Global, regional, and national burden of Parkinson's disease, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016

1Neurological disorders are now the leading source of disability globally. Among neurological disorders examined in the Global Burden of Disease, Injuries, and Risk Factors Study (GBD) 2015, Parkinson's disease was the fastest growing in prevalence, disability, and deaths. In that study, the overall number of people affected by the disease was estimated to have more than doubled globally from 1990 to 2015. Previous studies have examined the epidemiology of Parkinson's disease for different parts of the world, including systematic reviews on the prevalence of Parkinson's disease., However, none has examined change in prevalence, disability, and deaths in detail over the past generation for the entire world and across all countries. In GBD 2015, we identified larger variation in Parkinson's disease death rate estimates over time and between countries than we observed in prevalence estimates. This pattern suggested that coding practices rather than real changes over time and location were responsible, similar to what was observed for dementia.

2The prevalence of a disease reflects both the incidence and the duration of disease. The incidence of Parkinson's disease is linked to risk and protective factors.,, The most important risk factor is age, but the risk of Parkinson's disease also appears to be associated with industrial chemicals and pollutants, such as pesticides, solvents, and metals., Conversely, smoking is associated with a decreased risk of Parkinson's disease, but whether this association is causal is debatable. The factors that affect disease duration are less well known, but increasing longevity also translates into longer disease duration., Therefore, as ageing and industrialisation increase globally and smoking decreases in some regions, the prevalence of Parkinson's disease seems poised to increase., Detailed estimates of the disease burden can help to evaluate the effect of these risk factors and inform efforts to prevent the disease and to care for and treat those affected by the condition.

3Evidence before this studyThe Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2015 examined the epidemiology of Parkinson's disease for different parts of the world and showed that the number of people affected by the condition had more than doubled globally from 1990 to 2015. The increase in deaths from Parkinson's disease was greater than the increase in prevalence, and the large variation in death rates between countries was suggestive of a change in coding practices rather than greater death rates among Parkinson's disease cases. For pragmatic reasons, systematic reviews for Parkinson's disease are scheduled every other iteration of GBD. For GBD 2016, we updated our GBD 2013 PubMed search without language restrictions using the terms (((“Parkinson disease” AND “epidemiology”) AND (“2011/01/01”[PDat]: “2015/12/31”[PDat])) AND (“Parkinson disease” AND “epidemiology”)) to identify articles published between Jan, 1, 2011, and Dec, 31, 2015. Papers were selected if representative of the general population and identification of cases was based on our reference case definition (the presence of at least two of four primary symptoms: rest tremor, bradykinesia, stiffness of limbs and torso, and postural instability) or alternative case definitions (UK Parkinson's Disease Society Brain Bank criteria, and doctor's diagnosis based on International Classification of Diseases codes and prescription of medications specifically for Parkinson's disease).Added value of this studyWe used the results of this search to obtain the data needed to estimate global, regional, and country-specific prevalence and years lived with disability for Parkinson's disease from 1990 to 2016. To address the possible measurement error in Parkinson's disease death rates as reported by vital registration systems, we used a method that was previously applied to dementia in GBD 2015. In a natural history modelling approach, we assume a constant risk of death in Parkinson's disease cases over time and between locations and let the death rates be determined by variations in prevalence. Although the assumption of similar mortality risk in all time periods and countries is problematic, it produces less error than the large variation in death rates estimated previously. We also explored variation in the burden by age, sex, country, region, and Socio-demographic Index. This study showed that counts of prevalence, mortality, and disability-adjusted life-years more than doubled from 1990 to 2016, and that this increase was not solely due to increasing numbers of older people because age-standardised rates also increased in most regions. In addition, the burden of Parkinson's disease increased with increasing Socio-demographic Index.Implications of all the available evidenceNeurological disorders are now the leading source of disability in the world, and Parkinson's disease is the fastest growing of these disorders. As the population ages and life expectancy increases, the number of individuals with Parkinson's disease will continue to increase as well as the duration of the disease, leading to more patients with advanced Parkinson's disease. To address this burden, primary prevention strategies based on the underlying causes of Parkinson's disease and more effective treatments than are currently available are required. Additional incidence and prevalence studies are needed, especially in areas in which little data are available, to examine time trends and the factors that drive them.

4Evidence before this study

5The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2015 examined the epidemiology of Parkinson's disease for different parts of the world and showed that the number of people affected by the condition had more than doubled globally from 1990 to 2015. The increase in deaths from Parkinson's disease was greater than the increase in prevalence, and the large variation in death rates between countries was suggestive of a change in coding practices rather than greater death rates among Parkinson's disease cases. For pragmatic reasons, systematic reviews for Parkinson's disease are scheduled every other iteration of GBD. For GBD 2016, we updated our GBD 2013 PubMed search without language restrictions using the terms (((“Parkinson disease” AND “epidemiology”) AND (“2011/01/01”[PDat]: “2015/12/31”[PDat])) AND (“Parkinson disease” AND “epidemiology”)) to identify articles published between Jan, 1, 2011, and Dec, 31, 2015. Papers were selected if representative of the general population and identification of cases was based on our reference case definition (the presence of at least two of four primary symptoms: rest tremor, bradykinesia, stiffness of limbs and torso, and postural instability) or alternative case definitions (UK Parkinson's Disease Society Brain Bank criteria, and doctor's diagnosis based on International Classification of Diseases codes and prescription of medications specifically for Parkinson's disease).

6Added value of this study

7We used the results of this search to obtain the data needed to estimate global, regional, and country-specific prevalence and years lived with disability for Parkinson's disease from 1990 to 2016. To address the possible measurement error in Parkinson's disease death rates as reported by vital registration systems, we used a method that was previously applied to dementia in GBD 2015. In a natural history modelling approach, we assume a constant risk of death in Parkinson's disease cases over time and between locations and let the death rates be determined by variations in prevalence. Although the assumption of similar mortality risk in all time periods and countries is problematic, it produces less error than the large variation in death rates estimated previously. We also explored variation in the burden by age, sex, country, region, and Socio-demographic Index. This study showed that counts of prevalence, mortality, and disability-adjusted life-years more than doubled from 1990 to 2016, and that this increase was not solely due to increasing numbers of older people because age-standardised rates also increased in most regions. In addition, the burden of Parkinson's disease increased with increasing Socio-demographic Index.

8Implications of all the available evidence

9Neurological disorders are now the leading source of disability in the world, and Parkinson's disease is the fastest growing of these disorders. As the population ages and life expectancy increases, the number of individuals with Parkinson's disease will continue to increase as well as the duration of the disease, leading to more patients with advanced Parkinson's disease. To address this burden, primary prevention strategies based on the underlying causes of Parkinson's disease and more effective treatments than are currently available are required. Additional incidence and prevalence studies are needed, especially in areas in which little data are available, to examine time trends and the factors that drive them.

10As part of GBD 2016, we aimed to examine the changes from 1990 to 2016 in counts and age-standardised rates of Parkinson's disease for prevalence, disability, and deaths by location and by the Socio-demographic Index (SDI), a composite measure of income per capita, education, and fertility.

11The general methods for the studies on the global, regional, and national burden of neurological disorders have been published previously, and key aspects are summarised in the appendix. Additional information on derivation of non-fatal and fatal estimates are provided in the appendix as well as on Global Health Data Exchange.

12The International Classification of Diseases ninth revision (ICD-9) codes used in cause of death analyses for Parkinson's disease are 332 (Parkinson's disease), 332.0 (paralysis agitans), and 332.1 (secondary parkinsonism), and the corresponding ICD-10 codes are G20 (Parkinson's disease), G21 (secondary parkinsonism), and G22 (parkinsonism in diseases classified elsewhere). The reference case definition for Parkinson's disease used in many epidemiological studies is the presence of at least two of the four primary symptoms: rest tremor, bradykinesia, stiffness of the limbs and torso, and postural instability. We also accepted alternative definitions, including the UK Parkinson's Disease Society Brain Bank criteria, in addition to ICD-9 and ICD-10 codes, a doctor's diagnosis of Parkinson's disease, and the prescription of Parkinson's disease-specific medications. We also added 3 years of medical claims data (years 2000, 2010, and 2012) from the USA; for a disease such as Parkinson's disease, the data from claims sources would be expected to match true prevalence under the expectation that most patients would be receiving medical attention each year. If datapoints from epidemiological studies spanned ages of more than 20 years, we split the datapoints using the age pattern from the USA, for which the most detailed age data were available.

13For Parkinson's disease, we have seen large inconsistencies between the cause of death data and the non-fatal data. For example, US vital registration data show a greater than three times increase in the age-standardised rates of death from Parkinson's disease since 1980 without a corresponding increase in the prevalence data over the same time period (appendix). Likewise, we found a greater than 25 times difference across different countries in age-standardised mortality rates for the most recent year of vital registration data available at the time of GBD 2016 (see Causes of Death Visualization), and we did not see such heterogeneity between countries in our non-fatal data. Therefore, these differences are probably the result of changes and inconsistencies in coding practices for certifying deaths from Parkinson's disease. To correct for this bias, we jointly modelled the prevalence and mortality from Parkinson's disease. First, we ran an initial cause of death model using CODEm, the cause of death ensemble model used throughout GBD, and an initial non-fatal model using DisMod-MR 2.1, the Bayesian meta-regression tool developed for GBD. The initial CODEm model used 14 990 site-years of data (ie, a unique combination of calendar year and country) as well as predictive covariates of SDI, cumulative cigarette consumption, health-care access and quality, education, and mean cholesterol level (a full list of predictive covariates is in the appendix). The initial DisMod-MR 2.1 model included settings of no remission (ie, no cure) and no incidence before the age of 20 years because the disease is exceptional before that age. We excluded all incidence data from the model, since we saw inconsistencies between the available prevalence and incidence data, and we considered measurement error less likely to occur with prevalence data than with incidence data. We let DisMod-MR 2.1 adjust medical claims data to correct for any systematic under-reporting and datapoints with case definitions that differed from the reference. Smoking prevalence and SDI were used as predictive covariates in the model.

14We used these initial model results to identify countries with high-quality vital registration systems, age-standardised prevalence of more than five per 10 000, and a population of more than 1 million that also had the highest ratios of cause-specific mortality to prevalence, or highest likelihood to code to Parkinson's disease as a cause of death per prevalent case in the most recent year of estimates. For GBD 2016, these countries were Austria, Finland, and the USA. We then used the log-transformed ratios of cause-specific mortality rate to prevalence in 2016 to run a fixed-effects regression with dummy variables on age and sex. Because the ratio between cause-specific mortality rate and prevalence is equivalent to an excess mortality rate or excess rate of dying among people with Parkinson's disease compared with the general population, we used the results of this regression as input data for a second DisMod-MR 2.1 model. The excess mortality data obtained from the regression model were used as data for the entire 1990–2016 period and for every country except for the three used in the regression model, which retained their own data for 2016, and data for these countries were assumed to be constant over the entire time series. Apart from this addition of excess mortality data, the second DisMod-MR 2.1 model was identical to the initial model and used the same settings and covariates. We used the cause-specific mortality and prevalence results from this model as final outputs because they ensured consistency between the available non-fatal input data and the excess mortality rate in 2016 from the three countries most likely to code to Parkinson's disease as a cause of death.

15To calculate years lived with disability (YLDs) for Parkinson's disease, we split the overall prevalence from the second DisMod-MR 2.1 model into three severity categories using data reporting on the Hoehn and Yahr stages., We used 30 unique sources, covering nine of 21 world regions, and equated a score of 2·0 or less on the Hoehn and Yahr scale to mild Parkinson's disease, a score of 2·5–3·0 to moderate Parkinson's disease, and a score of 4·0–5·0 to severe Parkinson's disease (appendix). These data informed meta-analyses of the proportion of Parkinson's disease that is mild, moderate, and severe. We then used these proportions to split the overall prevalence of Parkinson's disease into the severity categories. Finally, we multiplied the prevalence of each severity category by severity-specific disability weights (see appendix for a detailed description) to calculate YLDs. YLDs were then corrected for comorbidity with a simulation that assigned all non-fatal outcomes to hypothetical individuals and adjusted disability in patients who had multiple conditions.

16Of the 84 risks quantified in GBD 2016, only smoking was judged to have sufficient evidence for a relationship with Parkinson's disease, with smoking associated with decreased risk. The main sources of exposure data were population-based surveys. We used the estimates of exposure, relative risk, and a theoretical minimum level of exposure of zero lifetime cigarettes smoked to calculate population attributable fractions. Further information on risk factor calculations can be found in the GBD 2016 risk factors paper.

17We calculated years of life lost (YLLs) by multiplying the number of deaths in an age group by the remaining life expectancy in that age group, taken from the GBD standard life table. Disability-adjusted life-years (DALYs) were then calculated as the sum of YLLs and YLDs.

18Through each computational step, uncertainty was propagated by sampling 1000 draws, which allowed us to combine the uncertainty from input data, corrections to the data, and residual non-sampling error. Uncertainty intervals (UIs) were defined as the 25th and 975th values of the ordered draws.

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

20The results of our analyses can be downloaded from the Global Health Data Exchange and Institute for Health Metrics and Evaluation (Seattle, WA, USA) results tools. Through the systematic analysis we identified 127 data sources on Parkinson's disease, including 91 sources on prevalence covering 16 of the 21 GBD world regions, 34 sources on incidence covering nine world regions, and 11 sources on mortality risk covering two world regions (appendix). The 11 sources on mortality risk are used in the non-fatal modelling process and therefore are marked as belonging to “non-fatal” in the Global Health Data Exchange input data tool. The 11 sources with data on mortality risk are easily identified by the words “mortality” or “survival” in the title. For prevalence, 40 (44·0%) studies were done in western Europe, nine (9·9%) in east Asia, seven (7·7%) each in high-income Asia Pacific, high-income North America, and North Africa and Middle East. 21 (23·0%) studies were from other regions, except for central Asia, central Latin America, tropical Latin America, central sub-Saharan Africa, and southern sub-Saharan Africa, for which no data were available.

21In 2016, 6·1 million (95% UI 5·0–7·3) individuals worldwide had Parkinson's disease, of whom 2·9 million (47·5%) were women and 3·2 million (52·5%) were men. 2·1 million (34·4%) of these individuals were from high SDI countries, 3·1 million (50·8%) from high-middle or middle SDI countries, and 0·9 million (14·8%) from low-middle or low SDI countries (table). The number of individuals with Parkinson's disease in 2016 was 2·4 times higher than in 1990 (2·5 million, 95% UI 2·0–3·0). In 1990, 1·1 million (44·0%) cases were in high SDI countries, 1·1 million (44·0%) in high-middle or middle SDI countries, and 0·3 million (12·0%) in low-middle or low SDI countries. The increase in the number of patients with Parkinson's disease worldwide between 1990 and 2016 was not explained exclusively by an increasing number of older people, because global age-standardised prevalence rates increased by 21·7% (95% UI 18·1–25·3) from 1990 to 2016 compared with an increase of 74·3% (69·2–79·6) for crude prevalence rates. The increase in the number of patients with Parkinson's disease between 1990 and 2016 was less pronounced in high SDI countries (9·2%, 95% UI 5·5–13·2) than in other countries, and the largest increase was seen in middle SDI countries (59·8%, 53·2–66·1). The increase in age-standardised prevalence rates between 1990 and 2016 was similar in men (21·4%, 95% UI 17·6–24·9) and women (19·3%, 15·7–22·7). Age-standardised prevalence rates of Parkinson's disease by country varied greater than five times, with the highest rates generally in high-income North America and lowest rates in sub-Saharan Africa (figure 1).

22Globally, Parkinson's disease caused 211 296 deaths (95% UI 167 771–265 160; 93 512, 95% UI 73 702–118 421, in women and 117 784, 93 729–147 607, in men) and 3·2 million DALYs (95% UI 2·6–4·0; 1·4 million, 95% UI 1·1–1·7, in women; 1·8 million, 1·5–2·3, in men) in 2016. Of these, high SDI countries accounted for 84 911 (40·2%) deaths and 1·1 million (34·4%) DALYs, high-middle or middle SDI countries for 98 820 (46·8%) deaths and 1·6 million (50·0%) DALYs, and low-middle or low SDI countries for 27 470 (13·0%) deaths and 0·5 million (15·6%) DALYs. The number of deaths was 2·6 times higher and the number of DALYs was 2·5 times higher in 2016 than in 1990. These increases were not explained exclusively by an increasing number of older people, because age-standardised rates increased from 1990 to 2016 for both deaths and DALYs by about 20% (table). Similar to prevalence, the increases in deaths and DALYs were lowest in high SDI countries and highest in middle SDI countries, and were seen in both men and women.

23Parkinson's disease was uncommon before 50 years of age. Prevalence in 2016 increased with age thereafter and peaked between 85 years and 89 years (1·7% for men; 1·2% for women) and decreased after that age (figure 2). Age-standardised prevalence of Parkinson's disease in 2016 was 1·40 times (95% UI 1·36–1·43) higher in men than in women; the male-to-female ratio was similar in 1990 (1·37, 95% UI 1·34–1·40). A similar pattern was seen for the rates of YLLs and YLDs according to age, although the age-related increase was considerably steeper for YLLs than for YLDs, suggesting that Parkinson's disease-related case fatality rises with age (figure 3).

24The age-standardised rate of DALYs of Parkinson's disease by the 21 GBD world regions generally increased with SDI (figure 4). Sub-Saharan Africa, Latin America and the Caribbean (with the exception of southern Latin America), south Asia, and high-income Asia Pacific had lower age-standardised DALY rates than other regions with similar SDI. Southern Latin America and high-income North America were the regions with highest estimates relative to expected values based on SDI.

25Smoking was found to have a small, protective effect on Parkinson's disease, and would have been expected to prevent 461 194 DALYs (95% UI 324 745–599 845) globally in 2016 if the association was truly causal.

26Over the past generation, the number of individuals with Parkinson's disease globally has more than doubled to over 6 million. Of all the neurological disorders included in GBD 2015, Parkinson's disease was the fastest growing. Ageing populations contributed to much of that growth as crude prevalence rates increased by about 74% from 1990 to 2016 and age-standardised prevalence rates increased by about 22%. However, because age-standardised prevalence, DALYs, and death rates all increased from 1990 to 2016, additional factors are probably important.

27First, changes in study methods, availability of higher-quality studies, and greater awareness of diagnosis might have led to better estimates of prevalence, DALYs, and deaths since 1990. For example, door-to-door studies are less likely to miss individuals who have never been diagnosed and would be missed in health records. Our DisMod-MR 2.1 model did not show evidence in favour of a systematic bias between door-to-door surveys and studies based on administrative records; however, establishing such evidence in a model with relatively sparse data is difficult. Although many regions have seen improvements in study methods, these changes alone are probably insufficient to explain the rising burden of Parkinson's disease. For example, the rates of Parkinson's disease have also increased in high-income countries without substantial changes in study methodology.

28Second, increasing life expectancy is probably contributing to longer disease duration in individuals with Parkinson's disease and thus to higher prevalence, even if incidence remains constant and individuals with Parkinson's disease show the same time trends in mortality as the general population. Indeed, in a meta-analysis of ten studies, recent cohorts showed longer disease duration, with an increase of 2·5 years per decade. That study showed no clear evidence that the introduction of levodopa and improvements in Parkinson's disease care have led to improvement in survival of individuals with Parkinson's disease compared with similar individuals without Parkinson's disease. A recent study estimated the burden of Parkinson's disease in France from 2010 to 2030 under a constant incidence scenario and assuming that the relative risk of death of individuals with Parkinson's disease relative to controls had not changed over time. It showed that the life expectancy of individuals with Parkinson's disease would be expected to increase by approximately 3 years and the age-standardised and sex-standardised prevalence rate by 12% over 20 years. As patients live with Parkinson's disease for more years and the number of individuals with advanced Parkinson's disease increases, studies will be needed to inform the distribution of the severity of the disease in representative samples with simple instruments such as the Hoehn and Yahr scale.

29Third, the increase in Parkinson's disease burden might be linked to environmental factors tied to the growing industrialisation of the world. In general, better health is positively associated with socioeconomic level.,, However, with Parkinson's disease, the opposite is true; age-standardised DALY rates due to Parkinson's disease increased with SDI. The reason for this association is not clear. Some environmental exposures tied to industrialisation, including pesticides, solvents, or metals,, which are also more common in high SDI countries, might contribute to the increased incidence of Parkinson's disease. For example, in China (a middle SDI country), which has undergone rapid industrial growth since 1990, the age-adjusted prevalence rates of Parkinson's disease more than doubled between 1990 and 2016, the largest increase worldwide. If environmental factors related to industrialisation played a part, an increase in incidence over time would be expected. A few studies have examined time trends in incidence of Parkinson's disease, with inconsistent findings. A study in the USA suggested that incidence increased by 24% (95% CI 8–43) per decade between 1976 and 2005 in men but not in women. By contrast, in the Netherlands, the Rotterdam Study reported a substantial decrease in Parkinson's disease incidence between 1990 and 2011, without any obvious explanation. One study in Canada and another in the USA study showed no significant time trends. In Finland, using the Finnish National Prescription Register, the incidence of Parkinson's disease increased between 1997 and 2014 both in rural and urban regions. High-quality prospective cohort studies and detailed registries are needed to survey time trends in the worldwide incidence of Parkinson's disease more accurately and understand the factors that might be driving time trends. Alternative explanations for the positive association between the burden of Parkinson's disease and SDI include better ascertainment of Parkinson's disease in higher SDI countries through better study methods or health-care access and disease recognition. However, in lower SDI countries, we included door-to-door studies, when available, that are considered to be less prone to underestimation.

30Fourth, declining smoking rates in some countries, although a global health boon, might contribute to higher incidence of Parkinson's disease. The risk of Parkinson's disease is decreased by approximately 40% among smokers. Whether this association is truly causal or explained by reverse causation or other biases is still debated. If the association between smoking and Parkinson's disease were causal, decreasing smoking rates would lead to an increase in the incidence of Parkinson's disease in the future. Assuming a causal inverse association and a 10-year lag to account for the temporal effect of smoking on the incidence of Parkinson's disease, one study in the USA estimated that declining smoking rates in the country might increase the projected burden of Parkinson's disease in 2040 by 10%. However, because the lag time between exposure and the actual effect on disease risk is unknown and might actually be longer than 10 years, the timing of the potential effect of declining smoking rates on Parkinson's disease incidence is uncertain, and additional studies are needed to examine the effects of changing smoking habits in different parts of the world with different smoking rates and time trends. Regardless of the results of such studies, the adverse health consequences of smoking far outweigh any potential benefit on Parkinson's disease. Finally, changes in the prevalence of other known (eg, head trauma) or unknown risk or protective factors that were not included in GBD 2016 might contribute to changing incidence rates of Parkinson's disease.

31This study confirms that Parkinson's disease is about 1·4 times more frequent in men than women, and this ratio did not change substantially over the study period. Environmental (eg, occupational) exposures to which men are more frequently exposed might contribute to this pattern. The prevalence of Parkinson's disease increased with age. Underascertainment at older ages owing to underdiagnosis, comorbidities, or institutional care might explain the decrease seen in the oldest age groups after the peak between 85 years and 89 years.

32The current estimates for the global burden of Parkinson's disease are generated from imperfect data and models that are refined in each iteration of the GBD study. Estimates from GBD studies can vary from year to year as revised estimates are generated on the basis of refined methods and inclusion of more and higher-quality studies that are less likely to underestimate the true burden of the disease. Nonetheless, high-quality epidemiological studies (especially on incidence and disease severity) are still rare for large portions of the world, especially in low-income regions, where such studies are needed to understand trends and guide efforts to reduce the disease burden. Methodological differences for determining prevalence and study shortcomings might result in estimates that vary considerably and underestimate the true burden of Parkinson's disease. This under-reporting is well known for Parkinson's disease in studies based on death certificates,,,,, whereas population-based door-to-door studies are considered a better approach because they are able to capture undiagnosed cases. However, disease frequency estimates from population-based studies might be affected by selection bias resulting from non-response, particularly if individuals affected by the disease under investigation are less likely to participate. Non-response is an important issue as participation rates in epidemiological studies have considerably decreased over the past 30 years. Another limitation is that because we lack strong predictors for the occurrence of Parkinson's disease, some of the variation between countries is probably due to measurement error that we have been unable to correct. Because we rely on prevalence data to derive our cause of death estimates, any residual measurement error in the prevalence estimates is transposed onto the death estimates for Parkinson's disease. Although the the bias in mortality is an unwanted property, it is less than would have been the case if we had based our estimates on the observed rates of death with Parkinson's disease as the underlying cause from vital registration data. The large variation in death rates over time within the same countries and the even larger variation between countries are implausible and probably explained by changing death coding practices.

33Neurological disorders are now the leading source of disability in the world, and Parkinson's disease is the fastest growing of these disorders. As the population ages and life expectancy increases, the doubling of the number of individuals with Parkinson's disease between 1990 and 2016 is projected to occur again in the coming generation.,,, To address this great health challenge will require action aimed at preventing the disease where feasible and improving the lives of those affected by the condition. Among the potential responses available are preventing the disease (eg, by increasing physical activity earlier in adulthood and reducing exposure to pesticides), improving worldwide access to care and effective treatments (eg, levodopa), increasing funding for research (eg, to understand the underlying causes), and development of new therapies.