Introduction

Fractures following inadequate trauma are the hallmark of osteoporosis [1]. More than 50 % of post-menopausal women and 30 % of men over the age of 60 years will suffer at least one minimal trauma fracture during their remaining lifetime [2, 3]. Any osteoporotic fracture predisposes to further fractures, significant morbidity and premature death [4, 5]. Importantly, following a first minimal trauma fracture, both men and women have a two- to threefold increased risk of subsequent fracture [6–8]. Not surprisingly, timely diagnosis and optimal treatment of osteoporosis prevent further fractures.

In recent years, a number of systematic interventions have been designed and locally implemented to improve the management of patients with osteoporotic fracture. A recent systematic review evaluating models of care for the secondary prevention of osteoporotic fractures demonstrated that intensive, co-ordinated programmes (“type A models”) are more effective in increasing treatment initiation rates than interventions based solely on patient or doctor education (“type C, D models”) [9, 10]. Whilst predictors of fracture amongst treatment-naïve patients have been well characterised (e.g. older age, previous fractures, falls, low bone mineral density, female gender), the factors associated with re-fracture in patients treated and managed within a Secondary Fracture Prevention (SFP) programme have not been determined. As early identification of patients who might re-fracture despite optimal therapy would justify additional, targeted interventions to lower the re-fracture risk, we aimed to determine the predictors of re-fracture amongst patients managed long-term for osteoporosis within the Concord SFP programme, an ongoing prospective observational study based at Sydney, Australia.

Materials and methods

Study population and design

The current analysis includes 234 patients who, following an incident osteoporotic fracture, were managed by the Concord Hospital SFP programme for a minimum of 3.5 years (Fig. 1). Relevant anthropometric, clinical and technical data were documented annually during a comprehensive clinical review. The first patient was recruited on 23 May 2005, and the last follow-up visit for the population in the current analysis was on 30 August 2012.

Fig. 1
figure 1

Study flow diagram. MTF minimal trauma fracture, SFP secondary fracture prevention

Full size image

The study population consisted of men and women aged 45 years and over who had sustained a symptomatic fracture due to minimal trauma (defined as a fall from a standing height or lesser impact). Patients were excluded from the trial if they were unable to provide informed consent, resided in a nursing home or hostel at the time of the index fracture or were diagnosed with malignant or metabolic bone disease. The study was approved by the Sydney Local Health District Human Research Ethics Committee (CH62/6/2009-021). All patients provided written consent to take part in the study, including consent for extraction of pharmaceutical claims data.

Patients identified by and/or referred to the Concord SFP programme were clinically assessed as described previously [9, 11, 12]. The following socio-demographic and clinical measures were obtained using a standardised questionnaire for each patient at baseline: date of birth, gender, smoking status (current smoker, former smoker, never smoked), alcohol intake, dietary calcium intake (in serves/week), index fracture site, family history of osteoporosis, maternal history of hip fracture, co-morbidities, past or current corticosteroid use, previous minimal trauma fracture (prior to index fracture) and ethnicity (Caucasian vs. non-Caucasian). Age was recorded at the time of the index fracture. Weight was measured in clothing without shoes, and height was measured using a Harpenden stadiometer (Dyfed, UK). All minimal trauma fractures apart from the face and skull were included. Hip, pelvis, wrist, humerus, vertebral, tibia and fibula fractures were classified as major fractures, whilst all other fracture sites were considered minor.

Total hip (TH), femoral neck (FN) and lumbar spine (L2–4) areal bone mineral density (BMD) (g/cm2) were measured at baseline and annually thereafter by dual X-ray absorptiometry using a GE/Lunar Prodigy (Lunar Corp. Madison, WI, USA; Software version 13.6) or a QDR4500-W Acclaim scanner (Hologic Inc., Bedford, Mass., USA; Software version 3.2). Osteoporosis was defined based on WHO diagnostic criteria [13]. For the Lunar densitometer, the coefficient of variation (%) and least significant change (g/cm2) were 0.8 % and 0.019 g/cm2 for the total hip, 0.5 % and 0.013 g/cm2 for the femoral neck and 1.5 % and 0.044 g/cm2 for the lumbar spine, respectively. For the Hologic densitometer, the corresponding numbers were 1.5 % and 0.034 g/cm2 for the total hip, 3.6 % and 0.069 g/cm2 for the femoral neck and 1.4 % and 0.033 g/cm2 for the lumbar spine, respectively. The standardised BMD was calculated using previously published equations [14, 15], and annual percent change in BMD from baseline was analysed for each individual at each site.

Urinary deoxypyridinoline (uDPD) concentrations were measured in a second morning void sample, using an enzyme-labelled chemiluminescent immunoassay (Pyrilinks-D, Siemens Healthcare Diagnostics, UK). The intra-assay coefficient of variation at a mean uDPD level of 30 and 100 nmol/L was 15 and 10 %, respectively. The inter-assay coefficient of variation was 7.1 % at uDPD concentrations between 11.6 and 110.3 nmol/L. Results were corrected for urinary creatinine levels and expressed as the ratio of uDPD to urinary creatinine (uDPD/cr).

The decision to treat was based on individual risk factors, co-morbid conditions and patient preference, as described previously [12, 16]. Patient management followed the current Australian osteoporosis guidelines as approved by the National Health and Medical Research Council and the Royal Australasian College of Physicians [17]. All patients were recommended non-pharmacological measures regarding bone health and falls prevention, including physical activity (e.g. weight-bearing and muscle strengthening exercises) and balance training (e.g. Tai Chi), as well as lifestyle changes such as sensible sunlight exposure and dietary calcium intake, as appropriate.

Patients deemed at high risk of re-fracture (group 1) were initiated on specific pharmacotherapy and were supplemented with calcium (600–1,200 mg/day) and vitamin D (1,000 IU/day) if and as required. Pharmacotherapy mainly consisted of oral bisphosphonates, while a smaller number of patients received intravenous bisphosphonates, strontium ranelate, denosumab, teriparatide and raloxifene. Pharmacotherapy was subsidised in all patients by the Australian Pharmaceutical Benefits Scheme (PBS). Patients considered at low risk of re-fracture (group 2) were advised to optimise their calcium intake and given instructions to help maintain sufficient vitamin D levels (Fig. 1).

Compliance and persistence with osteoporosis medication were calculated from pharmaceutical dispensing data obtained through Medicare Australia, as described previously [16]. The term compliance was defined as the extent to which patients act in accordance with the prescribed interval and dose of a treatment regimen [18]. In the present study, compliance was measured by calculating the medication possession ratio (MPR), i.e. the ratio of the number of days a patient is in possession of a medication over the observation period, with a maximum possible value of 1. A patient was considered persistent during the study period if there were no gaps in therapy of more than 30, 60 or 90 days over a 12-month observation period. Compliance and persistence were measured starting from the date that the patient was advised to commence therapy, which in the majority of cases was the baseline visit.

The primary outcome measure was a further minimal trauma fracture while being managed within the SFP programme. All new fractures were radiographically confirmed. Re-fractures sustained in the first 6 months after treatment initiation were excluded due to the known delay in the anti-fracture effect of most osteoporosis medications [19–21]. Traumatic fractures and fractures of the toes, fingers and skull were also excluded from the analysis. There were no pathological fractures observed during the study period.

Statistical analyses

Demographic characteristics and osteoporosis risk factors at baseline were described using means and standard deviations for normally distributed continuous variables and percentages for categorical variables. In order to obtain an accurate measure of how compliance (MPR) and persistence changed over time, each individual’s MPR and persistence were measured as a moving average over the previous 12 months, recalculated every 30 days. These measures of MPR and persistence were used as time-varying covariates to predict re-fracture amongst those with complete PBS data over the period from baseline to end of follow-up or re-fracture (i.e. time to event).

Employing Cox proportional hazards models, predictors of time to re-fracture were analysed separately for patients who were commenced on specific osteoporosis pharmacotherapy (group 1) and those maintained on calcium and vitamin D (group 2). Patients were censored on the date of the first fracture following the index fracture. In patients who did not suffer a further fracture during the follow-up period, data was censored on the date of the last clinic visit or the end of the study (30 August 12). Univariate (unadjusted) and multivariate (adjusted) analyses using clinical and socio-demographic variables (described above) were used to determine predictors of further fracture in both group 1 and group 2. Three multivariate analyses using a forward sequential method were conducted utilising variables with a p value <0.05 on univariate analysis:

  1. 1.

    All patients in group 1.

  2. 2.

    Patients in group 1 with complete PBS data (N = 69), using MPR and persistence as time-varying covariates.

  3. 3.

    All patients in group 2.

An additional sensitivity analysis included only patients treated with oral bisphosphonates. This analysis was performed to evaluate potential differences in anti-fracture efficacy between oral bisphosphonates and other osteoporosis therapies [19–25]. A second sensitivity analysis included re-fractures sustained within the first 6 months following treatment initiation.

Data were analysed using SPSS Statistics version 21 and SAS version 9.3.

Results

Baseline characteristics

The study population (n = 234; mean age 65 years) was mostly female (80 %), with 39 % having sustained a minimal trauma fracture prior to the index fracture. Approximately, 70 % of index fractures were classified as major. Over one third of patients in groups 1 and 2 sustained an index fracture of the wrist. Mean follow-up time was 5.2 years with a range of 3.5 to 7.3 years.

Individuals deemed at high risk of fracture and hence commenced on specific pharmacotherapy (group 1, N = 171) represented 73 % of the total study population. As expected, these patients were older, had more co-morbidities and prior fragility fractures, a lower body weight and lower lumbar spine and hip bone mineral density, and a higher urinary DPD/creatinine ratio than subjects in group 2 (Tables 1 and 2). Patients with incomplete (n = 102) or complete (n = 69) pharmaceutical claims data were similar at baseline for all variables listed in Tables 1 and 2, except that subjects with complete claims data had fewer falls (10.8 vs. 40.4 %, p < 0.001) and a higher frequency of oral steroid use (14.5 vs. 4.9 %, p = 0.030) and maternal history of hip fracture (5.8 vs. 0 %, p = 0.014). Within group 1, baseline characteristics of patients treated with oral bisphosphonates (N = 143, 84 %) were not different from those treated with other osteoporosis agents (N = 28, 16 %), except for a higher frequency of peptic ulcer disease in the latter group.

Table 1 Baseline characteristics—clinical and anthropometric data
Full size table
Table 2 Baseline characteristics—investigational results
Full size table

Re-fractures

Over a mean follow-up time of 5.2 years (range 3.5–7.3 years), 20.9 % of all subjects had sustained a further fracture, with an incidence of 44.4 per 1,000 person-years. In group 1, 26.3 % of patients sustained at least one further fracture, with an incidence of 57.1 fractures per 1,000 person-years (6.4 % had 2 further fractures and 5.3 % had 3 or more fractures). In contrast, 6.3 % of patients in group 2 sustained at least one further fracture, with an incidence of 12.7 fractures per 1,000 person-years (1.6 % sustained 2 further fractures). In group 1, the majority (60 %) of re-fractures occurred within the first 3 years from baseline (Fig. 2).

Fig. 2
figure 2

Re-fractures over 7 years of follow-up for groups 1 and 2. Each vertical line represents an incident fracture

Full size image

Amongst the 69 patients in group 1 with complete PBS data, 26.1 % of subjects had sustained a further fracture over a mean (SD) follow-up of 5.0 (0.8) years. In comparison, 26.5 % of subjects in group 1 with incomplete PBS data sustained a further fracture.

Predictors of re-fracture

Univariate predictors of re-fracture in patients considered at high risk of fracture (group 1) are listed in Table 3. Of note, age, gender, a minimal trauma fracture prior to the incident fracture, current smoking, a maternal history of hip fracture or lumbar spine bone mineral density were not associated with re-fracture in this high-risk group. In multivariate analyses, more than three co-morbidities, oral corticosteroid use (past or current) and lower total hip BMD remained significantly associated with re-fracture in group 1 (Table 4).

Table 3 Univariate predictors of re-fracture, group 1 (N = 171)
Full size table
Table 4 Multivariate predictors of re-fracture, group 1 (MPR and persistence data excluded, n = 171)
Full size table

Amongst patients with complete PBS data, a MPR ≤0.5 and lower body weight were significantly associated with re-fracture (Table 5). In contrast, a MPR ≤0.8 (vs. >0.8) was not a predictor of re-fracture (HR 1.10, 95 % CI 0.44–2.77, p = 0.842). Non-persistence, defined as a gap in therapy of either more than 30 days (HR 0.54, 95 % CI 0.21–1.40, p = 0.205), 60 days (HR 0.62, 95 % CI 0.23–1.69, p = 0.353) or 90 days (HR 0.69, 95 % CI 0.24–1.99, p = 0.490), was not a predictor of re-fracture. Proton pump inhibitor use was not a predictor of re-fracture (HR 1.31, 95 % CI 0.52–3.30, p = 0.566).

Table 5 Multivariate predictors of re-fracture, group 1 (patients with complete MPR data only, n = 68)
Full size table

Results were similar when the analysis was limited to patients treated with oral bisphosphonates only, with the exception that the number of co-morbidities was no longer associated with re-fracture. Results were also similar when re-fractures sustained in the first 6 months following treatment initiation were included in the analysis.

In patients considered at low risk of fracture (group 2), re-fractures were only associated with a higher baseline uDPD/cr (HR per unit increase 1.66, 95 % CI 1.02–2.70, p = 0.042).

Discussion

Secondary fracture prevention programmes significantly reduce the risk of further fragility fractures in patients with osteoporosis [9, 12, 26, 27]. However, even with the optimised treatment offered within a SFP programme, a proportion of patients will fracture again. The present long-term prospective, observational study identified poor compliance with osteoporosis therapy, significant co-morbidity, low hip BMD, low body weight and therapy with corticosteroids as major predictors of re-fracture amongst patients receiving high-intensity interventions within a SFP programme. Of note, risk factors known to be associated with the risk of fracture in treatment-naïve individuals, such as age, gender, prevalent fractures, previous falls, smoking, family history of hip fracture and lumbar spine bone mineral density [28–30], did not predict further fractures in our population. Although the sample size in our study was limited, it appears that the factors determining the risk of re-fracture differ significantly between treated and treatment-naïve populations.

These results contrast with those of a recent observational cohort study into the predictors of multiple (i.e. two or more) re-fractures in post-menopausal women treated for osteoporosis by their primary care physician [31]. Over 3 years of follow-up, prior fractures, two or more falls in the last 12 months and a worse score on a physical functioning/vitality questionnaire, were associated with further fractures. However, only 1.3 % of patients sustained multiple fractures while on treatment, incident fractures and adherence to therapy were self-reported and no data on baseline BMD, bone turnover or vitamin D status were available. Moreover, data were analysed using logistic regression rather than incorporating time-to-event information.

The current study demonstrates that even in the optimised setting of a SFP programme, low compliance to osteoporosis pharmacotherapy remains one of the most important predictors of re-fracture. Using MPR as a measure of compliance, previous reports found an increased risk of fracture at a MPR of less than 0.8 [32–35] or 0.5 [36]. However, in these studies, only 4 to 5 % of patients had sustained previous fractures; hence, the majority of events were not re-fractures per se [34–36]. Only one report has analysed MPR as a predictor of re-fracture in patients receiving oral bisphosphonate treatment [37]. In this study, a MPR of less than 0.5 was a significant predictor of re-fracture.

Reporting on the outcomes of a 2-year RCT within the setting of the Concord SFP programme, we recently demonstrated that compliance and persistence remained high once patients were initiated on treatment within the programme, independent of whether follow-up occurred with their primary care physician or a specialist within the SFP clinic [16]. Of note, one of the strengths of the latter and the present study is that MPRs were obtained from a claims database (rather than self-reported compliance) and analysed as a time-varying covariate to take into account the variation in MPR over time. Given the long duration of follow-up and inter-individual variation in time to event, a single measure of MPR (e.g. over the first 12 months) would not have accurately reflected compliance. Thus, our findings highlight the fact that encouraging patient compliance with therapy remains one of the major goals of physician follow-up.

The only predictor of re-fracture for group 2 was a higher urinary DPD/creatinine ratio, indicative of accelerated bone resorption. Both population-based and clinical studies have demonstrated that the rate of bone resorption (as measured by bone markers) is associated with the risk of hip and non-hip fractures in treatment-naïve post-menopausal women [38–40] and men [41], independent of BMD [38, 39]. Moreover, the degree of suppression of bone turnover during treatment with anti-resorptive agents appears to be associated with fracture risk reduction [42, 43]. Thus, in patients perceived as “low risk”, high bone turnover may still indicate the need for more intensive intervention.

The re-fracture rate in our total cohort was 20.9 % over a mean follow-up time of approximately 5 years. This number is significantly lower than the re-fracture rates reported in untreated/unmanaged populations. For example, in a statewide analysis of hospital admission data from New South Wales, Australia, 35 % of patients with incident osteoporotic fractures suffered another fracture within 6 years [44]. The latter figure is likely to be an underestimate, as the current coding system in New South Wales fails to capture all osteoporotic fractures. In comparison, the incidence of re-fracture reported in the Dubbo Osteoporosis Epidemiology Study (DOES) was 69 per 1,000 person-years in women and 71 per 1,000 person-years in men again significantly higher than in our population (44.4 per 1,000 person-years) [45]. This is consistent with the fact that the majority of patients in the current study received specific pharmacotherapy for osteoporosis whereas the DOES population remained largely untreated with 14 % of women and 4 % of men treated with anti-resorptive therapy.

Our study has several strengths. Apart from the large range of clinical and technical variables assessed during direct patient contact, the long follow-up of up to 7 years and the availability of PBS claims data in a subgroup of subjects, all re-fractures were confirmed radiologically and the fracture mechanism was ascertained by the study physician to exclude traumatic fractures. Furthermore, the cohort of patients in this study is highly representative of the populations managed by most SFP programmes; our results are therefore clinically relevant and widely applicable.

There are also several limitations of this study. Firstly, our sample size was relatively small and only 40 % of patients in group 1 had complete PBS data, limiting the power of our compliance analysis. However, baseline characteristics of those with and without complete PBS data were similar, indicating that the results of the compliance analysis may be valid for the entire group. Secondly, the study design is observational rather than a randomised controlled trial, which amongst a population at high risk of re-fracture would have been non-ethical. Thirdly, patients in the Concord SFP programme were prescribed different medications approved for the treatment of osteoporosis in Australia. As we cannot exclude that some of these agents may differ in terms of their anti-fracture efficacy, we performed a sensitivity analysis including only patients treated with oral bisphosphonates. This subgroup comprised 84 % of the total population and results did not differ significantly from the initial analysis. However, similar studies with other agents may have resulted in different outcomes. Furthermore, the majority of patients in this study were of Caucasian ethnicity and results may not be transferrable to other ethnicities. Lastly, pharmaceutical claims data provide no information whether the medication was actually taken by the patient and whether it was taken with the correct technique.

Overall, this study provides clinically applicable evidence of the predictors of re-fracture amongst patients with incident osteoporotic fractures managed within the setting of a SFP programme. While significant co-morbidity, low bone mineral density or body weight, and corticosteroid use may identify patients at high risk of re-fracture despite optimised post-fracture follow-up, poor compliance to therapy is the major driver of re-fracture in this population. Hence, improving and encouraging compliance to therapy remains the predominant task in any clinical setting.