A model that allows predicting the occurrence of an insured event within the next 12 months has been added to the Webiomed predictive analytics platform.
Predicting patient visits to medical organizations is important in health insurance, as it directly affects the costs of health insurance policies. Assessing the patient's risks in terms of the likelihood of an insured event is a key task that is usually performed by specifically trained employees of insurance companies, underwriters.
The profit of insurance companies is determined by the cost of compensation for insured events. In this regard, the use of algorithms for individual prediction of a patient's request for medical care can be a useful tool for increasing the economic efficiency of an insurance company.
We have created a simple model using machine learning, which is based on several features that can be obtained directly from the patient. It assesses how likely the patient is to go to a clinic or hospital.
The selection of features was carried out on the basis of the study of domestic and foreign experience in the application of machine learning methods in underwriting, consultations with medical experts and a series of preliminary experiments. The results of our research showed that the most significant parameters for predicting the occurrence of insured events are the history of medical care requests and patient's social data, while the contribution of medical information that can be obtained from the patient questionnaire is noticeably smaller.
The model predicts three facts of insured event occurrence:
- outpatient visit in the future
- hospitalization in the future
- calling an ambulance in the future
We invite you to learn more with a demo version of the model: https://webiomed.ai/machine-learning/prognozirovanie-obrashcheniia-za-12-mesiatsev/
You can read more about the prospects for digital underwriting here: https://webiomed.ai/blog/perspektivy-tsifrovogo-anderraitinga-na-osnove-ML/
Our studies have shown that additional consideration of data from an electronic health record, such as anamnesis, history of visits, and a number of characteristic medical information, can significantly increase the accuracy of model predictions. Therefore, we are going to prepare the second version of the model in the nearest future, which can work in conjunction with anonymized data from the patients' EHR.