Abstract
Background
In febrile comatose patients living in malaria-endemic areas, overlapping symptoms and limited laboratory capacity make it difficult to distinguish parasitic, bacterial, and viral central nervous system infections. We evaluated electroencephalography (EEG) as a biomarker to differentiate the microbiologic etiology of pediatric febrile coma at a major referral center in Malawi.
Methods
This was a retrospective case-control study comparing EEG recordings of Malawian children with cerebral malaria to those with febrile coma of nonmalarial cause (bacterial meningitis, viral encephalitis, or unknown cause). Participants were admitted to Queen Elizabeth Central Hospital (Blantyre, Malawi) between 2013 and 2021. Inclusion criteria were fever, coma (Blantyre Coma Score ≤2), and coma etiology (malarial or nonmalarial) defined by laboratory testing. Four supervised machine learning algorithms were used to train a balanced ensemble classifier, SuperLearner, generating test characteristics of the diagnostic ability of EEG features.
Results
Two hundred three children with cerebral malaria and 87 children with nonmalarial coma were included. Univariate analysis of qualitative (visual) EEG interpretations revealed higher voltage, slower background frequency, more sleep elements, less variability, more abnormal organization, and less continuity in cerebral malaria. Quantitative waveform analysis showed greater power in cerebral malaria. Both quantitative and qualitative EEG interpretation distinguished coma etiology (area under the receiver operating characteristic curve [AUROC] = 0.85 and 0.86, respectively). Combining qualitative and quantitative interpretation methods, the test characteristic improved (AUROC = 0.90).
Conclusions
EEG features distinguish malarial from nonmalarial coma in febrile Malawian children. This technology may aid in distinguishing the microbiologic etiology of febrile coma in malaria-endemic areas.
Keywords: central nervous systems infection, coma, cerebral malaria, electroencephalography
Electroencephalography (EEG) is an efficient diagnostic biomarker in distinguishing cases of febrile coma of malarial origin from nonmalarial coma. EEG might serve to support a diagnosis of malarial coma (cerebral malaria) in febrile comatose children.
Neuroinvasive infectious diseases, often presenting with febrile coma, remain a continuing cause of death and disability worldwide. Electroencephalography (EEG) monitoring is primarily used in the clinical care of children with febrile coma to detect nonconvulsive seizures. To date, the role of EEG in elucidating coma etiology and in furthering understanding of pathophysiologic mechanisms of neuroinvasive pathogens in children with febrile coma is less well defined.
Typically, EEG findings are considered less specific in pathogen identification compared to laboratory technologies such as polymerase chain reaction (PCR). Lateralized periodic discharges seen in acute herpes encephalitis, for instance, are common in other forms of acute brain injury and overshadowed by the 98% sensitivity of cerebrospinal fluid (CSF) PCR for establishing disease etiology [1–3].
Approximately half the world's population is at risk for malaria infection [4]. In high-transmission areas, malaria is the most common cause of febrile coma. Cerebral malaria (CM), defined as an otherwise unexplained coma in someone with Plasmodium parasitemia, is the most severe form of malarial illness with a mortality rate of 17%–25%, and high rates of permanent neurologic or cognitive disabilities [5–8]. Annually, over half a million children die of malaria, many with CM [4].
In the district hospitals and regional health centers where children present with fever and coma, treatment is guided by a clinical examination and the malaria rapid diagnostic test (mRDT) alone. The mRDT is prone to false positives when malaria parasitemia is caused by past exposure and consequent asymptomatic parasitemia because in these situations, circulating parasites are present (and detected by mRDT) but are etiologically unrelated to the febrile illness at hand [9]. False negatives on mRDT may arise when high parasite burden causes target antigen saturation and blocks immune complex formation on the assay [10, 11], when genes responsible for target antigen generation are deleted in parasitic populations under selection pressure [12], and when low parasite densities—potentially clinically significant to nonimmune individuals—fall below the mRDT detection threshold [11, 13].
While more sensitive than mRDT, malaria blood smear is also susceptible to error when used to establish the etiology of febrile coma. False-negative blood smears may result if blood smears are not collected before presumptive treatment because artemisinin-based antimalarials rapidly clear parasites from the blood. Like mRDTs, malaria smears also fail to distinguish between children with CM for whom malaria is responsible for the illness, and those with asymptomatic parasitemia plus a nonmalarial coma etiology [9]. Malaria smear preparation and interpretation require training (eg, smear preparation, storage, and staining; laboratory technician training in parasite detection) that may not be readily available at peripheral health centers where febrile comatose children initially present for medical care, whether in endemic or low-transmission areas.
If EEG could sensitively discriminate the microbiologic etiology of febrile coma when CM is in the differential diagnosis, it could be helpful for clinical care in both high- and low-transmission settings, supporting other diagnostic etiologic testing or where disease etiology is unclear. We aimed to evaluate whether EEG is such a diagnostic biomarker.
METHODS
We systematically analyzed and compared EEG records from children 3 months to 14 years old with CM and or nonmalarial febrile coma. Children were admitted to the Pediatric Research Ward of Queen Elizabeth Central Hospital in Blantyre, Malawi, between May 2013 and June 2021. EEGs were collected as part of 2 observational studies: one of CM pathogenesis, and one investigating the etiology and outcomes of children with both CM and nonmalarial febrile coma.
Clinical and Laboratory Data
CM was defined as coma (Blantyre Coma Score [BCS] ≤2), malaria parasitemia on peripheral blood smear, and no other known coma etiology (eg, head trauma). Children were treated with 3 doses of intravenous artesunate followed by 3 days of enteral lumefantrine-artemether, per Malawi Ministry of Health guidelines. Broad-spectrum intravenous antibiotics (typically ceftriaxone) were added at the discretion of the treating clinician. After clinical stabilization, children underwent lumbar puncture. All patients had blood cultures. Clinical complications, including seizures, hypoglycemia, and organ system failure, were treated as they arose.
Nonmalarial coma was defined as BCS ≤2, fever (or history of fever), and negative malaria peripheral smear with biochemical profile, neuroimaging, and/or positive CSF profile indicating a nonmalarial coma etiology (eg, bacterial meningitis). CSF was analyzed using microscopy, Gram staining, and culture as well as molecular testing for detection of 14 typical neuroinvasive organisms via PCR. Brain magnetic resonance imaging (MRI) was performed within 6 hours of admission using a 0.35-T Signa Ovation Excite MRI scanner (General Electric) (Supplementary Appendix Methods and Figure 1).
Figure 1.
Flowchart indicating test timing in relation to clinical care in children with nonmalarial coma. Abbreviations: CM, cerebral malaria; CRAG, cryptococcal antigen detection; CSF, cerebrospinal fluid; EEG, electroencephalogram; HIV, human immunodeficiency virus; LAM, urinary lipoarabinomannan; LFA, lateral flow assay; LP, lumbar puncture; MRI, magnetic resonance imaging (*verbal report available in real time in a subgroup of the cases, but not in the majority, due to staffing and timing capacity); PCR, polymerase chain reaction; POC, point of care; q12h, every 12 hours. This figure is adapted with permission from Ray et al. [14].
Most children with nonmalarial coma had negative malaria microscopy, negative mRDTs, and a confirmed pathogenic nonmalarial etiology. However, a subgroup of children with no confirmed nonmalarial pathogenic cause had a negative blood smear at admission, but either positive mRDT or report of recent positive blood smear at the referring district hospital. For this subgroup, a histidine-rich protein 2 (HRP-2) quantitative assay was performed. If the HRP-2 burden was >200 ng/L (an established threshold for true malaria infection), these children were deemed to have postparasitemia CM [14]. Participants with HRP-2 levels <200 ng/L were included in the cohort and classified as having coma of nonmalarial, unknown cause. Children whose coma was nonmalarial (BCS ≤2 with no peripheral parasitemia and negative mRDT were treated with intravenous ceftriaxone and escalated on clinician discretion to intravenous carbapenems, intravenous adjunctive aminoglycosides, or both if clinically indicated and available.
EEG Acquisition and Processing
Thirty-minute EEG recordings were obtained within 6 hours of Pediatric Research Ward admission and interpreted clinically. After standard 10–20 electrode placement by trained technologists, EEGs with a sampling frequency of 200 Hz were obtained on a CeeGraph (Biologic/Natus Medical Corporation, Pleasanton, California) through 2016 or an XLTEC 32 channel machine (Natus Medical Corporation) thereafter. Post-hospitalization, EEG recordings were reinterpreted independently by 2 board-certified pediatric neurophysiologists, and discrepant ratings were adjudicated by real-time consensus including a third neurophysiologist.
Qualitative EEG features included background organization (anterior-posterior gradient), voltage, continuity, sleep architecture, focal slowing, hemispheric symmetry of voltage and frequency, spontaneous variability, reactivity, rhythmic and periodic patterns, and seizures. We estimated dominant frequency and maximum voltage based on visual scanning of the entire record. We estimated average voltage based on the amplitude of the predominant background activity. Epileptiform discharges and electrographic seizures were localized and quantified. We used American Clinical Neurophysiology Society Standardized Critical Care EEG terminology in all classifications [15].
Quantitative processing was performed as an average of the first 5 seizure-free minutes of EEG recording using Persyst 14 (Persyst Corporation, Prescott, Arizona). Fast Fourier Transformation of EEG waveforms yielded the following values: absolute power (µV2) of frequency bands of interest, including delta (0.5–3.9 Hz), theta (4.0–7.9 Hz), and alpha (8.0–12.9 Hz); relative spectral power (power of a select frequency band of interest divided by total power of all frequency bands); and power ratios (a quotient of the power of frequency band(s) of interest divided by delta power). Measured power ratios included low-delta to delta ratio (0.5–1.9 Hz/0.5–3.9 Hz), theta-plus-alpha to delta ratio (4.0–12.9/0.5–3.9 Hz), and alpha to delta ratio (8.0–12.9 Hz/0.5–3.9 Hz). Symmetry between the cerebral hemispheres was assessed using the absolute asymmetry index function and median waveform amplitude using the peak envelope function. We calculated theta-alpha variability, previously shown to be associated with outcomes in children with CM [16]. Beta activity was excluded from quantitative analysis due to a desire to avoid capturing potential benzodiazepine effect on the EEG background. Qualitative EEG metrics did, however, include beta activity to conform with prior work on EEG in CM [17].
Statistical Analysis
We compared baseline demographic, clinical, and laboratory characteristics between children with CM and those with nonmalarial coma using t tests, Wilcoxon rank-sum tests, or χ2 tests, as appropriate. Qualitative and quantitative EEG factors between children with malarial and nonmalarial coma were similarly compared. We fit logistic regression models to assess the independent association between qualitative or quantitative EEG features and whether a child had CM or nonmalarial coma.
We built classifiers based on these 2 types of EEG data (qualitative or quantitative) based on their ability to predict a participant's CM versus nonmalarial coma status. Classifiers were built using an ensemble machine learning method consisting of logistic regression, generalized additive model, and random forest classifiers. The performance of classification was quantified and compared based on average out-of-sample area under the receiver operating characteristic curve (AUROC). Acknowledging age-related changes in frequency and amplitude on EEG recording, subanalysis was performed stratifying participants by age 6–23 months, 24–60 months, and ≥61 months.
Analyses were performed with the statistical computing software R version 4.0. Ensemble learning methods were performed using the package SuperLearner. Statistical testing was 2-sided and P < .05 was considered a prespecified, statistically significant difference between groups.
Standard Protocol Approvals, Registrations, and Patient Consents
Ethical review of the parent natural history studies was granted by Michigan State University (malaria patients), University of Liverpool (nonmalarial patients), and the University of Malawi College of Medicine (now Kamuzu University of Health Sciences) (both studies). At the time of enrollment into parent studies, families agreed to the secondary use of deidentified data, as were used in these analyses. This retrospective review of EEG data was approved by the Institutional Review Board of Children's National Hospital (Washington, District of Columbia).
Data Availability Statement
Quantitative and quantitative EEG features not provided here due to space limitation can be made available (anonymized) at the request of a qualified investigator.
Supplementary Material
Contributor Information
Alexander Andrews, Division of Neurology, The George Washington University/Children's National Hospital, Washington D.C., USA.
Dana B Harrar, Division of Neurology, The George Washington University/Children's National Hospital, Washington D.C., USA.
Tesfaye Zelleke, Division of Neurology, The George Washington University/Children's National Hospital, Washington D.C., USA.
Jiahui Zhang, Division of Statistics, University of Washington, Seattle, Washington, USA.
Bo Zhang, Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Center, Seattle, Washington, USA.
John Barber, Division of Biostatistics and Study Methodology, Children's National Research Institute, Washington D.C., USA.
Rohini Dasan, The George Washington University, School of Medicine and Health Sciences, Washington D.C., USA.
Nathaniel Amoah, The George Washington University, School of Medicine and Health Sciences, Washington D.C., USA.
Olivia Findorff, University of Virginia, Charlottesville, Virginia, USA.
Sean Krieger, College of Osteophatic Medicine, Michigan State University, East Lansing, Michigan, USA.
Emmalee Barrett, The George Washington University, School of Medicine and Health Sciences, Washington D.C., USA.
Nicole O’Brien, Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi; Division of Critical Care, The Ohio State University College of Medicine/Nationwide Children's Hospital, Columbus, Ohio, USA.
Charlotte E Fuller, Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi; Institute of Infection, Veterinary and Ecological Sciences, University of Liverpool, Liverpool, United Kingdom.
Christopher A Moxon, Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi; Wellcome Centre for Molecular Parasitology, Institute of Infection, Immunity and Inflammation, University of Glasgow, Glasgow, United Kingdom.
Terrie E Taylor, College of Osteophatic Medicine, Michigan State University, East Lansing, Michigan, USA; Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi.
Michael J Griffiths, Centre for Child and Adolescent Health Research, Western Sydney (Baludarri) Precinct, Faculty of Medicine and Health, University of Sydney, Sydney, New South Wales, Australia; The Brain Infection and Inflammation Group Institute of Infection Veterinary, and Ecological Sciences, University of Liverpool, Liverpool, United Kingdom.
Stephen T J Ray, Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi; The Brain Infection and Inflammation Group Institute of Infection Veterinary, and Ecological Sciences, University of Liverpool, Liverpool, United Kingdom; Oxford Vaccine Group, Department of Paediatrics, University of Oxford, Oxford, United Kingdom; Malawi-Liverpool-Wellcome Trust Clinical Research Programme, Blantyre, Malawi; Department of Paediatrics and Child Health, Kamzu University of Health Sciences, Blantyre, Malawi; Department of Paediatric Infectious Disease and Immunology, Oxford University Hospitals NHS Foundation Trust, Oxford, United Kingdom; Wilson Laboratory, Weil Institute for Neurosciences, University of San Francisco, San Francisco, California, USA.
Douglas G Postels, Division of Neurology, The George Washington University/Children's National Hospital, Washington D.C., USA; Blantyre Malaria Project, Kamuzu University of Health Sciences/University of Malawi College of Medicine, Blantyre, Malawi.
RESULTS
EEGs were available from 203 children with CM and 87 with nonmalarial coma (Table 1). On univariate analysis, children with CM were more likely to be younger, to have had a longer coma duration prior to EEG connection, to have lower platelet counts, and to have less severe coma (BCS 2 as compared to 0–1; Table 1).
Table 1.
Comparison of Baseline Demographic, Clinical, and Laboratory Characteristics Between Children With Cerebral Malaria and Nonmalarial Coma
| Characteristic | Cerebral Malaria (n = 203) | Nonmalarial Coma (n = 87) | P Value for Difference |
|---|---|---|---|
| Age, m, mean (SD) | 53.20 (31.36) | 67.24 (47.10) | .003 |
| Sex, male, No. (%) | 100 (49.3) | 48 (55.2) | .427 |
| Duration of coma before admission, h, median (IQR) | 50.97 (66.54) | 17.47 (15.03) | <.001 |
| Admission glucose, mmol/L, mean (SD) | 6.20 (2.48) | 6.17 (3.21) | .928 |
| Admission lactate, mmol/L, mean (SD) | 4.79 (3.26) | 4.71 (3.86) | .863 |
| Platelets, ×103, mean (SD) | 86.90 (92.02) | 298.33 (181.67) | <.001 |
| Blantyre Coma Score, No. (%) | .004 | ||
| 0 | 13 (6.4) | 16 (18.4) | |
| 1 | 90 (44.3) | 39 (44.8) | |
| 2 | 100 (49.3) | 32 (36.8) |
Abbreviations: IQR, interquartile range; SD, standard deviation.
Univariate analysis of both qualitative and quantitative EEG features in children with CM and nonmalarial coma showed several differences in EEG factors between groups. Using qualitative interpretation methods, children with CM were more likely to have higher average and maximum voltages, less fast (beta) frequency activity, presence of sleep spindles, absence of background variability, abnormal (reversed) anterior-posterior gradient, and a less continuous record (Figure 2, Table 2). Children with nonmalarial coma were more likely to have periodic discharges. Quantitative EEG outputs showed statistically significant increases in all absolute power values in children with CM. Significantly greater quantitative peak envelope indicated higher median voltage in CM cases. The low-delta/delta ratio was lower in CM cases.
Figure 2.
Electroencephalography tracings from children with cerebral malaria (CM) (A and C) and nonmalarial coma (B and D). A, High-amplitude recording of a child with CM. B, Periodic discharges in the recording of a child with nonmalarial coma. C, Suppressed recording with cardio-ballistic artifact in a child with CM. D, A background voltage of modest amplitude with greater mixture of fast frequency activity in nonmalarial coma (low frequency filter 0.16, high frequency filter 30 Hz, sensitivity 5 μV/mm).
Table 2.
Univariate Comparison of Qualitative and Quantitative Electroencephalography Factors From Children With Nonmalarial Coma and Those With Cerebral Malaria
| EEG Factor | Cerebral Malaria (n = 203) | Nonmalarial Coma (n = 87) | P Value for Difference |
|---|---|---|---|
| Qualitative | |||
| Average voltage, µV, mean (SD) | 96.04 (36.69) | 68.02 (37.18) | <.001 |
| Maximum voltage, µV, mean (SD) | 199.24 (69.32) | 159.22 (81.37) | <.001 |
| Dominant frequency in first 10 sec, No. (%) | |||
| Delta (< 4 Hz) | 191 (94.1) | 78 (89.7) | .277 |
| Theta (4 to <8 Hz) | 59 (29.1) | 21 (24.1) | .474 |
| Alpha (8 to <13 Hz) | 13 (6.4) | 2 (2.3) | .247 |
| Beta (≥13 Hz) | 26 (12.8) | 20 (23.0) | .046 |
| Dominant frequency, Hz, mean (SD) | 3.15 (1.50) | 2.84 (1.21) | .093 |
| Spindles, No. (%) | <.001 | ||
| Not present (and patient 2 m or older) | 103 (50.7) | 61 (70.1) | |
| Asymmetric | 2 (1.0) | 4 (4.6) | |
| Asynchronous (and patient <2 y old) | 3 (1.5) | 1 (1.1) | |
| Asynchronous (and patient 2 y or older) | 3 (1.5) | 6 (6.9) | |
| Fragments | 39 (19.2) | 2 (2.3) | |
| Present | 53 (26.1) | 13 (14.9) | |
| Absent variability, No. (%) | 198 (97.5) | 52 (59.8) | <.001 |
| Reactivity to painful stimulus, No. (%) | .713 | ||
| Absent | 91 (44.8) | 43 (49.4) | |
| Present | 97 (47.8) | 37 (42.5) | |
| Not assessed | 15 (7.4) | 7 (8.0) | |
| Anterior-posterior gradient, No. (%) | .036 | ||
| Normal | 33 (16.3) | 21 (24.1) | |
| Reversed | 24 (11.8) | 3 (3.4) | |
| None | 146 (71.9) | 63 (72.4) | |
| Periodic discharges present, No. (%) | 2 (1.0) | 5 (5.7) | .045 |
| Asymmetric background, No. (%) | 26 (12.8) | 15 (17.2) | .418 |
| Number of electrographic seizures, No. (%) | .090 | ||
| 0 | 198 (97.5) | 82 (94.3) | |
| 1 | 1 (0.5) | 4 (4.6) | |
| 3 | 3 (1.5) | 1 (1.1) | |
| 5 | 1 (0.5) | 0 (0.0) | |
| Background voltage, No. (%) | .152 | ||
| Normal | 197 (97.0) | 80 (92.0) | |
| Low (most or all activity 10–20 µV) | 3 (1.5) | 4 (4.6) | |
| Suppressed (all activity ≤10 µV) | 3 (1.5) | 3 (3.4) | |
| Continuity of the background, No. (%) | <.001 | ||
| Continuous | 116 (57.1) | 68 (78.2) | |
| Near continuous (≤10% attenuated or suppressed) | 69 (34.0) | 8 (9.2) | |
| Discontinuous (10%–49% attenuated or suppressed) | 10 (4.9) | 7 (8.0) | |
| Burst attenuation/suppression (≥50% attenuated or suppressed) | 8 (3.9) | 4 (4.6) | |
| Quantitative, mean (SD) | |||
| Total power (µV2) | 25.74 (9.88) | 18.14 (9.33) | <.001 |
| Delta power (µV2) | 24.84 (9.94) | 17.60 (9.27) | <.001 |
| Theta power (µV2) | 4.85 (2.18) | 3.20 (1.86) | <.001 |
| Alpha power (µV2) | 2.02 (0.98) | 1.45 (0.98) | <.001 |
| Peak envelope | 34.93 (14.91) | 22.92 (0.09) | <.001 |
| Low-delta/delta ratio (0.5–1.9 Hz/0.5–3.9 Hz) | 0.58 (0.08) | 0.62 (0.09) | <.001 |
| Alpha/delta ratio (8.0–11.9 Hz/0.5–3.9 Hz) | 0.12 (0.07) | 0.12 (0.08) | .830 |
| Theta/delta ratio (4.0–7.9 Hz/0.5–3.9 Hz) | 0.28 (0.10) | 0.24 (0.19) | .035 |
| Theta + alpha/delta ratio (4.0–12.9 Hz/0.5–3.9 Hz) | 0.35 (0.14) | 0.32 (0.20) | .169 |
| Absolute asymmetry index (EASI) | 24.80 (9.42) | 26.05 (9.62) | .306 |
| Theta-alpha variability | 0.34 (0.17) | 0.37 (0.17) | .205 |
Abbreviations: EASI, EEG asymmetry index; EEG, electroencephalography; SD, standard deviation.
Multivariate modeling revealed independent associations between several qualitative EEG features and a CM versus nonmalarial coma (Table 3). CM versus non-CM status positively correlated with higher average voltage (odds ratio [OR], 1.384, 95% confidence interval (CI) [1.138–1.692] per 10 volts increase; P < .001), negatively correlated with beta frequency prevalence (OR, 0.353 [ 95% CI.134–.930]; P = .035), and negatively correlated with variability (OR, 35.730 [95% CI 8.103–157.546] when variability was absent; P < .001) (Table 3). Results of ensemble machine learning methods showed that qualitative EEG features together had an average out-of-sample AUROC of 0.86 (ranging from 0.75 to 0.94; Figure 3A) and the result from a simple logistic regression classifier is 0.86 (ranging from 0.76 to 0.93), showing excellent predictive power of qualitative EEG features in discriminating cases of CM.
Table 3.
Multivariate Comparison of Qualitative and Quantitative Electroencephalography Factors and Cerebral Malaria Versus Non- Malarial Coma
| EEG Factor | Estimate (β) | Standard Error | P Value |
|---|---|---|---|
| Qualitative | |||
| Average voltage (µV) | .033 | 0.010 | .001 |
| Maximum voltage (µV) | −.004 | 0.004 | .325 |
| Dominant frequency among all channels in first 10 sec | |||
| Delta (<4 Hz) | .976 | 0.709 | .169 |
| Theta (4 to <8 Hz) | −.221 | 0.530 | .677 |
| Alpha (8 to <13 Hz) | 1.695 | 1.560 | .277 |
| Beta (≥13 Hz) | −1.041 | 0.494 | .035 |
| Dominant frequency (Hz) | .454 | 0.244 | .063 |
| Spindles | |||
| Asymmetric | −.449 | 1.077 | .677 |
| Asynchronous (and patient <2 y old) | −.822 | 0.911 | .367 |
| Asynchronous (and patient 2 y or older) | −.535 | 1.453 | .713 |
| Fragments | 1.747 | 0.911 | .055 |
| Present | .913 | 0.491 | .063 |
| Absent variability | 3.576 | 0.757 | <.001 |
| Reactivity to painful stimulus | |||
| Present | −.136 | 0.415 | .743 |
| Not assessed | .664 | 0.748 | .374 |
| Presence of periodic discharges | −.381 | 1.027 | .711 |
| Asymmetric background | −.141 | 0.526 | .789 |
| Electrographic seizures | |||
| 1 | −2.045 | 1.971 | .299 |
| 3 | 1.941 | 2.375 | .414 |
| 5 | 15.290 | 2400.00 | .995 |
| Background voltage | |||
| Low (most or all activity 10–20 µV) | −11.680 | 963.700 | .990 |
| Suppressed (all activity less than 10 µV) | −10.750 | 963.700 | .991 |
| Continuity of the background | |||
| Near continuous (≤10% attenuated or suppressed) | .966 | 0.510 | .058 |
| Discontinuous (10%–49% attenuated or suppressed) | −.032 | 0.684 | .963 |
| Burst attenuation/suppression (≥50% attenuated or suppressed) | 15.890 | 963.700 | .987 |
| Quantitative | |||
| Total power (µV2) | .188 | 0.250 | .451 |
| Delta power (µV2) | −.038 | 0.245 | .876 |
| Theta power (µV2) | −.605 | 0.300 | .044 |
| Alpha power (µV2) | .675 | 0.392 | .085 |
| Peak envelope | .013 | 0.040 | .751 |
| Low-delta/delta ratio (0.5–1.9 Hz/0.5–3.9 Hz) | −9.760 | 2.919 | .001 |
| Alpha/delta ratio (8.0–11.9 Hz/0.5–3.9 Hz) | −5.563 | 3.750 | .138 |
| Theta/delta ratio (4.0–7.9 Hz/0.5–3.9 Hz) | 6.011 | 3.944 | .128 |
| Theta + alpha/delta ratio (4.0–12.9 Hz/0.5–3.9 Hz) | .336 | 2.532 | .894 |
| Absolute asymmetry index (EASI) | −.002 | 0.017 | .920 |
| Theta-alpha variability | −1.474 | 0.975 | .131 |
Abbreviations: EASI, EEG asymmetry index ; EEG, electroencephalography; SD, standard deviation.
Figure 3.
Accuracy of differentiating between cerebral malaria and nonmalarial coma as measured by area under the curve using qualitative electroencephalography interpretation (A) or quantitative methods (B).
Multivariate modeling (Table 3) based on quantitative EEG features showed that CM diagnosis was independently associated with theta power (OR, 0.546 [95% CI.303–.983]; P = .044) and low-delta to delta ratio (OR, 5.771*e-05 [95% CI 1.890*e-07 to .018]; P = .001). Quantitative EEG features together achieved an average out-of-sample AUROC of 0.85 (ranging from 0.71 to 0.98; Figure 3B) and the result from a simple logistic regression classifier is 0.84 (ranging from 0.68 to 0.98), showing excellent predictive power of quantitative EEG features in differentiating children with CM from those with nonmalarial coma. Adding qualitative to quantitative EEG features slightly improved the ability of EEG to distinguish CM from nonmalarial coma cases (average AUROC = 0.90, ranging from 0.83 to 0.99; not depicted) and the result from a simple logistic regression classifier is 0.90 (ranging from 0.80 to 1.00). Stratification of both qualitative and quantitative EEG features by age did not impact test performance (Figure 4).
Figure 4.
Test efficiency (area under the receiver operating characteristic curve) of qualitative and quantitative electroencephalography features stratified by participant age. Abbreviations: AUC, area under the curve; EEG, electroencephalography; ROC, receiver operating characteristic.
DISCUSSION
In comatose febrile Malawian children, EEG interpretation using either qualitative or quantitative methods readily differentiates cases of CM from those of nonmalarial etiology. The sharp differentiation in EEG findings between these 2 categories identifies EEG as a diagnostic biomarker of pediatric febrile coma in malaria-endemic areas. These findings may be useful in areas of both high and low malaria transmission intensity.
In high-transmission areas, a child with clinical CM may have a positive mRDT due to recent previous malaria infection, or partial immunity, with secondary asymptomatic malaria parasitemia. If such a child becomes ill with bacterial meningitis, for example, a positive malaria test may prompt administration of antimalarials alone or provision of antimalarial agents without antibiotics of sufficient dose and duration to penetrate the central nervous system, failing to address the true cause of coma.
People living in new areas of resurgent malaria transmission may become gravely clinically ill at very low malaria parasite densities. In these clinical scenarios, EEG may serve as a complement to standard microbiological testing or blood and CSF to assess coma etiology.
Other laboratory factors that may also lead clinicians to suspect malarial coma (eg, thrombocytopenia) can be nonspecific. Low platelets may be seen in a variety of systemic neuroinvasive infections. As with EEG, thrombocytopenia may support the diagnosis of CM, but, like EEG, would not supplant eventual definitive malaria testing whether by blood smear or PCR.
Offering a real-time window into brain function, EEG provides clues to in vivo pathophysiologic differences between neurotropic pathogens. Viruses and bacteria usually invade brain tissue. Lower waveform amplitudes and lower quantitative power, with greater proportion of participants having faster (theta, beta) frequencies among the nonmalarial coma group, may reflect localized encroachment, which does not disrupt more extensively connected neural circuits.
Plasmodium falciparum, in contrast, impacts the brain without crossing the blood-brain barrier, via both local and global metabolic effects of sequestered, parasitized erythrocytes in cerebral vasculature [5] as well as systemic sequestration and high rates of multiorgan system dysfunction [18]. Higher amplitude waveforms among CM cases—supported by measures of total power, maximum voltage, average voltage, and peak envelope—may reflect a global derangement of neural networks that is further supported by a qualitative lack of variability and lack of background organization. Indeed, EEG recordings from encephalopathic patients with concomitant brain MRI reveal that diffuse delta-range slowing is associated with compromise of diffuse white matter tracts [19]. Abnormal (fragmented) sleep spindles in EEG recordings of CM potentially implicate impaired cortico-thalamic communication as well.
Not all effects of CM in the brain are seen on a global scale. It is possible that in more severe CM cases, brain perfusion may be compromised by parasite-mediated dearth of vasodilating nitric oxide [20] or obstruction to blood flow via thrombin-mediated coagulation [20, 21]. This may in turn lead to focal EEG abnormalities and, consequently, a degree of overlap in more localized EEG findings between some children with CM and those with nonmalarial coma etiologies.
This study has many advantages. Determination of whether coma etiology was due to malaria or a nonmalarial etiology was systematic and rigorous, supported by determination of plasma levels of parasite-encoded P. falciparum HRP-2. EEG recordings were acquired by the same technicians through the entire duration of the study, decreasing the likelihood of temporal bias. Clinicians caring for study participants also remained constant through the study enrollment period. EEG interpretations were performed systematically by trained pediatric experts providing independent ratings, and the consensus interpretations of the 2 neurophysiologists were matched in test efficient (AUROC) by quantitative (software-based) EEG interpretation. This holds promise for potential automation of EEG interpretation for these purposes. Additionally, test efficiency was not impacted when participants were stratified by age.
Our analyses have several limitations. Although EEGs were recorded at hospital admission, the time of EEG recording after illness (or coma) onset was not standardized. Certainly, it would be more valid to collect and compare EEG data collected at the same time point after patients lose consciousness, though logistically this would be challenging. Here, EEGs were collected for clinical care purposes upon hospital unit admission. Since seizures are common in both CM and nonmalarial coma, it is possible that children received prehospital treatments (including benzodiazepines) that could have modified EEG findings. Our findings are valid only for Malawian children and future studies should evaluate whether similar differences in EEG patterns occur in adults or in nonimmune child or adult visitors to malaria-endemic areas, or those living in areas of low malaria transmission intensity.
EEG services are not currently available in most levels of the Malawian healthcare system where febrile and comatose children present to care. Although more resource-intensive than PCR testing and malaria smear, which could arguably be scaled up as well, EEG may ultimately serve in the dual role of seizure detector (eg, treating status epilepticus and avoiding treatment of nonepileptic events) and as a confirmation of disease etiology in the not-infrequent cases of uncertain microbiological cause of coma.
Acquisition of recordings by trained technologists and interpretation by neurophysiologist staff are barriers to scaling of EEG services in many malaria-endemic countries. Easily applied EEG devices can address the former and systems of remote EEG interpretation may address the latter. In high-income contexts, continuous EEG recording is typically used in children in coma. Our results indicate that actionable information in coma can be obtained with a 30-minute EEG recording. A desire to meet the 24- to 48-hour seizure detection standard on EEG recording [22, 23] need not curtail efforts to scale EEG services in areas where a point-of-care test to diagnose a febrile coma etiology in 30 minutes may still be useful.
In summary, EEG is a valid, powerful diagnostic biomarker in pediatric febrile coma in Malawian children, able to readily differentiate children with CM from those with nonmalarial coma. Further studies are warranted to evaluate EEG's role in the clinical care of patients presenting in febrile coma in both malaria-endemic and nonendemic areas.
Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Notes
Author contributions. A. A. participated in data gathering and data analysis, drafted the manuscript, and edited the manuscript. D. B. H. and T. Z. participated in data gathering and editing of the manuscript. J. Z., B. Z., and J. B. participated in data analysis and editing of the manuscript. R. D., N. A., O. F., S. K., and E. B. participated in background research and editing of the manuscript. C. E. F., C. A. M., T. E. T., and N. O. participated in data gathering for participants in the parent natural history studies and edited the present manuscript. S. T. J. R. conceived of and oversaw the parent natural history studies and edited the present manuscript. D. G. P. conceived of the present study, participated in data gathering and data analysis, and edited the manuscript.
Acknowledgments. The authors acknowledge the contributions of Karl Seydel, MD, PhD (Michigan State University, Malawi Liverpool Wellcome Program), Elizabeth Kalanga, RN (Blantyre Malaria Project), Monica Sapuwa, RN (Blantyre Malaria Project), and Cynthia Manyozo, RN (Blantyre Malaria Project) in making this work possible.
Data availability. De-identified data (including raw electroencephalography [EEG] recordings and extracted quantitative and qualitative EEG features) can be supplied to qualified investigators who write to the corresponding author.
Financial support. This work was supported by the US National Institutes of Health (grant number R03NS124536 to A. A., T. Z., D. B. H., J. B., and D. P.).
Potential conflicts of interest. The authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.
References
- 1. Fitzpatrick W, Lowry N. PLEDs: clinical correlates. Can J Neurol Sci 2007; 34:443–50. [PubMed] [Google Scholar]
- 2. Lakeman FD, Whitley RJ. Diagnosis of herpes simplex encephalitis: application of polymerase chain reaction to cerebrospinal fluid from brain-biopsied patients and correlation with disease. National Institute of Allergy and Infectious Diseases Collaborative Antiviral Study Group. J Infect Dis 1995; 171:857–63. [DOI] [PubMed] [Google Scholar]
- 3. Lin L, Drislane FW. Lateralized periodic discharges: a literature review. J Clin Neurophysiol 2018; 35:189–98. [DOI] [PubMed] [Google Scholar]
- 4. World Health Organization (WHO) . WHO world malaria report. Geneva, Switzerland: WHO, 2023. [Google Scholar]
- 5. Birbeck GL, Molyneux ME, Kaplan PW, et al. Blantyre Malaria Project Epilepsy Study (BMPES) of neurological outcomes in retinopathy-positive paediatric cerebral malaria survivors: a prospective cohort study. Lancet Neurol 2010; 9:1173–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Postels DG, Birbeck GL. Cerebral malaria. Handb Clin Neurol 2013; 114:91–102. [DOI] [PubMed] [Google Scholar]
- 7. Postels DG, Taylor TE, Molyneux M, et al. Neurologic outcomes in retinopathy-negative cerebral malaria survivors. Neurology 2012; 79:1268–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Snow RW. Global malaria eradication and the importance of Plasmodium falciparum epidemiology in Africa. BMC Med 2015; 13:23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Hendriksen IC, White LJ, Veenemans J, et al. Defining falciparum-malaria-attributable severe febrile illness in moderate-to-high transmission settings on the basis of plasma PfHRP2 concentration. J Infect Dis 2013; 207:351–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Nolder D, Stewart L, Tucker J, et al. Failure of rapid diagnostic tests in Plasmodium falciparum malaria cases among travelers to the UK and Ireland: identification and characterisation of the parasites. Int J Infect Dis 2021; 108:137–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Schlabe S, Reiter-Owona I, Nordmann T, et al. Rapid diagnostic test negative Plasmodium falciparum malaria in a traveller returning from Ethiopia. Malar J 2021; 20:145. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Berhane A, Anderson K, Mihreteab S, et al. Major threat to malaria control programs by Plasmodium falciparum lacking histidine-rich protein 2, Eritrea. Emerg Infect Dis 2018; 24:462–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Perez-Ugarte A, Rubio Maturana C, Silgado A, et al. Unexpected false-negative result in a traveller's malaria diagnosis. J Travel Med 2023; 30:taac124. [DOI] [PubMed] [Google Scholar]
- 14. Ray STJ, Fuller CE, Ahmadu A, et al. Aetiology, neuroradiological features, long-term neurosequelae and risk factors for mortality of febrile coma in Malawian children: a prospective cohort study. Lancet Global Health 2024; 13:e1057–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Hirsch LJ, Fong MWK, Leitinger M, et al. American clinical neurophysiology society's standardized critical care EEG terminology: 2021 version. J Clin Neurophysiol 2021; 38:1–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Andrews A, Zelleke T, Harrar D, Izem R, Gai J, Postels D. Theta alpha variability on admission EEG is associated with outcome in pediatric cerebral malaria. J Clin Neurophysiol 2023; 40:136–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Postels DG, Wu X, Li C, et al. Admission EEG findings in diverse paediatric cerebral malaria populations predict outcomes. Malar J 2018; 17:208. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Bhutani A, Kaushik RM, Kaushik R. A study on multi-organ dysfunction syndrome in malaria using sequential organ failure assessment score. Tropical Parasitol 2020; 10:86–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Kaplan PW, Rossetti AO. EEG patterns and imaging correlations in encephalopathy: encephalopathy part II. J Clin Neurophysiol 2011; 28:233–51. [DOI] [PubMed] [Google Scholar]
- 20. Eisenhut M. The evidence for a role of vasospasm in the pathogenesis of cerebral malaria. Malar J 2015; 14:405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Chen Q, Schlichtherle M, Wahlgren M. Molecular aspects of severe malaria. Clin Microbiol Rev 2000; 13:439–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Claassen J, Mayer SA, Kowalski RG, Emerson RG, Hirsch LJ. Detection of electrographic seizures with continuous EEG monitoring in critically ill patients. Neurology 2004; 62:1743–8. [DOI] [PubMed] [Google Scholar]
- 23. Struck AF, Ustun B, Ruiz AR, et al. Association of an electroencephalography-based risk score with seizure probability in hospitalized patients. JAMA Neurol 2017; 74:1419–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Quantitative and quantitative EEG features not provided here due to space limitation can be made available (anonymized) at the request of a qualified investigator.