Deep Learning-Based Recurrent Delirium Prediction in Critically Ill Patients.

OBJECTIVES: To predict impending delirium in ICU patients using recurrent deep learning. DESIGN: Retrospective cohort study. SETTING: Fifteen medical-surgical ICUs across Alberta, Canada, between January 1, 2014, and January 24, 2020. PATIENTS: Forty-three thousand five hundred ten ICU admissions from 38,426 patients. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We used ICU and administrative health data to train deep learning models to predict delirium episodes in the next two 12-hour periods (0-12 and 12-24 hr), starting at 24 hours after ICU admission, and to generate new predictions every 12 hours. We used a comprehensive set of 3,643 features, capturing patient history, early ICU admission information (first 24 hr), and the temporal dynamics of various clinical variables throughout the ICU admission. Our deep learning architecture consisted of a feature embedding, a recurrent, and a prediction module. Our best model based on gated recurrent units yielded a sensitivity of 0.810, a specificity of 0.848, a precision (positive predictive value) of 0.704, and an area under the receiver operating characteristic curve (AUROC) of 0.909 in the hold-out test set for the 0-12-hour prediction horizon. For the 12-24-hour prediction horizon, the same model achieved a sensitivity of 0.791, a specificity of 0.807, a precision of 0.637, and an AUROC of 0.895 in the test set. CONCLUSIONS: Our delirium prediction model achieved strong performance by applying deep learning to a dataset that is at least one order of magnitude larger than those used in previous studies. Another novel aspect of our study is the temporal nature of our features and predictions. Our model enables accurate prediction of impending delirium in the ICU, which can potentially lead to early intervention, more efficient allocation of ICU resources, and improved patient outcomes.

Impact of Family Presence on Delirium in Critically Ill Patients: A Retrospective Cohort Study.

OBJECTIVE: To assess the effect of family presence on the prevalence and duration of delirium in adults admitted to an ICU. DESIGN: Retrospective cohort study. SETTING: Medical-surgical ICUs in Alberta, AB, Canada. PATIENTS: A population of 25,537 unique patients admitted at least once to an Alberta ICU. METHODS: We obtained electronic health records of consecutive adults (≥ 18 yr) admitted to one of 14 medical-surgical ICU in Alberta, Canada, from January 1, 2014, to December 30, 2018. Family presence was quantified using a validated algorithm and categorized as: 1) physical presence in ICU, 2) telephone call only, and 3) no presence (reference group). Delirium was measured using the Intensive Care Delirium Screening Checklist (ICDSC) and defined as an ICDSC greater than or equal to 4. Multivariable mixed-effects logistic and linear regression were used to evaluate the association between family presence and prevalence (binary) and duration (d) of delirium, respectively. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: The association between family presence and delirium prevalence differed according to admission type and admission Glasgow Coma Scale (GCS). Among medical and emergency surgical patients irrespective of admission GCS, physical presence of family was not significantly associated with the prevalence of delirium. In elective surgical patients, physical presence of family was associated with decreased prevalence of delirium in patients with intact Glasgow Coma Scale (GCS = 15; adjusted odds ratio, 0.60; 95% CI, 0.39-0.97; p = 0.02). Physical presence of family (adjusted mean difference [AMD] -1.87 d; 95% CI, -2.01 to -1.81; p < 0.001) and telephone calls (AMD -1.41 d; 95% CI, -1.52 to -1.31; p < 0.001) were associated with decreased duration of delirium in all patients. CONCLUSIONS: The effects of family presence on delirium are complex and dependent on type of visitation, reason for ICU admission, and brain function on ICU admission.

Surgery scheduling heuristic considering OR downstream and upstream facilities and resources.

BACKGROUND: Surgical theater (ST) operations planning is a key subject in the healthcare management literature, particularly the scheduling of procedures in operating rooms (ORs). The OR scheduling problem is usually approached using mathematical modeling and made available to ST managers through dedicated software. Regardless of the large body of knowledge on the subject, OR scheduling models rarely consider the integration of OR downstream and upstream facilities and resources or validate their propositions in real life, rather using simulated scenarios. We propose a heuristic to sequence surgeries that considers both upstream and downstream resources required to perform them, such as surgical kits, post anesthesia care unit (PACU) beds, and surgical teams (surgeons, nurses and anesthetists). METHODS: Using hybrid flow shop (HFS) techniques and the break-in-moment (BIM) concept, the goal is to find a sequence that maximizes the number of procedures assigned to the ORs while minimizing the variance of intervals between surgeries' completions, smoothing the demand for downstream resources such as PACU beds and OR sanitizing teams. There are five steps to the proposed heuristic: listing of priorities, local scheduling, global scheduling, feasibility check and identification of best scheduling. RESULTS: Our propositions were validated in a high complexity tertiary University hospital in two ways: first, applying the heuristic to historical data from five typical ST days and comparing the performance of our proposed sequences to the ones actually implemented; second, pilot testing the heuristic during ten days in the ORs, allowing a full rotation of surgical specialties. Results displayed an average increase of 37.2% in OR occupancy, allowing an average increase of 4.5 in the number of surgeries performed daily, and reducing the variance of intervals between surgeries' completions by 55.5%. A more uniform distribution of patients' arrivals at the PACU was also observed. CONCLUSIONS: Our proposed heuristic is particularly useful to plan the operation of STs in which resources are constrained, a situation that is common in hospital from developing countries. Our propositions were validated through a pilot implementation in a large hospital, contributing to the scarce literature on actual OR scheduling implementation.

Psychiatric Outcomes in ICU Patients With Family Visitation: A Population-Based Retrospective Cohort Study.

BACKGROUND: Lack of family visitation in the ICU can have long-term consequences on patients in the ICU after discharge. The effect of family visitation on the incidence of patient psychiatric disorders is unknown. RESEARCH QUESTION: What is the association between family visitation in the ICU and incidence of psychiatric outcomes in patients in the ICU 1 year after hospital discharge? STUDY DESIGN AND METHODS: This study assessed a population-based retrospective cohort of adult patients admitted to the ICU from January 1, 2014, through May 30, 2017, surviving to hospital discharge with ICU length of stay of ≥ 3 days. To be eligible, patients needed to have minimum of 5 years of administrative data before ICU admission and a minimum of 1 year of follow-up data after hospital discharge. An internally validated algorithm that interpreted natural language in health records determined patients with or without in-person family (ie, relatives, friends) visitation during ICU stay. The primary outcome was risk of an incidence of psychiatric disorder (composite outcome), including anxiety, depressive, trauma- and stressor-related, psychotic, and substance use disorders, identified using coding algorithms for administrative databases. Propensity scores were used in inverse probability weighted logistic regression models, and average treatment effects were converted to risk ratios (RRs) with 95% CIs. Secondary outcomes were incidences of diagnoses by type of psychiatric disorder. RESULTS: We included 14,344 patients with (96% [n = 13,771]) and without (4.0% [n = 573]) in-person family visitation who survived hospital discharge. More than one-third of patients received a diagnosis of any psychiatric disorder within 1 year after discharge (34.9%; 95% CI, 34.1%-35.6%). Patients most often received diagnoses of anxiety disorders (17.5%; 95% CI, 16.9%-18.1%) and depressive disorders (17.2%; 95% CI, 16.6%-17.9%). After inverse probability weighting of 13,731 patients, in-person family visitation was associated with a lower risk of received a diagnosis of any incident psychiatric disorder within 1 year after discharge (RR, 0.79; 95% CI, 0.68-0.92). INTERPRETATION: ICU family visitation is associated with a decreased risk of psychiatric disorders in critically ill patients up to 1 year after hospital discharge.

Natural language processing to measure the frequency and mode of communication between healthcare professionals and family members of critically ill patients.

OBJECTIVE: To apply natural language processing (NLP) techniques to identify individual events and modes of communication between healthcare professionals and families of critically ill patients from electronic medical records (EMR). MATERIALS AND METHODS: Retrospective cohort study of 280 randomly selected adult patients admitted to 1 of 15 intensive care units (ICU) in Alberta, Canada from June 19, 2012 to June 11, 2018. Individual events and modes of communication were independently abstracted using NLP and manual chart review (reference standard). Preprocessing techniques and 2 NLP approaches (rule-based and machine learning) were evaluated using sensitivity, specificity, and area under the receiver operating characteristic curves (AUROC). RESULTS: Over 2700 combinations of NLP methods and hyperparameters were evaluated for each mode of communication using a holdout subset. The rule-based approach had the highest AUROC in 65 datasets compared to the machine learning approach in 21 datasets. Both approaches had similar performance in 17 datasets. The rule-based AUROC for the grouped categories of patient documented to have family or friends (0.972, 95% CI 0.934-1.000), visit by family/friend (0.882 95% CI 0.820-0.943) and phone call with family/friend (0.975, 95% CI: 0.952-0.998) were high. DISCUSSION: We report an automated method to quantify communication between healthcare professionals and family members of adult patients from free-text EMRs. A rule-based NLP approach had better overall operating characteristics than a machine learning approach. CONCLUSION: NLP can automatically and accurately measure frequency and mode of documented family visitation and communication from unstructured free-text EMRs, to support patient- and family-centered care initiatives.

Digital public health surveillance: a systematic scoping review.

The ubiquitous and openly accessible information produced by the public on the Internet has sparked an increasing interest in developing digital public health surveillance (DPHS) systems. We conducted a systematic scoping review in accordance with the PRISMA extension for scoping reviews to consolidate and characterize the existing research on DPHS and identify areas for further research. We used Natural Language Processing and content analysis to define the search strings and searched Global Health, Web of Science, PubMed, and Google Scholar from 2005 to January 2020 for peer-reviewed articles on DPHS, with extensive hand searching. Seven hundred fifty-five articles were included in this review. The studies were from 54 countries and utilized 26 digital platforms to study 208 sub-categories of 49 categories associated with 16 public health surveillance (PHS) themes. Most studies were conducted by researchers from the United States (56%, 426) and dominated by communicable diseases-related topics (25%, 187), followed by behavioural risk factors (17%, 131). While this review discusses the potentials of using Internet-based data as an affordable and instantaneous resource for DPHS, it highlights the paucity of longitudinal studies and the methodological and inherent practical limitations underpinning the successful implementation of a DPHS system. Little work studied Internet users' demographics when developing DPHS systems, and 39% (291) of studies did not stratify their results by geographic region. A clear methodology by which the results of DPHS can be linked to public health action has yet to be established, as only six (0.8%) studies deployed their system into a PHS context.

Delirium prediction in the intensive care unit: a temporal approach.

The incidence of delirium in intensive care units is high and associated with poor outcomes; therefore, its prediction is desirable to establish preventive treatments. This retrospective study proposes a novel approach for delirium prediction. We analyzed static and temporal data from 10,475 patients admitted to one of 15 intensive care units (ICUs) in Alberta, Canada between January 1, 2014 and June 30, 2016. We tested 168 different combinations of study design parameters and five different predictive models (logistic regression, support vector machines, random forests, adaptive boosting and neural networks). The area under the receiver operating characteristic curve (AUROC) ranged from 0.754 (CI 95% ± 0.018) to 0.852 (± 0.033), with sensitivity and specificity respectively ranging from 0.739 (CI 95% ± 0.047) to 0.840 (CI 95% ± 0.064), and 0.770 (CI 95% ± 0.030) to 0.865 (CI 95% ± 0.038). These results are similar to previous studies; however, our approach allows for continuous updates and short-term prediction horizons which might provide major advantages.

Man vs. machine: Predicting hospital bed demand from an emergency department.

BACKGROUND: The recent literature reports promising results from using intelligent systems to support decision making in healthcare operations. Using these systems may lead to improved diagnostic and treatment protocols and to predict hospital bed demand. Predicting hospital bed demand in emergency department (ED) attendances could help resource allocation and reduce pressure on busy hospitals. However, there is still limited knowledge on whether intelligent systems can operate as fully autonomous, user-independent systems. OBJECTIVE: Compare the performance of a computer-based algorithm and humans in predicting hospital bed demand (admissions and discharges) based on the initial SOAP (Subjective, Objective, Assessment, Plan) records of the ED. METHODS: This was a retrospective cohort study that compared the performance of humans and machines in predicting hospital bed demand from an ED. It considered electronic medical records (EMR) of 9030 patients (230 used as a testing set, and hence evaluated both by humans and by an algorithm, and 8800 used as a training set exclusively by the algorithm) who visited the ED of a tertiary care and teaching public hospital located in Porto Alegre, Brazil between January and December 2014. The machine role was played by Support Vector Machine Classifier and the human prediction was performed by four ED physicians. Predictions were compared in terms of sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUROC). RESULTS: All graders achieved similar accuracies. The accuracy by AUROC for the testing set was 0.82 [95% confidence interval (CI) of 0.77-0.87], 0.80 (95% CI: 0.75-0.85), 0.76 (95% CI: 0.71-0.81) for novice physicians, machine, experienced physicians, respectively. Processing time per test EMR was 0.00812±0.0009 seconds. In contrast, novice physicians took on average 156.80 seconds per test EMR, while experienced physicians took on average 56.40 seconds per test EMR. CONCLUSIONS: Our data indicated that the system could predict patient admission or discharge states with 80% accuracy, which was similar the performance of novice and experienced physicians. These results suggested that the algorithm could operate as an autonomous and independent system to complete this task.

Text mining approach to predict hospital admissions using early medical records from the emergency department.

OBJECTIVE: Emergency department (ED) overcrowding is a serious issue for hospitals. Early information on short-term inward bed demand from patients receiving care at the ED may reduce the overcrowding problem, and optimize the use of hospital resources. In this study, we use text mining methods to process data from early ED patient records using the SOAP framework, and predict future hospitalizations and discharges. DESIGN: We try different approaches for pre-processing of text records and to predict hospitalization. Sets-of-words are obtained via binary representation, term frequency, and term frequency-inverse document frequency. Unigrams, bigrams and trigrams are tested for feature formation. Feature selection is based on χ2 and F-score metrics. In the prediction module, eight text mining methods are tested: Decision Tree, Random Forest, Extremely Randomized Tree, AdaBoost, Logistic Regression, Multinomial Naïve Bayes, Support Vector Machine (Kernel linear) and Nu-Support Vector Machine (Kernel linear). MEASUREMENTS: Prediction performance is evaluated by F1-scores. Precision and Recall values are also informed for all text mining methods tested. RESULTS: Nu-Support Vector Machine was the text mining method with the best overall performance. Its average F1-score in predicting hospitalization was 77.70%, with a standard deviation (SD) of 0.66%. CONCLUSIONS: The method could be used to manage daily routines in EDs such as capacity planning and resource allocation. Text mining could provide valuable information and facilitate decision-making by inward bed management teams.

Forecasting Daily Volume and Acuity of Patients in the Emergency Department.

This study aimed at analyzing the performance of four forecasting models in predicting the demand for medical care in terms of daily visits in an emergency department (ED) that handles high complexity cases, testing the influence of climatic and calendrical factors on demand behavior. We tested different mathematical models to forecast ED daily visits at Hospital de Clínicas de Porto Alegre (HCPA), which is a tertiary care teaching hospital located in Southern Brazil. Model accuracy was evaluated using mean absolute percentage error (MAPE), considering forecasting horizons of 1, 7, 14, 21, and 30 days. The demand time series was stratified according to patient classification using the Manchester Triage System's (MTS) criteria. Models tested were the simple seasonal exponential smoothing (SS), seasonal multiplicative Holt-Winters (SMHW), seasonal autoregressive integrated moving average (SARIMA), and multivariate autoregressive integrated moving average (MSARIMA). Performance of models varied according to patient classification, such that SS was the best choice when all types of patients were jointly considered, and SARIMA was the most accurate for modeling demands of very urgent (VU) and urgent (U) patients. The MSARIMA models taking into account climatic factors did not improve the performance of the SARIMA models, independent of patient classification.