Our results suggest that in the UK primary care-based population of older women, self-reported falls has the strongest association with self-reported fracture. Our Scree Plot suggests that reported falls explains the most variance in reported fractures, and it is at least twice as important as the other risk factors. As far as we are aware, this is the first report of the benefits, and quantification, of the addition of falls to the list of clinical risk factors using self-reported data, and suggests that previous analyses using computerised GP records have underestimated the importance of falls in fracture prediction.
The prospective cohort study of 366,104 women using the THIN database [16
] reported age, previous fracture, falls, BMI, smoking, any chronic disease diagnosis, and recent use of central nervous system medications, as characteristics that independently contributed to fracture risk at any site. However, in this paper, relative weights given to each risk factor show that for hip fractures for example, age was the most important determinant followed by falls, which agrees with the results of our study. Our study was unable to identify the other risk factors found in THIN and QResearch [17
] that add small additional predictive abilities, as we were only powered to find strong risk factors. However, in the THIN study [16
], a simple scheme that only included age and weight found fracture risks similar to their more complex scheme. It is likely that falls did not add a large additional benefit to age and weight, as seen in our study, because of lack of recording of falls on GP medical records. Our results support the view that in fallers, fracture prediction tools which take into account a falls history may have advantages over the FRAX tool which currently does not incorporate falls [20
We also showed, in agreement with all other studies, that age is the most important risk factor for fracture, despite the fact that age varied over a relatively narrow range in this study. Furthermore, we showed that height was associated with reported fractures, with taller women more likely to fracture. This has been shown previously for hip fracture [21
] with the explanation suggested involving biomechanical parameters of hip axis length [22
]. This is unlikely to be the explanation of the association with fractures in our cohort, as most were of the upper limb. It is possibly more to do with falling from a greater height, and a greater impact due to greater velocity, again reinforcing the importance of falls. Many other studies, using patient groups other than primary care-based populations agree with our results that low weight is an important predictor of fracture risk [23
], but we extend these results to suggest that falls is more important.
It is interesting that our results confirm that although age and height/weight are important risk factors for fracture risk, there was little evidence that other clinical risk factors used in FRAX such as previous or family history of fracture were associated with fractures. This is most likely to be because we were only powered to find strong risk factors. Other explanations include differences in the study population and methods of data collection. For example, studies used in the FRAX meta-analyses were prospective cohorts, while ours employed a cross-sectional analysis, suggesting that our data may be less robust.
Being a cross-sectional analysis means our study will suffer from bias, confounding and chance, and could be explained by reverse causality. In particular, the reported falls may predate the reported fracture. Nonetheless our study utilised a population-based cohort of elderly women from primary care and is roughly representative of the local area. Our response rate of 45.2%, while initially appearing low, is considered reasonable for a population-based recruitment strategy [25
]. Alternative explanations for our finding that 69% of frequent fallers had no record of this with their GP could include false positive reporting by study participants, but we think this is unlikely.
A further limitation is the use of reported rather than verified fractures as the main outcome. This will reduce the strength of any association found, but is unlikely to be a source of bias. We were unable to include rheumatoid arthritis as a secondary cause of osteoporosis due to poor validity and this is likely to have reduced the strength of any potential association between secondary causes and fractures in our analysis. There is a concern that people are far more likely to remember falls where they injured themselves such as fracturing a bone, resulting in recall bias. Furthermore, there is the potential for a large temporal difference between fractures since aged 50 and reported falls over the past 5 years which may introduce bias. The estimation of quantification of the additional predictive ability of adding falls to the clinical risk factor model using a Fit Statistic isn't ideal, as this penalises models with increasing numbers of variables.
So in conclusion, our study has shown that age and falls are the most important risk factors for fracture. We estimate that a history of more than one fall per year is at least twice as important as height or weight. Furthermore, using self-reported falls data are essential as computerised GP records underestimate falls prevalence. A risk-factor assessment just using age and falls is likely to be a useful means for simple fracture risk assessment for women within primary care in the UK.
- In post-menopausal women from primary care, age and falls are the strongest risk factors for fracture.
- Falling more than once per year is at least twice as important as other risk factors for fracture.
- Use of self-reported falls data is essential as computerised GP records underestimate the prevalence of falls in post-menopausal women.