Machine Learning Information | Identifying Palliative Care Patients Based on Deep Learning

“Hiring a Machine Learning engineer or Data Scientist in Silicon Valley is becoming like hiring a professional athlete. That’s how demanding it is” — The New York Times

Identifying palliative care patients based on deep learning

Stanford ML Group built a program that uses deep learning algorithms to identify hospitalized patients at high risk of dying in the next 3-12 months based on Electronic Health Record (EHR) data. Early warning information for these patients is sent to the palliative care team, which helps the palliative care team to intervene and provide services earlier.

Palliative care (Palliative Care, translated as palliative medicine in Japan and Taiwan) originated from the hospice movement and originated in the fourth century AD. According to the definition of the World Health Organization, palliative care emphasizes the control of pain and patient-related symptoms, and pays attention to psychological, social and spiritual problems, in order to obtain the best quality of life for patients and their families.

The prediction model is an 18-layer deep neural network, the input parameter is the EHR data of a patient, and the output is the probability of death in the next 3-12 months. The training data uses historical data from the Stanford Hospital EHR database, which contains data from more than 2 million patients. The EHR data includes the patient's diagnosis conclusions, treatment procedures, prescriptions and related details (desensitized and technically processed, and represented in the form of surrogate codes) for the past 12 months, and all data are converted into 13654-dimensional feature vectors. The trained model achieves an AUROC score of 0.93 and an average cross-validation accuracy of 0.69.

For machine learning systems, enabling users to act on predictions requires providing detailed explanations of the predictions, which is critical to building user confidence. Stanford's program can automatically generate a report that highlights items in the patient's EHR data that are important for predicting outcomes.

Classification

Mybridge AI selected the top 50 out of 20,000 articles on creating machine learning applications. Learning from data scientists with hands-on experience is a great way to share lessons learned about building and operating machine learning applications. The 50 articles can be roughly divided into 15 topics as follows: