While making real time diagnosis decisions, it is difficult for most of the doctors to integrate and monitor all the data involving charts, test results, and other metrics for multiple patients. Having concern over this, MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) find out the ways for computers that will improve patient care.
Scientists have created machine learning approach that takes detailed intensive-care-unit (ICU) data from labs, demographics, etc. The system is named as ‘ICU Intervene’.
It actually makes use of deep learning algorithm and thus predicts real time diagnosis decision based on past ICU cases. It also explains the reasoning behind these decisions.
Importantly, the ICU Intervene focus on an hourly prediction of five different interventions that cover a wide assortment of basic care needs. At each hour, the system extracts data values and represent vital signs and clinical notes. All the data represents values that indicate how far off a patient is from the average.
PhD student Harini Suresh said, “The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment. The goal is to leverage data from medical records to improve health care and predict actionable interventions.”
“In addition, the system is able to use a single model to predict many outcomes and actionable treatments.”
Moreover, the system can predict whether a patient will need a ventilator six hours later rather than just 30 minutes or an hour later. Scientists developed this by focusing on providing reasoning for the model’s predictions, giving physicians more insight.
While testing it on patients, scientists found that the system outperformed previous work in predicting interventions. They found that the system is especially good at predicting the need for vasopressors. Vasopressor is the medication that tightens blood vessels and raises blood pressure.
According to researchers, the ICU Intervene will improve patient care and provide more advanced reasoning for decisions. For example, why one patient might be able to taper off steroids, or why another might need a procedure like an endoscopy.
EHR Model Transfer
In addition to this ICU Intervene, another team of scientists developed ‘EHR Model Transfer’ approach. This approach can facilitate the application of predictive models on an electronic health record (EHR) system.
According to scientists, using this approach will predict the models for mortality and prolonged length of stay can be trained on one EHR system.
The approach works by involving different versions of EHR platforms, using natural language processing. Thus, it is able to identify differently encoded clinical concepts across systems and then map them into a common set of clinical concepts.
Influencing ICU data involves how it’s stored and what happens when that storage method gets changed. Currently, available models require data to be encoded in a consistent way. Due to this, many hospitals often change their EHR systems as they generate problems while data analysis.
That’s where EHR Model Transfer comes in.
Nigam Shah, an associate professor of medicine at Standford University said, “Machine-learning models in healthcare often suffer from the low external validity and poor portability across sites. The authors devise a nifty strategy for using prior knowledge in medical ontologies to derive a shared representation across two sites that allow models trained at one site to perform well at another site. I am excited to see such creative use of codified medical knowledge in improving portability of predictive models.”
When testing their model with EHR systems, scientists found two outcomes: Mortality and prolonged stay. Based on that, scientists trained it on one EHR platform and then tested its predictions on a different platform. They found that the system outperformed baseline approaches and demonstrated the better transfer of predictive models.
Although, Both models were trained using data from the critical care database MIMIC. Scientists involved data from roughly 40,000 critical care patients.