Real-World Applications in Healthcare

MST’s vectorized knowledge graph technology has been deployed across multiple high-impact use cases, offering measurable improvements in accuracy, compliance and operational efficiency as well as significant ROI.

Risk Adjustment

Problem: Accurately assessing a patient’s health risk is critical for proper reimbursement and population health management. Traditional models often miss nuanced relationships and return too many false positives from unstructured charts.

Solution: MST’s VKG captures and contextualizes both structured and unstructured patient data, preserving clinical and temporal relationships (e.g., distinguishing between actual diagnoses versus differential diagnoses). The result is improved identification of risk factors and more accurate coding, all with full traceability.

Real-World Results: Here’s an example of results for a limited number of HCC codes.

See our white paper on risk adjustment.

Payment Integrity and Fraud Detection

Problem: Fraudulent, wasteful, or incorrect claims cost billions annually. Rules based or manual efforts are not able to detect subtle anomalies or patterns across disparate datasets due to volume and complexity of medical diagnoses.

Fraud schemes are similar in nature to computer viruses in that new frauds constantly appear which is also why MST’s technology is superior to other methods because it learns from frauds across multiple plans and payers and does not use rules based methods.

Recovery is more difficult post payment and to a lesser extent once a claim has been submitted. Prevention is the best strategy which is possible with AI for payment integrity.

Solution: By linking medical events, diagnoses, procedures, and timelines, MST’s VKG uncovers inconsistent or suspicious patterns that rule-based systems miss.

Real-World Results: MST has identified numerous payment integrity (PI) concerns, ranging from incorrect or duplicative billing to outright fraud. The fraudulent activities vary in scope by provider, with estimated annual impacts ranging from $2 million to $10 million. These cases include both collusive schemes involving members and instances of non-collusive provider fraud.

See our white paper on fraud.

De-identification

Problem: Ensuring patient privacy while maintaining data utility for research and analysis is a persistent challenge.

Solution: MST’s VKG-based de-identification system identifies and classifies protected health information (PHI) within complex, unstructured records. Unlike LLM-based approaches that may rewrite or alter data, VKGs remove identifiers while preserving the surrounding context—enabling safe and meaningful analysis.

Real-World Results: In a recent blinded study at an academic medical center, MST successfully identified and masked a broad spectrum of PHI, including common elements (e.g. names, dates) and challenging local PHI. The results were compliant with HIPAA Safe Harbor Standards.

See our white paper on de-identification.