Improving Prediction of High-Cost Health Care Users with Medical Check-Up Data.

Studies found that a small portion of the population spent the majority of health care resources, and they highlighted the importance of predicting high-cost users in the health care management and policy. Most prior research on high-cost user prediction models are based on diagnosis data with additional cost and health care utilization data to improve prediction accuracy. To further improve the prediction of high-cost users, researchers have been testing various new data sources such as self-reported health status data. In this study, we use three categories of medical check-up data, laboratory tests, self-reported medical history, and self-reported health behavior data to build high-cost user prediction models, and to assess the medical check-up features as predictors of high-cost users. Using three data-mining models, logistic regression, random forest, and neural network models, we show that under the diagnosis-based approach, medical check-up data marginally improve diagnosis-based prediction models. Under the cost-based approach, we find that medical check-up data improve cost-based prediction models marginally and medical check-up data can be a viable alternate data source to diagnosis data in predicting high-cost users.

The Incidences and Characteristics of Various Cancers in Patients on Dialysis: a Korean Nationwide Study.

BACKGROUND: The numbers of patients on dialysis and their life expectancies are increasing. Reduced renal function is associated with an increased risk of cancer, but the cancer incidence and sites in dialysis patients compared with those of the general population require further investigation. We investigated the incidences of various cancers in dialysis patients in Korea and used national health insurance data to identify cancers that should be screened in dialysis clinics. METHODS: We accessed the Korean National Health Insurance Database and excerpted data using the International Classification of Disease codes for dialysis and malignancies. We included all patients who commenced dialysis between 2004 and 2013 and selected the same number of controls via propensity score matching. RESULTS: A total of 48,315 dialysis patients and controls were evaluated; of these, 2,504 (5.2%) dialysis patients and 2,201 (4.6%) controls developed cancer. The overall cancer risk was 1.54-fold higher in dialysis patients than in controls (adjusted hazard ratio, 1.71; 95% confidence interval, 1.61-1.81). The cancer incidence rate (incidence rate ratio [IRR], 3.27) was especially high in younger dialysis patients (aged 0-29 years). The most common malignancy of end-stage renal disease patients and controls was colorectal cancer. The major primary cancer sites in dialysis patients were liver and stomach, followed by the lung, kidney, and urinary tract. Kidney cancer exhibited the highest IRR (6.75), followed by upper urinary tract (4.00) and skin cancer (3.38). The rates of prostate cancer (0.54) and oropharyngeal cancer (0.72) were lower than those in the general population. CONCLUSION: Dialysis patients exhibited a higher incidence of malignancy than controls. Dialysis patients should be screened in terms of colorectal, liver, lung, kidney and urinary tract malignancies in dialysis clinics.

Stratified sampling design based on data mining.

OBJECTIVES: To explore classification rules based on data mining methodologies which are to be used in defining strata in stratified sampling of healthcare providers with improved sampling efficiency. METHODS: We performed k-means clustering to group providers with similar characteristics, then, constructed decision trees on cluster labels to generate stratification rules. We assessed the variance explained by the stratification proposed in this study and by conventional stratification to evaluate the performance of the sampling design. We constructed a study database from health insurance claims data and providers' profile data made available to this study by the Health Insurance Review and Assessment Service of South Korea, and population data from Statistics Korea. From our database, we used the data for single specialty clinics or hospitals in two specialties, general surgery and ophthalmology, for the year 2011 in this study. RESULTS: Data mining resulted in five strata in general surgery with two stratification variables, the number of inpatients per specialist and population density of provider location, and five strata in ophthalmology with two stratification variables, the number of inpatients per specialist and number of beds. The percentages of variance in annual changes in the productivity of specialists explained by the stratification in general surgery and ophthalmology were 22% and 8%, respectively, whereas conventional stratification by the type of provider location and number of beds explained 2% and 0.2% of variance, respectively. CONCLUSIONS: This study demonstrated that data mining methods can be used in designing efficient stratified sampling with variables readily available to the insurer and government; it offers an alternative to the existing stratification method that is widely used in healthcare provider surveys in South Korea.