Editor,—It has regularly been noted that familial cases of Alzheimer's disease (AD) tend to be of relatively early onset. This implies that there must be at least one further aetiological class, comprising sporadic cases that share some common exposure, and that tend to develop later in life than the familial cases. It can now be reported that cases of AD attributable to residence in areas where the public water supply has a relatively high content of aluminum do constitute such a class.

The largest data set so far studied in relation to waterborne aluminum comprises records of 2258 age and sex matched pairs of patients discharged from hospitals in the Province of Ontario during 1985, one member of each pair having a diagnosis of AD. Age at diagnosis was not documented for these cases but it is reasonable to assume that hospital patients with this diagnosis who were older at the time of discharge had also been older, on the average, at the time of onset.

Preliminary analysis of the Ontario data showed that the relative prevalence of hospitalised AD cases increased monotonically through four ranges of aluminum concentration in the public water supply (from <0.1 mg/1 to >0.200 mg/11). With the four ranges of aluminum collapsed to two (under and over 0.1 mg/1) these data have now been broken out by age and sex to obtain the results shown in table 1. Although the overall excess of AD in areas with the higher aluminum concentration was only 28%, five of the six estimates for ages beyond 75 were higher than this.

The statistical significance of the apparent age pattern is readily assessed by means of a permutational test procedure. Considering that there are 4+3+2+1 = 10 older–younger comparisons available within each sex, and a further 20 between older and younger age groups of unlike sex, we note that 29 of 40 such comparisons in table 1 show a higher relative risk in the older age group. A computer program was used to generate re-assignments of the estimated relative risk values to the 10 non-overlapping cells of table 1 and to count, for each such permutation, how many of the 40 older–younger comparisons had the higher value for the older group. After drawing a random sample (but with replacement) of 30 000 from the total of 3 628 800 possible permutations, there were found to have been only 1209 (4.03%) in which 29 or more comparisons had the higher value in the older group. Thus there is conventionally significant evidence against the null hypothesis of no age related aluminum effect and favouring the alternative hypothesis of an aluminum effect more strongly expressed in the age range where it has been judged—on completely separate evidence—that cases of AD are less genetically determinate.