J7ournal ofNeurology, Neurosurgery, and Psychiatry 1996;61:4-16

NEUROEPIDEMIOLOGY J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from

How far are we in understanding the cause of Parkinson's disease?

Y Ben-Shlomo

generate suffi- ... the writer will repine at no censure which to have a large sample size to the precipitate publication of mere conjectural cient cases for analysis. The combined alumni suggestions may incur: but shall think himself cohorts of Harvard and Pennsylvania colleges fully rewarded by having excited the attention of consisted of 50 000 people and only generated those, who may point out the most appropriate Older cohorts will be more efficient means of relieving a tedious and most distress- 160 cases.6 ing malady. in terms of statistical power. Case- James Parkinson' control studies, although far more practical, are It is almost 180 years since James Parkinson more prone to bias. Particularly problematic is described his series of patients with what is the recruitment of control subjects and the now known as Parkinson's disease. His own measurement of putative exposures. Associa- astute clinical observations were unable to dis- tions must be treated with some caution and if cern any obvious aetiology. "On the subject possible verified using either historical data or a indeed of remote causes, no satisfactory cohort design. account has yet been obtained from any of the sufferers."' He noted an indulgence in alcohol, CASE DEFINITION lying on damp ground, and the possibility of The case definition for Parkinson's disease is trauma. He also reported the occupations of based on clinical criteria. Only two thirds of three of his patients; one was a gardener, potential cases of Parkinson's disease recruited another a sailor, and the third a magistrate. directly from primary care may truly have the Even in 1817, the disease affected a broad disease as validated by postmortem diagnosis.7 social range of society. We have made some Prevalence studies from various countries have progress in elucidating the cause of Parkinson's shown that potential cases will include patients disease over this period but this has been lim- with benign essential tremor, dementia, and ited. For instance, the role of trauma in the cerebrovascular disease.8 1' In Aberdeen, 15% aetiology of Parkinson's disease remains uncer- of potential cases were excluded after specialist http://jnnp.bmj.com/ a neuro- tain and disputed.2 4 Epidemiological research examination.8 A clinical diagnosis by has steadily increased, particularly over the logist or geriatrician is also limited. The United past 10 years, and more research has been Kingdom Parkinson's Disease Society Brain directed towards analytical rather than just Bank data have shown that the predictive value descriptive studies. More recently, work in the of an expert diagnosis is only around 75%.11 area of genetic epidemiology5 has begun and The use of stricter clinical criteria can improve value but this is at the cost researchers are now focusing on both molecu- the predictive on October 2, 2021 by guest. Protected copyright. lar and environmental risk factors. This review of excluding more genuine cases and is there- interprets the existing evidence and provides a fore only really of value for therapeutic trials personal view on what we definitively know rather than epidemiological studies.'2 The use about the epidemiology of Parkinson's disease, of complex investigations, such as PET, is what remains speculative, and what questions not suitable for large epidemiological studies. remain to be answered. The result of including other disorders as well as true cases of Parkinson's disease in a study will depend on the relation between Problems facing a neuroepidemiologist these diseases and the exposures of interest. If Department of It is important to appreciate the problems and this is no different to that found for Parkinson's Epidemiology and limitations faced by neuroepidemiologists disease, the results will not be biased. It is Public Health, 1-19 that do not Torrington Place, attempting to study Parkinson's disease. These more reasonable to assume they University College, drawbacks help to explain some of the rather share the same aetiological factors. In this case, London Medical inconclusive and often contradictory findings misclassification will bias the results towards School, London it will be harder to show a relation WC1E 6BT, UK between studies. the null; Y Ben-Shlomo between exposure and disease, even if it truly Correspondence to: STUDY DESIGN exists. Dr Y Ben-Shlomo, Department of Social Most analytical studies use a case-control Medicine, University of approach. This is because Parkinson's disease MEASUREMENT OF EXPOSURE Bristol, Canynge Hall, The measurement of exposure is also problem- Whiteladies Road, Bristol is relatively rare. Any prospective study needs BS8 2PR, UK. atic. It is unclear as to when in the lifecourse an Howfar are we in understanding the cause ofParkinson's disease? 5

Figure 1 Nigral cell loss: 100 neurons at a constant rate. It did, possible periods during the however, lifecourse for the role ofan exclude a process which resulted in either an J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from environmentalfactor. Cell accelerating or decelerating cell death. loss is shown at a constant 80 Most exposures are measured by subjective ratefor simplicity; see text for details. (Modifiedfrom recall. This has two problems: low impact Langston"7.) exposure, such as past diet, is likely to be M 60 0 poorly recalled by both cases and controls ' 6 (non-differential misclassification). This will dilute any true relation and lead to erroneously _F 40 concluding that there is no relation. High 0 z impact exposures might be differentially recalled by cases and not controls, particularly if 20..._...... there is pre-existing knowledge of a possible Clinical symptoms link between the exposure and disease. For oN example, a previous head injury might be more 0 20 40 60 80 100 salient in the minds of patients than in con- Age (y) trols, as they may link the trauma with their neurological disease (recall bias). exposure may be important (fig 1). Clinical GEOGRAPHICAL PATlERNS symptoms are thought to start once cell loss International comparisons have in the past has reached a threshold level of 70% to 80%. relied on mortality rates adjusted for demo- Normal age-related cell loss (a) is thought not graphic differences between countries. Early to result in disease unless current life findings showed a sevenfold to eightfold varia- expectancy is prolonged. A genetic model tion in mortality rates2223 suggesting an impor- might propose that cell loss from birth tant role for environmental factors as well as occurred in some people at a faster than nor- possible genetic differences. Mortality rates are mal rate (b) due to "aging genes"." susceptible to variations in diagnosis, survival, Intrauterine events, such as infection or pla- and certification practice, which could produce cental abnormalities,'4 15 might result in a per- large artefactual differences. This is best illus- son being born with a depleted cell count and trated by the fivefold mortality differences normal age related loss (c) would ensure that between Scotland and Japan.22 However, there disease would appear in later life. This devel- is little difference in the prevalence rates for opmental model has been proposed to explain patients between 60 and 69 years of age, when the discordance rate seen for monozygotic underascertainment should be less problematic twins.'5 Environmental factors could act (Scotland 254/100 0008; Japan 245/100 000).9 through an acute mechanism-for example, a More complex approaches differentiate head injury (d) destroying a finite number of between "multisource" prevalence studies, cells-followed by nornal age related loss (two which rely on the complete identification of stage hypothesis).'6 Alternatively a low level patients with a pre-existing diagnosis, and chronic exposure (e)-for example, a diet rela- "population based" surveys which screen a tively deficient in antioxidants-may simply act total defined population identifying both pre- http://jnnp.bmj.com/ by slightly accelerating the rate of cell death. existing and new cases.24 The population based Finally an acute event-for example, exposure survey method is particularly important for to a toxic chemical-may act by triggering cell developing countries because of limited med- death which results in a cascade of events and ical services or for communities in which there accelerated cell death (f). This final model may be differential access to health care. For would suggest that a long latency period may example, differential access to health care for not be essential for disease aetiology and would ethnic minorities is well recognised in the on October 2, 2021 by guest. Protected copyright. focus attention on later life. Most people United States25 and in the Copiah county believe that a long latency period for study, more previously undiagnosed black Parkinson's disease is most likely.'7-'9 However, patients (58%) than white patients (32%) were pathological data challenge the hypothesis that found.26 The population based survey method age related cell loss on top of an external insult also enables more valid comparisons between is important.20 It is also possible that causation is studies as ascertainment is more standardised heterogenous and that any one of the various than conventional multisource studies. models might operate in different persons to Unfortunately, as the prevalence rates are usu- produce cell loss and this would be clinically ally based on relatively few cases they are less indistinguishable. robust, with wide confidence intervals. Recently, the temporal profile of neurode- Age adjusted prevalence rates suggest that generation in Parkinson's disease has been a threefold difference may exist, with Libya compared by various mathematical models to (57/100 000) at one end and Iceland (182/ distinguish between different patterns of cell 100 000) at the other.24 Prevalence rates are loss.2' Several important, and possibly ques- also susceptible to differences in diagnosis and tionable, assumptions had to be made to survival. Artefactual differences can be reduced enable the calculations to be performed. The by limiting the comparison to European coun- empirical data seemed to be compatible with tries with good healthcare systems, using inci- both an "event" that kills some neurons and dence rates as these are independent of reduces the life expectancy of others and a differences in survival, and comparing rates for "process" which continuously kills healthy subjects under 70 years, in whom underascer- 6 Ben-Shlomo

Table 1 Age specific incidence rates per 100 000 for Parkinson's disease from selected European studies

Rates per age group J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from Relative rates * Country (penrod) 30-39 40-49 50-59 60-69 70-79 80-89 50-69 (95% CIs) Iceland 1-4 5-6 32-9 97 2 136 2 69-2 61 3 2-75 (2-31-3-27) (1954-63) Southwest Finland 0 7 2-6 19 8 62-3 92 6 47.9 40.1 1 80 (1-46-2 21) (1968-70) Ferrara, Italy 0 7 7-2 25.1 49.6 45-8 1.1 35 9 1 61 (1 40-1 86) (1967-87) Aarhus, Denmark 0.0 1 0 15 8 39-8 58-5 28 1 26.8 1-20 (0-90-1 60) (1967-70) Netherlands 0.0 0-0 14 5 25-7 79.3 117 2 19 5 0-88 (0-54-1 42) (1983-5) Sardinia, Italy 1-3 7-1 26-4 18-0 4 3 0.0 22-3 1 00 *Relative rate for 50-69 years compared with Sardinia, Italy. Data taken from de Pedro-Cuesta.2

tainment should not be important.27 These some countries such as China and Africa. This results (table 1) suggest threefold variations, may reflect racial differences in genetic suscep- but this is reduced to less than twofold (80%) if tibility or the absence of specific environmental Iceland is excluded. No obvious geographical factors related to industrialisation. In China pattern emerges, and the rates for Denmark there is some evidence to support the role of and Holland are more similar to Sardinia than industrialisation as residing in a village resi- rates from Ferrara, Italy. Community door dence was seen to be protective.3' There is a to door studies suggest that prevalence rates need for further migrant studies to elucidate (fig 2) are fairly similar between both devel- the role of environmental factors.32 Across oped and some developing countries except for Europe there is relatively little variation, and China and west Africa. A recent pilot study what there is may be explained by methodolog- from Taiwan (not shown in fig 2) found much ical differences. Current standardised popula- higher rates than in China, even though it was tion based prevalence surveys will determine carried out in a rural area.28 This suggests that whether these differences are artefactual. The black persons in America and Chinese in surprisingly high rates seen in Iceland may be Taiwan have higher rates of disease than idiosyncratic or reflect a higher degree of ascer- their counterparts in west Africa or China. tainment when surveying a small population.3" Environmental factors may thus be more Analysis of more recent drug utilisation data important than racial factors in explaining for Iceland supports this finding.34 Further these variations.26 work should confirm this observation and A recent report on the prevalence of Lewy explore possible reasons, be they genetic or bodies in brains from Nigerians examined at environmental. postmortem has shown similar rates to those in western populations.29 Assuming that inciden- tal Lewy bodies reflect preclinical disease, this Temporal patterns finding supports the notion that the propensity MORTALITY RATES http://jnnp.bmj.com/ to develop Parkinson's disease is universal. It Descriptions of what might have been also supports the role of environmental fac- Parkinson's disease existed well before the 19th tor(s), as it is still necessary to explain why century. The ancient ayurvedic literature some Afro-Americans go on to develop clinical (4500-1000 BC) of India describes an illness disease whereas their African counterparts "kampavata" consisting of tremor and akine- remain asymptomatic.3' sia.35 Analysis of temporal trends have mostly relied on long term patterns in mortality rates. on October 2, 2021 by guest. Protected copyright. COMMENT Studies from several different countries have all It is clear that Parkinson's disease is found shown similar patterns of mortality over time: throughout the world, but is less common in the United States,3668 United Kingdom,39 42

Figure 2 Age adjusted Montevideo, Uruguay r prevalence rates/I00 000 from door to door surveys. Parsi, Bombay, India K * Indicates crude prevalence and not age Rural Kashmir, India (> 49 years) adjusted. The study by Tison et al66from Gironde, Thugbah, Saudi Arabia h France is not shown. (Data amendedfrom Copiah Co, Mississippi, USA - Zhang and Roman24.) Sicily K Six cities, China H 29 provinces, China K FH lgbo-Ora, Nigeria* lgbo-Ora, Nigeria* (2) 10 100 1000 Prevalence rate/100 000 Howfar are we in understanding the cause ofParkinson's disease? 7

Ireland,4' Italy,44 Norway,45 Denmark,36 and for cohorts born between 1875 and 1895.

Japan.46 Mortality rates for both men and Standardised mortality ratios for those aged J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from women have shown increases for older age 30-50 in 1920 in England and Wales were groups (> 75 years), and a decline for younger raised compared with those born earlier or age groups (< 65 years). These patterns are later.40 More recent multivariate modelling remarkably consistent and predate the wide- methods have been developed to disentangle spread introduction of levodopa in the 1970s. age, period, and cohort effects.55 Initial analy- There are several possible explanations for ses using these methods support these findings these patterns. (1) Awareness, diagnosis, and for both England and Wales and United States certification of Parkinson's disease, particularly data (C Martyn, personal communication). in elderly patients, may have increased over Despite these complex methods, there is still time. This might explain some of the increase an essential problem that cohorts born after the seen for older patients but is unlikely to explain epidemic from the mid-i 930s are still relatively the rise seen before the 1970s. It is unlikely to young (< 65 years) and have not reached the explain the differential increase in mortality age when mortality rates have been shown to seen for men compared with women.'8 White be increasing. The cohort hypothesis is also men over 80 showed a 100% greater increase unlikely to explain the differential mortality than women when comparing mortality increase seen for men (see above) as there was changes between 1962 and 1984. This is little difference in encephalitis lethargica mor- unlikely to reflect diagnosis or treatment and tality rates for men and women.56 could reflect occupational risk factors.38 This pattern has also been seen in data from INCIDENCE RATES England and Wales (unpublished personal Incidence rates are far superior to mortality observation). (2) Levodopa treatment may rates, but are still potentially biased by temporal have increased survival resulting in a shift of trends in consulting behaviour and doctor the age specific mortality curve to the right by diagnosis. Data from Rochester, Minnesota are about five years.36 There is some evidence to unique in covering a long period suggest that the initial introduction of levodopa (1935-79).57-5 The annual incidence rate has treatment may have delayed death.47 However, increased from 11-4/100 000 in 1935-44 to more recent survival cohorts48 49 suggest that 18 2/100 000 between 1967 and 1979. The mortality from Parkinson's disease compared age adjusted rates have remained fairly con- with that of the general population has not stant over the past 30 years.24 Age adjusted altered to a large degree from the prelevodopa rates can mask heterogeneity in age specific days.50 There has been a secular increase in life patterns as rates in older age groups could expectancy for the whole population, so that a increase whereas those in younger groups patient with Parkinson's disease today would decreased leaving the adjusted rate unchanged. expect to live longer like anyone else in the The use of different age groupings in the two population. (3) There may have been a gen- previous publications58 59 makes it impossible to uine change in the incidence of disease: (a) compare the age specific rates over this entire The incidence of disease in elderly people may period. Comparing 1945-54 with 1955-66, it have increased because of a reduction in heart does seem that the rate for patients aged

disease and stroke mortality and hence a between 40-69 has decreased whereas that for http://jnnp.bmj.com/ decline in competing causes of death.5" If the those over 70 years has increased.60 risk of Parkinson's disease was also associated with heart disease or stroke, a reduction in COMMENT deaths from heart disease would selectively Temporal data are limited and difficult to increase the pool of potential subjects who interpret. Mortality rates among elderly people could develop Parkinson's disease. Only one have increased substantially but the interpreta- study has shown such an association49 and this tion of this finding is controversial. The most has not been replicated.52 If these diseases are reliable data to answer this question come from on October 2, 2021 by guest. Protected copyright. independent of each other, then whereas a Rochester, Minnesota. A reanalysis of the reduction of heart disease will increase the Rochester data including more recent data absolute number of cases of Parkinson's dis- between 1980 and 1995 is essential to help us ease, it will not alter the age specific rates. (b) discriminate between various competing There has been a decrease in the incidence of hypotheses. disease in young age groups due to a decrease in environmental exposures.'7 (c) The diver- gence of age specific mortality rates reflects a Individual risk factors cohort effect due to a self limiting exposure, SOCIODEMOGRAPHIC FACTORS: AGE, SEX, AND such as the encephalitis lethargica epidemic.5'54 SOCIAL CLASS Such a hypothesis would predict an increase in The risk of developing Parkinson's disease mortality as "exposed" cohorts aged but a increases as a person gets older. But, it is decline in the mortality of younger age groups unclear whether the increase in risk is continu- who would have been born after the exposure. ous or eventually declines for the oldest age As these later unexposed cohorts reached old groups. The first pattern would favour the age there would be a subsequent reduction in notion of "aging-related" degeneration as has mortality for all ages. There have been several been argued for Alzheimer's dementia.6' attempts to examine the cohort hypothesis Multisource prevalence studies suggest that using mortality data.40414 Plotting mortality prevalence rates generally fall for the oldest age rates by birth cohort suggests excess mortality groups (table 1). It is likely that older subjects 8 Ben-Shlomo

are less likely to present for medical attention, Almost every study has failed to show any are more difficult to diagnose, and even when relation with 75 years of education.72 One study J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from under residential care may be missed.62 which did show that cases were better edu- Population based studies consistently show cated76 failed to adjust for age or sex, two that prevalence rates increase continuously in important confounders as controls were both an exponential fashion.26 Because broad older and were more likely to be women. age groups are used to calculate the age specific Similarly, a study of early onset Parkinson's rates, it is possible that these studies could disease showed an initial twofold increased risk mask a flattening of the age curve. However, associated with having completed high school the study from Gironde, France calculated graduation, but this was almost completely rates for five year age groups between 65 and attenuated after adjustment for smoking habit older than 90 years and also found an expo- and other variables.77 A case-control study with nential rise.66 discordant twin pairs suggested that cases were Most studies support an excess of male to less likely to be manual workers but this was female cases. The average ratio of male to not significant (odds ratio 0 50; 95% confi- female standardised rates is 1 35 from preva- dence interval (95% CI) 0-10-2-23).i' Only lence studies and 1 31 from incidence studies, one prospective study has reported social class but the range is wide.24 This excess is seen for and again found no relation.79 Farming has both multisource and population based stud- been examined as a specific occupation with a ies. Hospital based studies may be misleading, possible raised risk. Some75 "I 81 but not all stud- as there is evidence that the survival of women ies82 83 have shown an increased risk. This could recruited from hospitals is worse than men and relate to pesticide use or other exposures that hence will bias the results in favour of increasing are more common in rural areas (see later). the prevalence of men compared with women.49 COMMENT Adult social class is a variable often mea- Parkinson's disease is more common among sured in epidemiological studies67 as it is so older subjects and men. Limited evidence sug- strongly related to many diseases,68 and is often gests that risk increases continuously with age. used as a confounding variable when examin- Further evidence is required to test the "aging ing other exposures. American studies tend to related" hypothesis by focusing on population use a classification based on years of education surveys of elderly people. It is important to or income, whereas European studies are more determine why men have an increased risk likely to use an occupation based classifica- compared with women. This could be related tion.69 As a variable it is non-specific, and acts as to smoking behaviour or other toxic exposures. a proxy marker for many other exposures such Most case-control studies match cases and as smoking, diet, obesity, occupational expo- controls by sex, and are unable to examine the sures, sanitary conditions, overcrowding, risk odds ratio associated with sex before and after of infection, income, etc. Many potential risk adjustment for other risk factors. A population factors are distributed in a graded fashion by based case-control study, which does not socioeconomic class. Showing a gradient in match subjects by sex, would be useful to clarify disease by socioeconomic class is often not very this issue. There seems to be little social class

helpful as it fails to indicate which one of the or educational gradient. If anything, the risk http://jnnp.bmj.com/ many potential exposures is of importance. seems increased with higher socioeconomic However, the lack of a gradient is of relevance as class or more years of education. The lack of a it challenges conventional hypotheses. For gradient is surprising as exposures to toxic example, if it is speculated that Parkinson's dis- agents, infectious agents, etc are likely to differ ease is the result of a chemical exposure, it is by social class. Adult social class is moderately likely that such an exposure would be greater correlated with parental social class but over for manual rather than non-manual jobs, the past 60 years there has been upward social unless the chemical is commonly found in the mobility. Exposures acting in childhood may on October 2, 2021 by guest. Protected copyright. domestic home environment. Similarly, risk of be more relevant than adult factors and future infection will be associated with crowding, san- studies need to examine childhood circum- itary conditions, and hence socioeconomic stances by collecting data on parental social class. class as well as other childhood variables. Mortality data from the United Kingdom show no social class gradient for deaths between 20-64 years. For older ages there is a Specific exposures gradient with higher ratios in social class I and II Many potential exposures have been cited as than IV and V.70 However, proportional mor- possible risk factors for Parkinson's disease. tality ratios are influenced by the strong social These can broadly be divided into three class gradient for other more common causes groups: toxic exposures, infectious exposures, of death, which would artefactually increase and a heterogenous group of miscellaneous the proportion of mortality from Parkinson's exposures. Table 2 lists some of the studies disease in higher social classes.7' More reliable that have investigated these exposures. Cohort data can be obtained from case-control studies. studies provide the most robust observational Unfortunately, few studies have either exam- data. ined or reported the results for adult social class and none have reported parental social TOXIC EXPOSURES class, which would reflect childhood living con- The report that 1-methyl-4-phenyl- 1,2,5,6- ditions. tetrahydropyridine (MPTP) has neurotoxic Howfar are we in understanding the cause ofParkinson's disease? 9

Table 2 Relation between environmental riskfactors and Parkinson's disease according to study design

Types ofstudy design J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from Ecological studies Case control studies Cohort studies Increased No Increased Decreased No Decreased No Risk factor risk relation risk risk relation risk relation Toxic exposures: Pesticides, herbicides 84 75,77,80,81,85 74,82,83,86,87 Solvents 78,86 Mercury 88 86 Industrial residence 89 31 90 Infectious exposures: Shingles 78 Measles 72 6 Chicken pox 72 6 Mumps 72 6 Influenza 14 91 Whooping cough 34 6 Miscellaneous exposures: Smoking 72,74,77,78,92 76,80,88,99 79,90,101- 104 -98 100 103 Personality 94,105-107 Head trauma 74,95 77,81 108 Rural residence 84 80,82,83,85,87, 31 74,77,81,90 88 109 Well water 82,83,87,90 31,74,856,109 Raw vegetables 80 Dietary antioxidants 77 85,110,111 112 Alcohol 90,100 76,80,88,98 79 113 Coffee 98,100 79 If a study has a non-significant result, it has been classified as showing no relation even if the results are in the same direction as other studies. effects in humans"4 has led to the hunt for a an over-representation of rural cases compared more commonly occurring environmental with rural controls (see section on rural resi- toxin. The validity of using the MPTP model dence for detailed discussion). However, in to generate clues relating to idiopathic some studies the association with pesticide use Parkinson's disease is generally thought to be was present75 despite no association with rural reasonable.'"5 Paraquat is similar in structure to residence'09 and vice versa.8' The study from MIPTPI6 and therefore most attention has Calgary, Canada75 '09 is particularly impressive focused on herbicides and pesticides, although as its design should have minimised any selec- other toxic compounds, such as manganese tion bias and it also obtained detailed occupa- and mercury, which are known to have neuro- tional histories. One possibility is that logical effects, have also been examined. An pesticides differ between studies and only some early ecological study from Canada showed a types show an association with Parkinson's dis- positive relation between a rural area with the ease. However, this explanation is unlikely as greatest pesticide usage and both Parkinson's most of these studies cover a similar period and disease mortality and drug utilisation levels.84 examine populations in North America. Case-control studies can be divided into those Cross sectional data show an association

that show an association75 77 80 81 85 and those between exposure to organophosphate pesti- http://jnnp.bmj.com/ that do not.74 8283 86 87 Several possibilities exist cides and subtle differences in both cognitive to explain these discrepant results. Most of the processing and depression."8 Depression has studies that show an association, suggest a been shown to be common in Parkinson's dis- threefold increased risk associated with pesti- ease and may predate clinical symptoms."9 120 cide use. A weak study may have insufficient It is not clear whether these results have any power to detect such a risk and would lead to a relevance to the pathophysiology of spurious negative finding (type II error). Parkinson's disease. As this study was cross However, several of the negative studies were sectional, it is uncertain whether such subtle on October 2, 2021 by guest. Protected copyright. sufficiently large to have significantly detected cognitive differences may not have existed such an increased risk.8387 For example, the before the exposure. Only limited data are Kansas study was sufficiently powerful to available on occupational cohorts. The "Iowa detect a twofold increased risk.83 Most of the 65+ rural health study" showed that farmers, studies have measured occupational pesticide either still working or retired, were less likely to exposures75 83 but some have included domestic have Parkinson's disease compared with other exposure.74 80 Recall bias is unlikely to explain rural controls.'2' No information was provided, this association as there is no reason that it however, as regards their pesticide exposure should operate in some but not all studies and smoking habit was not taken into account. given the similarity of their populations. The hypothesis that toxins are involved has Empirical evidence suggests that differential been given much more credence by the recent recall for pesticides between cases and controls discovery that patients with Parkinson's disease is not a serious consideration." 7 Studies that have relatively less effective detoxification sys- show an increased risk with pesticides also tend tems.'22-'26 This enhances the plausibility of the to be the same studies with an increased risk "ecogenetic" hypothesis that Parkinson's dis- for rural residence. This is to be expected as ease may result from the combination of an occupational pesticide use is likely to be more inherited susceptibility and an environmental common in rural areas. An artefactual associa- toxin. If susceptibility is relatively rare-for tion between pesticides and Parkinson's dis- example, the proportion of the general popula- ease could exist if "selection bias" resulted in tion with poor metaboliser status for debriso- 10 Ben-Shlomo

quine is thought to be between 5% and 10% non, but might relate to other factors, such as the exposure may be common as most exposed infection in very early life. One study suggested J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from subjects will not experience any adverse results. the possibility of "in utero" infection but this From an epidemiological perspective, a com- was based on ecological data'4 and failed to be mon exposure experienced by 80% or more of replicated.9' Alternatively, it may reflect varia- the general population will be much harder to tions in immune function or the possibility of detect and require far larger sample sizes. confounding by another factor such as socioe- Studies will either need to be restricted to sub- conomic class or family size. The non-speci- jects with the susceptibility genes or must be ficity of the results favours the role of a able to test for interactions between genetic confounding factor, but this remains unre- and exposure state. solved. The public health importance of pesticide use as a cause of Parkinson's disease is limited. MISCELLANEOUS EXPOSURES Assuming a causal relation, the population Smoking and personality attributable risk-that is, the percentage of all The inverse relation between smoking and cases of Parkinson's disease that might be Parkinson's disease has been seen consistent related to occupational pesticide use is only for both case-control and cohort studies. Dorn, around 10%, although this could vary between in one of the earliest smoking cohort studies,'' 2% and 25% (95% CI).75 This estimate was so surprised by the association that he excludes non-occupational exposure, which is sought and received confirmation from both far more common although less intense. If Doll and Hill'l03 and Hammond. 1"2 Some publi- some subjects are particularly susceptible to cations have failed to show a statistically signif- even low level exposure, then pesticides could icant association but in most studies the results pose a far more serious public health risk. have been in the same direction. The consis- tency of this finding in different populations INFECTIOUS EXPOSURES and using different methods, makes it reason- The notion that Parkinson's disease could be able to exclude chance as an explanation for related to an infectious agent is far older and this association. Other possible explanations has existed since the recognition of posten- are (1) Bias Subjects may biologically age at cephalitic parkinsonism. Other neurological different rates due to genetic differences in the diseases, such as subacute sclerosing panen- ability to repair cellular damage. Smoking may cephalitis, also provide an analogous model of adversely affect DNA repair mechanisms and disease related to latent infection. Whereas a therefore smokers with less efficient mecha- toxic hypothesis has risen in popularity, an nisms would be selectively eliminated (selective infectious hypothesis has declined and work mortality). Older surviving cigarette smokers examining infections is generally older.43472789' would have relatively more effective DNA Some authorities consider that "the disease is repair mechanisms and would therefore be not of viral origin .. ."127 and hence there is no protected from developing either Parkinson's further need to test this hypothesis. Indeed, the disease or Alzheimer's disease.'3 12 This argu- inconsistent seroepidemiological results'28 128 ment is flawed for two reasons."' Monozygotic and failure to detect specific viral particles, twin pairs, when both twins are alive, still show

inclusions, or antigens in brain necropsy mater- the inverse association with smoking. Case- http://jnnp.bmj.com/ ial'30 suggest that pursuit of this hypothesis may control studies using early onset cases show be fruitless. However, several intriguing obser- similar findings, even though selective mortal- vations remain which require an answer. ity is unlikely to be of any importance for popu- Serological analyses suggest, if anything, that lations at this young age. (2) Causality-There serum antibody titres in cases of Parkinson's are several biologically plausible reasons why disease are often no different or lower than for smoking may protect or retard the develop- controls. For example, titres for measles, ment of Parkinson's disease. Hydrazine found rubella, herpes simplex 1, herpes simplex 2, in tobacco smoke protects mice from damage on October 2, 2021 by guest. Protected copyright. mumps, and cytomegalovirus were all lower for to nigrostriatal cells from MPTP. 1'4 Carbon patients, with the first three being statistically monoxide or other factors in cigarette smoke significant.'28 Although serological data are may scavenge free radicals. '" Smoking may objective, they cannot identify the age at infec- have an effect on mitochondrial activity and tion, which might be a more important predic- oxidative damage.'36 Polyaromatic hydrocar- tor than infection itself.34 13' Studies that have bons in tobacco smoke may induce used subjective reporting of infections also cytochrome P-450 enzymes and lead to show no significant differences between cases increased metabolism of xenobiotics. l 7 If and controls, except for measles infection, smoking did in fact alter the rate of nigral cell which has been reported to be less frequent death it could be predicted that smokers who among college students who went on to to developed Parkinson's disease would both be develop Parkinson's disease.6 However, pertus- older than non-smokers and would have a sis, chicken pox, and mumps infection also slower rate of disease progression. Both of seem to be less common,672 consistent with the these predictions are refuted by empirical serological results. It is unlikely that recall bias evidence.98-100 138 (3) Reverse causality could explain this as these results were found Parkinson's disease makes people give up in both retrospective and prospective studies. smoking as a secondary phenomenon. But the Simple misclassification would bias the results inverse association is still found if smoking towards the null. A reduced frequency of child- habits are examined for a period before onset hood infections could be a chance phenome- of clinical disease.773 It is possible, however, Howfar are we in understanding the cause ofParkinson's disease? I11

that subtle changes in personality occur well graphical prevalence rates between countries, before overt clinical disease as part of a more yet within countries residing in a rural area has J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from insidious onset. Such changes might effect been noted to increase risk.77 80 82 83 85 87 88 One of smoking behaviour. Personality may therefore the first reports of such an association came either act as a proxy marker of disease, or is a from a Canadian series of 21 patients whose confounding variable unrelated to the disease disease started under 41 years of age.145 This process. (4) Confounding-There are premor- series noted an excess of rural patients who had bid personality differences associated with both drunk well water. Rural residence, like social the likelihood of becoming a smoker and con- class, is a rather non-specific variable as it tinuing to smoke and other exposures/genes could simply be a proxy marker for various that are directly related to the risk of develop- other exposures, such as well water consump- ing Parkinson's disease. Smokers seem to differ tion, farming, pesticide exposure, diet, contact in personality traits from non-smokers and are with farm animals, or age at infection. Any more likely to be extrovert or exhibit type A genetic susceptibility to develop Parkinson's behaviour.'39 Non-smokers have higher levels disease may also differ between urban and of shyness and defensiveness.'40 Similar person- rural populations, as isolated rural populations ality differences may also determine alcohol might have a greater degree of inbreeding. consumption. Most studies show that patients Some studies have suggested that rural with Parkinson's disease either are no different exposure in early life is more important,82 or have reduced alcohol consumption.90 100 whereas others suggest that it is the total num- Patients with Parkinson's disease are reported ber of years.80 This finding is not consistent, to show personality changes, such as introver- with some studies showing a decreased risk3' or tism, that would make them less likely to be no significant risk.74 77 8190 109 Demographic vari- smokers.94 105-107 We do not know, however, to ables, such as rural residence, are particularly what degree personality, such as being the sensitive to the method of recruiting controls'46 more dominant twin,'07 could influence other and the possibility of selection bias. Most stud- risk factors, such as migration from rural to ies recruit patients from neurology or move- urban areas or reducing the risk of exposure to ment disorder clinics based in urban centres. pesticides. Alternatively certain genes may These specialist centres will often have a wide influence both the tendency to smoke and the catchment area and see patients from both risk of developing Parkinson's disease. This urban and rural areas. It is likely that elderly argument has been suggested to explain the rural patients might be managed locally and similar association found with Alzheimer's dis- not get referred on because of transport and ease.'41 access difficulties.'47 This is not likely for atypi- cal, severe, and young patients and specialist Head trauma clinics will have an overrepresentation of Some studies have noted an excess risk of head these.'48 Studies showing an increased risk trauma among patients with Parkinson's dis- (except one) selected controls from the hospi- ease than among controls.7495142 This associa- tal which recruited the cases. Controls were tion is usually doubted because of the usually subjects with more common medical likelihood that it reflects recall bias. One conditions such as heart or respiratory disease

method to reduce this possible bias is to only and would therefore be more likely to include http://jnnp.bmj.com/ measure severe head injuries, such as those local urban residents. Such a bias would arte- resulting in loss of consciousness. Even though factually increase the proportion of patients such events should be recalled equally well by with Parkinson's disease from rural areas. The both cases and controls, patients with only exception85 used spouses as controls. It is Parkinson's disease still show an increased unclear whether spouses might be more or less risk.74 Another strategy to control for recall bias likely to come from the same area as the patient is to ask patients about their own "lay beliefs" and this would depend on the level of migra- as to what caused their disease.'43 Patients who tion and mixing of urban and rural residents. on October 2, 2021 by guest. Protected copyright. report trauma as a possible cause can then be Studies which failed to show any increased risk removed from the analysis to determine the used various methods: "buddy" controls,74 degree of potential bias, although this might neurological clinic controls,90 stratified sam- underestimate the true risk.144 Only one study pling,8' rheumatoid arthritis controls,77 and has examined a cohort of subjects with population based control selection.'09 The use recorded head injuries and followed up their of buddy controls is not to be recommend as it risk of Parkinson's disease. This study failed to may well result in overmatching.'49 Selecting find any increased risk but had a 30% probabil- patients from a specialist clinic may diminish ity ofnot detecting a twofold relative risk even if selection bias but is also problematic especially one existed.'08 Susceptibility to trauma, rather if a single disease entity is chosen. Only the than severity itself may be more important. If population based study ensured non-biased head injury triggers a cascade of biological ascertainment of both cases and controls'09 and events that lead to cell death in only a few sus- this study failed to show an association. In fact, ceptible people, then it may be more difficult to rural residence in this study may have been detect an increased risk in a head injury cohort associated with a decreased risk for those sub- than in a case-control study. jects not exposed to pesticides. As this study showed an increased risk for occupational pesti- Rural residence cide use75 and this would be more common in It is paradoxical that industrialisation has been rural areas, the risk for rural subjects not proposed to explain the differences in geo- exposed to pesticides must have been reduced 12 Ben-Shlomo

so that the overall rural risk was not raised. It is males.85 Nuts and seeds seemed to be harmful interesting to note that the incidence study in yet another study.77 The most impressive from Ferrara, Italy did observe a greater risk of data come from a large prospective cohort J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from Parkinson's disease in rural areas, but this was study of 41 836 women, who have been fol- restricted to patients under 50 years and was lowed up for six years. In this study there was not seen among housewives. 150 no association between vitamin E and Parkinson's disease but a significant protective Well water effect was seen for both vitamin C and man- Well water, like rural residence, has also been ganese consumption, whereas vitamin A was seen to be associated with Parkinson's disease. associated with an increased risk."2 The failure This is not surprising as the two exposures are of the DATATOP trial to show any benefit closely correlated. Hence studies that show a with a-tocopherol'56 should not be interpreted relation between rural residence also tend to as excluding the potential role of dietary show a relation with well water and vice versa. antioxidants. Firstly, it is possible that the The strong correlation between these two vari- wrong intervention was chosen, and secondly, ables is evident in the Kansas case-control dietary factors may protect disease onset study. Well water had a significantly raised (induction) and yet have little effect on progno- odds ratio of 1 7 with Parkinson's disease, but sis once disease is established (promotion). For after adjusting for rural residence this was example, stopping smoking after the diagnosis attenuated to 1- 1.83 One study specifically used of lung cancer has little effect on survival. factor analysis in an attempt to overcome the Given the possible role of antioxidants in the problems of excessive collinearity.8' This is aetiology of Parkinson's disease, these results only a partial solution and does not overcome need confirmation using validated methods of the problems of determining "independent" dietary assessment such as seven-day weighed effects when measurement error is taken into diaries and prospective follow up. In view of account.15' Designing studies that specifically the long latency period, it will also be impor- break the confounding between two closely tant to consider both adult and childhood diet. correlated exposures is better than using com- plex statistical methods, but is not always possi- COMMENT ble in reality. Various studies have raised the possibility that Well water could act as a carrier of a potential several different environmental exposures toxin or as a vector for an infective agent. No might be related to Parkinson's disease. A toxic relation has been noted, however, in either pes- exposure is perhaps the most attractive sugges- ticides or heavy metal composition of water tion as it would be consistent with both the evi- supply sampled for early onset cases and con- dence on oxidative damage as well as genetic trols.152 Other possible factors still need to be variations in detoxification systems. Several considered, such as the role of sulphur contain- studies suggest pesticides to be of importance, ing compounds. Patients with Parkinson's dis- yet this finding is not consistent and is still rela- ease have reduced S-oxidation capacity and tively non-specific. It is important that we col- sulphate conjugation.'23 Alternatively, well lect more data on this exposure and refine water, like rural residence, may again be simply current methods of exposure measurement, by

a proxy marker for another exposure which is combining self reporting measures with biolog- http://jnnp.bmj.com/ more common in rural environments. ical variables. Prospective occupational cohort studies among those known to have high expo- Dietary factors sures will need to confirm these associations Diet could play a part in the aetiology of and identify whether any risk is linked to a spe- Parkinson's disease. The rare disease of amy- cific agent or a broader class of exposures. otrophic lateral sclerosis/parkinsonism-demen- Infectious exposures seem, if anything, to be tia, seen on the Islands of Guam and Rota, is less frequent in cases of Parkinson's disease. thought to be due to a dietary toxin from the The interpretation of this finding is uncertain on October 2, 2021 by guest. Protected copyright. cycad plant,'53 which is not found in western and further work is required to examine not diets. Diet is complex and it is difficult to only exposure, but also age at exposure. Self quantify exposure.'54 Usually all subjects are recall data will be problematic and other proxy exposed to the hypothesised causal factors and markers, such as household size, measures of thus continuous variables need to be mea- crowding, etc should also be examined. The sured, often with a limited range of variation. inverse relation between smoking and Retrospective data are especially prone to inac- Parkinson's disease remains enigmatic but curacy and recall bias.'54 Recall of past diet in deserves further attention, particularly in rela- each case is seen to be strongly influenced by tion to premorbid personality differences. In current dietary intake.'55 This influence may be the future, cohort studies with data on person- especially important in the context of case-con- ality and behaviour of children will eventually trol studies, in which the possibility exists that have sufficient cases of Parkinson's disease to cases but not controls may alter their diets sub- test these questions definitively. At the sequent to the diagnosis of their disease. moment these cohorts are still relatively young Several studies have found a relation between and have too few patients with Parkinson's dis- specific food stuffs and Parkinson's disease. ease. Head trauma cannot be definitively These results are, however, still tentative. For excluded as a risk factor, despite a negative example, one study showed that a "protective" cohort study of subjects with past head injury. effect with peanut consumption was seen only Larger cohorts with documented injury or in females, and salad with dressing only in occupational cohorts at high risk of head Howfar are we in understanding the cause ofParkinson's disease? 13

trauma should be examined. Laboratory work Conclusions is also needed to examine the biological plausi- There is little doubt that Parkinson's disease bility of minor trauma precipitating a chain of and other neurodegenerative diseases will J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from biochemical events that result in cell death. increase in importance over the next few The association between rural residence, well decades as the population continues to age and water consumption, and Parkinson's disease many of the other important causes of morbid- initially seemed to be a consistent and poten- ity continue to decline.'65 Epidemiological tially important clue. However, the absence of research on Parkinson's disease has flourished any association in a population based case-con- over the past 10 years, but many findings are trol study is a serious concern as prior results still tentative or inconsistent. Future descrip- may have been influenced by selection bias. tive studies should focus on populations with This finding needs confirmation by further either anomalous high or low rates of suitable studies. Population based prevalence Parkinson's disease. High quality temporal studies comparing stable urban and rural popu- data is essential to determine whether inci- lations in the same country will definitively dence rates are heterogenous depending on age answer whether there is any excess risk associ- and sex. Better data on elderly cohorts will ated with residence in a rural area. The role of clarify whether incidence of Parkinson's dis- dietary factors is still unclear as current evi- ease increases exponentially with age. Specific dence is weak. More complex methods of exposures need to be tested more rigorously, dietary measurement using prospective cohorts either by using special "at risk" cohorts, or by would establish whether these results are arte- improving measurement of exposure, such as factual. prospective cohorts with reliable measures of dietary intake. Our growing understanding of both the basic pathology and the possible role Comorbidity of genetic factors suggests that certain sub- The association between Parkinson's disease groups ofthe population may be more suscepti- and other diseases can be a useful clue as to ble to potential environmental factors. shared common aetiological factors, be they Epidemiologists will need to collaborate with genetic or environmental. laboratory based researchers to disentangle the Cardiovascular disease, hypertension, and diabetes relative importance of each risk factor and Heart disease and stroke have been reported to examine for potential interactions. be less common in male patients with I thank the two anonymous referees for useful comments on the Parkinson's disease72 or to be no different from initial version of this paper. Much of my knowledge about the the general population.'57 Most studies which epidemiology of Parkinson's disease was acquired while funded have reported causes of death for patients with by a Weilcome Trust fellowship in clinical epidemiology. Parkinson's disease have used proportional 158159 1 Parkinson J. An essay on the shaking palsy. London: mortality50 and have therefore potentially Sherwood, Neely, and Jones, 1817:1-66 underestimated the relevance of heart disease 2 Koller WC, Wong GF, Lang A. Posttraumatic movement the disorders: a review. Mov Disord 1989;4:20-36. compared with general population.7' One 3 Factor SA, Sanchez Ramos J, Weiner WJ. Trauma as an cohort study has reported a twofold to three- etiology of parkinsonism: a historical review of the con- fold increased risk for cept. Mov Disord 1988;3:30-6. both heart disease and 4 Stern MB. Head trauma as a risk factor for Parkinson's stroke compared with normal controls.49 disease. Mov Disord 1991;6:95-7. http://jnnp.bmj.com/ 5 Khoury MJ, Beaty TH, Cohen BH. Fundamentals ofgenetic However, this has not been replicated.52 epidemiology. New York: Oxford University Press, 1993: Hypertension has either been less common99 157 3-383 or no different 79 80 6 Sasco AJ, Paffenbarger RS. Measles infection and from control groups.72 This Parkinson's disease. AmJ Epidemiol 1985;122: 1017-31. might relate to abnormalities in autonomic 7 Ben-Shlomo Y, Sieradzan K. Idiopathic parkinson's dis- ease: epidemiology, diagnosis and management. BrJ Gen control mechanisms. No increased risk has Pract 1995;45:261-8. been seen for diabetes79 80160 despite the experi- 8 Mutch WJ, Dingwall-Fordyce I, Downie AW, Paterson mental finding that fructose blocks MPTP tox- JG, Roy SK. Parkinson's disease in a Scottish city. BMJ 1986;292:534-6. on October 2, 2021 by guest. Protected copyright. icity in isolated hepatocytes.'6' 9 Harada H, Nishikawa S, Takahashi K. Epidemiology of Parkinson's disease in a Japanese city. Arch Neurol 1983; Malignancy 40:151-4. Several studies have shown that patients 10 Martilla RJ, Rinne UK. Epidemiology of Parkinson's dis- with Parkinson's ease in Finland. Acta Neurol Scand 1976;53:81-102. disease experience about half 11 Hughes AJ, Daniel SE, Kilford L, Lees AJ. Accuracy of the number of cancers than would be expected clinical diagnosis of idiopathic Parkinson's disease: a from general 159 162-164 clinico-pathological study of 100 cases. J Neurol population rates.49 Neurosurg Psychiatry 1992;55: 181-4. When this is examined in more detail, this 12 Hughes AJ, Ben-Shlomo Y, Daniel SE, Lees AJ. What fea- deficit is tures improve the accuracy of clinical diagnosis in specifically related to smoking related Parlinson's disease: a clinicopathological study. cancers49 162 and is simply explained by the Neurology 1992;42:1142-6. larger proportion of in 13 Riggs JE. Smoking and Alzheimer's disease: protective never smokers the effect or differential survival bias? Lancet 1993;342: Parkinson's disease population. 793-4. 14 Mattock C, Marmot M, Stern G. Could Parkinson's dis- ease follow intra-uterine influenza?: a speculative hypoth- COMMENT esis. JNeurol Neurosurg Psychiatry 1988;51:753-6. Patients with 15 Eldridge R, Ince SE. The low concordance rate for Parkinson's disease have a differ- Parkinson's disease in twins: a possible explanation. ent comorbidity profile than normal controls. Neurology 1984;34:1354-6. The risk of 16 Calne DB, Langston JW. Aetiology of Parkinson's disease. cancers is lower as smoking related Lancet 1983;2: 1457-9. malignancies are less common. This is simply 17 Langston JW. Predicting Parkinson's disease. Neurology explained by the lower 1990;40:70-4. prevalence rates of 18 Koller WC, Langston JW, Hubble JP, et al. Does a long smoking. It is unclear whether the risk of car- preclinical period occur in Parkinson's disease?. diovascular Neurology 1991;41:8-13. disease is increased in patients with 19 Koller WC. When does Parkinson's disease begin? Parkinson's disease. Neurology 1992;42 (suppl 4):27-31. 14 Ben-Shlomo

20 Gibb WR, Lees AJ. The progression of idiopathic 54 Brown EL, Knox EG. Epidemiological approach to Parkinson's disease is not explained by age-related Parkinson's disease. Lancet 1972;l:974-6. effects of changes. Clinical and pathological comparisons with 55 Holford TR. Analysing the temporal age, period J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from post-encephalitic parkinsonian syndrome. Acta and cohort. Statistical Methods in Medical Research Neuropathol (Berl) 1987;73:195-201. 1992;1:317-37. 21 Schulzer M, Lee CS, Mak EK, Vingerhoets JG, Calne DB. 56 Ravenholt RT, Foege WH. 1918 Influenza, encephalitis A mathematical model of pathogenesis in idiopathic lethargica, parkinsonism. Lancet 1982;ii:860-4. parkinsonism. Brain 1994;117:509-16. 57 Kurland LT, Hauser WA, Okazaki H, Nobrega FT. 22 Goldberg ID, Kurland LT. Mortality in 33 countries from Epidemiologic studies of Parkinsonism with special refer- diseases of the nervous system. World Neurology 1962; ence to the cohort hypothesis. In: Proceedings of the third 3:444-63. symposiumn on Parkinsonism. Edinburgh: Livingstone, 23 Williams GR. Morbidity and mortality with parkinsonism. _7 1969:12-6. Neurosurg 1966;24 (suppl):138-43. 58 Nobrega FT, Glattre E, Kurland LT, Okazaki H. 24 Zhang Z, Roman GC. Worldwide occurence of Comments on the epidemiology of parkinsonism includ- Parkinson's disease: an updated review. ing prevalence and incidence statistics for Rochester, Neuroepidemiology 1993;12: 195-208. Minnesota, 1953-1966. In: Barbeau A, Brunette JR, eds. 25 Hayward RA, Shapiro MF, Freeman HE, Corey CR. Progress in Neurogenetics. Amsterdam: Excerpta Medica, Inequities in health services among insured Americans. 1969:474-85. 59 Offord Beard CM, Kurland LT. N Engl _J Med 1988;318:1507-12. Rajput AH, KP, 26 Schoenberg BS, Anderson DW, Haerer AF. Prevalence of Epidemiology of parkinsonism: incidence, classification Parkinson's disease in the biracial population of Copiah and mortality. Ann Neurol 1984;16:278-82. County, Mississippi. Neurology 1985;35:841-5. 60 de Pedro-Cuesta J, Stawiarz L. Parkinson's disease inci- 27 de Pedro-Cuesta J. Parkinson's disease diagnostic criteria, dence: magnitude, comparability, time trends. Acta incidence and etiology: III Europarkinson Workshop. Neurol Scand 199 1;84:382-8. Madrid: Institute of Health, 1991. 61 Ritchie K, Kildea D. Is senile dementia "age-related" 28 Wang SJ, Fuh JL, Liu CY, et al. Parkinson's disease in or ageing-related"?-evidence from meta-analysis of Kin-Hu, Kinmen: a community survey by neurologists. dementia prevalence in the oldest old. Lancet 1995;346: Neuroepidemiology 1994;13:69-74. 931-4. 29 Jendroska K, Olasode Bi, Daniel SE, et al. Incidental Lewy 62 Larsen JP and the Norwegian Study Group of Parkinson's body disease in black Africans. Lancet 1994;344:882-3. disease in the elderly. Parkinson's disease as community 30 Ben-Shlomo Y, Wenning G. Incidental Lewy body dis- health problem: study in Norwegian nursing homes. ease. Lancet 1994;344:1503. BMJ 1991;303:741-3. 31 Tanner CM, Chen B, Wang W, et al. Environmental fac- 63 Li S, Schoenberg BS, Wang C, et al. A prevalence survey tors and Parkinson's disease: a case-control study in of Parkinson's disease and other movement disorders in China. Neurology 1989;39:660-4. the People's Republic of China. Arch Neurol 1985;42: 32 Marmot MG, Adelstein AM, Bulusu L. Lessons from the 655-7. study of immigrant mortality. Lancet 1984;i: 1455-8. 64 Morgante L, Rocca WA, Di Rosa AE, et al. Prevalence of 33 Gudmundsson KR. A clinical survey of Parkinsonism in Parkinson's disease and other types of parkinsonism: a Iceland. Acta Neurol Scand 1967;43 (suppl 33):9-61. door-to-door survey in three Sicillian municipalities. 34 de Pedro-Cuesta J, Petersen IJ, Stawiarz L, et al. High lev- Neurology 1992;42:1901-7. Bhise odopa use in periodically time-clustered, Icelandic birth 65 Bharucha NE, Bharucha EP, Bharucha AE, AV, cohorts. A vestige of parkinsonism etiology? Acta Neurol Schoenberg BS. Prevalence of Parkinson's disease in the Scand 1995;91:79-88. Parsi community of Bombay, India. Arch Neurol 1988; 35 Manyam BV. Paralysis agitans and levodopa in 45:1321-3. "Ayurveda": ancient Indian medical treatise. Mov Disord 66 Tison F, Dartigues JF, Dubes L, Zuber M, Alperovitch A, 1990;5:47-8. Henry P. Prevalence of Parkinson's disease in the elderly: 36 Kurtzke JF, Murphy FM. The changing patterns of death a population study in Gironde, France. Acta Neurol rates in parkinsonism. Neurology 1990;40:42-9. Scand 1994;90:111-5. 37 Treves TA, de Pedro-Cuesta J. Parkinsonism mortality in 67 Liberatos P, Link BG, Kelsey JL. The measurement of the US, 1. Time and space distribution. Acta Neurol social class in epidemiology. Epidemiol Rev 19881O: Scand 1991;84:389-97. 87-121. 38 Lilienfeld DE, Chan E, Ehland J, et al. Two decades of 68 Marmot MG, Shipley MJ, Rose G. Inequalities in death- increasing mortality from Parkinson's disease among the specific explanations of a general pattern? Lancet US elderly. Arch Neurol 1990;47:731-4. 1984;i: 1003-6. 39 Duvoisin RC, Schweitzer MD. Paralysis agitans mortality 69 Office of Population Census and Surveys. Classificationi of in England and Wales, 1855-1962. Br Y Prev Soc Med occupations. London: HMSO, 1980 1966;20:27-33. 70 McKeigue PM, Marmot MG. Epidemiology of 40 Marmot MG. Parkinson's disease and encephalitis: the Parkinson's disease. In: Stern G, ed. Parkinson's disease. cohort hypothesis re-examined. In: Rose FC, ed. London: Chapman and Hall, 1989:295-306. Clinical neuroepidemiology. Kent: Pitman Medical, 1980; 71 Breslow NE, Day NE. Statistical methods in cancer research: 391-401. Volume 2-the design and analysis of cohort studies. Lyon: 41 Li T, Swash M, Alberman E. Morbidity and mortality in IARC: 1987. disease. http://jnnp.bmj.com/ motor neuron disease: comparison with multiple sclero- 72 Kessler II. Epidemiologic studies of Parkinson's sis and Parkinson's disease: age and sex specific rates and III. A community based survey. Am _7 Epidemiol 1972; cohort analyses. Neurol Neurosurg Psychiatry 1985;48: 96:242-54. 320-7. 73 Martilla RJ, Rinne UK. Arteriosclerosis, hereditary and 42 Clarke CE. Mortality from Parkinson's disease in England some previous infections in the aetiology of Parkinson's and Wales 1921-1989. Neurol Neurosurg Psychiatry disease. A case control study. Clin Neurol Neurosurg 1993;56:690-3. 1 976;79:46-56. 43 Ben-Shlomo Y, Finnan F, Allwright S, Davey Smith G. 74 Stern M, Dulaney E, Gruber SB, et al. The epidemiology The epidemiology of Parkinson's disease in Ireland: of Parkinson's disease: a case-control study of young- observations from routine data sources. Ir Med _7 onset and old-onset patients. Arch Neurol 1991;48: 1993;86: 190-4. 903-7. 44 Vanacore N, Bonifati V, Bellatreccia A, Edito F, Meco G. 75 Semchuck KM, Love EJ, Lee RG. Parkinson's disease on October 2, 2021 by guest. Protected copyright. Mortality rates for Parkinson's disease and parkinsonism and exposure to agricultural work and pesticide chemi- in Italy (1969-1987). Neuroepidemiology 1992;11:65-73. cals. Neurology 1992;42:1328-35. 45 Kurtzke JF, Flaten TP, Murphy FM. Death rates from 76 Mayeux R, Tang MX, Marder K, Cote LJ, Stern Y. Parkinson's disease in Norway reflect increased survival. Smoking and Parkinson's disease. Mov Disord 1994;9: Neurology 199 1;41:1665-7. 207-12. 46 Imaizumi Y, Kaneko R. Rising mortality from Parkinson's 77 Butterfield PG, Valanis BG, Spencer PS, Lindeman CA, disease in Japan, 1950-1992. Acta Neurol Scand 1995; Nutt JG. Environmental antecedents of young- 91:169-76. onset Parkinson's disease. Neurology 1993;43: 47 Clarke CE. Does levodopa therapy delay death in 1150-8. Parkinson's disease? A review of the evidence. Mov 78 Bharucha NE, Stokes L, Schoenberg BS, et al. A case- Disord 1995;1O:250-6. control study of twin pairs discordant for Parkinson's 48 Ebmeier KP, Calder SA, Crawford JR, Stewart L, Beeson disease: a search for environmental risk factors. Mutch Parkinson's disease in Aberdeen: sur- Neurology 1986;36:284-8. JAO, WJ. D. vival after 3-5 years. Acta Neurol Scand 1990;81:294-9. 79 Grandinetti A, Morens DM, Reed D, MacEachern 49 Ben-Shlomo Y, Marmot MG. Survival and cause of death Prospective study of cigarette smoking and the risk of Parkinson's disease. Am in a cohort of patients with parkinsonism: possible clues to developing idiopathic _7 129-38. aetiology. .7 Neurol Neurosurg Psychiatry 1995;58:293-9. Epidemiol 1994;139:1 50 Hoehn MM, Yahr MD. Parkinsonism: onset, progression, 80 Ho SC, Woo J, Lee CM. Epidemiologic study of and mortality. Neurology 1967;17:427-42. Parkinson's disease in Hong Kong. Neurology 1989;39: 51 Riggs JE, Schochet SS. Rising mortality due to Parkinson's 1314-8. disease and amyotrophic lateral sclerosis: a manifestation 81 Hubble JP, Cao T, Hassanein RE, Neuberger JS, Koller WC. Risk factors for Parkinson's disease. Neurology of the competitive nature of human mortality. Clin Epidemiol 1992;45:1007-12. 1 993;43: 1693-7. 52 Bennett DA, Beckett LA, Murray AM, et al. Prevalence of 82 Wong GF, Gray CS, Hassanein RS, Koller WC. Parkinsonian signs and associated mortality in a commu- Environmental risk factors in siblings with Parkinson's disease. Arch nity population of older people. N Engl _J Med 1996; Neurol 1991;48:287-9. 334:71-6. 83 Koller W, Vetere-Overfield B, Gray C, et al. Environ- 53 Poskanzer DC, Schwab RS. Cohort analysis of Parkinson's mental risk factors in Parkinson's disease. Neurology syndrome. Evidence for a single etiology related to 1990;40:1218-21. subclinical infection about 1920. _7 Chron Dis 1963;16: 84 Barbeau A, Roy M, Bernier G, Campanella G, Paris S. 961-73. Ecogenetics of Parkinson's disease: prevalence and envi- How far are we in understanding the cause ofParkinson's disease? 15

ronmental aspects in rural areas. Can -7 Neurol Sci Parkinson's disease? A critical review. Neurology 1990; 1987;14:36-41. 40:suppl 30. 85 Golbe LI, Farrell TM, Davis PH. Follow-up study of 116 Sanchez-Ramos JR, Hefti F, Weiner WJ. Paraquat and J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from early-life protective and risk factors in Parkinson's dis- Parkinson's disease. Neurology 1987;37:728. ease. Mov Disord 1990;5:66-70. 117 Blair A, Zahm SH. Pattems ofpesticide use among farmers: 86 Ohlson C, Hogstedt C. Parkinson's disease and occupa- implications for epidemiologic research. Epidentiology tional exposure to organic solvents, agricultural chemi- 1993;4:55-62. cals and mercury-a case-referent study. Scand_7 Work 118 Stephens R, Spurgeon A, Calvert IA, et al. Neuro- Environ Health 1981;7:252-6. psychological effects of long-term exposure to organ- 87 Jimenez-Jimenez FJ, Mateo D, Gimenez-Roldan S. ophosphates in sheep dip. Lancet 1995;345:1135-9. Exposure to well water and pesticides in Parkinson's dis- 119 Robins AH. Depression in patients with parkinsonism. BrJ ease: a case-control study in the Madrid area. Mov Disord Psychiatry 1976;128:141-5. 1992;7: 149-52. 120 Cummings JL. Depression and Parkinson's disease: a 88 Ngim C, Devathasan G. Epidemiologic study on the asso- review. Am .7 Psychiatry 1992;149:443-54. ciation between body burden mercury level and idio- 121 Yesalis CEI, Lemke JH, Wallace RB, Kohonut FJ, Morris pathic Parkinson's disease. Neuroepidemiology 1989;8: MC. Health status of the rural elderly according to farm 128-41. work history: the Iowa 65+ rural health study. Arch 89 Aquilonius SM, Hartvig P. A Swedish county with unex- Environ Health 1985;40:245-50. pectedly high utilization of anti-parkinsonian drugs. Acta 122 Tanner C. Abnormal liver enzyme-mediated metabolism Neurol Scand 1986;74:379-82. in Parkinson's disease: a second look. Neurology 90 Wang WZ, Fang XH, Cheng XM, Jiang DH, Lin ZJ. A 1991;41:89-91. case-control study on the environmental risk factors of 123 Steventon GB, Heafield MTE, Waring RH, Williams AC. Parkinson's disease in Tianjin, China. Neuroepidemiology Xenobiotic metabolism in Parkinson's disease. Neurology 1993;12:209-18. 1989;39:883-7. 91 Ebmeier KP, Mutch WJ, Calder SA, Crawford JR, Stewart 124 Steventon GB, Heafield MTE, Sturman SG, Waring RH, L, Besson JO. Does idiopathic Parkinsonism in Williams AC, Ellingham J. Degenerative neurological Aberdeen follow intrauterine influenza? _7 Neurol disease and debrisoquine-4-hydroxylation capacity. Neurosurg Psychiatry 1989;52:911-3. Medical Science Research 1989;17:163-4. 92 Baumann RJ, Jameson HD, McKean HE, Haack DG, 125 Armstrong M, Daly AK, Cholerton S, Bateman DN, Idle Weisberg LM. Cigarette smoking and Parkinson disease: JR. Mutant debrisoquine hydroxylation genes in 1. A comparison of cases with matched neighbours. Parkinson's disease. Lancet 1992;339:1017-8. Neurology 1980;30:839-43. 126 Smith CAD, Gough AC, Leigh PN, et al. Debrisoquine 93 Godwin-Austen RB, Lee PN, Marmot MG, Stem GM. hydroxylase gene polymorphism and suceptibility to Smoking and Parkinson's disease. .7 Neurol Neurosurg Parkinson's disease. Lancet 1992;339: 1375-7. Psychiatry 1982;45:577-81. 127 Agid Y. Parkinson's disease: pathophysiology. Lancet 94 Kondo K. Epidemiological evaluations of risk factors in 199 1;337:1321-3. Parkinson's disease. Adv Neurol 1987;45:289-93. 128 Elizan TS, Madden DL, Noble GR, et al. Viral antibodies in 95 Semchuk KM, Love EJ, Lee RG. Parkinson's disease: a serum and CSF of parkinsonian patients and controls. test of the multifactorial etiologic hypothesis. Neurology Arch Neurol 1979;36:529-34. 1993;43:1173-80. 129 Martilla RJ, Arstila P, Nikostelainen J, Halonen PE, Rinne 96 Nefzger MD, Quadfasel FA, Karl VC. A retrospective UK. Viral antibodies in the sera from patients with study of smoking in Parkinson's disease. Am _7 Epidemiol Parkinson's disease. Eur Neurol 1977;15:25-33. 1968;88: 149-58. 130 Schwartz J, Elizan TS. Search for viral particles and virus- 97 Martilla RJ, Rinne UK. Smoking and Parkinson's disease. specific products in idiopathic Parkinson disease brain Acta Neurol Scand 1980;62:322-25. material. Ann Neurol 1979;6:261-3. 98 Haack DG, Baumann RJ, McKean HE, Jameson HD, 131 Martyn CN, Cruddas M, Compston DA. Symptomatic Turbek JA. Nicotine exposure and Parkinson disease. Epstein-Barr virus infection and multiple sclerosis. .7 AmJEpidemiol 1981;114:191-9. Neurol Neurosurg Psychiatry 1993;56: 167-8. 99 Rajput AH, Offord KP, Beard CM, Kurland LT. A case- 132 Riggs JE. Cigarette smoking and Parkinson disease: the control study of smoking habits, dementia, and other ill- illusion of a neuroprotective effect. Clin Neuropharmacol nesses in idiopathic Parkinson's disease. Neurology 1992;15:88-99. 1987;37:226-32. 133 Ben-Shlomo Y. Smoking and neurodegenerative diseases. 100 Jimenez Jimenez FJ, Mateo D, Gimenez Roldan S. Lancet 1993;342: 1239. Premorbid smoking, alcohol consumption, and coffee 134 Yong VW, Perry TL. Monoamine oxidase B, smoking and drinking habits in Parkinson's disease: a case-control Parkinson's disease. _7 Neurol Sci 1986;72:265-72. study. Mov Disord 1992;7:339-44. 135 Jenner P. Oxidative damage in neurodegenerative disease. 101 Kahn HA. The Dom study of smoking and mortality Lancet 1994;344:796-8. among U.S. Veterans: report on eight and one-half 136 Smith PR, Cooper JM, Govan GG, Harding AE, Schapira years of observation. Nat Cancer Inst Monogr 1966;19: AHV. Smoking and mitochondrial function: a model for 1-125. environmental toxins. QJ Med 1993,86:657-60. 102 Hammond EC. Smoking in relation to the death rates of 137 Gresham LS., Molgaard CA, Smith RA. Induction of a one million men and women. In: Haenzel W, ed. cytochrome P-450 enzymes via tobacco smoke: potential http://jnnp.bmj.com/ Epidemiological approaches to the study of cancer and other mechanism for developing resistance to environmental chronic diseases. Bethesda, MD: National Cancer toxins as related to parkinsonism and other neurologic Institute, 1966:127-204. (National Cancer Institute diseases. Neuroepidemiology 1993;12: 114-6. Monograph No 19.) 138 Golbe LI, Cody RA, Duvoisin RC. Smoking and 103 Doll R, Peto R, Wheatley K, Gray R, Sutherland I. Parkinson's disease: search for a dose-response relation- Mortality in relation to smoking: 40 years' observations ship. Arch Neurol 1986;43:774-80. on male British doctors. BM. 1994;309:901-1 1. 139 Eysenck HJ, Tarrant M, Woolf M, England L. Smoking 104 Sasco AJ, Paffenbarger RSJ. Smoking and Parkinson's dis- and personality. BM,7 1960;i:456-60. ease. Epidemiology 1990;1:460-5. 140 Forgays DG, Bonaiuto P, Wrzesniewski K, Forgays DK. 105 Poewe W, Karamat E, Kemmler GW, Gerstenbrand F. Personality and cigarette smoking in Italy, Poland and The of with Parkinson's the United States. 1993;28:399-413. premorbid personality patients Int.7Addict on October 2, 2021 by guest. Protected copyright. disease: a comparative study with healthy controls and 141 Plassman BL, Helms MJ, Welsh KA, Saunders AM, patients with essential tremor. Adv Neurol 1990;53: Breitner JCS. Smoking, Alzheimer' disease and con- 339-42. founding with genes. Lancet 1995;345:387. 106 Eatough VM, Kempster PA, Stem GM, Lees AJ. 142 Godwin Austen RB, Lee PN, Marmot MG, Stem GM. Premorbid personality and idiopathic Parkinson's dis- Smoking and Parkinson's disease. _7 Neurol Neurosurg ease. Adv Neurol 1990;53:335-7. Psychiatry 1982;45:577-81. 107 Ward CD, Duvoisin RC, Ince SE, Nutt JD, Eldridge R, 143 Raphael K. Recall bias: a proposal for assessment and con- Caine DB. Parkinson's disease in 65 pairs of twins and in trol. Int_. Epidemiol 1987;16:167-70. a set of quadruplets. Neurology 1983;33:815-24. 144 Weiss N. Should we consider a subject's knowledge of the 108 Williams DB, Annegers JF, Kokmen E, O'Brien PC, etiologic hypothesis in the analysis of case-control stud- Kurland LT. Brain injury and neurologic sequelae: a ies? AmJ Epidemiol 1994;139:247-9. cohort study of dementia, parkinsonism, and amy- 145 Rajput AH, Utti RJ, Stem W, Laverty W. Early onset otrophic lateral sclerosis. Neurology 1991;41:1554-7. Parkinson's disease in Saskatchewan-environmental 109 Semchuck KM, Love EJ, Lee RG. Parkinson's disease and considerations for etiology. Can .7 Neurol Sci 1986;13: exposure to rural environmental factors: a population 312-6. based case-control study. Can _7 Neurol Sci 1991;18: 146 Martyn CN. Choosing cases and controls in neurological 279-86. research. .7 Neurol Neurosurg Psychiatry 1990;53:453-4. 110 Golbe LI, Farrell TM, Davis PH. Case-control study of 147 Blazer DG, Landerman LR, Fillenbaum G, Homer R. early life dietary factors in Parkinson's disease. Arch Health services access and use among older adults in Neurol 1988;45:1350-3. North Carolina: urban vs rural residents. Am.7 Public 111 Tanner CM, Chen B, Cohen JA, et al. Dietary antioxidant Health 1995;85:1384-90. vitamins and the risk of developing Parkinson's disease. 148 Rybicki BA, Johnson CC, Gorell JM. Demographic differ- _Neurology 1989;39: 181 ences in referral rates to neurologists of patients with sus- 112 Cerhan AR, Wallace RB, Folsom AR. Antioxidant intake pected Parkinson's disease: implications for case-control and risk of Parkinson's disease (PD) in older women. Am study design. Neuroepidemiology 1995;14:72-81. JEpidemiol 1994;139:S65. 149 Schlesselman nI. Case-control studies: design, conduct, analysis. 1 3 Lang AE, Marsden CD, Obeso JA, Parkes JD. Alcohol and New York: Oxford University Press, 1982:1-354. Parkinson disease. Ann Neurol 1982;12:254-6. 150 Granieri E, Carreras M, Casetta I, et al. Parkinson's dis- 114 Langston JW, Ballard P, Tetrud JW, Irwin I. Chronic ease in Ferrara, Italy, 1967 through 1987. Arch Neurol parkinsonism in humans due to a product of meperidine- 1991 ;49:854-7. analog synthesis. Science 1983,219:970-80. 151 Phillips AN, Davey Smith G. How independent are "inde- 1 15 Tanner CM, Langston JW. Do environmental toxins cause pendent" effects? Relative risk estimation when corre- 16 Ben-Shlomo

lated exposures are measured imprecisely. Clin 159 Marmot MG. Mortality and Parkinson's disease. In: Rose Epidemiol 1991 ;44: 1223-31. FC, Capildeo R, eds. Research progress in Parkinson's dis- 152 Rajput AH, Uitti RJ, Stem W, et al. Geography, drinking ease. London: Pitman, 1988;9-16. J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.61.1.4 on 1 July 1996. Downloaded from water chemistry, pesticides and herbicides and the etiology 160 Rajput AH, Offord KP, Beard CM, Kurland LT. A case- of Parkinson's disease. Can Neurol Sci 1987;14:414-8. control study of smoking habits, dementia, and other ill- 153 Spencer P. Guam ALS/Parkinsonism-dementia: a long- nesses in idiopathic Parkinson's disease. Neurology 1987; latency neurotoxic disorder caused by "slow toxin(s)" in 37:226-32. food? Can Neurol Sci 1987;14:347-57. 161 Di Monte D, Sandy MS, Blank L, Smith MT. Fructose 154 Willett W. Overview of nutritional epidemiology. In: prevents 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine Willett W, ed. Nutritional epidemiology. New York: (MPTP)-induced ATP depletion and toxicity in isolated Oxford University Press, 1990;3-19. hepatocytes. Biochem Biophys Res Commun 1988;153: 155 Dwyer JT, Gardner J, Halvorsen K, Krall EA, Cohen A, 734-40. Valadian I. Memory of food intake in the distant past. 162 Moller H, Mellemkjaer L, McLaughlin JK, Olsen JH. Am Epidemiol 1989;130: 1033-46. Occurence of different cancers in patients with 156 The Parkinson Study Group. Effects of tocopherol and Parkinson's disease. BMJ 1995;310:1500-1. deprenyl on the progression of disability in early 163 Barbeau A, Joly JG. Parkinson et cancer. Union Med Can Parkinson's disease. N EnglJ Med 1993;328: 176-83. 1963;92: 169-74. 157 Marttila RJ, Rinne UK. Arteriosclerosis, heredity, and 164 Jansson B, Jankovic J. Low cancer rates among patients some previous infections in the etiology of Parkinson's with Parkinson's disease. Ann Neurol 1985;17:505-9. disease. A case-control study. Clin Neurol Neurosurg 165 Barrett-Connor E. Are we living longer or dying longer? In: 1976;79:46-56. Poulter N, Sever P, Thom S, eds. Cardiovascular disease: 158 Hoehn MMM. Parkinson's disease: progression and mor- risk factors and intervention. Oxford: Radcliffe Medical tality. Adv Neurol 1986;45:457-61. Press, 1993:89-99. http://jnnp.bmj.com/ on October 2, 2021 by guest. Protected copyright.