Inferential methods to assess the incremental value of new biomarkers in risk classification models

评估风险分类模型中新生物标志物增量价值的推理方法

• 批准号: 9309399
• 负责人: Glenn Heller
• 金额: $ 28.27万
• 依托单位: SLOAN-KETTERING INST CAN RESEARCH
• 依托单位国家: 美国
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-08-01 至 2020-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9309399
• 关键词: Agreement; Algorithms; Applications Grants; Area; Biological Markers; Cessation of life; Classification; Clinical; Confidence Intervals; Decision Making; Development; Diagnosis; Diagnostic tests; Discrimination; Disease; Evaluation; General Population; Genomics; Grant; Guidelines; Health; Individual; Intervention; Laboratories; Lead; Measures; Medical; Methodology; Methods; Modeling; Noise; Outcome; Patient risk; Patients; Physicians; Play; Population; Probability; Property; Proportional Hazards Models; Proteomics; Publishing; Receiver Operating Characteristics; Research; Research Personnel; Risk; Role; Sampling; Sensitivity and Specificity; Specific qualifier value; Specificity; Statistical Methods; Structure; Survival Analysis; Testing; Time; Update; Variant; base; biomarker panel; clinical practice; clinical predictors; hazard; improved; indexing; innovation; novel marker; oncology; patient population; simulation; statistics; stem; theories

Abstract This proposal aims to develop statistical methods to evaluate improvement in risk classification measures due to the inclusion of new biomarkers. Risk classification modeling has a direct impact on patient treatment and health. These models are used to identify the need for diagnostic testing in the general population, or the type of treatment to be administered in a specified patient population. The most common risk classification improvement measures include the difference in the area under the receiver operating characteristic curve, the net reclassification index, the integrated discrimination improvement, and the difference in sensitivity and specificity. The new statistical methods will provide a metric to decide whether a change in patient risk, stemming from the evaluation of new biomarkers, is random variation or a true improvement in the accuracy of a risk classification model. The metric is defined through asymptotic distribution theory within a nested models framework. Methods to assess improvement in the accuracy of risk models are not well-developed, with current methods relying on classical hypothesis tests of association. These tests may lead to dissonance, since the results of parametric association statistics may not align with the nonparametric risk improvement statistics used in the models being evaluated. The objective of this grant proposal is to develop a coherent statistical inferential strategy that directly measures the incremental value of new biomarkers in risk classification models with binary and survival endpoints. The aims of the proposal are: 1) To develop inferential methods for the difference in concordance probability measures from nested proportional hazards models with a survival endpoint.2) To develop inferential methods for the incremental value of Net Reclassification Index (NRI) and Integrated Discrimination Improvement (IDI) with binary and survival regression models. 3) To develop inferential methods for the incremental value of sensitivity and specificity with binary and survival regression models. Measures of risk classification improvement impact medical decision making. The statistical evaluation of the incremental benefit of a new panel of markers, in relation to clinical measures and laboratory tests that are acquired in the course of routine clinical practice, will play a role in the level of confidence physicians and medical researchers have in updating their medical intervention algorithm.

摘要