(P 065) AN APP TO PREDICT MITOTIC COUNT ON THE SURGICAL SPECIMEN OF GIST.

Objective: Preoperative risk stratification in Gastrointestinal Stromal Tumors (GISTs) is important in determining the indication to neoadjuvant therapy. Size and site can be defined by imaging, whereas mitotic count is done on biopsy. The latter is burdened by some limitations mainly due to tumor heterogeneity and sampling bias. We aim to build an App that, accounting for these limitations, could allow a more accurate stratification of patients in preoperative settings.

Methods: As modeling strategy, we used a multilevel hierarchical model; to select variables to train the model, we created a Directed Acyclic Graph (DAG) of the available clinico-pathological variables. Given the DAG, to estimate the mitotic count we identified as variables: tumor size, tumor site, mitotic count on biopsy, surface area assessed on biopsy and tumor response to therapy – if performed. We performed prior predictive simulations, tested the model on a mock dataset and finally trained on real GIST cases with paired biopsy and surgery (n=80) from IRCCS Humanitas Research Hospital (a total of 160 cases) after mitotic count re-evaluation and measurement of the biopsies’ surface.

Results: Model diagnostics were satisfactory. To choose the best model we estimated model performance through Widely Applicable Information Criteria and Pareto Smooth Important Sampling Leave-One-Out Cross-Validation Criteria. We selected the model with the lower deviance in out of sample performance. Posterior predictive check was used to highlight model performance against ground truth. We built an app that used the posterior probability generated by the model to dynamically compute the number of mitoses on the surgical specimen given tumor size, site and surface of, and mitotic count on, biopsy.

Conclusion: We created an App that predicts mitotic count on the surgical specimen. This might allow a better preoperative risk stratification of GISTs, leading to a tailored treatment approach.
________
Caption:
Example 1 stomach, 42 mm, 3 mitoses in 5 mm2 an available surface of biopsy of 1.98 mm2 (i.e. around 9 HPF).
Example 2 stomach, 90 mm, 3 mitoses in 5 mm2 an available surface of biopsy of 3.47 mm2 (i.e. around 16 HPF).