Early warning system model predicts deterioration of cancer patients

About 9% of cancer patients experience complications while in hospital that lead to their condition deteriorating, being transferred to the intensive care unit, or even death. A multidisciplinary team of researchers at Washington University in St. Louis is developing a machine learning-based early warning system model to predict this deterioration and improve patient outcomes.


Chenyang Lu, Fullgraf Professor at the McKelvey School of Engineering, with collaborators including Marin Kollef, MD, the Golman Professor of Medicine at the School of Medicine and Director of the Medical Intensive Care Unit and Respiratory Care Services at the Barnes-Jewish Hospital, and Patrick Lyons, MD, medical instructor in the Faculty of Medicine, recently developed a new predictive model for inpatient cancer patients that integrates heterogeneous data available in electronic health records (EHRs).


Using historical, anonymized data from more than 20,000 hospitalizations of cancer patients at Barnes-Jewish Hospital, Lu and Dingwen Li, a doctoral student in his lab and first author of the article, found a way to integrate two types of valuable data. in deep learning models that could offer clues about a patient’s condition: static data or data collected at the time of admission, such as demographics, other medical diagnoses, or information on previous hospitalizations; and time series data, which is collected multiple times during a hospital stay and includes body temperature, blood pressure, medications, and test results.

Li presented the results of their work on November 3 at the Association for Computing Machinery conference on information and knowledge management.

Because both static and time series data contain additional information related to clinical deterioration, it is important for a predictive model to exploit both types of variables to maximize its accuracy, Lu said.

“There are warning signs hidden in the data that suggest that a person will develop clinical deterioration within hours or days,” said Lu, an expert in the Internet of Things, cyber-physical systems, and technology. clinical artificial intelligence. “Humans can’t see these patterns or trends hidden in the data, so this is where machine learning is very effective in detecting these patterns. “

Lu and his team used a Recurrent Neural Network (RNN) model originally designed for time series data and improved it to incorporate static data using a multimodal fusion approach. Their end-to-end model, called CrossNet, learns to predict deterioration events while accurately imputing any missing static or time-series data. This new approach for incorporating both static and time series data combines the power of deep recurring models with the advantages of heterogeneous data in the EHR.

Ideally, an early warning system would learn from a patient’s data the signs that the patient is deteriorating, which would set off an alarm calling healthcare providers to the patient’s bedside. However, one of the risks of such a system is that an alarm would sound so frequently, possibly triggered by false alarms, that healthcare providers would develop alarm fatigue and eventually stop responding.

In a case study in realistic hospital care settings, Lu and the team set a threshold of 48 notifications over a 24-hour period, or one every 30 minutes. The team then implemented a more proactive early warning system where the alarm rate can be high, but the number of false alarms is limited to avoid alarm fatigue. With the same false alarm rate, the team’s CrossNet model captured 39.5% of clinical deterioration events, while an existing model used by many hospitals called Modified Early Warning Scores (MEWS) n ‘captured only 3.9% of the same events.

While the model has potential, Lu is working with the doctors on the team to determine how best to implement it in a hospital setting.


Kollef said Barnes-Jewish Hospital had been using a simpler early warning system for about 15 years. After an assessment, alarms from this system are sent to an early intervention team who can assess and triage patients.

“An alert only makes sense if it is linked to an intervention,” Kollef said. “It’s easy for someone to take data out of a machine and analyze it, but what do you do with it? This is the challenge.

Humans can’t see these patterns or trends hidden in the data, so this is where machine learning is very good at detecting these patterns.

Chenyang Lu

Kollef, who worked in the intensive care unit for 35 years, said that an early warning system is a step in the right direction and that collaborators such as Lyon are essential to the implementation of an such system.


“Cancer patients are often very sick and fragile and are already being monitored intensively,” said Lyons, a clinician specializing in informatics. “With chemotherapy and other treatments, they generate a lot of data that is difficult to sort out in any meaningful way. We would like to use this model to distill what data is going to point healthcare providers in a clear direction. “

Lyons said the team is seeking funds to build an infrastructure around their model and test it to see if it improves processes of care. In the meantime, he organizes focus groups with patients and nurses to determine their priorities.

Li D, Lyons P, Klaus J, Gage B, Kollef M, Lu C. Integration of static and time series data into deep recurrent models for early warning systems in oncology. Acts of 30e ACM International Conference on Information and Knowledge Management, November 1-5, 2021. https://doi.org/10.1145/3459637.3482441

The McKelvey School of Engineering at Washington University in St. Louis promotes independent research and education with an emphasis on scientific excellence, innovation and collaboration without borders. McKelvey Engineering offers top-notch research and graduate programs across all departments, particularly in Biomedical Engineering, Environmental Engineering, and Computer Science, and has one of the most selective undergraduate programs in the world. country. With 140 full-time faculty, 1,387 undergraduates, 1,448 graduate students and 21,000 alumni alive, we work to solve some of society’s greatest challenges; prepare students to become leaders and innovate throughout their careers; and be a catalyst for economic development for the Saint-Louis region and beyond.

Comments are closed.