Introduction

Extensive prior literature has established that previous osteoporotic fractures (OF) are a significant independent risk factor for future OF [1]. This information is currently utilized in the FRAX calculator, which estimates the future risk of OF and supports the decision to initiate preventive treatment for osteoporosis [2, 3]. The elevated risk associated with previous OF is influenced by the patient’s sex and age; the time elapsed since the fracture and the site of the initial fracture have been recently incorporated in the extended FRAXplus model [4, 5]. Additionally, it has been shown that even a sentinel minor-OF is associated with an increased risk of future major-OF, although to a lesser extent compared to a sentinel major-OF [4].

As early as 2000, Klotzbucher concluded that any fracture doubles the risk of future OF [1]. There is evidence to suggest that even fractures resulting from high-energy trauma are associated with an increased risk of future OF [6]. However, no evaluation has been conducted to investigate fractures in typical non-osteoporotic sites, specifically fractures of the ankle, face, hands, feet, patella, and men’s tibial fractures.

The aim of this study was to investigate whether sentinel fractures classified by fracture site as non-osteoporotic also increase the risk of future major-OF, compared to sentinel major-OF and minor-OF. If so, it should be incorporated in models calculating fracture risk for decision making, such as the FRAX model. To that end, we compared the risk of future major-OF in patients who had major-OF, minor-OF, or non-OF with the general population, in a large primary care electronic health record (EHR) database from the UK.

Methods

Study design

Our study is a retrospective cohort study using the IQVIA Medical Research Data (IMRD), of primary care EHRs from the UK (IMRD-UK, version: 2022–09). The database incorporates data from THIN, A Cegedim Database (reference made to THIN is intended to be descriptive of the data asset licensed by IQVIA). The database contains various demographic and clinical data, including diagnoses, drug prescriptions, and procedures of over 14 million patients, covering approximately 5% of the UK population, and is representative in terms of demographics and condition prevalence [7]. IMRD-UK data has been standardized to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). The study was approved by IQVIA Scientific Review Committee (SRC Reference Number: 23SRC016).

Study population

The eligible study population included all patients aged 50 years or older on the index date of January 1, 2011, with at least 1 prior year of database inclusion (to allow for sufficient baseline characterization). The index date was chosen according to the median year of major-OF in the database. The overall eligible population was defined as the “general population,” including all patients, with and without prior fractures, and served as the reference group for estimates of risk ratio. Patients were classified based on the location of the sentinel fracture prior to the index date, identified by diagnosis codes (detailed in Appendix 1). Three exposure groups were defined: major-OF—namely a fracture of the proximal femur, proximal humerus, distal radius, or vertebra; minor-OF—specifically a fracture of the ribs, pelvis, midshaft and distal femur, distal humerus, proximal forearm, clavicle, scapula, sternum or women’s tibia and fibula [4]; and non-OF (see Appendix 1 for details). A patient could be assigned to multiple exposure groups in case of multiple fractures prior to the index date. In addition, three negative-control exposure groups were defined: glaucoma, onychomycosis, and anal fissures (see Appendix 1 for details). These were selected because they have no known causal effect on the outcome and allow assessment of potentially unobserved confounding (such as health-seeking behavior). For each patient, the following covariates were extracted: age, sex, observation period start date, observation period end date, socioeconomic status based on residential Townsend deprivation index, baseline General Practitioner visit counts, and osteoporosis risk factors included in the FRAX calculator. The latter were defined according to diagnosis, observation, and medication codes (detailed in Appendix 1), including secondary osteoporosis, family history of OF, low BMI, low bone mineral density (BMD) score—which was available only as a diagnosis but without an exact value, rheumatoid arthritis, previous use of glucocorticoids for more than 3 months, excessive alcohol consumption, and smoking. For patients with missing data on baseline Townsend index, we used post-index Townsend index. The result was dichotomized into a binary variable of deprivation, according to residential Townsend index ≥ 4.

Outcome and follow-up

The outcome of interest was the occurrence of a major-OF after the index date. The diagnosis codes used to detect the outcome of major-OF were identical to the codes of the major-OF exposure group. A fracture event was considered new if there was a gap of at least 90 days from a previous fracture diagnosis. Patients were followed until the earliest of major-OF, change of primary care provider, mortality, or the end of the study period (September 1, 2022).

Statistical analysis

Age- (in 1-year bands) and sex-specific incidence density rates of the outcome in the general population were computed as reference rates. For each combination of exposure group, age, and sex, the observed incidence density rates of that group were divided by the reference rates to derive raw incidence rate ratios (IRRs). The overall IRR across all age groups was modeled via Poisson regression with the log of expected hazard as the offset. The standardized incidence rate ratios (sIRR) were computed both with and without adjustment for potential confounding by deprivation and FRAX components. The analysis was repeated with forced exclusivity between exposure groups. That is, in case of multiple assignments, patients were reassigned only to the exposure group of the most osteoporotic fracture, e.g., patients with major- and minor-OF fractures prior to the index date were assigned only to the major-OF group.

The age- and sex-specific incidence rates were used to calculate absolute risk differences between each exposure group and the general population and the resulting number needed to screen (NNS) to prevent one additional outcome.

Results

The general population group included 1,951,388 patients. Prior to the index date, a total of 39,931 patients had a major-OF, 19,397 had a minor-OF, and 50,115 had a non-OF. Patients’ characteristics by exposure groups are presented in Table 1 (negative controls in Appendix Table 3). The patients in the major-OF group were older, with a median age of 77 years, as compared with 64 years in the general population and non-OF groups and 68 years in the minor-OF group; the percentage of females in the major-OF group was higher than in the other exposure groups. The median follow-up time for all groups was 4–5 years. Socioeconomic status did not differ significantly between groups. The distribution of sentinel fractures at different sites within the exposure groups is detailed in Appendix Table 4.

Table 1 Study population characteristics
Full size table

During follow-up, there were a total of 57,671 major-OF in the general population, 3748 in the major-OF group, 1398 in the minor-OF group, and 2528 in the non-OF group. Table 2 presents the age and sex sIRR of the outcome major-OF for each exposure group relative to the general population. Expectedly, the highest sIRR was observed in the major-OF group (2.73, 95% confidence intervals: 2.64–2.82). While the sIRR of the non-OF group was lower (1.83, 1.74–1.92), it was still significantly above one. Of note, in females, the sIRR in the minor-OF group was higher than the major-OF group (1.81, 1.75–1.88, vs. 1.61, 1.57–1.65). The sIRRs in males were higher than in females in all groups.

Table 2 Age and sex standardized rate ratios by prior exposure
Full size table

The trend between exposure groups was preserved when age-specific IRRs were calculated. Figure 1 presents smoothed curves of age- and sex-specific IRRs by exposure group. In males of nearly all ages, non-OF bears lower risk relative to minor-OF, and both present lower risk relative to the major-OF group. In females, minor-OF poses greater risk than major-OF in the younger ages, yet with overlap** confidence intervals in older ages. In females as well as males, across most age groups, non-OF shows lower risk than minor-OF and major-OF, yet a significantly increased risk in comparison to the general population. Another clear trend is a decrease in the IRRs with age. Interestingly, though, the IRRs for major-OF rise until the early 60’s for both males and females, as well as for non-OF in males.

Fig. 1
figure 1

Smoothed curves of age- and sex-specific rate ratios of incident major osteoporotic fractures by prior exposure to fractures

Full size image

Figure 2 presents age- and sex-specific crude incidence rates by exposure group. The incidence in the general population presents an exponential growth rate pattern, with the incidence in females preceding the males by a decade (e.g., the incidence of females in the age of 70 equals males at the age of 80). The average NNS among non-OF to prevent one additional major-OF compared with the general population was 534 for males and 231 for females, with a trend of decreasing NNS with age, reaching about 150 for both males and females above the age of 85. The average NNS for the major-OF group was 95 and 63 for males and females, respectively.

Fig. 2
figure 2

Age- and sex-specific crude incidence rates by prior exposure to fractures

Full size image

The results were not affected when exclusivity between exposure groups was applied (See Methods. Results not shown).

All three negative exposure groups showed sIRR not significantly different from 1 (Table 2), including in the age-specific curves (curves not shown), indicating that the risk in these subgroups is not different from the general population.

Discussion

Previous osteoporotic fractures are a well-established and significant risk factor for future osteoporotic fractures, and this risk was shown to depend on additional factors, mainly age, sex, and sentinel fracture’s site and recency. In our study, we investigated the rate ratio for future major-OF posed by fractures in non-osteoporotic sites, as well as the influence of age and sex on this risk. Our results suggest that history of non-OF is significantly associated with future major-OF, with a 1.8-fold increased risk in males and a 1.4-fold increased risk in females, and a clear trend of risk increase, with minor-OF (2.4 and 1.8 in males and females, respectively) and major-OF (2.7 and 1.6 in males and females, respectively). When calculating age-specific rate ratios, the risk grading remained consistent, and the rate ratio for non-OF was significantly greater than the general population in both sexes and across almost all ages. The additional risk decreased in all exposure groups with advancing age.

We demonstrated a significantly increased risk ratio for a future major-OF following a sentinel fracture in a non-osteoporotic site, while sentinel minor- and major-OF elevated risk ratio for future major-OF even further. The effect of sentinel minor-OF was shown in a study by Kanis [4], in which minor-OF increased the risk for future major-OF, especially in males, albeit to a lesser extent than sentinel major-OF. Our results show a similar effect of minor-OF, thus, in accordance with Kanis’ results, with the additional information about the effect of non-OF. To the best of our knowledge, a study of fractures in non-osteoporotic sites has not been conducted. Leslie et al. [6] compared high-energy with low-energy fractures, showing that both groups had comparable increased risk for future fractures. Nevertheless, with similarly low BMD Z-scores in both groups, a description of a high-energy fracture mechanism might not indicate a non-OF. Therefore, to our understanding, the impact we describe of non-OF has not been shown before and is currently not included in models such as the FRAXplus model.

Regarding the influence of patients’ age, we found a trend of decreasing sIRR with age following all fracture types: major, minor, and non-osteoporotic, in most ages. A trend of initial increase in sIRR was found only in younger patients following a major-OF and in younger males following non-OF. Previous studies have shown a decrease in risk ratio with advancing age [8,9,10,11]. Kanis found a consistent decrease in risk ratio with age, regardless of sex or fracture site, including minor-OF [4].

The clinical consequences can also be described in terms of NNS. Despite the decrease in sIRR with age, the parallel increase in the incidence rate of major-OF in the general population reduces the NNS in the elderly to about 150 in both males and females. This value is on par with NNS among patients with major-OF, which are considered for further evaluation. The impact of non-OF on the risk of major-OF does not necessarily stem from undiagnosed osteoporosis. An alternative explanation might be lifestyle factors such as high-risk behaviors which caused the sentinel fracture. These can also explain the trend of increasing risk we found in younger patients or the convergence of the risks with age. Further validation of these findings is required.

There are several key strengths to the study, primarily the substantial size of the study population. Some of the relations we examined were studied before, enabling us to validate our results and further explore gradations between exposure groups described in previous studies. Additionally, we incorporated various significant risk factors appearing in the FRAX calculator, except for BMD. Finally, we included three negative exposure groups for validation.

Our study also has several limitations: First, retrospective studies based on EHRs may be prone to misclassification and selection biases and do not allow direct validation of fractures ascertained. The large population and the many risk factors we accommodated might minimize these biases. Second, we lacked information on BMD, which is an important confounder, as low density can both increase the risk for non-osteoporotic high energy fractures [12] as well as for major-OF. Therefore, our findings of elevated risk among patients with prior non-OF might reflect a subgroup with lower BMD. While BMD is generally included in fracture-risk calculators such as FRAX, its use is optional and not mandatory. Therefore, when BMD is unavailable, including prior non-OF can benefit the estimation of future major-OF risk. Third, we did not calculate IRR according to recency nor standardized for it, but the median follow-up time was similar between the general population and the non-OF groups, and the history period prior to the index date was longer in the non-OF group, so results were likely not biased by these parameters.

In conclusion, we have shown that non-osteoporotic fractures are associated with increased risk for future major-OF, although to a lesser extent compared to major and minor osteoporotic fractures. This supports the inclusion of all fractures as components in fracture risk assessment tools such as the FRAXplus calculator. Future studies on refracture risk should include prior events of non-OF as an exposure group. Further research is warranted to clarify if this is an independent risk factor after accounting for BMD.