B0993
Title: Pitfalls in statistical modeling of tumor-infiltrating immune cells from histological imaging
Authors: Finn Hamilton - BC Cancer (Canada) [presenting]
Abstract: The advent of successful immunotherapies for cancer has spurred much research to understand the contribution of tumor-infiltrating immune cells (TIL) to patient outcome and to develop new therapies. Despite advances, TIL data arising from histological imaging have multiple issues that are infrequently addressed in modeling studies. These include that TIL are subject to substantial measurement and sampling error, especially when sampled from high-throughput tissue microarrays (TMAs), and that TIL data arise from overdispersed count distributions. Failure to account for these issues can lead to bias in parameter estimation, reduced power, and the failure of findings to generalize to broader populations. We demonstrate how incorporating these considerations into modeling strategies can increase the reliability of inference in TIL studies, and propose key metrics and methods to consider when analyzing TIL or other biomarkers measured from histological imaging.
