At present, the main sources of medical data are medical institutions (for example, medical pharmacy laboratories, medical rehabilitation centers, etc. ) and the internet. The collected data has a wide range, high dimension and various types, and is not aimed at a specific problem.
2. Measurement of uncertainty.
At present, most mature and practical data models are aimed at pharmaceutical companies and insurance companies. In the application of medical big data in the United States, it is often difficult to face doctors and patients, and it is difficult to find a suitable starting point. Business oriented to enterprises is relatively easy, especially insurance companies and pharmaceutical companies, but relatively difficult. Due to the limited accuracy of the big data model, its practical value is very limited for doctors and doctors with extremely high security requirements. For example, a model with an accuracy rate of 95% may still be inaccurate for doctors, because doctors make decisions based on individual patients rather than statistical significance.
In addition, the interpretability of statistical learning model is poor. Only statisticians and computer scientists can explain the model accurately and completely, but there are great obstacles for real users of the model, such as doctors and government officials.