Researchers from Google and the University of California have developed an AI system that can help to prevent deaths due to incorrect prescriptions.
Only in rare cases, incorrect prescriptions can result in hospitalization and sometimes even death or can badly interfere with a patient’s existing medications.
As per the researchers, the AI system can identify the conditions that a patient is getting treatment for depending on certain parameters. For instance, if a doctor prescribed doxycycline and ceftriaxone for a patient with fever, cough, and an elevated temperature, the AI will consider these as signals that the person was given treatment for pneumonia.
In the future, an AI model may step in if a patient is prescribed a medication that looks incorrect for a particular condition in their current scenario.
As per the researchers, even though no doctor, pharmacist or nurse may want to make a mistake that can harm a patient, a study shows that 2% of the hospitalized patients are known to experience serious incidents related to medication that could cause permanent harm, be life-threatening or can even result in death.
However, doctors and pharmacists train for years to acquire the complex skill of prescribing medications appropriately for a given patient at all times.
The Ai model was trained using anonymized datasets that feature about 3 million medication records issued from about 100,000 hospitalizations.
Since the patient records vary significantly in density and length of data points, the researchers formulated 3 Deep Learning Neural Network model architectures that can leverage such data in different ways. One is based on long short-term memory or recurrent neural networks, the other one on a neural network having boosted time-based decision stamps and the third one on attention-based TANN.
Each architecture was trained on each task and multiple time points such as before admission, at the time of admission, 24 hrs after admission, and at discharge, and the results of each architecture were combined by using ensembling.