Calibrated Survival Curve and Treatment Comparison with Imperfect Survival Outcomes from Electronic Health Records

• Activity Number: 164 - Leveraging Real-World Data in the Drug Development Process
• Type: Topic Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Chuan Hong* and Liang Liang and Tianxi Cai
• Companies: Harvard Medical School and Harvard T. H. Chan School of Public Health and Harvard University
• Keywords: Electronic Medical Records; Survival Curve Calibration; Semi-Supervised Learning; Treatment Comparison; Measurement Errors; Progression Assessment

Abstract

In EHR-based studies with survival outcomes, true clinical event times are often not readily available for analysis. Previous studies have focused on the development of machine learning models to derive surrogate event times. However, most of the existing clinical literature that aims to derive progression prediction models with large-scale EMR databases directly use the surrogate event times and simply ignores the measurement errors. With censoring and lack of precise observation of the event time, survival curve estimations and progression assessment become substantially more challenging. In this work, we propose EHR-based calibrated algorithms for survival curve and treatment comparison with imperfect event times under a semi-supervised survival setting. We demonstrate via simulation studies that the proposed methods have substantial potential gains in efficiency in terms of estimation and power. We illustrate the proposed methods using an lung cancer dataset derived from Partners Healthcare EHR by estimating the progression free survival and comparing the difference between two treatment groups.