Patrick Lee Purdon

All Publications

• ICA-S3M: switching state-space model guided automatic EEG artifact removal from independent components. Frontiers in neuroscience Tou, S. L., He, M., Oshiro, S. Y., Das, P., Purdon, P. L. 2026; 20: 1865301

  Abstract

  Electromyographic (EMG) artifact is among the most challenging contaminants in scalp EEG: its broadband spectrum overlaps directly with neural activity, and sustained muscle contractions distribute across many independent components (ICs) rather than segregating cleanly-an effect we term EMG smearing.We propose ICA-S3M, a two-stage pipeline that first decomposes the recording via independent component analysis (ICA), then applies a switching state-space model to each retained IC. This model separates neural oscillations, modeled as damped oscillators fitted to the recording's own spectral content, from broadband artifact, modeled as an autoregressive (AR(2)) process. The method produces an explicit artifact probability at every time point and requires no training data. We evaluated ICA-S3M against a CNN from EEGdenoiseNet in three settings: simulated EEG with known ground truth; a semi-synthetic real-EEG dataset constructed from 29 participants through 11,025 expert IC-trial classifications, in which a trained rater classified every IC on a per-trial basis across interleaved rest and facial-movement segments; and a naturalistic recording of musical improvisation.In simulation, ICA-S3M reduced RRMSE from 0.83 to 0.23 and improved SNR by 12.4 dB, versus 1.2 dB for the CNN. On the semi-synthetic dataset, ICA-S3M significantly outperformed the CNN in 38 of 42 tested scenarios spanning seven metrics, two time periods, and three scalp regions, while leaving clean segments essentially untouched. Critically, the CNN exhibited an "alpha hallucination" failure mode, injecting spurious alpha-band peaks where none exist in the ground truth; we reproduced this with an alpha-free control simulation and observed it again on real scalp recordings.ICA-S3M avoids this failure by adapting to each recording's own oscillatory structure rather than imposing learned spectral templates. The method offers a principled, interpretable, and training-free alternative for EMG artifact removal in challenging EEG recordings.

  View details for DOI 10.3389/fnins.2026.1865301

  View details for PubMedID 42602229

  View details for PubMedCentralID PMC13473372

• Rethinking the Integration of Artificial Intelligence Into Surgery: Centralized Risk Assessment for System-Level Impact. Annals of surgery open : perspectives of surgical history, education, and clinical approaches Mather, R. V., Kunitake, H., Purdon, P. L. 2026; 7 (2): e661

  View details for DOI 10.1097/AS9.0000000000000661

  View details for PubMedID 42344479

  View details for PubMedCentralID PMC13290149

• A Prospective Study Characterizing Cognitive Function in Patients with Inflammatory Bowel Disease. Clinical and translational gastroenterology Raje, P., Stone, T. A., Alkourabi, J., Mather, R. V., Liu, R., Higuchi, M., Nipp, R. D., Purdon, P. L., Kochar, B., Kunitake, H. 2026

  Abstract

  Inflammatory bowel disease (IBD) may be associated with cognitive impairment. Cognitive decline is also linked to weaker anesthesia-induced alpha wave electroencephalographic (EEG) signals. We aimed to characterize the associations between cognition and EEG alpha power in patients with IBD.In this prospective cohort study, patients with IBD and controls undergoing diagnostic or screening colonoscopies underwent preprocedural cognitive testing using the tablet-based Brain Health Assessment (BHA), intraprocedural EEG monitoring, and follow-up testing. Primary outcomes were BHA scores and EEG alpha power. Secondary outcomes included within-participant changes in cognitive performance.We enrolled 40 patients with IBD and 42 control patients. Fifteen IBD patients and 17 controls completed follow-up cognitive testing 6-18 months after endoscopy. Patients with IBD were younger (mean age 42 vs. 56 years, p<0.001), more likely to screen positively for depression (p=0.004), and had fewer years of education (16.2 vs. 17.3 years, p=0.03). Fifteen IBD patients had active endoscopic inflammation. Adjusting for demographics, education level, and depression, EEG alpha power did not differ between groups. Median BHA scores indicated moderate likelihood of cognitive impairment in both groups. However, controls demonstrated improved within-participant follow-up performance (p<0.01), while IBD patients did not (p=0.16).IBD patients and controls demonstrate preprocedural cognitive impairment on BHA, but no differences in EEG alpha power. Lack of follow-up improvement in IBD patients may suggest lower baseline cognitive function while highlighting the importance of further investigations on cognition in this population.

  View details for DOI 10.14309/ctg.0000000000001036

  View details for PubMedID 41995584

• Intraoperative Electroencephalogram Alpha Power Associated with Mortality: Reply. Anesthesiology Mather, R. V., Nipp, R. D., Balanza, G., Stone, T. A., Gutierrez, R., Raje, P., Higuchi, M., Liu, R., Mercado, L. A., Bittner, E. A., Kunitake, H., Purdon, P. L. 2025; 143 (5): 1425-1427

  View details for DOI 10.1097/ALN.0000000000005676

  View details for PubMedID 41085317

• Association of peripheral nerve blocks with increased postoperative pain and opioid use in orthopaedic surgery: a single-centre retrospective cohort study. British journal of anaesthesia Chung, A. R., Mather, R. V., Gutierrez, R., Liu, R., Leung, C. F., Zhang, M., Santa Cruz Mercado, L. A., Houle, T. T., Bittner, E. A., Purdon, P. L. 2025

  Abstract

  Peripheral nerve blocks have become popular in orthopaedic surgeries to improve acute postoperative pain. However, studies are mixed on their effectiveness in decreasing postoperative opioid consumption. A more comprehensive analysis is necessary to understand if peripheral nerve blocks reduce postoperative opioid exposure and risk for opioid dependence.This retrospective cohort study evaluated electronic health record data for adults undergoing orthopaedic surgery with general anaesthesia from 2016 to 2020 at the Massachusetts General Hospital. Linear models were fitted on propensity-weighted data to characterise the association between single injection peripheral nerve blocks and clinical outcomes. Our primary outcomes were maximum pain score and cumulative opioid dose, quantified in morphine milligram equivalents, administered in the PACU. Post-discharge outcomes associated with pain and opioid consumption were also evaluated.Among 22 956 patients, peripheral nerve block administration was associated with lower maximum pain scores and lower probability of opioid administration in the PACU. However, it was associated with higher maximum pain scores and a 22.7% increase in opioid consumption during the hospital stay. Peripheral nerve blocks were associated with an increase in opioid prescriptions at 30 days after discharge, but no increase at 90 or 180 days, and with decreased chronic pain diagnoses 1 yr after operation.Although single injection peripheral nerve blocks were effective in reducing immediate postoperative pain and opioid consumption, they were associated with greater opioid consumption that could increase the risk for opioid dependence. Standardised protocols to mitigate the risk for rebound pain could help minimise postoperative opioid exposure.

  View details for DOI 10.1016/j.bja.2025.05.030

  View details for PubMedID 40610285

• Retrospective Study on Perioperative Opioid Administration and Maximum Perioperative Pain by Median Income of Patient Zip Code Sydloski, J., Mather, R., Johnson, J., Gutierrez, R., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 725-727

  View details for Web of Science ID 001551889100282

• State-Space Oscillator Modeling to Identify a Personalized Intraoperatiye EEG Marker for Opioid Effects Bharadwar, K., Mather, R., He, M., Das, P., Villegas, G., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 584-589

  View details for Web of Science ID 001551889100227

• EEG Patterns Correlated with Biomarkers of Alzheimer's Disease and Neurodegeneration after Cardiac Surgery Mather, R., Feinstein, I., Gutierrez, R., Young, C., McAllister, T., Mendiola, J., Gao, X., Porras, G., Kitsis, A., Hansine, A., Delaney, G., Wilson, E., Angst, M., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 485

  View details for Web of Science ID 001551889100191

• Phase-Amplitude Modulation Between Delta and Alpha Oscillations in the Electroencephalogram (EEG) During Pediatric Anesthesia Chao, J., Gutierrez, R., Chao, C., Zhang, L., Yozawitz, E., Legatt, A., Jackson, W., Nafiu, B., Kudchadkart, S., Li, G., Eikermann, M., Lo, Y., Shinnar, S., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 901-902

  View details for Web of Science ID 001551889100348

• Biomarkers of Alzheimer's Disease and Neurodegeneration After Cardiac Surgery: First Insights from the Ongoing CARDIAC-PND Study Feinstein, I., Young, C., McAllister, T., Mendiola, J., Gao, X., Porras, G., Kitsis, A., Huhn, A., Delaney, G., Mather, R., Gutierrez, R., Purdon, P., Wilson, E., Angst, M. LIPPINCOTT WILLIAMS & WILKINS. 2025: 530-533

  View details for Web of Science ID 001551889100207

• Association of Single-Shot Peripheral Nerve Block With Postoperative Pain and Opioid Use in Orthopedic Surgery Chung, A., Mather, R., Gutierrez, R., Liu, R., Leung, A., Zhang, M., Mercado, L., Houle, T., Bittner, E., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 1100-1105

  View details for Web of Science ID 001551889100422

• Comparison of Pain Assessments and Max Pain Scores by Language Group in Women Receiving Neuraxial Analgesia for Vaginal Delivery Pandit, A., Gutierrez, R., Leung, A., Mather, R., Johnson, J., Chung, A., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 607-609

  View details for Web of Science ID 001551889100233

• The Effects of Timing of Intraoperative Opioid Administration on Postoperative Pain and Opioid Use Outcomes Liu, R., Gutierrez, R., Mather, R., Stone, T., Bittner, E., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 734-738

  View details for Web of Science ID 001551889100285

• Intraoperative Frontal EEG Alpha Power is Associated with Postoperative Mortality and Other Adverse Outcomes Mather, R., Villegas, G., Stone, T., Gutierrez, R., Raje, P., Higuchi, M., Liu, R., Mercado, L., Bittner, E., Nipp, R., Kunitake, H., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2025: 554-558

  View details for Web of Science ID 001551889100215

• A ketogenic diet decreases sevoflurane-induced burst suppression in rats. Brain research bulletin Siegmann, M. J., Parry, S., Lark, A. R., Mir, F. A., Choi, J., Carpenter, A. H., Crowley, E. A., White, C. G., Kang, J., Purdon, P. L., Nehs, C. J. 2025: 111274

  Abstract

  The brain requires a continuous fuel supply to support cognition and can get energy from glucose and ketones. Dysregulated brain metabolism is thought to contribute to perioperative neurocognitive disorders and anesthesia-induced burst suppression. Therefore, we investigated the relationship between brain metabolites and neurophysiology during the behavioral states of sleep and anesthesia under a standard diet (SD) or a ketogenic diet (KD).We measured prefrontal cortex glucose, lactate, and electroencephalogram in Fischer344 rats during spontaneous sleep/wake followed by 3% sevoflurane anesthesia. Nine rats were fed a KD and 8 rats a SD. To assess the role of adenosine receptor-mediated ketone activity on burst suppression, 5 additional rats on the KD received an intraperitoneal injection of vehicle or the adenosine A1 receptor antagonist, DPCPX, before 3% sevoflurane.Sevoflurane induced larger fluctuations in glucose (p<0.001) and lactate (p=0.015) concentrations compared to sleep as measured by the standard deviation (glucose 0.085mM and lactate 0.16mM in sleep/wake and 0.25mM and 0.41mM during sevoflurane respectively). Changes in glucose and lactate were closely tied to electrophysiological oscillations. Animals on the KD had reduced burst suppression ratio (mean 10% in KD vs 30% in SD) (p=0.007) as well as increased time to loss of movement (mean 14min in KD vs 8min in SD) (p=0.003) compared to SD. DPCPX in KD rats showed a trend to increased burst suppression, reduced the time to start of burst suppression (45min in KD+vehicle to 37min KD+DPCPX) (p=0.007), and increased duration of burst suppression (49min in KD+vehicle to 90min in KD+DPCPX) (p=0.046) compared to KD+vehicle.It is thought that anesthesia-induced burst suppression reflects an underlying deficiency in brain energy. Accordingly, we found that upregulating ketones, which increase available brain ATP levels, delayed anesthetic induction and decreased burst suppression consistent with the idea that the underlying metabolic state of the brain influences an anesthetic's effect on the brain. These findings suggest that metabolic interventions could be useful therapeutic targets to modulate brain activity during sleep and anesthesia. Future studies will examine whether ketones can reduce the cognitive symptoms associated with postoperative delirium.

  View details for DOI 10.1016/j.brainresbull.2025.111274

  View details for PubMedID 40010575

• Electroencephalogram-Guided General Anesthesia in a Pediatric Patient With Alexander's Disease: A Case Report. A&A practice Zhang, M., Mather, R. V., Chung, A. R., Leung, C. F., Ramamurthi, R. J., Purdon, P. L. 2025; 19 (1): e01910

  Abstract

  In this case, the electroencephalogram (EEG) was used to guide anesthesia care for a pediatric patient with Alexander's Disease undergoing serial intrathecal injections. Previous procedures using a standard maintenance propofol dose of up to 225 µg/kg/min led to postanesthetic recovery times of over 6 hours, requiring a neurology consult for noncoherence. The EEG assisted in guiding maintenance propofol dosing to 75 µg/kg/min, decreasing postanesthetic wash-off and postanesthesia care unit (PACU) recovery time by 50%. This highlights the potential impact of astrocyte dysfunction on anesthetic sensitivity and robustness of EEG as a biomarker of anesthetic effect, including for pediatric patients with rare neurodevelopmental diseases.

  View details for DOI 10.1213/XAA.0000000000001910

  View details for PubMedID 39831714

• EEG Artifact Removal Using Switching State Space Models 2025 2025 59th Asilomar Conference on Signals, Systems, and Computers Stone, T. A., Tou, S., He, M., Beck, A., Purdon, P. L. 2025: 6

  View details for DOI 10.1109/IEEECONF67917.2025.11443890

• Intraoperative Frontal EEG Alpha Power is Associated with Post-Operative Mortality and Other Adverse Outcomes. Anesthesiology Mather, R. V., Nipp, R. D., Balanza, G., Stone, T. A., Gutierrez, R., Raje, P., Higuchi, M., Liu, R., Santa Cruz Mercado, L. A., Bittner, E. A., Kunitake, H., Purdon, P. L. 2024

  Abstract

  With estimated global post-operative mortality rates at 1-4% leading to approximately 3-12 million deaths per year, an urgent need exists for reliable measures of perioperative risk. Existing approaches suffer from poor performance, place a high burden on clinicians to gather data, or do not incorporate intraoperative data. Prior work demonstrated that intraoperative anesthetics induce prefrontal EEG oscillations in the alpha band (8-12Hz) that correlate with post-operative cognitive outcomes.We analyzed a retrospective cohort of 1,081 patients undergoing surgery with general anesthesia at Massachusetts General Hospital with intraoperative EEG recordings. The association between EEG alpha power and adverse outcomes were characterized using statistical models that were fitted on propensity weighted data. Our primary outcome was post-operative mortality, measured from date of surgery to date of death or last follow-up. Secondary outcomes included mortality within pre-specified time windows (30-days, 90-days, 180-days, and 1-year), hospital and PACU lengths of stay, discharge to long-term care, and 30-day hospital readmission.Alpha power was associated with mortality risk (HR = 0.92, 95% CI:[ 0.85, 0.99], p=0.039). Within specified time windows, alpha power was associated with 30-day mortality (OR = 0.81, 95% CI: [0.66, 0.95], p=0.010), 90-day mortality (OR = 0.68, 95% CI: [0.55, 0.79], p<0.001), 180-day mortality (OR = 0.75, 95% CI: [0.66, 0.83], p<0.001), and 1-year mortality (OR = 0.85, 95% CI: [0.79, 0.91], p<0.001). Additionally, alpha power was associated with discharge to long-term care (OR = 0.91, 95% CI: [0.86, 0.96], p<0.001). We did not find significant associations among alpha power and 30-day readmission and hospital or PACU lengths of stay.Intraoperative EEG alpha power is independently associated with post-operative mortality and adverse outcomes, suggesting it could represent a broad measure of post-operative physical resilience and provide clinicians with a low-burden, personalized measure of post-operative risk.

  View details for DOI 10.1097/ALN.0000000000005315

  View details for PubMedID 39601585

• Intraoperative EEG Alpha Power Predicts Postoperative Mortality and Adverse Postoperative Outcomes Mather, R., Villegas, G., Stone, T., Gutierrez, R., Raje, P., Higuchi, M., Liu, R., Mercado, L., Bittner, E., Kunitake, H., Nipp, R., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2024: 449-453

  View details for Web of Science ID 001349531300176

• Electroencephalographic (EEG) Characteristics in Children Undergoing Sevoflurane Anesthesia: Comparing EEG-Guided Anesthesia vs Standard Care Bong, C., Villegas, G., Stone, T., Gutierrez, R., Liang, S., Allen, J., Purdon, P. LIPPINCOTT WILLIAMS & WILKINS. 2024: 677

  View details for Web of Science ID 001349531300264

• Depth of anesthesia monitoring: an argument for its use for patient safety. Current opinion in anaesthesiology Brook, K., Agarwala, A. V., Li, F., Purdon, P. L. 2024

  Abstract

  PURPOSE OF REVIEW: There have been significant advancements in depth of anesthesia (DoA) technology. The Anesthesia Patient Safety Foundation recently published recommendations to use a DoA monitor in specific patient populations receiving general anesthesia. However, the universal use of DoA monitoring is not yet accepted. This review explores the current state of DoA monitors and their potential impact on patient safety.RECENT FINDINGS: We reviewed the current evidence for using a DoA monitor and its potential role in preventing awareness and preserving brain health by decreasing the incidence of postoperative delirium and postoperative cognitive dysfunction or decline (POCD). We also explored the evidence for use of DoA monitors in improving postoperative clinical indicators such as organ dysfunction, mortality and length of stay. We discuss the use of DoA monitoring in the pediatric population, as well as highlight the current limitations of DoA monitoring and the path forward.SUMMARY: There is evidence that DoA monitoring may decrease the incidence of awareness, postoperative delirium, POCD and improve several postoperative outcomes. In children, DoA monitoring may decrease the incidence of awareness and emergence delirium, but long-term effects are unknown. While there are key limitations to DoA monitoring technology, we argue that DoA monitoring shows great promise in improving patient safety in most, if not all anesthetic populations.

  View details for DOI 10.1097/ACO.0000000000001430

  View details for PubMedID 39248004

• A dynamic generative model can extract interpretable oscillatory components from multichannel neurophysiological recordings. eLife Das, P., He, M., Purdon, P. L. 2024; 13

  Abstract

  Modern neurophysiological recordings are performed using multichannel sensor arrays that are able to record activity in an increasingly high number of channels numbering in the 100s to 1000s. Often, underlying lower-dimensional patterns of activity are responsible for the observed dynamics, but these representations are difficult to reliably identify using existing methods that attempt to summarize multivariate relationships in a post hoc manner from univariate analyses or using current blind source separation methods. While such methods can reveal appealing patterns of activity, determining the number of components to include, assessing their statistical significance, and interpreting them requires extensive manual intervention and subjective judgment in practice. These difficulties with component selection and interpretation occur in large part because these methods lack a generative model for the underlying spatio-temporal dynamics. Here, we describe a novel component analysis method anchored by a generative model where each source is described by a bio-physically inspired state-space representation. The parameters governing this representation readily capture the oscillatory temporal dynamics of the components, so we refer to it as oscillation component analysis. These parameters - the oscillatory properties, the component mixing weights at the sensors, and the number of oscillations - all are inferred in a data-driven fashion within a Bayesian framework employing an instance of the expectation maximization algorithm. We analyze high-dimensional electroencephalography and magnetoencephalography recordings from human studies to illustrate the potential utility of this method for neuroscience data.

  View details for DOI 10.7554/eLife.97107

  View details for PubMedID 39146208

• For EEG-Guided Anesthesia, We Have to Go Beyond the Index. Anesthesia and analgesia Kreuzer, M., García, P. S., Gutierrez, R., Purdon, P. L. 2024

  View details for DOI 10.1213/ANE.0000000000007098

  View details for PubMedID 38885144

• A dynamic generative model can extract interpretable oscillatory components from multichannel neurophysiological recordings. bioRxiv : the preprint server for biology Das, P., He, M., Purdon, P. L. 2024

  Abstract

  Modern neurophysiological recordings are performed using multichannel sensor arrays that are able to record activity in an increasingly high number of channels numbering in the 100's to 1000's. Often, underlying lower-dimensional patterns of activity are responsible for the observed dynamics, but these representations are difficult to reliably identify using existing methods that attempt to summarize multivariate relationships in a post-hoc manner from univariate analyses, or using current blind source separation methods. While such methods can reveal appealing patterns of activity, determining the number of components to include, assessing their statistical significance, and interpreting them requires extensive manual intervention and subjective judgement in practice. These difficulties with component selection and interpretation occur in large part because these methods lack a generative model for the underlying spatio-temporal dynamics. Here we describe a novel component analysis method anchored by a generative model where each source is described by a bio-physically inspired state space representation. The parameters governing this representation readily capture the oscillatory temporal dynamics of the components, so we refer to it as Oscillation Component Analysis (OCA). These parameters - the oscillatory properties, the component mixing weights at the sensors, and the number of oscillations - all are inferred in a data-driven fashion within a Bayesian framework employing an instance of the expectation maximization algorithm. We analyze high-dimensional electroencephalography and magnetoencephalography recordings from human studies to illustrate the potential utility of this method for neuroscience data.

  View details for DOI 10.1101/2023.07.26.550594

  View details for PubMedID 37546851

  View details for PubMedCentralID PMC10402019

• Development and multicentre validation of the FLEX score: personalised preoperative surgical risk prediction using attention-based ICD-10 and Current Procedural Terminology set embeddings. British journal of anaesthesia Liu, R., Stone, T. A., Raje, P., Mather, R. V., Santa Cruz Mercado, L. A., Bharadwaj, K., Johnson, J., Higuchi, M., Nipp, R. D., Kunitake, H., Purdon, P. L. 2024

  Abstract

  BACKGROUND: Preoperative knowledge of surgical risks can improve perioperative care and patient outcomes. However, assessments requiring clinician examination of patients or manual chart review can be too burdensome for routine use.METHODS: We conducted a multicentre retrospective study of 243 479 adult noncardiac surgical patients at four hospitals within the Mass General Brigham (MGB) system in the USA. We developed a machine learning method using routinely collected coding and patient characteristics data from the electronic health record which predicts 30-day mortality, 30-day readmission, discharge to long-term care, and hospital length of stay.RESULTS: Our method, the Flexible Surgical Set Embedding (FLEX) score, achieved state-of-the-art performance to identify comorbidities that significantly contribute to the risk of each adverse outcome. The contributions of comorbidities are weighted based on patient-specific context, yielding personalised risk predictions. Understanding the significant drivers of risk of adverse outcomes for each patient can inform clinicians of potential targets for intervention.CONCLUSIONS: FLEX utilises information from a wider range of medical diagnostic and procedural codes than previously possible and can adapt to different coding practices to accurately predict adverse postoperative outcomes.

  View details for DOI 10.1016/j.bja.2023.11.039

  View details for PubMedID 38184474

• Simultaneous localization of multiple oscillations using dynamic solutions to the inverse problem in E/MEG source analysis He, M., Das, P., Beck, A., Hotan, G., Purdon, P. L. edited by Matthews, M. B. IEEE. 2024: 228-233

  View details for DOI 10.1109/IEEECONF60004.2024.10942699

  View details for Web of Science ID 001479671800044

• Development and prospective validation of postoperative pain prediction from preoperative EHR data using attention-based set embeddings. NPJ digital medicine Liu, R., Gutiérrez, R., Mather, R. V., Stone, T. A., Santa Cruz Mercado, L. A., Bharadwaj, K., Johnson, J., Das, P., Balanza, G., Uwanaka, E., Sydloski, J., Chen, A., Hagood, M., Bittner, E. A., Purdon, P. L. 2023; 6 (1): 209

  Abstract

  Preoperative knowledge of expected postoperative pain can help guide perioperative pain management and focus interventions on patients with the greatest risk of acute pain. However, current methods for predicting postoperative pain require patient and clinician input or laborious manual chart review and often do not achieve sufficient performance. We use routinely collected electronic health record data from a multicenter dataset of 234,274 adult non-cardiac surgical patients to develop a machine learning method which predicts maximum pain scores on the day of surgery and four subsequent days and validate this method in a prospective cohort. Our method, POPS, is fully automated and relies only on data available prior to surgery, allowing application in all patients scheduled for or considering surgery. Here we report that POPS achieves state-of-the-art performance and outperforms clinician predictions on all postoperative days when predicting maximum pain on the 0-10 NRS in prospective validation, though with degraded calibration. POPS is interpretable, identifying comorbidities that significantly contribute to postoperative pain based on patient-specific context, which can assist clinicians in mitigating cases of acute pain.

  View details for DOI 10.1038/s41746-023-00947-z

  View details for PubMedID 37973817

  View details for PubMedCentralID 8369227

• Anesthesia-induced Brain Oscillations and Vulnerability to Postoperative Neurocognitive Disorders. Anesthesiology Gutierrez, R., Purdon, P. L. 2023; 139 (5): 557-559

  View details for DOI 10.1097/ALN.0000000000004704

  View details for PubMedID 37815470

• Multilevel State-Space Models Enable High Precision Event Related Potential Analysis. Conference record. Asilomar Conference on Signals, Systems & Computers Das, P., He, M., Purdon, P. L. 2023; 2023: 1496-1499

  Abstract

  During cognitive tasks, the elicited brain responses that are time-locked to the stimulus presentation are manifested in electroencephalogram (EEG) as Event Related Potentials (ERPs). In general, ERPs are ~ 1 μ V signals embedded in the background of much stronger neural oscillations, and thus they are traditionally extracted by averaging hundreds of trial responses so that the neural oscillations can cancel out each other. However, often in cognitive science experiments, it is difficult to administer large number of trials due to physical constraints. Additionally, these excessive averaging can also blur fine structures of the ERPs signals, which might otherwise be indicative of various intrinsic factors. Here we propose to model the background oscillations using a novel oscillation state-space representation and identify their time-traces in a data-driven way. This allows us to effectively separate the oscillations from the response signals of interest, thus improving the signal-to-noise of the evoked response, and eventually increasing trial fidelity. We also consider a random-walk like continuity constraint for the ERP waveforms to recover smooth, de-noised estimates. We employ a generalized expectation maximization algorithm for estimating the model parameters, and then infer the approximate posterior distribution of ERP waveforms. We demonstrate the reduced reliance of our proposed ERP extraction technique via a simulation study. Finally, we showcase how the extracted ERPs using our method can be more informative than the traditional average-based ERPs when analyzing EEG data in cognitive task settings with fewer trials.

  View details for DOI 10.1109/IEEECONF59524.2023.10476951

  View details for PubMedID 39497917

  View details for PubMedCentralID PMC11534075

• Switching state-space modeling of neural signal dynamics. PLoS computational biology He, M., Das, P., Hotan, G., Purdon, P. L. 2023; 19 (8): e1011395

  Abstract

  Linear parametric state-space models are a ubiquitous tool for analyzing neural time series data, providing a way to characterize the underlying brain dynamics with much greater statistical efficiency than non-parametric data analysis approaches. However, neural time series data are frequently time-varying, exhibiting rapid changes in dynamics, with transient activity that is often the key feature of interest in the data. Stationary methods can be adapted to time-varying scenarios by employing fixed-duration windows under an assumption of quasi-stationarity. But time-varying dynamics can be explicitly modeled by switching state-space models, i.e., by using a pool of state-space models with different dynamics selected by a probabilistic switching process. Unfortunately, exact solutions for state inference and parameter learning with switching state-space models are intractable. Here we revisit a switching state-space model inference approach first proposed by Ghahramani and Hinton. We provide explicit derivations for solving the inference problem iteratively after applying a variational approximation on the joint posterior of the hidden states and the switching process. We introduce a novel initialization procedure using an efficient leave-one-out strategy to compare among candidate models, which significantly improves performance compared to the existing method that relies on deterministic annealing. We then utilize this state inference solution within a generalized expectation-maximization algorithm to estimate model parameters of the switching process and the linear state-space models with dynamics potentially shared among candidate models. We perform extensive simulations under different settings to benchmark performance against existing switching inference methods and further validate the robustness of our switching inference solution outside the generative switching model class. Finally, we demonstrate the utility of our method for sleep spindle detection in real recordings, showing how switching state-space models can be used to detect and extract transient spindles from human sleep electroencephalograms in an unsupervised manner.

  View details for DOI 10.1371/journal.pcbi.1011395

  View details for PubMedID 37639391

  View details for PubMedCentralID PMC10491408

• Multilevel State-Space Models Enable High Precision Event Related Potential Analysis Das, P., He, M., Purdon, P. L. edited by Matthews, M. B. IEEE. 2023: 1496-1499

  View details for DOI 10.1109/IEEECONF59524.2023.10476951

  View details for Web of Science ID 001207755100274