Just a moment, the page is loading...

Promising subgroup identification method based on biomarkers for precision medicine.

Promising subgroup identification method based on biomarkers for precision medicine.

Kang Li

Biostatistics department of Public Health school, Harbin Medical University, China

12 May 2021

The research aims at providing a new statistical method that can detect promising subgroups based on biomarkers for precision/tailored medicine. In those subgroups, the treatment effect is much higher than the population average.The main idea is to transfer the subgroup discovery problem to an optimization problem. Based on some optimal algorithms, the method has the ability to find a maximum treatment effect subgroup in the whole potential subgroup searching space.Compared to the traditional model-based or other tree-based subgroup identification methods, this new method has more ability to explore interactive biomarkers, and hence the identified subgroups may be more predictive. The new method also addresses the overfitting/false discovery issue inherent in data mining applications using a resampling-based method.DefinitionsTreatment effect: The difference (or ratio) of an outcome variable (endpoint) between the treatment and control group, e.g., the hazard ratio is always used as treatment effect in oncology studies to quantify the treatment difference. In this research, we want to find potential subgroups with higher treatment effects than the population average.Tree-based subgroup identification method: Creating a decision tree that classifies patients into subgroups with differential treatment effects using sequential splits based on dichotomous (or dichotomized) predictors. Many subgroup identification methods are utilizing this algorithm, e.g., interaction trees (Su, Tsai, Wang, Nickerson, & Bogong, 2009), generalized unbiased interaction detection and estimation (Loh, 2002, 2009), subgroup identification based on differential effect search (Lipkovich, Dmitrienko, Denne, & Enas, 2011), etc. Overfitting/false discovery issue: In this research, overfitting reflects that the identified subgroup's actual treatment effect is over-estimated. If that subgroup's actual treatment effect is equal to or lower than the population average, it is falsely discovered.

[{ "PostingID": 3767, "Title": "ASTELLAS-OSI-774-302", "Description": "A multicenter, randomized, double-blind, placebo-controlled, phase 3 study of single-agent Tarceva® (erlotinib) following complete tumor resection with or without adjuvant chemotherapy in patients with stage IB-IIIA non-small cell lung carcinoma who have EGFR-positive tumors" }]

Statistical Analysis Plan