Epidemiology of central nervous system infectious diseases: a meta-analysis and systematic review with implications for neurosurgeons worldwide

ABBREVIATIONS AFR = African Region; AMR-L = Region of the Americas–Latin America; AMR-US/Can = Region of the Americas–United States/Canada; BM = bacterial meningitis; CNS = central nervous system; EMR = Eastern Mediterranean Region; EUR = European Region; HIC = high-income country; HIV = human immunodeficiency virus; LIC = low-income country; LMICs = lowand middle-income countries; MIC = middle-income country; NCC = neurocysticercosis; SEAR = Southeast Asia Region; TB = tuberculosis; WPR = Western Pacific Region. SUBMITTED March 16, 2017. ACCEPTED October 24, 2017. INCLUDE WHEN CITING Published online June 15, 2018; DOI: 10.3171/2017.10.JNS17359. * F.C.R. and J.R.L. contributed equally to this work and share first authorship. Epidemiology of central nervous system infectious diseases: a meta-analysis and systematic review with implications for neurosurgeons worldwide

conducted.The WHO regions are classified as follows: African Region (AFR), Region of the Americas-United States/Canada (AMR-US/Can), Region of the Americas-Latin America (AMR-L), Southeast Asia Region (SEAR), European Region (EUR), Eastern Mediterranean Region (EMR), and Western Pacific Region (WPR).The World Bank database (2016) was used to characterize the income level for each country using gross national income per capita.Country and region populations were also obtained from World Bank population metadata (https://data.worldbank.org/data-catalog/population-projection-tables).

Meta-Analysis
Data were analyzed with Comprehensive Meta-Analysis Version 3 (Biostat Inc.) and Stata 14.0 software (Stata-Corp).To account for variation between and within studies, the DerSimonian and Laird random-effects model was used to obtain overall incidence estimates and 95% confidence intervals. 49Forest plots allowed visualization of individual and summary estimates.Heterogeneity among studies was evaluated using Cochran's Q test (p < 0.10) and I 2 to measure the proportion of between-study variance.An I 2 value > 50% was considered high. 31Subgroup analyses by categorical covariates were used to explore potential sources of heterogeneity, that is, disease type (NCC, bacterial meningitis [BM], intracranial abscess, tuberculosis [TB] meningitis/osteomyelitis, and non-TB spinal osteomyelitis), WHO region, and World Bank income classification.Total world incidence of a disease was estimated by multiplying our derived WHO regional incidence by public WHO region population data, and all seven regions were summed to provide a global total.
The criteria for surgical versus nonsurgical CNS infectious disease were decided unanimously among the authors.Publications on cerebral malaria, cryptococcal meningitis, unspecified CNS infections, and human immunodeficiency virus (HIV)-related CNS infections were not included as those disease entities are primarily managed medically, with less relevance for neurosurgical intervention.To explore sources of heterogeneity, stratification by WHO region and income level was done for each disease category.Once aggregate incidence data were obtained for each disease category by WHO region and income level, these values were summed to a total disease volume of CNS infections.When published data were not available for a disease in a given WHO region, the incidence was estimated using the weighted proportion of HIC/MIC/ LIC within that region multiplied by the incidence rate of the disease by income level.When this method yielded an estimate greater than one deviation from the averaged incidences, the value underwent logarithmic transformation to prevent significant outliers. 124There were rare instances in which the literature provided no incidence rate of a disease for a particular income level.In these cases, the incidence was then removed from the meta-analysis.Potential publication bias was assessed using Egger's linear regression test and Begg's correlation test.If publication bias was indicated, the number of missing studies was evaluated by the trim-and-fill method.A p < 0.05 was considered significant unless otherwise indicated.

Incidence, Demographics, and Subtype
A total of 508,078 cases of CNS infections across all studies were included, with a total sample size of 130,681,681 individuals.There was a high degree of heterogeneity across all disease categories (not shown), with I 2 values ranging from 42.9% for toxoplasmosis to 79.6% for TB-spinal/cranial to 91.7% for NCC.The incidence of studied CNS infection was consistently highest in LICs, followed by MICs and then HICs (Table 2).Regarding WHO regions, Africa had the highest rates of BM, NCC, and TBrelated disease, whereas Southeast Asia had the highest reported rates of intracranial abscess, and Europe had the highest reported rates of non-TB spinal osteomyelitis (Table 3).The global burden of these five CNS infections by WHO region and the proportion of infections by pathology are depicted in Fig. 2. Case fatality was calculated from the included studies; however, because of the sporadic and limited reporting of mortality, the resultant calculations may represent an over-or underestimation of the true rates.

Intracranial Abscess
Eight studies on intracranial abscesses, which include extradural and subdural empyema, were suitable for inclusion, with two studies each from Africa and the Western Pacific, and one each from Latin America, US/Canada, Eastern Mediterranean, and Europe (Supplemental Figs. 1  and 2); estimates for Southeast Asia data were calculated as described above in the Methods.The overall estimate of the global population affected by intracranial abscesses was 1,088,237 each year (Table 3).Reported incidence ranged from 0.15 cases/100,000 people to 49/100,000.Incidence was lowest in the Western Pacific (0.15 cases/100,000 people, 95% CI 0.05-0.25)and highest in Southeast Asia (49/100,000, 95% CI 6-62).Reported incidence in HIC and MIC was similar at 0.54 cases/100,000 and 0.52/100,000, respectively (p-interaction = 0.19).Only one study from an LIC was included and had a reported incidence of 240 cases/100,000 people (95% CI 86-672). 184ase fatality data were not available for intracranial abscess in the included studies.

Nontuberculous Osteomyelitis
Ten studies on spinal nontuberculous osteomyelitis and discitis were suitable for inclusion (Supplemental Figs. 3  and 4).Eight studies were performed in Europe, and one each was performed in the Eastern Mediterranean and Southeast Asia.The overall estimate of the total population affected was 108,426 per annum (Table 3).The reported incidence from individual studies ranged from 0.27 cases/100,000 people (95% CI 0.16-0.45) to 39.7/100,000 (95% CI 34.5-45.8).All included studies originated from HICs, and the overall pooled estimate of incidence was 4 cases/100,000 people (95% CI 1-9).Case fatality for non-TB osteomyelitis was available in six studies, 24,37,76,82,186,187 giving an overall rate of 13.6% (95% CI 3.8%-23.5%),with the highest report of 26.8% in an elderly population in New Zealand 76 and the lowest (< 0.01%) in England. 187tudy quality was not a source of heterogeneity in a metaregression analysis (p = 0.99).

Tuberculous Meningitis/Osteomyelitis
Four studies on tuberculous meningitis and/or osteomyelitis were suitable for inclusion (Supplemental Figs. 5 and  6).Two studies originated from HICs in Europe, and one study each originated from a MIC in Africa and Southeast Asia.The overall estimate of the global population affected by TB meningitis or osteomyelitis was 1,005,612 each year (Table 3).The reported incidence from individual studies ranged from 0.51 cases/100,000 in a European HIC (95% CI 0.16-1.66) to 54.8/100,000 in an African MIC (95% CI 11.4-263.8;Supplemental Figs. 5 and 6).Case fatality for TB osteomyelitis was only available in one Turkish study, with a case fatality of 21.1%. 78All deaths were associated with stage III recrudescent disease.

Study Quality
Study quality was assessed as a potential source of heterogeneity for each of the outcomes and hence was included in a meta-regression analysis.For BM, there was no evidence of an effect modification by study quality (slope = -0.82,p = 0.21).After adjusting for WHO region, income level, and study quality in the model, the multivariate meta-regression revealed that study quality was not a source of heterogeneity (p = 0.15).Similarly, for NCC, meta-regression revealed there was no evidence of an effect modification by study quality (slope = -0.65,p = 0.41).This remained unchanged even after adjusting for WHO region (p = 0.18) or income level (p = 0.14).Study quality   was not a source of heterogeneity for intracranial abscess (p = 0.53), nontuberculous osteomyelitis (p = 0.99), or tuberculous meningitis/osteomyelitis (p = 0.60).

Publication Bias
Begg's p value and Egger's p value were nonsignificant for each of the five CNS infection types, except for NCC where the Egger's test was significant (p = 0.02; Table 5).The funnel plot was asymmetrical for all of the different disease outcomes, which suggests that studies showing a higher incidence could be missing, and, consequently, our results could be an underestimation.Because at least 10 studies are needed to assess publication bias with a funnel plot, assessment of bias could not be completed for tuberculous meningitis/osteomyelitis (4 studies).

Discussion
This study represents the first comprehensive estimate of the global burden of neurosurgically relevant CNS infectious diseases.The results of this systematic review and meta-analysis underscore the epidemiological disproportions of disease by region, income status, and pathogen.Areas of lower income are particularly vulnerable to the persistence and spread of infection due to poverty, overcrowding, inadequate access to clean water and proper sanitation systems, and insufficient access to health care overall. 72,119,171In our study of the five categories of CNS infection, LICs had an overall incidence of 726 cases/100,000 people and MICs had 299/100,000, compared with approximately 11/100,000 in the high-income counterparts.Furthermore, LMICs have the smallest neurosurgical workforce, with many LMICs in Sub-Saharan Africa having only one neurosurgeon per 3-10 million constituents, in contrast to one per 20,000-60,000 in Europe and other HICs. 1,52,147,162The limited workforce and resources in LMICs compounded by the increased burden of CNS infection highlight the importance of this public health issue.

Bacterial Meningitis
Bacterial meningitis remains a significant cause of CNS infection worldwide, particularly in the region of Sub-Saharan African known as the "meningitis belt." 6,116Our study revealed an annual incidence of 65 cases/100,000 people for the AFR, corroborating findings from the 2014 WHO Global Health Observatory database, which reported an annual incidence ranging from 0.03/100,000 in Mauritania to 227/100,000 in the Democratic Republic of Congo (http://apps.who.int/gho/data).These similar conclusions served as validation for our methodology.Moreover, the significant contrast between the incidence in Africa and more Westernized regions such as the US/Canada and Europe (2 and 6/100,000, respectively) reinforces that BM is primarily a disease of the developing world.
Although BM is initially managed medically, unsuccessful or untimely treatment can result in sequelae, such as postinfectious hydrocephalus and subdural empyema, which can require neurosurgical intervention. 64Global efforts to better characterize the transmission patterns of BM in endemic regions are ongoing. 6,30,136

Neurocysticercosis
Neurocysticercosis was the most reported individual disease in publications of CNS infections worldwide (27 of 71 studies).The burden of NCC was most prominent in Africa and in LMICs, where neurosurgeons are few, 51 with an incidence of 650 and 401 cases/100,000 people, respectively.Latin American countries contributed the greatest amount of high-quality research on NCC, and WPR and SEAR had substantial rates of infection as well.Our results corroborate that NCC has been a known public health problem in these four WHO regions and in individual LMICs because of poor hygiene, methods of pig management and slaughter, and inadequate waste and water management. 151Importantly, our estimates may involve selection bias as many population-based studies were conducted in villages and regions of known endemnicity.For instance, Secka et al. conducted a population-based radiographic survey within the Soutou village in a historically endemic region of Senegal, which showed an incidence of 516 cases/100,000 people.In contrast, several studies focused only on symptomatic disease by screening within epileptic populations, which may have underestimated the total incidence of NCC infection.The lowest regional in-cidence of NCC was noted in the Eastern Mediterranean at 0.23 cases/100,000 people.One proposed explanation is that predominantly Muslim regions have a smaller pork industry and therefore a substantially lower rate of cysticercosis and subsequent fecal-oral transmission leading to NCC. 114 Neurocysticercosis can require neurosurgical intervention for biopsy, treatment of hydrocephalus, or alleviation of mass effect. 16,155Furthermore, as the most common cause of acquired seizures worldwide, NCC-related refractory epilepsy often necessitates surgery. 36,181Overall, NCC provokes sequelae that require surgical intervention and contributes to the burden of neurosurgical disease, particularly in endemic areas.

Intracranial Abscess
While our study showed a globally low incidence of intracranial abscess (1 case/100,000 people), LICs had a significantly higher burden of disease.Latin America had the highest incidence among the countries with available figures (19 cases/100,000 people), although this statistic was based on a single study in the West Indies. 36The FIG. 2. Global incidence and burden of CNS infection.For the five CNS infection types studied, the combined incidence (A) and global burden (B) of CNS infection are depicted, as are proportions of infection by pathology (C).Publications on cerebral malaria, cryptococcal meningitis, unspecified CNS infections, and HIV-related CNS infections were not included since those are primarily medically managed disease entities with less relevance for neurosurgical intervention.Map reproduced with permission from OpenStreetMap Contributors, CC BY-SA 2.0 (http://www.openstreetmap.org/copyright).Figure is available in color online only.calculation of incidence in Southeast Asia using income data estimates (see Methods) showed abscess rates of 49 cases/100,000 people.The etiology of brain abscess is believed to relate directly to poor socioeconomic conditions, 130 including both hematogenous and direct spread from pneumonia, poor dental hygiene and associated cardiac valvular vegetations, previous surgery, or other contiguous sites. 29,192Successful management generally requires surgical drainage for both diagnostic and thera-peutic purposes, with aerobic, anaerobic, mycobacterial, and fungal cultures to guide medical treatment; 13 however, LICs are often forced to rely on empirical medical management because of limitations in both radiographic and surgical resources.For instance, in a series of 112 cerebral abscesses in Burkina Faso, 47% of patients were treated with antibiotics alone, while 53% underwent surgery. 85his contrasts with higher-income countries, where nearly 90% of cases undergo surgical treatment 28,80,130 and medi-  cally managed patients are closely monitored with serial imaging. 61Overall, reduced mortality in this disease process with the advent of CT imaging and neurosurgical intervention has been corroborated in other studies, 29,130,191 emphasizing the importance of both imaging and surgery as standard care.

Tuberculosis-Related CNS Disease
Tuberculosis-related CNS disease includes TB meningitis and TB spondylitis.The WHO data repository reports an average annual incidence of systemic TB infection as 244 cases/100,000 people over the last 5 years, of which spondylodiscitis and meningitis represent approximately 2% and 3%, respectively. 175,197In our study, the region with the highest noted incidence was Africa with a rate of 55 cases/100,000 people based on the single study in Cameroon by Lootie et al. 101 Interestingly, our review demonstrated that much of the known epidemiological data for CNS TB is limited to MICs and HICs, with no LIC included in the quantitative meta-analysis.
These severe forms of extrapulmonary TB can cause significant morbidity and mortality via the development of hydrocephalus, encephalitis, vasculitis, thrombosis, infarc-tion, or spinal column invasion and compressive myelopathy.In a Danish study of patients with spinal TB, 54/133 (40.6%) had neurological deficits on admission and 17.3% presented with cauda equina syndrome, requiring surgical intervention. 84In a Filipino study of TB meningitis in children, the mortality rate was 16%, and 71% of patients exhibited signs of hydrocephalus. 93Thus, neurosurgeons have a role to intervene in CNS TB for biopsy, spinal decompression and stabilization, and CSF diversion.Furthermore, despite few publications on the epidemiology of CNS TB in LICs, it is suspected that in these countries, the demand for CNS TB treatment is highest and should be investigated further.

Nontuberculous Osteomyelitis
In our study, the nontuberculous vertebral spondylodiscitis was the least prevalent of the five diseases, affecting approximately 100,000 people each year worldwide.The surgical intervention rate ranged from 9.2% described by Grammatico et   111 Of note, all 10 included studies were reflective of HICs. 17,24,62,66,69,75,101,110,157,160,185This does not imply that the incidence of non-TB spondylodiscitis is zero for all LICs and MICs, but rather that it is simply not known.Note, however, that the high incidence in HICs indicates that vertebral osteomyelitis may be a disease afflicting the developed world.Known risk factors include injection drug use, diabetes, and a compromised immune system, 7 several of which are prevalent in more developed nations but are becoming increasingly common in the developing world. 131,149Our data indicate that vertebral osteomyelitis is not currently a major contributor to the global neurosurgical infectious volume; however, the continuation of ongoing epidemiological trends may show a rise in vertebral osteomyelitis in the near future.

Study Limitations and Future Directions
The disease estimations presented in this study are based on the most wide-ranging and comprehensive studies available.Strengths of this study include the comprehensive literature search, large number of included studies, and emphasis on the inclusion of data from developing countries.However, the findings are not without limitations.
While the five disease categories studied include pathology that often requires surgical intervention, this list is not comprehensive and omits multiple diseases that may indeed require neurosurgery, such as HIV-related infections.However, less common diseases that were omitted, such as cerebral schistosomiasis, prion disease, and neuroborreliosis, are unlikely to greatly affect the estimated volume of CNS infections.
Additionally, regions in which CNS infections are suspected to occur with the greatest frequency are also those regions with studies of the lowest methodological quality.The exclusion of large regional and national registries excludes a large body of data that could have greatly impacted the power of our approximations.To include data from the greatest possible number of regions and achieve our goal of describing the global CNS infectious disease volume, we used a lower methodological threshold for studies from LMICs.This may have resulted in either an overestimation or underestimation of incidence.Furthermore, many of the included studies focused on populations with a known high prevalence of disease, further decreasing the  generalizability to entire regions.However, this strategy of using a lower threshold for underrepresented regions follows precedent. 31Furthermore, gold-standard diagnostic testing is frequently unavailable in low-resource settings.Unsurprisingly, included studies with a broader definition of infection (such as seropositivity for cysticercosis and history of seizure) had a higher rate of infection than those studies with a stricter definition of infection.While we report the volume of CNS infections that are most likely to require surgery, our estimates for true neurosurgical demand were restricted because few publications specifically indicated the proportion of patients requiring surgery in their investigation.It would have been ideal to show this proportion of surgical cases for each disease category, as well as to describe and quantify the neurosurgical procedures performed in patients with neurosurgical infections on a global scale.This would allow comparison of neurosurgical treatment of CNS infections across WHO regions, which could have differing management options dictated by the availability of resources.Finally, heterogeneity remained high in most categories despite stratifying by trial-level covariates; this could be attributable to residual confounding or variance in true rates rather than sampling error.Study quality did not appear to be a source of heterogeneity, likely because the studies that provided incidence and were consequently entered in the meta-analysis were generally of moderate to high quality.While these limitations are substantial, they highlight the difficulty in obtaining high-quality epidemiological data on infectious diseases affecting the CNS, particularly in lowresource settings.Further research should specifically address the limitations of this study, with particular emphasis on population-based epidemiological studies in LMICs.

Conclusions
Here, we performed a systematic review and metaanalysis of more than 10,000 titles and ultimately included  154 articles to estimate the global volume of CNS infections that may require neurosurgical intervention.To our knowledge, this is the first study of its kind.In total, these studies included over 130,681,681 patients across 44 countries.While our study provides an approximation, our results can be used as a benchmark against which local disease incidence can be compared.Overall, we have described the estimated global and regional incidence of five CNS infectious diseases of neurosurgical relevance.The strength and quality of epidemiological information regarding neurosurgical disease in the developing world is certainly lacking, yet those populations are the ones most afflicted by the burden of infectious CNS disease.The data in this study provide the neurosurgical community with an inclusive estimate of the global incidence of CNS infectious disease.

FIG. 1 .
FIG. 1. PRISMA diagram summarizing the search process.From a total of 10,906 studies, 154 were incorporated into the review, with 71 in the quantitative analysis.

FIG. 3 .
FIG. 3. Forest plot demonstrating distribution of overall incidence of BM by WHO region.Twenty-two studies on BM were included, and data were analyzed according to WHO region.Overall incidence was highest in Africa and lowest in AMR-US/Can.Solid squares represent the point estimate of each study, and the diamonds represent the pooled estimate of the incidence for each subgroup.The width of the diamond denotes 95% CIs.The size of the solid squares is proportional to the weight of the study.ES = effect size. Figure is available in color online only.

F
. C. Robertson et al.

FIG. 4 .
FIG. 4. Forest plot demonstrating distribution of overall incidence of BM by World Bank income level.Twenty-two studies on BM were included, and data were analyzed according to LIC, MIC, and HIC World Bank indication.Figure is available in color online only.

HFIG. 5 .
FIG. 5. Forest plot demonstrating distribution of overall incidence of NCC by WHO region.Neurocysticercosis was the most reported individual disease in publications of CNS infections worldwide (27 of 71 included in the final statistical analysis).Overall incidence was highest in AFR and lowest in EMR.Solid squares represent the point estimate of each study, and the diamonds represent the pooled estimate of the incidence for each subgroup.The width of the diamond denotes 95% CIs.The size of the solid squares is proportional to the weight of the study.Figure is available in color online only.

FIG. 6 .
FIG. 6. Forest plot demonstrating distribution of overall incidence NCC by World Bank income level.Twenty-seven studies regarding NCC were analyzed according to LIC, MIC, and HIC World Bank indication.Solid squares represent the point estimate of each study, and the diamonds represent the pooled estimate of the incidence for each subgroup.The width of the diamond denotes 95% CIs.The size of the solid squares is proportional to the weight of the study.Figure is available in color online only.

TABLE 1 . Review of 71 studies included in the quantitative meta-analysis
F. C. Robertson et al.

TABLE 3 . Worldwide estimated incidence for CNS infectious diseases separated by WHO region per 100,000 people, and total number of individuals affected per annum worldwide
* Meningitis/osteomyelitis.†No income data available for estimation of mean incidence or number affected per year.Value counted as zero for worldwide estimate.‡ Calculated based on income data for WHO region.
al. in France 69 to 96.8% by Menon et al. in Oman.

TABLE 5 . Publication bias results for 5 disease outcomes
Boldface type indicates statistical significance.* Meningitis/osteomyelitis. † Need at least 10 studies to assess publication bias.