The clinician-artificial intelligence partnership in early sepsis identification: Leveraging predictive intelligence for enhanced financial outcomes

Background Diagnosing sepsis is a critical challenge due to complex clinical and systemic barriers; consequently, delayed and failed diagnoses result in poor patient outcomes, high mortality, and severe financial repercussions for healthcare systems. Inadequate documentation and coding frequently cause sepsis claim denials, leading to reimbursement loss. This analysis evaluates the potential clinical economic impact of using the U.S. Food and Drug Administration-authorized artificial intelligence-based Sepsis ImmunoScore software to achieve accurate severity of illness coding. Methods This retrospective, multisite, observational study included patients with suspected serious infection treated at four U.S. hospitals. Medical Severity Diagnosis Related Group (MS-DRG) assignments were determined following CMS Medicare v39 definitions using each patient’s ICD-10-CM diagnoses, ICD-10-PCS procedures, age, sex, and discharge status. For patients with High/Very High Sepsis ImmunoScore results lacking sufficient sepsis documentation, we compared reimbursement with and without adding a sepsis ICD-10-CM diagnosis. This resulted in two MS-DRG calculations per patient: the current MS-DRG based on existing documentation and the potential MS-DRG with sepsis diagnosis included. When the potential sepsis MS-DRG yielded higher reimbursement than the current MS-DRG, the case represented a revenue recovery opportunity. A subset analysis restricted to patients meeting Sepsis-3 criteria provided a conservative estimate of potential revenue recovery. Results The final analysis cohort included 4419 patients. The Sepsis ImmunoScore identified 745 cases (16.9%) where High/Very High risk results indicated undocumented sepsis with higher reimbursement than current MS-DRG assignments. This represents $4 684 373 (95% CI, $4 125 378–$5 248 933) in potential revenue recovery across all 4419 patients, or $1060 ($934–$1188) per patient tested. When restricted to cases meeting Sepsis-3 criteria for clinical validation, 516 cases (11.7%) represented $3 240 546 ($2 727 080–$3 782 855) in potential revenue recovery, or $733 ($617–$856) per patient tested. Conclusions This analysis demonstrates that implementation of an enhanced diagnostic tool can improve documentation accuracy and ensure it reflects the complexity of care provided, supporting full reimbursement for hospital services.