New statistical models developed using precise physiological data from over 100 surgeries provide objective and accurate insights into ‘nociception,’ which is the body’s subconscious perception of pain.
The way an anesthesiologist manages a surgical patient’s subconscious pain processing, known as “nociception,” greatly influences the extent of post-operative drug side effects and the need for additional pain relief. Measuring pain can be quite challenging since it is a subjective experience, particularly when a patient is unconscious. In a recent study, researchers from MIT and Massachusetts General Hospital (MGH) introduced a series of statistical models designed to objectively measure nociception during surgeries. Their aim is to assist anesthesiologists in optimizing medication dosages to reduce post-operative pain and side effects.
The newly created models analyze data that was carefully logged over a span of 18,582 minutes during 101 abdominal surgeries involving both men and women at MGH. Under the guidance of Sandya Subramanian, a former MIT graduate student now serving as an assistant professor at UC Berkeley and UC San Francisco, the researchers analyzed information from five physiological sensors while patients experienced a total of 49,878 unique “nociceptive stimuli” (e.g., surgical incisions or cauterization). Additionally, they documented the types of medications used, including dosage and timing, to understand how these impacted nociception and cardiovascular responses. This comprehensive data collection was then used to develop models that could effectively indicate bodily reactions to nociceptive stimuli retrospectively.
The main goal for the team is to provide anesthesiologists with real-time, precise, and objective physiological insights, allowing them to make better decisions for administering pain relief drugs during surgery. Administering too much medication can subject patients to side effects like nausea or delirium, while insufficient dosages may leave them in significant pain upon waking.
“Sandya’s research has enabled us to establish a solid foundation for understanding and measuring nociception (unconscious pain) during general anesthesia,” stated study senior author Emery N. Brown, who is the Edward Hood Taplin Professor of Medical Engineering and Computational Neuroscience at MIT, as well as an anesthesiologist at MGH and a Professor at Harvard Medical School. “Our upcoming goal is to transform the insights we’ve gathered from Sandya’s work into practical and reliable tools that anesthesiologists can utilize during surgeries.”
Surgery and statistics
This research, which was published in The Proceedings of the National Academy of Sciences, originated from Subramanian’s doctoral thesis project in Brown’s lab back in 2017. Previous attempts to objectively model nociception primarily relied on electrocardiogram (ECG, an indirect marker of heart-rate variability) or other systems, which either didn’t capture the true intensity of surgical pain or validated their findings through limited time points in multiple surgeries, according to Subramanian.
“The operating room is the only setting to examine surgical pain,” noted Subramanian. “We aimed not just to develop algorithms from surgical data but also to validate them in a real surgical context. If we want anesthesiologists to monitor nociception minute by minute, we must confirm its accuracy in that environment.”
Thus, she and Brown enhanced the existing methods by collecting data from various sensors throughout the entire surgery while also considering how the administered drugs affected the outcomes. Their objective was to create a model capable of making accurate predictions consistent throughout the surgical procedure.
Improvements stemmed partly from monitoring heart rate alongside skin conductance. Variations in these two physiological signals can reveal the body’s instinctual “fight or flight” response to pain, yet some surgical drugs can directly affect the cardiovascular system, while skin conductance (or “EDA,” electrodermal activity) typically does not. The study measured ECG, supplemented with PPG, an optical method for assessing heart rate (similar to the function of a smartwatch’s oxygen sensor), as ECG signals can be distorted by electrical equipment in the operating room. Subramanian also supported EDA readings with skin temperature measurements to ensure any detected changes in skin conductance were due to nociception rather than simply the patient being overheated. Respiration was also recorded.
Following data collection, the authors performed statistical analyses to generate physiological indices from the cardiovascular and skin conductance signals. Once these indices were established, additional statistical procedures enabled them to combine these indices together into models capable of accurately predicting moments of nociception and the body’s responses to it.
Nailing nociception
Subramanian developed four model versions, training them with data indicating when nociceptive stimuli occurred, allowing them to correlate physiological measurements with these painful events. Some models excluded drug information, while others applied different statistical techniques, such as “linear regression” or “random forest.” One model was based on a “state space” method, allowing it to independently identify nociception cues merely from the physiological indices. The performance of these models was compared against an existing industry standard known as the ANI model, which tracks ECG.
The output from each model can be displayed as a graph illustrating predicted nociception levels over time. The ANI model performed slightly better than random chance but operates in real-time. The unsupervised model outperformed ANI, albeit not as significantly as the supervised models, particularly one that factored in drug information and utilized a “random forest” approach. The findings also indicate that the unsupervised model’s performance above chance suggests that nociception can be objectively identified across different patients.
“Utilizing a state space framework with multi-sensory physiological observations effectively reveals this latent nociceptive state consistently across various subjects,” stated Subramanian, Brown, and their co-authors. “This marks an important advancement towards creating a measure for monitoring nociception without relying on nociception ‘ground truth’ data, making it more practical and scalable for clinical application.”
Future research will focus on expanding data sampling and fine-tuning the models for implementation in the operating room, enabling real-time predictions of nociception, which would assist anesthesiologists or intensivists in their medication dosing decisions. Looking further ahead, the modeling could also lead to automated systems that manage drug dosing under the anesthesiologist’s oversight.
“Our study is a crucial step forward in developing objective tools for tracking surgical nociception,” concluded the authors. “These tools could enable the objective evaluation of nociception in other complex clinical environments, including ICU settings, and pave the way for advanced closed-loop control systems for managing pain.”
In addition to Subramanian and Brown, the research team includes Bryan Tseng, Marcela del Carmen, Annekathryn Goodman, Douglas Dahl, and Riccardo Barbieri.
This study received funding from The JPB Foundation, The Picower Institute for Learning and Memory, George J. Elbaum (MIT ’59, SM ’63, PhD ’67), Mimi Jensen, Diane B. Greene (MIT, SM ’78), Mendel Rosenblum, Bill Swanson, Cathy and Lou Paglia, along with additional support from annual donors to the Anesthesia Initiative Fund, the National Science Foundation, and an MIT Office of Graduate Education Collabmore-Rogers Fellowship.
