This study was prospectively conducted (January 2003–August 2011) with inpatients on all wards of Mahosot Hospital in Vientiane (17.959431°N, 102.613144°E, 188 m above mean sea level), an ≈400-bed hospital providing primary, secondary, and tertiary care and admitting ≈2,000 patients/month. We recruited inpatients of all ages for whom diagnostic lumbar puncture was indicated for suspicion of CNS infection because of altered consciousness or neurologic findings and for whom lumbar puncture was not contraindicated. For patient inclusion, we used no formal definition for CNS infection; patient recruitment was at the discretion of the responsible physician, reflecting local clinical practice. We recorded patient history and examination findings on standardized forms.

We obtained verbal (2003–2006) or written (2006–2011) informed consent from all recruited patients or close relatives. Ethics clearance was granted by the Ethical Review Committee of the Faculty of Medical Sciences, National University of Laos (Vientiane, Laos), and the Oxford University Tropical Ethics Research Committee (Oxford, UK).

We classified febrile patients meeting the World Health Organization (WHO) criteria for encephalitis or meningitis (Figure 1) (18) as patients with WHO clinical CNS (hereafter WHO CNS) infection. Because of the overlapping WHO case definitions for encephalitis and meningitis, which both include a Glasgow Coma Scale (GCS) score <15 as criteria, we created additional classifications for febrile patients: those with stiff neck; reduced GCS score (<15), seizures, or both; stiff neck and reduced GCS score, seizures, or both; no stiff neck but reduced GCS score, seizures, or both; and stiff neck, a GCS score of 15, and no seizures (Table 1).

Cerebrospinal fluid (CSF) was collected from patients (≈8 mL for adults [defined as patients >15 years of age], ≈3.5 mL for children 1–14 years of age, and ≈2.5 mL for children <1 year of age), and opening pressure was recorded. Venous blood (≈18.5 mL for adults, 10 mL for children 1–14 years of age, and 5.5 mL for children <1 year of age) was drawn on the same day. We aimed to collect ≈2 mL follow-up serum 7–10 days after lumbar puncture. Specimens were transported to the laboratory within ≈30 minutes, and we aliquoted and immediately tested or stored them at −80°C. We submitted all patient samples for a panel of laboratory tests: complete blood count; biochemistry panel; culture; and serologic and molecular assays for a range of bacteria, viruses, parasites, and fungi (Table 2; Appendix). HIV-1 and HIV-2 rapid diagnostic tests were performed when indicated by the physician. Computed tomography brain scan was available starting in 2002 but rarely used, especially for intensive care patients, because of difficulties transferring patients. Magnetic resonance imaging and electroencephalographic facilities were not available.

The confirmed etiology was determined by the results of a panel of diagnostic tests (Table 2), which included tests for the direct detection of pathogens in CSF or blood, specific IgM in CSF, seroconversion, or a 4-fold rise in antibody titer between admission and follow-up serum samples. Pathogen detection was confirmed after critical analysis of test results to rule out possible contamination. When evidence of >1 pathogen was obtained for a patient, we defined the confirmed etiology as detection by direct tests over indirect tests (antibody-based tests) and prioritized CSF detection over blood detection (39). We defined a confirmed co-infection as the direct (or indirect, if only indirect tests were positive) detection of >1 pathogen in the same matrix (CSF or blood).

We compared patients with confirmed bacterial infection (including co-infections involving only bacteria) and patients with confirmed viral infection (including co-infections involving only viruses) with all other patients, excluding those with mixed co-infections (i.e., co-infections with fungi or infections with both bacteria and viruses). We investigated the factors associated with death (died in hospital or discharged moribund), bacterial infection, or viral infection by univariate analysis using the χ² or Fisher exact test for categorical variables or the Mann-Whitney U test for continuous variables. We analyzed the independent predictors of death, bacterial infection, and viral infection using multivariate logistic regression models. In multivariate analyses, we included all factors having a p value <0.010 in the univariate analysis.

For variables with 6%–20% of the values missing, we executed multiple imputation models using chained equations and used a number of imputations that exceeded the highest proportion of missing values (40). We added age and sex to imputation models as auxiliary variables. We specified the imputation methods as linear for continuous normally distributed variables, logistic for binary variables, ordered logistic for ordinal variables, and predictive mean matching for continuous skewed variables. We performed logistic regression with the dependent variable (death, bacterial or viral infection) and all relevant covariates on each imputed data set and combined results using Rubin rules to take into account the variability in estimates among imputed data sets (41). Only variables that were significant (p<0.050) were retained in the final models. For comparison of analysis methods, we provided the results using the corresponding complete case analysis. We conducted analyses using Stata/SE version 14.0 (StataCorp, https://www.stata.com).

In total, 1,065 inpatients with suspected CNS infection consented to study participation (Appendix Figure 1); 80% were recruited from the pediatric and adult intensive care wards and adult infectious disease ward. On each ward, ≈8 physicians were in charge of patient recruitment. All were general hospital or infectious disease physicians; none were neurologists. We collected information on patient demographics, clinical presentation, and blood and CSF parameters (Table 3–4; Figure 2). More patients were recruited during the rainy season (Figure 3). The median time between admission and follow-up blood collection was 9 (interquartile range [IQR] 6–16) days. One third (33.6%, 358/1,065) were children <15 years of age (Table 3; Appendix Table 2).

Of 476 adults tested for HIV, 118 (24.8%) were seropositive; of 227 children tested, 1 (0.4%) was seropositive. More than half (61.9%, 590/953) of patients had a history or hospital chart evidence of antibiotic use before lumbar puncture. Most (90.8%, 962/1,059) patients had a history of fever or documented admission fever. The median length of fever at admission was 4 (IQR 2–8) days. The most frequent symptoms and signs were headache (88.1%, 787/893); neck stiffness (64.2%, 683/1,064); confusion (57.4%, 608/1,060); GCS score <15 (52.6%, 551/1,047); and vomiting, diarrhea, or both (54%, 575/1,064). All symptoms and signs were more frequent in children than adults (p<0.05), except headache, which was more frequent in adults (p = 0.023). Most (93.6%, 832/889) patients had CSF findings outside reference ranges (elevated CSF white cell count, elevated CSF lactate, elevated CSF protein, low CSF glucose, or any combination of these parameters) (Table 4; Appendix Table 3, Figure 2).

Etiology was confirmed in 450 (42.3%) patients; 413 (38.8%) had monoinfections and 37 (3.5%) co-infections (Appendix Tables 4–8). The identified monoinfections were JEV (8.8%, 94/1,065), Cryptococcus spp. (6.6%, 70/1,065; 9 C. gattii), Orientia tsutsugamushi (2.9%, 31/1,065), dengue virus (2.5%, 27/1,065), Leptospira spp. (2.3%, 25/1,065), Rickettsia spp. (2.3%, 24/1,065), S. pneumoniae (2.1%, 22/1,065), M. tuberculosis (1.9%, 20/1,065), HSV-1 or HSV-2 (1.4%, 15/1,065), human cytomegalovirus (1.1%, 12/1,065), enterovirus (0.9%, 10/1,065), varicella zoster virus (0.6%, 6/1,065), mumps virus (0.5%, 5/1,065), and Plasmodium falciparum (0.4%, 4/1,065). Other bacteria were detected in 48 (4.5%) patients (Figure 4; Appendix Table 5). All samples were negative for West Nile virus, influenza A and B, Henipavirus, and measles virus by PCR. Infection by M. tuberculosis, Cryptococcus spp., or varicella zoster virus was not detected in children (Appendix Table 8). The median age of children with enterovirus infection was 4.5 (IQR 1–11) years and JEV infection 13 (IQR 8–20) years. The proportion of patients with JEV infection was higher for children (14%, 50/358) than adults (6%, 44/707, p<0.001). Significantly more enterovirus patients (80%) than nonenterovirus patients (33%; p = 0.002) were children.

We compared patients with single (n = 170) or multiple (n = 5) bacterial infections (excluding co-infections with viruses or fungi) with all other patients (n = 875). Factors significantly associated with bacterial infections on univariate analysis (p<0.01; Appendix Table 9) were included in multivariate analysis. Diabetes (adjusted odds ratio [aOR] 3.1, 95% CI 1.2−7.7), history of fever or fever at admission (aOR 3.9, 95% CI 1.4−11.1), higher serum C-reactive protein (aOR 1.08, 95% CI 1.05−1.11), and higher CSF lactate (aOR 3.5, 95% CI 2.3−5.4) were independent predictors of bacterial infection (Appendix Table 10).

We compared patients with single (n = 169) or multiple (n = 3) viral infections (excluding co-infections with bacteria or fungi) with all other patients (n = 867). Factors significantly associated with viral infections on univariate analysis (p<0.01; Appendix Table 11) were included in multivariate analysis. Neck stiffness (aOR 1.9, 95% CI 1.3−2.8) and higher hematocrit (aOR 1.4, 95% CI 1.1−1.9) were associated with viral infection, whereas higher CSF lactate (aOR 0.3, 95% CI 0.1–0.5), older age (aOR 0.8, 95% CI 0.7–0.9), and longer interval between admission and lumbar puncture (aOR 0.9, 95% CI 0.8–1.0) were negatively associated with viral infection (Appendix Table 12).

In total, 771 (74.1%) of 1,040 patients had WHO CNS infection; 44.2% of these patients had confirmed etiologies compared with 37.9% of patients not fulfilling WHO CNS infection criteria (p = 0.063; Appendix Table 13). Because of the considerable overlap between the WHO encephalitis and meningitis definitions, 551 (71.5%) patients were classified as having meningoencephalitis. Therefore, we analyzed the frequency of neck stiffness, reduced GCS score, and seizures among febrile patients with clinical CNS infection (Table 1).

When comparing viral and bacterial infections, we observed no significant differences (p>0.05) in the proportions of encephalitis and meningitis syndromes, although differences were observed for some specific etiologies (Figure 4). In total, 90 (53.6%) febrile patients with viral infection and 86 (49.7%) with bacterial infection had neck stiffness and reduced GCS score, seizures, or both; 17 (10.1%) patients with viral infection and 16 (9.3%) with bacterial infection had reduced GCS score, seizures, or both without neck stiffness; and 36 (21.4%) patients with viral infection and 38 (22.0%) with bacterial infection had neck stiffness, a GCS score of 15, and no seizures (Figure 4). We obtained similar results using the WHO definitions. In total, 25 (14.9%) patients with viral infection and 33 (19.1%) with bacterial infection did not fulfill the WHO CNS infection definition.

In comparison with the distribution of syndromes observed for all patients, the distribution in patients with some etiologies were significantly different (p<0.05). Of the 89 JEV patients with WHO CNS infection, 75.3% had fever; neck stiffness; and reduced GCS score, seizures, or both. Of the 26 O. tsutsugamushi patients with WHO CNS infection, 50% had fever, neck stiffness, a GCS score of 15, and no seizures. Of note, almost half (47.8%) of the patients with cryptococcal infection did not fulfill the definition for WHO CNS infection, and of the 36 who did, 55.6% had fever, neck stiffness, a GCS score of 15, and no seizures.

Of 893 patients, 235 (26.3%) died, including those discharged moribund; we compared them to the 658 (73.7%) patients discharged alive and not moribund. For factors significantly associated with death (p<0.01; Appendix Table 14) in the univariate analysis, we conducted multivariate analysis. The variables strongly associated with death were higher CSF lactate (aOR 1.1, 95% CI 1.0–1.1) and reduced GCS score (aOR 0.8, 95% CI 0.8–0.9). Patients with viral infection were less likely to die than those with other diagnoses (aOR 0.4, 95% CI 0.3–0.7) (Appendix Table 15). Diabetes and hyperglycemia (glucose >7.7 mmol/L) at admission were not associated with death.

In total, 56 patients (12.4% of the 450 patients with confirmed etiologies) were infected with bacteria treatable by ceftriaxone and 64 patients (14.2% of the patients with confirmed etiologies) with bacteria treatable by doxycycline but not ceftriaxone (Table 5). Twenty-eight patients were infected with a Leptospira spp. treatable by ceftriaxone or doxycycline, but 2 were co-infected with O. tsutsugamushi not treatable by ceftriaxone. Of 142 patients infected by bacteria treatable by ceftriaxone or doxycycline, 89 (62.7%) received appropriate treatment, 17% (13/77) of whom died. In comparison, 25% (12/48) of the patients who did not receive appropriate treatment died (p = 0.270).

Including the 450 patients with confirmed diagnoses, we analyzed the criteria for bacterial meningitis commonly considered when making decisions on antibiotic treatment: elevated CSF white cell count, elevated CSF lactate, elevated CSF protein, decreased CSF glucose, reduced GCS score, turbid CSF, and neck stiffness. A low percentage (<23%) of patients with any 1 of these criteria (except turbid CSF, 38.8%) or a combination of these criteria had bacterial infections treatable by ceftriaxone or doxycycline (Appendix Table 16). Furthermore, only 1 combination of criteria (elevated C-reactive protein, elevated CSF protein, or elevated CSF lactate or any combination of these criteria) could identify all patients infected with bacteria treatable by ceftriaxone (Table 5). However, because only 5% of our patient series did not display this combination of criteria, none of the analyzed clinical and biological results can be reliably used to guide decisions on antibiotic use.