Association Rules Analysis And Prediction Model Construction Of High Hospitalization Out-of-pocket Expenses For Patients With Neurological Diseases

Objective:Neurological disorders are universal diseases with high mortality rate and high disability rate,which seriously threaten people’s physical and mental health,and cause huge burden to society,families and patients.With population growth and aging,the overall burden will inevitably continue to increase,and have become great challenge in public health in China that cannot be ignored.It had positive promoting role in alleviating the medical burden of patients to calculate the treatment costs,explore the influencing factors of high hospitalization out-of-pocket expense,and build prediction models.Therefore,the main purpose was:(1)at the macro level,we calculated the treatment cost to clarify the finance,distribution,and usage to explicit the flow.At the micro level,we analyzed the distribution of outpatient and inpatient average expense as well as the out-ofpocket expense in order to obtain more abundant and accurate health economic data.It also provided baseline data to support the second part of association rules and the third part of prediction model construction.(2)we located the influencing factors of high hospitalization out-of-pocket expenses and clarified association rules to find the key points of cost control,as well as provided predictive factors for the construction of prediction model in the third part.(3)we constructed prediction models of high hospitalization outof-pocket expenses to strengthen detailed control over high medical expenses.Methods:First,based on the framework of SHA 2011,we collected macro and sample institution data,built database,and calculated treatment costs of patients with neurological diseases in S City in 2017-2020.Second,the K-means clustering algorithm was used to determine critical value of high hospitalization out-of-pocket expense.Apriori algorithm was used to analyze factors associated with high hospitalization out-of-pocket expenses.Last,we used Decision Tree(DT),Support Vector Machine(SVM),K-Nearest Neighbor(KNN),and Random Forest(RF)algorithms to build prediction models of high hospitalization out-of-pocket expenses,and explained the best model with SHAP algorithm.Results:Part Ⅰ:(1)The treatment cost of neurological diseases in S City rose from 1.955 billion yuan in 2017 to 2.389 billion yuan in 2019.Among them,outpatient expenses rose from 158 million yuan to 331 million yuan;hospitalization costs rose from 1.797 billion to 2.057 billion.In 2020,treatment costs decreased to 1.657 billion yuan,of which outpatient costs rose to 371 million yuan,while hospitalization costs fell to 1.286 billion yuan.(2)From the perspective of financing flow,the main sources of financing were public financing program and family health expenditure.(3)From the perspective of institutional flow,the institutional allocation mainly occurred in hospitals(88.40%-96.69%),while primary health care institutions accounted for a relatively small proportion(3.28%11.02%).(4)From the age distribution,the treatment cost was mainly in the 55-89 years group,and the peak value was in the 60-69 years group.(5)In the distribution of various types of neurological diseases,cerebrovascular diseases(160-169)were dominant in both outpatients and inpatients(59.69%-71.29%),and showed a fluctuating upward trend.(6)From the perspective of financing flow at different medical institutions,the financing of general hospitals was mainly depended on public financing program and family health expenditure.Primary health care institutions mainly depended on public financing program,while public health institutions mainly depended on family health expenditure.(7)From the distribution structure of neurological diseases at different ages,the treatment costs of most types were concentrated in patients aged 55-89.(8)The median cost per outpatient was 145.80 yuan,and the median out-of-pocket expense was 90.06 yuan.The median cost per inpatient was 7,309.75 yuan,and the median out-of-pocket expense was 1,827.84 yuan.Part Ⅱ:(1)Based on the K-means clustering algorithm,the critical value of high hospitalization out-of-pocket expense was determined as 18,322 yuan.(2)A total of 26 association rule models based on the Apriori algorithm were found to be correlated with high hospitalization out-of-pocket expense.There were 5 types of support were greater than 7%,and the highest was {type=1}?{high hospitalization out-of-pocket expense=1},with the support of 9.67%.There were 2 types of confidence greater than 50%,and the highest was {insurance=3;hospital type=1}?{high hospitalization out-of-pocket expense=1},with the confidence of 56.07%.There were 4 kinds of lift greater than 3,and the highest was {insurance=1;hospital type=1}?{high hospitalization out-of-pocket expense=1},with the lift of 5.67.The association rules showed that the attributes of general hospitals,municipal medical institutions,cerebrovascular disease,male,moderate/severe comorbidity,three or more comorbidities,surgery,out-of-pocket,and hospitalization duration of 11-15 days were significantly associated with high hospitalization out-ofpocket expense.Part Ⅲ:(1)The accuracy rate,recall rate,F1-score of DT model were 0.81,0.78,and 0.79 respectively,while overall accuracy was 0.93,and AUC value was 0.88.The SVM model were 0.33,0.59,and 0.43 respectively,while overall accuracy was 0.73,and AUC value was 0.61.The KNN model were 0.88,0.76 and 0.82 respectively,while overall accuracy was 0.94,and AUC value was 0.87.The RF model were 0.86,0.92 and 0.89 respectively,while overall accuracy was 0.96,and AUC value was 0.94.(2)Combining the performance evaluation and comparison of different models,RF model was more stable.(3)The SHAP algorithm was used to explain the RF model.The results showed that surgery was the most important,followed by hospitalization duration of≥21 days,out-of-pocket,cerebrovascular disease.Non-surgery,hospitalization duration of 6-10 days,urban workers medical insurance,no complications,and hospitalization duration of≤5 days were negatively predicting high hospitalization out-of-pocket expense.Surgery,hospitalization duration of≥21 days,out-of-pocket,cerebrovascular disease,hospitalization duration of 16-20 days,three kinds of complications or more,municipal medical institutions,general hospital treatment were positively predicting high hospitalization out-of-pocket expense.Conclusion:(1)The treatment cost of neurological diseases was at a high level,of which the hospitalization cost was the main component,and control of hospitalization expense was a key measure to ease the medical burden.Health financing was largely depended on public financing program and family health expenditure.The treatment cost mainly went to general hospitals and were concentrated in elderly.Cerebrovascular diseases dominated the treatment costs.(2)A total of 26 association rules were determined by the Apriori algorithm.There was a significant correlation between general hospitals,municipal medical institutions,cerebrovascular diseases,male,moderate/severe comorbidity,three or more comorbidities,surgery,out-of-pocket,hospitalization duration of 11-15 days and high hospitalization out-of-pocket expenses.(3)The performance of RF model was more stable,and it was suitable to construct the prediction model of high hospitalization out-ofpocket expenses for neurological diseases.