Data Science & Social Safety Net

Using Data Science to Strengthen

Social Safety Net

Emergency Rooms

The aim of this study is to construct a model that predicts the risk rate of subscribers to the public health insurance system using data science. To this end, this paper first looked at the concept of a data science approach and recent research trends. Next, applying these discussions, a model for predicting the hospitalization rate of insured persons under the US public health insurance system was constructed. When comparing six models, including the traditional regression model, general machine learning model, and deep learning model, deep learning models showed the highest reproducibility. In particular, the omnidirectional multiple neural network model was able to predict the patient’s hospitalization with 80% accuracy through sociodemographic information. In the conclusion, measures to increase the predictive power of the model and implications for implementing preventive interventions for high-risk subjects were discussed. These discussions could contribute as part of academic and policy research to reinforce social safety nets using artificial intelligence technology and to ensure that social welfare resources are properly delivered to families and individuals at risk.

..................................................................................................................................................................................

EXTRACTED FROM:

Using Data Science to Strengthen Social Safety Net (2020)

south-korea-icons-noun-project-339505.pn
Desk with Stethoscope

In this study, the insurance claims database for Medicare and Medicaid were analyzed and compared through various models to predict patient readmission rates. Traditionally, researchers have used regression models when analyzing observational data. In particular, logistic regression analysis uses clinical time series data to develop predictive models. However, this parametric method is based on certain assumptions, which may violate well-established theory. There may be limits to exploring unknown fields or unknown data structures. In addition, if the number of parameters is too large compared to the number of samples, and the parameter is a hyperdimensional functional model, the correlation between the variable is also large. If the bias, which is the difference, is large analysis becomes difficult. Thus, in this study, the focus in on first creating a model that predicts hospital admission through insurance claim data. In addition, it analyzes the variables that influence the prediction.

CHARACTERISTICS OF MEDICAID AND MEDICARE CLAIMANTS
Picture1.png

The data set for this study was deidentified and consisted of data from 45,000 recipients of Medicare and Medicaid, the public health coverage in the United States.

Model A:

Decision Tree

Screen Shot 2021-01-29 at 3.55.29 PM.png
Model A:  DECISION TREE
PREDICTIVE EVALUATION BY MODEL
Screen Shot 2021-01-29 at 3.47.10 PM.png