Photo Credit: Tonpor Kasa

Research shows that routine risk assessment of patients with upper GI bleeding could reduce healthcare costs, primarily through fewer hospital admissions.

Despite studies that prove the benefits of risk assessment with the Glasgow-Blatchford score (GBS) in patients with acute upper gastrointestinal bleeding (UGIB), this strategy has been slow to take hold in the ED.

Studies have shown that patients with UGIB identified as being very low risk (i.e., GBS=0) can often be discharged from the ED with outpatient management, significantly reducing associated costs In addition to the GBS, a machine learning model can be applied to assess risk in patients with UGIB.

Risk Assessment Potential

Although studies have shown the potential of incorporating these risk assessment strategies, the actual healthcare cost savings have yet to be examined. To that end, Dennis L. Shung, MD, MHS, PhD, and colleagues developed a study comparing the triage strategies of GBS and a validated machine learning model to usual care, applying a Markov chain model from a managed care perspective.

Dr. Shung and colleagues identified a group of hospital-ED patients with UGIB; they obtained this participant group from the 2019 National Emergency Department Sample. The researchers divided the patients into four quartiles, with Quartile 1 having the smallest hospital EDs and Quartile 4 having the largest (based on the number of patient encounters). The research team evaluated the cost savings, including confirmation, enactment, and maintenance of risk assessment scores.

Results of the study were published in the American Journal of Gastroenterology.

Savings Over 5 Years

Over 5 years, when a GBS=0 was established, and the patient was treated accordingly, the estimated savings were $6.99 million (95% uncertainty interval [UI], $2.06 million-$13.5 million) in the Quartile 4 hospital EDs. The estimated savings in the Quartile 1 hospital EDs were $137,138 (95% UI, $32,589-$270,581).

When machine-based learning risk assessment was applied, the estimated 5-year savings ranged from $8.82 million (95% UI, $3.72 million-$14.8 million) in the Quartile 4 hospital-based EDs to $174,400 (95% UI, $67,315-$302,438) in the Quartile 1 hospital-based EDs.

National savings over 5 years, applying a GBS = 0, were estimated to be $2.69 billion (95% UI, $0.79 billion-$5.21 billion).  With a machine learning model at 100% sensitivity, the national savings were estimated to be $3.40 billion (95% UI, $1.42 billion-$5.71 billion).

Adjusting the Threshold

When the very-low risk thresholds were expanded to GBS=0-1, the estimated total national savings over 5 years was $3.49 billion (95% UI, $1.45 billion-$5.81 billion) and $3.83 billion (95% UI, $1.77 billion-$6.11 billion) for the machine learning model at 99% sensitivity.

The research team noted that cost savings depended on averting unnecessary hospital admissions. To confirm this assessment, a one-way sensitivity analysis was applied. For Quartile 1 hospital EDs, this would mean reducing admission rates of patients with UGIB by 2%, and for all other quartiles, this would mean reducing admission rates by 1%.

Dr. Shung and his colleagues concluded that the results uncovered could positively impact practice.

“This finding provides a financial argument for insurers to provide incentives for health systems to invest in the infrastructure necessary to implement risk scores in routine clinical practice and for payers to consider developing novel payment structures to incentivize healthcare systems for the use of risk assessment tools in patients with UGIB.”