Method for predicting cancer risk based on hereditary gene mutation

Abstract

The invention discloses a method for predicting a cancer risk based on hereditary gene mutation, and belongs to the field of medical statistics. The method can be used to construct an individualized primary cancer risk prediction quantitative model by collecting medical history information and saliva samples of family members of primary cancer patients. On the premise of using a molecular biological detection experiment technology, a strategy for accurately determining hereditary mutation of related genes from oral epithelial cells of a consultant and constructing a statistical prediction model in combination with family medical history data is provided. According to the prediction model provided by the invention, multiple cancer medical history information can be integrated, family structure information is fully utilized to correct and confirm deviation, and residual correlation among family members is analyzed. Simulation experiments show that the method can correctly find model parameters; results of real patient data fitting show that the method can reach the highest accuracy known at present on a cross validation data set, and the primary cancer risk of the first time can alsobe more stably predicted in an independent data set. The method can be used for constructing a cancer risk assessment system in genetic counseling, and can also be used for screening key gene loci.

Description

technical field [0001] The invention belongs to the field of medical statistics, and in particular relates to a method for constructing a primary cancer risk prediction model based on genetic mutation data. Background technique [0002] Inherited gene mutation, also known as germline mutation, refers to the existence of this mutation in all cells of an individual (including somatic cells and germ cells), and the mutation has already existed in the fertilized egg period of the individual. Cancers caused by reproductive mutations in cancer-related genes are called hereditary cancers. About 10% of cancer patients are hereditary cancers. Because hereditary cancers are less random than non-hereditary cancers, the research on the pathogenic mechanism and preventive measures of known hereditary cancers is relatively mature. Numerous data have shown that early cancer surveillance for people at high risk of hereditary cancer can significantly improve their prognosis. If families a...

AI Technical Summary

Problems solved by technology

Although this method can predict the risk of first and second primary cancers, it has the following significant disadvantages: a) Due to the lack of a large number of mutation data in the training data, this method integrates a complicated missing data inference model, This makes the model itself too complex and difficult to converge when estimating parameters; b) this model can only integrate the mutation information of a single gene, and it is difficult to study more genetic factors related to pathogenicity; c) this method places too much emphasis on multiple primary cancers ( Such as the prediction of the second primary cancer), but in practical applications, the risk prediction of the first cancer is often more important; d) This method only includes the method of statistical model construction, and does not form a method system that can be used in practice

Therefore, in general, this method cannot be transformed into a clinical broad-spectrum hereditary cancer risk assessment tool, and it is difficult to apply to the construction of risk prediction methods other than LFS

Shinet.al （4） A semiparametric competing risk survival analysis model was used to fit LFS patient data to predict the risk of three different types of cancer. The biggest technical flaw of this method is that although it specifically predicts the first cancer risk, it does not use medical history data. The relevant information of the second cancer has no obvious advantage in prediction accuracy compared with the traditional method; in addition, this method also integrates a missing data inference model, and can only infer the overall mutation probability of a single gene. If sexual cancers are associated with multiple genes or loci, this model cannot be used

Images