Patrick James Tighe

Interest in the potential applications of artificial intelligence in medicine, anesthesiology, and the world at large has never been higher. The Anesthesia Research Council steering committee formed an anesthesiologist artificial intelligence expert workgroup charged with evaluating the current state of artificial intelligence in anesthesiology, providing examples of future artificial intelligence applications and identifying barriers to artificial intelligence progress. The workgroup’s findings are summarized here, starting with a brief introduction to artificial intelligence for clinicians, followed by overviews of current and anticipated artificial intelligence–focused research and applications in anesthesiology. Anesthesiology’s progress in artificial intelligence is compared to that of other medical specialties, and barriers to artificial intelligence development and implementation in our specialty are discussed. The workgroup’s recommendations address stakeholders in policymaking, research, development, implementation, training, and use of artificial intelligence–based tools for perioperative care.

Interest in the potential applications of artificial intelligence in medicine, anesthesiology, and the world at large has never been higher. The Anesthesia Research Council steering committee formed an anesthesiologist artificial intelligence expert workgroup charged with evaluating the current state of artificial intelligence in anesthesiology, providing examples of future artificial intelligence applications and identifying barriers to artificial intelligence progress. The workgroup’s findings are summarized here, starting with a brief introduction to artificial intelligence for clinicians, followed by overviews of current and anticipated artificial intelligence–focused research and applications in anesthesiology. Anesthesiology’s progress in artificial intelligence is compared to that of other medical specialties, and barriers to artificial intelligence development and implementation in our specialty are discussed. The workgroup’s recommendations address stakeholders in policymaking, research, development, implementation, training, and use of artificial intelligence–based tools for perioperative care.

During the early stages of hospital admission, clinicians use limited information to make decisions as patient acuity evolves. We hypothesized that clustering analysis of vital signs measured within six hours of hospital admission would reveal distinct patient phenotypes with unique pathophysiological signatures and clinical outcomes. We created a longitudinal electronic health record dataset for 75,762 adult patient admissions to a tertiary care center in 2014–2016 lasting six hours or longer. Physiotypes were derived via unsupervised machine learning in a training cohort of 41,502 patients applying consensus k-means clustering to six vital signs measured within six hours of admission. Reproducibility and correlation with clinical biomarkers and outcomes were assessed in validation cohort of 17,415 patients and testing cohort of 16,845 patients. Training, validation, and testing cohorts had similar age (54–55 years) and sex (55% female), distributions. There were four distinct clusters. Physiotype A had physiologic signals consistent with early vasoplegia, hypothermia, and low-grade inflammation and favorable short-and long-term clinical outcomes despite early, severe illness. Physiotype B exhibited early tachycardia, tachypnea, and hypoxemia followed by the highest incidence of prolonged respiratory insufficiency, sepsis, acute kidney injury, and short- and long-term mortality. Physiotype C had minimal early physiological derangement and favorable clinical outcomes. Physiotype D had the greatest prevalence of chronic cardiovascular and kidney disease, presented with severely elevated blood pressure, and had good short-term outcomes but suffered increased 3-year mortality. Comparing sequential organ failure assessment (SOFA) scores across physiotypes demonstrated that clustering did not simply recapitulate previously established acuity assessments. In a heterogeneous cohort of hospitalized patients, unsupervised machine learning techniques applied to routine, early vital sign data identified physiotypes with unique disease categories and distinct clinical outcomes. This approach has the potential to augment understanding of pathophysiology by distilling thousands of disease states into a few physiological signatures.

Background: Long-term postoperative pain (POP) and patient responses to pain relief medications are not yet fully understood. Although recent studies have developed an index for the nociception level of patients under general anesthesia based on multiple physiological parameters, it remains unclear whether these parameters correlate with long-term POP outcomes. Objective: This study aims to extract unbiased and interpretable descriptions of how the dynamics of physiological parameters change over time and across patients in response to surgical procedures and intraoperative medications using a multivariate-temporal analysis. We demonstrated that there is an association (correlation) between the main features of intraoperative physiological responses and long-term POP, which has a predictive value, even without claiming causality. Methods: We proposed a complex higher-order singular value decomposition method to accurately decompose patients’ physiological responses into multivariate structures evolving over time. We used intraoperative vital signs of 175 patients from a mixed surgical cohort to extract three interconnected, low-dimensional, complex-valued descriptions of patients’ physiological responses: multivariate factors, reflecting subphysiological parameters; temporal factors, reflecting common intrasurgery temporal dynamics; and patients’ factors, describing interpatient changes in physiological responses. Results: Adoption of the complex higher-order singular value decomposition method allowed us to clarify the dynamic correlation structure included in the intraoperative physiological responses. Instantaneous phases of the complex-valued physiological responses of 242 patients within the subspace of principal descriptors enabled us to discriminate between mild and not-mild (moderate-severe) levels of pain at postoperative days 30 and 90. Following rotation of physiological responses before projection to align with the common multivariate-temporal dynamic, the method achieved an area under curve for postoperative day 30 and 90 outcomes of 0.81 and 0.89 for thoracic surgery, 0.87 and 0.83 for orthopedic surgery, 0.87 and 0.88 for urological surgery, 0.86 and 1 for colorectal surgery, 1 and 1 for transplant surgery, and 0.83 and 0.92 for pancreatic surgery, respectively. Conclusions: By categorizing patients into different surgical groups, we identified significant surgery-related principal descriptors. Each of them potentially encodes different surgical stimulation. The dynamics of patients’ physiological responses to these surgical events were linked to long-term POP development.

Mistrust is a major barrier to implementing deep learning in healthcare settings. Entrustment could be earned by conveying model certainty, or the probability that a given model output is accurate, but the use of uncertainty estimation for deep learning entrustment is largely unexplored, and there is no consensus regarding optimal methods for quantifying uncertainty. Our purpose is to critically evaluate methods for quantifying uncertainty in deep learning for healthcare applications and propose a conceptual framework for specifying certainty of deep learning predictions. We searched Embase, MEDLINE, and PubMed databases for articles relevant to study objectives, complying with PRISMA guidelines, rated study quality using validated tools, and extracted data according to modified CHARMS criteria. Among 30 included studies, 24 described medical imaging applications. All imaging model architectures used convolutional neural networks or a variation thereof. The predominant method for quantifying uncertainty was Monte Carlo dropout, producing predictions from multiple networks for which different neurons have dropped out and measuring variance across the distribution of resulting predictions. Conformal prediction offered similar strong performance in estimating uncertainty, along with ease of interpretation and application not only to deep learning but also to other machine learning approaches. Among the six articles describing non-imaging applications, model architectures and uncertainty estimation methods were heterogeneous, but predictive performance was generally strong, and uncertainty estimation was effective in comparing modeling methods. Overall, the use of model learning curves to quantify epistemic uncertainty (attributable to model parameters) was sparse. Heterogeneity in reporting methods precluded the performance of a meta-analysis. Uncertainty estimation methods have the potential to identify rare but important misclassifications made by deep learning models and compare modeling methods, which could build patient and clinician trust in deep learning applications in healthcare. Efficient maturation of this field will require standardized guidelines for reporting performance and uncertainty metrics.

Background: Gabapentinoids are increasingly prescribed to manage chronic noncancer pain (CNCP) in older adults. When used concurrently with opioids, gabapentinoids may potentiate central nervous system (CNS) depression and increase the risks for fall. We aimed to investigate whether concurrent use of gabapentinoids with opioids compared with use of opioids alone is associated with an increased risk of fall-related injury among older adults with CNCP.Methods and findings: We conducted a population-based cohort study using a 5% national sample of Medicare beneficiaries in the United States between 2011 and 2018. Study sample consisted of fee-for-service (FFS) beneficiaries aged ≥65 years with CNCP diagnosis who initiated opioids. We identified concurrent users with gabapentinoids and opioids days’ supply overlapping for ≥1 day and designated first day of concurrency as the index date. We created 2 cohorts based on whether concurrent users initiated gabapentinoids on the day of opioid initiation (Cohort 1) or after opioid initiation (Cohort 2). Each concurrent user was matched to up to 4 opioid-only users on opioid initiation date and index date using risk set sampling. We followed patients from index date to first fall-related injury event ascertained using a validated claims-based algorithm, treatment discontinuation or switching, death, Medicare disenrollment, hospitalization or nursing home admission, or end of study, whichever occurred first. In each cohort, we used propensity score (PS) weighted Cox models to estimate the adjusted hazard ratios (aHRs) with 95% confidence intervals (CIs) of fall-related injury, adjusting for year of the index date, sociodemographics, types of chronic pain, comorbidities, frailty, polypharmacy, healthcare utilization, use of nonopioid medications, and opioid use on and before the index date. We identified 6,733 concurrent users and 27,092 matched opioid-only users in Cohort 1 and 5,709 concurrent users and 22,388 matched opioid-only users in Cohort 2. The incidence rate of fall-related injury was 24.5 per 100 person-years during follow-up (median, 9 days; interquartile range [IQR], 5 to 18 days) in Cohort 1 and was 18.0 per 100 person-years during follow-up (median, 9 days; IQR, 4 to 22 days) in Cohort 2. Concurrent users had similar risk of fall-related injury as opioid-only users in Cohort 1(aHR = 0.97, 95% CI 0.71 to 1.34,p= 0.874), but had higher risk for fall-related injury than opioid-only users in Cohort 2 (aHR = 1.69, 95% CI 1.17 to 2.44,p= 0.005). Limitations of this study included confounding due to unmeasured factors, unavailable information on gabapentinoids’ indication, potential misclassification, and limited generalizability beyond older adults insured by Medicare FFS program.Conclusions: In this sample of older Medicare beneficiaries with CNCP, initiating gabapentinoids and opioids simultaneously compared with initiating opioids only was not significantly associated with risk for fall-related injury. However, addition of gabapentinoids to an existing opioid regimen was associated with increased risks for fall. Mechanisms for the observed excess risk, whether pharmacological or because of channeling of combination therapy to high-risk patients, require further investigation. Clinicians should consider the risk–benefit of combination therapy when prescribing gabapentinoids concurrently with opioids.

BACKGROUND Long-term pain conditions after surgery and patient responses to pain relief medications are not yet fully understood. Although recent studies have developed an index for the nociception level of patients under general anesthesia based on multiple physiological parameters, it remains unclear whether and how dynamic of these parameters indicate long-term postoperative pain. OBJECTIVE To extract unbiased and interpretable descriptions of how the dynamics of physiological parameters change over time and across patients in response to surgical procedures and intraoperative medications, we used a multivariate-temporal analysis. We demonstrate that the main features of intraoperative physiological responses can be used to predict long-term postoperative pain. METHODS We propose to use a complex higher-order singular-value decomposition (complex-HOSVD) method to accurately decompose patients’ physiological responses into multivariate structures evolving over time. We used intraoperative vital signs of 175 patients from a mixed surgical cohort to extract three interconnected, low-dimensional, complex-valued descriptions of patients’ physiological responses: multivariate factors, reflecting subphysiological parameters; temporal factors, reflecting common intrasurgery temporal dynamics; and patients’ factors, describing interpatient changes in physiological responses. RESULTS Adoption of the complex-HOSVD allowed us to clarify the dynamic correlation structure included in intraoperative physiological responses. Instantaneous phases of the complex-valued physiological responses within the subspace of principal descriptors enabled us to discriminate between “mild” and “severe” levels of pain at postoperative days 30 and 90. CONCLUSIONS By abstracting patients into different surgical groups, we identified significant surgery-related principal descriptors. Each of them potentially encodes different surgical stimulation. The dynamics of patients’ physiological responses to these surgical events were linked to long-term postoperative pain development.

Established guidelines describe minimum requirements for reporting algorithms in healthcare; it is equally important to objectify the characteristics of ideal algorithms that confer maximum potential benefits to patients, clinicians, and investigators. We propose a framework for ideal algorithms, including 6 desiderata: explainable (convey the relative importance of features in determining outputs), dynamic (capture temporal changes in physiologic signals and clinical events), precise (use high-resolution, multimodal data and aptly complex architecture), autonomous (learn with minimal supervision and execute without human input), fair (evaluate and mitigate implicit bias and social inequity), and reproducible (validated externally and prospectively and shared with academic communities). We present an ideal algorithms checklist and apply it to highly cited algorithms. Strategies and tools such as the predictive, descriptive, relevant (PDR) framework, the Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence (SPIRIT-AI) extension, sparse regression methods, and minimizing concept drift can help healthcare algorithms achieve these objectives, toward ideal algorithms in healthcare.