Global, regional, and national burden of neurological disorders, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016

1The UN General Assembly report of December, 2017, underscored that progress in reducing the burden of non-communicable diseases, including neurological disorders, has been insufficient to meet the UN Sustainable Development Goal targets by 2030. Neurological disorders are the leading cause of disability and the second leading cause of death worldwide. Thus, strategies and programmes that reduce the burden from neurological disorders would potentially help to achieve the UN Sustainable Development Goal targets.

2Although age-standardised incidence, mortality, and prevalence rates of many neurological disorders declined for many countries from 1990 to 2015, the absolute number of people affected by, dying, or remaining disabled from neurological disorders over the past 25 years has been increasing globally. Regular assessments of incidence, prevalence, mortality, and disability associated with neurological disorders by cause and over time are important for evidence-based health-care planning, priority setting, and resource allocation. On a global scale, this type of information has facilitated between-country and between-region comparisons of the burden from neurological disorders and their trends. The information can also be used for hypothesis generation on causal relationships and effects of sociodemographic and health care factors on the burden from neurological disorders.

3Evidence before this studyFindings from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2015, showed that neurological disorders are the world's largest cause of disability. However, some disabling neurological disorders, specifically traumatic brain injury and spinal cord injury, were not included, thus underestimating the global burden of this group of disabilities. No other relevant efforts, beyond GBD, have been made to quantify the global burden of neurological disorders as a whole. The data sources identified after the GBD 2015 study are detailed in the other GBD papers in the series of the neurological disorders overvieved in this paper.Added value of this studyThis study improves upon GBD 2015 by including four significant considerations. First, compared with a previous analysis based on GBD 2015, we were able to add the non-fatal outcomes of traumatic brain injury and spinal cord injury. Thus, in this analysis, we have aggregated results from GBD 2016 for 15 disease and injury outcomes that are generally cared for by neurological services, such as CNS infectious conditions (tetanus, meningitis, and encephalitis), stroke, brain and other CNS cancers, traumatic brain injury and spinal cord injury, Alzheimer's disease and other dementias, Parkinson's disease, multiple sclerosis, motor neuron diseases, idiopathic epilepsy, migraine, tension-type headache, and a residual category of other less common neurological disorders. Second, we incorporated new data from sources identified after GBD 2015, including peer-reviewed publications, reports from statistical agencies or ministries of health, surveys, and administrative and hospital data. Third, in the individual papers of the burden of neurological disorders analysed, we extended our analysis of GBD results by Socio-demographic Index (SDI), a summary indicator of income per capita, years of schooling, and total fertility rate, with new ways of presenting and visualising changes over time and the relationship with SDI. Fourth, we extended the terminal age group used in our analysis from 80 years and older to 80–84 years, 85–89 year, 90–94 year, and 95 years and older.Implications of all the available evidenceNeurological disorders continue to be the leading cause of disability worldwide, and their contribution to the overall burden from all health conditions is increasing. The bulk of the burden from neurological disorders continues to be in low-income and middle-income countries. Ageing of the population, population growth, and ongoing epidemiological transition across all countries require regular updates on the global, regional, and national burden from neurological disorders to allow evidence-based health-care planning and resource allocation for these disorders.

4Evidence before this study

5Findings from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2015, showed that neurological disorders are the world's largest cause of disability. However, some disabling neurological disorders, specifically traumatic brain injury and spinal cord injury, were not included, thus underestimating the global burden of this group of disabilities. No other relevant efforts, beyond GBD, have been made to quantify the global burden of neurological disorders as a whole. The data sources identified after the GBD 2015 study are detailed in the other GBD papers in the series of the neurological disorders overvieved in this paper.

6Added value of this study

7This study improves upon GBD 2015 by including four significant considerations. First, compared with a previous analysis based on GBD 2015, we were able to add the non-fatal outcomes of traumatic brain injury and spinal cord injury. Thus, in this analysis, we have aggregated results from GBD 2016 for 15 disease and injury outcomes that are generally cared for by neurological services, such as CNS infectious conditions (tetanus, meningitis, and encephalitis), stroke, brain and other CNS cancers, traumatic brain injury and spinal cord injury, Alzheimer's disease and other dementias, Parkinson's disease, multiple sclerosis, motor neuron diseases, idiopathic epilepsy, migraine, tension-type headache, and a residual category of other less common neurological disorders. Second, we incorporated new data from sources identified after GBD 2015, including peer-reviewed publications, reports from statistical agencies or ministries of health, surveys, and administrative and hospital data. Third, in the individual papers of the burden of neurological disorders analysed, we extended our analysis of GBD results by Socio-demographic Index (SDI), a summary indicator of income per capita, years of schooling, and total fertility rate, with new ways of presenting and visualising changes over time and the relationship with SDI. Fourth, we extended the terminal age group used in our analysis from 80 years and older to 80–84 years, 85–89 year, 90–94 year, and 95 years and older.

8Implications of all the available evidence

9Neurological disorders continue to be the leading cause of disability worldwide, and their contribution to the overall burden from all health conditions is increasing. The bulk of the burden from neurological disorders continues to be in low-income and middle-income countries. Ageing of the population, population growth, and ongoing epidemiological transition across all countries require regular updates on the global, regional, and national burden from neurological disorders to allow evidence-based health-care planning and resource allocation for these disorders.

10In 2017, we published aggregated estimates of the burden from neurological disorders (as measured by prevalence, mortality, disability-adjusted life-years [DALYs], and years lived with disability [YLDs]), based on the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2015 data. However, neurological disorders analysed did not include traumatic brain injury (TBI) or spinal cord injury, which are significant sources of global disability. Additionally, newly available publications and other data sources were added to the GBD 2016 estimates.

11In the 2018–19 series of publications in The Lancet Neurology,,,,,,,,,, we presented global, regional, and national estimates of the burden of individual neurological disorders as measured by prevalence, mortality, DALYs, YLDs, years of life lost (YLLs), and, for selected disorders, also incidence, and their trends from 1990 to 2016 according to Socio-demographic Index (SDI), a summary indicator of income per capita, years of schooling, and total fertility rate. Here, we present updated aggregated estimates (compared with aggregated estimates presented in the GBD 2015 study) of neurological disorders and their trends from 1990 to 2016 by SDI, age, and sex.

12The Institute for Health Metrics and Evaluation produces annual updates of the GBD study and includes a growing collaboration of scientists. Reported estimates span the period from 1990 to 2016. Annual updates allow incorporation of new data and methodological improvements to ensure that the most up-to-date information is available to policy makers to help make resource allocation decisions. In this analysis, we have aggregated results from GBD 2016 for 15 disease and injury outcomes that are generally cared for by neurological services. These include infectious conditions (tetanus, meningitis, and encephalitis), stroke, brain and other CNS cancers, TBI, spinal cord injury, Alzheimer's disease and other dementias, Parkinson's disease, multiple sclerosis, motor neuron diseases, idiopathic epilepsy, migraine, tension-type headache, and a residual category of other less common neurological disorders. Compared to a previous analysis based on GBD 2015, we added non-fatal outcomes of TBI and spinal cord injury. Medication overuse headache is no longer included as a separate cause but quantified as a consequence of the underlying headache types. For all neurological disorders combined we report here estimates of deaths and DALYs because aggregate incidence and prevalence estimates of combined neurological disorders are not useful for policy making.

13In the methods section of this overview paper, we present a summary of the general methods of the GBD as they apply to neurological disorders. In the ten accompanying disease-specific papers, we have concentrated on methods that are specific to each disorder. Details on the methods of estimates for tetanus, encephalitis, and the residual category of other neurological disorders are provided in the appendix. The guiding principle of GBD is to assess health loss due to mortality and disability comprehensively, defining disability as any departure from full health. In GBD 2016, estimates were made for 195 countries and territories and 579 subnational locations, for 27 years starting from 1990, for 23 age groups, and for both sexes. Deaths were estimated for 264 diseases and injuries, whereas prevalence and incidence were estimated for 328 diseases and injuries. To allow meaningful comparisons between deaths and non-fatal disease outcomes as well as between diseases, data for deaths from and prevalence of neurological disorders are summarised in a single indicator, the DALY. DALYs are the sum of YLLs and YLDs. YLLs are estimated as the product of counts of deaths and a standard ideal remaining life expectancy at the age of death. The standard life expectancy is derived from the lowest observed mortality rates by age in any population in the world larger than 5 million people. YLDs are estimated as the product of prevalence of individual consequences of disease (or sequelae) multiplied by their corresponding disability weights, which quantify the relative severity of sequelae as a number between 0 (representing full health) and 1 (representing death). Disability weights have been estimated in nine population surveys and an open-access internet survey in which respondents were asked to choose the healthier option between random pairs of health states that were presented with short descriptions of their main features.

14All-cause mortality rates are estimated from vital registration data in countries with complete coverage as analysed by demographic growth balance methods comparing successive census population counts. For other countries, the probabilities of death before age 5 years and between ages 15 and 60 years are estimated from censuses and surveys asking mothers to provide a history of children ever born and those still alive, and surveys asking adults about siblings who are alive or have died. Using model life tables, these probabilities of death are transformed into age-specific death rates by location, year, and sex. GBD has collated a large database of cause-of-death data from vital registrations and verbal autopsy surveys in which relatives are asked a standard set of questions to ascertain the likely cause of death, supplemented with police and mortuary data for injury deaths in countries with no other data. Cause-of-death information is provided in a large number of different classification systems based on versions of the International Classification of Diseases (ICD) or bespoke classifications in some countries. All data are mapped into the disease and injury categories of GBD. All classification systems contain codes that are less informative than ICD because they do not have a specific diagnosis (eg, unspecified cancer) or refer to codes that cannot be the underlying cause of death (eg, low back pain or senility) or are intermediate causes (eg, heart failure or sepsis). Such deaths are redistributed to more precise underlying causes of death. After these redistributions and corrections for under-registration, the data are analysed with the Cause of Death Ensemble model (CODEm), a highly systematised tool that runs many different models on the same data and chooses an ensemble of models that best reflects all the available input data. The statistical performance of all models is tested by withholding 30% of the data and checking how well a model covers the data that were held out. To enforce consistency from CODEm, the sum of all-cause specific mortality rates is scaled to that of the all-cause mortality rates in each age, sex, location, and year category. All our estimates of causes of death are categorical: each death is assigned to a single underlying cause. This means that all estimates add to 100%.

15Non-fatal estimates are based on systematic reviews of published papers and unpublished documents, survey microdata, administrative records of health encounters, registries, and disease surveillance systems. These data sources are catalogued in our Global Health Data Exchange (GHDx), the largest repository of health data globally. We first set a reference case definition or study method, or both, that best quantifies each disease or injury or consequence thereof. If there is evidence of a systematic bias in data that used different case definitions or methods compared to reference data, we adjust those datapoints to reflect what their value would have been if measured as the reference. DisMod-MR 2.1, a Bayesian meta-regression tool, is our main method of analysing non-fatal data. It is designed as a geographical cascade where a first model is run on data from all countries, which produces an initial global fit and estimates coefficients for predictor variables and the adjustments for alternative study characteristics. The global fit adjusted by the values of random effects for each of seven GBD super-regions (appendix), the coefficients on sex, and country predictors, is passed down as data to a model for each super-region together with the input data for that geography. The same steps are repeated going from super-regions to 21 regions and then to 195 countries and, where applicable, a further level down to subnational units. Below the global level, all models are run separately by sex and for six time periods (1990, 1995, 2000, 2005, 2010, and 2016). During each analysis, all data for prevalence, incidence, remission (ie, cure rate), and mortality are forced to be internally consistent. For most diseases, the bulk of data for prevalence or incidence are at the disease level, with even fewer studies providing data on the proportions of cases of disease in each of the sequelae defined for the disease. The proportions in each sequela are pooled using DisMod-MR 2.1 or meta-analysis, or are derived from analyses of patient-level datasets. The multiplication of prevalent cases for each disease sequela and the appropriate disability weight produces YLD estimates that do not yet take into account comorbidity. To correct for comorbidity, these data are used in a simulation to create hypothetical individuals in each age, sex, location, and year combination who have none, one, or multiple sequelae simultaneously. We assume that disability weights are multiplicative rather than additive because this avoids assigning a combined disability weight value in any individual exceeding 1—ie, worse off than a year lost due to death.

16For risks, we use a different, counterfactual approach—ie, answering the question “what would the burden have been if the population had been exposed to a theoretical minimum level of exposure to a risk?” Thus, we need to define what level of exposure to a risk factor leads to the lowest amount of disease. We then analyse data on the prevalence of exposure to a risk and derive relative risks for any risk–outcome pair for which we find sufficient evidence of a causal relationship. Prevalence of exposure is estimated in DisMod-MR 2.1 using spatiotemporal Gaussian process regression, or from satellite imagery in the case of ambient air pollution. Relative risk data are pooled using meta-analysis of cohort, case-control, and intervention studies. From the prevalence and relative risk results, population attributable fractions are estimated relative to the theoretical minimum risk exposure level (TMREL). Criteria for inclusion of risks into GBD were the availability of sufficient evidence for a causal relationship between a risk and one or more disease or injury outcomes; evidence to support generalisability of an effect size beyond the populations included in epidemiological studies; availability of sufficient data and methods to enable estimation of exposure levels by country; and the likely importance of a risk factor to disease burden or policy considerations.

17Uncertainty is propagated throughout all these calculations by creating 1000 values for each prevalence, death, YLL, YLD, or DALY estimate and performing aggregations across causes and locations at the level of each of the 1000 values for all intermediate steps in the calculation. The lower and upper bounds of the 95% UI are the 25th and 975th values of the ordered 1000 values. Significance of differences was established if 975 or more of the ordered 1000 values of difference were on either side of zero.

18GBD uses a composite indicator of sociodemographic development, SDI, which reflects the geometric mean of normalised values of a location's income per capita, the average years of schooling in the population aged 15 years and older, and the total fertility rate. Countries and territories are grouped into quintiles of high, high-middle, middle, low-middle, and low SDI based on their 2016 values.

19The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication.

20In 2016, the neurological disorders included in this analysis were responsible for 276 million (95% UI 247–308) DALYs, comprising 11·6% (10·7–12·4) of global DALYs for all diseases. Combined, these disorders were the underlying cause in 9·0 million (8·8–9·4) deaths or 16·5% (16·1–17·0) of total global deaths in 2016. Incidence and prevalence estimates for each neurological disorder can be found in the GBD online results tool. Results and findings mentioned can also be viewed interactively through an online data visualisation tool. Neurological disorders were the leading cause group of global DALYs in 2016, followed by cardiovascular diseases (excluding stroke). In terms of deaths, these combined neurological disorders ranked second after cardiovascular diseases. Since 1990, the number of deaths from neurological disorders increased by 39% (34–44) and DALYs for neurological disorders increased by 15% (9–21). However, age-standardised mortality rates decreased by 28% (26–30) over the same time, indicating that population increase and ageing are driving global numbers higher, even though the global population is exposed to a lower risk of death from these causes, as indicated by the lower age-standardised rates. Similarly, the age-standardised DALY rates decreased by 27% (24–31) between 1990 and 2016. The only neurological disorders that had a decrease in rates and absolute numbers of deaths and DALYs were tetanus, meningitis, and encephalitis.

21Stroke was the largest contributor to global neurological DALYs, responsible for 42·2% (95% UI 38·6–46·1) of these DALYs in 2016 (table 1). Migraine was the second largest contributor (16·3% [11·7–20·8]), followed by Alzheimer's and other dementias (10·4% [9·0–12·1]) and meningitis (7·9% [6·6–10·4]). Stroke ranked first among neurological disorders in terms of age-standardised DALY rates in 19 of 21 world regions (figure 1). In Australasia and western Europe, migraine ranked first. Migraine and Alzheimer's disease and other dementias were ranked among the top four contributing neurological conditions in all 21 GBD world regions. Large variations in ranking were found for meningitis (ranked second in central, eastern, and western sub-Saharan Africa and 14th in high-income Asia Pacific), epilepsy (ranked second in southern sub-Saharan Africa and third in central Asia and eighth in eastern Europe), and encephalitis (ranked fifth in south Asia and 14th in Australasia, western Europe, and high-income North America). Other high rankings of note are for spinal cord injury (ranked fourth in all five high-income regions), TBI (ranked fourth in central and eastern Europe), and tetanus (ranked sixth in eastern sub-Saharan Africa; figure 1).

22Across the aggregate of 15 neurological disorder categories analysed, age-standardised DALY rates were significantly higher in males than in females (male-to-female ratio 1·12 [95% UI 1·05–1·20]). When examining individual disorders, a significantly higher burden (as measured by age-standardised DALY rates) in males than females was observed for TBI, Parkinson's disease, tetanus, motor neuron diseases, and stroke, with male-to-female ratios of at least 1·5. By contrast, migraine, multiple sclerosis, and tension-type headache had male-to-female ratios of less than 0·7 (table 1). In children younger than 5 years, infectious neurological conditions (ie, tetanus, meningitis, and encephalitis), particularly meningitis, were the main causes of neurological DALYs in both sexes (figure 2). DALYs from epilepsy were highest in people aged 5 years to 29 years. Migraine and tension-type headache were large contributors in young and middle-aged adults, with much higher numbers in females than males. Stroke burden rapidly increased up to the age of about 80 years, and was the dominant cause of neurological burden between ages 60 and 84 years, more so in males than females. Although Alzheimer's disease and other dementias were the dominant cause of neurological burden from age 90 years, the global number of DALYs from dementias was greatest between ages 80 and 89 years (figure 2).

23The patterns by age of YLLs and YLDs due to neurological conditions were very different. Deaths, and by extension YLLs, were the dominant feature of infectious causes, stroke, Alzheimer's disease and other dementias, and brain and other CNS cancer, whereas the burden estimated for headaches, TBI, and spinal cord injury were all YLDs, because headaches are not considered an underlying cause of death and the ICD attributes injury deaths to the cause and not the nature of injury. Epilepsy, stroke, and Alzheimer's disease and other dementias were other substantial contributors to YLDs of neurological disorders (figure 3).

24Five countries had age-standardised DALY rates of combined neurological disorders exceeding 7000 DALYs per 100 000 population in 2016: Afghanistan (9135 DALYs [95% UI 7788–10 709] per 100 000), Kiribati (7477 DALYs [6596–8483] per 100 000), Solomon Islands (7104 DALYs [6075–8391] per 100 000), Central African Republic (7283 DALYs [5974–8644] per 100 000), and Somalia (7349 DALYs [5980–9416] per 100 000; table 2). All of these countries have scarce general health data, and no data for any of the neurological conditions included in this analysis. Thus, estimates depended heavily on predictive covariate values and geographical proximity and, hence, are rather uncertain. 11 countries had age-standardised DALY rates less than 2500 DALYs per 100 000 population: Singapore, Peru, Ecuador, Costa Rica, Puerto Rico, Taiwan, Japan, Australia, Switzerland, Canada, and Bermuda (table 2). The reasons for the low rates in these countries were varied. For example, Taiwan and Japan have low rates of migraine, whereas Costa Rica's stroke rates were low. High stroke rates were common among the five countries with the highest neurological DALY rates, whereas, meningitis rates were also high in Somalia, Central African Republic, and Afghanistan and tetanus rates were much higher in Somalia than in any of the other four countries. The age-standardised rates of neurological DALYs decreased significantly in 171 of 195 countries and territories between 1990 and 2016. In 23 countries, the change did not reach statistical significance, and only in North Korea was there a significant increase, which was 20% (95% UI 7–34).

25Only three of the 15 neurological disorder categories had more than 10% of DALYs attributable to the 84 risks quantified in GBD 2016: risk-attributable DALYs accounted for 88·8% (86·5–90·9) of all stroke DALYs, 22·3% (11·8–35·1) of DALYs for Alzheimer's disease and other dementias, and 14·1% (10·8–17·5) of idiopathic epilepsy DALYs. For the other neurological disorders, the proportion of DALYs that were risk attributable was either very small (meningitis, encephalitis, and multiple sclerosis) or zero (tetanus, brain and other CNS cancer, migraine and tension-type headache, Parkinson's disease, and motor neuron diseases).

26Contributing 11·6% of global DALYs and 16·5% of deaths from all causes, neurological disorders remain the leading group cause of DALYs and the second leading group cause of deaths in the world. These estimates are close to those reported in GBD 2015 estimates on the global burden of neurological disorders. Furthermore, our findings reported in the individual neurological disorder papers of the significant geographic and sex variations and the effects of SDI on the patterns of the burden of neurological disorders were consistent with our previous observations. Although the age-standardised rates of deaths and DALYs from 1990 to 2016 have decreased significantly, the absolute numbers of deaths and DALYs have significantly increased over the same period. The only neurological disorders that had significant decreases in age-standardised incidence, prevalence, mortality, and DALY rates were infectious neurological disorders (meningitis, encephalitis, and tetanus). These diverging trends in communicable and non-communicable neurological disorders are in line with our previous estimates and consistent with the observed overall global burden shift from communicable to non-communicable disorders. The decline of age-standardised rates with concomitant increase in absolute numbers was observed in most countries. The increase in absolute numbers of non-communicable neurological disorders juxtaposed with a significant decrease in their age-standardised rates is consistent with the major role of population ageing and growth as the main drivers of this increase; it also suggests that intervention strategies to reduce non-communicable neurological disorders are not sufficiently deployed or effective. With continuing ageing of populations, the prevalence of non-communicable neurological disorders is likely to continue increasing, placing pressure on already overstretched health-care services.,, Therefore, development of new strategies to treat or prevent the major neurological disorders and the implementation of already proven effective prevention strategies for stroke and infectious neurological disorders are of paramount importance to achieve UN Sustainable Development Goal targets.

27The relationship between SDI and age-standardised DALY rates has been discussed in each of the neurological disorder-specific GBD 2016 papers.,,,,,,,,, For stroke, epilepsy, and meningitis we showed that age-standardised DALY rates by GBD regions over the period 1990–2016 follow a downward trajectory aligned with the predicted direction of decreasing burden of these conditions with socioeconomic development. Four other categories, brain and other CNS cancer, Parkinson's disease, multiple sclerosis, and motor neuron diseases, appear to follow a pattern of increasing DALY rates with increasing SDI. However, for all four of these conditions there were large variations in rates from the expected pattern according to SDI, and high-income Asia Pacific has much lower rates for all four conditions than other high-income regions. This suggests that the observed relationship with SDI is spurious, with SDI being a proxy for other factors (eg, genetic) explaining variations in disease rates. For Alzheimer's disease and other dementias and migraine and tension-type headache, the variation in age-standardised DALY rates and SDI between world regions and within regions over time suggests that socioeconomic development does not determine the extent of DALYs from these diseases.

28Our study findings have important health service implications. The increase in number of people affected by non-communicable neurological disorders implies a need for substantially increased resources for their management. Our detailed estimates for 195 countries and a growing number of subnational locations provides a basis for locality-specific priority setting and financing of health services, including workforce development. For example, by knowing the absolute numbers of new and existing cases of major disabling neurological disorders from the GBD study, one can estimate annual requirements for first specialist neurologist assessments (total number of incident cases of these disorders) and for follow-up neurological assessments (assuming an average number [eg, two] of follow-up visits per year), and from these, the number of neurologists required in a country for outpatient management. For example, in New Zealand in 2016, there were 23 132 incident and 152 210 prevalent cases of stroke, dementias, refractory migraines (perhaps 5% of all migraine cases), epilepsy, Parkinson's disease, multiple sclerosis, motor neuron diseases, and brain and other CNS cancers combined. Thus, the total time for required first specialist assessments (23 132 × 0·75 h for a new clinic patient as reported in New Zealand is 17 349 h) and neurology follow-ups (152 210 [number of prevalent cases] × two [average number of follow-ups per year] × 0·5 h for a follow-up is 152 210 h) was 169 559 h. By dividing 169 559 h by the average estimate of physicians' working hours per year (40 h per week × 48 weeks is 1920 h), one can estimate that approximately 88 full-time equivalent (FTE) neurologists are required (compared with the 36 that are available) to provide specialist neurological care for the most common disabling neurological disorders in New Zealand. This result is close to previous estimates of FTE neurologist requirements in New Zealand using a pragmatic analysis.

29This study was not free of the general limitations shared by all GBD estimates. First, to bring as much information as possible into the estimation of disease and death rates for GBD, an extensive effort is dedicated to making data sources comparable—eg, by mapping cause-of-death data across various iterations of ICD classification systems and the local idiosyncrasies in how these are applied; the redistribution of less informative codes for an underlying cause of death to specified causes in the GBD list; and adjustments of non-fatal and risk factor data to account for biases introduced by different case definitions or study methods. However, we cannot guarantee that we are aware of and can accurately correct for all sources of measurement bias, particularly for diseases like dementia, multiple sclerosis, motor neuron diseases, and Parkinson's disease, which require more complex survey procedures to establish a diagnosis. Additionally, diagnosis of headaches is entirely based on patient report and thus, prone to various sources of reporting bias. Second, despite extensive efforts to identify data sources for GBD, many countries have sparse data and missing information. In those countries, we rely heavily on predictive covariates and assume similarity through geographical proximity. However, many of the neurological disorders do not have strong predictors. The exception is stroke, for which the combination of all risks included in GBD predicts 88·8% of the total burden. The absence of strong predictors in the models of the other disorders makes it much harder to predict disease levels in countries with sparse or no data. It also makes it more essential to conduct good-quality epidemiological studies in those countries that are currently poorly represented in the disease models. Third, we were able to add spinal cord injury and TBI estimates to this analysis but only in terms of incidence and YLDs. The dominant tradition in cause-of-death attribution is to assign injury deaths to the cause of the injury rather than the consequent nature of injury. In future iterations of GBD we plan to analyse data sources with multiple causes listed for each death to approximate deaths from nature of injury categories like TBI and spinal cord injury. Fourth, in GBD, the correction for comorbidity is made on the basis of the independent probability that an individual will have combinations of disabling sequelae. Although we would like also to take into account the more commonly known comorbidities, computationally this has remained intractable for the large number of diseases and their sequelae estimated in GBD. For the neurological disorders that are highly prevalent in older people with known comorbidities such as stroke and diabetes, we are overestimating the non-fatal burden. Finally, the residual category of other neurological disorders tries to capture deaths and DALYs for neurological disorders that are not explicitly being estimated. For deaths, where vital registration systems provide data on all causes, this can be estimated with greater precision. For the non-fatal estimates, we approximate YLDs by assuming the same ratio of YLDs and YLLs estimated for the main fatal neurological disorders (Alzheimer's disease and other dementias, Parkinson's disease, multiple sclerosis, motor neuron diseases, and idiopathic epilepsy) and apply this ratio to the YLLs calculated for residual causes like Huntington's disease and muscular dystrophy. Although as a placeholder estimate this method has at least some validity for residual neurological disorders that lead to death, it does not capture the burden of any residual neurological disorders that are not a cause of death, such as peripheral neuropathies, neuropathic pain, or restless legs syndrome. Thus, our estimates of the burden of neurological disorders, although already large and increasing, are still an underestimate of the true burden of neurological disorders and will be until resources are found to explicitly estimate the burden of these residual disorders.

30In summary, the burden of neurological disorders (especially non-communicable disorders) is large and increasing, posing a challenge to the sustainability of health systems. Although a wealth of knowledge is available on how best to reduce the burden of stroke and infectious neurological diseases, intervention and reduction of the burden of other non-communicable neurological disorders is much more complex and requires the development of new evidence to design more effective treatments and prevention measures. Further epidemiological studies on neurological disorders in various populations are required to fill the gaps in the knowledge of distribution, frequency, outcomes, and determinants of major neurological disorders. We call on a wide range of clinicians, surgeons, and other health-care professionals with neurological expertise caring for individuals with neurological disorders to work collaboratively with the GBD to strengthen the accuracy of future GBD estimates by enhancing epidemiological research of neurological disorders.