A new study demonstrates that machine-learning strategies can be applied to routinely collected physiological data, such as heart rate and blood pressure, to provide clues about pain levels in people with sickle cell disease. Mark Panaggio of Johns Hopkins University Applied Physics Laboratory and colleagues present these findings in the open-access journal PLOS Computational Biology.
Pain is subjective, and monitoring pain can be intrusive and time-consuming. Pain medication can help, but accurate knowledge of a patient’s pain is necessary to balance relief against risk of addiction or other unwanted effects. Machine-learning strategies have shown promise in predicting pain from objective physiological measurements, such as muscle activity or facial expressions, but few studies have applied machine learning to routinely collected data.
Now, Panaggio and colleagues have developed and applied machine-learning models to data from people with sickle cell disease who were hospitalized due to debilitating pain. These statistical models classify whether a patient’s pain was low, moderate, or high at each point during their stay based on routinely collected measurements of their blood pressure, heart rate, temperature, respiratory rate, and oxygen levels.
The researchers found that these vital signs indeed gave clues into the patients’ reported pain levels. By taking physiological data into account, their models outperformed baseline models in estimating subjective pain levels, detecting changes in pain, and identifying atypical pain levels. Pain predictions were most accurate when they accounted for changes in patients’ vital signs over time.
“Studies like ours show the potential that data-driven models based on machine learning have to enhance our ability to monitor patients in less invasive ways and ultimately, be able to provide more timely and targeted treatments,” Panaggio says.
Looking ahead, the researchers hope to leverage more comprehensive data sources and real-time monitoring tools, such as fitness trackers, to build better models for inferring and forecasting pain.
Materials provided by PLOS. Note: Content may be edited for style and length.