Participants—This study was ancillary to the study of osteoporotic fractures, a multicentre study of risk factors for osteoporosis and fractures. During 1986-8 we recruited 9704 white women aged at least 65 years in Portland, Oregon; Minneapolis, Minnesota; Baltimore, Maryland; and the Monongahela Valley (an area with several small communities) in Pennsylvania. Recruitment sources included lists for jury selection and registration of voters, motor vehicle records, and membership records of health plans. Men and black women were excluded as were white women who were unable to walk without assistance and women who had a bilateral hip replacement.
Exposure to fluoridated water
—Exposure to fluoridated water was assessed with a questionnaire on residence history. Women were asked to list each address (street, city, state, and postal code), years they lived at that address, and the type of water supply (public, well, spring, etc) for each of their residences from 1950 to 1994. Water system maps and the 1992 fluoridation census2
were used to link street addresses with water system and fluoridation status. If a question arose, the appropriate water district was contacted to ascertain water source and fluoride content. For each year women were coded as being exposed or not exposed or having unknown exposure. The coding of unknown was used for residences outside the United States, incomplete addresses, and private wells in areas with naturally occurring fluoride. The questionnaire was sent to the 7612 women still active in the study and 94% (7129) completed it. To assess the reliability of the residence history, duplicate histories were completed by a randomly selected group of 103 women. The κ statistic for agreement between fluoride exposure was 0.94.
Measurement of bone mass—Bone mineral density (g/cm2) of the distal radius, proximal radius, and calcaneus were measured with single photon absorptiometry (Osteo-Analyzer, Siemens-Osteon, Wahiawa, Hawaii). Bone mineral density of the lumbar spine and proximal femur were measured with dual energy x ray absorptiometry (QDR 1000, Hologic Inc, Waltham, Massachusetts).
Assessment of risk factors
—Information on medical history, drugs and supplements, reproductive history, menopause, alcohol consumption, exercise, smoking, caffeine intake, and history of fractures was obtained through a questionnaire. Dietary calcium was assessed by a food frequency questionnaire administered by an interviewer.14
Women were also asked about walking, time spent sitting or lying down, and the amount of difficulty experienced with activities of daily living. Height and weight were also measured.15
Ascertainment of incident non-spinal fractures
—During the study participants were contacted every four months to inquire if a fracture had occurred (incident fracture). About 99% of these contacts were completed.16
If a fracture was reported the woman was interviewed and a copy of the radiographic report obtained. To be coded as a fracture the report had to mention the occurrence of an acute fracture. Fractures due to major trauma were excluded. All fractures that occurred up until 1 December 1995 were included (average follow up of 7.0 years).
Ascertainment of prevalent and incident vertebral fractures
—Lateral radiographs of the thoracic and lumbar spine were taken during the first clinical visit. A vertebral body was considered to have a prevalent fracture (fracture that had occurred before the study) if any of the following ratios were more than 3SD below the mean: the ratio of anterior to posterior height, mid-height to posterior height, and anterior height to the anterior height of the adjacent vertebrae.17
Repeat radiographs were obtained from 7238 women (average follow up of 4.0 years). The following definition of an incident vertebral fracture was used: a 20% reduction in the vertebral height of the anterior, middle, or posterior dimension of a vertebral body and at least a 4 mm decrease in the vertebral height of a dimension.
—To evaluate the effect of long term fluoride exposure we statified data by fluoride exposure. Women with no exposure during 1971-90 (n=2563) were compared with women with continuous exposure (n=3218) and women with mixed exposure (n=1348). A 20 year period was selected because information on residence history for dates before 1971 was less reliable. We used χ2
tests of homogeneity and analysis of variance and covariance to compare mean bone mineral density and other covariates across the exposure groups. We used proportional hazard and logistic regression models to assess the relation between fluoride and incident fractures. Multivariable models included those factors that differed between the fluoride exposure groups plus factors previously shown to be significantly related to skeletal health in the study of osteoporotic fractures.18,19