Study On Data Mining Based Personalized Treatment Strategy For AIDS Patients

Objective: AIDS is a serious infectious disease.The medicine compatibility regimen is the key of successful therapy during the initial highly active antiretroviral therapy(HAART).Because HIV patients maybe quite different in health status,baseline situation,disease progression and drug resistance,the effect of treatment with the same therapy regimen may vary.This study applied the data mining technique to explore the relationship between patients' clinical characteristics and medicine compatibility regimen and further developed the personalized treatment plan for individual patients,helping to avoid regimen switching,lessen the suffering from drug resistance and side effect,improve life quality,and prolong life.Method:Several demographic and clinical charcteristics for 721 patients were collected at the time of initiating the HAART treatment and follow-ups,including age,height and weight,and CD4+T lymphocyte count,CD8+T lymphocyte count,HIV viral load,leukocyte count,total lymphocyte count,platelet count,hemoglobin,serum creatinine,blood urea nitrogen,blood glucose,aspartate aminotransferase,alanine aminotransferase,and total bilirubin.All combinations of different antiviral drugs were also sorted.The best therapy remedy for individual AIDS patients was determined by analyzing the data of all patients with good immune reconstruction with three patients' similarity-based data mining algorithms,i.e.cluster analysis,case-based reasoning(CBR)and association rule mining.Finally,the derived personalized treatment regimens were validated and compared.Results:(1)Patients' similarity was measured by the Euclidean Distance.After cluster analysis,the best regimen was determined as the most popular regimen in eachcluster,reaching an agreement of 67.3% between the recommended and the actual regimen.In CBR,for a certain patient,the most popular adopted treatment regimen among the top 10% most similar patients with him was recommended.When using the total 16 indexes and 7 core indexes for computing the patients' similarity,the consistency rates of recommended and actual regiment were 83.3% and 86.0%,respectively.(2)Association rule mining was used to obtain the relationship between medications and 7 core patients' indexes and corresponding medicine compatibility.The highest support of regiment including Efavirenz was 0.02835,which was higher than those of Nevirapine(0.00622)and Lopinavir & Ritonavir(0.00138).The consistency rate of recommended and actual regiment was 71.4% with the validation by the leave-one-out method.Conclusion: By applying data mining technique of cluster analysis,CBR and association rule mining,the demography and laboratory indexes of AIDS patients could be used to recommend the personalized treatment regimen.Among them,CBR has the highest accuracy.The study verified the effectiveness and feasibility of the personalized treatment development with the data mining,providing the objective and scientific suggestion for the personalized treatment of AIDS patients.