Dr Schousboe serves as a consultant to Eli Lilly & Co. and receives support from Hologic, Inc. All other authors state that they have no conflicts of interest.
Published online on July 17, 2006;
In this large cohort of elderly women, prior non-spine non-hip fractures and radiographic vertebral deformities >10 years old were modestly associated with incident hip fracture, but the excess risks of hip fracture attributable to those prior fractures and deformities seem to wane over time.
Introduction: Whereas prior clinical fractures and prevalent radiographic vertebral deformities are well-documented predictors of incident hip fracture, the excess risks of incident fractures attributable to those prior fractures and deformities may decrease over time. Current guidelines regarding the assessment of fracture risk do not consider elapsed time since prior fracture or ascertainment of radiographic vertebral deformity.
Materials and Methods: We ascertained self-reported history of prior clinical fractures and calcaneal and total hip bone BMD and performed lateral spine radiographs in a cohort of 9516 community-dwelling elderly women who had not had a prior hip fracture. We prospectively followed them to assess incident hip fracture. Prevalent radiographic vertebral deformities were identified at baseline using morphometry, and incident hip fractures were confirmed by review of radiographic reports during three follow-up periods (0–5, >5–10, and >10 years after baseline exam).
Results: Among women who survived for 10 or more years after the baseline exam without having had a hip fracture, a history of non-spine non-hip fracture since age 50 reported at the baseline study examination was associated with a 21% age- and calcaneal BMD–adjusted excess risk (hazard ratio [HR], 1.21; 95% CI, 1.01–1.45) for subsequent incident hip fracture. Baseline radiographic vertebral deformity was associated with a 41% age- and BMD-adjusted excess risk (HR, 1.41; 95% CI, 1.15–1.73) of hip fracture after 10 years of follow-up. In comparison, the age- and BMD-adjusted HRs of incident hip fracture during the first 5 years of follow-up associated with prior non-spine non-hip fractures reported at the baseline study exam and prevalent radiographic vertebral deformities were 1.70 (95% CI, 1.30–2.22) and 2.10 (95% CI, 1.58–2.78), respectively.
Conclusions: Self-reported prior non-spine non-hip fractures and prevalent radiographic vertebral deformities known to be at least 10 years old are modestly associated with incident hip fracture. The association between these predictor fractures and subsequent hip fractures seems to wane with increased time after ascertainment of the predictor fracture. Hip fracture risk assessment strategies incorporating prior fracture history should also consider elapsed time since those prior fractures.
Prior fractures have been documented consistently to predict risk of subsequent incident fractures.(1–6) In particular, prevalent vertebral deformities are powerful predictors of incident vertebral fractures.(5,7–9) Virtually all published guidelines for diagnosis and treatment of postmenopausal osteoporosis recommend that those with a history of prior fracture be studied further for fracture risk.(10–19) Moreover, many also recommend that pharmacologic therapy to reduce fracture risk be considered for postmenopausal women with a history of prior fracture or radiographic vertebral deformity even in the absence of osteoporosis by BMD criteria.(10,12–15,17)
Several studies, however, have observed that fracture risk attributable to prior fractures is highest within the first couple years after their occurrence and that the risk attenuates over time.(8,20,21) The risk of incident fractures attributable to prior nonvertebral fractures may attenuate faster than the risk attributable to prevalent clinical vertebral fractures.(20) It is unclear how far into the future a prior clinical fracture or prevalent radiographic vertebral deformity still confers excess risk of incident fracture. Because hip fracture is the most costly(22) and one of the two most common fractures in the elderly,(23) this may be a particularly important issue with respect to incident hip fracture among the elderly, some of whom may have had a prior fracture that occurred in the distant past.
To address this issue, we ascertained prior history of clinical fracture and presence of prevalent radiographic vertebral deformity in a cohort of 9516 elderly women who had not had a prior hip fracture and followed them prospectively for incident hip fracture beyond 10 years of follow-up. Our primary aim was to assess whether prior radiographic vertebral deformities and clinical non-spine non-hip fractures still are risk factors for hip fracture if a hip fracture has not yet occurred after 10 years of follow-up.
MATERIALS AND METHODS
From 1986 to 1988, 9704 women at least 65 years old participated in the first examination of the prospective Study of Osteoporotic Fractures (SOF). Women were recruited from population-based listings at four metropolitan centers in the United States (Baltimore, MD; Minneapolis, MN; Pittsburgh, PA; and Portland, OR). The methods of recruitment have been described previously.(24) At the baseline examination, personal history of prior clinical fracture was ascertained, BMD at the calcaneus was measured, and lateral thoracic and lumbar spine films were obtained. Subsequent examinations were conducted every 2 years.
Ascertainment of prior clinical and prevalent radiographic vertebral fractures
At the baseline SOF examination, participants were asked if they had had any prior fractures, the skeletal location of those fractures, and their recollection of their age when they occurred. Among 9516 women who had not had a prior hip fracture, 3235 women self-reported a prior non-spine non-hip clinical fracture since age 50. Confirmation through review of radiographs or radiographic reports of these prior fractures was not sought.
Lateral thoracic and lumbar spine radiographs were taken at the baseline examination in the left lateral decubitus position; six-point digitations of the X-ray films for quantitative morphometry were done as described previously, so that the anterior (Ha), middle (Hp) and posterior (Hp) heights could be accurately measured.(25) Baseline vertebral deformities were defined using modified Melton/Eastell vertebral height ratio criteria.(26) A vertebral body was considered to be deformed if either of two height ratios within the vertebra (Ha/Hp, Hm/Hp) was >3 SD below the mean for that vertebral level or if both the anterior and posterior heights relative to the vertebra immediately inferior (Ha/Ha+1 and Hp/Hp+1) or relative to the vertebra immediately superior (Ha/Ha-1 and Hp/Hp-1) were >3 SD below the mean for that level. Mild radiographic vertebral deformities were those where the lowest height ratio was between 3 and 4 SD below the mean for that level, and severe radiographic vertebral deformities were those where the lowest height ratio was >4 SD below the mean for that level. The derivation of the normal mean height ratios and their SD have been described in a previous publication.(26)
A total of 9388 women who had not had a prior hip fracture had technically adequate radiographs at the baseline examination, and 1830 women were noted to have one or more prevalent radiographic vertebral deformities. For 640 women, the worst of these fractures was mild, and for 1190 women, the worst of these deformities was severe.
Measurement of BMD
BMD of the calcaneus was measured at the baseline examination (1986–1988) with OsteoAnalyzers (Siemens-Osteon, Wahiawa, HI, USA). At the second examination (1989–1990), BMD was measured at the proximal femur and lumbar spine with QDR-1000 scanners (Hologic, Bedford, MA, USA). CVs, determined by repeated measurement on study staff members who visited all four centers, were 1.2% for the calcaneus and femoral neck. Further details of densitometry measurements and quality control methods have been published previously.(27,28)
Ascertainment of incident hip fractures
Follow-up contact was made with participants every 4 months by mail or phone to ascertain incident fractures over a mean follow-up time of 12.8 years (range, 0–18.4 years), with a 97% response rate to these queries. Incident hip fractures were confirmed by physician adjudication of radiographic reports at the clinical centers and the study data coordinating center.
At the baseline examination, participants completed a questionnaire and were interviewed. They were asked to estimate the number of hours spent per day sitting and laying down and were queried about smoking status, self-rated health, estrogen use, maternal history of hip fracture since age 50, body weight and height at age 25, falls during the prior 12 months, and whether they walked for exercise. The ability to stand at least five times from a chair without using one's arms was evaluated by direct observation as a measure of proximal lower extremity strength. Participants' walking speed was measured in meters per second. Height and weight were measured during both baseline and the second clinic examinations with a Harpenden stadiometer and a balance beam scale, respectively. The change of weight (kg) from age 25 to the time of each examination was calculated from self-reported body weight at age 25 and measured weight at each of the examinations. Body mass index (BMI) was calculated as the ratio of current weight (kg) to height at age 25 (m) squared. Self-reported use of calcium and vitamin D supplements and exogenous estrogen was recorded at all exams, and alendronate at exam 6 (conducted during 1996–1997, a mean of 10.2 years after the baseline exam) was recorded shortly after it had become available for routine use in clinical practice.
Baseline characteristics of women with a self-reported non-spine non-hip fracture since age 50 before the baseline examination were compared with those of women without a prior clinical fracture using t-tests for continuous variables and χ2 statistics for categorical variables. Similarly, baseline characteristics of women with one or more radiographic vertebral deformities were also compared with those of women without a baseline deformity.
For the primary analyses, proportional hazards models were used to evaluate the risk of incident first hip fracture >10 years beyond the date of participants' baseline exam. Importantly, we excluded not only those who had died or terminated before the 10-year anniversary of their baseline exam, but also those who had suffered a hip fracture before that 10-year anniversary.
Four sets of regression analyses were performed. The first set estimated the hazard ratio for incident hip fracture >10 years after the baseline exam in those with baseline self-reported prior non-spine non-hip fracture after age 50 compared with those without baseline self-reported prior clinical fracture. The second set estimated the hazard ratio for incident hip fracture in those with prevalent vertebral deformities on baseline radiographs compared with those with no baseline vertebral deformity. The third set compared those with any mild (but no severe) radiographic vertebral deformities to those with no baseline vertebral deformity, and the fourth set compared those with one or more severe radiographic vertebral deformities to those with no baseline vertebral deformity.
Because spine radiographs were not repeated 10 years after the baseline exam, we did not exclude those who had an incident non-spine non-hip fracture or vertebral deformity between the baseline exam and the start of this follow-up period. Within each set of analyses, age-adjusted, age- plus calcaneal BMD–adjusted, and multivariate-adjusted models were constructed. The final models included those baseline covariates shown to be independently associated with both predictors (prior non-spine non-hip fracture and prevalent radiographic vertebral deformity) and with incident hip fracture in bivariate analyses at a p < 0.1. The included covariates were the ability to stand five times from a chair without using one's arms, falls within the last 12 months, BMI, walking speed, weight change since age 25 to the date of the baseline exam, and calcaneal BMD. For all proportional hazards models, participants were censored on death, termination, or occurrence of a hip fracture.
A secondary aim of this study was to examine whether the two pre-baseline fracture variables were more strongly associated with first incident hip fracture during the first 5 years after baseline than during the next two 5-year periods. We used proportional hazards models with time-dependent indicators for each of the two predictor fractures to estimate the period-specific relative hazards and compared the resulting estimates using χ2 tests for heterogeneity and pairwise differences. Separate age- and BMD-adjusted proportional hazards regressions were also done with incident hip fracture during the 0- to 5-year follow-up as the dependent variable, and again with incident hip fracture during the >5- to 10-year follow-up period as the dependent variable. For each set of analyses, participants who had suffered a hip fracture before the start of the follow-up period were excluded.
Because incident hip fracture has a stronger association with total hip BMD compared with calcaneal BMD,(29) we repeated age- and BMD-adjusted analyses of the associations between prior non-spine non-hip clinical fracture and prevalent radiographic vertebral deformity ascertained at the baseline exam and incident hip fracture occurring >5–10 or >10 years after the baseline exam, adjusting for total hip BMD measured at the second SOF examination. Additionally, we repeated the analyses between non-spine non-hip clinical fracture and first incident hip fracture occurring after 10 years of follow-up, this time also excluding those who had an incident non-spine non-hip clinical fracture between the baseline exam and the start of the follow-up period 10 years later.
Finally, to examine whether differential use of bone active agents between those with and without non-spine non-hip fractures or baseline vertebral deformities could have biased results obtained in our primary analyses, two additional sets of analyses were performed. First, the risk of incident hip fracture during the 0- to 5-year follow-up period was examined after further adjusting the age and calcaneal BMD adjusted model for baseline self-reported calcium, vitamin D, and estrogen. Then, incident hip fracture risk during the >10-year follow-up period was examined after further adjusting the age and total hip BMD adjusted model for calcium, vitamin D, estrogen, and alendronate use reported at exam 6.
Characteristics of the study population 10 years after the baseline exam
Among women who survived 10 years beyond the baseline exam without having a hip fracture (n = 8797), at baseline those women with a prior non-spine non-hip fracture were slightly older, had lower calcaneal BMD, were more likely to use their arms to stand up from a chair and to have had a fall in the prior 12 months, and had a slower walking speed compared with women who did not report a prior clinical fracture at baseline (Table 1). Among women who had survived 10 years after the baseline exam without suffering a hip fracture and who had evaluable radiographs from the baseline exam (n = 8765), those with baseline prevalent radiographic vertebral deformity were also slightly older, had lower BMD and BMI, were more likely use their arms to stand from a chair and to have had a fall within the prior 12 months, had a slower walking speed, and has less change of weight since age 25 compared with women without a prevalent radiographic vertebral deformity on baseline radiographs (Table 1).
Table Table 1.. Characteristics of Study Survivors With No Prior Hip Fracture 10 Years After Baseline Exam, With or Without Prior Fracture or Vertebral Deformity
Compared with those women with a prior non-spine non-hip fracture or baseline prevalent radiographic deformity, those without a history of either of these past fractures self-reported very slightly lower current or past use of calcium and vitamin D supplements but very slightly higher use of exogenous estrogen at the baseline examination. At the sixth examination, medication use data were available on 5964 women (68% of those surviving 10 years beyond the baseline exam without an incident hip fracture), on whom only 305 (5.1%) reported use of alendronate.
Associations between baseline predictor fractures or deformities and first incident hip fracture after >10 years of follow-up
Five hundred twenty-one hip fractures occurred after >10 years of follow-up, 392 occurred during the >5- to 10-year follow-up period, and 226 hip fractures occurred during the 0- to 5-year follow-up period. Those with one or more prevalent radiographic vertebral deformities at baseline had an age- and calcaneal BMD–adjusted 41% excess risk of first incident hip fracture after 10 years of follow-up (Table 2). Those with any moderate to severe radiographic vertebral deformities had an age- and BMD-adjusted 62% excess risk of incident hip fracture relative to those with no prevalent vertebral deformity, whereas there was no excess risk of incident hip fracture among those with only one or more mild vertebral deformities (Table 2). Adjustment for other covariates beyond BMD did not alter these results.
Table Table 2.. HRs of Incident Hip Fracture in Those With Compared With Those Without Baseline Predictor Fracture (or Vertebral Deformity) After 10 Years of Follow-Up
Those with self-reported non-spine non-hip clinical fractures had an age- and calcaneal-BMD adjusted 21% excess risk of first incident hip fracture after 10 years of follow-up (Table 2). Additional adjustment for covariates other than BMD again did not alter these results. Excluding those women who had an incident non-spine non-hip fracture between the baseline exam and the start of the follow-up period also had no effect on these results (data not shown).
Comparison of the associations between the predictors and first incident hip fracture occurring during different follow-up time periods
There was an age- and calcaneal BMD-adjusted 110% excess risk for first incident hip fracture occurring during the first 5 years of follow-up attributable to baseline prevalent radiographic vertebral deformities, compared with a 75% excess risk for first incident hip fracture during the time period 5–10 years after the baseline exam, and a 41% excess risk >10 years after the baseline exam (Fig. 1). The age and calcaneal BMD–adjusted excess risk for first incident hip fracture attributable to prior non-spine non-hip clinical fracture occurring 0–5 years after the baseline exam was 70% compared with a 32% excess risk of first incident hip fracture 5–10 years after the baseline exam, and a 21% excess risk of first incident hip fracture >10 years after the baseline exam (Fig. 1). The hazard ratios for incident hip fracture during the >5- to 10- and 0- to 5-year follow-up periods were not significantly different for women with compared with women without baseline prior non-spine non-hip fracture (p = 0.10) or for women with compared with women without baseline prevalent radiographic vertebral deformity (p = 0.24). However, the hazard ratios (HRs) for incident hip fracture during the >10-year follow-up period was significantly lower than that for incident hip fracture during the 0- to 5-year follow-up period in those with compared with those without baseline non-spine non-hip fracture (p = 0.02), and in women with compared with those without prevalent radiographic vertebral deformity (p = 0.007). In all of these analyses, adjustment for additional covariates beyond calcaneal BMD did not alter these results (data not shown).
Adjustment for total hip BMD compared with adjustment for calcaneal BMD
There was a 23% age- and total hip BMD-adjusted excess risk of first incident hip fracture >10 years after the baseline exam attributable to baseline radiographic vertebral deformity compared with an estimated 41% excess risk when adjusting for age and calcaneal BMD. There was little change in the estimated excess risk of incident hip fracture 5–10 years after the baseline exam attributable to baseline prevalent radiographic vertebral deformity when adjusting for total hip BMD instead of calcaneal BMD. Similarly, the estimated excess risks of first incident hip fracture during either the 5- to 10- or >10-year follow-up periods attributable to baseline prior non-spine non-hip fracture were changed little when adjusting for total hip BMD instead of calcaneal BMD.
Adjustment for use of calcium, vitamin D, estrogen, and alendronate
Alendronate use was associated with a multivariate-adjusted HR of 0.71 for incident hip fracture >10 years after baseline, but this was not statistically significant (95% CI, 0.44–1.14). Calcium, vitamin D, and estrogen use were not significantly associated with incident hip fracture during any follow-up time period. The estimated HRs for incident hip fractures attributable to either predictor fracture during the 0- to 5- or >10-year follow-up periods were not altered by additional adjustments for calcium and estrogen use assessed at baseline or for calcium, vitamin D, estrogen, or alendronate use assessed at exam 6 (data not shown).
Prior clinical fractures and radiographic vertebral deformities are strong risk factors for incident hip fractures, and guidelines generally emphasize including prior fracture history not only in case finding strategies for those with osteoporosis, but also in determining the magnitude of fracture risk and whether pharmacologic drug therapy to reduce fracture risk should be initiated. None of these guidelines, however, incorporate the element of time since prior fracture or since discovery of prevalent radiographic vertebral deformity in their recommendations of assessing future fracture risk.
In this large cohort of elderly women, we found that prior non-spine non-hip clinical fracture still confers a modest excess risk for incident hip fracture independent of BMD after 10 years of follow-up. That excess risk, however, was only about one third the excess risk attributable to prior fracture during the first 5 years of follow-up. Similarly, the excess risk or incident hip fracture attributable to prevalent radiographic vertebral deformity after 10 years of follow-up was less than one half of the excess risk evident through the first 5 years of follow-up. As others have reported,(30) we found that severe prevalent radiographic vertebral deformities were more strongly associated with incident hip fracture than mild deformities, but the strength of that relationship also waned with follow-up time. For those with one or more mild (but no severe) prevalent radiographic vertebral deformities, there was no excess risk for incident hip fracture attributable to those fractures independent of BMD beyond 10 years of follow-up. We found no evidence that differential use of calcium, vitamin D, estrogen, or alendronate between those with and those without prior fracture or prevalent vertebral deformity could account for these findings.
These results are consistent with those of Johnell et al.,(8) who noted that the risk of incident hip fracture was highest during the first year after a clinical vertebral fracture and declined each year over the subsequent 3 years, but remained slightly elevated over an additional 4 years. The incidence of subsequent radiographic vertebral deformity has also noted to be particularly high in the first year after an incident radiographic vertebral deformity.(31) Similarly, a previous study showed that self-reported prior wrist fracture since age 50 was a predictor of incident hip fracture up to but not beyond 7 years of follow-up.(21) Other studies that have assessed excess incident fracture risk attributable to prior fractures over long periods of time have not segmented the follow-up period,(32–34) and therefore have been unable to provide estimates of excess fracture risk in those with a prior fracture but who have not as of yet had the incident fracture after a number of years.
Survival bias might explain the waning association that we observed between prior fracture and incident hip fracture that we observed in two ways. First, a higher risk of fracture is associated with a higher mortality rate,(35) and in our study, those at highest risk of fracture may not have survived through the initial 10-year period after the baseline examination. Second, there may be individual heterogeneity in the association between prior fractures and skeletal fragility, and those women whose prior non-spine non-hip fracture or prevalent radiographic vertebral deformity were most indicative of skeletal fragility may have been at higher risk of fracture and therefore more likely to have a hip fracture soon after baseline exam. In essence, the population of those who have survived >10 years after a prior fracture without having a hip fracture may have “stood the test of time,” such that the prior fracture that occurred (or prevalent radiographic vertebral deformity that was documented) in the distant past is much less indicative of their current hip fracture risk than within the population of those who just had that prior fracture. The association between any predictor and incident fracture may wane with increasing follow-up time. For example, the association between baseline BMD and incident fracture also has been shown to weaken with follow-up time.(36)
The association between follow-up time and excess risk of incident fracture attributable to a prior clinical fracture or prevalent radiographic vertebral deformity may be important to recognize in clinical settings. Those individuals may not be at particularly high risk of incident hip fracture if the prior fracture occurred or radiographic vertebral deformity was ascertained >10 years in the past, and significant additional fracture risk factors are not present. Moreover, these individuals are commonly encountered in clinical practice. We found that nearly one half of all incident hip fractures occurred after 10 years of follow-up, and of those who had a baseline history of prior non-spine non-hip fracture, 90.8% (representing 30.9% of the original cohort) did not have any subsequent clinical fractures through the first 10 years of follow-up. Future guidelines may tether recommendations for pharmacologic therapy to estimated absolute fracture risk for the subsequent 5- or 10-year period, which in turn is estimated based on a number or clinical factors that when present independently confer fracture risk.(37) Clearly, the accuracy of estimated absolute fracture risk in those with a prior fracture will be enhanced if time since the occurrence of prior fracture or since ascertainment of prevalent radiographic vertebral deformity is also considered.
Our study has a number of important strengths. To our knowledge, our study is the first to assess excess incident fracture risk attributable to prior non-spine non-hip or prevalent radiographic vertebral deformities in segmented follow-up periods such that excess risks could be assessed specifically in those who have survived without incident hip fracture having occurred up to 10 years. Second, this study was conducted in a very large cohort of elderly women where incident hip fractures are determined with 99% accuracy.(38) This minimizes misclassification and contributes to the high internal validity of this study. Third, response rates to follow-up surveys sent every 4 months to assess incident fractures exceed 97% among those who have not died in this cohort. Fourth, we were able to adjust the estimated risks of incident hip fractures for multiple confounders, including BMD.
Our study has several limitations. First, we considered follow-up time from the baseline exam rather than from the actual date of prior non-spine non-hip clinical fracture because accurate dates for the occurrence of the prior fracture were not available. Hence, a precise estimate of the excess risk for incident hip fracture attributable to a prior non-spine non-hip clinical fracture and time since that fracture occurred cannot be made from our data. However, because the majority of radiographic vertebral deformities are not clinically evident at the time of their occurrence, we believe that our estimates for this study population of the association between excess risk for incident hip fracture attributable to prevalent radiographic vertebral deformity and time since ascertainment of that prevalent fracture are reasonably accurate. Second, we only considered incident hip fracture as the outcome of interest in this study, but the relationship between excess risk for incident fractures attributable to prior fracture and follow-up time may differ according to the skeletal sites of the prior and subsequent incident fractures. For example, Johnell et al.(20) found that the excess risk for incident clinical vertebral fracture attributable to prior hip or shoulder fracture waned significantly faster over a 5-year follow-up time period than the excess risk attributable to a prior clinical vertebral fracture. Third, self-reported prior clinical fractures were not confirmed by review of radiographic reports or medical records. With the exception of prior rib, face, skull, and hand fractures, self-report of prior non-spine non-hip fractures has been shown to be reasonably accurate, with acceptably low false-positive and false-negative rates in this cohort.(38) Nonetheless, modest misclassification of prior clinical fracture status at the baseline exam may have slightly biased toward the null our estimates of the association between prior non-spine non-hip clinical fractures and incident hip fractures. Fourth, radiographic vertebral deformities were defined morphometrically, and the degree to which morphometric vertebral deformities, especially mild ones, are genuine vertebral fractures remains controversial.(25,39–41) On the other hand, even mild prevalent radiographic vertebral deformities defined morphometrically were still associated with incident hip fracture in early follow-up time periods. Fifth, our ability to rule out any bias in our estimates of the association between baseline prior fractures or prevalent vertebral deformities and incident hip fractures >10 years later by use of calcium, vitamin D, estrogen, or alendronate is limited by the assessment of use of these agents around the start of the >10-year follow-up period in only 68% of the cohort. Finally, our findings were derived form a cohort of elderly white U.S. women ≥65 years of age and may not apply to other populations.
In conclusion, prior non-spine non-hip fractures and prevalent moderate to severe radiographic vertebral fractures that are known to be at least 10 years old are still modestly associated with incident hip fracture. These associations, however, wane with time after ascertainment of the predictor fractures. Mild radiographic vertebral fractures defined morphometrically and ascertained >10 years in the past are not associated with incident hip fracture. Fracture risk assessments using prior fracture history to estimate incident hip fracture risk should also consider the elapsed time since prior fracture occurrence or ascertainment.