This study has a number of strengths: the sample size was large; the finding of dementia and its differential diagnosis was made by experts following nationally agreed upon criteria; elimination of subjects whose examination was prior to age 65 lowered the number of AD cases primarily of genetic origin; the occupational information included job title and a description of the duties of the job; the staff collecting the data had no idea that one day the data may be used to study MF exposure as a risk factor; the study sites were spread across California, but used a single protocol; and the criteria for medium and high MF exposure used field measurements, was conservative, and were the same as in the previous Sobel et al
. studies [4
]. The relative frequencies of high and medium MF exposure were rather low because of the conservative nature of the criteria: 1.8% high and 4.9% medium.
The OR estimates are unlikely to be biased upwards, but may certainly be biased towards 1.0. We have taken hundreds of MF measurements of sewing machines, operators using both industrial and home sewing machines, and the ambient MF levels in the apparel manufacturing environment. We have also taken several measurements of the MF exposure of pilots and welders. These were the only occupations of the subjects which were classified as having high MF exposure. There is really no argument that the 35 subjects in the high MF classification most likely had high and longterm MF exposure. There is more room for error in the medium MF classification, however. Some of these 93 subjects may have had low MF exposure, but it is unlikely that they had high exposure. Among the 1770 study subjects classified as having low MF exposure (Table ), some may have had medium or even high MF exposure because there are many, perhaps somewhat uncommon or perhaps even common, reasons for extended high MF exposure, e.g., sitting very close to AC/DC transformers which are common in an office environment, having an office next to a communications equipment room, sleeping on the other side of a wall with a circuit breaker box, spending significant time in a kitchen with a microwave oven operating or using a handheld mixer, blow drying hair (e.g., dog's hair), using leaf blowers carried as backpacks, operating a car with certain electronic or electrical equipment actually near the driver's seat. These errors in classification all would tend to bias the OR estimators towards 1.0.
The use of demented subjects as controls may seem somewhat problematic. The initial Sobel et al
. study [5
], included 3 independent studies with varying controls: subjects hospitalized for an illness other than a neurologic problem; non-demented "neighborhood" subjects; patients with vascular dementia. The odds ratio estimates, for both females and males combined, were 2.9, 3.0, and 3.1 for M/H exposure. The combined sample odds ratio was 3.0 (p < 0.001). In the second Sobel et al
. study [4
], the controls were precisely as defined in the current study and the initial study. The M/H exposure odds ratio estimate was 3.9 (p = 0.006). A population-based study from Turkey by Harmanci et al
] used the same methodology as in our studies and expert diagnoses. The M/H odds ratio estimate was 4.0 (p < 0.05). In current study, the univariate M/H odds ratio was 2.2 (p < 0.02).
Three other studies have been published which used expert diagnoses. The Feychting et al
. study [12
] was small, used 2.0 mG as the cutpoint for M/H exposure, and had two control groups. They found odds ratios of 2.4 and 2.7 (p > 0.05). The Qui et al
. study [13
] also used 2.0 mG as the cutpoint, a Swedish "job exposure matrix" and expert diagnoses. They found an odds ratio of 0.9 or 1.0 (p > 0.05), depending upon the type of statistical adjustment. However, the odds ratio for males was 2.3 (p < 0.05) and for females was 0.8 (p > 0.05). Qui et al
. classified seamstresses who used a home sewing machine as having low exposure. In addition, 23.8% of the females in the study were classified as having M/H exposure, as opposed to 3.3% in the current study. Qui et al
. certainly classified some occupations as having M/H exposure which we classified as low exposure. They did not provide any information, except for seamstresses and telephone operators, as to which occupations were classified as M/H exposed.
Contrary to the statements in the Qui et al
. study, the classification of occupations in the Sobel et al
. studies [4
] and the current study were based on extensive occupational measurements, which is particularly true for seamstresses. Graves et al
] studied unionized workers and their families who subscribed to a large HMO. AD cases were expertly diagnosed. Their study was small and certainly found no relationship between MF exposure and AD. However, their operational definition of exposure was unusual and resulted in more than 19% of the subjects being classified as exposed. There were no seamstresses or tailors in their study, probably because workers in these occupations are seldom in unions.
The other published studies on AD and MF exposure used death certificate or hospital discharge records to determine AD status and were thus primarily not based on expert diagnosis. Among these studies, 4 had somewhat positive aspects [15
] and 3 were negative [19
]. Because other dementias are very often misdiagnosed as AD by community-based physicians, death certificate and even hospital discharge records are often incorrect when AD is provided as a cause of death or hospitalization. In addition, AD, when present, is often not specified as a cause or contributing cause of death. These two errors will bias odds ratio estimators towards 1.0.
The current study also has weaknesses. These include the following: occupational information would have been more detailed if MF exposure had been of interest, with interviewers and subjects blinded as to the study hypothesis; information on the use of equipment generating magnetic fields in hobbies or housework was not collected; lifetime occupational information was not collected; subjects were not sampled from a well-defined population; ApoE genotyping was not performed; specific measurements of exposure were not conducted for study subjects. Future studies can overcome these weaknesses if funding becomes available.