Estimating the global incidence of traumatic brain injury

ABBREVIATIONS AFR = African Region; AMR-L = Region of the Americas–Latin America; AMR-US/Can = Region of the Americas–United States and Canada; EMR = Eastern Mediterranean Region; EUR = European Region; GBD = Global Burden of Disease; GBD 2015 = GBD Study 2015; HI = head injury; HIC = high-income country; IHME = Institute for Health Metrics and Evaluation; LIC = low-income country; LMICs = lowand middle-income countries; MIC = middle-income country; MOI = mechanism of injury; P(RTI) = probability that a member of the population will sustain an RTI annually; P(RTI|TBI) = probability that TBI is secondary to RTI; P(TBI|RTI) = probability that an RTI will lead to a TBI; PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses; RTI = road traffic injury; RTITOTAL = total number of RTIs in a country annually; RTI∩TBI = intersection of RTITOTAL and TBITOTAL; SEAR = Southeast Asian Region; TBI = traumatic brain injury; TBITOTAL = total number of TBI cases in a region annually; WB = World Bank; WHO = World Health Organization; WPR = Western Pacific region; ∩ = intersection of 2 events; | = conditional on 1 event. SUBMITTED February 8, 2017. ACCEPTED October 18, 2017. INCLUDE WHEN CITING Published online April 27, 2018; DOI: 10.3171/2017.10.JNS17352. Estimating the global incidence of traumatic brain injury

T raumaTic brain injury (TBI), often referred to as the "silent epidemic," 208,253 remains a growing public health concern and represents the greatest contributor to death and disability globally among all trauma-related injuries. 206 Previous studies from the United States and New Zealand have estimated approximately 500-800 new cases of TBI per 100,000 people each year. 83,209 However, estimates of the TBI burden from low-and middle-income countries (LMICs) are scarce. A large survey-based study in 8 LMICs identified a lifetime prevalence of TBI from < 1% (China) to nearly 15% (Mexico and Venezuela) of the studied population, with most estimates approximating those from high-income countries (HICs). 134 Efforts to identify reliable epidemiological data on the incidence of and the disability and mortality from TBI in resourcepoor settings are still needed.
Road traffic collisions are a significant source of TBI. 24,161,216,231,257,259 Using national registries, populationbased literature, and statistical modeling, the Global Burden of Disease (GBD) Study 2015 (GBD 2015) 116 estimated the incidence of road traffic injuries (RTIs) in countries worldwide. By understanding the relationship between RTI and TBI, the incidence of TBI can be estimated. Because the interaction between RTI and TBI probably differs across regions of various populations, regulations, and infrastructures, a region-specific estimate of this relationship is essential to ensure accurate TBI estimates.
Beyond a fundamental disparity in quality data, a majority of the global population resides in LMICs, underscoring the need for reliable estimates of the TBI burden in resource-poor settings. In this comprehensive review, we provide estimates for TBI across geographic regions and income groups to deliver a global estimate of the volume and burden of TBI worldwide.

Overview
Incidence figures and overall disease volume were calculated from literature reviews, national registries, the GBD initiative, and the World Bank (WB). A similar methodology of estimating the frequency of traumatic injuries in LMICs has been described elsewhere. 4,149 A flowchart illustrates the contribution of relevant data sources and a stepwise progression in our methodology (Fig. 1). Initially, the incidences of RTI in different countries were obtained from the Institute for Health Metrics and Evaluation (IHME) GBD dataset. The incidences of RTIs were converted to population-based rates, that is, P(RTI), which represents the proportion of RTIs in a given population (that is, the probability that a person living in a country will sustain an RTI in a year; Fig. 1 II). By multiplying P(RTI) by the country population, we were able to obtain the RTI TOTAL , representing the total number of RTIs in a country annually (Fig. 1 III): We next sought to obtain the number of patients who had sustained a TBI or head injury (HI) from total RTIs, represented by RTI∩TBI (that is, the intersection of RTI TOTAL and TBI TOTAL ; Fig. 1). To this end, we conducted a systematic review and meta-analysis of studies reporting the proportion of RTIs that had resulted in TBIs in different WHO regions and income groups. This proportion is expressed as a probability value, P(TBI|RTI), that is, the proportion sustaining TBI after RTI (Fig. 1 IV): . By multiplying the RTI TOTAL by P(TBI|RTI), we obtained the total number of patients who sustained TBI after RTI ( Fig. 1 V): Understanding that traffic collisions are one of many TBI mechanisms of injury (MOIs), we sought the proportion of TBIs that are caused by RTIs. Accordingly, we conducted another systematic review and meta-analysis, this time to quantify the proportion of TBIs with RTI as the MOI in different WHO regions and income groups. This proportion is expressed as a probability value, P(RTI|TBI), that is, the proportion of TBIs secondary to RTIs ( Fig. 1 VI): . Multiplying RTI∩TBI by the inverse of P(RTI|TBI) and thereby accounting for the non-RTI causes of TBI, we obtained the total number of TBI cases annually in different WHO regions and income groups ( Fig. 1 VIII): . A more detailed explanation of our methodology is outlined below.

Incidence of RTIs
To identify the proportion of the population that sus-tains an RTI every year, we obtained and extracted relevant data on the incidence of RTI from the IHME GBD dataset by using the open-access site vizhub.healthdata. org/epi/. 116 Region-specific data sources included the World Health Organization (WHO) regional office, ministries of health, and so forth, while mixed-effects-modeled IHME GBD data were excluded. Regions not recognized by the WB or the WHO were excluded (for example, Tibet). In a few instances, the incidence for only one sex was provided; therefore, to maintain uniformity and generalizability, incidence data that included male and female sex were selected for analysis over incidence data for just one FIG. 1. Methodology flow diagram. P(RTI) = probability that a member of the population will sustain an RTI annually; P(RTI|TBI) = probability that a TBI is secondary to an RTI; P(TBI|RTI) = probability that an RTI will lead to TBI; RTI TOTAL = total cases of RTI, with or without TBI; RTI∩TBI = intersection of RTIs and TBIs, thus representing either the number of RTIs that lead to TBI or the number of TBIs secondary to RTIs; TBI TOTAL = total cases of TBI, whether the mechanism is an RTI or a non-RTI. Figure is available in color online only.
sex. Incremental age and sex values were averaged, and mean incidence values were estimated for each country. Each incidence value was expressed as the probability that a person living in a country would sustain an RTI annually [P(RTI); Fig. 1 II and III]. The total number of RTIs (RTI TOTAL ) per WHO region and WB income group was calculated as the product of the WB 2015 population metadata 260 and the GBD RTI data. The WB metadata are modeled figures to project population changes over time.

Proportion of RTIs Causing TBIs
The probability of sustaining a TBI after an RTI is represented by P(TBI|RTI). This is equal to the ratio of RTIs with TBI to all RTI cases with or without TBI in a population ( Fig. 1 IV). To identify this proportion, a systematic literature search of PubMed was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for studies reporting the proportion of RTIs resulting in TBI or HI (Fig. 2). 173 The aim of the search was to comprehensively identify population-and hospital-based studies quantifying the injuries resulting from traffic accidents. Thus, a series of searches were performed to capture a wide range of relevant studies to calculate the proportion of TBIs or HIs resulting from the primary event of an RTI. A full description of search parameters, as well as inclusion and exclusion criteria, can be found in the Appendix. Briefly, search terms included "traffic accident," "vehicular crash," "vehic* accident," "brain injury," and "head injury." Our search focused on "accident" as opposed to "crash" or "collision" because most epidemiological studies on road injuries have historically followed this notation convention until recently, when studies began to shift to more objective terminology. Thus, our use of "vehicular crash" was an attempt to capture more recent studies using this new convention. Articles were included if epidemiological data categorized the types of injuries sustained from RTIs and if the proportion of RTIs resulting in HI or brain injury was discernible. Studies that included only TBI or HI and those that only reported a certain severity of injury (for example, severe TBI only) were excluded to minimize selection bias. Two reviewers (A.R., S.G.) and a single arbiter (M.C.D.) conducted this search.
Next, we conducted a search of governmental traffic injury registries that reported HIs. The Organisation for Economic Co-operation and Development (OECD) Health Statistics report ("Injuries in Road Traffic Accidents 2016") and citation information from the IHME GBD data on road injury incidence were queried. 116 Registries from 15 different countries were screened: United States, United Kingdom, Canada, Mexico, Brazil, Australia, New Zealand, Taiwan, China, India, South Africa, Belgium, Chile, France, Italy. A single registry (United Kingdom) yielded compatible information on HI and was incorporated into the model with the peer-reviewed study data.
Study results were then pooled using MedCalc software version 15.1 to conduct a meta-analysis. Data were pooled with inverse probability random-effects weighting to estimate the proportion of RTIs resulting in TBI, represented by P(TBI|RTI) for each WB income group and WHO region.
The number of cases of RTI that resulted in TBI, represented by RTI∩TBI, was calculated as the product of RTI TOTAL and P(TBI|RTI) for all WHO regions and WB income groups (Fig. 1 V). was conducted to estimate the relative distribution and proportions of MOI for TBI. The goal was to calculate the proportion of TBI cases that were attributable to RTI, represented by P(RTI|TBI), or the probability that TBI resulted from RTI as a mechanism. This is equal to the ratio of TBI caused by RTI to TBI from all causes in a population ( Fig. 1 VI).
Following the PRISMA guidelines, 173 we searched the PubMed and Cochrane Database of Systematic Reviews to identify studies reporting country-specific epidemiological data on TBI MOI. A full list of search terms and the search methodology can be found in the Appendix, and a detailed breakdown of the article screening process is illustrated in Fig. 3. In summary, MeSH and title/abstract terms were included to maximize the inclusion of studies related to TBI epidemiology (incidence, prevalence, burden, mortality, and so forth) . The methodological quality of individual studies was scored on a 6-point scale from lowest (0 = small sample size, hospital based) to highest (5 = large, ideal population based) to allow quality comparisons among regions and income groups. 238 As described by Feigin et al., 82 less rigorous study quality was permitted for studies from LMICs, from which data would otherwise be unavailable.
Mechanism of injury studies were first selected based on completeness of data (that is, the sum of individual MOI cases equaled the total number of TBI cases reported). Studies were then reviewed for study design and subject selection; studies reporting incidence within a population that could be extended beyond a hospital (that is, at least the regional level) were included for data analysis. Mechanism of injury studies were excluded if they had narrow selection criteria (only pediatric patients, only severe TBI, and so forth).

Incidence of TBIs
The total number of TBI cases from all MOIs (TBI TOTAL ) in a population annually was computed by dividing the number of TBI cases secondary to RTI (RTI∩TBI) by the proportion of TBIs caused by RTIs [P(RTI|TBI); Fig. 1 VII]. The WB population data and TBI TOTAL were then used to calculate the incidence of TBI in a given population [P(TBI); Fig. 1 VIII]. The calculations for confidence intervals are outlined in the Supplement.

Severity of Injury
Finally, we sought to characterize the distribution of mild, moderate, and severe TBI. Studies identified in the systematic review for MOI (Methods, Traumatic Brain Injury MOI) were queried for the reporting of TBI severity. Population-based studies categorizing TBI severity with Glasgow Coma Scale scores of mild (13-15), moderate (9-12), and severe (3)(4)(5)(6)(7)(8) were extracted. Severity distribution values were pooled using random-effects inverse weighting meta-analysis in accordance with the methodology and reasoning outlined above (Methods, Proportion of RTIs Causing TBIs). From these data, the proportion of different TBI severities was calculated and multiplied by the total TBI incidence to arrive at an estimated incidence of mild, moderate, and severe TBI across geographic regions and income brackets.

Incidence of RTIs
A total of 66 countries were represented in the GBD RTI mean incidence rate data, including all 7 WHO regions-African Region (AFR) = 20 countries, Region of the Americas-Latin America (AMR-L) = 6 countries, Region of the Americas-United States and Canada (AMR-US/Can) = 1 country, Eastern Mediterranean Region (EMR) = 8 countries, European Region (EUR) = 18 countries, Southeast Asian Region (SEAR) = 6 countries, Western Pacific Region (WPR) = 7 countries-and all WB income groups (high = 16, middle = 40, low = 10). The annual incidence is displayed as a proportion of the population [P(RTI)] and was highest in the SEAR (1.5%) and lowest in the AMR-US/Can (1.1%; Table 1). Despite differences in the proportion of motor vehicle users across WHO regions, there was relatively minimal variability in the risk of RTI.

Proportion of RTIs Resulting in TBI
A total of 12 large RTI studies reporting data on the proportion of HIs or TBIs were identified. 9,29,36,46,65,103,149,170,203,216,229,249 Five WHO regions were represented: AFR = 5 studies (4 countries), AMR-L = 0 studies, AMR-US/Can = 1 study, EMR = 0 studies, EUR = 2 studies (2 countries), SEAR = 2 studies (1 country), WPR = 2 studies (2 countries). All income groups were also represented (studies: HIC = 3, middle-income country [MIC] = 6, low-income country [LIC] = 3). Methodology, sample size, and cohort characteristics for each study can be found in Supplemental Table S1. The pooled proportion of RTIs and TBIs for each region and income group is listed in Table 1. The greatest P(TBI|RTI) was found in the AFR and SEAR (34%), whereas AMR-US/Can (29%) had the lowest proportion. This equated to 4,404,063 TBI cases related to RTI in AFR and 1,157,181 in AMR-US/Can. Despite having an equivalent or lower P(TBI|RTI) than in the AFR, the SEAR and WPR carry the greatest absolute caseload of TBIs secondary to RTIs, at 10.1 and 8.9 million new cases each year, respectively.
Unauthenticated | Downloaded 11/17/21 10:30 PM UTC istics, patient demographics, and study setting and design were captured in a comprehensive database of the most relevant TBI studies worldwide over the last decade. After further filtering by study population, injury setting, completeness of data, and broad/representative subject selection, we identified 10 studies reporting on 11 unique cohorts (1 study 52 counted twice) that provided representative TBI mechanism distributions (Supplemental Table S3). 18 Table 1. The P(RTI|TBI) was lowest in the AMR-US/Can (25%) and highest in the SEAR and AFR (56% each). A total of 6 studies, comprising 7 distinct cohorts, were incorporated into the severity of injury estimate (4 WHO regions). 33,52,125,218,228,240 Mild TBI accounted for 81.02% of injuries, moderate TBI for 11.04%, and severe TBI for 7.95% (Table 2).

Worldwide Incidence and Volume of TBI
The incidence of TBI was highest in the AMR-US/Can (1299 cases per 100,000 people, 95% CI 650-1947) and EUR (1012 cases per 100,000 people, 95% CI 911-1113) and lowest in the AFR (801 cases per 100,000 people, 95% CI 732-871; Table 1). Extrapolating onto regional populations, the greatest volume of TBI annually was observed in the SEAR (18.3 million) and WPR (17.3 million; Fig. 4).
The global incidence of all-cause, all-severity TBI is estimated at 939 cases per 100,000 people (95% CI 874-1005); thus, an estimated 69.0 million (95% CI 64.2-73.8 million) people worldwide will suffer TBI each year (Table 1). Mild TBI affects approximately 740 cases per 100,000 people, or a total of 55.9 million people each year, whereas 5.48 million people are estimated to suffer severe TBI each year (73 cases per 100,000 people).

Discussion
In this report, we have amassed a comprehensive overview of global TBI, with a focus on quantifying injury incidence and volume. Our model estimates that between 64 and 74 million new cases of TBI will occur worldwide each year. While the incidence of TBI was highest in the AMR-US/Can and EUR, the greatest overall burden of TBI is seen in the SEAR and WPR.
Estimates provided here are generally higher than those in previous efforts to quantify the volume of TBI. In 2010, it was estimated that 1.7 million people in the United States sustain a TBI each year, 45 far fewer than our estimate of 4.6 million in the United States and Canada. However, the former report primarily examined individuals presenting to an emergency department for care and thus probably underestimated the overall population burden of TBI. Indeed, many patients who sustain a mild TBI (sports concussions, falls, low-velocity RTI) probably never seek medical attention. Zhang et al. estimated that up to 1650 adolescents per 100,000 patients suffer a concussion each year-a figure that does not include the many other types and severities of HI. 264 Preliminary results from the 2010 GBD Project 39 suggested that the global incidence rate of TBI was 200 cases per 100,000 people per year, equating to approximately 15 million persons affected. For comparison, an estimated 3.8 million new cases of human immunodeficiency virus, 287 million cases of malaria, and 8.8 million cases of tuberculosis will develop each year. 116 In terms of other common conditions causing neurological morbidity and mortality, recent estimates suggest that each year there are 16.9 million new cases of stroke and 30 million new cases of central nervous system infection. 82 Contextualizing their figure of 15 million new cases, the GBD Project authors acknowledged the uncertainty of their estimate and that it was "a likely underestimate" because they had used hospital-and emergency department-based studies. The updated GBD study (GBD 2015) used here provides more accurate figures for RTI, and our ratio estimates are based on more recent TBI studies. Furthermore, the higher prevalence of mild TBI in our estimate probably explains some of these disparities.
In calculating the incidence figures herein, we relied on RTI data from the IHME since reliable, population-based incidence figures for TBI in the majority of LMICs were unavailable-both in the literature and via national registries. Relative ratios of TBI occurrence, as well as the contribution of RTI to overall TBI, obtained via systematic review and meta-analysis are subject to multiple sources of bias. Any small difference or inaccuracy in the TBI/RTI ratio is compounded when applied to P(RTI) and regional populations. For example, in this model, a low P(RTI|TBI) will boost regional TBI incidence because incidence is calculated from the product of P(RTI) and the inverse of P(RTI|TBI). The incidence of TBI in the AMR-US/Can probably stands as an outlier in part for this very reason. While also relatively high, the EUR incidence (1012 cases per 100,000 people) is somewhat diluted by MICs, in which less robust, hospital-derived data tend to produce lower overall TBI incidence rates because some cases of mild TBI are never reported. The lower TBI incidence in the AFR is probably explained in part by lower-quality road traffic data from member countries, as well as by the overwhelming contribution of RTIs to TBI. In this model, the contribution of all other MOIs (recreation, falls, assault, and so forth) is incorporated indirectly by the inverse proportion of TBIs from RTIs (Fig. 1). Nevertheless, several observations lending credence to our estimates warrant elaboration. First, our meta-regression suggests that P(TBI|RTI) is highest in the AFR (34%) and SEAR (34%) and lowest in the AMR-US/Can (29%). This is an intuitive finding given the abundance of traffic regulations and safety laws in places like the United States relative to many LICs. 52,203 Additionally, RTI refers to injuries sustained not only by motorists, but also by pedestrians and cyclists. A dearth of sidewalks and traffic lights and poor helmet compliance among cyclists and motor-cyclists in LMICs probably translate to a higher rate of HI following RTI. 66,100,220 Moreover, inadequate on-board safety technology or overcapacity vehicles can compound an otherwise trivial collision. The single collision of a cargo truck full of unrestrained occupants in LMICs can result in dozens of cases of TBI-a scenario not frequently observed in most HICs.
Second, we found that P(RTI|TBI) was lowest in the AMR-US/Can (25%) and highest in the SEAR (56%) and AFR (56%). A larger proportion of studies from the SEAR and AFR represented hospital-based analyses relative to studies from regions with a predominance of HICs (Supplemental Table S2). As a result, TBI presenting to hospitals in LMICs likely overrepresented high-severity injuries, which are known to be associated with road traffic collisions more than other MOIs. 5,185,187,252 Furthermore, in many HICs, in which life expectancy exceeds that in LMICs, injury secondary to falls, especially in the elderly, tends to dilute the overall mechanistic proportion of TBI; our results suggest that this phenomenon may exist. 18,112 Mild TBI occurs with far greater frequency than moderate or severe TBI-nearly 10-fold the burden of both moderate and severe injury. When establishing health care priorities in the setting of limited resources, some may dismiss this mild TBI burden as being of nominal consequence. However, the disabling effects of even mild TBI probably translate into economic, societal, and qual- ity of life detriments. Nearly a quarter of patients describe disabling symptoms several months after injury. 195,268 And despite the normalization of neuropsychological and functional scales by 1 year, half of TBI victims report 3 or more persistent posttraumatic symptoms. 68 The volume and extent of our literature review attempts not only to address our stated hypotheses, but also to aid researchers interested in exploring these hypotheses further. The tremendous amount of data found within these studies cannot possibly be summarized in a single paper; instead, highlights and major patterns are described in the text and tables. Readers are encouraged to consult the Supplemental Tables to gain a more granular understanding of the nature of TBI in regions around the world. Supplemental Table S2 is organized by WHO region and country to serve as a quick reference for the interested reader and those seeking an understanding of the data available inand absent from-the literature.

Limitations and Future Directions
The conclusions of this report must be examined in the context of our study design. First, all TBI estimates were modeled after the GBD estimates for RTI. Therefore, assumptions or errors made in the GBD methodology would be carried over into these estimates. Second, by nature of the data available from the literature, we assumed uniform disease susceptibility across age groups and sexes. We also assumed that member countries of a particular WHO region or WB income group share the same injury incidence and proportion. The gold-standard alternative to this limitation is a series of large, population-based sampling studies conducted in every representative population worldwide; the feasibility and cost of such an effort is problematic, though no less important.
Next, the literature reviews and meta-analyses conducted to obtain RTI and TBI relative ratios rely on heterogeneous and often biased study designs. Naturally, a topic as broad as TBI yields results from non-uniform populations, thereby making aggregations and direct comparisons challenging. For example, in the latter systematic review, some studies only examined severe TBI 84,243,257 and some only reported on TBI in young 188,194,197,241,243,245,254,258,265 or elderly 48,49,134,225 cohorts. Combining epidemiological data across disparate cohorts risks misrepresentation of the disease burden and volume. Moreover, the methodological quality from LMICs was, on average, lower than that from HICs; the TBI estimates from HICs may be more reliable than those from LMICs. This limitation is inherent to most global epidemiological surveys, wherein data derive from sources of disparate quality. Lastly, even basic discrepancies, such as differing definitions of TBI or conflicting injury severity scores, encountered across studies may have influenced our results.
Despite the limitations of this model and its underlying methodology, justification for its use resides in the scientific estimation of TBI burden in countries and regions in which data are otherwise entirely unavailable. Our aim of estimating the volume of TBI on a global scale in a transparent and quantitative fashion has been realized, albeit with the aforementioned considerations. Concrete estimates of TBI with region and income-level specificity are provided.
A tremendous burden-approximately 69 million cases-of TBI can be expected each year. The vast majority of this burden affects populations in LMICs, in which adequate health care resources are often either inaccessible or nonexistent. Examining the disease burden between regions and comparing against available resources allow identification of such deficits in existing health care coverage. More robust research, especially in LMICs where high-quality data are deficient, is essential for a targeted campaign. A logical first step in this effort is the establishment of an international TBI registry to improve our understanding of the nature and scope of TBI worldwide. Such a registry should be intuitive, secure, electronic, transferable across heterogeneous institutional informational technologies, and free. While there are multiple such platforms available, we have extensive experience with REDCap 105 (Research Electronic Data Capture, Vanderbilt University) and its successful application in international data collection for clinical neurosurgery in LMICs. Secondarilyand concurrently-a series of targeted, community-based epidemiological surveys of representative populations would allow for the generation of more reliable incidence and mortality figures for TBI in low-resource settings. Ultimately, curbing the silent epidemic of TBI will require a multipronged effort toward public awareness, safety legislation and enforcement, injury prevention campaigns, health care capacity strengthening, and community-based efforts to promote recovery and rehabilitation.

Conclusions
Each year an estimated 69 million individuals will suffer a TBI, the vast majority of which will be mild (81%) and moderate (11%) in severity. Per capita, the highest annual incidence of all-cause TBI is observed in the AMR-US/Can and EUR (1299 and 1012 cases per 100,000 people, respectively). Taking into account regional populations, however, the greatest burden of HIs is in the SEAR (18.3 million) and WPR (17.3 million). The health care systems in LMICs encounter nearly 3 times as many total TBIs than those in HICs. These estimates are limited by relatively low-quality data from LMICs and suggest the need for more robust and accurate injury reporting. The global disparity in health care between regions with fewer resources and a high disease burden and those with greater assets and a lower burden deserves attention and action.