# Integrating Genomic and Clinical Data in AML: Real-World Application of the Sanger Multistage Model

## TL;DR

This study evaluates a genomic-based model for predicting outcomes in AML patients and finds that TP53 mutations strongly influence prognosis.

## Contribution

The study validates the Sanger AML multistage model in real-world data and highlights the prognostic significance of TP53 mutations.

## Key findings

- TP53 mutations were the only independent predictor of death in AML patients.
- The Sanger model showed moderate concordance with observed outcomes at 12 and 36 months.
- Discrepancies in the model were mainly in patients who relapsed after initial remission.

## Abstract

Background: Acute myeloid leukemia (AML) is genomically heterogeneous, and translating baseline molecular data into individualized prognosis remains difficult. We assessed real-world outcomes and externally validated the Sanger Institute AML multistage prognostic model. Methods: This single-center, retrospective study included 73 AML patients who underwent targeted NGS profiling. In intensively treated patients, the published, validated Sanger AML multistage prognostic model was compared with observed 12- and 36-month clinical outcomes using quadratic-weighted Cohen’s kappa. Results: Median age was 61 years, and median overall survival was 13 months, with the most significant survival differences driven by treatment intensity. TP53 mutations occurred in 7 patients (9.6%) and were linked to primary refractoriness and extremely poor survival. TP53 was the only independent predictor of death (HR 8.07, 95% CI 2.23–29.13; p = 0.0014). Model concordance was moderate at 12 months (29 evaluable cases; weighted κ = 0.52; alive/dead κ = 0.52) and fair-to-moderate at 36 months (23 cases; weighted κ = 0.46). The tool performed best for predicted death without remission, while most discrepancies involved patients expected to remain in first remission who later relapsed and died. Conclusions: TP53 disruption dominates prognosis in real-world AML. The multistage tool supports early high-risk identification but shows limited long-term calibration, motivating the development of dynamic models integrating contemporary therapies and longitudinal min/serial NGS data.