Fifteen years of the Garvan Fracture Risk Calculator: a Personal Reflection
A personal reflection on the development of prediction models for personalized fracture risk assessment.
Fifteen years ago today (Mar 7), we introduced the Garvan Fracture Risk Calculator (FRC), marking a pivotal moment in our commitment to advancing global osteoporosis and bone health initiatives. As we celebrate the 15th anniversary today, I take a moment to contemplate the journey that resulted in the creation and deployment of FRC, followed by its evolution into the upgraded mark II version known as BONEcheck.
Two catalysts prompted me to create a novel model for evaluating fracture risk: an insightful commentary by Dr. Richard Wasnich and our observation on the relationship between bone mineral density and fracture risk.
Dr Wasnich's commentary
At the time, the diagnosis of osteoporosis was (and still) relied on a measurement of bone mineral density expressed in terms of a T-score. A consensus conference convened by the WHO established the definition that an individual with a T-score below -2.5 is classified as having 'osteoporosis.' This definition appears sensible because, through a series of studies, we know that those individuals are at high risk of fracture.
After that consensus, Dr. Richard Wasnich, a prominent figure in bone research, wrote a dissenting editorial (Wasnich R. Consensus and the T-score fallacy. Clin Rheumatol 1997;16(4):337-9). I was captivated by this commentary where he asserted eloquently:
“What are the issues surrounding the use of T-scores, as recommended by the WHO panel? On the one side, they seemingly offer simplicity, which is sorely needed. However they are not readily translated into interventional guidelines. The opposing viewpoint is that T-scores are a major step backwards into the realm of 'fracture thresholds.'
Fundamental to this debate is the fact that bone density is a risk factor, and not a diagnostic test. So making a 'diagnosis' of osteoporosis based on the presence of a single risk factor, at a single point in time, is already a tenuous concept."
And, the subsequent passage truly inspires and drives my motivation.
“We need is an estimate of absolute fracture rate […] There is no need to obscure this useful information by inventing a new statistic, e.g. the "T-score.”
Absolute risk. That is what we need!
Our observation: most fracture cases are not osteoporotic
I thought the dichotomous classification of T-score was absurd. Consider having a T-score of -2.48, which is not deemed osteoporotic, yet a T-score of -2.51 is categorized as such. In our previous studies we discovered that the relationship between T-score and fracture risk was continuous, lacking a distinct 'break point' suitable for straightforward dichotomization.
What led to the choice of the -2.5 threshold? On reviewing the literature, I learned that this threshold was selected to align the prevalence of osteoporosis with the lifetime risk of fracture among Caucasian women, which is approximately 30% (Kanis et al. The Diagnosis of Osteoporosis. JBMR 1994;9:1137-1141). However, it's worth noting that this threshold faced criticism from Regina C. Elandt-Johnson and Gayle E. Lester (JBMR 1996;8:1198-1200).
Crucially, we noted that over 50% of women and 70% of men who experienced fractures did not exhibit 'osteoporosis' (meaning that their BMD T-scores were above the -2.5 threshold). In other words, if we treat those with T-score < -2.5, we miss a lot of high risk people.
Dr. Nguyen D. Nguyen, my Ph.D. student at the time, and I were fascinated by the observation. I tasked Nguyen with conducting a sophisticated analysis known as "Bayesian Model Averaging" to identify factors beyond the T-score that were linked to fracture risk. He found that apart from old age and low BMD, the number of falls and the number of prior fractures were very important risk factors for fracture. When he presented the result in a lab meeting, I said to myself: this makes sense!
We then wrote a series of papers to describe our predictive models and how they could be used for individualized fracture risk assessment. I decided to send the papers to Osteoporosis International because the journal was (and still is) a highly clinically oriented venue. In the papers, we made a point that:
“The ultimate aim of developing a prognostic model is to provide clinicians and each individual with their risk estimate to guide clinical decisions. At present, individuals with low bone mineral density (i.e., T-scores being less than -2.5) or with a history of prior low trauma fracture are recommended for therapeutic intervention. This recommendation is logical and appropriate, since these individuals – as shown in this study and previous studies – have higher risk of fracture, and treatment can reduce their risk of fracture. However, because fracture is a multifactorial event, there is more than one way that an individual can attain the risk conferred by either low BMD or a prior fracture. Indeed, virtually all women aged 70 years with BMD T-scores less than -1.5 and all 80-year-old men with BMD T-scores less than -1.0 can be considered ‘high risk’. On the other hand, no 60 year old men or women without a prior fracture and a fall are considered high risk, even when their BMD T-scores are below -2.5. This demonstrates the informativeness of a multivariable prognostic model, and the limitation of a risk stratification-based approach for risk assessment for an individual.”
We also made another point re the uniqueness of fracture risk:
“Each individual is important and unique. […] Prognosis is about imparting information of fracture risk to an individual and each individual is a unique case, because there exists no ‘average individual’ in the population. The more risk factors are considered, the greater likelihood of uniqueness of an individual’s profile being defined. Therefore, by modeling risk factors in their continuous scale the present models can be uniquely tailored to an individual.”
Actually, my idea of ‘individualization‘ was not new; I learned it from colleagues in the cancer research field. At the time, cancer researchers were busily developing nomograms for predicting the risk of having cancer, and it appeared that these nomograms worked well for many cases. Why do these probabilistic tools work well? Now, we know that highly experienced clinicians can also make good prognoses, and that is a fact. Unfortunately, their predictions are highly variable and less consistent, or in scientific language, clinicians’ predictions are irreproducible. But reproducibility is a bedrock of science. So, scientifically, we cannot rely on a clinician’s judgment. Statistical prognostic models have been shown to outperform clinical judgment because these models can objectively incorporate many risk data. Moreover, any prognosis from a statistical model is unbiased, consistent, and completely reproducible.
One year after the publication of our papers, the FRAX model — developed under the sponsorship of the World Health Organization — was published. So, doctors and patients in the world now have at least two tools to assess their own risk of fracture in their convenience. The two models, Garvan and FRAX, have helped transform the management of osteoporosis worldwide.
BONEcheck
After more than 10 years of experience with FRC and recent advances in research, I have identified several features that could improve the utility and relevance of the tool:
Prediction timeframe: FRC and FRAX primarily offer a 10-year forecast for fracture risk. I believe that managing a 10-year risk is more challenging for elderly individuals than a 5-year risk. Therefore, a shift in the prediction timeframe is warranted.
Risk presentation: all current fracture risk assessment tools present numerical probabilities, which can pose a challenge for the general population to comprehend. There is a need for a more user-friendly presentation.
Treatment context: existing risk estimates produced by risk assessment models don't provide the benefit in terms of fracture reduction and increased survival (and potential risk) if a high-risk patient opts for treatment; limiting the communication of risk and clinically useful discussions between patients and their physicians.
Refracture: an existing fracture significantly elevates the risk of subsequent fractures, yet existing fracture risk prediction models do not estimate the risk of refracture.
Mortality: most fractures, especially hip fractures, are linked to an increased risk of mortality. However, existing tools do not incorporate mortality into their predictions.
To address the aforementioned issues, our team at the University of Technology Sydney (UTS), supported by an NHMRC grant, has overhauled the original Garvan Fracture Risk Calculator, introducing a novel and advanced version named BONEcheckTM. This updated version of BONEcheck incorporates features absent in existing tools, including:
Five-year frame prediction.
Treatment contextualization: BONEcheck incorporates data from randomized controlled trials (RCTs) to inform patients about the specific reduction in fracture risk associated with medication use, tailored to their age and risk profile.
Risk of refracture: BONEcheck features a dedicated module for predicting the probability of refracture.
Mortality: BONEcheck draws from previous research to provide users with the risk of mortality following a fracture.
Skeletal Age: BONEcheck introduces the concept of Skeletal Age, representing an individual's skeleton age due to a fracture or exposure to risk factors that heighten fracture risk.
Remeasurement of bone mineral density (BMD): BONEcheck incorporates a module predicting the time required to reach osteoporosis in individuals not currently classified as osteoporotic. This feature assists clinicians in advising patients on the timing for repeat BMD measurements.
Predicting osteoporosis: for individuals without BMD data, BONEcheck includes a module to predict the risk of osteoporosis (T-score less than -2.5).
Preventive information: BONEcheck is designed with a patient-centric approach, aiming to empower individuals with information that enables them to actively reduce their risk of fractures.
Several validation studies have conclusively shown that the rebranded BONEcheck, formerly the FRC, demonstrates fracture risk prediction accuracy equal to or surpassing that of FRAX. The predicted probability of fracture derived from BONEcheck/FRC also aligns closely with clinical decisions. Notably, FRC/BONEcheck comes recommended by the Royal Australian College of General Practitioners, Healthy Bone Australia, Osteoporosis New Zealand, and the Asia Pacific Consortium for Osteoporosis for application in clinical practice. Our FRC tool is a key component of “Know Your Bones” that helps people self-assess their bone health. (You can click on the above link to have your test now!)
Lastly, I am pleased to announce that BONEcheck is entirely free of charge. Since its launch in May 2023, BONEcheck has garnered usage from over 15,000 users across 165 countries worldwide.
The development and implementation of personalised fracture risk assessment has been considered a revolution in the management of osteoporosis (Saag and Geusens, Arthritis Res & Ther 2009). I am delighted to have played a pioneering role in this transformative journey.
_____
Post Script: BONEcheck is now accessible to users through multiple platforms. Users can access it directly from our website or download the app from the Apple Store or Google Play. Please click on the links below to start utilizing the BONEcheck tool:
Website: https://bonecheck.org
Apple Store: https://apps.apple.com/app/bonecheck/id6447424513.
Google Play: https://play.google.com/store/apps/details?id=org.saigonmec.bonecheck.
The development of BONEcheck and its features can be found in the following article: https://www.sciencedirect.com/science/article/pii/S2405525523000481?via%3Dihub
_____
There have been hundreds of articles on our fracture risk assessment model in the literature, including ours. Here are some of the recent reviews that I have written for journals and books:
Nguyen TV. Personalized fracture risk assessment: where are we at? Expert Review in Endocrinology and Metabolism 2021;16(4):191-200.
Nguyen TV (2020). Toward the era of precision fracture risk assessment. J Clin Endocrinol Metab. pii: dgaa222.
Nguyen TV, Eisman JA. Post-GWAS Polygenic Risk Score: Utility and Challenges. JBMR Plus 2020;4: e10411.
Nguyen TV. Personalised assessment of fracture risk: which tool? Aust J Gen Pract 2022;51:189-190.
Nguyen TV. Individualized Fracture Risk Assessment: State-of-the-Art and Room for Improvement. Osteoporosis and Sarcopenia 2018;4(1):2-10.
Nguyen TV. Individualized Assessment of Fracture Risk: Contribution of “Osteogenomic Profile”. J Clin Densitom 2017;20:353-359.
Nguyen TV, Eisman JA. Assessment of fracture risk: population association vs individual prediction. J Bone Miner Res 2017 Dec 27.
Nguyen TV, Eisman JA. Genetic profiling and individualized assessment of fracture risk. Nature Review Endocrinolology 2013 Mar;9(3):153-61.
Nguyen TV, Center JR, Eisman JA. Individualized fracture risk assessment: progresses and challenges. Curr Opin Rheumatol. 2013 Jul;25(4):532-41.
Nguyen TV, Eisman JA. Genetics and the individualized prediction of fracture. Curr Osteoporos Rep 2012 Sep;10(3):236-44.
Nguyen TV. Mapping translational research into individualized prognosis of fracture risk. International Journal of Rheumatic Diseases 2008; 11:347-358.
Tran B, Center JR, Nguyen TV. Translational genetics of osteoporosis: from population association to individualized risk assessment. In Primer on the Metabolic Bone Diseases and Disorders of Mineral Metabolism, Seventh Edition, Ed: Clifford Rosen. ASBMR 2017 Edition.
Nguyen TV, Eisman JA. Pharmacogenetics and pharmacogenomics of osteoporosis: personalized medicine outlook. In Genetics of Bone Biology and Skeletal Disease, Edited by RJ Thakker, MP Whyte, JA Eisman, T Igarashi. Academic Press Amsterdam 2017 Edition.
Nguyen TV. Individualized Progress of Fractures in Men. In Osteoporosis in Men – the effect of gender on skeletal health, 2nd Ed, Edited by ES Orwoll, JP Biezikian, and D Vanderschueren. Academic Press, 2011.