Accurate prediction of disease-risk factors from volumetric medical scans by a deep vision model pre-trained with 2D scans.

The application of machine learning to tasks involving volumetric biomedical imaging is constrained by the limited availability of annotated datasets of three-dimensional (3D) scans for model training. Here we report a deep-learning model pre-trained on 2D scans (for which annotated data are relatively abundant) that accurately predicts disease-risk factors from 3D medical-scan modalities. The model, which we named SLIViT (for 'slice integration by vision transformer'), preprocesses a given volumetric scan into 2D images, extracts their feature map and integrates it into a single prediction. We evaluated the model in eight different learning tasks, including classification and regression for six datasets involving four volumetric imaging modalities (computed tomography, magnetic resonance imaging, optical coherence tomography and ultrasound). SLIViT consistently outperformed domain-specific state-of-the-art models and was typically as accurate as clinical specialists who had spent considerable time manually annotating the analysed scans. Automating diagnosis tasks involving volumetric scans may save valuable clinician hours, reduce data acquisition costs and duration, and help expedite medical research and clinical applications.

A review of machine learning methods for non-invasive blood pressure estimation.

Blood pressure is a very important clinical measurement, offering valuable insights into the hemodynamic status of patients. Regular monitoring is crucial for early detection, prevention, and treatment of conditions like hypotension and hypertension, both of which increasing morbidity for a wide variety of reasons. This monitoring can be done either invasively or non-invasively and intermittently vs. continuously. An invasive method is considered the gold standard and provides continuous measurement, but it carries higher risks of complications such as infection, bleeding, and thrombosis. Non-invasive techniques, in contrast, reduce these risks and can provide intermittent or continuous blood pressure readings. This review explores modern machine learning-based non-invasive methods for blood pressure estimation, discussing their advantages, limitations, and clinical relevance.

Assisted Fluid Management and Sublingual Microvascular Flow During High-Risk Abdominal Surgery: A Randomized Controlled Trial.

BACKGROUND: Implementation of goal-directed fluid therapy (GDFT) protocols remains low. Protocol compliance among anesthesiologists tends to be suboptimal owing to the high workload and the attention required for implementation. The assisted fluid management (AFM) system is a novel decision support tool designed to help clinicians apply GDFT protocols. This system predicts fluid responsiveness better than anesthesia practitioners do and achieves higher stroke volume (SV) and cardiac index values during surgery. We tested the hypothesis that an AFM-guided GDFT strategy would also be associated with better sublingual microvascular flow compared to a standard GDFT strategy. METHODS: This bicenter, parallel, 2-arm, prospective, randomized controlled, patient and assessor-blinded, superiority study considered for inclusion all consecutive patients undergoing high-risk abdominal surgery who required an arterial catheter and uncalibrated SV monitoring. Patients having standard GDFT received manual titration of fluid challenges to optimize SV while patients having an AFM-guided GDFT strategy received fluid challenges based on recommendations from the AFM software. In all patients, fluid challenges were standardized and titrated per 250 mL and vasopressors were administered to maintain a mean arterial pressure >70 mm Hg. The primary outcome (average of each patient's intraoperative microvascular flow index (MFI) across 4 intraoperative time points) was analyzed using a Mann-Whitney U test and the treatment effect was estimated with a median difference between groups with a 95% confidence interval estimated using the bootstrap percentile method (with 1000 replications). Secondary outcomes included SV, cardiac index, total amount of fluid, other microcirculatory variables, and postoperative lactate. RESULTS: A total of 86 patients were enrolled over a 7-month period. The primary outcome was significantly higher in patients with AFM (median [Q1-Q3]: 2.89 [2.84-2.94]) versus those having standard GDFT (2.59 [2.38-2.78] points, median difference 0.30; 95% confidence interval [CI], 0.19-0.49; P < .001). Cardiac index and SVI were higher (3.2 ± 0.5 vs 2.7 ± 0.7 l.min-1.m-2; P = .001 and 42 [35-47] vs 36 [32-43] mL.m-2; P = .018) and arterial lactate concentration was lower at the end of the surgery in patients having AFM-guided GDFT (2.1 [1.5-3.1] vs 2.9 [2.1-3.9] mmol.L-1; P = .026) than patients having standard GDFT strategy. Patients having AFM received a higher fluid volume but 3 times less norepinephrine than those receiving standard GDFT (P < .001). CONCLUSIONS: Use of an AFM-guided GDFT strategy resulted in higher sublingual microvascular flow during surgery compared to use of a standard GDFT strategy. Future trials are necessary to make conclusive recommendations that will change clinical practice.

Restrictive versus Decision Support Guided Fluid Therapy during Major Hepatic Resection Surgery: A Randomized Controlled Trial.

BACKGROUND: Fluid therapy during major hepatic resection aims at minimizing fluids during the dissection phase to reduce central venous pressure, retrograde liver blood flow, and venous bleeding. This strategy, however, may lead to hyperlactatemia. The Acumen assisted fluid management system uses novel decision support software, the algorithm of which helps clinicians optimize fluid therapy. The study tested the hypothesis that using this decision support system could decrease arterial lactate at the end of major hepatic resection when compared to a more restrictive fluid strategy. METHODS: This two-arm, prospective, randomized controlled, assessor- and patient-blinded superiority study included consecutive patients undergoing major liver surgery equipped with an arterial catheter linked to an uncalibrated stroke volume monitor. In the decision support group, fluid therapy was guided throughout the entire procedure using the assisted fluid management software. In the restrictive fluid group, clinicians were recommended to restrict fluid infusion to 1 to 2 ml · kg-1 · h-1 until the completion of hepatectomy. They then administered fluids based on advanced hemodynamic variables. Noradrenaline was titrated in all patients to maintain a mean arterial pressure greater than 65 mmHg. The primary outcome was arterial lactate level upon completion of surgery (i.e., skin closure). RESULTS: A total of 90 patients were enrolled over a 7-month period. The primary outcome was lower in the decision support group than in the restrictive group (median [quartile 1 to quartile 3], 2.5 [1.9 to 3.7] mmol · l-1vs. 4.6 [3.1 to 5.4] mmol · l-1; median difference, -2.1; 95% CI, -2.7 to -1.2; P < 0.001). Among secondary exploratory outcomes, there was no difference in blood loss (median [quartile 1 to quartile 3], 450 [300 to 600] ml vs. 500 [300 to 800] ml; P = 0.727), although central venous pressure was higher in the decision support group (mean ± SD of 7.7 ± 2.0 mmHg vs. 6.6 ± 1.1 mmHg; P < 0.002). CONCLUSIONS: Patients managed using a clinical decision support system to guide fluid administration during major hepatic resection had a lower arterial lactate concentration at the end of surgery when compared to a more restrictive fluid strategy. Future trials are necessary to make conclusive recommendations that will change clinical practice.

Closing the Chasm: Understanding and Addressing the Anesthesia Workforce Supply and Demand Imbalance.

The imbalance in anesthesia workforce supply and demand has been exacerbated post-COVID due to a surge in demand for anesthesia care, especially in non-operating room anesthetizing sites, at a faster rate than the increase in anesthesia clinicians. The consequences of this imbalance or labor shortage compromise healthcare facilities, adversely affect the cost of care, worsen anesthesia workforce burnout, disrupt procedural and surgical schedules, and threaten academic missions and the ability to educate future anesthesiologists. In developing possible solutions, one must examine emerging trends that are affecting the anesthesia workforce, new technologies that will transform anesthesia care and the workforce, and financial considerations, including governmental payment policies. Possible practice solutions to this imbalance will require both short- and long-term multifactorial approaches that include increasing training positions and retention policies, improving capacity through innovations, leveraging technology, and addressing financial constraints.

An algorithm to detect dicrotic notch in arterial blood pressure and photoplethysmography waveforms using the iterative envelope mean method.

BACKGROUND AND OBJECTIVE: Detection of the dicrotic notch (DN) within a cardiac cycle is essential for assessment of cardiac output, calculation of pulse wave velocity, estimation of left ventricular ejection time, and supporting feature-based machine learning models for noninvasive blood pressure estimation, and hypotension, or hypertension prediction. In this study, we present a new algorithm based on the iterative envelope mean (IEM) method to detect automatically the DN in arterial blood pressure (ABP) and photoplethysmography (PPG) waveforms. METHODS: The algorithm was evaluated on both ABP and PPG waveforms from a large perioperative dataset (MLORD dataset) comprising 17,327 patients. The analysis involved a total of 1,171,288 cardiac cycles for ABP waveforms and 3,424,975 cardiac cycles for PPG waveforms. To evaluate the algorithm's performance, the systolic phase duration (SPD) was employed, which represents the duration from the onset of the systolic phase to the DN in the cardiac cycle. Correlation plots and regression analysis were used to compare the algorithm against marked DN detection, while box plots and Bland-Altman plots were used to compare its performance with both marked DN detection and an established DN detection technique (second derivative). The marking of the DN temporal location was carried out by an experienced researcher using the help of the 'find_peaks' function from the scipy Python package, serving as a reference for the evaluation. The marking was visually validated by both an engineer and an anesthesiologist. The robustness of the algorithm was evaluated as the DN was made less visually distinct across signal-to-noise ratios (SNRs) ranging from -30 dB to -5 dB in both ABP and PPG waveforms. RESULTS: The correlation between SPD estimated by the algorithm and that marked by the researcher is strong for both ABP (R2(87,343) =0.99, p<.001) and PPG (R2(86,764) =0.98, p<.001) waveforms. The algorithm had a lower mean error of DN detection (s): 0.0047 (0.0029) for ABP waveforms and 0.0046 (0.0029) for PPG waveforms, compared to 0.0693 (0.0770) for ABP and 0.0968 (0.0909) for PPG waveforms for the established 2nd derivative method. The algorithm has high rate of detectability of DN detection for SNR of >= -9 dB for ABP waveforms and >= -12 dB for PPG waveforms indicating robust performance in detecting the DN when it is less visibly distinct. CONCLUSION: Our proposed IEM- based algorithm can detect DN in both ABP and PPG waveforms with low computational cost, even in cases where it is not distinctly defined within a cardiac cycle of the waveform ('DN-less signals'). The algorithm can potentially serve as a valuable, fast, and reliable tool for extracting features from ABP and PPG waveforms. It can be especially beneficial in medical applications where DN-based features, such as SPD, diastolic phase duration, and DN amplitude, play a significant role.

Comparison of Differences in Cohort (Forward) and Case Control (Backward) Methodologic Approaches for Validation of the Hypotension Prediction Index.

BACKGROUND: The Hypotension Prediction Index (the index) software is a machine learning algorithm that detects physiologic changes that may lead to hypotension. The original validation used a case control (backward) analysis that has been suggested to be biased. This study therefore conducted a cohort (forward) analysis and compared this to the original validation technique. METHODS: A retrospective analysis of data from previously reported studies was conducted. All data were analyzed identically with two different methodologies, and receiver operating characteristic curves were constructed. Both backward and forward analyses were performed to examine differences in area under the receiver operating characteristic curves for the Hypotension Prediction Index and other hemodynamic variables to predict a mean arterial pressure (MAP) less than 65 mmHg for at least 1 min 5, 10, and 15 min in advance. RESULTS: The analysis included 2,022 patients, yielding 4,152,124 measurements taken at 20-s intervals. The area under the curve for the index predicting hypotension analyzed by backward and forward methodologies respectively was 0.957 (95% CI, 0.947 to 0.964) versus 0.923 (95% CI, 0.912 to 0.933) 5 min in advance, 0.933 (95% CI, 0.924 to 0.942) versus 0.923 (95% CI, 0.911 to 0.933) 10 min in advance, and 0.929 (95% CI, 0.918 to 0.938) versus 0.926 (95% CI, 0.914 to 0.937) 15 min in advance. No variable other than MAP had an area under the curve greater than 0.7. The areas under the curve using forward analysis for MAP predicting hypotension 5, 10, and 15 min in advance were 0.932 (95% CI, 0.920 to 0.940), 0.929 (95% CI, 0.918 to 0.938), and 0.932 (95% CI, 0.921 to 0.940), respectively. The R2 for the variation in the index due to MAP was 0.77. CONCLUSIONS: Using an updated methodology, the study found that the utility of the Hypotension Prediction Index to predict future hypotensive events is high, with an area under the receiver operating characteristics curve similar to that of the original validation method.

Clinical Knowledge and Reasoning Abilities of AI Large Language Models in Anesthesiology: A Comparative Study on the American Board of Anesthesiology Examination.

BACKGROUND: Over the past decade, artificial intelligence (AI) has expanded significantly with increased adoption across various industries, including medicine. Recently, AI-based large language models such as Generative Pretrained Transformer-3 (GPT-3), Bard, and Generative Pretrained Transformer-3 (GPT-4) have demonstrated remarkable language capabilities. While previous studies have explored their potential in general medical knowledge tasks, here we assess their clinical knowledge and reasoning abilities in a specialized medical context. METHODS: We studied and compared the performance of all 3 models on both the written and oral portions of the comprehensive and challenging American Board of Anesthesiology (ABA) examination, which evaluates candidates' knowledge and competence in anesthesia practice. RESULTS: Our results reveal that only GPT-4 successfully passed the written examination, achieving an accuracy of 78% on the basic section and 80% on the advanced section. In comparison, the less recent or smaller GPT-3 and Bard models scored 58% and 47% on the basic examination, and 50% and 46% on the advanced examination, respectively. Consequently, only GPT-4 was evaluated in the oral examination, with examiners concluding that it had a reasonable possibility of passing the structured oral examination. Additionally, we observe that these models exhibit varying degrees of proficiency across distinct topics, which could serve as an indicator of the relative quality of information contained in the corresponding training datasets. This may also act as a predictor for determining which anesthesiology subspecialty is most likely to witness the earliest integration with AI. CONCLUSIONS: GPT-4 outperformed GPT-3 and Bard on both basic and advanced sections of the written ABA examination, and actual board examiners considered GPT-4 to have a reasonable possibility of passing the real oral examination; these models also exhibit varying degrees of proficiency across distinct topics.

An algorithm to detect dicrotic notch in arterial blood pressure and photoplethysmography waveforms using the iterative envelope mean method.

Background and Objective: Detection of the dicrotic notch (DN) within a cardiac cycle is essential for assessment of cardiac output, calculation of pulse wave velocity, estimation of left ventricular ejection time, and supporting feature-based machine learning models for noninvasive blood pressure estimation, and hypotension, or hypertension prediction. In this study, we present a new algorithm based on the iterative envelope mean (IEM) method to detect automatically the DN in arterial blood pressure (ABP) and photoplethysmography (PPG) waveforms. Methods: The algorithm was evaluated on both ABP and PPG waveforms from a large perioperative dataset (MLORD dataset) comprising 17,327 patients. The analysis involved a total of 1,171,288 cardiac cycles for ABP waveforms and 3,424,975 cardiac cycles for PPG waveforms. To evaluate the algorithm's performance, the systolic phase duration (SPD) was employed, which represents the duration from the onset of the systolic phase to the DN in the cardiac cycle. Correlation plots and regression analysis were used to compare the algorithm with an established DN detection technique (second derivative). The marking of the DN temporal location was carried out by an experienced researcher using the help of the 'find_peaks' function from the scipy PYTHON package, serving as a reference for the evaluation. The marking was visually validated by both an engineer and an anesthesiologist. The robustness of the algorithm was evaluated as the DN was made less visually distinct across signal-to-noise ratios (SNRs) ranging from -30 dB to -5 dB in both ABP and PPG waveforms. Results: The correlation between SPD estimated by the algorithm and that marked by the researcher is strong for both ABP (R2(87343) =.99, p<.001) and PPG (R2(86764) =.98, p<.001) waveforms. The algorithm had a lower mean error of dicrotic notch detection (s): 0.0047 (0.0029) for ABP waveforms and 0.0046 (0.0029) for PPG waveforms, compared to 0.0693 (0.0770) for ABP and 0.0968 (0.0909) for PPG waveforms for the established 2nd derivative method. The algorithm has high accuracy of DN detection for SNR of >= -9 dB for ABP waveforms and >= -12 dB for PPG waveforms indicating robust performance in detecting the DN when it is less visibly distinct. Conclusion: Our proposed IEM- based algorithm can detect DN in both ABP and PPG waveforms with low computational cost, even in cases where it is not distinctly defined within a cardiac cycle of the waveform ('DN-less signals'). The algorithm can potentially serve as a valuable, fast, and reliable tool for extracting features from ABP and PPG waveforms. It can be especially beneficial in medical applications where DN-based features, such as SPD, diastolic phase duration, and DN amplitude, play a significant role.

Differences in the Receipt of Regional Anesthesia Based on Race and Ethnicity in Colorectal Surgery.

BACKGROUND: Health equity in pain management during the perioperative period continues to be a topic of interest. The authors evaluated the association of race and ethnicity with regional anesthesia in patients who underwent colorectal surgery and characterized trends in regional anesthesia. METHODS: Using the American College of Surgeons National Surgical Quality Improvement Program database from 2015 to 2020, the research team identified patients who underwent open or laparoscopic colorectal surgery. Associations between race and ethnicity and use of regional anesthesia were estimated using logistic regression models. RESULTS: The final sample size was 292,797, of which 15.6% (n = 45,784) received regional anesthesia. The unadjusted rates of regional anesthesia for race and ethnicity were 15.7% white, 15.1% Black, 12.8% Asian, 29.6% American Indian or Alaska Native, 16.3% Native Hawaiian or Pacific Islander, and 12.4% Hispanic. Black (odds ratio [OR] 0.93, 95% confidence interval [CI] 0.90-0.96, p < 0.001) and Asian (OR 0.76, 95% CI 0.71-0.80, p < 0.001) patients had lower odds of regional anesthesia compared to white patients. Hispanic patients had lower odds of regional anesthesia compared to non-Hispanic patients (OR 0.72, 95% CI 0.68-0.75, p < 0.001). There was a significant annual increase in regional anesthesia from 2015 to 2020 for all racial and ethnic cohorts (p < 0.05). CONCLUSION: There was an annual increase in the use of regional anesthesia, yet Black and Asian patients (compared to whites) and Hispanics (compared to non-Hispanics) were less likely to receive regional anesthesia for colorectal surgery. These differences suggest that there are racial and ethnic differences in regional anesthesia use for colorectal surgery.

Control of mean arterial pressure using a closed-loop system for norepinephrine infusion in severe brain injury patients: the COMAT randomized controlled trial.

Brain injury patients require precise blood pressure (BP) management to maintain cerebral perfusion pressure (CPP) and avoid intracranial hypertension. Nurses have many tasks and norepinephrine titration has been shown to be suboptimal. This can lead to limited BP control in patients that are in critical need of cerebral perfusion optimization. We have designed a closed-loop vasopressor (CLV) system capable of maintaining mean arterial pressure (MAP) in a narrow range and we aimed to assess its performance when treating severe brain injury patients. Within the first 48 h of intensive care unit (ICU) admission, 18 patients with a severe brain injury underwent either CLV or manual norepinephrine titration. In both groups, the objective was to maintain MAP in target (within ± 5 mmHg of a predefined target MAP) to achieve optimal CPP. Fluid administration was standardized in the two groups. The primary objective was the percentage of time patients were in target. Secondary outcomes included time spent over and under target. Over the four-hour study period, the mean percentage of time with MAP in target was greater in the CLV group than in the control group (95.8 ± 2.2% vs. 42.5 ± 27.0%, p < 0.001). Severe undershooting, defined as MAP < 10 mmHg of target value was lower in the CLV group (0.2 ± 0.3% vs. 7.4 ± 14.2%, p < 0.001) as was severe overshooting defined as MAP > 10 mmHg of target (0.0 ± 0.0% vs. 22.0 ± 29.0%, p < 0.001). The CLV system can maintain MAP in target better than nurses caring for severe brain injury patients.

A Prospective Observational Cohort Study of Language Preference and Preoperative Cognitive Screening in Older Adults: Do Language Disparities Exist in Cognitive Screening and Does the Association Between Test Results and Postoperative Delirium Differ Based on Language Preference?

BACKGROUND: A greater percentage of surgical procedures are being performed each year on patients 65 years of age or older. Concurrently, a growing proportion of patients in English-speaking countries such as the United States, United Kingdom, Australia, and Canada have a language other than English (LOE) preference. We aimed to measure whether patients with LOE underwent cognitive screening at the same rates as their English-speaking counterparts when routine screening was instituted. We also aimed to measure the association between preoperative Mini-Cog and postoperative delirium (POD) in both English-speaking and LOE patients. METHODS: We conducted a single-center, observational cohort study in patients 65 years old or older, scheduled for surgery and evaluated in the preoperative clinic. Cognitive screening of older adults was recommended as an institutional program for all patients 65 and older presenting to the preoperative clinic. We measured program adherence for cognitive screening. We also assessed the association of preoperative impairment on Mini-Cog and POD in both English-speaking and LOE patients, and whether the association differed for the 2 groups. A Mini-Cog score ≤2 was considered impaired. Postoperatively, patients were assessed for POD using the Confusion Assessment Method (CAM) and by systematic chart review. RESULTS: Over a 3-year period (February 2019-January 2022), 2446 patients 65 years old or older were assessed in the preoperative clinic prior. Of those 1956 patients underwent cognitive screening. Eighty-nine percent of English-speaking patients underwent preoperative cognitive screening, compared to 58% of LOE patients. The odds of having a Mini-Cog assessment were 5.6 times higher (95% confidence interval [CI], 4.6-7.0) P < .001 for English-speaking patients compared to LOE patients. In English-speaking patients with a positive Mini-Cog screen, the odds of having postop delirium were 3.5 times higher (95% CI, 2.6-4.8) P < .001 when compared to negative Mini-Cog. In LOE patients, the odds of having postop delirium were 3.9 times higher (95% CI, 2.1-7.3) P < .001 for those with a positive Mini-Cog compared to a negative Mini-Cog. The difference between these 2 odds ratios was not significant ( P = .753). CONCLUSIONS: We observed a disparity in the rates LOE patients were cognitively screened before surgery, despite the Mini-Cog being associated with POD in both English-speaking and LOE patients. Efforts should be made to identify barriers to cognitive screening in limited English-proficient older adults.

Closing the loop: automation in anesthesiology is coming.

Anesthesiology and intensive care medicine provide fertile ground for innovation in automation, but to date we have only achieved preliminary studies in closed-loop intravenous drug administration. Anesthesiologists have yet to implement these tools on a large scale despite clear evidence that they outperform manual titration. Closed-loops continuously assess a predefined variable as input into a controller and then attempt to establish equilibrium by administering a treatment as output. The aim is to decrease the error between the closed-loop controller's input and output. In this editorial we consider the available intravenous anesthesia closed-loop systems, try to clarify why they have not yet been implemented on a large scale, see what they offer, and propose the future steps towards automation in anesthesia.

Tight control of mean arterial pressure using a closed loop system for norepinephrine infusion after high-risk abdominal surgery: a randomized controlled trial.

Intensive care unit (ICU) nurses frequently manually titrate norepinephrine to maintain a predefined mean arterial pressure (MAP) target after high-risk surgery. However, achieving this task is often suboptimal. We have developed a closed-loop vasopressor (CLV) controller to better maintain MAP within a narrow range. After ethical committee approval, fifty-three patients admitted to the ICU following high-risk abdominal surgery were randomized to CLV or manual norepinephrine titration. In both groups, the aim was to maintain MAP in the predefined target of 80-90 mmHg. Fluid administration was standardized in the two groups using an advanced hemodynamic monitoring device. The primary outcome of our study was the percentage of time patients were in the MAP target. Over the 2-hour study period, the percentage of time with MAP in target was greater in the CLV group than in the control group (median: IQR25-75: 80 [68-88]% vs. 42 [22-65]%), difference 37.2, 95% CI (23.0-49.2); p < 0.001). Percentage time with MAP under 80 mmHg (1 [0-5]% vs. 26 [16-75]%, p < 0.001) and MAP under 65 mmHg (0 [0-0]% vs. 0 [0-4]%, p = 0.017) were both lower in the CLV group than in the control group. The percentage of time with a MAP > 90 mmHg was not statistically different between groups. In patients admitted to the ICU after high-risk abdominal surgery, closed-loop control of norepinephrine infusion better maintained a MAP target of 80 to 90 mmHg and significantly decreased postoperative hypotensive when compared to manual norepinephrine titration.

Novel Preoperative Risk Stratification Using Digital Phenotyping Applying a Scalable Machine-Learning Approach.

BACKGROUND: Classification of perioperative risk is important for patient care, resource allocation, and guiding shared decision-making. Using discriminative features from the electronic health record (EHR), machine-learning algorithms can create digital phenotypes among heterogenous populations, representing distinct patient subpopulations grouped by shared characteristics, from which we can personalize care, anticipate clinical care trajectories, and explore therapies. We hypothesized that digital phenotypes in preoperative settings are associated with postoperative adverse events including in-hospital and 30-day mortality, 30-day surgical redo, intensive care unit (ICU) admission, and hospital length of stay (LOS). METHODS: We identified all laminectomies, colectomies, and thoracic surgeries performed over a 9-year period from a large hospital system. Seventy-seven readily extractable preoperative features were first selected from clinical consensus, including demographics, medical history, and lab results. Three surgery-specific datasets were built and split into derivation and validation cohorts using chronological occurrence. Consensus k -means clustering was performed independently on each derivation cohort, from which phenotypes' characteristics were explored. Cluster assignments were used to train a random forest model to assign patient phenotypes in validation cohorts. We reconducted descriptive analyses on validation cohorts to confirm the similarity of patient characteristics with derivation cohorts, and quantified the association of each phenotype with postoperative adverse events by using the area under receiver operating characteristic curve (AUROC). We compared our approach to American Society of Anesthesiologists (ASA) alone and investigated a combination of our phenotypes with the ASA score. RESULTS: A total of 7251 patients met inclusion criteria, of which 2770 were held out in a validation dataset based on chronological occurrence. Using segmentation metrics and clinical consensus, 3 distinct phenotypes were created for each surgery. The main features used for segmentation included urgency of the procedure, preoperative LOS, age, and comorbidities. The most relevant characteristics varied for each of the 3 surgeries. Low-risk phenotype alpha was the most common (2039 of 2770, 74%), while high-risk phenotype gamma was the rarest (302 of 2770, 11%). Adverse outcomes progressively increased from phenotypes alpha to gamma, including 30-day mortality (0.3%, 2.1%, and 6.0%, respectively), in-hospital mortality (0.2%, 2.3%, and 7.3%), and prolonged hospital LOS (3.4%, 22.1%, and 25.8%). When combined with the ASA score, digital phenotypes achieved higher AUROC than the ASA score alone (hospital mortality: 0.91 vs 0.84; prolonged hospitalization: 0.80 vs 0.71). CONCLUSIONS: For 3 frequently performed surgeries, we identified 3 digital phenotypes. The typical profiles of each phenotype were described and could be used to anticipate adverse postoperative events.

Medical Informatics Operating Room Vitals and Events Repository (MOVER): a public-access operating room database.

Objectives: Artificial intelligence (AI) holds great promise for transforming the healthcare industry. However, despite its potential, AI is yet to see widespread deployment in clinical settings in significant part due to the lack of publicly available clinical data and the lack of transparency in the published AI algorithms. There are few clinical data repositories publicly accessible to researchers to train and test AI algorithms, and even fewer that contain specialized data from the perioperative setting. To address this gap, we present and release the Medical Informatics Operating Room Vitals and Events Repository (MOVER). Materials and Methods: This first release of MOVER includes adult patients who underwent surgery at the University of California, Irvine Medical Center from 2015 to 2022. Data for patients who underwent surgery were captured from 2 different sources: High-fidelity physiological waveforms from all of the operating rooms were captured in real time and matched with electronic medical record data. Results: MOVER includes data from 58 799 unique patients and 83 468 surgeries. MOVER is available for download at https://doi.org/10.24432/C5VS5G, it can be downloaded by anyone who signs a data usage agreement (DUA), to restrict traffic to legitimate researchers. Discussion: To the best of our knowledge MOVER is the only freely available public data repository that contains electronic health record and high-fidelity physiological waveforms data for patients undergoing surgery. Conclusion: MOVER is freely available to all researchers who sign a DUA, and we hope that it will accelerate the integration of AI into healthcare settings, ultimately leading to improved patient outcomes.