You Might Also Like
Explore This IssueSeptember 2019
MADRID—Each year in the U.S., more than 300,000 people—mostly older than 65—are hospitalized for hip fracture.1 The problem is a worldwide phenomenon and will likely increase as life expectancy increases.
Multiple risk factors for hip fracture are known, including older age, use of medications that cause bone loss, excessive alcohol or caffeine consumption, smoking and frailty. An issue often faced by clinicians is the proactive identification of patients at the highest risk for hip and other fractures.2 Systematic approaches to identify these patients have varied by doctor and practice, but the widespread use of electronic health records (EHRs) has now created an opportunity to develop automated systems
At the 2019 European Congress of Rheumatology (EULAR), June 12–15, Bo Abrahamsen, PhD, professor and consultant endocrinologist at the University of Southern Denmark, led an intriguing discussion about his work to create an automated short-term fracture risk prediction model based on existing information collected in EHR systems. In explaining the rationale for his work, Prof. Abrahamsen noted that such record systems provide a unique opportunity to study the epidemiology of fractures because they catalog information from a large number of patients, are used in clinical practice, are convenient and speedy, and do not require self-reporting. Interestingly, information on fracture data has been collected for many decades by Olmsted County, Minn., since 1928, in the Malmö Radiology Register since 1950, and elsewhere.
Several osteoporotic fracture risk calculators have been developed for clinical use. In 2009, researchers in the U.K. analyzed primary care data from more than 2 million patients to develop and validate two new fracture clinical risk scores that can be used to identify patients at high risk for fracture who may benefit from interventions to reduce their risk.3
In 2018, Danish researchers, including Prof. Abrahamsen, used public health registries with information on the total population of Denmark aged 45 and older—nearly 2.5 million patients—to develop a fracture risk evaluation model (FREM) to automate the process of identifying individuals at high risk for hip or major osteoporotic fractures.4 For this study, the researchers identified patients with an osteoporotic fracture in 2013 and looked retrospectively at patient information from the preceding 14 years to identify risk factors that could be used to estimate one-year risk of major osteoporotic and hip fracture. Although the positive predictive value of this algorithm was low, given that most patients won’t have a second fracture within a year of the first fracture, the negative predictive value was very high, at 99.2% for women and 99.5% for men.
The FREM tool & others are being evaluated for integration directly into EHR systems & nationwide patient information registries.
Many physicians may ask: Isn’t the Fracture Risk Assessment Tool (FRAX; an online calculator launched by the University of Sheffield, South Yorkshire, England, in 2008 to calculate the 10-year risk of major osteoporotic and hip fracture) already being used by physicians to counsel patients on fracture risk?5 Yes, however, Prof. Abrahamsen pointed out a few important caveats that may influence the utility of this tool. For example, the fracture risk estimate is for the subsequent 10 years, which may not be as valuable as shorter term risk estimates for some patients, specifically octogenarians or nonagenarians.
Additionally, such tools as FRAX require a clinician to physically find the necessary information in a patient’s chart and manually enter these data into the online form. The FREM tool and others are being evaluated for integration directly into EHR systems and nationwide patient information registries. This feature will automate the process, not relying on individual human effort. The possibility of discovering new fracture risk factors by combining big data repositories also exists and, to this point, has not been used to evaluate fracture risk in a systematic way.
On the subject of automated assistance in identifying patients at risk for recurrent fracture, Prof. Abrahamsen noted that neural networks are already being evaluated to recognize incidental radiographic findings that may quite dramatically change a patient’s risk estimate for future fracture. In 2017, Israeli researchers used an automated method to detect spine compression fractures in computed tomography (CT) scans. This research was based on the established concept that the presence of a vertebral compression fracture is indicative of osteoporosis and represents a robust predictor for development of a second osteoporotic fracture in the spine or elsewhere. Such a system could easily be integrated into radiology reports in which such fractures are not commented on because they were not the indication for the CT study.6
Opportunity & Challenges Ahead
The question-and-answer segment of this clinical practice session highlighted the opportunities and potential challenges faced by the enactment of such automated systems in the real world. Example: Even if the algorithms presented by Prof. Abrahamsen and colleagues use open-source software, the vendors who operate EHR systems may choose not to incorporate these algorithms into their programs, because doing so will cost time and money.
The question of how patients will view this fairly proactive means of health data evaluation also remains. Concerns include giving the impression patient health information is being reviewed without necessary consent, as well as the problem of providing information about a low, but not negligible, risk of fracture that may induce anxiety and uncertainty in a patient, causing them to second-guess the next, best therapeutic steps.
Ultimately, similar to many aspects of medicine, the integration of new technology and the use of data analytics are likely to continue to increase. It is up to physicians, administrators, patients and all stakeholders to determine how best to implement these new solutions and deal with the issues they may create.
Jason Liebowitz, MD, recently completed his fellowship in rheumatology at Johns Hopkins University, Baltimore, where he also earned his MD. He is currently in practice with Arthritis, Rheumatic, and Back Disease Associates, New Jersey.
- U.S. Department of Health & Human Services. Agency for Healthcare Research and Quality: Healthcare cost and utilization project (HCUP). HCUPnet. 2012.
- Stanford Health Care. Risk factors for hip fracture. 2019.
- Hippisley-Cox J, Coupland C. Predicting risk of osteoporotic fracture in men and women in England and Wales: Prospective derivation and validation of QFractureScores. BMJ. 2009 Nov 19;339:b4229.
- Rubin KH, Moller S, Holmberg T, et al. A new fracture risk assessment tool (FREM) based on public health registries. J Bone Miner Res. 2018 Nov;33(11):1967–1979.
- University of Sheffield Centre for Metabolic Bone Diseases. FRAX: Fracture risk assessment tool. 2019.
- Armato SG III, Petrick NA (eds). Proceedings of SPIE. Medical Imaging 2017: Computer-Aided Diagnosis. 2017 Mar 3;10134.
The September article, “If It’s Broken, Fix It,” mistakenly stated: “The FRAX risk calculator requires the femoral neck bone mineral density score from a dual-energy X-ray absorptiometry (DXA) scan to calculate risk estimates. If a patient has not had a DXA scan, then this tool cannot be used.” However, the FRAX score can actually be calculated with or without the bone mineral density, corrected above. We regret the error.