‘Skeletal Age’ for mapping the impact of fracture on mortality

  1. Thach Tran  Is a corresponding author
  2. Thao Ho-Le
  3. Dana Bliuc
  4. Bo Abrahamsen
  5. Louise Hansen
  6. Peter Vestergaard
  7. Jacqueline R Center
  8. Tuan V Nguyen  Is a corresponding author
  1. School of Biomedical Engineering, University of Technology Sydney, Australia
  2. Garvan Institute of Medical Research, Australia
  3. Faculty of Medicine, UNSW Sydney, Australia
  4. Department of Medicine, Holbæk Hospital, Denmark
  5. Department of Clinical Research, Odense Patient Data Explorative Network, University of Southern Denmark, Denmark
  6. Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences University of Oxford, United Kingdom
  7. Kontraktenheden, North Denmark Region, Denmark
  8. Department of Clinical Medicine, Aalborg University, Denmark
  9. Department of Endocrinology, Aalborg University Hospital, Denmark
  10. Steno Diabetes Center North Jutland, Denmark
  11. School of Medicine Sydney, University of Notre Dame Australia, Australia
  12. School of Population Health, UNSW Medicine, UNSW Sydney, Australia

Abstract

Background:

Fragility fracture is associated with an increased risk of mortality, but mortality is not part of doctor-patient communication. Here, we introduce a new concept called ‘Skeletal Age’ as the age of an individual’s skeleton resulting from a fragility fracture to convey the combined risk of fracture and fracture-associated mortality for an individual.

Methods:

We used the Danish National Hospital Discharge Register which includes the whole-country data of 1,667,339 adults in Denmark born on or before January 1, 1950, who were followed up to December 31, 2016 for incident low-trauma fracture and mortality. Skeletal age is defined as the sum of chronological age and the number of years of life lost (YLL) associated with a fracture. Cox’s proportional hazards model was employed to determine the hazard of mortality associated with a specific fracture for a given risk profile, and the hazard was then transformed into YLL using the Gompertz law of mortality.

Results:

During the median follow-up period of 16 years, there had been 307,870 fractures and 122,744 post-fracture deaths. A fracture was associated with between 1 and 7 years of life lost, with the loss being greater in men than women. Hip fractures incurred the greatest loss of life years. For instance, a 60-year-old individual with a hip fracture is estimated to have a skeletal age of 66 for men and 65 for women. Skeletal Age was estimated for each age and fracture site stratified by gender.

Conclusions:

We propose ‘Skeletal Age’ as a new metric to assess the impact of a fragility fracture on an individual’s life expectancy. This approach will enhance doctor-patient risk communication about the risks associated with osteoporosis.

Funding:

National Health and Medical Research Council in Australia and Amgen Competitive Grant Program 2019.

Editor's evaluation

This important study presents the idea of "Skeletal Age", defined as the age of one's skeleton as a consequence of fragility fracture, as a potential new tool to raise awareness about the increased risk of mortality following a fracture (particularly hip fractures) and thus improve the medical management of osteoporosis. The evidence is convincing and is derived from a very large database from the Danish National Hospital Discharge Registry. The proposed approach is of general interest and might represent a starting point for making the health risks of an osteoporotic fracture more intuitive and possibly more effective.

https://doi.org/10.7554/eLife.83888.sa0

eLife digest

Osteoporosis is a ‘silent disease’ which often has no immediate symptoms but gradually weakens bones and makes them more likely to break. A bone fracture caused by osteoporosis in people over the age of 50 is linked to long-term health decline and in some cases, even early death. However, poor communication of the mortality risk to patients has led to a low uptake of treatment, resulting in a crisis of osteoporosis management.

The impact of a fracture on life expectancy is typically conveyed to patients and the public in terms of probability (how likely something is to occur) or the relative risk of death compared to other groups. However, statements such as “Your risk of death over the next 10 years is 5% if you have suffered from a bone fracture” can be difficult to comprehend and can lead to patients underestimating the gravity of the risk.

With the aim of devising a new way of conveying risks to patients, Tran et al. analyzed the relationship between fracture and lifespan in over 1.6 million individuals who were 50 years of age or older. The findings showed that one fracture was associated with losing up to 7 years of life, depending on gender, age and fracture site. Based on this finding, Tran et al. proposed the idea of 'skeletal age' as a new metric for quantifying the impact of a fracture on life expectancy.

Skeletal age is the sum of the chronological age of a patient and the estimated number of years of life lost following a fracture. For example, a 60-year-old man with a hip fracture is predicted to lose an estimated 6 years of life, resulting in a skeletal age of around 66. Therefore, this individual has the same life expectancy as a 66-year-old person that has not experienced a fracture.

Skeletal age can also be used to quantify the benefit of osteoporosis treatments. Some approved treatments substantially reduce the likelihood of post-fracture death and translating this into skeletal age could help communicate this to patients. For instance, telling patients that “This treatment will reduce your skeletal age by 2 years” is easier to understand than “This treatment will reduce your risk of death by 25%”.

Given the current crisis of osteoporosis management, adopting skeletal age as a new measure of how the skeleton declines after a fracture could enhance doctor-patient communication regarding treatment options and fracture risk assessment. Tran et al. are now developing an online tool called ‘BONEcheck.org’ to enable health care professionals and the public to calculate skeletal age. Future work should investigate the effectiveness of this new metric in conveying risk to patients, compared with current methods.

Introduction

Fragility fracture is a direct consequence of osteoporosis, just as stroke is a consequence of hypertension. Fracture, especially hip fracture, is associated with an increased risk of mortality. Indeed, patients with a fragility fracture have on average a twofold increase in the risk of mortality (Center et al., 1999). Between 22% (Brauer et al., 2009) and 58% (Rapp et al., 2008) of patients with a hip fracture die within 12 months post fracture. The identification of high-risk individuals for early intervention is a major priority in the control of osteoporotic fractures in the general community. In randomized controlled trials, treating high-risk individuals reduces the risk of fracture (Black et al., 2020), though whether the reduction translates into reduced mortality risk remains contentious (Bolland et al., 2010; Cummings et al., 2019). However, the treatment uptake has been low, with only 40% of hip fracture patients in the United States being given osteoporosis treatment within 1 month after discharge in 2002, which was then halved in 2011 (Solomon et al., 2014). This undertreatment is considered a crisis in the management of osteoporosis globally.

Despite the excess mortality after fracture, mortality is not part of doctor-patient communication about treatment or risk assessment, because there is a lack of an intuitive method of conveying risk. Traditionally, relative risk (e.g. "the risk of mortality is increased by twofold") has been used as a metric of excess risk, but this metric often gives an exaggerated impression (Trevena et al., 2013), and is perceived as larger than the absolute risk (Akl et al., 2011). On the other hand, absolute risk in terms of probability over a period of time is much harder to understand (Zipkin et al., 2014), even for doctors and patients (Gigerenzer et al., 2007). The underappreciation of post-fracture mortality’s gravity has caused patients to be hesitant towards treatment and prevention, contributing to the crisis of osteoporosis management. Thus, there is an urgent need for a more informative metric to internalize the combined risks of fracture and mortality in a way that patients and doctors can comprehend easily.

The concept of ‘Effective Age’ (Spiegelhalter, 2016) is useful here. In engineering, effective age is the age of a structure based on its current conditions; whereas in medicine, the effective age of an individual is the age of a typical healthy person who matches the specific risk profile of this individual (Spiegelhalter, 2016). Depending on whether the risk profile is not healthy (i.e. presence of risk factors) or healthy (i.e. absence of risk factors), the effective age is older or younger, respectively than the chronological age (Spiegelhalter, 2016). The best-known effective age in medicine is ‘Heart Age’ or ‘Vascular Age’ (NHS Health Check, 2022) and ‘Lung Age’ (Morris and Temple, 1985). The use of heart age and lung age has resulted in a better clinical impact than the traditional absolute risk metric (Kulendrarajah et al., 2020).

We consider that patients with osteoporosis and the general public are not sufficiently informed about the risk of post-fracture mortality, and this might have contributed to the current crisis of under management of osteoporosis (Roux and Briot, 2020; Beaudoin et al., 2019). In an effort to improve the management of osteoporosis, we advance the idea of the ‘Skeletal Age’ which is conceptually defined as the age of one’s skeleton as a consequence of fragility fracture. This new metric is not only one that quantifies the impact of fracture on mortality but also one that captures the risk of fracture and the risk of post-fracture mortality.

The aim of this study was to analyze the relationship between fracture and mortality, and then translate this relationship into the ‘Skeletal Age’ for each fracture site by leveraging data from the Danish National Hospital Discharge Registry (NHDR). The NHDR data are ideal for this analysis because, apart from comorbidities at the individual patient level, it has documented the incidence of fractures and post-fracture mortality for the entire Danish population.

Methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Software, algorithmR Project for Statistical ComputingR Project for Statistical ComputingRRID:SCR_001905
Software, algorithmStata Statistical SoftwareA Software resource for statistical analysis and presentation of graphicsRRID:SCR_012763

Study design

The Danish National Hospital Discharge Registry can be viewed as a retrospective population-based cohort. This analysis included all adults aged 50 years old and older as of January 1, 2001 in Denmark whose health status had been followed up until December 31, 2016 for mortality. Individuals who had sustained a fracture at 45+ years old between 1996 and 2000 were excluded to avoid potential bias that the incident fracture analyzed in this study was a second fracture (Figure 1). This is not a clinical trial. This analysis (Statistics Denmark project number 706667) was approved by the National Board of Health, the Danish Data Protection Agency, and Statistics Denmark, and is subject to independent control and monitoring by The Danish Health Data Authority. Written informed consent is waived for routinely collected, pseudonymized registry data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Flowchart of recruitment and follow-up.

Ascertainment of fracture and mortality

The initial incident fracture was defined as the first low-trauma fracture reported between January 1, 2001 and December 31, 2014. When more than one fracture occurred during a single event, only the fracture at the most proximal site was considered. We used the International Statistical Classification of Disease and Related Health Problems, tenth version (ICD-10) codes to identify individuals with specific fracture sites including hip, femur, pelvis, vertebrae, humerus, rib, clavicle (collectively known as proximal fractures), forearm, lower leg, knee, ankle, foot and hand (collectively known as distal fractures) from the Danish NHDR (Supplementary file 1). All fractures included in the analysis were radiologically ascertained. Face, skull, finger, or toe fractures and high-trauma fractures due to traffic accidents were excluded.

The study participants were followed up to December 31, 2016 for mortality, allowing at least 2 years of follow-up post fracture. Death was ascertained from the Danish Register on Causes of Death.

Covariates assessment

The predefined covariates included age and comorbidities. We used the ICD-10 from the NHDR that includes any diagnosis documented between 1996 and 2000 to operationally define comorbidities at the study entry (i.e. January 1, 2001), and those within 5 years prior to the initial fracture to define comorbidities at fracture time. The severity of comorbidities was summarised using the Charlson comorbidity index (Quan et al., 2011).

Statistical analysis

Skeletal age is operationally defined as the sum of the chronological age and the change in life expectancy associated with fracture. The changes in life expectancy resulting from a specific fracture were computed incorporating (i) the association between individual fracture sites and mortality from a Cox’s proportional hazards regression and (ii) the baseline hazards described by Gompertz distribution and the population life expectancy from the national lifetable data (Kulinskaya et al., 2020).

For the first aim, a Cox’s proportional hazards regression was used to quantify the association between an initial specific fracture and mortality, in which both fracture and confounding variables were analyzed in a time-dependent manner to minimize the risk of immortal time bias (Suissa, 2008). The analyses did not account for the possibility of subsequent or recurrent fractures as this falls outside the scope of the study. The models adjusted for confounding effects of age and severity of comorbidities (Quan et al., 2011). Age and comorbidities at baseline and those at the time of fracture were also used. For individuals without fracture, the follow-up time was calculated from the study entry to the study end (December 31, 2016) or date of death, whichever came first; while their covariates at baseline were adjusted in the analysis model (Figure 2). By contrast, the follow-up time for those with an incident fracture was split into the pre-fracture (i.e. unexposed) and post-fracture (i.e. exposed) period for which the clustering effect was accounted for in the Cox’s proportional hazards regression. We conducted a time-dependent analysis to quantify the relationship between an incident exposure (i.e. an incident fracture) and the primary outcome of interest (i.e. post-fracture mortality) accounting for the risk of immortal time bias (Suissa, 2008). The pre-fracture period was the time interval between the study entry and date of fracture, and used the covariates at study entry; whereas the post-fracture period was between the date of fracture and date of death or study end, whichever came first, and included the covariates at the time of fracture. The proportional hazards assumption was graphically checked using the Schoenfeld’s residuals.

Schematic representation of time-dependent analysis.

For the second aim, we determined the skeletal age of an individual based on the individual’s specific risk profile. First, we calculated a prognostic index for an individual as a single-number summary of the combined effects of his/her specific fracture site and comorbidity severity (Henderson and Keiding, 2005). The prognostic index is a linear combination of the risk factors with weights derived from the regression coefficients. The individualized fracture-mortality association for an individual with a specific risk profile was then the difference between his/her prognostic index and the mean prognostic index of ‘typical’ people in the study population (Henderson and Keiding, 2005). In the second step, we used the Gompertz law of mortality and the Danish national lifetable data to transform the fracture-mortality association into life expectancy as a result of a fracture (Kulinskaya et al., 2020). The Gompertz law of mortality indicates the annual risk of dying at the age t can be expressed as ht=Bekt in which B~0.0000189 in women and 0.0000347 in men (Gompertz, 1825). Under the Gompertz law of mortality, the annual risk of dying associated with aging 1 year is remarkably consistent between ages 50 and 95 across ethnicities and over time (Brenner et al., 1993; Vaupel, 2010). Assuming the Gompertz baseline hazard, the log-hazards of mortality to specific fracture site i at the fracture time t is a+bt, where a and b are the estimated regression coefficients. The life expectancy for a specific fracture site at the fracture age z under the Gompertz distribution G(a,b) is obtained as eGa,bz=b-1exp(b-1ea)E1(b-1ea+bz)exp(-eab-1(ebz-1)) , where E1(b-1ea+bz) denotes the exponential integral. The loss of life years associated with specific fracture site was then calculated as the difference between the estimate life expectancy associated with fracture eGa,bz and the population life expectancy (Kulinskaya et al., 2020). The skeletal age for each individual site of fracture was the sum of the individual’s chronological age at the time of fracture and the loss of life years as a result of the fracture. The 95% CI of the skeletal age was computed using the confidence limits of the adjusted hazards ratios to account for uncertainty of the magnitude of association between individual site of fracture and mortality. The analyses were performed using Stata (Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC) and the R statistical environment on a Windows platform (R Development Core Team, 2020).

Results

Incidence of fractures

The present analysis was based on data from 793,815 men and 873,524 women aged 63.9 (standard deviation [SD] 10.3) and 65.5 (11.3) years old as of January 1, 2001, respectively (Figure 1). During a median of 14 years of follow-up (interquartile range [IQR]: 6.5, 14.0 years), 95,372 men and 212,498 women had sustained a fracture. The incidence of fractures was 10.9 fractures/1000 person-years (95% confidence interval [CI]: 10.8, 11.0) in men and 23.2 fractures/1000 person-years (23.1, 23.3) in women. Collectively, forearm, hip, and humerus fractures accounted for 55% of all fractures in men and 70% in women.

As expected, men and women with a fracture were on average older than those without a fracture. Individuals with a fracture had more comorbidities than those who did not sustain a fracture (Table 1). For instance, the prevalence of myocardial infarction among patients with a fracture (9.3% in men and 4.7% in women) was higher than that among individuals without a fracture (4% in men and 1.5% in women). Similar trend was also observed for stroke, diabetes, and cancer.

Table 1
Characteristics of the study population at baseline stratified by gender, fracture, and mortality status.
CharacteristicNon-fracture groupFracture group
AliveDeadAliveDead
Men
Number of individuals466,936231,50754,35541,017
Age at baseline159.8 (7.7)71.1 (10.4)61.3 (8.9)72.4 (10.4)
Age at fracture168.2 (9.9)77.8 (10.5)
Comorbidities*
Charlson comorbidity index20 (0, 0)0 (0, 1)0 (0, 1)1 (0, 2)
 0404,415 (86.7%)140,293 (60.6%)32,287 (59.4%)12,592 (30.7%)
 1–256,701 (12.1%)70,841 (30.6%)17,557 (32.3%)19,073 (46.5%)
 3–43,328 (0.7%)9,260 (4.0%)2,881 (5.3%)5,209 (12.7%)
 5+2,492 (0.5%)1,112 (4.8%)1,631 (3.0%)4,143 (10.1%)
Specific comorbidities3
Myocardial infarction18,734 (4.0%)22,225 (9.6%)5,055 (9.3%)6,071 (14.8%)
Congestive heart failure12,814 (2.7%)26,623 (11.5%)5,109 (9.4%)8,655 (21.1%)
Stroke21,792 (4.7%)31,253 (13.5%)9,458 (17.4%)11,567 (28.2%)
Peripheral vascular disease10,912 (2.3%)18,058 (7.8%)4,240 (7.8%)5,660 (13.8%)
Arrhythmias10,835 (2.3%)18,289 (7.9%)4,892 (9.0%)5,747 (18.4%)
Hypertension14,340 (3.1%)15,279 (6.6%)7,718 (14.2%)7,875 (19.2%)
Diabetes mellitus with no chronic complications19,805 (4.2%)21,067 (9.1%)6,305 (11.6%)6,317 (15.4%)
Diabetes mellitus with chronic complications6,120 (1.3%)8,103 (3.5%)2,446 (4.5%)2,748 (6.7%)
Renal disease4,290 (0.9%)8,797 (3.8%)2,120 (3.9%)3,609 (8.8%)
Cancer26,126 (5.6%)43,523 (18.8%)8,914 (16.4%)11,813 (28.8%)
Metastatic tumors1,894 (0.4%)9,492 (4.1%)1,033 (1.9%)3,035 (7.4%)
Dementia2,869 (0.6%)7,640 (3.3%)2,663 (4.9%)6,071 (14.8%)
Chronic lung disease16,600 (3.6%)25,697 (11.1%)6,740 (12.4%)8,696 (21.2%)
Rheumatologic4,881 (1.1%)4,630 (2.0%)1,685 (3.1%)1,600 (3.9%)
Mild liver disease2,647 (0.6%)3,704 (1.6%)1,468 (2.7%)1,805 (4.4%)
Moderate/severe liver disease589 (0.1%)1,389 (0.6%)435 (0.8%)779 (1.9%)
Hemiplegia661 (0.1%)927 (0.4%)461 (0.9%)480 (0.9%)
HIV/AIDS108 (0.02%)81 (0.03%)37 (0.07%)23 (0.06%)
Women
Number of individuals465,559195,467130,77181,727
Age at baseline160.5 (8.3)74.0 (11.1)63.9 (9.7)76.3 (9.9)
Age at fracture170.9 (10.2)81.3 (9.6)
Comorbidities*
Charlson comorbidity index20 (0, 0)0 (0, 1)0 (0, 1)1 (0, 2)
 0400,005 (85.9%)121,776 (62.3%)81,574 (62.4%)32,093 (39.3%)
 1–259,923 (12.9%)57,663 (29.5%)41,114 (31.4%)35,866 (44.0%)
 3–42,975 (0.6%)6,255 (3.2%)5,196 (4.0%)7,960 (9.8%)
 5+2,656 (0.6%)9,773 (5.0%)2,887 (2.2%)5,695 (7.0%)
Specific comorbidities3
Myocardial infarction7,164 (1.5%)10,946 (5.6%)6,109 (4.7%)7,562 (9.3%)
Congestive heart failure7,307 (1.6%)18,374 (9.4%)8,056 (6.2%)13,615 (16.7%)
Stroke16,329 (3.5%)22,870 (11.7%)17,143 (13.1%)19,247 (23.6%)
Peripheral vascular disease6,517 (1.4%)10,555 (5.4%)6,196 (4.7%)7,309 (9.0%)
Arrhythmias7,671 (1.7%)13,096 (6.7%)7,947 (6.1%)11,440 (14.0%)
Hypertension12,910 (2.8%)12,510 (6.4%)17,970 (13.7%)15,727 (19.3%)
Diabetes mellitus with no chronic complications14,151 (3.0%)14,074 (7.2%)9,960 (7.6%)9,214 (11.3%)
Diabetes mellitus with chronic complications3,367 (0.7%)4,496 (2.3%)2,969 (2.3%)3,126 (3.8%)
Renal disease2,386 (0.5%)4,300 (2.2%)2,697 (2.1%)3,561 (4.4%)
Cancer30,913 (6.6%)34,989 (17.9%)20,246 (15.5%)18,785 (23.0%)
Metastatic tumors2,302 (0.5%)8,796 (4.5%)2,221 (1.7%)4,632 (5.7%)
Dementia3,035 (0.7%)7,428 (3.8%)7,169 (5.5%)13,136 (16.1%)
Chronic lung disease18,759 (4.0%)20,915 (10.7%)15,065 (11.5%)14,018 (17.2%)
Rheumatologic10,392 (2.2%)7,037 (3.6%)7,482 (5.7%)5,256 (6.4%)
Mild liver disease2,599 (0.6%)2,346 (1.2%)2,165 (1.7%)1,802 (2.2%)
Moderate/severe liver disease296 (0.1%)782 (0.4%)422 (0.3%)643 (0.8%)
Hemiplegia524 (0.1%)573 (0.3%)577 (0.4%)475 (0.6%)
HIV/AIDS14 (0.0%)9 (0.0%)12 (0.01%)4 (0.0%)
  1. Notes: 1: mean (SD); 2: median (IQR); 3: number (%). *: at baseline for individuals without fracture or at fracture time for fracture patients. Comorbidities included not only chronic diseases that require hospitalization, but also those documented as either secondary diagnoses or at outpatient or emergency visits.

Incidence of mortality

During a median follow-up of 6.5 years (6.0 years (IQR: 2.5, 10.0) in men, 6.7 years (3.3, 10.8) in women), 272,524 men and 277,194 women had died, yielding unadjusted mortality incidence rates of 6.55 deaths/100 person-years (95% CI: 6.48, 6.61) in men and 5.42 deaths/100 person-years (5.39, 5.46) in women. More importantly, the unadjusted rate of mortality among patients with a fracture (6.54/100 person-years in men and 5.42/100 person-years in women) was greater than among those without a fracture (2.56/100 person-years in men and 2.24/100 person-years in women). Analysis by fracture site revealed that men and women with a fracture at the hip, pelvis, or vertebra had a much greater risk of mortality than other fractures (Table 2).

Table 2
Incidence of mortality following specific fracture sites stratified by gender.
FractureNumberAge at fracture (years)Number of deathsFollow-up time (person-years)Rate* of mortality (95% CI)
Men
No fracture698,44363.6** (10.2)231,5079,049,1942.6% (2.5, 2.6)
Any fracture95,37272.3 (11.2)41,017626,7336.5% (6.5, 6.6)
Hip fracture25,70679.5 (9.7)16,890107,78915.7% (15.4, 15.9)
Femur1,91074.5 (10.8)99710,4689.5% (8.9, 10.1)
Pelvis1,30577.0 (10.9)7516,46511.6% (10.8, 12.5)
Vertebrae6,92472.7 (10.5)3,12541,7467.5% (7.2, 7.7)
Humerus10,12672.6 (10.7)4,82762,3507.7% (7.5, 8.0)
Rib6,86769.7 (10.4)2,35050,1864.7% (4.5, 4.9)
Clavicle5,20168.2 (10.5)1,63639,7314.1% (3.9, 4.3)
Lower leg6,29667.0 (9.5)1,79652,9153.4% (3.2, 3.6)
Forearm15,26869.4 (10.2)4,675120,7193.9% (3.8, 4.0)
Knee1,38270.2 (10.1)44710,8864.1% (3.7, 4.5)
Ankle1,08467.6 (9.8)2759,0813.0% (2.7, 3.4)
Hand8,30967.7 (10.1)2,17870,2633.1% (3.0, 3.2)
Foot4,99465.3 (9.7)1,07044,1362.4% (2.3, 2.6)
Women
No fracture661,02664.5** (11.1)195,4678,723,8632.2% (2.2, 2.3)
Any fracture212,49874.9 (11.2)81,7271,506,9405.4% (5.4, 5.5)
Hip fracture51,66982.1 (9.2)31,816259,08712.3% (12.1, 12.4)
Femur3,52978.9 (10.9)1,92119,3229.9% (9.5, 10.4)
Pelvis4,92081.2 (10.0)2,89326,76110.8% (10.4, 11.2)
Vertebrae9,73276.6 (10.6)4,65860,2957.7% (7.5, 7.9)
Humerus28,29874.4 (10.5)10,330202,6625.1% (5.0, 5.2)
Rib3,15374.1 (11.9)1,21822,2855.5% (5.2, 5.8)
Clavicle4,48072.7 (11.5)1,57731,7555.0% (4.7, 5.2)
Lower leg11,46070.1 (10.8)3,13094,3083.3% (3.2, 3.4)
Forearm68,33372.0 (10.3)18,470559,4063.3% (3.2, 3.3)
Knee2,60471.3 (9.7)62921,0853.0% (2.8, 3.2)
Ankle1,93670.0 (10.6)49915,7263.2% (2.9, 3.5)
Hand12,48969.6 (10.2)2681108,1122.5% (2.4, 2.6)
Foot9,91667.8 (9.6)190486,1362.2% (2.1, 2.3)
  1. Notes: *: rates were calculated as number of deaths/100 person-years; **: age at baseline (years).

However, the above observation could be confounded by age and comorbidities, because individuals with a fracture were on average older and had more comorbidities than those without a fracture. Therefore, we employed multivariable-adjusted Cox’s proportional hazards model to estimate the strength of the association between fracture and mortality, adjusting for age and severity of comorbidities (Figure 3). Individuals with any fragility fracture were associated with a 30–45% increased hazard of death (adjusted hazard ratio = 1.46 (95% CI: 1.45, 1.48) in men; 1.28 (1.27, 1.30) in women). For the same age and comorbidity profile, men and women with a hip or femur fracture had an almost twofold greater risk of death than those without a fracture.

Association between specific fracture and mortality risk, adjusted for age and severity of comorbidities*: hazard ratio and 95% confidence interval (CI).

Legend * severity of comorbidities is assessed using the Charlson Comorbidity Index. Hazard ratio and the corresponding 95% confidence interval were estimated from a multivariable-adjusted Cox's proportional hazards regression.

The increased risk of mortality was also observed among those with a proximal fracture, such as the pelvis, vertebrae, humerus, rib, clavicle, and lower leg fracture after controlling for age and comorbidities. However, there was no significantly increased risk of death following a forearm, knee, ankle, hand, or foot fracture (Figure 3).

Skeletal age

Based on the association between fracture and mortality and using the Gompertz law, we estimated the skeletal age for each fracture site and each chronological age (from the age of 50) in men and women (Figure 4; Supplementary file 2). Patients with a fracture - any fracture at all - have lost years of life, and hence their skeletal age was greater than their chronological age. For a given age, men with a fracture on average had greater skeletal age than women by approximately 1 year. In either sex, the loss of years of life was more pronounced in younger age groups and gradually converged in the older age groups.

Skeletal age by specific fracture site and chronological age at fracture.

As expected, patients with a hip fracture had the highest skeletal age than those with other fractures. For example, a 70-year-old man who had sustained a hip fracture would have a skeletal age of 75 years (i.e. a loss of 5 years of life); however, if a 50-year-old man with a hip fracture would have a skeletal age of 56.8 years which is equivalent to almost 7 years of life lost. Other fractures such as the femur, pelvis, vertebrae, and humerus also signified a significant loss of years of life (around 5 years); thus, 50-year-old patients with one of these fractures are estimated to have a skeletal age of around 55 years.

However, fractures are the rib, clavicle, and lower leg were associated with lower years of life lost, and the skeletal age of patients with one of these fractures was generally lower than patients with a more serious fracture (e.g. hip fracture). For instance, a 60-year-old patient with a lower leg fracture would be expected to have a skeletal age of 62.4 years for men or 61.9 years for women (Supplementary file 3).

Discussion

It has been well established that fracture, especially hip fracture, is associated with an increased risk of mortality, and this excess risk is commonly expressed in terms of the relative risk metric. Here, we proposed a new effective age metric called ‘Skeletal Age’ to quantify the impact of fracture on mortality. Using data from a nationwide cohort of 1.7 million adults aged 50+ years old in Denmark, we showed that almost all types of fracture were associated with a loss of years of life, indicating that the skeletal age of individuals suffering from a fracture is greater than their chronological age. This finding has important implications that we discuss below.

Our finding confirmed the previous studies (Tran et al., 2018; Bliuc et al., 2009; Browner et al., 1996; Melton et al., 2013; Haentjens et al., 2010; Tran et al., 2017) that patients with a fragility fracture were associated with a significantly greater risk of mortality than their similarly aged and gender counterparts without fracture who had the same comorbidity profile. Our finding is also consistent with an earlier Danish study demonstrating reduced life expectancy in patients, with or without fractures, at the time of beginning osteoporosis treatment (Abrahamsen et al., 2015). However, the magnitude of the association between individual sites of fracture observed in our study is slightly lower than that documented in several previous cohort studies (Bliuc et al., 2009; Browner et al., 1996; Melton et al., 2013; Haentjens et al., 2010; Tran et al., 2017), with hip fractures being associated with a 1.7-to-6-fold increased risk of death (Haentjens et al., 2010). This discrepancy could be due to the difference in study populations. While cohort studies usually recruit healthy participants, our national registry-based study was able to document all individuals in Denmark. As a result, our control group (i.e. those without a fracture) included individuals in poor health conditions with multiple comorbidities who are usually unable to be recruited in a cohort study. Whether the association between fracture and mortality is causal or not is a matter of contention. It is commonly assumed that the association is confounded by comorbidities, but multiple studies have found that comorbidities contributed little to the excess risk of mortality among fractured patients (Vestergaard et al., 2007; Chen et al., 2018). In the present analysis, we have also adjusted for comorbidities, and the significant fracture – mortality association remained unchanged, suggesting that the association is unlikely due to comorbidities.

Regardless of the nature of the association, mortality is not a component of doctor-patient communication. Existing fracture risk assessment models such as Garvan (Nguyen et al., 2007) and FRAX (Kanis et al., 2007) provide the estimated probability (i.e. absolute risk) of fracture over a 10 year period without any information on mortality. However, doctors and patients have difficulty in understanding and interpreting probability (Gigerenzer et al., 2007). Only a fifth of a sample of highly educated American adults could understand one in 1000 is equivalent to 0.1% (Lipkus et al., 2001). Conveying a low, though clinically significant absolute risk (e.g. ‘Your risk of hip fracture over the next 10 years is 5%’) might provide a false peace of mind and underestimated risk perception, leading to increasing possibilities of refusing the recommended treatment (Trevena et al., 2013). Thus, poor communication about fracture risk and mortality consequences might have contributed to the global crisis of undertreatment of osteoporosis.

Based on the concept of effective age, we proposed the idea of ‘Skeletal Age’ as a new metric for communicating the risk of mortality following a fracture. Unlike the current fracture risk assessment tools (Beaudoin et al., 2019) which estimate the probability of fracture over time using probability-based metrics, such as relative risk and absolute risk, skeletal age quantifies the consequence of a fracture using a natural frequency metric. A natural frequency metric has been consistently shown to be easier and more friendly to doctors and patients than the probability-based metrics (Akl et al., 2011; Gigerenzer et al., 2007; Ahmed et al., 2012). It is not straightforward to appreciate the importance of the twofold increased risk of death (i.e. relative risk = 2.0) without knowing the background risk (i.e. twofolds of 1% would remarkably differ from twofolds of 10%). By contrast, for the same twofold mortality risk of hip fracture, telling a 60 year man with a hip fracture that his skeletal age would be 66 years old, equivalent to a 6 year loss of life, is more intuitive. The skeletal age of 66 years old can also be interpreted as the individual being in the same risk category as a 66-year-old with ‘favorable risk factors’ or at least the ones that are potentially modifiable. Hence, an older skeletal age also means a greater risk of fracture. In addition, the skeletal age can be also used to convey the possible benefit of treatment, providing an alternative metric to the conventional metrics such as relative risk or relative risk reduction. For instance, a patient might find the statement ‘Zoledronic acid treatment helps a patient with a hip fracture gain 3 years of life’ much easier to understand and probably more persuasive than ‘Zoledronic acid treatment reduced the risk of death by 28%’ (Lyles et al., 2007).

Skeletal age is an addition to the already available metrics such as ‘Heart Age,’ ‘Lung Age,’ and more recently ‘Covid Age’ which have been demonstrated to be superior to the conventional risk metrics (Grover et al., 2007; Lowensteyn et al., 1998; Lopez-Gonzalez et al., 2015; Bonner et al., 2015; Svendsen et al., 2020) in terms of positive behavioral changes. For instance, compared with usual care, individuals who were given heart age had a greater smoking cessation rate (Lopez-Gonzalez et al., 2015; Bonner et al., 2015), substantial improvement in weight, body mass index, and physical activity (Lopez-Gonzalez et al., 2015), and a greater proportion of high-risk patients returning for a follow-up appointment (Lowensteyn et al., 1998). The use of heart age also led to significantly greater improvement at 12 months in metabolic parameters (Lopez-Gonzalez et al., 2015) or lipid profiles (Grover et al., 2007). A recent cluster randomized controlled trial in Norway also found that heart age was a good way to communicate cardiovascular risk and to motivate individuals to reduce cardiovascular risk factors (Svendsen et al., 2020). Similarly, the use of lung age could also lead to a clinically significant increase in the reported smoking cessation rates (Kaminsky et al., 2011; Takagi et al., 2017; Parkes et al., 2008). Collectively, these data suggest that effective age metrics can help patients better understand their risk and lead to preventive changes. A web-based calculator ‘BONEcheck.org’ has been developed to assist healthcare professionals and the public to estimate the risk of fracture and skeletal age for an individual with specific risk profiles, potentially improving doctor-patient risk communication.

Our findings should be interpreted within the context of their strengths and limitations. First, we included a whole-country population with long follow-up and robust diagnostic data, minimizing potential selection bias and misclassification (Frank, 2000), and allowing us to examine mortality risk following individual fracture sites. Second, the current study analyzed both fracture and covariates, such as aging and the presence of comorbidities in a time-dependent manner, making it statistically rigorous in minimizing an immortal time bias and sufficiently accounting for confounding effects. The analysis thus provided accurate estimates of the association between specific fracture sites and mortality risk as it was able to control for confounding effects at both the study entry and the time of fracture.

However, the use of registry-based data, originally documented and coded for administrative or reimbursement purposes is prone to variable data accuracy, the lack of specific clinical information, and non-medical factors (Mazzali and Duca, 2015). Fortunately, all essential study variables (i.e. comorbidities, fracture, and death) were systematically obtained from the Danish NHDR that includes excellent, complete medical records and precise diagnoses for all individuals living in Denmark since 1995 (Andersen et al., 1999; Vestergaard and Mosekilde, 2002). The use of diagnosis-based comorbidities, though possibly associated with lower sensitivity than the self-reported ones (Lujic et al., 2017) is able to capture the severe health disorders that prompt the participants to seek for medical assistance. Second, the analyses could not make adjustments for lifestyle factors and physical activity, which are usually obtained in a cohort study. As they are closely related to aging and the presence of comorbidities which were already accounted for in our analysis, making further adjustments for these lifestyle factors is unlikely to substantially modify the findings. Third, mild, asymptomatic fractures or chronic diseases that did not require medical attention might not be registered in the Danish NHDR. It is nevertheless unlikely that these unregistered fractures and comorbidities would modify the findings significantly, as previous studies have indicated a high concordance between self-reported fractures and those registered in the NHDR (Hundrup et al., 2004), and a high positive predictive value of the self-reported comorbidities (Lujic et al., 2017). Finally, our analysis did not include the examination of the effect of bone metabolism-affecting medications on the assessment of skeletal health. However, these medications were prescribed to only a few individuals aged 50 years or over in Denmark (Vestergaard et al., 2005), even for those with a hip fracture (Roerholt et al., 2009), and there is no strong evidence from randomized controlled trials of their impact on post-fracture mortality (Cummings et al., 2019).

In conclusion, we advanced the concept of skeletal age as the age of an individual’s skeleton resulting from a fragility fracture. Unlike existing metrics (e.g. relative risk, probability of fracture), skeletal age combines the risk that an individual will sustain a fracture and the risk of mortality once a fracture has occurred, making the doctor-patient communication more intuitive and possibly more effective. Given the evidence of the successful implication of similar effective age metrics, skeletal age is expected to improve risk communication and ultimately improve treatment uptake among patients who are indicated for treatment. A randomized controlled trial aiming to compare the use of skeletal age and the current metrics in fracture risk communication is warranted.

Data availability

The study used data from the Danish National Patient Register which individual patient data cannot be shared without authorised access, as per Danish Legislation (https://www.dst.dk/en/TilSalg/Forskningsservice/Dataadgang). To be able to access data interested researchers need to be employees of Danish university, university hospital or other specified Danish organizations (https://www.dst.dk/ext/594574631/0/forskning/Regler-for-adgang-til-pseudonymiserede-mikrodata-under-Danmarks-Statistiks-forskerordninger--pdf). Non-Danish citizens may obtain access if employed at said institutions. The interested researchers will need to be registered as an authorized research group under Statistics Denmark (https://www.dst.dk/da/TilSalg/Forskningsservice/Dataadgang/Autorisering) and fulfill the criteria mentioned above. The research proposal expected to present the specific aims and the feasibility is examined and approved by the Danish Data Protection Agency and the Danish Health Data Authority [https://sundhedsdatastyrelsen.dk/da/forskerservice/ansog-om-data]. However, commercial research cannot be performed (https://www.dst.dk/ext/594574631/0/forskning/Regler-for-adgang-til-pseudonymiserede-mikrodata-under-Danmarks-Statistiks-forskerordninger--pdf). Deidentified data may not be removed from the server (https://www.dst.dk/da/TilSalg/Forskningsservice/hjemtagelse-af-analyseresultater). Aggregated data can only be presented if no individuals can be identified (e.g., counts in a contingency table must not be below five). Supplementary File 1 provides the ICD-10 codes from the Danish National Patient Discharge Register used to define fractures and comorbidities in this analysis. The aggregated data and the R analysis codes utilized to estimate skeletal age for each individual fracture site and to generate the graphs and charts in the manuscript (Skeletal-Age.html), and the analysis output (Skeletal Age_Rmarkdown.pdf) are publicly accessible (https://github.com/ThachSTran/Skeletal-Age.git, copy archived at swh:1:rev:db2c39ebf11f55612a745140ea078d7afa279bb0).

References

    1. Andersen TF
    2. Madsen M
    3. Jørgensen J
    4. Mellemkjoer L
    5. Olsen JH
    (1999)
    The Danish national Hospital register. A valuable source of data for modern health sciences
    Danish Medical Bulletin 46:263–268.
  1. Software
    1. R Development Core Team
    (2020) R: A language and environment for statistical computing
    R Foundation for Statistical Computing, Vienna, Austria.

Decision letter

  1. Luigi Gennari
    Reviewing Editor; University of Siena, Italy
  2. Mone Zaidi
    Senior Editor; Icahn School of Medicine at Mount Sinai, United States
  3. Elena Tsourdi
    Reviewer; University Clinic Dresden, Germany

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "'Skeletal Age' as a measure of the impact of fracture on mortality: a nationwide population-based study" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Mone Zaidi as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Elena Tsourdi (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) Better clarification of study design and statistical analysis should be required in order to address the conceptual problems raised by Reviewer 1 (e.g. concerning the assessment of comorbidities, or how the information about low-energy fracture obtained from ICD registries, or whether the risk of a second fracture/recurrent fractures was taken into account in the analyses). Additional information is also required about the identification of fractures, particularly vertebral fractures, as outlined by Reviewer 2 (e.g. were all fractures radiographically confirmed?).

2) As underlined by Reviewer 1, the analysis performed did not show that individuals with and without fractures matched for co-morbidities had different mortality risks. Given the large and comprehensive sample, this could be explored in more detail.

3) Additional information is required to better explain how the assessment of skeletal age may capture the risk of a future fracture.

4) The discussion should be more balanced, since, apart from a zoledronic acid trial and the quoted metanalysis by Bolland et al. (JCEM 2010;95:1174-81), there is no clear information from randomized controlled trials demonstrating a significant effect of treatments for osteoporosis on mortality. Indeed, as also outlined by Reviewer 2, it seems that the impact of medications affecting bone metabolism on the assessment of skeletal health (osteoporosis drugs, glucocorticoids, PPI, aromatase-inhibitors, etc.) in the Danish National Hospital Discharge Registry was not considered. Moreover, in their conclusive remarks, the Authors propose to replace the fracture probability with skeletal age for individualized fracture risk assessment. Again, given the limited information about the effects of anti-osteoporotic agents on mortality, I'd suggest somehow recommending the "integration" rather than "replacement" of the individual fracture risk assessment with skeletal age, at least at this stage.

Reviewer #1 (Recommendations for the authors):

General comments:

The individuals who fracture have more co-morbidities with high mortality risk than the individuals who do not. In order to understand the contribution of a fracture to mortality risk, it would be more appropriate to establish populations of individuals with and without fracture matched with respect to age, gender, and co-morbidities and compare mortality between these groups.

Specific comments:

Results:

l.151-: As the period of fracture identification was from 2001 to 2014, how can the median follow-up be 16 years?

l.163-: Why was the follow-up period for mortality shorter than for fracture? To make the occurrence of fracture and death comparable, I suggest providing both per 100 or 1000 patient-years.

l.166-: It should be clearly stated that the mortality rates in individuals with and without fractures are unadjusted.

Material and methods:

l.320-: Why only data to 2016, this is 6 years ago?

l.330-: How was low-energy fracture determined? Fracture mechanism is usually not captured in registers?

l.343-: Were only comorbidities leading to hospitalisation registered? What about comorbidities diagnosed and treated by GPs?

l.363-: it is not clear how the mortality rate pre-fracture in the patients who fractured was taken into account. The individuals had a higher prevalence of many dead-prone comorbidities and would therefore most likely also have had a higher risk of death before the fracture than the no-fracture population. In addition, how was the risk of a second fracture/recurrent fractures taken into account in the analyses?

Figure 2: the HR was adjusted for the Charlson Co-morbidity index, not specific co-morbidities.

Reviewer #2 (Recommendations for the authors):

A few additional points which should be addressed too:

Ascertainment of fractures: The authors should provide some information as to whether these fractures (esp. vertebral fractures) were self-reported or radiographically assessed. In addition to this, was it ascertained that these were fragility fractures?

Patients: Did the authors have any information on the BMD of these subjects? In addition, were medications that affect bone metabolism (osteoporosis drugs, glucocorticoids, PPI, aromatase inhibitors, etc.) documented? If so, why were they not imputed in Cox's proportional hazards model?

https://doi.org/10.7554/eLife.83888.sa1

Author response

Essential revisions:

1) Better clarification of study design and statistical analysis should be required in order to address the conceptual problems raised by Reviewer 1 (e.g. concerning the assessment of comorbidities, or how the information about low-energy fracture obtained from ICD registries, or whether the risk of a second fracture/recurrent fractures was taken into account in the analyses). Additional information is also required about the identification of fractures, particularly vertebral fractures, as outlined by Reviewer 2 (e.g. were all fractures radiographically confirmed?).

Thank you for your suggestions. We have now clarified the study design and statistical analysis.

Comorbidities were ascertained based on the primary diagnosis and all secondary diagnoses of chronic health disorders in the Danish National Hospital Discharge Registry (NHDR) that includes a complete record of diagnoses for all patients treated at Danish hospitals, whether on an outpatient or an inpatient basis. The time frame was within 5 years of the recruitment (i.e., 1996-2000) or of the index fracture. Therefore, comorbidities included not only chronic diseases in the primary diagnosis that require medical attention but also those documented as either secondary diagnoses or at outpatient or emergency visits.

The 5-year window allowed us to capture important chronic diseases. Importantly, the registry-based data of chronic diseases have been found to yield higher positive predictive value than the self-reported ones1, indicating the accuracy of the registry-based data in capturing of chronic diseases.

All fractures were ascertained based on the primary diagnosis in the Danish NHDR that required medical attention between 2001 and 2014. All fractures were radiologically confirmed. The fractures included in the current analysis were considered low trauma, as we excluded all high-trauma fractures, including skull, face, nose, finger or toes fractures or those due to traffic accidents.

The nationwide Danish NHDR has included all medical contacts in Denmark since 1995 with excellent completeness of medical records and precision of diagnoses2,3. The concordance between fracture reports in the NHDR and patient files was 97%3. Importantly, our Danish co-authors (Prof. Bo Abrahamsen, Prof. Peter Vestergaard and Dr. Louise Hansen), who are health-care experts in Denmark confirm that fractures in Denmark are always treated in hospitals, unless they are asymptomatic or overlooked such as some osteoporotic vertebral compressions or small "march" fractures of the smaller bones in the feet which are beyond our analysis scopes. Additionally, previous studies4 have shown that self-reported fractures among Danish female nurses were highly concordant, up to 84% for upper hand or wrist fractures and 94% for hip fractures with fractures registered in the NHDR. A sentence for fracture ascertainment has been added:

“All fractures included in the analysis were radiologically ascertained.”

We have then discussed it as a limitation of the study by adding the following sentences to the current manuscript:

“Thirdly, mild, asymptomatic fractures or chronic diseases that did not require medical attention might not be registered in the Danish NHDR. It is nevertheless unlikely that these unregistered fractures and comorbidities would modify the findings significantly, as previous studies have indicated a high concordance between self-reported fractures and those registered in the NHDR45 and high positive predictive value of the self-reported comorbidities44.”

The analysis did not account for the possibility of subsequent or recurrent fractures, which is indeed beyond the study scope which primarily aimed to quantify the relationship between individual sites of an initial fracture and mortality risk. Mortality risk among patients with a fracture was compared with that of non-fracture individuals, accounting for confounding effect of age and comorbidity severity. Instead, a study that aims to examine the relationship between subsequent or recurrent fractures and mortality should be conducted in a study population of patients who have already sustained an initial fracture. We have added the following sentence to the Statistical analysis section:

“For the first aim, a Cox’s proportional hazards regression was used to quantify the association between an initial specific fracture and mortality, in which both fracture and confounding variables were analysed in a time-dependent manner to minimise the risk of immortal time bias50. The analyses did not account for the possibility of subsequent or recurrent fractures as this falls outside of the scope of the study. The models adjusted for confounding effects of age and severity of comorbidities48. Age and comorbidities at baseline and those at the time of fracture were also used.”.

2) As underlined by Reviewer 1, the analysis performed did not show that individuals with and without fractures matched for co-morbidities had different mortality risks. Given the large and comprehensive sample, this could be explored in more detail.

We agree that a matched design could be a possible design option. However, such a matched design might be underpowered as it is not always straightforward to obtain sufficient matched pairs to make the planned analysis robust, especially for the analysis that aims to examine mortality risk following each individual fracture site.

We consider that the multivariable model that we used is methodologically rigorous and statistically equivalent to the matched design in the control for confounding effects. Specifically, the multivariable-adjusted Cox’s proportional hazard regression used in this study is capable of accounting for the confounding effect of age and comorbidities as a matched design can do. Moreover, the adjusted analysis allows us to make use of data of all participants in the study population, ultimately improving the analysis power and robustness that a matched design might not be able to achieve.

3) Additional information is required to better explain how the assessment of skeletal age may capture the risk of a future fracture.

Thank you for your suggestion. Traditional fracture risk assessment tools (i.e., FRAX, Garvan or QFracture)5 provide an estimated absolute risk of fracture over a period of time. By contrast, skeletal age reflects the excess mortality as a consequence of a fracture for an individual with a specific risk profile. For instance, a 60-year man with a hip fracture is estimated to the skeletal age of a 66-year old, indicating a 6-year loss of life. This operational definition is consistent with the definition of 'Effective Age' in the medical literature6.

We have added more details to explain how the assessment of skeletal age would provide the conceptual risk of both fracture and post-fracture mortality as follows:

“Unlike the current fracture risk assessment tools17 which estimate the probability of fracture over a period of time using probability-based metrics, such as relative risk and absolute risk, skeletal age quantifies the consequence of a fracture using a natural frequency metric. A natural frequency metric has been consistently shown to be easier and more friendly to doctors and patients than the probability-based metrics9 11 30. It is not straightforward to appreciate the importance of the two-fold increased risk of death (i.e., relative risk = 2.0) without knowing the background risk (i.e., 2 folds of 1% would remarkably differ from 2 folds of 10%). By contrast, for the same 2-fold mortality risk of hip fracture, telling a 60-year man with a hip fracture that his skeletal age would be 66 years old, equivalent to a 6-year loss of life, is more intuitive. The skeletal age can also be interpreted as the individual being in the same risk category as a 66-year-old with 'favorable risk factors' or at least the ones that are potentially modifiable. Hence, an older skeletal age means a greater risk of fracture.”.

4) The discussion should be more balanced, since, apart from a zoledronic acid trial and the quoted metanalysis by Bolland et al. (JCEM 2010;95:1174-81), there is no clear information from randomized controlled trials demonstrating a significant effect of treatments for osteoporosis on mortality.

Thank you for your suggestion. We agree that the effect of osteoporosis treatments on mortality remains contentious, but this is not our primary focus. Our focus is to demonstrate the use of skeletal age to convey a positive impact of a treatment, and the mention of zoledronic acid's effect was simply for illustration purpose. We have decided to remove the section related to the benefit of pharmacological treatment on post-fracture mortality.

Indeed, as also outlined by Reviewer 2, it seems that the impact of medications affecting bone metabolism on the assessment of skeletal health (osteoporosis drugs, glucocorticoids, PPI, aromatase-inhibitors, etc.) in the Danish National Hospital Discharge Registry was not considered.

Our analysis could not factor in the impact of medications that impact bone metabolism. Nevertheless, these medications unlikely have a substantially effect on the risk of post-fracture mortality. First, only 0.3% of men and 2.2% of women aged 50+ in Denmark were prescribed bisphosphonate, and bisphosphonate or raloxifene, respectively in 19997, while only 4.1% of men and 9.2% of women with hip fracture began bisphosphonates in 20048. Secondly, the most recent meta-analysis of 38 randomised clinical trials of drug therapies that included 101,642 patients with osteoporosis suggested these bone metabolism-affecting medications were not associated with improved survival (RR: 0.98; 95% CI: 0.91 to 1.05; I2= 0%)9. Finally, it is not ideal to examine the impact of a pharmacological intervention using our observational data which are inevitably confounded by uncontrolled biases and residual confounding effects.

We have acknowledged it as a study limitation as follows:

“Finally, our analysis did not include the examination of the effect of bone metabolism-affecting medications on the assessment of skeletal health. However, these medications were prescribed to only a few individuals aged 50 and over in Denmark46, even for those with hip fractures47, and there is no substantial evidence from randomized controlled trials of their impact on post-fracture mortality6.”

Moreover, in their conclusive remarks, the Authors propose to replace the fracture probability with skeletal age for individualized fracture risk assessment. Again, given the limited information about the effects of anti-osteoporotic agents on mortality, I'd suggest somehow recommending the "integration" rather than "replacement" of the individual fracture risk assessment with skeletal age, at least at this stage.

We agree with the reviewer that the assessment of skeletal age should not 'replace' the current fracture risk assessment. We have removed the word 'replace' from the conclusion statement.

Reviewer #1 (Recommendations for the authors):General comments:

The individuals who fracture have more co-morbidities with high mortality risk than the individuals who do not. In order to understand the contribution of a fracture to mortality risk, it would be more appropriate to establish populations of individuals with and without fracture matched with respect to age, gender, and co-morbidities and compare mortality between these groups.

We have commented on the issue of matching versus multivariable regression model. Both approaches are designed to control for the effect of potential confounders, but each approach has strengths and limitations, with the matching design often having lower power than the regression adjustment. Some studies have shown that Cox’s regression model applied to the entire cohort is statically more powerful than a matched design. In our study, we chose Cox’s model to factor in the effects of age, gender, and comorbidities on the fracture-mortality association. This analytic strategy also allowed us to estimate the skeletal age for each individual based on the individual’s comorbidity profile.

Specific comments:

Results:

l.151-: As the period of fracture identification was from 2001 to 2014, how can the median follow-up be 16 years?

Thank you for pointing it out. The median follow-up was 16 years for the whole study period (from 2001 to 2016), but 14 years for fracture identification (from 2001 to 2014). We have corrected this typo as:

“During a median of 14 years of follow up (interquartile range [IQR]: 6.5, 14.0 years)”.

l.163-: Why was the follow-up period for mortality shorter than for fracture? To make the occurrence of fracture and death comparable, I suggest providing both per 100 or 1000 patient-years.

The follow-up time for mortality was shorter for fracture patients who were more likely to die sooner following a fracture than those who did not sustain a fracture. We reported the mortality rate as the number of deaths per 100 person-years for both fracture and non-fracture groups.

In our analysis, the contribution of initial fractures to mortality was examined in a time-dependent analysis in which both the occurrence of fracture (as the main exposure) and the confounding effect of age and comorbidities (as the predefined covariates) were analysed in a time-dependent manner to minimise the risk of immortal time bias21. Statistically, the follow-up time for fracture patient was split into the pre- and post-fracture periods. The pre-fracture period (i.e., the follow-up time between the study entry and the date of fracture for those with an incident fracture) is considered unexposed time, while its clustering effect is accounted for in a Cox’s proportional hazards model21.

We have added details and a reference to explain the concept of time-dependent analysis further as follows:

“By contrast, the follow-up time for those with an incident fracture was split into the pre-fracture (i.e., unexposed) and post-fracture (i.e., exposed) period for which their clustering effect was accounted for in the Cox’s proportional hazards regression. We conducted a time-dependent analysis to quantify the relationship between an incident exposure (i.e., an incident fracture) and the outcome of interest (i.e., post-fracture mortality) and to minimize the risk of immortal time bias50.”

l.166-: It should be clearly stated that the mortality rates in individuals with and without fractures are unadjusted.

Thank you. We have revised the text to indicate that the mortality rates were unadjusted.

Material and methods:

l.320-: Why only data to 2016, this is 6 years ago?

This study is a part of our collaborative project (FCEK 706667) that was approved in 2019 using medical records between 1996 and 2016 in the Danish NHDR. There is usually a time gap between the actual data and when the data are linked and available for a research project.

l.330-: How was low-energy fracture determined? Fracture mechanism is usually not captured in registers?

Please refer to the above response for more detail. Briefly, we used the ICD-10 codes in the primary analysis for any healthcare contact from the Danish NHDR to capture all fractures among individuals aged 50+ years that require medical attention between 2001 and 2014. Fractures in Denmark are always treated in hospitals, unless they are asymptomatic or overlooked such as some osteoporotic vertebral compressions or small "march" fractures of the smaller bones in the feet which are beyond our analysis scopes. All fractures included in the analysis were radiologically confirmed. We excluded high-trauma fractures, including skull, face, nose, finger or toes fractures or those due to traffic accidents. We have additionally acknowledged the possibilities, though unlikely that the analysis might have missed mild, asymptomatic fractures that did not receive medical attention.

l.343-: Were only comorbidities leading to hospitalisation registered? What about comorbidities diagnosed and treated by GPs?

No, comorbidities were identified using not only the primary diagnosis that requires medical attention but also all secondary diagnoses in any healthcare contact in both inpatient and outpatient visits. Unfortunately, the Danish NHDR does not include comorbidities diagnosed and treated by GPs. However, given the high concordance between the self-reported comorbidities and those in the registry1, it is unlikely that the current analysis might have missed substantial number of comorbidities. We have additionally acknowledged it as a study limitation that mild comorbidities that did not require medical attention might have not been registered in the Danish NHDR.

l.363-: it is not clear how the mortality rate pre-fracture in the patients who fractured was taken into account. The individuals had a higher prevalence of many dead-prone comorbidities and would therefore most likely also have had a higher risk of death before the fracture than the no-fracture population.

For individuals who sustained an incident fracture during the follow-up period, the follow-up time was split into two time periods, i.e., the pre-fracture (~pre-exposure) and post-fracture (post-exposure) periods to minimise the risk of immortal time bias16. As a result, no death event, as the primary endpoint of interest occurs during the pre-exposure period of the fracture patients since they have to survive to sustain a fracture. As mentioned earlier, we have added sentences and a reference to elaborate the concept of a time-dependent analysis for controlling immortal time bias.

The issue of competing risk of death, i.e., some participants might have died before they had a chance to sustain a fracture, is not a statistical issue in the current analysis that models all-cause death as the primary endpoint.

In addition, how was the risk of a second fracture/recurrent fractures taken into account in the analyses?

As explained earlier, the current analysis did not account for the possibility of subsequent or recurrent fractures which falls outside of the scope of the study.

Figure 2: the HR was adjusted for the Charlson Co-morbidity index, not specific co-morbidities.

Thank you. We have corrected the sentence as suggested.

Reviewer #2 (Recommendations for the authors):

A few additional points which should be addressed too:

Ascertainment of fractures: The authors should provide some information as to whether these fractures (esp. vertebral fractures) were self-reported or radiographically assessed. In addition to this, was it ascertained that these were fragility fractures?

Please refer to the above comment for details. Briefly, we only included the fractures that required medical attention and were radiologically ascertained. We excluded high-trauma fractures among participants aged 50+ years.

Patients: Did the authors have any information on the BMD of these subjects?

Unfortunately, we did not have any BMD data of the participants.

In addition, were medications that affect bone metabolism (osteoporosis drugs, glucocorticoids, PPI, aromatase inhibitors, etc.) documented? If so, why were they not imputed in Cox's proportional hazards model?

As clarified above, we did not consider the effect of medication in our analysis, and we acknowledge that this is a potential weakness. Nevertheless, it is very unlikely that these medications would have a substantial impact on the risk of post-fracture mortality.

References:

1. Lujic S, Simpson JM, Zwar N, Hosseinzadeh H, Jorm L. Multimorbidity in Australia: Comparing estimates derived using administrative data sources and survey data. PloS one 2017; 12(8): e0183817.

2. Andersen TF, Madsen M, Jorgensen J, Mellemkjoer L, Olsen JH. The Danish National Hospital Register. A valuable source of data for modern health sciences. Dan Med Bull 1999; 46(3): 263-8.

3. Vestergaard P, Mosekilde L. Fracture risk in patients with celiac Disease, Crohn's disease, and ulcerative colitis: a nationwide follow-up study of 16,416 patients in Denmark. Am J Epidemiol 2002; 156(1): 1-10.

4. Hundrup YA, Hoidrup S, Obel EB, Rasmussen NK. The validity of self-reported fractures among Danish female nurses: comparison with fractures registered in the Danish National Hospital Register. Scand J Public Health 2004; 32(2): 136-43.

5. Beaudoin C, Moore L, Gagne M, et al. Performance of predictive tools to identify individuals at risk of non-traumatic fracture: a systematic review, meta-analysis, and meta-regression. Osteoporos Int 2019; 30(4): 721-40.

6. Spiegelhalter D. How old are you, really? Communicating chronic risk through 'effective age' of your body and organs. BMC Med Inform Decis Mak 2016; 16: 104.

7. Vestergaard P, Rejnmark L, Mosekilde L. Osteoporosis is markedly underdiagnosed: a nationwide study from Denmark. Osteoporos Int 2005; 16(2): 134-41.

8. Roerholt C, Eiken P, Abrahamsen B. Initiation of anti-osteoporotic therapy in patients with recent fractures: a nationwide analysis of prescription rates and persistence. Osteoporos Int 2009; 20(2): 299-307.

9. Cummings SR, Lui LY, Eastell R, Allen IE. Association Between Drug Treatments for Patients With Osteoporosis and Overall Mortality Rates: A Meta-analysis. JAMA Int Med 2019; 179(11): 1491-500.

10. Chen W, Simpson JM, March LM, et al. Comorbidities Only Account for a Small Proportion of Excess Mortality After Fracture: A Record Linkage Study of Individual Fracture Types. J Bone Miner Res 2018; 33(5):795-802

11. Vestergaard P, Rejnmark L, Mosekilde L. Increased mortality in patients with a hip fracture-effect of pre-morbid conditions and post-fracture complications. Osteoporos Int 2007; 18(12): 1583-93.

12. Tran T, Bliuc D, Hansen L, et al. Persistence of Excess Mortality Following Individual Nonhip Fractures: A Relative Survival Analysis. J Clin Endocrinol Metab 2018; 103(9): 3205-14.

13. Tran T, Bliuc D, Ho-Le T, et al. Association of Multimorbidity and Excess Mortality After Fractures Among Danish Adults. JAMA Netw Open 2022; 5(10): e2235856.

14. Henderson R, Keiding N. Individual survival time prediction using statistical models. J Med Ethics 2005; 31(12): 703-6.

15. Kulinskaya E, Gitsels LA, Bakbergenuly I, Wright N. Calculation of changes in life expectancy based on proportional hazards model of an intervention. Insur Math Econ 2020; 93: 27-35.

16 Lopez-Gonzalez AA, Aguilo A, Frontera M, et al. Effectiveness of the Heart Age tool for improving modifiable cardiovascular risk factors in a Southern European population: a randomized trial. Eur J Prev Cardiol 2015; 22(3): 389-96.

17. Lyles KW, Colon-Emeric CS, Magaziner JS, et al. Zoledronic acid and clinical fractures and mortality after hip fracture. N Engl J Med 2007; 357(18): 1799-809.

18. Reid IR, Horne AM, Mihov B, et al. Fracture Prevention with Zoledronate in Older Women with Osteopenia. N Engl J Med 2018; 379(25): 2407-16.

19. Bonner C, Batcup C, Cornell S, et al. Interventions Using Heart Age for Cardiovascular Disease Risk Communication: Systematic Review of Psychological, Behavioral, and Clinical Effects. JMIR Cardio 2021; 5(2): e31056.

20. Svendsen K, Jacobs DR, Morch-Reiersen LT, et al. Evaluating the use of the heart age tool in community pharmacies: a 4-week cluster-randomized controlled trial. Eur J Public Health 2020; 30(6): 1139-45.

21. Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008; 167(4): 492-9.

https://doi.org/10.7554/eLife.83888.sa2

Article and author information

Author details

  1. Thach Tran

    1. School of Biomedical Engineering, University of Technology Sydney, Sydney, Australia
    2. Garvan Institute of Medical Research, Sydney, Australia
    3. Faculty of Medicine, UNSW Sydney, New South Wales, Australia
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    For correspondence
    SonThach.Tran@uts.edu.au
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6454-124X
  2. Thao Ho-Le

    School of Biomedical Engineering, University of Technology Sydney, Sydney, Australia
    Contribution
    Conceptualization, Formal analysis, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  3. Dana Bliuc

    1. Garvan Institute of Medical Research, Sydney, Australia
    2. Faculty of Medicine, UNSW Sydney, New South Wales, Australia
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared
  4. Bo Abrahamsen

    1. Department of Medicine, Holbæk Hospital, Holbæk, Denmark
    2. Department of Clinical Research, Odense Patient Data Explorative Network, University of Southern Denmark, Odense, Denmark
    3. Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    has received institutional research grants from UCB, Kyowa-Kirin and Pharmacosmos; consulting fees from UCB and Kyowa-Kirin; and fees for lectures from Amgen, Gedeon-Richter, Kyowa-Kirin, Eli Lily and Pharmacosmos. The author is also on the Advisory Board for UCB and is president of European Calcified Tissue Society. The author has no other competing interests to declare
  5. Louise Hansen

    Kontraktenheden, North Denmark Region, Denmark, Denmark
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4400-2732
  6. Peter Vestergaard

    1. Department of Clinical Medicine, Aalborg University, Aalborg, Denmark
    2. Department of Endocrinology, Aalborg University Hospital, Aalborg, Denmark
    3. Steno Diabetes Center North Jutland, Aalborg, Denmark
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared
  7. Jacqueline R Center

    1. Garvan Institute of Medical Research, Sydney, Australia
    2. Faculty of Medicine, UNSW Sydney, New South Wales, Australia
    3. School of Medicine Sydney, University of Notre Dame Australia, Sydney, Australia
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    has received honoraria for educational talks and Advisory boards from Amgen and honoraria for an Advisory board from Bayer. The author has no other competing interests to declare
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5278-4527
  8. Tuan V Nguyen

    1. School of Biomedical Engineering, University of Technology Sydney, Sydney, Australia
    2. School of Medicine Sydney, University of Notre Dame Australia, Sydney, Australia
    3. School of Population Health, UNSW Medicine, UNSW Sydney, Kensington, Australia
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Visualization, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    TuanVan.Nguyen@uts.edu.au
    Competing interests
    has received grants from Australian National Health and Medical Research Council and Amgen Competitive Grant Program; fees for lectures from Amgen, Bridge Health Care (VN), DKSH Pharma, MSD and VT Health Care (VN); and support from Amgen for attending the annual meeting of APCO and from VT Health Care (VN) for attending the Vietnam Osteoporosis Society Annual Scientific Meeting. The author is also Chair of the Research Committee for Australian and New Zealand Bone and Mineral Society, a Senior Advisor for Vietnam Osteoporosis Society, and an Executive Member of Asia Pacific Consortium on Osteoporosis. The author has no other competing interests to declare
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3246-6281

Funding

National Health and Medical Research Council (APP1195305)

  • Tuan V Nguyen

Amgen Competing Grant Program

  • Tuan V Nguyen

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Ethics

Human subjects: This analysis (Statistics Denmark project number 706667) was approved by the National Board of Health, the Danish Data Protection Agency, and Statistics Denmark, and subject to independent control and monitoring by The Danish Health Data Authority. Written informed consent is waived for routinely collected, pseudonymized registry data.

Senior Editor

  1. Mone Zaidi, Icahn School of Medicine at Mount Sinai, United States

Reviewing Editor

  1. Luigi Gennari, University of Siena, Italy

Reviewer

  1. Elena Tsourdi, University Clinic Dresden, Germany

Version history

  1. Preprint posted: September 10, 2022 (view preprint)
  2. Received: October 1, 2022
  3. Accepted: March 31, 2023
  4. Version of Record published: May 16, 2023 (version 1)

Copyright

© 2023, Tran et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,421
    Page views
  • 126
    Downloads
  • 2
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Thach Tran
  2. Thao Ho-Le
  3. Dana Bliuc
  4. Bo Abrahamsen
  5. Louise Hansen
  6. Peter Vestergaard
  7. Jacqueline R Center
  8. Tuan V Nguyen
(2023)
‘Skeletal Age’ for mapping the impact of fracture on mortality
eLife 12:e83888.
https://doi.org/10.7554/eLife.83888

Further reading

    1. Epidemiology and Global Health
    2. Medicine
    Jeffrey Thompson, Yidi Wang ... Ulrich H von Andrian
    Research Article Updated

    Background:

    Although there are several efficacious vaccines against COVID-19, vaccination rates in many regions around the world remain insufficient to prevent continued high disease burden and emergence of viral variants. Repurposing of existing therapeutics that prevent or mitigate severe COVID-19 could help to address these challenges. The objective of this study was to determine whether prior use of bisphosphonates is associated with reduced incidence and/or severity of COVID-19.

    Methods:

    A retrospective cohort study utilizing payer-complete health insurance claims data from 8,239,790 patients with continuous medical and prescription insurance January 1, 2019 to June 30, 2020 was performed. The primary exposure of interest was use of any bisphosphonate from January 1, 2019 to February 29, 2020. Bisphosphonate users were identified as patients having at least one bisphosphonate claim during this period, who were then 1:1 propensity score-matched to bisphosphonate non-users by age, gender, insurance type, primary-care-provider visit in 2019, and comorbidity burden. Main outcomes of interest included: (a) any testing for SARS-CoV-2 infection; (b) COVID-19 diagnosis; and (c) hospitalization with a COVID-19 diagnosis between March 1, 2020 and June 30, 2020. Multiple sensitivity analyses were also performed to assess core study outcomes amongst more restrictive matches between BP users/non-users, as well as assessing the relationship between BP-use and other respiratory infections (pneumonia, acute bronchitis) both during the same study period as well as before the COVID outbreak.

    Results:

    A total of 7,906,603 patients for whom continuous medical and prescription insurance information was available were selected. A total of 450,366 bisphosphonate users were identified and 1:1 propensity score-matched to bisphosphonate non-users. Bisphosphonate users had lower odds ratios (OR) of testing for SARS-CoV-2 infection (OR = 0.22; 95%CI:0.21–0.23; p<0.001), COVID-19 diagnosis (OR = 0.23; 95%CI:0.22–0.24; p<0.001), and COVID-19-related hospitalization (OR = 0.26; 95%CI:0.24–0.29; p<0.001). Sensitivity analyses yielded results consistent with the primary analysis. Bisphosphonate-use was also associated with decreased odds of acute bronchitis (OR = 0.23; 95%CI:0.22–0.23; p<0.001) or pneumonia (OR = 0.32; 95%CI:0.31–0.34; p<0.001) in 2019, suggesting that bisphosphonates may protect against respiratory infections by a variety of pathogens, including but not limited to SARS-CoV-2.

    Conclusions:

    Prior bisphosphonate-use was associated with dramatically reduced odds of SARS-CoV-2 testing, COVID-19 diagnosis, and COVID-19-related hospitalizations. Prospective clinical trials will be required to establish a causal role for bisphosphonate-use in COVID-19-related outcomes.

    Funding:

    This study was supported by NIH grants, AR068383 and AI155865, a grant from MassCPR (to UHvA) and a CRI Irvington postdoctoral fellowship, CRI2453 (to PH).

    1. Epidemiology and Global Health
    Tianyi Huang
    Insight

    A large observational study has found that irregular sleep-wake patterns are associated with a higher risk of overall mortality, and also mortality from cancers and cardiovascular disease.