Treatment of Postmenopausal Osteoporosis with Slow-Release Sodium Fluoride: Final Report of a Randomized Controlled Trial

Methods

Clinical Data

Demographic and baseline presentations were described in the interim report [6]. We recruited 110 women with postmenopausal osteoporosis into the trial. All had radiologic evidence of osteopenia and osteoporosis; one or more vertebral fractures believed to be nontraumatic; and no secondary cause of bone loss. They were randomly assigned to one of two groups and stratified according to estrogen treatment. All study personnel were unaware of group assignment while data were being gathered. Ninety-nine patients completed at least 1 study cycle (1 year of actual treatment). The demographic or baseline presentations of these 99 patients did not differ according to treatment group [6] (Table 1). The two groups were similar in age, time since menopause, dietary calcium intake, height, weight, and number of spinal fractures. Both groups had moderate to severe osteoporosis: The average L2-L4 bone density was approximately 30% less than of a normal 30-year-old woman, and each group had a median of two spinal fractures at baseline.

Treatment

Patients in the fluoride group received slow-release sodium fluoride (Slow Fluoride, Mission Pharmacal Co., San Antonio, Texas), 25 mg twice daily, orally before breakfast and at bedtime in repeated 14-month cycles (12 months receiving treatment followed by 2 months not receiving treatment). They also received calcium citrate (Citracal, Mission Pharmacal), 400 mg calcium twice daily, before breakfast and at bedtime continuously throughout the study. Those in the placebo group received placebo (identical in appearance to Slow Fluoride but containing excipient only [provided by Mission Pharmacal]) on the same time schedule. The Mission Pharmacal Company had no role in the design of the study or in data retrieval, analysis, or interpretation. Thirteen of 48 patients in the fluoride group and 16 of 51 patients in the placebo group received concurrent treatment with estrogen. Nine of the 29 patients treated with estrogen were recruited at the primary site at Dallas; the other 20 were enrolled and evaluated at the Scott and White Clinic, Temple, Texas.

Fracture Quantitation

Before treatment and at 12 months of each cycle, a lateral spine roentgenogram was obtained for the assessment of spinal fractures. In the interim analysis [6], prevalent fractures (fractures present at baseline) were identified with the aid of radiology reports. For this final report, prevalent fractures were also analyzed using a computer program that calculated the vertebral dimensions of clearly unaffected vertebrae from landmarks (anterior and posterior corners and midpoints). By comparing these dimensions with published normal values [14], we obtained a correction factor. Using this correction factor, we estimated idealized vertebral dimensions before a fracture had occurred for the remaining vertebrae in the given baseline radiograph. A reduction in any height of more than 20% (from idealized to actual) accompanied by a decrease of at least 10% in vertebral area represented a prevalent fracture.

Incident spinal fractures (fractures occurring during the trial) were identified as described previously [6], using a computer-derived method. A reduction in any vertebral height of more than 20% accompanied by a decrease in vertebral area of more than 10% from one year to the next constituted a fracture [15]. A new incident fracture was a fracture that occurred during treatment in a previously unaffected vertebrae. A recurrent fracture was one that developed on a previously fractured vertebra.

Bone Mass Measurements

The use of different densitometers prompted us to calculate and use percentage changes per year rather than absolute values. The method for calculating changes in L2-L4 bone mineral content and bone density of the femoral neck and the radial shaft was described previously [6].

Safety Variables

Serum fluoride concentrations were measured before the morning dose of the test drug at 0, 3, 6, 9, and 12 months of each cycle, and they were analyzed using an ion-specific electrode. At the same visits, a history was taken for gastrointestinal and musculoskeletal side effects. A microfracture was defined clinically as moderate to severe lower-extremity pain that persisted for more than 6 weeks despite a reduction in treatment dose and objectively as changes on bone scan or radiograph.

The relation of each side effect to treatment was assessed. A symptom was considered to be related to treatment if it was moderate to severe in intensity, had no other cause, had newly appeared and persisted during the treatment phase, or had disappeared during the withdrawal period or with dose reduction. It was considered to be unrelated if it was present at baseline or during the late withdrawal phase, or if it had newly appeared but was not persistent.

The severity and frequency of side effects were also quantitated as adverse symptom scores. We identified 10 gastrointestinal items (symptoms such as nausea, vomiting, and diarrhea), 4 rheumatic items (pain in the foot, knee, hip, and other joints), and 3 skeletal items (pain in the lower, mid-, or upper back). Each item was given a numerical value of 1 to 3 for frequency (infrequent, frequent, or very frequent) and a numerical value of 1 to 3 for severity (mild, moderate, or severe). Side-effect score was the product of the value for frequency and the value for severity for each item. Thus, a constant, severe back pain yielded a score of 9 (3 × 3). A gastrointestinal score was derived for each patient by adding the scores of the 10 gastrointestinal items for all relevant visits and dividing the sum by the number of visits. A similar computation was done to derive rheumatic and skeletal scores for each patient.

Statistical Analysis

The data for incident spinal fractures were compared between the two groups-using three methods.

Individual Vertebral Fracture Rate

For each patient, the individual vertebral fracture rate was obtained by dividing the total number of new fractures by the duration of treatment. Because the data were skewed, this rate was compared between the two groups using the Wilcoxon rank-sum test.

Fracture-Free Rate

This rate was the percentage of patients without new fractures, unadjusted for covariates. The two groups were compared using the log-rank test to account for differential follow-up.

Survival

The Cox proportional-hazards regression model [16] was constructed to estimate the relative risk for a new spinal fracture while adjusting for covariates (treatment group, age, prevalent spinal fractures, years since menopause, height, weight, estrogen treatment, and stratum of baseline L2-L4 bone density). Time (in years) to the first fracture was considered to be the survival time. Analyses of fracture rates and logistic regression were also done [6]; the data are not presented because findings were similar to those obtained using the above methods.

The arithmetic difference in height from baseline to the end of treatment for each patient was compared between groups using a two-sample t-test and a two-way analysis of variance with the following factors: 1) treatment [fluoride vs placebo] and 2) fracture status (fracture-free vs one or more new or recurrent fractures).

For each patient, we calculated the percentage change per year for L2-L4 bone mineral content and bone density of femoral neck and radial shaft. The individual mean change for each patient was calculated as the average of yearly changes. The group mean was obtained by averaging the individual means. One-sample t-tests were then used to compare the percentage change to zero for each year or for the mean. Comparisons between groups were made using two-sample t-tests. Missing data precluded implementing a repeated-measures analysis of variance.

For related adverse events, the frequency of each event was compared between the two groups by using the Fisher exact test. Adverse symptom scores were compared between the groups by using the Wilcoxon rank-sum test and within the groups by using the Wilcoxon signed-rank test. For nonvertebral fractures, the exact tests based on the binomial distribution using person-year data were used to compare the two groups.

Most analyses were done using BMDP Statistical Software (BMDP, Los Angeles, California). Programs for analyzing person-time data were developed by the authors. Data are presented as mean ±SD unless otherwise indicated. All reported P values are two-sided.

Results

Duration of Treatment

The total duration of follow-up, including withdrawal periods, was 193 patient-years in the fluoride group (a mean of 3.57 years per patient for the 54 patients) and 189 patient-years in the placebo group (a mean of 3.38 years per patient for the 56 patients). Total duration was used in the side-effect analysis.

For patients completing at least one study cycle, the actual duration of fluoride or placebo treatment, corresponding to the number of completed cycles (1 year per cycle, exclusive of 2-month withdrawal periods), was 3.31 years per patient for the 48 patients in the fluoride group and 3.06 years per patient for the 51 patients in the placebo group. Actual duration was used to calculate spinal fracture rate and change in bone mass.

Computation of Incident Spinal Fracture Rates from the Visual Detection of Prevalent Fractures

Using spinal fracture data computed from spinal fractures visually detected at baseline, as in the interim analysis [6], we found that the individual vertebral fracture rate for new fractures was lower in the fluoride than in the placebo group (0.072 ± 0.184 compared with 0.209 ± 0.303 per patient-year; P = 0.006). Thirty-nine of 48 patients (81.3%) receiving fluoride compared with 29 of 51 patients (56.9%) receiving placebo had no new spinal fractures (P = 0.005).

Computation of Incident Spinal Fracture Rates from the Computer-Derived Detection of Prevalent Fractures

There was good agreement between the results of the visual and computer-derived methods for detecting prevalent fractures (κ coefficient, 0.946). A minor disagreement affected the incident fracture outcome slightly. Three fractures (in three patients) identified as new fractures in the interim analysis [6] were found to be recurrent fractures in our current analysis, because the vertebrae in question had been considered free of fractures by the visual method at baseline but were already identified as fractured by the computer-derived method. Baseline radiographs from the three patients were examined by two of the investigators and an experienced radiologist. Their consensus view agreed with the findings of the computer-derived method, validating the computation of incident spinal fracture rates from computer-derived detection of prevalent fractures. The results of this approach are presented below.

New Fractures

Table 2 compares spinal fracture data for new fractures between the fluoride and placebo groups. Eleven new spinal fractures developed over 159 cumulative patient-years in the fluoride group, whereas 35 new spinal fractures developed over 156 patient-years in the placebo group. In the fluoride group, 3 patients had 1 fracture and 4 patients had 2 fractures. In the placebo group, 14 patients had 1 fracture, 5 patients had 2 fractures, 2 patients had 3 fractures, and 1 patient had 5 fractures.

The individual vertebral fracture rate for new fractures was lower in the fluoride than in the placebo group (Table 2). Forty-one of 48 patients (85.4%) receiving fluoride were free of new spinal fractures compared with 29 of 51 patients (56.9%) receiving placebo. The difference in unadjusted fracture-free rate was statistically significant (P = 0.001).

Recurrent Fractures

Seven patients in the fluoride group and eight in the placebo group had recurrent spinal fractures (Table 2). The difference in the individual vertebral fracture rate between the two groups was not statistically significant.

Survival Estimates

Survival estimates for new spinal fractures, done using the Cox proportional-hazards model, showed adjusted fracture-free rates of 84.5% in the fluoride group and 60.4% in the placebo group after 4 years of treatment (Figure 1). The relative risk for new spinal fractures (fluoride/placebo) was 0.30 (95% CI, 0.12 to 0.76).

Estrogen treatment was not an independent predictor of new vertebral fractures (relative risk, 1.06 for patients receiving estrogen compared with patients not receiving estrogen). Among patients receiving estrogen, the mean value for the individual vertebral fracture rate was lower, and that for the fracture-free rate was higher, in the fluoride than in the placebo group. However, the changes were not significant, probably because of the small sample size. In addition, the study site was not found to be an important variable.

Dependence of Spinal Fracture Outcome on Baseline L2-L4 Bone Density

In Figure 2, top, the individual vertebral fracture rate for new fractures in each patient treated with fluoride is plotted against the corresponding baseline L2-L4 bone density (expressed as the percentage of the mean bone density of a normal 30-year-old woman). Among 28 patients whose baseline L2-L4 bone density was at least 65% of the mean bone density of a normal 30-year-old woman (previously defined as mild to moderate bone loss [17]), only 1 had a new fracture during fluoride treatment. In contrast, 6 of 20 patients with severe bone loss (baseline L2-L4 bone density less than 65% of the mean bone density of a normal 30-year-old woman) developed new fractures during fluoride therapy.

The relation between the individual new vertebral fracture rate and baseline L2-L4 bone density in the placebo group is shown in Figure 2, bottom. Nine of 28 patients with mild to moderate bone loss and 13 of 23 patients with severe bone loss developed new spinal fractures.

Of patients with mild to moderate bone loss, those in the fluoride group had a markedly lower individual vertebral fracture rate for new fractures (0.009 ± 0.047 compared with 0.188 ± 0.310 per patient-year; P = 0.004) and a substantially higher unadjusted fracture-free rate (96.4% compared with 67.9%; P = 0.004) compared with those in the placebo group. Thus, fluoride treatment almost abolished spinal fractures, whereas fractures still occurred at a substantial rate in the placebo group. Of patients with severe bone loss, those in the fluoride group had a marginally lower individual vertebral fracture rate (0.140 ± 0.262 compared with 0.225 ± 0.285 per patient-year; P = 0.19) and a marginally higher unadjusted fracture-free rate (43.5% compared with 70.0%; P = 0.06) than those in the placebo group.

Change in Bone Mass

In the fluoride group, the L2-L4 bone mineral content increased substantially by 4% to 6% per year during all 4 years (Figure 3). In contrast, the L2–L4 bone mineral content of the placebo group did not change substantially in any year. The difference between the two groups was statistically significant at each year. The mean change for all years was +4.82% ± 5.14% per year for the fluoride group compared with 0.15% ± 3.47% per year for the placebo group (P < 0.0001).

In the fluoride group, there was a statistically significant increase in femoral neck bone density during the first 2 years and for the mean (mean, +2.38% ± 3.33% per year; P < 0.001) (Figure 3). However, the placebo group had no statistically significant change in any year or in the mean (mean, 0.98% ± 5.04% per year; P = 0.73). There was no statistically significant difference in the change in femoral neck bone density between the fluoride and placebo groups.

There was no statistically significant change in radial shaft bone density in any year or for the mean in either group (mean, 0.48% ± 3.03% per year in the fluoride group compared with 0.67% ± 2.14% per year in the placebo group; P = 0.74) (Figure 3). The values in the two groups did not differ statistically at corresponding time periods.

Change in Body Height

The change in body height from baseline to the last visit for which data were available was calculated. The height decreased by 0.52 cm per patient in the fluoride group and decreased by 1.53 cm per patient in the placebo group; the difference between the groups was statistically significant (P = 0.03). Most of the loss in height occurred in patients who developed spinal fractures during the trial. A two-way analysis of variance showed that the critical determinant for loss of height was the presence of new or recurrent fractures (P < 0.0001) rather than treatment group (P = 0.34).

Serum Fluoride Concentration

In the placebo group, the serum fluoride concentration at the start of the study was less than 5 µmol/L, which is probably the therapeutic threshold [8] (Figure 4). It remained below that level throughout treatment. In the fluoride group, serum fluoride concentrations at 3, 6, 9, and 12 months of each year (when fluoride was given) were significantly greater than those of the placebo group (P < 0.0001); they ranged from 5 to 10 µmol/L (therapeutic window) and were significantly higher than the baseline value. However, before treatment (0 months) and at 14 months of each study period (after 2 months of not receiving fluoride), the serum fluoride concentration was less than 5 µmol/L.

Safety

Compliance was assessed by pill count and was 95.2% ± 7.9% in the fluoride group and 94.1% ± 6.4% in the placebo group. Minor gastrointestinal side effects were encountered in 9.3% of patients in the fluoride group and 7.1% of patients in the placebo group (P = 0.74). Minor musculoskeletal side effects occurred in 11.1% of patients receiving fluoride and in 14.3% of those receiving placebo (P = 0.78). There were no statistically significant differences in side-effect frequencies.

Gastrointestinal, rheumatic, and skeletal scores were calculated before treatment and during treatment visits (exclusive of withdrawal periods) for both groups. The scores before treatment did not differ significantly between the two groups for the three organ systems. The gastrointestinal scores did not change substantially during treatment relative to before treatment in either group (0.352 to 0.369 per patient visit in the fluoride group; P = 0.78; 0.411 to 0.297 per patient visit in the placebo group; P = 0.78). There was a statistically marginal decrease in rheumatic scores during treatment in both groups (0.611 to 0.278 per patient visit in the fluoride group [P = 0.07]; 0.411 to 0.221 per patient visit in the placebo group [P = 0.09]).

In the placebo group, there was no statistically significant change in the skeletal score during treatment relative to before treatment (0.839 to 0.664 per patient visit; P = 0.19). However, the skeletal score of the fluoride group was substantially lower during treatment than before treatment (0.870 to 0.422 per patient visit; P = 0.001).

No one in the fluoride group and one person in the placebo group (1.8%) had a hip fracture. Three patients in the fluoride group (5.6%) had nonaxial fractures other than of the hip. In two patients (3.7%), fractures of the pelvis and wrist were considered nontraumatic in origin. One patient had a traumatic fracture of the toe and fibula (fall from a ladder). In the placebo group, four patients (7.1%) had nonaxial fractures other than of the hip: Three patients (5.4%) had nontraumatic fractures (rib, humerus, and metatarsal bone, respectively), and one had a traumatic fracture of the sternum (automobile accident). Thus, the hip fracture rate in the placebo group was 0.0053 per patient-year (it was zero in the fluoride group). The rate of nontraumatic, nonaxial fractures other than of the hip was 0.0104 per patient-year in the fluoride group and 0.0159 per patient-year in the placebo group. For all nonspinal fractures other than those of the hip, the incidence rate was 0.0207 per patient-year in the fluoride group and 0.0211 per patient-year in the placebo group. All comparisons were nonsignificant (P > 0.2).

No one developed microfractures in either group. Three patients from the fluoride group and four from the placebo group had positive tests for occult blood in the feces, which in each case was believed to be unrelated to the test drug (hemorrhoids, diverticulitis, arteriovenous malformation in colon, and colonic polyp, respectively, were indicated). One patient from the placebo group had gastric submucosal bleeding from a hiatal hernia that was believed to be unrelated to the test drug. No one had gastric ulcers.

Withdrawals

Seventeen patients (31.5%) withdrew from the fluoride group: 6 because of travel problems, 6 because of lack of interest, 1 because of pelvic fracture, and 4 because of unrelated medical problems (recurrent seizures, persistent hypocalciuria, lung mass in a heavy smoker, and death from stroke, respectively). Seventeen patients (30.4%) withdrew from the placebo group: 3 because of travel problems, 6 because of lack of interest, 1 because of nausea and vomiting, 2 because of a spouse's illness, and 5 because of unrelated medical problems (stroke, interstitial lung disease, primary hyperparathyroidism, and deaths from myocardial infarction and stroke, respectively). The dropout rate was unrelated to the severity of spinal bone loss at baseline or to spinal fractures during treatment.

Discussion

This final report shows that the favorable effects of fluoride previously reported from an interim analysis [6] have been sustained through 20 additional months of follow-up. Thus, the survival estimate shows that a substantially higher percentage of patients in the fluoride group than in the placebo group remained free of new spinal fractures (relative risk, 0.3). The L2-L4 bone mass increased substantially, by 4% to 6% per year for 4 years, for a mean increase of 4.82% per year in the fluoride group; it did not change in the placebo group. The femoral neck bone density increased by a mean of 2.38% per year in the fluoride group but did not increase in the placebo group. The radial shaft bone density did not change in either group. Serum fluoride concentrations were kept within 5 to 10 µmol/L, which is believed to be the therapeutic window [8]. Sodium fluoride was well tolerated, and the two groups did not differ significantly in side-effect frequencies or appendicular fracture rates. No one from either group developed microfractures or gastric ulcers.

In this report, we update information from the interim report [6] in three important areas. First, by documenting that the effect on spinal fracture rate has been sustained through 4 years, we show that the action of slow-release sodium fluoride is different from that of etidronate [15]. Etidronate, which was reported to be effective in inhibiting spinal fractures during the first 2 years of treatment [15], was found during the third year of treatment to be ineffective [18].

Second, the spinal fracture data have now been separately analyzed at two different levels of baseline L2-L4 bone density. This has shown that the ability of slow-release sodium fluoride to inhibit spinal fractures is dependent on the baseline L2-L4 bone density. In patients with L2-L4 bone density at least 65% that of a normal 30-year-old woman (defined previously as mild to moderate bone loss [17]), fluoride therapy almost abolished spinal fractures. Thus, it may be possible to predict the response to fluoride using the level of L2-L4 bone density.

Third, we have examined supportive measures of effectiveness. Thus, the fluoride group has been shown to have one third the loss in cumulative height of the placebo group, probably because fluoride inhibits vertebral fractures. Moreover, the fluoride group, but not the placebo group, had a substantial reduction in the skeletal adverse-symptom score, which is a quantitative measure of back pain. Thus, by preventing spinal fractures, slow-release sodium fluoride may have reduced the severity and frequency of back pain.

The findings of this report are relevant to the results of other trials in osteoporosis. First, in contrast to this report, the trial of etidronate [18] showed a significant reduction in spinal fracture rate during treatment in the high-risk group with low spinal bone density, whereas the fracture rate did not decrease significantly in the whole group. However, it is not possible to ascertain whether the high-risk group in the etidronate trial belonged to the severe bone loss group as defined in our report.

Second, concern has been raised that fluoride treatment might exaggerate appendicular fractures and cause micro-fractures [4]. Our results suggest otherwise. Despite the limited number of persons evaluated, our trial showed no clinically or statistically significant difference in the incidence of appendicular fractures between the fluoride and placebo groups. In our total experience with slow-release sodium fluoride, which has involved 442 patients followed for 1213 cumulative years, the hip fracture rate was 0.0066 per patient-year and the rate of nonaxial fractures other than of the hip was 0.0206 per patient-year. In a randomized, controlled study using immediate-release sodium fluoride [4], the corresponding values during fluoride treatment were fourfold and fivefold greater than those seen with slow-release sodium fluoride. Immediate-release sodium fluoride lowers radial shaft bone density [4], but slow-release sodium fluoride has been shown to maintain it.

The positive clinical data in this report are supported by published reports of laboratory studies. Intact mineralization of bone has been shown by light microscopy [10, 13], quantitative histomorphometry [5, 10], and backscattered electron microscopy [13]. Improved trabecular connectivity was found by strut analysis [17] and nuclear magnetic resonance microscopy [19]. Moreover, an improved quality of cancellous bone without a deterioration in cortical bone was shown by critical-angle reflection ultrasound analysis [10-13].

The preservation of structural appearance and functional properties of cortical bone [6] could explain the lack of microfractures and exaggerated appendicular fractures. An increased strength of cancellous bone and a significant increase in trabecular connectivity [17] may account for the virtual abolishment of spinal fractures by fluoride treatment in the group with mild to moderate bone loss. Conversely, a nonsignificant increase in trabecular connectivity [17] may allow for a marginal inhibition of spinal fractures.

The positive response to slow-release sodium fluoride treatment described here is due not only to the delayed release of fluoride but also to the intermittent withdrawal of fluoride and the provision of calcium citrate. A short-term (2-month) withdrawal of fluoride (after 12 months of treatment) has been shown to overcome the attenuated osteoblastic activity that occurs with continuous fluoride therapy [10]. Optimum calcium supplementation with calcium citrate [20, 21] may satisfy the need for more calcium to accommodate the fluoride-induced stimulation of bone formation [22]. Moreover, calcium citrate may reduce bone resorption by inhibiting parathyroid hormone secretion (unpublished data). This conclusion is supported by examinations of bone biopsy specimens, which found an intact mineralization process and fewer eroded surfaces [13]. Calcium carbonate has been reported to decrease radial shaft bone density in patients with postmenopausal osteoporosis [4], but we found that calcium citrate alone stabilized it. The overall effect of this treatment regimen is therefore sustained stimulation of the formation of normally mineralized bone combined with reduced bone resorption. Thus, this treatment differs from treatment with bisphosphonates and calcitonin, which produce a low turnover state (with reduced resorption and formation), and from conventional fluoride therapy, which may enhance bone resorption.

Nevertheless, certain limitations of slow-release sodium fluoride treatment should be emphasized. First, as described in the interim analysis [6], this treatment does not alter the rate of fractures occurring on already fractured vertebrae (recurrent fractures). Second, in patients with severe bone loss, fluoride treatment only marginally affected spinal fracture rate. Third, this treatment has not been shown to inhibit appendicular fractures, although it does not appear to exaggerate them.

Intermittent slow-release sodium fluoride plus continuous administration of calcium citrate augments vertebral and femoral neck bone mass without decreasing radial shaft bone density, inhibits vertebral fractures without exaggerating appendicular fractures, attenuates a loss of height, reduces the severity and frequency of back pain, and is safe to use.

Dr. Piziak: Scott and White Clinic, 2401 South 31st Street, Temple, TX 76508.