BACKGROUND
Accurate preoperative risk stratification in Gastrointestinal Stromal Tumors (GISTs) is crucial for determining the need for neoadjuvant therapy. However, existing methods, such as biopsy-based mitotic count assessment, suffer from limitations such as tumor heterogeneity and sampling bias.
OBJECTIVE
Our aim was to develop an innovative App that enhances patient stratification accuracy.
METHODS
Utilizing a Bayesian Network as a predictive model, we constructed a Directed Acyclic Graph (DAG) incorporating relevant clinico-pathological variables. Key variables for mitotic count estimation were identified, including tumor size, site, mitotic count on biopsy, surface area assessed on biopsy, and tumor response to therapy (if applicable). The model underwent rigorous testing through prior predictive simulations, validation on a mock dataset, and training on real GIST cases with paired biopsy and surgery (n=80) from IRCCS Humanitas Research Hospital, encompassing a total of 160 cases.
RESULTS
Our model demonstrated excellent diagnostic performance, with selection based on lower deviance and robust out-of-sample performance. The posterior predictive check further confirmed its accuracy against ground truth. We successfully developed an App that dynamically computes the number of mitoses on the surgical specimen based on tumor size, site, surface area, and mitotic count on biopsy, using posterior probabilities generated by the model.
CONCLUSIONS
Our novel App enables precise prediction of mitotic count on surgical specimens, significantly improving preoperative risk stratification in GISTs. This advancement empowers clinicians to adopt tailored treatment approaches, resulting in enhanced patient outcomes.