The Study On Identification Of Critical Attributes For Disease Diagnosis And Assessment Of Treatment Alternatives Based On Individual Differences Of Patients

Based on the practical problems of clinical diagnosis and treatment decision andthe current research status of treatment alternatives assessment，the dissertationstudies the basic theory and decision mode of clinical diagnosis and treatment fromthe point of individual differences of patients. Then the identification of criticalfeatures for disease diagnosis and the treatment alternatives assessment involvingpatients’ preferences are studied in depth. The objective is to improve the accuracand validity of clinical decision making and provide valuable guidance for utilizinghealth resources reasonably and better patient-centred care. The major results aresummarized as follows:Firstly, the dissertation analyses the diagnosis factors, treatment modalities andcharacteristic of medicine information, studies the system structure and theoreticalframework of clinical diagnosis and treatment decision from the point of biologicalvariability and humanistic differences of patients. And then it introduces thecomponents and critical technical problems of the framework.Secondly, in order to better identify the critical features of disease diagnosis withthe characteristics of high dimensional features and small sample sizes, a method ofextracting rules for disease diagnosis based on elastic net and support vector machine(SVM) is proposed. Compared with other algorithms, the proposed method has higherclassification accuracy and is more effective in reducing the irrelevant and uselesscharacteristics. Then, considering the imbalanced and multi-cass classification dataproblems, synthetic minority over-sampling technique (SMOTE) and randomunder-sampling method are used to deal with the original data sets. After that, Elasticnet and multi-cass classification method based on SVM are used to identify thecritical features from imbalanced data sets for disease diagnosis. Compared with otheralgorithms, the results show that the proposed method is more effective method forfeature identification from imbalanced data sets for disease diagnosis.Thirdly, health care professionals and patients often present substantialdifferences in the preferences for efficacy and adverse events, and it is difficult toestablish a mutual agreement in treatment decisions. Considering the differentpreferences between health care professionals and patients elicited by discrete choice experiment (DCE), the desirability function method, the technique for orderpreference by similarity to ideal solution (TOPSIS) and the loss function method areused as structured methodologies to study the subjective rationale behind the choiceof treatments respectively. The application examples are given to test the effectivenessand practicability of each model.Fourthly, based on the consideration of processing imprecise observations andpreference information with the stakeholders’ attitude toward ambiguity in healthcaredecision making, the dissertation proposes a fuzzy model for medicines appraisalusing an EVIDEM framework and fuzzy multi-criteria group decision by integratingan improved fuzzy AHP based on fuzzy linguistic preference relations (FuzzyLinPreRa) and a modified TOPSIS method.Finally, for the shortage of professionals-patients share decision making, inwhich there is no effective ways for patient to participate in clinical decision, thetwo-dimension classified evaluative index system of treatment alternatives based onbenefits and risks is presented. The treatment alternatives are classified into differentcategories by considering the risk attitude and preference of patient individual, whichcould provide effective guidance for clinical decision making.