E-News Exclusive

AI Insight for Population Health

By Orit Wimpfheimer, MD

Prevention is better than cure, says Arnon Makori, MD, MHA, director of imaging informatics at Clalit Health Services in Israel, the second largest health maintenance organization in the world. He points out that this is a basic tenet of population health, which in part seeks to improve care by assessing the risk of chronic disease, managing that risk, and driving prevention in specific populations.

Across the globe at the National Health Service’s Oxford University Hospitals in the United Kingdom (UK) as well as in the United States at Intermountain Health, the largest health care provider in the Utah, Idaho, and Nevada area, physicians emphatically agree. They view this growing approach to health care as a crucial way to treat patients more effectively, efficiently, and at lower cost. They also agree that in today’s data-driven health care age, AI-driven population health tools can help this burgeoning health care strategy live up to its full potential. In fact, some believe that AI will become a new standard of care in population health and is poised to change the face of medicine.

AI has already achieved large-scale acceptance as a means of assisting radiologists in the analysis of images from CT to X-ray, most often for evaluating acute conditions and standard screening exams such as mammograms. However, in these leading-edge hospitals and others worldwide, AI is now emerging as an important means of identifying previously unrecognized patient risk and disease at a very early stage. A population health program can then address these findings to enable more timely patient management and clinical interventions that can potentially avert a serious problem or reduce its impact.

An Emerging Opportunity
While saving significant time and cost, population health AI need not require additional patient visits or imaging exams. It can be employed opportunistically to search recent or current images for incidental findings indicative of a chronic disease or other medical issue. What’s more, a well-designed AI population health solution can integrate seamlessly into existing radiology workflow and call upon the radiologist’s trained eyes when an image is flagged for chronic disease signs, allowing the department to maintain peak efficiency. The impact of identifying previously undiagnosed or underdiagnosed conditions can be dramatic in improving patient morbidity and mortality. It can also significantly increase economic value, given that in the United States alone, $3.5 trillion is spent on caring for patients with chronic diseases. 

Already, the emerging application of AI in population health is addressing early detection of heart disease, osteoporosis, and more. In today’s data-intensive health care age, it is helping physicians make the most of the ever-growing volume of patients' medical imaging data. Estimates are that this quantity is doubling every three months and, with it, AI’s importance.

Today, sifting through patient information for relevant data is almost impossible and certainly not the best use of a trained specialist’s time, notes Michael Phillips, MD, MBA, a partner and managing director of Intermountain Ventures, which is the venture arm of Intermountain Healthcare. A dedicated proponent of AI in population health, he believes the technology enhances radiologists’ effectiveness by bringing pertinent findings to their attention for evaluation and, ultimately, makes more data globally available for patient care.

Intermountain Healthcare demonstrated the value of AI in a recent study using Zebra Medical Vision’s automated Imaging Analytics Engine, which leverages the technology for population health. Using CT exams performed for acute spinal conditions, the study applied AI analysis to identify incidental findings of vertebral compression fractures (VCFs) and then quantified the results. With AI, detection rates for VCFs increased from 30% to 90% on a global average, enabling proactive early treatment to lower the morbidity and mortality associated with osteoporotic hip fractures. At-risk patients treated for osteoporosis are 25% less likely to fracture a hip.

Similarly, a study at Oxford University Royal College of Physicians in the UK found that the AI population health technology boosted the number of patients treated by their Fracture Liaison Services four-fold, without increasing staffing levels. The study also found that these diagnoses and resultant early interventions saved the hospital 570,000 British pounds ($760,000) per year.

Improving Outcomes
On the cardiovascular front, AI detection and analysis of coronary artery calcium on noncontrast chest CTs is showing significant value in the stratification of patient risk for a cardiovascular event in a five-year time span. Examining the impact of this, a Zebra Medical Vision study found that risk detection and intervention prevented 40 cardiovascular events in every 100,000 patients annually, saving $620,000 in treatment costs, based on a spend of $15,500 per cardiac event. Ongoing surveillance of a designated patient population for a 10-year period cut cardiovascular events by 20%, with a similar relative cost savings and a dramatic savings in human life, given that 1 in 3 Americans dies from a cardiovascular event.

In Israel, under Makori’s direction, Clalit Health implemented a similar cardiovascular screening program in its 14 hospitals serving 4.5 million insured members. After examining tens of thousands of existing patient images, the facility identified improvements similar to the Zebra study. Clalit also uses the population health solution for VCF screening and plans to expand to other applications shortly.

Based on his experience, Makori notes that tight AI integration with both radiology and overall hospital workflow is crucial to ensure that AI population health data are communicated accurately and efficiently and fit within existing departmental processes and procedures. He recommends using a fully automated population health AI solution that pulls archived images directly from a PACS and new images from the modality to its AI engine. After analysis, it is also important to capture results as deidentified data, to maximize the flexibility to share and store them without limits.

Other important features to streamline information flow are automatic transmission of exams with positive findings to the radiology department for verification, with complete worklist and RIS integration. Result synopses in the report can then be sent to primary care physicians or appropriate specialists, and patients can be enrolled to receive follow-up care.

Along with advanced algorithms and IT infrastructure, this communication is crucial in a population health AI solution. After all, in the final analysis, AI-driven population health isn’t just about a sophisticated, automated interpretation of a single image; it’s about integrating findings into total patient care, to plan care and improve outcomes.

— Orit Wimpfheimer, MD, is chief medical officer for Zebra Medical Vision.