Competitive risk survival analysis method based on causal inference

Abstract

The invention discloses a competitive risk survival analysis method based on causal inference. The method comprises the following steps: building a structured causal model according to a competitive risk survival analysis model; confusion factors existing in the competitive risk survival analysis model and back door paths generated by the confusion factors are identified according to the structured causal model; performing causal intervention on the competitive risk survival analysis model through backdoor adjustment to remove confusion factors in the model; defining a loss function of the competitive risk survival analysis model, and correcting the loss function to obtain a loss function after causal intervention; and minimizing a loss function after causal intervention to realize training optimization of the competitive risk survival analysis model. According to the competitive risk survival analysis method based on causal inference, an existing competitive risk survival analysis model is corrected from a causal angle by using a structured causal model, and a deviation-removed survival model is learned through a backdoor adjustment formula in a causal inference mode.

Description

technical field [0001] The invention belongs to the field of data processing, and in particular relates to a survival analysis method of competition risk based on causal inference. Background technique [0002] Survival analysis is a collection of data analysis techniques whose goal is to analyze the relationship between covariates and hit times to events of interest. Survival analysis methods range from statistical methods to machine learning and, in recent years, deep learning methods. Various survival analysis methods are now widely used in various fields, including medicine, recommender systems, and economics. [0003] Traditional methods of statistical survival analysis, such as the Cox proportional hazards model (CPH), while achieving great success, lack the ability to deal with the problem of competing risks, that is, environments where there are multiple events of interest. Competing risk is a class of events that either hinders the observation of an event of inter...

AI Technical Summary

Problems solved by technology

[0004] Despite much progress, existing survival analysis models for competing risks suffer from a major flaw in that competing risks are confounding factors that can mislead survival analysis models to learn about covariates when capturing causal relationships between covariates and events of interest Spurious correlation between X and event Y, leading to poor performance of the model

Although confounding factors are good for capturing the underlying relationship between X and Y through model calculations P(Y|X), it can erroneously extract event-unrelated, as well as not event-related covariates

Images