Prognostic Significance and Course of Retinopathy in Children WithSevere Malaria FREE

Severe forms of malaria due to Plasmodium falciparum infection still have high mortality and morbidity in sub-Saharan Africa. Plasmodium falciparum malaria causes 1.5 to 2.7 milliondeaths every year,¹ preponderantly in Africanchildren who have cerebral malaria (CM) or severe malarial anemia (SMA). Inchildren, CM is characterized by acute coma, associated with metabolic acidosisand anemia; while in SMA anemia and acidosis predominate. Complications seenin adults, such as renal failure, pulmonary edema, and disseminated intravascularcoagulation are uncommon. Even with optimum hospital care, childhood CM hasa mortality rate of 15%, and about 10% suffer neurological sequelae, suchas epilepsy, ataxia, hemiplegia, and cortical blindness.^{2 - 3}

Severe malaria is associated with a unique cluster of retinal signsthat have been described in children in Malawi, Kenya, and Gambia.^{4 - 6} These signs are bestseen by indirect ophthalmoscopy and have been found in 77% of patients withCM.⁴ The characteristic features include whiteningof the macula with sparing of the foveola; whitening of the peripheral retina;retinal vessel discoloration to pale orange or white, including whiteningof the capillary network (Figure 1);multiple retinal hemorrhages, which are preponderantly white centered; andpapilledema.⁶ The vessel whitening can be inthe form of delineation of the central blood column.

Vessel whitening may be due to the presence of cytoadhered, or sequestered,erythrocytes in which hemoglobin has been metabolized by intracellular parasites.Sequestration of dehemoglobinized, parasitized erythrocytes has been shownon histopathologic examination to occur in retinal vessels of eyes with vesselwhitening.⁷ The number of retinal hemorrhageshas been shown to correlate with cerebral hemorrhages at postmortem examination.⁸ The pathophysiology of retinal whitening remains unclear,although intracellular swelling in response to hypoxia is suspected.⁹

The relationship between retinal lesions and outcome in CM has not beenfully explored. Lewallen et al¹⁰ found papilledemaand the presence of retinal edema (whitening) to be associated with deathin Malawian children. Other studies have focused on retinal hemorrhages andpapilledema observed by direct ophthalmoscopy.^{11 - 14} Retinalhemorrhages, edema, and papilledema have been recorded in adults with CM.^{13 ,15} To our knowledge, the course of retinalsigns in CM has not been formally investigated.

We conducted a prospective study of all children admitted to an Africancentral hospital and treated for CM or SMA during 2 consecutive malaria seasons.We aimed to relate the presence and severity of retinal signs to clinicaloutcome and to determine the course of the retinal signs.

The study cohort was all children with CM or SMA admitted to the MalariaResearch Project ward at the Queen Elizabeth Central Hospital, Blantyre, Malawi.They were studied prospectively during 2 malaria seasons, from January toJune 1999 and 2000. Cerebral malaria was defined as an acute onset coma maintainedfor at least 4 hours after admission, scoring 2 or less on the Blantyre ComaScale,³ with P falciparum parasitemia. Other explanations for altered consciousness, such ashypoglycemia, postictal state, or meningitis were excluded. Severe malarialanemia was defined in patients with parasitemia, a hematocrit reading of 15%or less, and a Blantyre Coma Scale score of 3 or higher. The CM group includedall patients with SMA who were unconscious (Blantyre Coma Scale score <3).

All children were treated according to a protocol with intravenous quinineand fluids by weight and also glucose, anticonvulsants, and antipyretics accordingto need. Intravenous antibiotics were given for concurrent infections, orif a lumbar puncture to exclude meningitis was not done. Screened blood transfusionswere given for life-threatening anemia. All children followed the same treatmentprotocol with no differentiation between those with and without retinal signs.

Clinical observations, including the Blantyre Coma Scale score, weremade every 2 hours until coma resolved. Coma resolution time (the time takento recover to a Blantyre Coma Scale score of 5) was noted. Outcome on dischargewas recorded as full recovery, recovery with neurological sequelae, or death.The Malawi College of Medicine Research and Ethics Committee gave ethicalapproval. Patients were recruited after their parents or guardians had givenpermission in their own language.

Recruited children underwent indirect and direct ophthalmoscopy by anophthalmologist (N.A.B. and C.S. in 1999 and N.A.B in 2000) following pupildilatation with a combination of 1.0% tropicamide hydrochloride and 2.5% phenylephrinehydrochloride. Findings were recorded on standardized charts with acceptableinterobserver concordance.¹⁶ Hemorrhages weregraded using an ordinal frequency scale; macular whitening (MW) and fovealwhitening (FW) by estimated area of involvement, peripheral whitening (PW)by an ordinal severity scale (range, 0, 1+, 2+, and 3+) for each quadrantand averaged to give a composite score, and vessel changes (VC) by numberof quadrants involved.

Children underwent ophthalmoscopy on admission, or after their clinicalcondition had stabilized. During 1999, examinations were repeated daily subjectto the patient's cooperation, or until the results of 2 consecutive examinationswere normal. Ophthalmoscopy was attempted in children attending a follow-upappointment with particular attention paid to those with retinal signs duringadmission.

Data were analyzed using Stata (release 6, Stata Corp, College Station,Tex). Groups with and without retinopathy were compared to establish whetherthey were comparable for demographic features and treatment received. Associationswere investigated between clinical variables on admission and the presenceof retinopathy, using the Mann-Whitney and χ² tests. The relationshipbetween severity of retinal signs and hematocrit reading was examined by Cuzick'snonparametric test for trend.¹⁷ The relationshipbetween severity of retinal signs and death was examined by χ² fortrend and univariate logistic regression.

Multivariate logistic regression analysis was performed, with deathas the dependent variable, and including all predictor variables that wererelated to death with a P<.10 in univariate analysis.This was done to establish whether retinal signs were independent of eachother and other clinical features in predicting death. The other outcome measurewas coma resolution time that was used as the dependent variable in a multivariatelinear regression with retinal signs as predictor variables.

Changes in retinal signs over time were investigated for relationshipswith outcome. Change was defined according to the following preset criteriafor identifying change in retinopathy over time: (1) a change in the severitygrade of hemorrhages; (2) the development or resolution of papilledema; (3)a change in MW or FW by 2 severity grades based on size of area affected;(4) a change in PW by 1 grade in 2 or more quadrants, or by 2 or more gradesin any quadrant; or (5) the appearance or resolution of VC in 2 or more quadrants.The eye with the most consistent view was used, or the right eye was usedif both had a similar view.

The clinical course and outcome of CM and SMA were different; therefore,the results are given separately.

During the study period 304 patients were admitted with CM and 278 patients(91%) were recruited. Three children with prior neurological deficits wereexcluded from the study, and 23 were not recruited because of their deathbefore ophthalmoscopy (9 patients), denial of consent (1 patient), or unavailabilityof observers (13 patients).

The median age of recruited patients was 2.8 years (age range, 4 monthsto 14 years) and 48.2% were males. The median time of ophthalmoscopy was 1hour after admission (range, 0-48 hours); 73% were within 4 hours and 96%within 24 hours.

Forty-one patients (14.8%) died; 17 (6.1%) were discharged with neurologicalsequelae. All patients who were ultimately discharged with neurological sequelaewere recruited. Neurological sequelae included ataxia, dysphasia, corticalblindness, deafness, hemiparesis, gaze palsy, and persistent coma.

The frequency of retinal changes in CM are given in Table 1. When patients with retinopathy were compared with thosewith normal fundi, there was no difference in age, sex, nutritional status,and medications given (Table 2).Patients with retinopathy were more likely to have respiratory distress (P = .02), had a higher average parasite density (P = .02), and were less likely to have repeated or prolongedconvulsions (>5 recorded episodes during admission) than those without retinopathy(P = .01).

In addition, patients with retinal changes were more anemic (medianhematocrit reading, 20%) than those without retinopathy (median hematocritreading, 28%; P<.001) and were more likely toreceive a blood transfusion. The severity of each individual retinal signwas negatively related to hematocrit reading (P<.005),except for papilledema (P = .96). There was no associationbetween anemia and fatal outcome, patients who died were no more anemic thansurvivors (median hematocrit reading, 21% vs 24%; Mann-Whitney test, P = .45). The differences of other clinical parameterswere not statistically significant.

The difference between the frequency of retinopathy in children whodeveloped neurological sequelae (7/17; 41%) and those who fully recovered(128/220; 58%) was not statistically significant (P =.17). Survivors were analyzed as a group.

The presence and severity of retinal findings were analyzed in relationto fatal outcome (Table 3). Retinopathywas present in a significantly higher proportion of those who died comparedwith survivors (P = .001). The relative risk of deathconferred by the presence of any abnormalities of the fundus was 3.7 (95%confidence interval, 1.6-8.5) and was also increased in the presence of papilledema,retinal hemorrhages, or vessel abnormalities.

The severity of retinal hemorrhages, VC, FW, and MW all showed significant,positive association with death by χ² for trend analysis (Table 3). The presence of moderate to severegrades of hemorrhages, VC, FW, and MW conferred a significantly increasedrelative risk of death.

Univariate logistic regression on the clinical and laboratory variablesin Table 2 identified factorsrelated to death with P≤.10. These were retainedfor inclusion in a multivariate logistic regression of fatal outcome. Theresults of univariate logistic regression on the signs of the fundus are presentedin Table 4 and these were alsoincluded in the model. The clinical and laboratory variables consistentlyrelated to death in studies of CM in children are deep coma and hypoglycemia.^{3 ,10 ,18 - 20} Respiratorydistress or acidosis was also predictive of death in several studies of severemalaria.^{21 - 23} Unfortunately,blood gas evaluation was unavailable because of equipment failure. However,respiratory distress, along with Blantyre Coma Scale score and blood glucoselevel on admission were included in the model. The full model is listed in Table 5. Papilledema and hemorrhages wereindependent predictors of death (P≤.01) with greaterodds ratios than any of the other factors. The other retinal signs were correlatedwith hemorrhages (r>0.5, P<.001).

In survivors coma resolution time was significantly longer in patientswith retinopathy (median, 42 hours) than those with none (median, 29 hours; P<.001). Each sign was significantly related to thecoma resolution time by univariate linear regression (Table 6) (P<.005). A minimal-effectsregression analysis of fundus signs showed that hemorrhages, PW, and papilledemawere independent in predicting prolonged coma (Table 7)(P<.05).

All patients had blood cultures taken on admission, and in 9 patients(3%) with CM pathogens were cultured, mostly Salmonella species. Three patients with septicemia died; 1 had normal fundi and2 had papilledema only.

Forty-seven patients with SMA were recruited, 82% of admissions forSMA. Their median age was 2 years (age range, 5 months to 12 years). Theirmedian hematocrit reading was 10%, and median hemoglobin concentration was3.1 g/dL. There were 2 deaths (4%), in 1 of which the patient had had concurrentpneumonia.

Retinopathy was less common in SMA (53%) than in CM (61%) and tendedto be less severe with fewer of the component changes (Table 1). In SMA all MW was in the mildest category, and only 1patient had more than 5 hemorrhages. However PW and VC were almost as commonas in CM. There were 5 patients admitted with critically severe anemia (hematocritreading, 7%-8%) without malaria detectable by parasitemia; none had any retinalchanges.

The fundi of 116 children with CM were examined daily during a meanof 2.7 days. A single observer (N.A.B. or C.S.) performed all examinationson 103 (89%) of 116 children. Results of the initial examination showed noabnormality in 44 children (38%), all of whom survived and only 2 (5%) ofwhom developed 1 to 5 hemorrhages.

Seventy-two children (62%) had retinal changes initially. The differencebetween the mortality in those whose retinopathy worsened (7/39 [18%]) andthose whose retinopathy did not (2/33 [6%]) was not statistically significant(P = .17) (Table 8).

Patients with a large increase in the number of retinal hemorrhages(defined as progression by ≥2 grades, or to the most severe grade) wereat an increased risk of death (P = .02) (Table 8). The relative risk of death inthis group was 5.1 (95% confidence interval, 1.5-17.1). The intervals betweenthese observations and subsequent death were three-quarter hour and, 2, 5,and 7 hours. There was no association between death and either the progressionof other retinal signs or an increase in hemorrhages by 1 grade only.

During the 1999 malaria season, 102 patients (60%) who had had CM attendedfor follow-up within 4 weeks of discharge. Twenty-seven children (26%) weresufficiently cooperative for fundus examination. Retinal signs had been presentin 22 of these patients during admission; all were observed to be improvingwithout any secondary retinal sequelae.

The progress of retinal signs after discharge is given in Table 9. Vessel changes persisted for up to 3 weeks. Macular whiteningand FW resolved over 2 weeks. Peripheral whitening was not seen in any patientfor longer than 7 days after discharge. This is longer than the acute clinicalepisode that generally resolves with therapy during 24 to 48 hours in thosewho recover fully.

This study has shown that retinal changes in CM are significantly associatedwith poor outcome. The time to regain consciousness is longer in patientswith retinopathy, and the risk of death is increased 3.5-fold (95% confidenceinterval, 1.5-8.0). Coma recovery time and risk of death increase with increasingseverity of retinal signs.

The pathophysiology of CM is poorly understood, but these results, byrelating severity of retinal signs to length of coma and to fatal outcome,support the hypothesis that retinal signs relate to cerebral pathophysiology.Papilledema and retinal hemorrhages were independent predictors of death.This suggests that papilledema and retinal changes are indicative of differentpathophysiological processes that can occur together in CM but are independentlyrelated to poor outcome.

Our results support the findings of Lewallen et al¹⁰ whostudied 141 Malawian children with CM and found a relative risk of death withpapilledema of 6.7, and with vascular abnormalities of 3.2, compared withour findings of 4.5 and 2.4, respectively. They found no association betweendeath and the presence of retinal hemorrhages but had not evaluated the severityof retinal signs. Data from Lewallen et al were from patients enrolled ina therapeutic drug trial and more specific entry criteria may account forsome variation in results. Other studies of retinal changes in CM are notdirectly comparable as they did not use indirect ophthalmoscopy^{12 ,14} orwere in adults.^{11 ,13 ,15}

In patients with SMA, retinopathy was milder and disease outcomes werebetter. The number of deaths in this group were too small to analyze in relationto retinopathy.

We have found an important relationship between retinopathy and anemiain CM. There was a highly significant association between a low hematocritreading and the severity of each retinal sign (P<.005),except papilledema. The relationship with anemia was independent of death,and anemia alone without parasitemia is insufficient to cause these retinalchanges. None of 13 children with meningitis and severe anemia (hematocritreading, <20%) examined during the study period had these specific retinalsigns.

An association between retinal hemorrhages and anemia in complicatedmalaria has been previously noted,⁴ but ourfindings show anemia to be related to the other retinal signs. The pathogenesisof acute anemia is complex, but its severity is related to the maximum parasitaemia.²⁴ Severe anemia is likely to be associated with highconcentrations of sequestered, parasitized erythrocytes within the microvasculature,thought to cause retinal VC.⁸ Similarly, highdensities of rapidly metabolizing parasites in the retinal vasculature mayproduce relative hypoxia leading to intracellular edema. This has been proposedas a mechanism causing retinal whitening.⁴

To our knowledge, retinal changes in CM have not previously been followedup systematically over time. We found that a large increase in the numberof retinal hemorrhages after admission was associated with death. This findingis based on only 4 deaths in each group and so this needs to be treated withcaution. However, its clinical significance can be illustrated by the factthat the interval between the observation of a substantial increase in hemorrhagesand subsequent death was short (three quarters of an hour and 2, 5, and 7hours). This premorbid development of many retinal hemorrhages is a new observationand may indicate the development of coagulopathy with fibrin deposition aspart of the terminal process in these children. Fibrin deposition in cerebralcapillaries has been noted in autopsy studies,^{25 - 27} includingstudies in this research programme (Richard Carr, MRCPath, written communication,March 12, 2001).

The remaining analysis of our longitudinal data failed to show statisticallysignificant associations between deterioration in retinal signs, either individuallyor collectively, and fatal outcome of the disease. The changes in retinalsigns were limited after admission, and our analysis was restricted by thefew deaths in this group. Deaths tend to occur soon after admission to thehospital, precluding serial examinations of the fundus.

Retinal changes in CM took longer to resolve than the clinical episode.Vascular changes were persistent for up to 3 weeks. This has an importantbearing on theories regarding the mechanism of coma and death in CM.²⁸ Retinal vascular changes are associated with parasitizederythrocytes sequestered in retinal vessels.⁸ Retinalvascular changes and sequestered erythrocytes (which may contain dead parasites)are still present long after the child has regained consciousness. Assumingretinal vascular changes occur in parallel with cerebral sequestration, thisfinding of prolonged retinal vessel whitening suggests that the physical presenceof parasitized erythrocytes alone is insufficient to maintain coma.

There were no secondary retinal changes or ischemic sequelae in the4 weeks after discharge. This supports the findings of Hero et al⁴ that showed no evidence of microvascular occlusionor leakage on fluorescein angiography.

Parasitemia is common in the population served by the Queen ElizabethCentral Hospital, and so incidental parasitemia can occur in any comatosepatient. Three children having a clinical diagnosis of CM died with a septicemia,while none had retinal changes other than papilledema. In the context of thisstudy, this weakens the power of retinal signs to predict death. In the widercontext, patients with apparently fatal CM and no malarial retinopathy may,in fact, have other factors contributing to coma and death. We emphasize theneed to look for these factors in patients without retinopathy who do nothave rapid recovery from coma. Conversely, retinal changes, by nature of theirspecificity, are valuable in confirming a diagnosis of CM in a comatose childwith parasitemia. Cerebral malaria is a clinical diagnosis complicated bythe presence of incidental parasitemia in this population, and ophthalmoscopyprovides valuable diagnostic information.

This study demonstrates the value of ocular examination by indirectand direct ophthalmoscopy in patients with CM. We have shown that the presenceand severity of retinal signs are significant predictors of prolonged comaand death. The previously reported link between retinopathy and outcome inCM has been strengthened and clarified by these findings. The presence ofretinopathy, albeit in a milder form, is reported in SMA. It will be importantto assess the extent to which nonspecialized practitioners in malarial areascan identify some or all of the changes described.

Our findings are consistent with the hypothesis that retinal changesin CM relate to cerebral pathophysiological processes. Retinal features arean integral part of the clinical picture, and ophthalmic observations cancontribute to continuing studies of pathophysiological processes and therapeuticinterventions.

Correspondence: Nicholas A. Beare, FRCOphth, St Paul's Eye Unit, RoyalLiverpool University Hospital, Prescot Street, Liverpool L7 8XP, England(nbeare@btinternet.com).

Submitted for publication July 19,2002; final revision received January15, 2004; accepted January 15, 2004.

This study was supported by grants from the Beit Trust, Surrey, England;The British Council for Prevention of Blindness, London, England; and TheFoundation for the Prevention of Blindness, West Sussex, England. The enrollmentand care of the study patients was part of a program funded by The WellcomeTrust, London.

We thank Robin Broadhead, FRCP, and staff of the Department of Paediatrics,College of Medicine, University of Malawi, Blantyre; and the patients, parents,and guardians for their participation in this study. We would like to acknowledgethe work of the Malaria Project staff, especially Rose Mazengera, SRN, andFlora Nyalo, EN. Mandy Walsh, BSc, provided data and advice on blood cultureresults.

Correspondence: Nicholas A. Beare, FRCOphth, St Paul's Eye Unit, RoyalLiverpool University Hospital, Prescot Street, Liverpool L7 8XP, England (nbeare@btinternet.com).