In a recent article in THE JOURNAL, Jacobsen et al (1) demonstrated a positive ecologic association between hip fracture discharge rates in the United States and fluoride content in water supplies. In a similar study, we recently reported the ecological association of discharge rates for hip fracture and water fluoride levels in 39 county districts in England. (2) Our study was performed in response to the suggestion that fluoridation of water might serve to stem the rising tide of hip fracture in western populations. Such a strategy was supported by laboratory evidence that fluoride was a potent inducer of bone formation, and by epidemiologic studies from Finland and the United States demonstrating lower rates of hip fractures associated with higher water fluoride levels. Our analysis demonstrated no significant association between discharge rate and total fluoride concentration (r = .16, P = .34). This lack of association was found for both men and women, as well as in a restricted analysis eliminating the smaller counties. However, the results from the recent US study prompted us to reexamine our data.
Our original statistical methods did not adequately account for differences in precision of the county-specific rate estimate. (3) We reanalyzed the data using a weighted least-squares technique (weighting each county by the size of the population aged 45 years and older) to allow for these differences. We found a significant positive correlation between fluoride levels and discharge rates for hip fracture (r = .41, P = .009, Figure). This relationship persisted for both women (r = .39, P = .014) and men (r – .42, P = .007). The respective regression coefficients and their standard errors were 0.46 (0.17) (discharges per thousand/milligrams per liter of fluoride) for total rate, 0.65 (0.25) for women, and 0.23 (0.08) for men.
We present these data for two purposes. First, given the widespread use of fluoridated water in public water supplies for the prevention of dental caries, any risk or benefit associated with this practice will affect extremely large numbers of persons. Using an appropriately weighted regression model, there appears to be a positive ecologic association between fluoride levels of county water supplies and fracture discharge rates. This ecologic association is consistent with a recently published study (1) and others currently in progress. Of course, this approach remains hampered by the problems common to all ecologic studies. (4) The relationship observed may be spurious due to the confounding of some other other factor that has not been accounted for in our analysis. Furthermore, an adverse impact of such low levels of fluoride appears biologically implausible, despite the recent trials suggesting such a consequence at much higher doses than in our study. (5) Nevertheless, this positive association demands further investigation at the individual level.
Our second purpose is to stress the methodologic issue of weighting in this type of analysis. The precision with which each county-specific rate is estimated is directly related to the size of the population. Analyses that fail to adequately account for this variation in precision give inappropriate emphasis to counties in which there is greater error in measurement of the rate. These data provide a striking example of such a bias obscuring the detection of potentially important associations.
1. Jacobsen SJ, Goldberg J, Miles TP, Brody JA, Stiers W, Rimm AA. Regional variation in the incidence of hip fracture: US white women aged 65 years and older. JAMA. 1990; 264:500-502.
2. Cooper C, Wickham C, Lacey RF, Barker DJP. Water fluoride concentration and fracture of the proximal femur. J Epidemiol Community Health. 1990; 44:17-19.
3. Rothman KJ. Modern Epidemiology. Boston: Little Brown & Co. Inc; 1988: 304-305.
4. Morgenstern, H. Uses of ecologic analysis in epidemiologic research. Am J Public Health. 1982; 72:1336-1344.
5. Riggs BL, Hodgson SF, O’Fallon WM. et al. Effect of fluoride on the fracture rate in postmenopausal women with osteoporosis. N Eng J Med. 1990; 322:302-309.