OUP user menu

Risk factors for fracture in a UK population: a prospective cohort study

J. Porthouse, Y.F. Birks, D.J. Torgerson, S. Cockayne, S. Puffer, I. Watt
DOI: http://dx.doi.org/10.1093/qjmed/hch097 569-574 First published online: 18 August 2004


Background: Common clinical risk factors for fracture in older women have been identified. To date, most of these risk factors have not been confirmed in a UK population.

Aim: To confirm the important risk factors for fracture in older women.

Design: Comprehensive cohort study (CCS) with a nested randomized controlled trial.

Methods: The CCS included 4292 women aged >70 years. We assessed potential risk factors for fracture, and followed-up participants for 24 months for incidence of non-vertebral fractures.

Results: Odds ratios (ORs) for predicting any non-vertebral fracture were: previous fracture, 2.67 (95%CI 2.10–3.40); a fall in the last 12 months, 2.06 (95%CI 1.63–2.59); and age (per year increase), 1.03 (95%CI 1.01–1.05). ORs for predicting hip fracture were: previous fracture, 2.31 (95%CI 1.31–4.08); low body weight (<58 kg), 2.20 (95%CI 1.28–3.77); maternal history of hip fracture, 1.68 (95%CI 0.85–3.31); a fall in the last 12 months, 2.92 (95%CI 1.70–5.01); and age (per year increase), 1.09 (95%CI 1.04–1.13). ORs for predicting wrist fracture were: previous fracture, 2.29 (95%CI 1.56–3.34); and a fall in the last 12 months, 1.60 (95%CI 1.10–2.31). Being a current smoker was not associated with an increase in risk, and was consistent across all fracture types.

Discussion: Older women with the clinical risk factors identified in this study should be investigated for osteoporosis or offered preventive treatment.


Osteoporotic fractures represent a huge social burden in terms of both illness and financial implications.1,,2 However, unless preventive treatments are given to people at increased risk of fracture, they are unlikely to be cost-effective.3

In systematic reviews, bone mineral density (BMD) using dual-energy X-ray absorptiometry (DXA) predicts fracture risk.4 Guidelines recommend that the use of DXA should be restricted primarily to those presenting with fracture risk factors.5,,6 The largest study of fracture risk factors, the Study of Osteoporotic Fractures (SOFt), from the US, identified 16 independent risk factors for hip fracture.7 These 16 risk factors were subsequently reduced to four: personal history of previous fracture; family history of hip fracture; low body weight (participants in the lowest quartile of weight (up to 57.8 kg) had a significant increase in risk of future fracture) and current smoker.8 Risk factor studies in the UK have either had a small sample size9 or examined only a limited number of risk factors.10

Extrapolating the SOFt findings to a different population is problematic; risk factors require confirmation in a second population, because they tend to be optimistic in their predictive value within the original study population. Indeed, recent NICE guidelines have highlighted the necessity of further research on fracture risk factors.11 The aim of this study was to confirm the validity of the ‘SOFt’ risk factors within a UK primary care population.


The study design was a prospective, comprehensive cohort study (CCS) with a nested randomized controlled trial (RCT). The RCT was a study of the effectiveness of hip protectors, the methods and results of which have been reported elsewhere.12 In brief, this multi-centre RCT recruited women with one or more of the SOFt risk factors from GP lists. Two centres also collected data on women who either had risk factors but declined to take part in the RCT, or had no risk factors.


The participants were women aged ≥70 years, living in North Yorkshire and North Cumbria, UK, and were recruited between November 1999 and March 2001. Women (n = 35 000) were contacted via their primary care general practitioner (GP). Practices that agreed to take part were asked to mail to all women aged ≥70 years who were on their age-sex registers, after excluding those who were either terminally ill or who could not give informed consent. The envelopes contained information about the study and the risk factor questionnaire. The study was approved by the regional MREC and appropriate LRECs and all women provided written, informed consent.

Risk factor questionnaire

The questionnaire, as well as collecting data on the four SOFt risk factors also asked about whether the participant had fallen in the last 12 months and their self-reported health status (excellent, good, fair or poor). In addition, they were asked about current medication, both those prescribed by their doctor and those purchased by themselves over the counter. Figure 1 presents a flowchart illustrating participant recruitment.

Figure 1.

Flowchart showing recruitment of participants into the study.

Outcome assessment

The primary outcome was self-reported fractures, since the baseline information had been collected. All participants were contacted by postal questionnaire at 24 months and were asked about incident fractures over the previous 2 years. For participants in the RCT, hip fractures were also confirmed by the participant's GP. For all non-hip fractures in trial participants plus fractures of any type in the epidemiological study confirmation was not sought.

Statistical analysis

To identify independent risk factors for fracture, we used logistic regression, presented as odds ratio (OR) with 95%CIs. A regression was done with the four SOFt risk factors (adjusted for age) to enable a direct comparison with those data. A second regression was also done, including those four risk factors plus other variables.


Descriptive statistics

Of the 35 000 women contacted, 25.5% completed the baseline risk factor questionnaire (n = 8933), and 4.76% (n = 1666) had one or more risk factors and consented to be randomized into the trial. At 24 months, 8933 women were sent a follow-up questionnaire. At follow-up, 248 women had died, and 4393 did not respond or had withdrawn from the study. Consequently, the study population for which we had complete data comprised 4292 women (12.3%). Table 1 describes the characteristics of the women.

View this table:
Table 1

Baseline characteristics of subjects (n = 4292)

Age (years) (mean, SD)76.9 (5.14)
Weight (kg) (mean, SD)64.25 (11.05)
Previous fracture43.6% (n = 1867)
Maternal hip fracture11.6% (n = 498)
Current smoker7.8% (n = 335)
Fall in last 12 months29.3% (n = 1253)
Low body weight (<58 kg)29.1% (n = 1247)
Poor or fair self-reported health27.9% (n = 1197)
Anti-fracture treatments
HRT2.5% (n = 107)
Calcium/vitamin D10% (n = 429)
Bisphosphonates6% (n = 257)
SERMS0.2% (n = 9)

During follow-up, 330 women (7.7%) reported a non-vertebral fracture of any type (fingers, toes and ribs were excluded), 2.9% (125) had a wrist fracture and 1.3% (57) reported a hip fracture.

Identification of women at increased risk of fracture

In women with none of the SOFt risk factors, 4.6% (n = 49) had a fracture, vs. 8.7% (n = 281) of those with one or more risk factors (p<0.0001). The risk of fracture also increased with the number of risk factors present (Table 2). Using the four SOFt risk factors,8 a previous fracture was significantly predictive of any future fracture, hip and wrist fracture after adjustment for age (Table 3). In addition, low body weight was also an independent risk factor for hip fracture. A history of maternal hip fracture was associated with increased risk, but was not statistically significant at the p = 0.10 level. There was no association between smoking and increased risk of fracture.

View this table:
Table 2

Risk of fracture according to number of risk factors present (n = 4292, p for trend <0.0001)

Risk factors presentFractured (%, n)
0 (n = 1069)4.6 (n = 49)
1 (n = 1737)7.2 (n = 126)
2 (n = 1108)10.1 (n = 112)
3 (n = 307)11.3 (n = 35)
4 (n = 67)12.3 (n = 8)
View this table:
Table 3

Risk factors identified by the SOFt and their association with fracture risk in this study population

Any non-vertebral fracture
Previous fracture2.942.31–3.730.000
Low body weight (<58 kg)1.070.83–2.530.597
History of maternal hip fracture0.860.60–1.240.424
Current smoker0.700.42–1.170.171
Hip fracture
Previous fracture (1.8)*2.631.49–4.650.001
Low body weight (1.9)*2.201.28–3.780.004
History of maternal hip fracture (1.4)*1.670.85–3.260.135
Current smoker (1.7)*0.790.24–2.570.695
Wrist fracture
Previous fracture2.421.67–3.500.000
Low body weight (<58 kg)0.850.56–1.280.428
History of maternal hip fracture0.840.47–1.500.556
Current smoker0.760.35–1.650.483
  • *Figures in brackets represent RR of SOFt risk factors.

A second analysis was performed that included other possible risk factors for fracture (Table 4). Other risk factors were either not statistically significant, or did not appear to be associated with risk in this population (Table 4). A history of maternal hip fracture and poor or fair self-reported health were associated with increased risk of future hip fracture in subjects but did not reach statistical significance. No association was found between current smoking and any of the fracture categories.

View this table:
Table 4

All risk factors for fracture

Any non-vertebral fracture
Age (per year increase)1.031.01–1.050.003
Previous fracture2.672.10–3.400.000
Fall in the last 12 months2.061.63–2.590.000
Current smoking0.690.41–1.150.157
Maternal hip fracture0.860.59–1.250.422
Poor/fair self-reported health1.040.81–1.340.758
Low body weight (<58 kg)1.060.82–1.360.674
Hip fracture
Age (per year increase)1.091.04–1.130.000
Previous fracture2.311.31–4.080.004
Fall in the last 12 months2.921.70–5.010.000
Current smoking0.690.21–2.250.536
Maternal hip fracture1.680.85–3.310.134
Poor/fair self-reported health1.400.82–2.400.223
Low body weight (<58 kg)2.201.28–3.770.004
Wrist fracture
Age (per year increase)1.000.97–1.030.820
Previous fracture2.291.56–3.340.000
Fall in the last 12 months1.601.10–2.310.012
Current smoking0.790.36–1.730.556
Maternal hip fracture0.840.47–1.500.557
Poor/fair self-reported health0.810.54–1.220.319
Low body weight (<58 kg)0.840.55–1.260.396


These data confirm three (a personal history of previous fracture, low body weight and maternal history of hip fracture) of the four common SOFt risk factors as independent predictors of hip fracture. However, the data confirm only a personal history of previous fracture as a predictor of wrist fracture. In addition to the SOFt risk factors, a fall in the last 12 months was predictive of any non-vertebral, hip and wrist fracture, and increasing age was predictive of any non-vertebral and hip fracture, but not wrist fracture.

Personal history of previous fracture

Prior fracture more than doubled the risk of future fracture and was consistent across all fracture types. These findings are consistent with existing studies.8,,10 Even when BMD is known, studies suggest that a previous fracture may remain a strong risk factor. For instance, Cummings et al. found that a history of fracture since the age of 50 years remained significantly predictive, even after adjustment for BMD (RR 1.5, 95%CI 1.1–2.0).7 Similarly, prior vertebral deformity was a strong significant risk factor for future fracture, particularly hip fracture (RR 1.86, 95%CI 1.53–2.26) and vertebral deformity (RR 4.09, 95%CI 3.29–5.08).13

Family history of hip fracture

In both of our analyses, a history of maternal hip fracture was associated with a 70% increase (OR 1.67, 95%CI 0.85–3.26) in the risk of hip fracture. The size of the risk increase in this study is similar to the SOFt results, which identified a doubling of risk of hip fracture in participants.7 Despite maternal history of hip fracture not reaching the level of statistical significance in this study, the estimate is similar to that identified from the SOFt study. The link between maternal hip fracture and future fracture was specific to future hip fracture only, and no association was found with wrist fracture (RR 0.84) or any non-vertebral fracture (0.86). This finding accords with the SOFt study, where the risk associated with maternal history of hip fracture was specific to the site of fracture both for the hip and wrist.14 Thus a maternal history of hip fracture did not increase the risk of wrist fracture and visa versa. These findings, in conjunction with its possible independence from BMD, suggest that the effect of maternal hip fracture may be due to factors specific to the hip, such as the geometric properties of the bone, rather than a general predisposition to skeletal fragility.14

Low body weight (<58 kg)

Low body weight more than doubled the risk of hip fracture in the analysis based on the four SOFt risk factors (OR 2.20, 95%CI 1.28–3.78) and this did not alter when adjusted for other covariates (OR 2.20, 95%CI 1.28–3.77). These data accord with other studies where BMD was unknown. McGrother et al. and SOFt both found similar risk estimates for the association between low body weight and hip fracture.8,,9 However, low body weight did not appear to be associated with any risk for wrist fracture in this population (OR 0.84, 95%CI 0.55–1.26) and only slight elevation in any non-vertebral fracture that was not statistically significant (OR 1.06, 95%CI 0.82–1.36).

The SOFt study consistently demonstrated that when BMD is known, the risk association with low body weight disappears, suggesting a high correlation between these two variables (RR 1.0, 95%CI 0.6–1.5).15 Further, in the EPIDOS data, low body weight was the best predictor of very low BMD in elderly women.16 Consequently, low body weight is a very useful predictor of future fracture where BMD is unknown or unavailable.

Current smoking

We found no evidence to suggest that smoking is associated with increased fracture risk. This observation is in contrast to the SOFt study, which estimated a RR of 1.7 for current smoking and fracture.8 Recent meta-analyses have also demonstrated a significant association between smoking and all combined, hip and spine fracture, but not wrist fracture.17,,18

The reasons for the lack of association between smoking and fracture in this study are not entirely clear. The percentage of women who were smokers in this study was 7.8%, which is lower than in the SOFt study (9.9%). Women in this study were also older (mean age 76.9 years) than those in SOFt (mean age 72 years) and this, along with cultural differences in the starting age of smoking in the UK and USA, may have contributed to the finding, although this can only be speculation. The lack of association could also be a chance finding. However, it is worth noting that other studies have also shown non-significant associations between smoking status and fracture risk.9,,19

A fall in the last 12 months

The risk estimate in the presence of a falls history was nearly tripled for future hip fracture (OR 2.92, 95%CI 1.70–5.01) and for any non-vertebral fracture saw a doubling of risk (OR 2.06, 95%CI 1.63–2.59). For wrist fracture, the risk estimate was lower but again, was statistically significant (OR 1.60, 95%CI 1.10–2.31).

Other studies support the notion that a falls history may be an important risk factor for fracture. In the EPIDOS study, a falls history was predictive of fractures.20 Similarly, in the Dubbo study, a history of falls was significantly predictive of future forearm and wrist fracture.19 Further, in the SOFt study, falls were predictive of fractures of the distal radius and hip, although for the latter it just failed to reach statistical significance.7,,21

Self-reported poor or fair health

The risk of fracture in the presence of poor or fair health was not statistically significant. The risk estimate was greatest for hip fracture, but there was no evidence of an association with wrist fracture. Evidence from other studies is also strongest for hip fracture. Initially the SOFt data estimated a significant association, but in a subsequent analysis marginally failed to reach statistical significance.7,,22 Furthermore, McGrother et al. identified poor or fair health as an independent predictor of hip fracture, although the confidence intervals were wide around the risk estimate.9

On a point of general interest, it is worth drawing attention to the low baseline levels of anti-fracture treatment use in the women participating in this study, particularly since over 40% of them reported having had a previous fracture. However, this is not entirely surprising, since other studies have also shown that the majority of patients sustaining an osteoporotic fracture are not prescribed any pharmaceutical agents aimed at reducing fracture risk, such as bisphosphonates, calcium with vitamin D or hormone replacement therapy.23,,24

Strengths and limitations of the study

This study has several limitations. First, the results are not generalizable to women living in nursing or residential care, younger women or men. We had complete data on only 12% of the women that were contacted. This low response rate could lead to bias, if those who did not respond have different risks compared with those that did respond. However, the study by van Staa et al., which looked at previous fracture as a risk factor, used the general practice research database and, therefore, managed to identify virtually an entire population (i.e. all those registered with a GP, which will be nearly all).10 The risk estimates for future hip fracture in that study (SIR 2.8 for women aged 75–84 years) were similar to those in this study (OR 2.31), suggesting that bias is not a significant problem for that risk factor.

In terms of strengths, whilst our response rate was low, it is more than double that of the SOFt study (5%) which is the best available evidence on the role of multiple fracture risk factors to date (Cummings SR, personal communication). Furthermore, the SOFt sampling frame included driver license lists, which may not be as representative of the general population as our sampling strategy. Our sample was more than double the size of the only other UK-based study (n = 1864)9 and was large enough to allow us to confirm with a fair degree of confidence the importance of these common risk factors in a primary care population.

The method of fracture ascertainment in this study relied on self-report via a postal questionnaire. While a practical and relatively inexpensive method of data collection, particularly for large-scale studies, it may be a less accurate form of outcome assessment, with participants under- or over-reporting the outcome of interest. Other methods, such as collecting data from GP records or hospital admissions may be more accurate, but can be more expensive and time-consuming. The accuracy of self-report varies across different conditions, type of illness and ages.25–27 However, several studies have found self-report of fracture in the elderly to be a reasonably accurate means of fracture data collection, particularly where the recall interval does not exceed 2 years.26,28,,29 Another UK-based study that used GP and hospital records estimated a hip fracture incidence of 2% over three years, which is similar to this study (1.3% over two years), suggesting that self-report is a reasonably accurate means of data collection.


This study has confirmed that a personal history of previous fracture and a fall in the last 12 months are predictive of any future non-vertebral, hip and wrist fracture in a UK population. Increasing age is also predictive of any non-vertebral and hip fracture, but not wrist fracture. In addition, low body weight is also predictive of hip fracture. Older women with these clinical risk factors should be investigated for osteoporosis or offered preventive treatment.


View Abstract