|Home | About | Journals | Submit | Contact Us | Français|
Epidemiological evidence of an association between Alzheimer's disease (AD) and the most frequently studied occupational exposures—pesticides, solvents, electromagnetic fields (EMF), lead and aluminium—is inconsistent. Epidemiological studies published up to June of 2003 were systematically searched through PubMed and Toxline. Twenty‐four studies (21 case–control and 3 cohort studies) were included. Median GQI was 36.6% (range 19.5–62.9%). Most of the case–control studies had a GQI of <50%. The study with the highest score was a cohort study. Likelihood of exposure misclassification bias affected 18 of the 24 studies. Opportunity for bias arising from the use of surrogate informants affected 17 studies, followed by disease misclassification (11 studies) and selection bias (10 studies). Eleven studies explored the relationship of AD with solvents, seven with EMF, six with pesticides, six with lead and three with aluminium. For pesticides, studies of greater quality and prospective design found increased and statistically significant associations. For the remaining occupational agents, the evidence of association is less consistent (for solvents and EMF) or absent (for lead and aluminium).
Alzheimer's disease (AD) is the most common cause of dementia in the elderly, accounting for 60–70% of the cases of progressive cognitive impairment. The prevalence of AD is up to 40% in those aged 85 years and older. The population of patients with AD will nearly quadruple in the next 50 years if the current trend continues.1 The diagnosis of this disease is considered probable when other alternative causes of dementia have been excluded, but only necropsy allows a definitive diagnosis of AD.2,3
Several risk factors for AD have been identified in epidemiological studies in addition to age and female sex. The strongest and most consistent risk factor is the apolipoprotein E genotype epsilon 4 allele (APOE4). Other risk factors evaluated include head injury, low serum levels of folate and vitamin B12, raised plasma and total homocysteine levels, a family history of AD or dementia, fewer years of formal education, lower income and lower occupational status.1 The evidence for increased risk of AD for occupational exposures is generally not consistent.4,5 The most widely studied occupational agents have been pesticides, solvents, electromagnetic fields (EMF), lead and aluminium.
The lack of evidence between AD and occupational exposures might be explained by the problem of validity, given the presence of characteristic biases in epidemiological studies on AD such as the bias derived from use of surrogate informants in retrospective studies, as the cognitive state of the patients makes it necessary to gather relevant information from the family or close friends, and diagnosis misclassification bias due to the difficulties of differential diagnosis for AD.6,7,8
Epidemiological research of occupational risks for AD has also frequent precision problems, because occupational exposures of interest are relatively uncommon and large studies would be required to show even relatively moderate risks. In 1991, under the assumption of an insufficient sample size, 11 case–control studies were reanalysed to assess with increased statistical power potential risk factors, including environmental factors. However the heterogeneity of the studies prevented the pooling of data.9,10
Different reanalysis and meta‐analysis of observational studies have been carried out, but as far as we know, except for the reanalysis mentioned above, none into occupational risk factors and AD. This study aimed at assessing, with a standardised and systematic approach, the strength of the associations between AD and pesticides, solvents, EMF, lead and aluminium in the workplace, evaluating the quality of published studies.
Epidemiological studies on the association between AD and occupational exposure were located through electronic searches on PubMed and Toxline and further searching the references of relevant articles found.
The search was carried out in June 2003. For the search in PubMed a combination of the MeSH terms “Occupational exposure” and “Alzheimer disease” was used, with a further search with “Alzheimer*” and “occupatio*” in free text. In Toxline “Alzheimer*” and “occupatio*” in free text were used as searching terms. In both databases no limit was applied to the search strategy. Two hundred and forty‐one references were obtained in PubMed and 199 in Toxline. A first selection of relevant articles was made, including all epidemiological studies with individualised data, written in English, Spanish, French or Italian, in which it was possible to calculate measurements of relative risk for AD between those exposed at least once and those never exposed. Therefore ecological studies or studies focusing exclusively on aetiopathogenic mechanisms were excluded. Studies assessing environmental exposures which did not occur in the workplace were also excluded. Only original articles assessing specific exposures to those most widely studied agents (pesticides, solvents, EMF, aluminium and lead) were considered. Papers in which exposure was related to the starting age of AD or the evolution of the disease, but not to its aetiology, were also excluded. Also, only original papers, in which the effect analysed was a specific diagnosis of AD, were included. The clinical diagnosis of AD was based, therefore, on the application of the criteria of the National Institute of Neurologic and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS‐ADRDA), the Diagnostic and Statistical Manual for mental disorders, revised 3rd and 4th editions (DSM III‐R, DSM IV), and the International Classification of Diseases, 9th and 10th revisions (ICD‐9, ICD‐10) or equivalent criteria. Hence, studies not applying diagnostic criteria of AD or studies limited to the assessment of cognitive impairment or presenile dementia were also excluded.
Inclusion and exclusion criteria were applied to the retrieved references, by reading the abstracts or, when necessary, full paper. We found two studies which had carried out reanalysis of previous data. The first one10 was conducted with data from four previously published studies.11,12,13,14 This reanalysis was excluded, as the original articles complying with our inclusion criteria were already included in our selection. The second reanalysis15 was based on three independent studies, unpublished at the time. The quality of these studies was therefore analysed separately in our review.
If there was more than one publication of the same study, we included the most recent, provided that it included the information of the previous studies. If these related papers presented different aspects of the study they were all selected but a note was made in the description explaining that all came from the same research.
A specially designed questionnaire was applied to each of the selected articles in order to assess the quality of each study and determine the presence of the main types of biases6,7,8 which might affect the results. On the basis of the design of the studies (cases and controls or cohorts) appropriate specific questionnaires were drawn up. The questionnaires were designed on the basis of protocols and questionnaires used previously with similar aims.16,17,18,19,20,21,22
Data collection followed the recommendations of Chalmers23 and Delgado‐Rodriguez and Sillero‐Arenas16 in order to minimise observer bias: each article was allocated an identification number and the details of the journal, authors and affiliations were removed. Every article was evaluated independently by two expert epidemiologists (FB and AMG). In cases of disagreement, the final evaluation was obtained by a consensus meeting between them.
The questionnaire for evaluation of the case–control studies contained 39 items measuring the quality of studies, with a maximum score of 111 points. These items were distributed in seven sections: (1) selection of cases and controls; (2) inclusion and exclusion criteria; (3) occupational exposure measurement; (4) control of confounding variables; (5) precision of the study; (6) internal and external validity of the study; and (7) general assessment of the presence or absence of biases (table 11 and Appendix I, available at http://ard.bmjjournals.com/supplemental). Thirty‐seven of these 39 items were distributed in five dimensions. Each dimension assessed a specific type of bias. Some of the items contributed to more than one dimension. The potential of the study for the presence of selection bias (17 items, maximum score of 44 points), disease misclassification (8 items, 23 points), exposure misclassification (11 items, 29 points), bias arising from the use of surrogate informants (7 items, 21 points) and misclassification bias of the confounding variables (5 items, 25 points) was analysed. Lower scores mean more potential for bias, while higher scores point to a smaller potential for bias in the study (table 22 and Appendix I, available at http://ard.bmjjournals.com/supplemental).
The questionnaire for cohort studies contained some common items with the questionnaire for case–control studies and specific items for prospective and retrospective cohorts, distributed in eight sections: (1) definition and follow‐up of the cohort; (2) follow‐up losses; (3) measurement of disease incidence; (4) occupational exposure measurement; (5) control of confounding variables; (6) precision of the study; (7) internal and external validity; (8) general assessment of the presence or absence of biases. Thirty items measure the quality of the prospective cohort studies with a maximum score of 90 points, and 28 items measure the quality in retrospective cohorts with a maximum score of 93 points (table 11 and Appendix II, available at http://ard.bmjjournals.com/supplemental). A total of 27 items (25 in retrospective cohorts) were distributed in four dimensions. The potential of the study for the presence of selection bias (14 items for prospective cohorts and 13 items for retrospective cohorts, with maximum scores of 40 and 43 points respectively), disease misclassification (8 items, 24 points, for prospective cohorts and 9 items, 31 points, for retrospective cohorts), exposure misclassification (7 items, 25 points, for prospective cohorts and 8 items, 32 points, for retrospective cohorts), and misclassification of the confounding variables (5 items, 25 points, for prospective cohorts and 7 items, 33 points, for retrospective cohorts) was analysed (table 22,, Appendix II, available at http://ard.bmjjournals.com/supplemental).
For each study, a Global Quality Index (GQI) was calculated according to the total sum of the points for each item mentioned above. As a result of the different number of items with different maximum scores, this index is presented as a percentage of the maximum possible value (100%) that each study can achieve (GQI= (score obtained/corresponding maximum score) ×100).
To determine the presence or absence of bias in each study, we calculated in the same way the percentage of the possible maximum score in each dimension which assessed each specific type of bias ((score obtained in the dimension/maximum score for that dimension) × 100). Then, these percentages were grouped in five categories for each particular dimension (referred to each particular source of bias): highly probable (when the percentage of the maximum score for that dimension was <20%), probable (20–40%), possible (>40–60%), improbable (>60–80%) and highly improbable (>80%).
Lastly, in the last section of the questionnaires the experts reviewing the studies should establish, for each bias identified as probable or highly probable, the pattern for the bias (non‐differential or differential for classification biases) and the effect or direction of the bias on the associations observed in the study (Appendices I and II, available at http://ard.bmjjournals.com/supplemental).
For the data management we used the statistical packet SPSS, version 11.0 and the Excel spreadsheet.
According to selection criteria, 22 original articles11,12,13,14,15,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40 (19 case–control studies and three cohort studies) were included. One of the articles, as mentioned above, is a reanalysis of three independent series of case–control studies on AD and EMF.15 Therefore, the questionnaires were finally applied to 24 original studies.
Table 33 presents a summary of the main methodological aspects of the studies analysed, including population studied, data collection period, assessment of exposure and control of confounding bias.
In Table 44 the results of the expert evaluation of the studies are presented. For the 24 studies, the median for the GQI was 36.6%. The article with the highest score reached a GQI of 62.9%. There was great variability in the quality of the different studies, with a range of 43.4% between the papers with the highest and lowest score. All the case–control studies but one27 showed a GQI below 50%. Five case–control studies scored below 25%, and the lowest score in the total sample was for a case–control study (GQI=19.4%). Quality in the three cohort studies was greater and more homogeneous than that seen in the case–control studies. The lowest value in the cohort studies corresponded to a prospective cohort study (GQI=50.5%).40
The most common potential bias is that of misclassification in the exposure, present in 18 of the 24 studies analysed (75.0%). The second in order of frequency is the potential bias arising from the use of surrogate informants, present in 12 of the 17 studies (70.6%). The third potential bias is that of misclassification of the disease, which appeared in 11 of the 24 studies (45.8%), followed by bias of selection present in 10 studies (41.7%). Confounding was considered the less frequent potential type of bias (fig 11).
In only one case, in bias arising from the use of surrogate informants, was it judged that the effect of bias might at least probably increase the association between AD and the assessed exposure, in this case to EMF. For the remaining studies the experts either judged that the observed association was probably underestimated or they failed to reach a conclusion about the effect of the potential biases under consideration. In 16 of the 18 studies affected by potential misclassification in the exposure (88.9%), it was judged that this effect could give rise to a non‐differential misclassification which would bias the associations towards the null (fig 11 and table 44).
In the reviewed papers, for the specific occupational exposures considered, 11 studies explored the relationship of AD with solvents, seven with electromagnetic fields (EMF), six with pesticides, six with lead and three with aluminium. Solvents and pesticides are the exposures with the highest number of high quality studies (five and four studies, respectively, with a score above the median GQI), followed by EMF (three studies) and lead and aluminium (two studies for each exposure) (table 44).
For pesticides, research of greater quality and prospective design found increased and statistically significant associations with AD. Tyas et al38 reported adjusted relative risk (aRR) of 4.35 (95% CI 1.05 to 17.90) for exposure to defoliants and fumigants (a smaller and non‐significant association was found for exposure to the wider category of “pesticides, fertilisers”: aRR=1.45, 95% CI 0.57–3.68) and Baldi et al40 found aRR for occupational exposure to pesticides in men of 2.39 (95% CI 1.02 to 5.63). The two case–control studies assessing risk associated with pesticide exposure and with GQI above the median26,37 found evidence of smaller and non‐significant associations, supporting the hypothesis that potential biases might have affected these results, decreasing the associations towards the null (table 44).). Finally, one of the remaining two case–control studies assessing exposure to pesticides30 found an unadjusted RR of 2.54 (95% CI 0.41 to 27.06) for organophosphates.
For the remaining occupational agents considered in this review the evidence of an association is less consistent. For solvents, only two out of the 11 studies analysing this exposure found a significant association with AD. The two studies focused on the same population base. The first27 is a high quality case–control study (the case–control study with the highest score for GQI), an example of proper selection and diagnostic bias control, where only the aRR in exposed men was significantly increased (aRR=6.3, 95% CI 2.2 to 18.1). The second paper32 included only cases with a spouse who was willing to collaborate in the interview, which might increase the likelihood of selection bias. With this restriction the aRR for solvent exposure during more than 18 years reached statistical significance (aRR=2.62, 95% CI 1.07 to 7.43), although it fell to 1.77 (95% CI 0.81 to 3.90) when exposure was classified as ever/never. However, a prospective cohort study,38 also assessing exposure to solvents and with a high quality ranking (GQI) in our evaluation, did not find association with this exposure focused in degreasers (aRR=0.88, 95% CI 0.31 to 2.50), and neither did the other two case–control studies with above the median GQI scores.24,26
There are three studies assessing risk for occupational exposure to EMF with high quality, well above that of the other studies,33,35,36 including the study with the highest GQI score in our ranking. However, the highest odds ratio (OR) values for this exposure correspond to the lower quality studies. Our analysis suggests that these studies are likely to be biased and that selection bias might explain these results.
For lead exposure there are no data supporting any association. All the studies are case–control studies, with a relatively low level of quality according to our classification. For aluminium, one of the three studies about this exposure is the second in the quality ranking of the case–control studies.28 Results from this study show no association (aRR=0.95, 95% CI 0.5 to 1.9). In the other two studies associations are also non‐significant (table 44).
The meta‐analysis protocol recommends evaluation of the quality of the primary studies included in the research16,41 through ad hoc developed questionnaires for the assessment. Recommendations have been proposed for quality assessment of observational studies.16,17,18,19,20,21,22 The recent initiative named STROBE (STrengthening the Reporting of OBservational studies in Epidemiology)42 should help to improve the quality of published epidemiological research, and some of the papers included in this review might have been improved if STROBE recommendations had been considered for their publication.
The quality of the reviewed studies was assessed with a blinded, standardised and systematic approach. Two epidemiologists independently reviewed all the studies and discrepancies were solved through consensus meetings. The percentage of global agreement between the two epidemiologists was 83.5% for all reviewed case–control studies, 93.3% for the two prospective cohort studies and 85.7% for the retrospective cohort study. Therefore, reproducibility was reasonably good.
Considering only the studies with higher quality, occupational exposure to pesticides is the risk for which, according to our analysis, there is the greatest evidence of association with AD. The quantitative synthesis of the data in a meta‐analysis, including studies with higher methodological quality, might enable a more accurate quantification of the size of this suggested potential risk. For the remaining occupational agents, the evidence of an association is less consistent. Contradictory results are found among studies assessing occupational exposure to solvents and EMF, and a lack of association in studies for lead and aluminium.
Valid assessment of exposure is always a problem in occupational epidemiology. Most of the studies evaluated did not have a good occupational exposure assessment, with too wide a range for exposure, which can cause no differential bias. Hence, it is necessary to measure occupational exposures of interest with increased specificity according to probability, intensity, and duration and time period of exposure allowing for latency periods for the disease and dose–response relationships. Also, solvents, pesticides and EMF are categories that have too wide an exposure, including exposures with highly different biological effects (e.g., benzene and toluene, organochlorines and organophosphates, radiofrequencies and extremely low‐frequency EMF). More specific definition of exposure should also be considered in research. Prospective cohort studies are a more adequate design for proper consideration of occupational exposure characteristics.
For evidence of an association for occupational exposures according to sex, 15 of the 22 articles included both men and women. In four of these 15 studies15,27,29,40 the results differed by sex. In the associations with pesticides40 and solvents,27 the strongest associations are found among men, which would be compatible with the hypothesis of an association between the exposure and AD, as it is probable that men perform activities with higher exposure to the occupational agents considered. Contrary to these results, in EMF exposures the risk in women is greater than in men in one study,15 while in another it is greater among men,29 both associations being statistically significant. These results should be interpreted with caution as the quality of these studies is relatively low. However, the possible role of sex as an interaction variable has not been fully explored.
Lastly, another well established risk factor for AD, as previously mentioned, is the allele APOE4. The genetic characteristics of individual people may modulate the expression of different environmental exposures. In our review, only one study37 introduces the APOE in the model as a confounding factor. But the interaction of this allele with the different occupational exposures was not investigated in any of the studies. This interaction should be considered for further research, but given the relatively low prevalence of this allele, even in cases of AD, this analysis will require larger samples.43
The results of our quality analysis, together with the associations observed in reviewed studies, suggest that there is evidence of an association between AD and occupational exposure to pesticides. The quantitative synthesis of the data in a meta‐analysis including studies with higher methodological quality might enable a more accurate quantification of the size of this suggested potential risk.
We thank Dr Jennifer Prieto House for her support in the translation of this paper.
AD - Alzheimer's disease
APOE4 - apolipoprotein E genotype epsilon 4 allele
aRR - adjusted relative risk
EMF - electromagnetic fields
GQI - Global Quality Index
Competing interests: None