AI in healthcare: Google advances in predictive analytics
Nobel Laureate Herbert Simon once said:
“What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it.”
Google quoted this saying in its patent filing that will be discussed in this post.
Google is aggressively working to push the use of AI in healthcare industry. It is applying its expertise in artificial intelligence to diagnose and treat diseases. It is leveraging on deep learning methods in the areas of ophthalmology, oncology and genomics to name a few.
A patent application of Google related to predicting future clinical events, including death was published by the USPTO on January 31st, 2019.
The invention makes use of deep learning models to do the prediction.
Google notes in its filing, “In the clinical setting, the management and presentation of information regarding a patient is an important aspect of patient care and healthcare decision making, for example how to treat a patient or when to discharge a patient from a hospital.
Management of information is a particularly acute issue in a busy hospital or clinic situation where a healthcare provider, such as a nurse or physician, is attending to many patients simultaneously.
Information, for example, contained within the electronic health records of a patient, consumes the attention of the recipient (e.g., nurse or physician). A wealth of information, for example as contained in an extensive medical history for a particular patient over many years, or more usually the medical history of a multitude of patients, creates a poverty of attention.”
It further notes, “There is a need for systems and methods to assist healthcare providers to allocate their attention efficiently among the overabundance of information from diverse sources, as well as to provide predictions of future clinical events and highlighting of relevant underlying medical events contributing to these predictions in a timely manner.”
To solve this problem, Google came up with a method for predicting future clinical events. This invention addresses a pressing question facing the physician in the hospital, namely which patients have the highest need for my attention now and, at an individual level, what information in the patient’s chart should I attend to?
Method for predicting clinical events from electronic health records
Step 1: The first step of the invention is to aggregate electronic health records (EHR) from a large group of patients of diverse age, health conditions, and demographics. The EHRs are obtained in a de-identified form from hospitals or medical systems. The EHRs contain medications, laboratory values, diagnoses, vital signs, and medical notes.
Step 2: The aggregated electronic health records are converted into a single standardized data structure format and into an ordered arrangement per patient. This is done because the data from hospitals or medical institutions may be in different data formats due to lack of standardization in the industry. The standardized data structure format is the Fast Health Interoperability Resources (FHIR) format.
Step 3: Deep learning models are trained on the aggregated health records converted into the FIHR format.
Google’s invention uses the following three deep learning models.
- A Long-Short-Term Memory (LSTM) model
- A time aware Feed-Forward Model (FFM), also referred to in the patent application as a feedforward model with Time-Aware Attention
- An embedded boosted time-series model, also referred to in the patent application as an Embedded Time-Aware Boosting model.
Step 4: Using the trained deep learning models to predict future clinical events. The deep learning models also summarize or highlight past medical events related to the predicted future clinical event.
These models predict an unplanned transfer to intensive care unit, length of stay in a hospital greater than 7 days, unplanned hospitalization, ER visit or readmission within 30 days after discharge of the patient, inpatient mortality, primary diagnosis, or a complete set of primary and secondary billing diagnoses at patient discharge.
These models could also predict atypical laboratory values, including potentially things such as acute kidney injury, hypokalemia, hypoglycemia, or hyponeutrimia.
Step 5: Generating an alert of the predicted future clinical event to the healthcare provider through an electronic device like a smartphone or tablet.
The deep learning models contain “attention mechanisms” also known as “attribution mechanisms” which, when invoked, indicate how much attention the models gave to tokens corresponding to elements (individual words in a note, lab measurements, medications, etc.) in the EHR to predict the future clinical event.
The attention mechanism results are displayed on the electronic device to provide the healthcare provider with confidence in the prediction and its basis.
FIG. 1 is a schematic diagram of the overall system including aggregated electronic health records, computer executing trained deep learning models, and electronic device used by a healthcare provider which receives predictions and pertinent relevant past medical events related to the prediction from the deep learning models and has an interface to present such information on its display.
This patent application shows how Google is betting on AI in healthcare industry. Its CEO Sundar Pichai feels AI is more important than fire or electricity.
Meanwhile, scientists at University of Nottingham created an AI algorithm that can predict death.
Let us know your thoughts in the comments section.
Subscribe to our newsletter to stay updated on latest inventions.