Prolonged patient stays are extremely costly for providers, insurers, and even patients as this generates higher health care consumption and increases the possibility of developing high-risk conditions during the hospital stay. This is a major financial concern for the government. In the public health care system in Colombia, cost increases due to preventable hospitalizations have raised many questions regarding the use and effectiveness of prevention programs.
The objective of this project predict the length of hospitalization of users of the public health system in Colombia and to estimate the potential cost reduction by using the proposed prevention program. The project proposes a hospitalization prevention program in which a decision is made whether or not to intervene with a patient based on a risk prediction model of the length of hospitalization each year using machine learning.
For the purpose of predicting patient length of stay and evaluating an intervention program we did two things: for the first objective we used automated machine learning and for the second, we used a decision rule based on the first stage predictions which indicates when to intervene with the patient in order to reduce the risk and expected cost of next year's hospitalization. Finally, we measure the potential cost savings of such a prevention program relative to several other baseline scenarios.
First, reinforced trees, random forests, and artificial neural networks were used for the prediction program. These methods demonstrate better performance than linear regression methods in predicting annual patient length of stay because when calculating the errors of each method using the square root of the root mean square error, better known as RMSE, in the final model they have a smaller sampling error. Then, to evaluate a program of when to intervene with patients based on the above predictions in order to reduce costs, we used a decision rule model. To achieve this model we measured the probability or risk of hospitalization taking into account differences in age, gender, and location of patients.
The results show that the prevention program contributes significantly to the cost-effectiveness of the system. Relative to several baseline scenarios, the prediction program shows efficiencies greater than or equal to 40 %, and reductions of between $100,000 COP and $700,000 COP in the cost of intervention per patient (approximately a reduction of between 14% and 100% of the average cost per patient in the Colombian health system). An automated decision rule based on these predictive models is an important source of cost savings for any insurer in Colombia's contributory health care system.
In Projects, you can learn about real applications of our services, use cases and examples of them.
