Skip to main content

Case study

Identifying patients at risk for a bone fracture within 2 years

Optum Life Sciences experts used real-world data to help test an algorithm designed to assess for osteoporosis risk.

February 15, 2024 | 4-minute read


Collaboration is key when it comes to moving the needle for health care. That’s why Amgen, a global biotechnology firm, turned to Optum Life Sciences when it needed validation testing for a potential new population health tool — an algorithm leveraging natural language processing (NLP) to identify patients at risk for a bone fracture from osteoporosis within 2 years.

Often called the silent disease, osteoporosis can weaken and thin bones while remaining undetected for years. The first sign of disease is often a debilitating fracture. In the U.S., an estimated 10 million adults aged 50 and older have osteoporosis — most are women, but about 2 million are men.

In addition, over 43 million more people have low bone mass, putting them at increased risk for osteoporosis and fractures. Given our aging U.S. population, fractures are projected to increase by nearly 70% by 2040.  

The Amgen algorithm — Crystal Bone — could become a powerful risk management tool in the fight against osteoporosis. It borrows methods from NLP, an application of artificial intelligence to process and analyze human language. Crystal Bone was directly inspired by the same type of NLP that enables predictive text on digital devices. But instead of looking at language to predict what someone will type next, Crystal Bone analyzes diagnosis codes from a patient’s electronic health records (EHR) to help assess fracture risk.

Designed to interface with an EHR system, Crystal Bone may advance a physician’s ability to identify a patient’s risk of bone fractures within a 2-year time frame. This ideally enables the physician and patient to start preventative measures before a bone fracture happens.

To test a tool such as Crystal Bone in a clinical setting, the developer (in this case, Amgen) can begin by generating evidence to support the model’s generalizability using real-world data. Optum Life Sciences, with its established relationships to Optum Care and UnitedHealthcare, helped Amgen test Crystal Bone using the de-identified EHR data set from Optum.

“Amgen’s deep scientific knowledge to support a tool like Crystal Bone, along with our ability to validate their findings using real-world data, helps us get closer to helping doctors do even more to identify osteoporosis-related fracture risk and potentially support patients before fractures even occur,” says Dr. Brian Solow, chief medical officer of Optum Life Sciences.

The urgency to address osteoporosis early on

The U.S. Preventive Services Task Force advises osteoporosis screening for women over 65. It also suggests screening for postmenopausal women younger than 65 years who are at increased risk of osteoporosis based on a formal clinical risk assessment tool. A common screening tool is the bone mineral density testing by dual-energy x-ray absorptiometry (DXA).

However, real-world data estimates show that overall osteoporosis screening rates remain quite low, ranging from 12% among those over age 80 to 26% among those age 65–79. This suggests a potential opportunity for new tools or technologies to help bridge the gap between guidelines and patient care.

Undiagnosed or poorly managed osteoporosis can have both clinical and economic consequences. Patients who experience an osteoporosis-related fracture face significant health and economic burdens. In the 12 months following the fracture, patients can incur health care costs of more than $30,000. And an average of $3,000 will be paid by the patient. Costs also increase incrementally with subsequent fractures.

“Crystal Bone could be a new pre-screening tool that helps providers by identifying patients that may benefit from screening because they are at the highest risk,’" says Tina Kelley, senior director at Optum Life Sciences. “Given low screening rates, Crystal Bone may enable more conversations that lead to treatment plans that can reduce osteoporotic fracture risk.”

What may differentiate Crystal Bone from similar tools?

Clinicians today can use a risk assessment test called FRAX®, which calculates absolute fracture risk within 10 years and identifies patients who should start anti-osteoporosis medications. The online tool requires input of patient information — including previous fractures, bone mineral density (BMD) and other risk factors to calculate fracture risk scores.

Without manual data entry, Crystal Bone utilizes a patient’s ICD diagnosis codes, age and sex to identify fracture risk within 2 years, considering data such as fractures at various sites.

Amgen researchers wanted to predict fractures on a shorter time frame to aid clinicians in identifying patients at the highest risk of fracture. More importantly, the automated analysis of existing EHR data can help clinicians identify at-risk patients without imposing any added data-entry burdens on them.

“Crystal Bone has the potential to plug into existing electronic health records and use diagnosis codes to help predict osteoporotic risk, which is unique,” says Dr. Mary Oates, global medical lead for Bone Health at Amgen.

The algorithm was first trained and tested in 2020, which found that the accuracy of the predictive model was comparable to the performance of other fracture risk calculators currently available to physicians, including FRAX. Amgen researchers used the Optum EHR data set for model training, which at the time covered about 100 million individuals.

Validating Crystal Bone with quality real-world data

With these promising results, the next step involved validation testing for Crystal Bone. To help Amgen advance their work, Optum Life Sciences brought in Reliant Medical Group, a large multispecialty group within the Optum Care network. While maintaining all necessary patient privacy protections, Optum Life Science researchers ran the Crystal Bone algorithm on Reliant’s EHR data in order to identify patients likely to have a fracture within the next 2 years.

In the validation testing, Kelley and her team evaluated data from more than 105,000 patients over 50 years old with EHR codes for more than a 2-year consecutive history from December 2014 to November 2020. Records included demographics, diagnosis and procedure codes, and prescriptions. The validation study was presented as a poster at the American Society of Bone and Mineral Research (ASBMR) 2022 Annual Meeting.

Noteworthy findings included:

  • Crystal Bone identified patients at risk of fracture within 2 years, with the ability to stratify patients into 3 times, 4 times, 7 times and 9 times relative fracture risk groups with model performance consistent with FRAX®.
  • Among the 19,100 patients that Crystal Bone identified as at 3 times greater risk of future fracture, 69% had no prior osteoporosis diagnosis.

With these promising findings, Amgen plans to further evaluate the Crystal Bone algorithm in the US to better understand its potential clinical use as an emerging technology that can innovatively transform care.  

“Our next steps involve assisting Amgen to investigate the clinical utility of the Crystal Bone in a clinical study. It’s exciting to help test a risk assessment tool that may benefit patients,” Kelley says.

Olivia Banyon

Olivia Banyon, MPH
Vice President
Optum Clinical Enterprise Programs

Discover how Optum can help improve your business strategies

Request more information

Related content

Case study: Improving SGLT2i therapy guideline adherence

Article that covers OLS and Boehringer Ingelheim collaboration to increase adherence for patients with diabetes and CVD.

Pilot Program Lowers Cardiovascular Disease Risk

An innovative program involving patient awareness and provider outreach improves patient use of lipid-lowering therapies.

Natural language processing (NLP) methods for RWD insights

NLP systems allow us to draw insights from billions of unstructured medical notes. Learn more about our NLP methodology in this white paper.