Evaluation and application of assessment measure for risk prediction model when independent censoring assumption is not hold

• Project/Area Number: 26870564
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Epidemiology and preventive medicine; Statistical science
• Research Institution: St. Marianna University School of Medicine (2017); National Center for Child Health and Development (2014-2016)
• Principal Investigator: Inoue Eisuke 聖マリアンナ医科大学, 医学部, 教授 (50528338)
• Project Period (FY): 2014-04-01 – 2018-03-31
• Project Status: Completed (Fiscal Year 2017)
• Budget Amount: ¥3,510,000 (Direct Cost: ¥2,700,000、Indirect Cost: ¥810,000); Fiscal Year 2016: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000); Fiscal Year 2015: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000); Fiscal Year 2014: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
• Keywords: 予測モデル / 生存時間解析 / 独立打ち切り / NRI / RMST / c統計量 / Restricted Survival Time

Outline of Final Research Achievements

The assessment measures of risk prediction model using time-to-event data were proposed and evaluated when the independent censoring assumption is not hold.
In the evaluation of the c-statistic, the weight of the censoring probability in the estimator was estimated by strata. Based on this statistic, the objective function of the Maximum Rank Correlation estimator was improved and evaluated by the simulation studies. In the NRI study, the consistency and asymptotic normality of the proposed estimators were shown. In addition, simulation studies were conducted to show the properties of the estimators. As for the regression model research for the Restricted Mean Survival Time, the cause of bias in the IPCW method was found by the theoretical consideration. Since this phenomenon clearly appears in the group comparison problem, it is recommended to use the Pseudo Observation method rather than IPCW.

Research Products

• [Presentation] Combining Biomarkers for Risk Prediction Using Approximated Rank Correlation Statistic with Censored Survival Data (2017)
  • Author(s): Eisuke Inoue
  • Organizer: Joint Statistical Meeting 2017
  • Int'l Joint Research
• [Presentation] Impact of the Violation of Independent Censoring Assumptions in Restricted Mean Survival Data Analysis for Time-to-Event Data (2016)
  • Author(s): Eisuke Inoue
  • Organizer: Joint Statistical Meeting 2016
  • Place of Presentation: Chicago, Ilinois, US
  • Year and Date: 2016-07-30
  • Int'l Joint Research
• [Presentation] Multiple imputation and inverse probability weighting for regression models using restricted mean survival time under covariate missing (2016)
  • Author(s): Eisuke Inoue
  • Organizer: The XXVIIIth International Biometric Conference
  • Place of Presentation: Victoria, British Columbia, Canada
  • Year and Date: 2016-07-10
  • Int'l Joint Research