Table 3 shows inferences for this lower bound under the main model that assumes the traits do not protect against malaria parasitemia incidence and sensitivity analyses. Under the main model, the lower bound is estimated to be .93 with 95% confidence interval (.68, 1). For the sensitivity analyses, although the systematic review of Taylor et al. (2012) found no consistent evidence that HbAS reduces malaria parasitemia incidence, some studies reviewed found protection and we consider sensitivity analyses that allow for a small amount of protection (10%) and the largest amount of protection found in all the studies reviewed (41%) (Ntoumi et al., 1997). Also, the sensitivity analyses vary the false discovery rate and false omission rate between 0 and the upper bound estimated in Materials and methods. Under all scenarios considered, we found evidence for a substantial contribution of malaria parasites to the pathogenesis of Ret- CM with lower 95% confidence bounds ranging from .37 to .77 and point estimates for the lower bound ranging from .86 to .95.

Table 3

Inferences for lower bound on malaria parasitemia attributable fraction of Ret- CM (fraction of Ret- CM cases that would be prevented if malaria parasitemia were to be eliminated) under the sufficient-component cause model based on Figure 1 presented in Materials and methods. Inferences under the main model and sensitivity analyses that vary the effect of HbAS on malaria parasitemia incidence rate, the false discovery rate ( $F D R$ ) and the false omission rate ( $F O R$ ) for malarial retionopathy.

https://doi.org/10.7554/eLife.23699.006

Effect of HbAS on malaria parasitemia incidence rate	$F D R$	$F O R$	Lower bound on malaria parasitemia attributable fraction of Ret- CM Estimate (95% CI)
Main Model
No Effect	.07	.05	.93 (.68, 1)
Sensitivity Analyses
Reduce 10%	.07	.05	.92 (.64, 1)
Reduce 41%	.07	.05	.88 (.46, 1)
No Effect	.30	.11	.94 (.75, 1)
Reduce 10%	.30	.11	.94 (.72, 1)
Reduce 41%	.30	.11	.91 (.58, 1)
No Effect	0	.11	.92 (.62, 1)
Reduce 10%	0	.11	.91 (.58, 1)
Reduce 41%	0	.11	.86 (.37, 1)
No Effect	.30	0	.95 (.77, 1)
Reduce 10%	.30	0	.94 (.74, 1)
Reduce 41%	.30	0	.91 (.61, 1)
No Effect	0	0	.92 (.66, 1)
Reduce 10%	0	0	.92 (.63, 1)
Reduce 41%	0	0	.87 (.44, 1)

Discussion

We have studied a natural experiment that alters the level of malaria illness and found evidence that children with genetic traits associated with resistance to malaria illness are underrepresented in admissions with both Ret+ and Ret- CM (HbAS) or in admissions with Ret+ CM only (BGO).

Figure 1 shows a model of how HbAS and BGO affect Ret- and Ret+ CM. HbAS protects children with malaria parasitemia from developing uncomplicated malaria illness (e.g., fever) and severe malaria illness (Taylor et al., 2012). By contrast, current evidence suggests that BGO has no effect on developing uncomplicated malaria illness (Uneke, 2007), but does inhibit the cytoadherence of parasitized red blood cells to endothelial cells in the microcirculation, e.g., by affecting rosetting (Rowe et al., 2007) and physical properties of the red cell membrane (Méndez et al., 2012), thereby preventing severe malaria illness (Cserti and Dzik, 2007; Migot-Nabias et al., 2000; Uneke, 2007). Consistent with this current evidence, in Figure 1, HbAS prevents Ret- CM through pathway (b), which involves two components (uncomplicated malaria illness + other illness), by preventing one of the components, uncomplicated malaria illness, whereas BGO does not protect against Ret- CM through pathway (b) because it does not prevent uncomplicated malaria illness (Figure 1).

In Figure 1 pathway (b), Ret- CM results from an interaction between malaria illness and other illness. While the interactions between malaria parasites and other pathogens are incompletely understood and not well investigated, there is evidence for interaction (Obaro and Greenwood, 2011; Mallewa et al., 2013; Postels and Birbeck, 2011). In studies of sepsis and malaria, the effect of microvascular parasite sequestration on the integrity of the gut mucosa is thought to allow bacterial seeding into the blood stream and hence bacteremia (Scott et al., 2011). Another example is that following antigenic challenge from a Plasmodium falciparum candidate vaccine, children coinfected with schistosomiasis had lower acquired specific immune responses than those not infected (Diallo et al., 2010).

Another way in which malaria parasitemia could play a causal role in Ret- CM besides pathway (b) in Figure 1 is that the parasitemia could be the sole cause of coma. Taylor et al. (2004)’s autopsy study results suggest that among children dying from Ret- CM, malaria parasitemia is typically not the sole cause of death (see Causes of Death in Autopsy study in Appendix 1), but it is possible that in survivors from Ret- CM, malaria parasitemia is the major contributor to acute illness. Whether malarial retinopathy is present or absent in CM could be affected by factors such as the child’s genome, the parasite’s genome and the child’s previous exposures to malaria (Postels and Birbeck, 2011). The protection provided against Ret- CM by HbAS but not BGO could be explained by an interaction between the factor(s) affecting whether malarial retinopathy is present and BGO, e.g., a parasite genotype which causes malarial retinopathy to be absent could interfere with the mechanism by which BGO provides protection against severe malaria.

Our study has several limitations. We only considered the genetic variants of sickle cell trait and blood group because these were the only statistically significant (p<.05) protective variants in the Malawi sample on which malarial retinopathy was measured, but future work could further test the model in Figure 1 by looking at additional genetic variants that have been found to affect the risk of malaria in other sites in Africa (Malaria Genomic Epidemiology Network, 2014). Our study only addresses CM pathogenesis in children. Adult and pediatric CM have important clinical differences and our results may not be generalizable to adults with Ret- CM. Detection of the presence or absence of malarial retinopathy was determined by several ophthalmologists over the 11 year duration of data collection, the sensitivity of detection of retinal changes may have varied between practitioners. Malarial retinopathy was determined on the basis of ophthalmoscopy alone, but since the time of our study, techniques such as optical coherence tomography that may increase sensitivity have been developed (Joshi et al., 2017). We measure malarial retinopathy at the time of admission, but patients are admitted at different points on the disease trajectory. Malarial retinopathy can change over time; it doesn’t resolve during the 2–4 days of hospitalization but can become worse, which is a poor prognostic sign. Although there are limitations in our study to the sensitivity with which malarial retinopathy is measured, even if a moderate number of Ret- CM cases should be Ret+ CM cases in Table 2, e.g., if the false omission rate is .5 and the false discovery rate is 0, there is still strong evidence that HbAS has a protective effect for Ret- CM (p-value=0.007; odds ratio: 6.81, 95% CI: (1.50, $\infty$ )). Our main analysis assumed that possession of HbAS or BGO does not affect malaria parasitemia incidence; however, sensitivity analyses showed that our results were not sensitive to plausible violations of this assumption. Our analysis assumed that HbAS and BGO do not have selection effects on which CM cases are admitted to the Paediatric Research Ward as opposed to dying before reaching the Paedieatric Research Ward or being cured before needing to be referred to the ward; another interpretation of our study is that instead of studying ‘cerebral malaria’ as such, we have studied, ‘cerebral malaria mild enough to make it to the ward but severe enough not to recover without being taken to the ward.’ Some of the non-malaria hospital controls had malaria parasitemia; since children can sometimes develop severe malaria with relatively low levels of parasitemia, it is possible that malaria contributed to the illness in these control participants. Furthermore, older non-malaria hospital controls may have a selection bias for HbAS since they have survived to an older age (Ackerman et al., 2005). Our results are robust to considering only the community controls (Appendix 1—table 4) which do not suffer from these potential biases of the non-malaria hospital controls.

In summary, we studied a natural experiment that alters the level of malaria illness experienced by children through genetic variation and found evidence that malaria parasitemia is on the causal pathway to a substantial proportion of Ret- CM cases. Our approach of using genetically inherited traits to study CM pathogenesis could be adapted to illuminate the pathogenesis of other diseases.

Materials and methods

Setting and study participants

Request a detailed protocol

The study sample includes children with WHO-defined CM from 1997 to 2007 who were admitted to the Paediatric Research Ward at Queen Elizabeth Central Hospital, a tertiary referral center and teaching hospital in Blantyre, Malawi. The WHO definition of CM requires coma, circulating Plasmodium falciparum parasites, and no other cause of coma evident either by history or physical examination. Patients were enrolled during the rainy season, the time of annual peak incidence of CM. All patients were treated with intravenous quinine as standard antimalarial therapy. Enrollment in the study required explicit written consent from the parent or guardian. There were 947 enrolled patients. Malarial retinopathy was assessed on 726 of these patients and we subsequently limited our analysis to these 726 patients. To assess retinal changes, direct and indirect ophthalmoscopy were performed by an ophthalmologist well versed in findings typical of malarial retinopathy.

We consider two types of controls in the study. The first type are community controls – 3704 children intended to be representative of the populations to which the cases belonged. Community controls were cord blood samples from Queen Elizabeth Central Hospital, mostly from the newborn nursery. The second type of controls are 194 patients who were admitted during the study period to the Paediatric Research Ward at Queen Elizabeth Central Hospital with a non-malarial cause of illness – meningitis, non-malarial anemia or other non-malaria illness.

Advantages of considering two control groups

Request a detailed protocol

Two sources of bias in case-control studies are hidden bias (unmeasured confounding) and selection bias (Rosenbaum, 1987, Rosenbaum, 2002). Hidden bias occurs when there are unmeasured variable(s) that are associated with the exposure and the outcome. Selection bias occurs when control subjects are selected based on a variable that is affected by the exposure. Different control groups might be subject to different amounts of hidden bias and selection bias. If two such control groups have similar exposure rates, this provides evidence against hidden bias and selection bias (Rosenbaum, 1987). Further, if the case group has a different exposure rate than both control groups and the control groups have similar exposure rates, this provides stronger evidence that the exposure has a causal effect on the outcome, rather than the results being biased by hidden bias or selection bias, as compared to a study with a single control group (Rosenbaum, 1987).

Community controls are less likely to suffer from selection bias than the hospital controls. Community controls were randomly selected from the same population as the cases. In contrast, hospital controls could suffer from selection bias if the genetic variant of interest was associated with a non-malaria illness -- this type of selection bias is known as Berkson’s bias (Westreich, 2012; Berkson, 1946). Although the hospital controls might suffer from more selection bias than the community controls, the hospital controls have the potential advantage that they would reduce hidden bias if there was population stratification such that the genetic variant was associated with an unmeasured population stratification feature (e.g., housing conditions) that caused both malaria and non-malaria illness.

Genotypes

Request a detailed protocol

Participants were genotyped using the Sequenom iPLEX MassARRAY platform (Malaria Genomic Epidemiology Network, 2014). The genotyping included 55 SNPs for which there were previously reported associations with severe malaria. Two of the SNPs were found to be statistically significantly (p<.05) associated with cerebral malaria in Malawi -- HBB rs334, which encodes the sickle cell trait, and ABO rs8176719, which encodes the blood type (O vs. A, B or AB); see Figure 1 in Malaria Genomic Epidemiology Network (2014). These two SNPs were used in our analysis.

Hypotheses

For a given genetically inherited trait, we consider two competing hypotheses for testing whether malaria parasitemia plays a pathogenic role in Ret- CM. We assume a biological model under which the trait does not affect other illnesses that cause Ret- CM and under which the trait does not have a selection effect on which CM cases are admitted to the Paedieatric Research Ward vs. dying before reaching the Paediatric Research Ward or being cured before needing to be referred to the ward.

Null hypothesis

Request a detailed protocol

Malaria parasitemia is an incidental finding in children with true Ret- CM and/or the genetic trait affects an aspect of disease not causal to development of true Ret- CM.

Let $λ_{t . r +}$ be the probability that a child with true Ret+ CM has a trait t (e.g., t could be the sickle cell trait or blood type O), $λ_{t . r -}$ the probability that a child with true Ret- CM has the trait t and $λ_{t . c o}$ the probability that a control child has the trait t. For the observed data, we take into account that malarial retinopathy may be measured with error. Among children with WHO-defined CM, let $F D R$ be the false discovery rate that a child who is found by the examining ophthalmologist to have Ret+ CM actually has Ret- CM and let $F O R$ be the false omission rate that a child who is found by the examining ophthalmologist to have Ret- CM actually has Ret+ CM.

Suppose malaria parasitemia is an incidental finding in patients with true Ret- CM and/or the trait affects an aspect of malaria infection not experienced by Ret- CM patients. Then the probability of the trait would be the same in controls as in patients with true Ret- CM, $λ_{t . r -} = λ_{t . c o}$ . If this is the case, we have the following probabilities for observing the trait in the observable groups:

\begin{matrix} P (C o n t r o l h a s t r a i t) = λ_{t . c o} \\ P (Ret+ CM ​ p a t i e n t h a s t r a i t) = (1 - F D R) * λ_{t . r +} + F D R * λ_{t . c o} \\ P (Ret- CM ​ p a t i e n t h a s t r a i t) = F O R * λ_{t . r +} + (1 - F O R) * λ_{t . c o} \end{matrix}

Alternative hypothesis

Request a detailed protocol

Malaria parasitemia is pathogenic for true Ret- CM and the trait reduces the pathogenic potential of malaria infection for true Ret- CM.

If malaria parasitemia is pathogenic for true Ret- CM and the trait reduces the pathogenic potential of malaria infection for true Ret- CM, then the trait will reduce the probability of true Ret- CM, $λ_{t . r -} < λ_{t . c o}$ . We have the following probabilities for observing the trait in the observable groups:

\begin{matrix} P (C o n t r o l h a s t r a i t) = λ_{t . c o} \\ P (Ret+ CM ​ p a t i e n t h a s t r a i t) = (1 - F D R) * λ_{t . r +} + F D R * λ_{t . r -} \\ P (Ret- CM ​ p a t i e n t h a s t r a i t) = F O R * λ_{t . r +} + (1 - F O R) * λ_{t . r -} \end{matrix}

To estimate $F D R$ and $F O R$ , we use data from Beare et al., 2002 who compared two ophthalmologists’ concordance in grading malarial retinopathy. As described in the next section, we obtain point estimates of .07 and .05 and upper bound estimates of .30 and .11 for $F D R$ and $F O R$ respectively; the main analyses use the point estimates and additional analyses use the upper bound estimates. For fixed values of $F D R$ and $F O R$ , we estimate the parameters $λ_{t . r +}$ , $λ_{t . r -}$ and $λ_{t . c o}$ by maximum likelihood.

Estimating false discovery and false omission rates for malarial retinopathy

Request a detailed protocol

We use data on 65 patients at Queen Elizabeth Central Hospital in Blantyre, Malawi who were each examined by two ophthalmologists (Beare et al., 2002). Malarial retinopathy is diagnosed if any of the following six signs are present: retinal hemorrhages (RH), macular whitening (MW), foeval whitening (FW), peripheral whitening (PW), vessel changes (VC) and capillary whitening (CW). RH, VC and CW require little observer judgement if they are seen, but they may not be seen, depending on the degree of pupillary dilation and the presence/absence of spontaneous eye movements; for these consider the specificity to be 1 (Beare et al., 2002). Identifying MW, FW and PW requires more experience on the part of the observer; MW and FW are more reproducible, and PW is less so (Beare et al., 2002).

We first consider the false discovery rate. To estimate an upper bound on the false discovery rate, we assumed that if RH, VC or CW was detected by either ophthalmologist, there was true malarial retinopathy, but if MW, FW or PW was detected without RH, VC or CW, then it was a false discovery; this is likely an overestimate since in some cases where MW, FW or PW was detected without RH, VC or CW, there is likely true malarial retinopathy that was missed by RH, VC and CW. There were 39 patients diagnosed by ophthalmologist 1 with malarial retinopathy, 33 of whom had RH, VC or CW by at least one ophthalmologist and there were 36 patients diagnosed by ophthalmologist 2 with malarial retinopathy, 33 of whom had RH, VC or CW by at least one ophthalmologist. To find a conservative upper bound on the false discovery rate, we consider ophthalmologist 1 and found the 95% Wilson binomial confidence interval (Wilson, 1927) based on 6 out of 39 false discoveries, resulting in an 95% confidence interval of (0.07, 0.30), so resulting in an upper bound of 0.30 for the false discovery rate. To find a point estimate for the false discovery rate, we assume that there was true malarial retinopathy if RH, VC or CW was detected by either ophthalmologist or if FW was detected by both ophthalmologists or if MW was detected by both ophthalmologists since FW and MW were found to be reproducible by Beare et al., 2002, and then we averaged the resulting point estimates for the false discovery rate for the two ophthalmologists ((4/39 + 1/36)/2), to obtain an estimate of .07.

We next consider the false omission rate. To estimate an upper bound on the false omission rate, we estimate an upper bound on the false omission rate if we were to only use RH, VC and CW to diagnose malarial retinopathy. The actual false omission rate is likely to be at least as small because we also diagnose malarial retinopathy if there’s any of MW, FW or PW and we think that the majority of time when we find MW, FW or PW but not RH, VC and CW, there is true malarial retinopathy, whereas a false omission will be relatively rare. We assume that the false positive rate for RH, VC and CW is 0 and use a model similar to model 1 in Nedelman (1988). Specifically, let $ψ$ denote the prevalence of true malarial retinopathy and $ζ$ the false negative probability that an ophthalmologist will fail to detect RH, VC or CW in a child with true malarial retinopathy. Assume that we have two independent ophthalmologists. Then, the probability that both ophthalmologists will detect at least one of RH, VC or CW is $ψ {(1 - ζ)}^{2}$ , the probability that one but not the other ophthalmologist will detect RH, VC or CW is $2 ψ ζ (1 - ζ)$ and the probability that both ophthalmologists will detect none of RH, VC or CW is $ψ ζ^{2} + 1 - ψ$ . The false omission rate is $F O R = \frac{ψ ζ}{1 - ψ + ψ ζ}$ . We found a point estimate of 0.05 for the false omission rate and a 95% confidence interval (using the percentile bootstrap) of (0, .11). Thus we take .11 as an estimated upper bound for the false omission rate and .05 as a point estimate for the false omission rate, recognizing that it is likely to be an upwardly biased point estimate.

Statistical inference

Request a detailed protocol

We will test the null hypothesis $H_{0}^{t} : λ_{t . r -} = λ_{t . c o}$ (the probability of the malaria resistance trait t is the same in true Ret- CM cases as controls) versus the alternative hypothesis $H_{a}^{t} : λ_{t . r -} < λ_{t . c o}$ (the probability of the malaria resistance trait t is lower in true Ret- CM cases than controls). To test these hypotheses, we will estimate the parameters under the null and alternative hypotheses by maximum likelihood and use the generalized likelihood ratio test with 1 degree of freedom (Rice, 2007). The maximum likelihood estimation takes into account the assumed false discovery rate and false omission rate for malarial retinopathy detection. We will form confidence intervals for the odds ratios of the trait among controls vs. true Ret- CM patients ( $[{λ_{t . c o} / (1 - λ_{t . c o})} / {λ_{t . r -} / (1 - λ_{t . r -})}]$ and the odds ratio of the trait among controls vs. true Ret+ CM patients $[{λ_{t . c o} / (1 - λ_{t . c o})} / {λ_{t . r +} / (1 - λ_{t . r +})}]$ by inverting the generalized likelihood ratio test. We will use Fisher’s exact to test whether the odds ratio of the trait differs among the community controls vs. the non-malaria illness controls and construct a 95% confidence interval for this odds ratio.

Inference for malaria parasitemia attributable fraction for coma among Ret- CM Cases

Request a detailed protocol

The malaria parasitemia attributable fraction for coma among Ret- CM cases is the fraction of coma that would be prevented if malaria parasitemia were to be eliminated among Ret- CM cases. We formulate a model using the sufficient-component cause framework (Rothman, 1976) and estimate the malaria parasitemia attributable fraction for Ret- CM using this model. A sufficient cause for a disease is a set of conditions that inevitably produces the disease. We assume that Ret+ CM can be represented by one sufficient cause: malaria parasitemia + factors that lead the malaria parasitemia to develop into uncomplicated malaria illness (e.g., lack of immunity) + factors that lead the uncomplicated malaria illness to further progress to complicated malaria illness with coma and malarial retinopathy (e.g., genetic complexity of the malaria infection that overwhelms a child’s ability to control the infection). We assume that Ret- CM can be represented by two sufficient causes: (a) malaria parasitemia + another illness that is sufficient, in and of itself, to produce coma without malarial retinopathy; (b) malaria parasitemia + factors that lead the malaria parasitemia to develop into uncomplicated malaria illness + second insult (innate or acquired) that combined with the uncomplicated malarial illness leads to a coma without malarial retinopathy but which would not in and of itself be sufficient to produce coma. These two sufficient causes (a) and (b) for Ret- CM correspond to pathways (a) and (b) to Ret- CM in Figure 1. Let $p$ be the proportion of Ret- CM from sufficient cause (b), which is the malaria parasitemia attributable fraction for coma among Ret- CM cases.

Let $r_{t, p}$ be the factor by which the trait t multiplies the risk of malaria parasitemia (i.e., relative risk of malaria parasitemia for individuals with trait t compared to individuals without trait t), $r_{t, u}$ be the factor by which the trait multiplies the risk of factors that lead the malaria parasitemia to develop into uncomplicated malaria illness conditional on having malaria parasitemia, $r_{t, c}$ be the factor by which the trait multiplies the risk of factors that lead uncomplicated malaria illness to further progress to complicated malaria illness with coma and malarial retinopathy, $r_{t, a}$ be the factor by which the trait multiplies the risk of another illness that is sufficient, in and of itself, to produce coma without malarial retinopathy in the presence of malaria parasitemia and $r_{t, s}$ be the factor by which the trait multiplies the risk of a second insult that combined with uncomplicated malaria illness leads to a coma without malaria retinopathy but which would not in and of itself be sufficient to produce coma. We assume that $r_{t, u}, r_{t, c} \leq 1$ , i.e., that the trait has no effect or a beneficial effect in preventing uncomplicated and complicated malaria.

Let $λ_{t}$ be the proportion of trait t in the population. For a rare disease, this proportion is approximately the proportion of trait t among the controls, $λ_{t} \approx λ_{t . c o}$ . Ret+ CM and Ret- CM are both rare diseases – the vast majority of malaria infections do not progress to cerebral malaria; subsequently we assume $λ_{t} = λ_{t . c o}$ . We then have

λ_{t . r^{+}} = \frac{r_{t, p} r_{t, u} r_{t, c} λ_{t . c o}}{r_{t, p} r_{t, u} r_{t, c} λ_{t . c o} + 1 - λ_{t . c o}}

λ_{t . r^{-}} = (1 - p) \frac{r_{t, p} r_{t, a} λ_{t . c o}}{r_{t, p} r_{t, a} λ_{t . c o} + 1 - λ_{t . c o}} + p \frac{r_{t, p} r_{t, u} r_{t, s} λ_{t . c o}}{r_{t, p} r_{t, u} r_{t, s} λ_{t . c o} + 1 - λ_{t . c o}}

For our main model, we make the following assumptions: (i) $r_{t, p} = 1$ -- the trait has no effect on malaria parasitemia; (ii) $r_{t, a} = 1$ -- the trait has no effect on illnesses that are sufficient in and of themselves to produce coma without malarial retinopathy in the presence of malaria parasitemia and (iii) $r_{t, s} = 1$ -- the trait has no effect on an insult that combined with uncomplicated malaria illness leads to a coma without malaria retinopathy but which would not in and of itself be sufficient to produce coma. We relax the assumption that $r_{t, p} = 1$ in sensitivity analyses. Under assumptions (i)-(iii),

\begin{matrix} λ_{t . r^{+}} = \frac{r_{t, u} r_{t, c} λ_{t . c o}}{r_{t, u} r_{t, c} λ_{t . c o} + 1 - λ_{t . c o}} \\ λ_{t . r^{-}} = (1 - p) λ_{t . c o} + p \frac{r_{t, u} λ_{t . c o}}{r_{t, u} λ_{t . c o} + 1 - λ_{t . c o}} \end{matrix}

The parameters $λ_{t . c o}, λ_{t . r^{+}}, λ_{t . r^{-}}$ can be identified from the data and thus Equation (1.1) involve two equations in three unknowns ( $p, r_{t, u}, r_{t, c}$ ). The $p$ is minimized (and hence the malaria parasitemia attributable fraction for coma among Ret- CM cases, is minimized) by letting $r_{t, c} = 1$ , specifically the minimizing $p$ solves $λ_{t . r^{-}} = (1 - p) λ_{t . c o} + p λ_{t . r^{+}}$ . Let $p_{l b, t}$ represent this lower bound on $p$ based on trait $t$ , which is a function of $(λ_{t . c o}, λ_{t . r^{+}}, λ_{t . r^{-}})$ ; let $p_{l b, t'}$ represent this lower bound on $p$ based on another trait $t'$ , which is a function of $(λ_{t' . c o}, λ_{t' . r^{+}}, λ_{t' . r^{-}})$ and let $p_{l b} = \min (p_{l b, t}, p_{l b, t'})$ be the lower bound on $p$ based on both traits, which is a function of $(λ_{t . c o}, λ_{t . r^{+}}, λ_{t . r^{-}}$ , $λ_{t' . c o}, λ_{t' . r^{+}}, λ_{t' . r^{-}})$ . We estimate $p_{l b}$ by maximum likelihood and form a 95% confidence interval by inverting the generalized likelihood ratio test for $p_{l b}$ .

We conduct sensitivity analyses that allow for $r_{t, p} < 1$ . Maintaining the assumptions (ii) $r_{t, a} = 1$ and (iii) $r_{t, s} = 1$ , we have that the lower bound on $p$ solves

λ_{t . r^{-}} = (1 - p) \frac{r_{t, p} λ_{t . c o}}{r_{t, p} λ_{t . c o} + 1 - λ_{t . c o}} + p λ_{t . r^{+}}

Software

The supplemental file rcode_for_paper.R contains R (R Development Core Team, 2016) code for replicating the analyses in our paper.

Appendix 1

Supplemental results

Sample sizes and missing data proportions

Appendix 1—table 1 shows the sample sizes of the Ret+ CM, Ret- CM, non-malaria hospital controls and community controls. Appendix 1—table 2 shows the proportion of missing data on the genetic traits for each of these four groups. Appendix 1—table 3 shows the proportion of missing data on the demographic and clinical variables for the Ret+ CM, Ret- CM and non-malaria hospital control groups.

Appendix 1—table 1

Sample sizes.

https://doi.org/10.7554/eLife.23699.008

	Retinopathy-positive CM	Retinopathy-negative CM	Non-malaria hospital controls	Community controls
Sample Size	438	288	204	3704

Appendix 1—table 2

Missing data proportions for genetic traits.

https://doi.org/10.7554/eLife.23699.009

	Retinopathy-positive CM	Retinopathy-negative CM	Non-malaria hospital controls	Community controls
Sickle Cell Trait	.002	0	.054	.010
Blood Group	.011	.003	.025	.043

Appendix 1—table 3

Missing data proportions for demographic and clinical variables.

https://doi.org/10.7554/eLife.23699.010

	Retinopathy-positive CM	Retinopathy-negative CM	Non-malaria hospital controls
Female	.018	.021	.005
Age (months)	0	0	.005
Mid-upper arm circumference (cm)	.016	.014	.054
Weight (kg)	0	0	0
Height (cm)	.009	.024	.034
Temperature (°C)	0	0	0
Pulse rate – beats/minute	.002	0	.010
Respiratory rate – breaths/minute	0	0	.005
Liver size – cm below costal margin	.009	.024	.010
Spleen size – cm below costal margin	.005	.014	.010
Deep breathing	.007	.021	0
Blantyre Coma Score:	0	0	0
CSF opening pressure – mm of water	.420	.330	.623
Hematocrit -- %	.009	.024	.034
Platelets	.153	.160	.132
Malaria parasitemia – parasites/mm³	.039	.042	.025
White blood cells	.082	.097	.118
Lactate – mmol/liter	.653	.753	.564
Blood glucose – mmol/liter	.014	.003	0
CSF white cell count – % ≥ 5	.277	.170	.275
Blood culture positive for pathogen	.039	.024	.059
HIV positive	.144	.153	.353
Discharge outcome	0	.007	0

Causes of death in autopsy study

For children dying with Ret+ CM, among the children with cerebral sequestration, one had pneumonia and one had pneumonia and meningoencephalitis. The one child with Ret+CM but without cerebral sequestration had a likely cause of death of anemia. Six children with Ret+CM and cerebral sequestration had severe anemia which was thought to be due to malaria parasitemia. Among the 15 children dying with Ret- CM, three had Reye’s syndrome and pneumonia, three had pneumonia alone, one had pneumonia with spread to meninges, one had pneumonia with meningoencephalitis, one had hepatic necrosis, one had hepatitis, one had skull fractures, one had subdural/intracerebral hematomas, one had left ventricular failure with pulmonary edema, one had septicemia and one had an unknown cause of death.

Sensitivity to assumptions about false discovery rate and false omission rates for detecting malarial retinopathy

For testing the null hypothesis vs. the alternative hypothesis for the sickle cell trait, the p-value remains below. 0001 as $F D R$ and $F O R$ are varied up to the upper bounds of .30 and .11 respectively. For blood group O, the p-value increases to 0.94 when $F D R$ and $F O R$ are set at their upper bounds of .30 and .11 respectively.

Results using community controls only

Appendix 1—table 4 shows the inferences from our model for comparing true Ret+ CM to controls and true Ret- CM to controls (Materials and methods) when we use only the community controls.

Appendix 1—table 4

Odds ratio comparing community controls to true Ret+ CM and Ret- CM groups, which account for the fact that there is measurement error in observed retinopathy status (false discovery rate = 0.07 and false omission rate = 0.05). The p-values are two-sided p-values for testing that the odds ratio equals 1.

https://doi.org/10.7554/eLife.23699.011

	Odds ratio (95% CI)	p-value
HbAS
Controls vs. true Ret- CM	14.43 (3.23, 258.94)	<0.0001
Controls vs. true Ret+ CM	1234.96 (9.93, $\infty$ )	<0.0001
BGO
Controls vs. true Ret- CM	1.03 (0.83, 1.29)	0.79
Controls vs. true Ret+ CM	1.23 (1.01, 1.51)	0.04

Data availability

The following previously published data sets were used

1. Malaria Genomic Epidemiology Network
(2015) Genome-wide study of resistance to severe malaria
Access to this data is available through application to the MalariaGen Independent Data Access Committee. Please see https://www.malariagen.net/data/terms-use/human-gwas-data for more information.

https://www.malariagen.net/data/genome-wide-study-resistance-severe-malaria-eleven-populations

References

(2005) A comparison of case-control and family-based association methods: the example of sickle-cell and malaria
Annals of Human Genetics 69:559–565.
https://doi.org/10.1111/j.1529-8817.2005.00180.x
- PubMed
- Google Scholar
1. Barrera V
2. Hiscott PS
3. Craig AG
4. White VA
5. Milner DA
6. Beare NA
7. MacCormick IJ
8. Kamiza S
9. Taylor TE
10. Molyneux ME
11. Harding SP
(2015) Severity of retinopathy parallels the degree of parasite sequestration in the eyes and brains of malawian children with fatal cerebral malaria
Journal of Infectious Diseases 211:1977–1986.
https://doi.org/10.1093/infdis/jiu592
- PubMed
- Google Scholar
1. Bearden D
(2012) Retinopathy in cerebral malaria: new answers, new puzzles
Neurology 79:1196–1197.
https://doi.org/10.1212/WNL.0b013e31826aad9d
- PubMed
- Google Scholar
(2002) Inter-observer concordance in grading retinopathy in cerebral malaria
Annals of Tropical Medicine & Parasitology 96:105–108.
https://doi.org/10.1179/000349802125000565
- PubMed
- Google Scholar
(2006) Malarial retinopathy: a newly established diagnostic sign in severe malaria
The American Journal of Tropical Medicine and Hygiene 75:790–797.
https://doi.org/10.4269/ajtmh.2006.75.790
- PubMed
- Google Scholar
(1998)
Encyclopedia of Biostatistics

Attributable Risk, Encyclopedia of Biostatistics, United States, John Wiley & Sons.
- Google Scholar
1. Berkson J
(1946) Limitations of the application of fourfold table analysis to hospital data
Biometrics Bulletin 2:47–53.
https://doi.org/10.2307/3002000
- PubMed
- Google Scholar
1. Cserti CM
2. Dzik WH
(2007) The ABO blood group system and plasmodium falciparum malaria
Blood 110:2250–2258.
https://doi.org/10.1182/blood-2007-03-077602
- PubMed
- Google Scholar
1. Diallo TO
2. Remoue F
3. Gaayeb L
4. Schacht AM
5. Charrier N
6. De Clerck D
7. Dompnier JP
8. Pillet S
9. Garraud O
10. N'Diaye AA
11. Riveau G
(2010) Schistosomiasis coinfection in children influences acquired immune response against Plasmodium falciparum malaria antigens
PLoS One 5:e12764.
https://doi.org/10.1371/journal.pone.0012764
- PubMed
- Google Scholar
1. Greenland S
2. Robins JM
(1988) Conceptual problems in the definition and interpretation of attributable fractions
American Journal of Epidemiology 128:1185–1197.
https://doi.org/10.1093/oxfordjournals.aje.a115073
- PubMed
- Google Scholar
1. Joshi V
2. Agurto C
3. Barriga S
4. Nemeth S
5. Soliz P
6. MacCormick IJ
7. Lewallen S
8. Taylor TE
9. Harding SP
(2017) Automated detection of malarial retinopathy in digital fundus images for improved diagnosis in malawian children with clinically defined cerebral malaria
Scientific Reports 7:42703.
https://doi.org/10.1038/srep42703
- PubMed
- Google Scholar
1. Lewallen S
2. Harding SP
3. Ajewole J
4. Schulenburg WE
5. Molyneux ME
6. Marsh K
7. Usen S
8. White NJ
9. Taylor TE
(1999) A review of the spectrum of clinical ocular fundus findings in P. falciparum malaria in African children with a proposed classification and grading system
Transactions of the Royal Society of Tropical Medicine and Hygiene 93:619–622.
https://doi.org/10.1016/S0035-9203(99)90071-8
- PubMed
- Google Scholar
1. MacCormick IJ
2. Maude RJ
3. Beare NA
4. Borooah S
5. Glover S
6. Parry D
7. Leach S
8. Molyneux ME
9. Dhillon B
10. Lewallen S
11. Harding SP
(2015) Grading fluorescein angiograms in malarial retinopathy
Malaria Journal 14:367.
https://doi.org/10.1186/s12936-015-0897-7
- PubMed
- Google Scholar
1. Malaria Genomic Epidemiology Network
(2008) A global network for investigating the genomic epidemiology of malaria
Nature 456:732–737.
https://doi.org/10.1038/nature07632
- PubMed
- Google Scholar
1. Mallewa M
2. Vallely P
3. Faragher B
4. Banda D
5. Klapper P
6. Mukaka M
7. Khofi H
8. Pensulo P
9. Taylor T
10. Molyneux M
11. Solomon T
(2013) Viral CNS infections in children from a malaria-endemic area of Malawi: a prospective cohort study
The Lancet Global Health 1:e153–e160.
https://doi.org/10.1016/S2214-109X(13)70060-3
- PubMed
- Google Scholar
1. Migot-Nabias F
2. Mombo LE
3. Luty AJ
4. Dubois B
5. Nabias R
6. Bisseye C
7. Millet P
8. Lu CY
9. Deloron P
(2000) Human genetic factors related to susceptibility to mild malaria in Gabon
Genes and Immunity 1:435–441.
https://doi.org/10.1038/sj.gene.6363703
- PubMed
- Google Scholar
1. Milner DA
2. Lee JJ
3. Frantzreb C
4. Whitten RO
5. Kamiza S
6. Carr RA
7. Pradham A
8. Factor RE
9. Playforth K
10. Liomba G
11. Dzamalala C
12. Seydel KB
13. Molyneux ME
14. Taylor TE
(2015) Quantitative assessment of multiorgan sequestration of parasites in fatal pediatric cerebral malaria
Journal of Infectious Diseases 212:1317–1321.
https://doi.org/10.1093/infdis/jiv205
- PubMed
- Google Scholar
1. Miura K
(2016) Progress and prospects for blood-stage malaria vaccines
Expert Review of Vaccines 15:765–781.
https://doi.org/10.1586/14760584.2016.1141680
- PubMed
- Google Scholar
1. Modiano D
2. Luoni G
3. Sirima BS
4. Simporé J
5. Verra F
6. Konaté A
7. Rastrelli E
8. Olivieri A
9. Calissano C
10. Paganotti GM
11. D'Urbano L
12. Sanou I
13. Sawadogo A
14. Modiano G
15. Coluzzi M
(2001) Haemoglobin C protects against clinical plasmodium falciparum malaria
Nature 414:305–308.
https://doi.org/10.1038/35104556
- PubMed
- Google Scholar
1. Méndez D
2. Hernáez ML
3. Kamali AN
4. Diez A
5. Puyet A
6. Bautista JM
(2012) Differential carbonylation of cytoskeletal proteins in blood group O erythrocytes: potential role in protection against severe malaria
Infection, Genetics and Evolution 12:1780–1787.
https://doi.org/10.1016/j.meegid.2012.06.013
- Google Scholar
1. Nedelman J
(1988) The prevalence of malaria in Garki, Nigeria: double sampling with a fallible expert
Biometrics 44:635–655.
https://doi.org/10.2307/2531580
- PubMed
- Google Scholar
1. Ntoumi F
2. Mercereau-Puijalon O
3. Ossari S
4. Luty A
5. Reltien J
6. Georges A
7. Millet P
(1997) Plasmodium falciparum: sickle-cell trait is associated with higher prevalence of multiple infections in gabonese children with asymptomatic infections
Experimental Parasitology 87:39–46.
https://doi.org/10.1006/expr.1997.4173
- PubMed
- Google Scholar
1. Obaro S
2. Greenwood B
(2011) Malaria and bacteraemia in African children
The Lancet 378:1281–1282.
https://doi.org/10.1016/S0140-6736(11)61146-X
- Google Scholar
1. Postels DG
2. Birbeck GL
(2011) Children with retinopathy-negative cerebral malaria: a pathophysiologic puzzle
The Pediatric Infectious Disease Journal 30:953–956.
https://doi.org/10.1097/INF.0b013e3182271c69
- PubMed
- Google Scholar
Software
1. R Development Core Team
(2016)
R: A Language and Environment for Statistical Computing

R Foundation for Statistical Computing, Vienna , Austria.
1. Rice JA
(2007)
Mathematical Statistics and Data Analysis Vol. 72

Mathematical Statistics and Data Analysis Vol. 72, 3rd edn, Belmont, CA, United States: Duxbury Press.
- Google Scholar
1. Malaria Genomic Epidemiology Network
(2014) Reappraisal of known malaria resistance loci in a large multicenter study
Nature Genetics 46:1197–1204.
https://doi.org/10.1038/ng.3107
- PubMed
- Google Scholar
1. Rosenbaum PR
(1987) The role of a second control group in an observational study
Statistical Science 2:292–306.
https://doi.org/10.1214/ss/1177013232
- Google Scholar
Book
1. Rosenbaum P
(2002) Observational Studies (2nd edn)
New York: Springer-Verlag.
https://doi.org/10.1007/978-1-4757-3692-2
- Google Scholar
1. Rothman KJ
(1976) Causes
American Journal of Epidemiology 104:587–592.
https://doi.org/10.1093/oxfordjournals.aje.a112335
- PubMed
- Google Scholar
1. Rowe JA
2. Handel IG
3. Thera MA
4. Deans AM
5. Lyke KE
6. Koné A
7. Diallo DA
8. Raza A
9. Kai O
10. Marsh K
11. Plowe CV
12. Doumbo OK
13. Moulds JM
(2007) Blood group O protects against severe plasmodium falciparum malaria through the mechanism of reduced rosetting
PNAS 104:17471–17476.
https://doi.org/10.1073/pnas.0705390104
- PubMed
- Google Scholar
1. Scott JA
2. Berkley JA
3. Mwangi I
4. Ochola L
5. Uyoga S
6. Macharia A
7. Ndila C
8. Lowe BS
9. Mwarumba S
10. Bauni E
11. Marsh K
12. Williams TN
(2011) Relation between falciparum malaria and bacteraemia in kenyan children: a population-based, case-control study and a longitudinal study
The Lancet 378:1316–1323.
https://doi.org/10.1016/S0140-6736(11)60888-X
- PubMed
- Google Scholar
1. Seydel KB
2. Fox LL
3. Glover SJ
4. Reeves MJ
5. Pensulo P
6. Muiruri A
7. Mpakiza A
8. Molyneux ME
9. Taylor TE
(2012) Plasma concentrations of parasite histidine-rich protein 2 distinguish between retinopathy-positive and retinopathy-negative cerebral malaria in malawian children
Journal of Infectious Diseases 206:309–318.
https://doi.org/10.1093/infdis/jis371
- PubMed
- Google Scholar
1. Seydel KB
2. Kampondeni SD
3. Valim C
4. Potchen MJ
5. Milner DA
6. Muwalo FW
7. Birbeck GL
8. Bradley WG
9. Fox LL
10. Glover SJ
11. Hammond CA
12. Heyderman RS
13. Chilingulo CA
14. Molyneux ME
15. Taylor TE
(2015) Brain swelling and death in children with cerebral malaria
New England Journal of Medicine 372:1126–1137.
https://doi.org/10.1056/NEJMoa1400116
- PubMed
- Google Scholar
1. Smith GD
2. Ebrahim S
(2003) 'Mendelian randomization': can genetic epidemiology contribute to understanding environmental determinants of disease?
International Journal of Epidemiology 32:1–22.
https://doi.org/10.1093/ije/dyg070
- PubMed
- Google Scholar
1. Smith TA
(2007) Measures of clinical malaria in field trials of interventions against plasmodium falciparum
Malaria Journal 6:53.
https://doi.org/10.1186/1475-2875-6-53
- PubMed
- Google Scholar
1. Taylor TE
2. Fu WJ
3. Carr RA
4. Whitten RO
5. Mueller JS
6. Fosiko NG
7. Lewallen S
8. Liomba NG
9. Molyneux ME
10. Mueller JG
(2004) Differentiating the pathologies of cerebral malaria by postmortem parasite counts
Nature Medicine 10:143–145.
https://doi.org/10.1038/nm986
- PubMed
- Google Scholar
(2012) Haemoglobinopathies and the clinical epidemiology of malaria: a systematic review and meta-analysis
The Lancet Infectious Diseases 12:457–468.
https://doi.org/10.1016/S1473-3099(12)70055-5
- PubMed
- Google Scholar
1. Uneke CJ
(2007) Plasmodium falciparum malaria and ABO blood group: is there any relationship?
Parasitology Research 100:759–765.
https://doi.org/10.1007/s00436-006-0342-5
- PubMed
- Google Scholar
1. WHO
(2000) Severe falciparum malaria
Transactions of the Royal Society of Tropical Medicine and Hygiene 94:1–90.
https://doi.org/10.1016/S0035-9203(00)90300-6
- Google Scholar
1. WHO
(2014) Severe malaria
Tropical Medicine & International Health 19:7–131.
https://doi.org/10.1111/tmi.12313_2
- PubMed
- Google Scholar
1. WHO
(2015)
World Malaria Report 2015

World Malaria Report 2015, 243, 978 92 4 1564403.
- Google Scholar
1. Westreich D
(2012) Berkson's bias, selection bias, and missing data
Epidemiology 23:159–164.
https://doi.org/10.1097/EDE.0b013e31823b6296
- PubMed
- Google Scholar
(1983) A case-control study in Northern Liberia of plasmodium falciparum malaria in haemoglobin S and beta-thalassaemia traits
Annals of Tropical Medicine & Parasitology 77:239–246.
https://doi.org/10.1080/00034983.1983.11811704
- PubMed
- Google Scholar
1. Williams TN
2. Mwangi TW
3. Wambua S
4. Alexander ND
5. Kortok M
6. Snow RW
7. Marsh K
(2005) Sickle cell trait and the risk of plasmodium falciparum malaria and other childhood diseases
The Journal of Infectious Diseases 192:178–186.
https://doi.org/10.1086/430744
- PubMed
- Google Scholar
1. Wilson EB
(1927) Probable Inference, the law of Succession, and statistical inference
Journal of the American Statistical Association 22:209–212.
https://doi.org/10.1080/01621459.1927.10502953
- Google Scholar

Decision letter

Ben Cooper

Reviewing Editor; Mahidol Oxford Tropical Medicine Research Unit, Thailand

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Evidence from a Natural Experiment that Malaria Parasitemia is Pathogenic in Retinopathy-Negative Cerebral Malaria" for consideration by eLife. Your article has been favorably evaluated by Prabhat Jha (Senior Editor) and three reviewers, one of whom, Ben Cooper (Reviewer #1), is a member of our Board of Reviewing Editors. The following individual involved in review of your submission have agreed to reveal their identity: Chandy John (Reviewer #3).

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

The paper uses data on the distribution of the sickle cell trait and blood type O in different populations to make inferences about the causal role that malaria parasitemia plays in retinopathy-negative cerebral malaria. The work provides strong evidence that malaria parasitemia does contribute to the pathogenesis of retinopathy-negative CM and provides a lower bound malaria parasitemia attributable fraction in retinopathy-negative cerebral malaria.

Summary:

The reviewers agreed that this is potentially important work, though there was a consensus that it would be far more interesting if it could go beyond the focus on hypothesis testing and expand on the work on estimating the attributable fraction (at present only a lower bound on the lower 95% confidence bound is given). It was felt that this was particularly important given that the null hypothesis (no causal role of malaria in ret-CM) is widely seen as a quite an eccentric position in light of autopsy studies (though, admittedly, it is held by some). In addition to this, there was agreement that the work needed to be better motivated with a more complete discussion of relevant previous work in the Introduction and a more complete consideration of alternative hypotheses that might explain the findings in the Discussion. For example, are there other factors (human or parasite) that interact between uncomplicated malaria and blood group O but not HbAS to alter risk of retneg CM?

Essential revisions:

The reviewers agreed that all the points made in the full reviews below should be addressed. In particular:

i) The Introduction should contain a full explanation of relevant autopsy studies with numbers, logic of hypothesis, and the motivation for the work more clearly explained.

ii) Unless there are compelling reasons why this is not possible (if there are, these should be given), the authors should seek to provide point estimates of the attributable fraction and upper and lower uncertainty bounds. This is likely to require additional assumptions (and may well be facilitated by working within a Bayesian framework), in which case analysis of sensitivity to these assumptions/priors should be provided.

iii) The Discussion should be expanded to contain a fuller discussion of alternative hypotheses explaining the data (see in particular comment 3 from reviewer #3) and limitations of the study (see in particular comments from reviewer #2).

iv) Points 4 and 5 raised by reviewer #1 require at the minimum some clarification.

Reviewer #1:

The paper uses data on the distribution of the sickle cell trait and blood type O in different populations to make inferences about the causal role that malaria parasitemia plays in retinopathy-negative cerebral malaria. This is a novel and methodologically interesting application of Mendelian randomization ideas which have become very popular in non-communicable disease epidemiology but which are still little used in the communicable disease community. Despite evidence from autopsy studies of patients with retinopathy-negative cerebral malaria (ret-CM), there is still some debate about the role malaria plays in ret-CM so this work does address a real problem. Overall the paper is well-written and does a fairly good job of explaining quite a difficult subject. I believe the findings do represent a substantial advance over previous work.

1) "An autopsy study found […] children dying with ret- CM mostly lacked cerebral sequestration and mostly had non-malarial etiologies of death (Taylor et al. 2004)"

But I think some of the those dying with ret-CM did have cerebral sequestration which seems to contradict the null. Perhaps as well as emphasising the small sample size, this would be a good time to point out that false positives can occur in retinopathy so the small histopathology studies do not provide conclusive evidence against the null. It also suggests to me that rather more prominence should be given to the estimated lower bound on the malaria parasitemia attributable fraction of ret- CM which currently doesn't get a mention in the abstract. To me, this seems like the most important result.

2) I think it's a good idea to illustrate the ideas here with an idealised unethical RCT, but the example given (Introduction, third paragraph) seems to add to the confusion as it depends on patient outcomes of those with ret-CM, which are not considered in this paper. I think a more relevant (if more expensive) impossible RCT would be to randomize a very large number patients with uncomplicated malaria illness to treatment vs. no-treatment. If the null is true there would be no difference in the incidence of ret-CM in the two arms. This leads on to the natural experiment, as HbAS/BGO are effectively doing similar randomisations.

3) The work rests on the assumption that HbAS has no effect on asymptomatic parasitemia (if it did, different prevalences of HbAS might be expected in those with Ret-CM and the non-malaria controls even if malaria played no role in the CM). While this seems approximately true (and if there is an effect it is likely to be small) this assumption may not be completely rock solid. A meta analysis on protection by HbAS against parasitemia (pubmed id 22445352) concluded "Taken together, HbAS does not consistently protect from P. falciparum parasitemia." Some studies have found protection though, but not consistently. This is something the authors might want to consider addressing in a sensitivity analysis.

4) The FP and FN rate as defined in the subsection “Hypotheses” are the conventional definitions (i.e. FP = Pr(test +ve | true -ve), FN=Pr(test -ve| true +ve). Equations in the second to last paragraph of the aforementioned subsection and elsewhere: what is written as FP seems to be Pr(true -ve | test+ve) and similarly for FN. So, I think what is written as FP is actually the FDR (false discovery rate) while what is written as 1-FP is the PPV (prob true +ve | test +ve). Similarly, I think what is given as FN is the FOR (false omission rate) and 1-FN is really the NPV. Fortunately, it looks like the calculations later on for "FP" are actually calculating FDR (which is just what is needed). So, I think the numbers are all OK, but the labels need some attention. Note that checking the calculations to see if what is labelled as FN is actually FOR is more difficult and I haven't done this, but it should be checked.

5) Subsection “Inference for Proportion of Retinopathy-Negative CM due to path (a) in Figure 1:” onwards confused me. If p=0 (so all ret-CM comes from path B – that is, all are causally linked to malaria illness) then the proportion of HbAS in the ret-CM patients will be the same as in the control group. This doesn't make sense. I couldn't make sense of the next equation either. Is there a typo here?

Reviewer #2:

In a previous autopsy study, the authors observed that cases of cerebral malaria with retinal changes children had cerebral sequestration of parasitized red blood cells in the cerebral microvasculature and no other identified causes of death besides the malaria parasitemia but in those who were retinopathy negative, few had cerebral sequestration and most had non-malaria etiologies of death. Here, the authors use the fact that HbAS (and blood group O) have been found to protect against symptomatic malaria (or cerebral sequestration) to test whether malaria parasitemia contributes to the deaths in CM ret- by comparing the frequencies of sickle cell in CM ret- vs. controls. Studies using sickle cell and other genetic markers have been used in several studies to test hypotheses, but here this is placed on a more formal footing in the framework of Mendelian randomization.

1) The paper frames the problem as being binary in the Introduction and Methods: e.g. "If a given trait decreases the probability of developing ret+ but not ret- CM, this would suggest that malaria parasites are pathogenic for ret+ CM but not ret- CM or the trait affects an aspect of disease not causal to the development of ret- CM".e.g. "we test the null hypothesis that the trait frequency is the same in controls and true ret- CM cases vs. the alternative hypothesis that the trait frequency is higher in control than true ret- cases."

However, it is likely that a proportion of CM ret- and a very high proportion of CM ret+ cases were caused or co-caused by the parasitemia. In both groups, there could be a non-zero proportion of coma caused by non-malaria, unless having CM ret+ excludes other causes.

So, it seems that using significance to test for an association between CM ret- and controls and their sickle cell status does not capture the question. Estimation would be more useful.

The binary stance is softened later, in the Results section, "have found evidence that malaria parasitemia is on the causal pathway to a substantial proportion of ret- CM cases". However, we do not have an estimate of what the proportion is, and this seems to be mostly a guess. A model which estimates the attributable fraction could be considered.

2) The main results are not inconsistent with the hypothesis of the authors, but neither are they strongly supportive of it.

For HbAS, there was a lower frequency of HbAS in the CM cases compared to the controls, which fits with HbAS being protective against symptomatic malaria. However, there is little evidence that the proportions of HbAS for CM ret+ and CM ret- are similar: there is only 1 HbAS in the two groups, so a non-significant result is inevitable, and the confidence interval stretches from 0.18 to infinity.

For blood group O, the authors stated that they would expect to see a higher proportion of children with O in the CM ret- compared to CM ret+ group if parasites did not cause CM ret- symptoms since blood group O protects against adherence and severe malaria in the brain. The estimate is in the right direction but the CI includes 1 (0.88, 1.62), and so the result in inconclusive. There is a fair amount of overlap in the CIs for controls vs. CM ret- and controls versus CM ret+, and comparing the p-values would not be appropriate due to a non-significant result indicating "no evidence of a difference" rather than "no difference", and being partly reliant on the sample sizes, which differ.

Comparability of the groups

3) There is a little discussion of the biases for the controls, but none for the CM cases. More discussion would be helpful. It may perhaps be that some CM cases do not reach hospital because they die more quickly, or that the older controls have selection bias for HbAS.

Both coma score and retinopathy may differ over the course of the illness, and what is seen is just a snapshot at the time of admission.

Can the differences in the Blantyre coma score and other clinical indicators for ret- and ret+ CM in Table 1 be explained?

The controls are mostly from cord blood samples and some from non-malaria hospital admissions. They are combined in the analysis. While this is not unreasonable, it does make it hard to interpret the results.

4) There would be a need to explain the results of the Taylor et al. 2004 study in the light of the findings.

Reviewer #3:

This is an elegant study that uses the genetic traits of HbS trait and blood group O, both well described as protecting against severe malaria, to assess whether malaria contributes to the disease process (i.e., development of coma) in children with cerebral malaria (CM) who are retinopathy positive (ret+) vs. retinopathy negative (ret-).

The authors find that HbS is protective against ret+ and ret- CM, while blood group O is protective only against ret+ CM. They posit that this is because HbS protects in the step where asymptomatic parasitemia develops into uncomplicated malaria, while blood group O only protects only in the step from uncomplicated malaria to severe malaria. In this scenario, ret-CM is due to the presence of uncomplicated malaria plus some other factor that leads to coma. Children with HbS are protected from CM, including ret- CM, because they are protected at the stage where asymptomatic parasitemia becomes uncomplicated malaria. However, children with blood group O are protected only against ret+ CM, because in ret-CM, children with blood group O can still develop uncomplicated malaria, and need only the other factor to tip them into coma. They conclude that malaria likely does play a role in some/many cases of ret- CM.

I find the idea behind this innovative assessment compelling. The paper is well written, and the data appear solid, with much larger numbers than most CM studies are able to obtain. So, overall I think this is a well conducted study and of significant interest to the field. I have only a few comments.

1) Some of the "non-malaria" hospital controls had malaria parasitemia. Since children can sometimes develop severe malaria with relatively low levels of parasitemia, how can the authors be sure that malaria did not contribute to illness in these control participants?

2) The authors pioneered assessment of PfHRP2 levels in children with CM, and these were shown to distinguish nicely between RP and RN CM. Do they have PfHRP2 levels on even a subset of these individuals? It would be nice to further differentiate whether, for example, those with low or undetectable PfHRP2 had lower mortality, or less/no protection from HbS, for example.

3) I think the authors' hypothesis is the most satisfactory one, but it seems possible that the results could be explained by protection afforded by HbS against development of severe malaria from uncomplicated malaria that is not afforded by BGO in children with ret- CM. In other words, there may be interaction with ret- CM and BGO such that in ret- children, whatever mechanism BGO provides that affords protection in going from uncomplicated malaria to CM is interfered with or less effective. I am not sure what that factor could be, but it could potentially relate to things like parasite genotype. Have the authors considered this possibility as an explanation?

4) Are there any clinical or other indicators that suggest to the authors what the other factor(s) may be in ret- CM that lead children with uncomplicated malaria to go on to develop coma?

https://doi.org/10.7554/eLife.23699.014

Author response

Essential revisions:

The reviewers agreed that all the points made in the full reviews below should be addressed. In particular:

i) The Introduction should contain a full explanation of relevant autopsy studies with numbers, logic of hypothesis, and the motivation for the work more clearly explained.

We now provide results in the Introduction and Appendix 1 from the autopsy study described by Taylor et al. (2004) that update the results reported in Taylor et al. (2004) to include patients enrolled after 2004. In the Introduction, we now say:

“Autopsy data show that children dying of Ret+ CM have a high degree of sequestration of parasitized erythrocytes in cerebral vasculature, considered the pathological hallmark of this condition (Taylor et al. 2004). […] In the only autopsy study among children dying with CM we are aware of (Taylor et al. 2004; follow-up results in Milner et al. 2015; Barrera et al. 2015), among children for whom retinopathy was assessed, 41 of 42 children dying with Ret+ CM had substantial cerebral sequestration of parasitized red blood cells in the cerebral microvasculature (defined as ≥23% of cerebral capillaries had sequestration) and mostly lacked other identified potential causes of death besides the malaria parasitemia, whereas all 15 children dying with Ret- CM lacked substantial cerebral sequestration (<23% of cerebral capillaries had sequestration) and mostly had non-malarial etiologies of death (see Appendix 1 for causes of death); these numbers update those in Taylor et al. (2004) to include patients enrolled after 2004.”

Taylor et al. (2004) and follow-up papers is the only autopsy study we are aware of among children dying with CM.

We have sought to explain more clearly the motivation for our work in the Introduction. In particular, first, we have added to the Introduction the following motivation for our work:

“Incidental malaria parasitemia is common in people living in areas of high malaria transmission. […] Our aim in the research presented here was to assess the contribution of acute malaria infection in the pathophysiology of retinopathy negative CM. “

Second, as suggested by reviewer 1, we have considered a different, clearer infeasible randomized experiment, which would address our hypothesis of interest that malaria parasites are involved in the pathogenesis of some Ret- CM cases. We then explain how considering the relationship between inherited genetic traits and cerebral malaria provides a natural analogue of this infeasible randomized experiment. See our response to comment 2 of reviewer 1.

We have added Table 3 that provides results for point estimates and uncertainty bounds for the attributable fraction. We have rewritten the section in the supplement “Inference for Proportion of Retinopathy-Negative CM due to path (a) in Figure 1” and now name it as “Inference for Malaria Parasitemia Attributable Fraction for Coma among Ret- CM Cases.” The section now formulates in more detail the causal assumptions underlying our inference about the attributable fraction using the sufficient-component cause framework (Rothman, 1976).

We have added fuller discussion and alternative hypotheses explaining the data to the Discussion as follows:

“In Figure 1 pathway (b), Ret- CM results from an interaction between malaria illness and other illness. […] The protection provided against Ret- CM by HbAS but not BGO could be explained by an interaction between the factor(s) affecting whether malarial retinopathy is present and BGO, e.g., a parasite genotype which causes malarial retinopathy to be absent could interfere with the mechanism by which BGO provides protection against complicated malaria.”

iv) Points 4 and 5 raised by reviewer #1 require at the minimum some clarification.

See our responses to reviewer #1 in which we have made corrections to account for the good points raised by reviewer #1.

Reviewer #1:

[…] 1) "An autopsy study found […] children dying with ret- CM mostly lacked cerebral sequestration and mostly had non-malarial etiologies of death (Taylor et al. 2004)"

We now provide results from the autopsy study described by Taylor et al. (2004) that update the results reported in Taylor et al. (2004) to include patients enrolled after 2004. We now say, “In the only autopsy study among children dying with CM we are aware of (Taylor et al. 2004; follow-up results in Milner et al. 2015; Barrera et al. 2015), among children for whom retinopathy was assessed, 41 of 42 children dying with Ret+ CM had substantial cerebral sequestration of parasitized red blood cells in the cerebral microvasculature (defined as ≥23% of cerebral capillaries had sequestration) and mostly lacked other identified potential causes of death besides the malaria parasitemia, whereas all 15 children dying with Ret- CM lacked substantial cerebral sequestration (<23% of cerebral capillaries had sequestration) and mostly had non-malarial etiologies of death (see Appendix 1 for causes of death); these numbers update those in Taylor et al. (2004) to include patients enrolled after 2004.”

Thanks for the good suggestion to put more emphasis on the malaria parasitemia attributable fraction of Ret- CM. We have now added a sentence to the Abstract, “A lower bound for the fraction of retinopathy-negative CM that would be eliminated if malaria parasitemia were to be eliminated is estimated to be 0.93 (95% confidence interval: 0.68, 1).”

We have also added Table 3 to present the main model and sensitivity analysis results for the malaria parasitemia attributable fraction of Ret- CM and have expanded the description in the results to:

“In Methods, we formulate a sufficient-component cause model (Rothman 1976a) based on Figure 1 and describe how to make inferences about the fraction of Ret- CM cases that are due to pathway (b) in Figure 1, i.e., the malaria parasitemia attributable fraction of Ret- CM (the fraction of Ret- CM cases that would be prevented if malaria parasitemia were eliminated (Benichou et al. 1998)). […] Under all scenarios considered, we found evidence for a substantial contribution of malaria parasites to the pathogenesis of Ret- CM with lower 95% confidence bounds ranging from.37 to.77 and point estimates for the lower bound ranging from.86 to.95.”

Thanks for this valuable suggestion. We agree with your comment that the idealized unethical RCT originally given added confusion as it depended on outcomes of those with Ret- CM which are not considered in the paper. As an alternative, we considered the following impossible RCT based on your suggestion – randomize a large number of children to either (i) be followed up and treated with ACT as soon as signs of uncomplicated malaria illness start to develop or (ii) left untreated when signs of uncomplicated malaria illness start to develop. However, in thinking about this trial, we ran into the following difficulty. Once a child is treated with ACT, the ACT will clear the malaria parasites. Consequently, even if the child has a non-malarial infection that would lead to coma without malarial retinopathy, the child’s coma would be classified as a non-malarial coma rather than retinopathy-negative CM because the child no longer has malaria parasitemia. Thus, even if malaria parasites play no causal role in retinopathy-negative CM, we would expect the treatment arm in this trial to have less retinopathy-negative CM.

To address your valuable comment about the problem with our original idealized, unethical RCT, we formulated a different impossible RCT based on a hypothetical blood stage malaria vaccine as follows:

“Whether malaria parasites play a role in the pathogenesis of Ret- CM could in principle be tested by a randomized experiment. […] If the trait decreases the probability of developing Ret+ CM but not Ret- CM, this would suggest that malaria parasites are pathogenetic for Ret+ CM but are either not pathogenetic for Ret-CM or the trait affects an aspect of disease not causal to the development of Ret- CM.”

3) The work rests on the assumption that HbAS has no effect on asymptomatic parasitemia (if it did, different prevalences of HbAS might be expected in those with Ret-CM and the non-malaria controls even if malaria played no role in the CM). While this seems approximately true (and if there is an effect it is likely to be small) this assumption may not be completely rock solid. A meta analysis on protection by HbAS against parasitemia (pubmed id 22445352) concluded "Taken together, HbAS does not consistently protect from P. falciparum parasitemia." Some studies have found protection though, but not consistently. This is something the authors might want to consider addressing in a sensitivity analysis.

Thanks for bringing up this important point. We have now made the assumption that HbAS has no effect on asymptomatic parasitemia clear in the Introduction as follows:

“Analogous to the randomized trial described above, possession vs. lack of a malaria protective trait assigns children to an arm in which some malaria illness is prevented vs. not prevented. […] For both BGO and HbAS, it is plausible that the traits do not protect against malaria parasitemia incidence as systematic reviews have not found consistent evidence for protection (Uneke 2007; S. M. Taylor, Parobek, and Fairhurst 2012); we will assume no protection for our main analysis but do sensitivity analyses that allow for protection.”

We now provide a sensitivity analysis to violations of the assumption that HbAS has no effect on asymptomatic parasitemia in Table 3. We discuss this sensitivity analysis in the Results as follows:

“For the sensitivity analyses, although the systematic review of Taylor et al. (2012) found no consistent evidence that HbAS reduces malaria parasitemia incidence, some studies reviewed found protection and we consider sensitivity analyses that allow for a small amount of protection (10%) and the largest amount of protection found in all the studies reviewed (41%).”

Thanks for pointing out these errors on our part. We have now corrected them. We have checked that what was labelled as FN is actually FOR and labeled it correctly. We have rewritten the section “Estimating False Discovery and False Omission Rates for Malarial Retinopathy” (and renamed it) to account for the fact that we need to estimate the false discovery and false omission rates rather than the false positive and false negative rates.

5) Subsection “Inference for Proportion of Retinopathy-Negative CM due to path (a) in Figure 1:” onwards confused me. If p=0 (so all ret-CM comes from path B – that is, all are causally linked to malaria illness) then the proportion of HbAS in the ret-CM patients will be the same as in the control group. This doesn't make sense. I couldn’t make sense of the next equation either. Is there a typo here?

Thanks for pointing out these errors. We have rewritten this section in the supplement and now name it as “Inference for Malaria Parasitemia Attributable Fraction for Coma among Ret- CM Cases.” The section now formulates in more detail the causal assumptions underlying our inference about the attributable fraction using the sufficient-component cause framework (Rothman, 1976)

Reviewer #2:

[...] 1) The paper frames the problem as being binary in the Introduction and Methods: e.g. "If a given trait decreases the probability of developing ret+ but not ret- CM, this would suggest that malaria parasites are pathogenic for ret+ CM but not ret- CM or the trait affects an aspect of disease not causal to the development of ret- CM".e.g. "we test the null hypothesis that the trait frequency is the same in controls and true ret- CM cases vs. the alternative hypothesis that the trait frequency is higher in control than true ret- cases."

So, it seems that using significance to test for an association between CM ret- and controls and their sickle cell status does not capture the question. Estimation would be more useful.

Thanks for raising these good points. We have now formulated a model for estimating the attributable fraction in the section “Inference for Malaria Parasitemia Attributable Fraction for Coma among Ret- CM Cases.” The results are presented in Table 3. We have added a description of these results as follows:

2) The main results are not inconsistent with the hypothesis of the authors, but neither are they strongly supportive of it.

Thanks for making these points. We realized from your comment that we had not been clear that our hypotheses of interest are about whether a trait affects Ret- CM and not about how a trait affects Ret- CM vs. Ret- CM. We have removed the comparisons of Ret- CM vs. Ret+ CM.

Comparability of the groups

Thanks for raising these good points. We have added the following discussion of biases for the controls in the limitations and also conducted an additional analysis in Appendix 1—table 4 which only considers the community controls for which the results are similar to the analysis with all controls:

“Our analysis assumed that HbAS and BGO do not have selection effects on which CM cases are admitted to the Paediatric Research Ward as opposed to dying before reaching the Paedieatric Research Ward or being cured before needing to be referred to the ward; another interpretation of our study is that instead of studying ‘cerebral malaria’ as such, we have studied, ‘cerebral malaria mild enough to make it to the ward but severe enough not to recover without being taken to the ward.’ […] Our results are robust to considering only the community controls (Appendix 1—table 4) which do not suffer from the potential biases of the non-malaria hospital controls.”

Both coma score and retinopathy may differ over the course of the illness, and what is seen is just a snapshot at the time of admission.

Thanks for pointing this out. We have added this as a limitation, while noting that the results for HbAS having a protecting effect on Ret- CM are somewhat insensitive:

“We measure malarial retinopathy at the time of admission, but patients are admitted at different points on the disease trajectory. […] Although there are limitations in our study to the sensitivity with which malarial retinopathy is measured, even if a moderate number of Ret- CM cases should be Ret+ CM cases in Table 2, e.g., if the false omission rate is.5 and the false discovery rate is 0, there is still strong evidence that HbAS has a protective effect for Ret- CM (p-value =.007; odds ratio: 6.81, 95% CI: (1.50, ∞)).”

Can the differences in the Blantyre coma score and other clinical indicators for ret- and ret+ CM in Table 1 be explained?

We have added discussion of the results in Table 1 as follows:

“The Blantyre coma score is statistically significantly higher in Ret+ CM patients than Ret- CM patients, but we do not regard the difference as clinically significant. […] Comparing CM cases to non-malaria controls, as expected, laboratory abnormalities associated with malaria infection (e.g. low hematocrit and platelet count) were more frequent in the CM cases.”

We have added an analysis that uses only the community controls (cord blood samples) in Appendix 1—table 4 and the results are similar to those including all controls.

4) There would be a need to explain the results of the Taylor et al. 2004 study in the light of the findings.

Thanks for the suggestion. We have added to the Discussion an explanation of how our results relate to the Taylor et al. 2004 study:

“In Figure 1 pathway (b), Ret- CM results from an interaction between malaria illness and other illness[…] The protection provided against Ret- CM by HbAS but not BGO could be explained by an interaction between the factor(s) affecting whether malarial retinopathy is present and BGO, e.g., a parasite genotype which causes malarial retinopathy to be absent could interfere with the mechanism by which BGO provides protection against complicated malaria.”

Reviewer #3:

[…] 1) Some of the "non-malaria" hospital controls had malaria parasitemia. Since children can sometimes develop severe malaria with relatively low levels of parasitemia, how can the authors be sure that malaria did not contribute to illness in these control participants?

Thanks for raising this good point. We have added this as a limitation, “Some of the non-malaria hospital controls had malaria parasitemia; since children can sometimes develop severe malaria with relatively low levels of parasitemia, it is possible that malaria contributed to the illness in these control participants.” Furthermore, we have added an analysis that uses only the community controls (cord blood samples) in Appendix 1—table 4 and the results are similar to those including all controls.

This is a good idea to consider. Unfortunately we currently mostly only have the PfHRP2 data for the Ret+ CM patients and not the Ret- CM patients in the cohort years being considered in this paper (1996-2007, which is the period that intersects with when the control data for the Malaria Genomic Epidemiologic Study data was collected). Because there is only one HbAS case among both the Ret+ and Ret- CM cases, we do not think there will be a large enough sample size to study the correlation between HbAS and PfHRP2 with any accuracy.

Regarding the relationship between PfHRP2 and mortality, Dr. Seydel has a paper currently under review that studies this relationship. He finds that there is not a significant correlation between plasma pfHRP2 and mortality but there is a significant correlation between cerebrospinal fluid pfHRP2 and mortality.

Thanks for raising this interesting explanation that we had not thought of. We have added discussion of this explanation to the Discussion:

“Another way in which malaria parasitemia could play a causal role in Ret- CM besides pathway (b) in Figure 1 is that the parasitemia could be the sole cause of coma. […] The protection provided against Ret- CM by HbAS but not BGO could be explained by an interaction between the factor(s) affecting whether malarial retinopathy is present and BGO, e.g., a parasite genotype which causes malarial retinopathy to be absent could interfere with the mechanism by which BGO provides protection against complicated malaria.”

4) Are there any clinical or other indicators that suggest to the authors what the other factor(s) may be in ret- CM that lead children with uncomplicated malaria to go on to develop coma?

This is an important question. Dr. Postels has an R21 grant in which he is studying factors associated with Ret- CM. He is still analyzing the data, but hopes to have results to report soon.

https://doi.org/10.7554/eLife.23699.015

Article and author information

Author details

Dylan S Small

Department of Statistics, The Wharton School, University of Pennsylvania, Philadelphia, United States

Contribution
DSS, Conceptualization, Software, Formal analysis, Investigation, Methodology, Writing—original draft

For correspondence
dsmall@wharton.upenn.edu

Competing interests
The authors declare that no competing interests exist.

"This ORCID iD identifies the author of this article:" 0000-0003-4928-2646
Terrie E Taylor
1. Department of Osteopathic Medical Specialties, College of Osteopathic Medicine, Michigan State University, East Lansing, United States
2. Blantyre Malaria Project, Blantyre, Malawi
Contribution
TET, Conceptualization, Resources, Data curation, Investigation, Project administration, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.
Douglas G Postels

Department of Neurology and Ophthalmology, College of Osteopathic Medicine, Michigan State University, East Lansing, United States

Contribution
DGP, Investigation, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.
Nicholas AV Beare
1. Department of Eye and Vision Science, University of Liverpool, Liverpool, United Kingdom
2. St. Paul’s Eye Unit, Royal Liverpool University Hospital, Liverpool, United Kingdom
Contribution
NAVB, Investigation, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.
Jing Cheng

Department of Preventive and Restorative Dental Sciences, University of California, San Francisco, San Francisco, United States

Contribution
JC, Methodology, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.
Ian JC MacCormick
1. Department of Eye and Vision Science, University of Liverpool, Liverpool, United Kingdom
2. Malawi-Liverpool-Wellcome Trust Clinical Research Programme, Blantyre, Malawi
Contribution
IJCM, Investigation, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.
Karl B Seydel
1. Department of Osteopathic Medical Specialties, College of Osteopathic Medicine, Michigan State University, East Lansing, United States
2. Blantyre Malaria Project, Blantyre, Malawi
Contribution
KBS, Conceptualization, Resources, Data curation, Investigation, Project administration, Writing—review and editing

Competing interests
The authors declare that no competing interests exist.

Funding

The authors declare that there was no funding for this work.

Acknowledgements

We thank the MalariaGEN consortium and Gavin Band and Dominic Kwiatkowski for facilitating access to the MalariaGEN consortium data. MalariaGEN is supported by the Wellcome Trust (WT077383/Z/05/Z) and by the Foundation for the National Institutes of Health (566) as part of the Bill and Melinda Gates’ Grand Challenges in Global Health Initiative. The Resource Centre for Genomic Epidemiology of Malaria is supported by the Wellcome Trust (090770/Z/09/Z). Support was also provided by the Medical Research Council (G0600718; G0600230; MR/M006212/1). The Wellcome Trust also provides core awards to The Wellcome Trust Centre for Human Genetics (075491/Z/04; 090532/Z/09/Z) and the Wellcome Trust Sanger Institute (077012/Z/05/Z and 098051). We thank Sam Wassmer for helpful insights for Figure 1. We thank the reviewers and the senior editor for insightful comments that improved the paper.

Ethics

Human subjects: We analyzed deidentified data from the Blantyre Malaria project and the Malaria Genomic Epidemiology Network which was previously collected prior to our study; informed consent and consent to publish was previously obtained for these data. The use of human subjects in our research project falls under Exemption 4 for Exempt Human Subjects Research.

Reviewing Editor

Ben Cooper, Mahidol Oxford Tropical Medicine Research Unit, Thailand

Publication history

Received: November 27, 2016
Accepted: May 4, 2017
Version of Record published: June 7, 2017 (version 1)

Copyright

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.