Search

Abstract Background Electronic health records (EHR) contain large volumes of unstructured free-form text notes that richly describe a patient’s health and medical comorbidities. It is unclear if perioperative risk stratification can be performed directly from these notes without manual data extraction. We conduct a feasibility study using natural language processing (NLP) to predict the American Society of Anesthesiologists Physical Status Classification (ASA-PS) as a surrogate measure for perioperative risk. We explore prediction performance using four different model types and compare the use of different note sections versus the whole note. We use Shapley values to explain model predictions and analyze disagreement between model and human anesthesiologist predictions. Methods Single-center retrospective cohort analysis of EHR notes from patients undergoing procedures with anesthesia care spanning all procedural specialties during a 5 year period who were not assigned ASA VI and also had a preoperative evaluation note filed within 90 days prior to the procedure. NLP models were trained for each combination of 4 models and 8 text snippets from notes. Model performance was compared using area under the receiver operating characteristic curve (AUROC) and area under the precision recall curve (AUPRC). Shapley values were used to explain model predictions. Error analysis and model explanation using Shapley values was conducted for the best performing model. Results Final dataset includes 38,566 patients undergoing 61,503 procedures with anesthesia care. Prevalence of ASA-PS was 8.81% for ASA I, 31.4% for ASA II, 43.25% for ASA III, and 16.54% for ASA IV-V. The best performing models were the BioClinicalBERT model on the truncated note task (macro-average AUROC 0.845) and the fastText model on the full note task (macro-average AUROC 0.865). Shapley values reveal human-interpretable model predictions. Error analysis reveals that some original ASA-PS assignments may be incorrect and the model is making a reasonable prediction in these cases. Conclusions Text classification models can accurately predict a patient’s illness severity using only free-form text descriptions of patients without any manual data extraction. They can be an additional patient safety tool in the perioperative setting and reduce manual chart review for medical billing. Shapley feature attributions produce explanations that logically support model predictions and are understandable to clinicians.

Background and Objectives: We analyzed delirium testing, delirium prevalence, critical care associations outcomes at the time of hospital discharge in patients with acute brain injury (ABI) due to acute ischemic stroke (AIS), non-traumatic subarachnoid hemorrhage (SAH), non-traumatic intraparenchymal hemorrhage (IPH), and traumatic brain injury (TBI) admitted to an intensive care unit. Materials and Methods: We examined the frequency of assessment for delirium using the Confusion Assessment Method for the intensive care unit. We assessed delirium testing frequency, associated factors, positive test outcomes, and their correlations with clinical care, including nonpharmacological interventions and pain, agitation, and distress management. Results: Amongst 11,322 patients with ABI, delirium was tested in 8220 (726%). Compared to patients 18–44 years of age, patients 65–79 years (aOR 0.79 [0.69, 0.90]), and those 80 years and older (aOR 0.58 [0.50, 0.68]) were less likely to undergo delirium testing. Compared to English-speaking patients, non-English-speaking patients (aOR 0.73 [0.64, 0.84]) were less likely to undergo delirium testing. Amongst 8220, 2217 (27.2%) tested positive for delirium. For every day in the ICU, the odds of testing positive for delirium increased by 1.11 [0.10, 0.12]. Delirium was highest in those 80 years and older (aOR 3.18 [2.59, 3.90]). Delirium was associated with critical care resource utilization and with significant odds of mortality (aOR 7.26 [6.07, 8.70] at the time of hospital discharge. Conclusions: In conclusion, we find that seven out of ten patients in the neurocritical care unit are tested for delirium, and approximately two out of every five patients test positive for delirium. We demonstrate disparities in delirium testing by age and preferred language, identified high-risk subgroups, and the association between delirium, critical care resource use, complications, discharge GCS, and disposition. Prioritizing equitable testing and diagnosis, especially for elderly and non-English-speaking patients, is crucial for delivering quality care to this vulnerable group.

Background and objective: There is no report of the rate of opioid prescription at the time of hospital discharge, which may be associated with various patient and procedure-related factors. This study examined the prevalence and factors associated with prescribing opioids for head/neck pain after elective craniotomy for tumor resection/vascular repair. Methods: We performed a retrospective cohort study on adults undergoing elective craniotomy for tumor resection/vascular repair at a large quaternary-care hospital. We used univariable and multivariable analysis to examine the prevalence and factors (pre-operative, intraoperative, and postoperative) associated with prescribing opioids at the time of hospital discharge. We also examined the factors associated with discharge oral morphine equivalent use. Results: The study sample comprised 273 patients with a median age of 54 years [IQR 41,65], 173 females (63%), 174 (63.7%) tumor resections, and 99 (36.2%) vascular repairs. The majority (n = 264, 96.7%) received opioids postoperatively. The opiate prescription rates were 72% (n = 196/273) at hospital discharge, 23% (19/83) at neurosurgical clinical visits within 30 days of the procedure, and 2.4% (2/83) after 30 days from the procedure. The median oral morphine equivalent (OME) at discharge use was 300 [IQR 175,600]. Patients were discharged with a median supply of 5 days [IQR 3,7]. On multivariable analysis, opioid prescription at hospital discharge was associated with pre-existent chronic pain (adjusted odds ratio, aOR 1.87 [1.06,3.29], p = 0.03) and time from surgery to hospital discharge (compared to patients discharged within days 1–4 postoperatively, patients discharged between days 5–12 (aOR 0.3, 95% CI [0.15; 0.59], p = 0.0005), discharged at 12 days and later (aOR 0.17, 95% CI [0.07; 0.39], p < 0.001)). There was a linear relationship between the first 24 h OME (p < 0.001), daily OME (p < 0.001), hospital OME (p < 0.001), and discharge OME. Conclusions: This single-center study finds that at the time of hospital discharge, opioids are prescribed for head/neck pain in as many as seven out of ten patients after elective craniotomy. A history of chronic pain and time from surgery to discharge may be associated with opiate prescriptions. Discharge OME may be associated with first 24-h, daily OME, and hospital OME use. Findings need further evaluation in a large multicenter sample. The findings are important to consider as there is growing interest in an early discharge after elective craniotomy.

OBJECTIVES:. To evaluate factors predictive of clinical progression among coronavirus disease 2019 patients following admission, and whether continuous, automated assessments of patient status may contribute to optimal monitoring and management. DESIGN:. Retrospective cohort for algorithm training, testing, and validation. SETTING:. Eight hospitals across two geographically distinct regions. PATIENTS:. Two-thousand fifteen hospitalized coronavirus disease 2019–positive patients. INTERVENTIONS:. None. MEASUREMENTS AND MAIN RESULTS:. Anticipating Respiratory failure in Coronavirus disease (ARC), a clinically interpretable, continuously monitoring prognostic model of acute respiratory failure in hospitalized coronavirus disease 2019 patients, was developed and validated. An analysis of the most important clinical predictors aligns with key risk factors identified by other investigators but contributes new insights regarding the time at which key factors first begin to exhibit aberrency and distinguishes features predictive of acute respiratory failure in coronavirus disease 2019 versus pneumonia caused by other types of infection. Departing from prior work, ARC was designed to update continuously over time as new observations (vitals and laboratory test results) are recorded in the electronic health record. Validation against data from two geographically distinct health systems showed that the proposed model achieved 75% specificity and 77% sensitivity and predicted acute respiratory failure at a median time of 32 hours prior to onset. Over 80% of true-positive alerts occurred in non-ICU settings. CONCLUSIONS:. Patients admitted to non-ICU environments with coronavirus disease 2019 are at ongoing risk of clinical progression to severe disease, yet it is challenging to anticipate which patients will develop acute respiratory failure. A continuously monitoring prognostic model has potential to facilitate anticipatory rather than reactive approaches to escalation of care (e.g., earlier initiation of treatments for severe disease or structured monitoring and therapeutic interventions for high-risk patients).

Opioids play a critical role in acute postoperative pain management. Our objective was to develop machine learning models to predict postoperative opioid requirements in patients undergoing ambulatory surgery. To develop the models, we used a perioperative dataset of 13,700 patients (≥ 18 years) undergoing ambulatory surgery between the years 2016-2018. The data, comprising of patient, procedure and provider factors that could influence postoperative pain and opioid requirements, was randomly split into training (80%) and validation (20%) datasets. Machine learning models of different classes were developed to predict categorized levels of postoperative opioid requirements using the training dataset and then evaluated on the validation dataset. Prediction accuracy was used to differentiate model performances. The five types of models that were developed returned the following accuracies at two different stages of surgery: 1) Prior to surgery-Multinomial Logistic Regression: 71%, Naïve Bayes: 67%, Neural Network: 30%, Random Forest: 72%, Extreme Gradient Boost: 71% and 2) End of surgery-Multinomial Logistic Regression: 71%, Naïve Bayes: 63%, Neural Network: 32%, Random Forest: 72%, Extreme Gradient Boost: 70%. Analyzing the sensitivities of the best performing Random Forest model showed that the lower opioid requirements are predicted with better accuracy (89%) as compared with higher opioid requirements (43%). Feature importance (% relative importance) of model predictions showed that the type of procedure (15.4%), medical history (12.9%) and procedure duration (12.0%) were the top three features contributing to model predictions. Overall, the contribution of patient and procedure features towards model predictions were 65% and 35% respectively. Machine learning models could be used to predict postoperative opioid requirements in ambulatory surgery patients and could potentially assist in better management of their postoperative acute pain.