Real-time Clinical Decision Support Research Articles

Introducing surgical intelligence in gynecology: Automated identification of key steps in hysterectomy.

The analysis of surgical videos using artificial intelligence holds great promise for the future of surgery by facilitating the development of surgical best practices, identifying key pitfalls, enhancing situational awareness, and disseminating that information via real-time, intraoperative decision-making. The objective of the present study was to examine the feasibility and accuracy of a novel computer vision algorithm for hysterectomy surgical step identification. This was a retrospective study conducted on surgical videos of laparoscopic hysterectomies performed in 277 patients in five medical centers. We used a surgical intelligence platform (Theator Inc.) that employs advanced computer vision and AI technology to automatically capture video data during surgery, deidentify, and upload procedures to a secure cloud infrastructure. Videos were manually annotated with sequential steps of surgery by a team of annotation specialists. Subsequently, a computer vision system was trained to perform automated step detection in hysterectomy. Analyzing automated video annotations in comparison to manual human annotations was used to determine accuracy. The mean duration of the videos was 103 ± 43 min. Accuracy between AI-based predictions and manual human annotations was 93.1% on average. Accuracy was highest for the dissection and mobilization step (96.9%) and lowest for the adhesiolysis step (70.3%). The results of the present study demonstrate that a novel AI-based model achieves high accuracy for automated steps identification in hysterectomy. This lays the foundations for the next phase of AI, focused on real-time clinical decision support and prediction of outcome measures, to optimize surgeon workflow and elevate patient care.

Digital Twins for More Precise and Personalized Treatment.

The use of Digital Twins (DTs) or the digital replicas of physical entities has provided benefits to several industry sectors, most notably manufacturing. To date, the application of DTs in the healthcare sector has been minimal, however. But, as pressure increases for more precise and personalized treatments, it behooves us to investigate the potential for DTs in the healthcare context. As a proof-of-concept demonstration prior to working with real patients, we attempt in this paper, to explore the potential for creating and using DTs. We do this in a synthetic environment at this stage, making use of data that is all computer-generated. DTs of synthetic present patients are created making use of data of synthetic past patients. In the real world, the clinical objective for creating such DTs of real patients would be to enable enhanced real-time clinical decision support to enable more precise and personalized care. The objective of the numerical experiment reported in this paper, is to envisage the possibilities and challenges of such an approach. We attempt to better understand the strengths and weaknesses of applying DTs in the healthcare context to support more precise and personalized treatments.

ICU Utilization After Implementation of Minor Severe Pneumonia Criteria in Real-Time Electronic Clinical Decision Support.

To determine if the implementation of automated clinical decision support (CDS) with embedded minor severe community-acquired pneumonia (sCAP) criteria was associated with improved ICU utilization among emergency department (ED) patients with pneumonia who did not require vasopressors or positive pressure ventilation at admission. Planned secondary analysis of a stepped-wedge, cluster-controlled CDS implementation trial. Sixteen hospitals in six geographic clusters from Intermountain Health; a large, integrated, nonprofit health system in Utah and Idaho. Adults admitted to the hospital from the ED with pneumonia identified by: 1) discharge International Classification of Diseases , 10th Revision codes for pneumonia or sepsis/respiratory failure and 2) ED chest imaging consistent with pneumonia, who did not require vasopressors or positive pressure ventilation at admission. After implementation, patients were exposed to automated, open-loop, comprehensive CDS that aided disposition decision (ward vs. ICU), based on objective severity scores (sCAP). The analysis included 2747 patients, 1814 before and 933 after implementation. The median age was 71, median Elixhauser index was 17, 48% were female, and 95% were Caucasian. A mixed-effects regression model with cluster as the random effect estimated that implementation of CDS utilizing sCAP increased 30-day ICU-free days by 1.04 days (95% CI, 0.48-1.59; p < 0.001). Among secondary outcomes, the odds of being admitted to the ward, transferring to the ICU within 72 hours, and receiving a critical therapy decreased by 57% (odds ratio [OR], 0.43; 95% CI, 0.26-0.68; p < 0.001) post-implementation; mortality within 72 hours of admission was unchanged (OR, 1.08; 95% CI, 0.56-2.01; p = 0.82) while 30-day all-cause mortality was lower post-implementation (OR, 0.71; 95% CI, 0.52-0.96; p = 0.03). Implementation of electronic CDS using minor sCAP criteria to guide disposition of patients with pneumonia from the ED was associated with safe reduction in ICU utilization.

Survey of the Impact of Decision Support in Preoperative Management of Anemia (i-Anemia): Survey Study.

Major surgery on patients with anemia has demonstrated an increased risk of perioperative blood transfusions and postoperative morbidity and mortality. Recent studies have shown that integrating preoperative anemia treatment as a component of perioperative blood management may reduce blood product utilization and improve outcomes in both cardiac and noncardiac surgery. However, outpatient management of anemia falls outside of daily practice for most anesthesiologists and is probably weakly understood. We conducted a simulated case survey with anesthesiologists to accomplish the following aims: (1) evaluate the baseline knowledge of the preoperative optimization of anemia and (2) determine the impact of real-time clinical decision support on anemia management. We sent a digital survey (i-Anemia) to members of the French Society of Anaesthesia and Critical Care. The i-Anemia survey contained 7 simulated case vignettes, each describing a patient's brief clinical history and containing up to 3 multiple-choice questions related to preoperative anemia management (12 questions in total). The cases concerned potential situations of preoperative anemia and were created and validated with a committee of patient blood management experts. Correct answers were determined by the current guidelines or by expert consensus. Eligible participants were randomly assigned to control or decision support groups. In the decision support group, the primary outcome measured was the correct response rate. Overall, 1123 participants were enrolled and randomly divided into control (n=568) and decision support (n=555) groups. Among them, 763 participants fully responded to the survey. We obtained a complete response rate of 65.6% (n=364) in the group receiving cognitive aid and 70.2% (n=399) in the group without assistance. The mean duration of response was 10.2 (SD 6.8) minutes versus 7.8 (SD 5) minutes for the decision support and control groups, respectively (P<.001). The score significantly improved with cognitive aid (mean 10.3 out of 12, SD 2.1) in comparison to standard care (mean 6.2 out of 12, SD 2.1; P<.001). Management strategies to optimize preoperative anemia are not fully known and applied by anesthesiologists in daily practice despite their clinical importance. However, adding a decision support tool can significantly improve patient care by reminding practitioners of current recommendations.

A proposed framework for advancing acute kidney injury risk stratification and diagnosis in children: a report from the 26th Acute Disease Quality Initiative (ADQI) conference

Acute kidney injury (AKI) in children is associated with increased morbidity, reduced health-related quality of life, greater resource utilization, and higher mortality. Improvements in the timeliness and precision of AKI diagnosis in children are needed. In this report, we highlight existing, novel, and on-the-horizon diagnostic and risk-stratification tools for pediatric AKI, and outline opportunities for integration into clinical practice. We also summarize pediatric-specific high-risk diagnoses and exposures for AKI, as well as the potential role of real-time risk stratification and clinical decision support to improve outcomes. Lastly, the key characteristics of important pediatric AKI phenotypes will be outlined. Throughout, we identify key knowledge gaps, which represent prioritized areas of focus for future research that will facilitate a comprehensive, timely and personalized approach to pediatric AKI diagnosis and management.

Improving McDiarmid Tree Performance for Predicting Heart Disease from Data Streams with Missing and Meaningless Values

Building machine learning algorithms for real-time clinical decision support systems has become a current research hotspot. The success of these algorithms depends on their ability to handle data stream characteristics. For example, we cite as characteristics the large amounts of data, the high speed and rate of incoming data, and the change in data nature and distribution over time. The Very Fast Decision Tree (VFDT) is a method for incrementally building decision trees. Since its proposition in the literature, it has become one of the most popular tools for data stream classification. This paper aims to optimize a new version of VFDT called McDiarmid Tree (MT) for the early prediction of heart diseases in real-time clinical decision support systems. The proposed method for improving MT performance consists of two main mechanisms: detecting the presence of missing and meaningless values in data attributes and handling the impact of this presence. The proposed MT has been compared with MT and VFDT. Simulation results show that the proposed MT attains significantly higher prediction accuracy with less time and model cost (RAM-Hours) than the other two algorithms.

Natural Language Processing — A Surveillance Stepping Stone to Identify Child Abuse

ObjectiveWe aimed to refine a natural language processing (NLP) algorithm that identified injuries associated with child abuse and identify areas in which integration into a real-time clinical decision support (CDS) tool may improve clinical care. MethodsWe applied an NLP algorithm in “silent mode” to all emergency department (ED) provider notes between July 2021 and December 2022 (n = 353) at 1 pediatric and 8 general EDs. We refined triggers for the NLP, assessed adherence to clinical guidelines, and evaluated disparities in degree of evaluation by examining associations between demographic variables and abuse evaluation or reporting to child protective services. ResultsSeventy-three cases falsely triggered the NLP, often due to errors in interpreting linguistic context. We identified common false-positive scenarios and refined the algorithm to improve NLP specificity. Adherence to recommended evaluation standards for injuries defined by nationally accepted clinical guidelines was 63%. There were significant demographic differences in evaluation and reporting based on presenting ED type, insurance status, and race and ethnicity. ConclusionsAnalysis of an NLP algorithm in “silent mode” allowed for refinement of the algorithm and highlighted areas in which real-time CDS may help ED providers identify and pursue appropriate evaluation of injuries associated with child physical abuse.

Machine learning for cryosection pathology predicts the 2021 WHO classification of glioma

Timely and accurate intraoperative cryosection evaluations remain the gold standard for guiding surgical treatments for gliomas. However, the tissue-freezing process often generates artifacts that make histologic interpretation difficult. In addition, the 2021 WHO Classification of Tumors of the Central Nervous System incorporates molecular profiles in the diagnostic categories, so standard visual evaluation of cryosections alone cannot completely inform diagnoses based on the new classification system. To address these challenges, we develop the context-aware Cryosection Histopathology Assessment and Review Machine (CHARM) using samples from 1,524 glioma patients from three different patient populations to systematically analyze cryosection slides. Our CHARM models successfully identified malignant cells (AUROC= 0.98± 0.01 in the independent validation cohort), distinguished isocitrate dehydrogenase (IDH)-mutant tumors from wild type (AUROC= 0.79-0.82), classified three major types of molecularly defined gliomas (AUROC= 0.88-0.93), and identified the most prevalent subtypes of IDH-mutant tumors (AUROC= 0.89-0.97). CHARM further predicts clinically important genetic alterations in low-grade glioma, including ATRX, TP53, and CIC mutations, CDKN2A/B homozygous deletion, and 1p/19q codeletion via cryosection images. Our approaches accommodate the evolving diagnostic criteria informed by molecular studies, provide real-time clinical decision support, and will democratize accurate cryosection diagnoses. Supported in part by the National Institute of General Medical Sciences grant R35GM142879, the Google Research Scholar Award, the Blavatnik Center for Computational Biomedicine Award, the Partners' Innovation Discovery Grant, and the Schlager Family Award for Early Stage Digital Health Innovations.

Predicting Cardiac Arrest in Children with Heart Disease: A Novel Machine Learning Algorithm

Children with congenital and acquired heart disease are at a higher risk of cardiac arrest compared to those without heart disease. Although the monitoring of cardiopulmonary resuscitation quality and extracorporeal resuscitation technologies have advanced, survival after cardiac arrest in this population has not improved. Cardiac arrest prevention, using predictive algorithms with machine learning, has the potential to reduce cardiac arrest rates. However, few studies have evaluated the use of these algorithms in predicting cardiac arrest in children with heart disease. We collected demographic, laboratory, and vital sign information from the electronic health records (EHR) of all the patients that were admitted to a single-center pediatric cardiac intensive care unit (CICU), between 2010 and 2019, who had a cardiac arrest during their CICU admission, as well as a comparator group of randomly selected non-cardiac-arrest controls. We compared traditional logistic regression modeling against a novel adaptation of a machine learning algorithm (functional gradient boosting), using time series data to predict the risk of cardiac arrest. A total of 160 unique cardiac arrest events were matched to non-cardiac-arrest time periods. Using 11 different variables (vital signs and laboratory values) from the EHR, our algorithm's peak performance for the prediction of cardiac arrest was at one hour prior to the cardiac arrest (AUROC of 0.85 [0.79,0.90]), a performance that was similar to our previously published multivariable logistic regression model. Our novel machine learning predictive algorithm, which was developed using retrospective data that were collected from the EHR and predicted cardiac arrest in the children that were admitted to a single-center pediatric cardiac intensive care unit, demonstrated a performance that was similar to that of a traditional logistic regression model. While these results are encouraging, future research, including prospective validations with multicenter data, is warranted prior to the implementation of this algorithm as a real-time clinical decision support tool.

Implementation of a myasthenia gravis drug-disease interaction clinical decision support tool reduces prescribing of high-risk medications.

High-risk medication exposure is a modifiable risk factor for myasthenic exacerbation and crisis. We evaluated whether real-time electronic clinical decision support (CDS) was effective in reducing the rate of prescribing potentially high-risk medications to avoid or use with caution in patients with myasthenia gravis. An expert panel reviewed the available drug-disease pairings and associated severity levels to activate the alerts for CDS. All unique alerts activated in both inpatient and outpatient contexts were analyzed over a two-year period. Clinical context, alert severity, medication class, and alert action were collected. The primary outcome was alert override rate. Secondary outcomes included the percentage of unique medication exposures avoided and predictors of alert override. During the analysis period, 2817 unique alerts fired, representing 830 distinct patient-medication exposures for 577 unique patients. The overall alert override rate was 85% (80.3% for inpatient alerts and 95.8% for outpatient alerts). Of unique medication-patient exposures, 19% were avoided because of the alert. Assigned alert severity of "contraindicated" were less likely to be overridden (odds ratio [OR] 0.42, 95% confidence interval [CI] 0.32-0.56), as well as alerts activated during evening staffing (OR 0.69, 95% CI 0.55-0.87). Implementation of a myasthenia gravis drug-disease interaction alert reduced overall patient exposure to potentially harmful medications by approximately 19%. Future optimization includes enhanced provider and pharmacist education. Further refinement of alert logic criteria to optimize medication risk reduction and reduce alert fatigue is warranted to support clinicians in prescribing and reduce electronic health record time burden.

An Accessible Clinical Decision Support System to Curtail Anesthetic Greenhouse Gases in a Large Health Network: Implementation Study.

Inhaled anesthetics in the operating room are potent greenhouse gases and are a key contributor to carbon emissions from health care facilities. Real-time clinical decision support (CDS) systems lower anesthetic gas waste by prompting anesthesia professionals to reduce fresh gas flow (FGF) when a set threshold is exceeded. However, previous CDS systems have relied on proprietary or highly customized anesthesia information management systems, significantly reducing other institutions' accessibility to the technology and thus limiting overall environmental benefit. In 2018, a CDS system that lowers anesthetic gas waste using methods that can be easily adopted by other institutions was developed at the University of California San Francisco (UCSF). This study aims to facilitate wider uptake of our CDS system and further reduce gas waste by describing the implementation of the FGF CDS toolkit at UCSF and the subsequent implementation at other medical campuses within the University of California Health network. We developed a noninterruptive active CDS system to alert anesthesia professionals when FGF rates exceeded 0.7 L per minute for common volatile anesthetics. The implementation process at UCSF was documented and assembled into an informational toolkit to aid in the integration of the CDS system at other health care institutions. Before implementation, presentation-based education initiatives were used to disseminate information regarding the safety of low FGF use and its relationship to environmental sustainability. Our FGF CDS toolkit consisted of 4 main components for implementation: sustainability-focused education of anesthesia professionals, hardware integration of the CDS technology, software build of the CDS system, and data reporting of measured outcomes. The FGF CDS system was successfully deployed at 5 University of California Health network campuses. Four of the institutions are independent from the institution that created the CDS system. The CDS system was deployed at each facility using the FGF CDS toolkit, which describes the main components of the technology and implementation. Each campus made modifications to the CDS tool to best suit their institution, emphasizing the versatility and adoptability of the technology and implementation framework. It has previously been shown that the FGF CDS system reduces anesthetic gas waste, leading to environmental and fiscal benefits. Here, we demonstrate that the CDS system can be transferred to other medical facilities using our toolkit for implementation, making the technology and associated benefits globally accessible to advance mitigation of health care-related emissions.

Making Complex Cancer Reports Easier Using Self-Built Interactive Templates

Abstract Surgical pathology reports are becoming increasingly complex, especially with neoplastic cases. Pathologists need to verify all elements are addressed in the CAP synoptics, order proper reflex molecular testing or immunohistochemistry depending on care standards and institutional guidelines, and address key points in the report for clinicians. For example, breast cancer cases at our institution need repeat biomarkers for all cases status post neoadjuvant therapy, oncotype reflex testing for pT2N0, estrogen receptor positive tumors, and tumor bed size with residual tumor burden calculations for all breast cancer only after neoadjuvant chemotherapy. For residents and faculty alike, these expansive reports are challenging, especially for those new to the institution’s policies and conventions. Previous workflow included using Microsoft Word documents with previously signed out cases had significant risk for errors that stem from new updates in the CAP protocols, missing or different reflex testing from the previous case, and updated institutional guidelines. We set out to create a customized library of decision support and interactive templates using a software called PhraseExpander. We created PhraseExpander templates for all cancer resections. The templates included the associated synoptics required for accreditation by CAP, implemented automated staging using logic statements, and provided conditional reminders when reflex testing was indicated. Variable labels were utilized to help clarify notes in the CAP synoptic such as how to grade specific tumors and how many high-power fields to count. The tools were distributed to all surgical pathology fellows, residents, and select faculty who expressed interest in participating. The templates were successfully implemented into the routine workflow for surgical pathology sign out. End-user feedback was overwhelmingly positive. First year and upper level residents appreciate the detailed descriptions on how to complete the report, most notably the reflex molecular testing reminders and many conditional aspects such as in our breast cancer reports. Fellows and attendings have also appreciated these aspects as well as the automated staging and reminders to do administrative tasks, such as adding the case to the Laboratory Information System tumor registry. Interactive templates with built in conditions and calculators are powerful tools that can be used for generating the increasingly complex reports in pathology. They provide pathologists real-time and non-interruptive clinical decision support, allow them to spend less time typing and reduce reliance on remembering the required reflex testing. The most notable disadvantage in coding these complex templates is the considerable build and maintenance time. We plan to expand use to flow cytometry reports, biopsy reports, and routine non-neoplastic cases while incorporating labels to assist in resident education.

Artificial Intelligence Vision Based Analysis and Key Steps Identification Feasibility Study in Laparoscopic Hysterectomies

<h3>Study Objective</h3> The analysis of surgical videos and key steps identification using artificial intelligence (AI) holds great promise for the future of healthcare. In this paper, we present a novel computer vision algorithm for surgical steps identification in laparoscopic hysterectomy. <h3>Design</h3> A retrospective analysis of surgical videos of laparoscopic hysterectomies. <h3>Setting</h3> Gynecological department in a tertiary hospital with an average of 6500 procedures a year. <h3>Patients or Participants</h3> 190 laparoscopic hysterectomies from September 2020 to April 2022. <h3>Interventions</h3> Artificial intelligence-driven surgical platform that uses advanced computer vision technology to capture video data during all surgeries, de-identify it, and upload it to a secure cloud infrastructure. <h3>Measurements and Main Results</h3> A total of 190 full-length hysterectomies videos were manually annotated with sequential steps of surgery. Of these, 115 cases served as a training dataset for algorithm development, 28 cases were used for internal validation, and 47 were used as a separate testing cohort for evaluating algorithm accuracy. Concordance between AI-enabled automated video analysis and manual human video annotation was 92%. Algorithm accuracy was highest for the vaginal vault closure step (97%) and lowest for the adhesiolysis step (71%). <h3>Conclusion</h3> A variety of AI applications are expanding in clinical systems, from databases to intraoperative video analysis. The unique nature of surgical practice puts surgeons in a strong position to contribute to the next phase of AI, focused on real-time clinical decision support designed to optimize surgeon workflow and patient care. This work validates the ability to annotate the different steps of hysterectomy using an AI vision-based algorithm. Automated surgical video analysis has immediate impact in this field and the implementation of such a system holds enormous promise for future surgical education and improvement.

A time-aware attention model for prediction of acute kidney injury after pediatric cardiac surgery.

Acute kidney injury (AKI) is a common complication after pediatric cardiac surgery, and the early detection of AKI may allow for timely preventive or therapeutic measures. However, current AKI prediction researches pay less attention to time information among time-series clinical data and model building strategies that meet complex clinical application scenario. This study aims to develop and validate a model for predicting postoperative AKI that operates sequentially over individual time-series clinical data. A retrospective cohort of 3386 pediatric patients extracted from PIC database was used for training, calibrating, and testing purposes. A time-aware deep learning model was developed and evaluated from 3 clinical perspectives that use different data collection windows and prediction windows to answer different AKI prediction questions encountered in clinical practice. We compared our model with existing state-of-the-art models from 3 clinical perspectives using the area under the receiver operating characteristic curve (ROC AUC) and the area under the precision-recall curve (PR AUC). Our proposed model significantly outperformed the existing state-of-the-art models with an improved average performance for any AKI prediction from the 3 evaluation perspectives. This model predicted 91% of all AKI episodes using data collected at 24h after surgery, resulting in a ROC AUC of 0.908 and a PR AUC of 0.898. On average, our model predicted 83% of all AKI episodes that occurred within the different time windows in the 3 evaluation perspectives. The calibration performance of the proposed model was substantially higher than the existing state-of-the-art models. This study showed that a deep learning model can accurately predict postoperative AKI using perioperative time-series data. It has the potential to be integrated into real-time clinical decision support systems to support postoperative care planning.

Prediction of postoperative cardiac events in multiple surgical cohorts using a multimodal and integrative decision support system

Postoperative patients are at risk of life-threatening complications such as hemodynamic decompensation or arrhythmia. Automated detection of patients with such risks via a real-time clinical decision support system may provide opportunities for early and timely interventions that can significantly improve patient outcomes. We utilize multimodal features derived from digital signal processing techniques and tensor formation, as well as the electronic health record (EHR), to create machine learning models that predict the occurrence of several life-threatening complications up to 4 hours prior to the event. In order to ensure that our models are generalizable across different surgical cohorts, we trained the models on a cardiac surgery cohort and tested them on vascular and non-cardiac acute surgery cohorts. The best performing models achieved an area under the receiver operating characteristic curve (AUROC) of 0.94 on training and 0.94 and 0.82, respectively, on testing for the 0.5-hour interval. The AUROCs only slightly dropped to 0.93, 0.92, and 0.77, respectively, for the 4-hour interval. This study serves as a proof-of-concept that EHR data and physiologic waveform data can be combined to enable the early detection of postoperative deterioration events.

Process Mining the Performance of a Real-Time Healthcare 4.0 Systems Using Conditional Survival Models

As the world moves into the exciting age of Healthcare 4.0, it is essential that patients and clinicians have confidence and reassurance that the real-time clinical decision support systems being used throughout their care guarantee robustness and optimal quality of care. However, current systems involving autonomic behaviour and those with no prior clinical feedback, have generally to date had little focus on demonstrating robustness in the use of data and final output, thus generating a lack of confidence. This paper wishes to address this challenge by introducing a new process mining approach based on a statistically robust methodology that relies on the utilisation of conditional survival models for the purpose of evaluating the performance of Healthcare 4.0 systems and the quality of the care provided. Its effectiveness is demonstrated by analysing the performance of a clinical decision support system operating in an intensive care setting with the goal to monitor ventilated patients in real-time and to notify clinicians if the patient is predicted at risk of receiving injurious mechanical ventilation. Additionally, we will also demonstrate how the same metrics can be used for evaluating the patient quality of care. The proposed methodology can be used to analyse the performance of any Healthcare 4.0 system and the quality of care provided to the patient.

Machine learning algorithms as early diagnostic tools for pancreatic fistula following pancreaticoduodenectomy and guide drain removal: A retrospective cohort study

ObjectiveClinically relevant postoperative pancreatic fistula (CR-POPF) remains the major cause of morbidity following pancreaticoduodenectomy (PD). Several model score systems such as the Fistula Risk Score (FRS) have been developed to predict CR-POPF using preoperative and intraoperative data. Machine learning (ML) algorithms are increasingly applied in the medical field and they could be used to assess the risk of CR-POPF, identify clinically meaningful data and guide drain removal. MethodsData from consecutive patients who underwent PD between January 1, 2010 and March 31, 2021 at a single high-volume center was collected retrospectively in this study. Demographics, clinical features, intraoperative parameters, and laboratory values were used to conduct the ML model. Four different ML algorithms (CatBoost, lightGBM, XGBoost and Random Forest) were used to train this model with cross-validation. ResultsA total of 2421 patients with 62 clinical parameters were enrolled in this ML model. The majority of patients (76.3%) underwent open PD while others underwent robot-assisted PD. CR-POPF occurred in 424 (17.5%) patients. The CatBoost algorithm outperformed other algorithms with a mean area under the receiver operating characteristic curve (AUC) of 0.81 (95% confidence interval: 0.80–0.82) from the 5-fold cross-validation procedure. In the test dataset, the CatBoost algorithm also achieved the best mean-AUC of 0.83. The most important value was mean drain fluid amylase (DFA) in the first seven postoperative days (POD). The performance of models that used only preoperative data and intraoperative data was marginally lower than that of models that used combined data. ConclusionOur ML algorithms could be applied as early diagnostic tools for CR-POPF in patients who underwent PD. Such real-time clinical decision support tools can identify patients with a high risk of CR-POPF, help in developing the perioperative management plan and guide the optimal timing of drain removal.

A Pragmatic, Stepped-Wedge, Cluster-controlled Clinical Trial of Real-Time Pneumonia Clinical Decision Support.

Rationale: Care of emergency department (ED) patients with pneumonia can be challenging. Clinical decision support may decrease unnecessary variation and improve care. Objectives: To report patient outcomes and processes of care after deployment of electronic pneumonia clinical decision support (ePNa): a comprehensive, open loop, real-time clinical decision support embedded within the electronic health record. Methods: We conducted a pragmatic, stepped-wedge, cluster-controlled trial with deployment at 2-month intervals in 16 community hospitals. ePNa extracts real-time and historical data to guide diagnosis, risk stratification, microbiological studies, site of care, and antibiotic therapy. We included all adult ED patients with pneumonia over the course of 3 years identified by International Classification of Diseases, 10th Revision discharge coding confirmed by chest imaging. Measurements and Main Results: The median age of the 6,848 patients was 67 years (interquartile range, 50-79), and 48% were female; 64.8% were hospital admitted. Unadjusted mortality was 8.6% before and 4.8% after deployment. A mixed effects logistic regression model adjusting for severity of illness with hospital cluster as the random effect showed an adjusted odds ratio of 0.62 (0.49-0.79; P < 0.001) for 30-day all-cause mortality after deployment. Lower mortality was consistent across hospital clusters. ePNa-concordant antibiotic prescribing increased from 83.5% to 90.2% (P < 0.001). The mean time from ED admission to first antibiotic was 159.4 (156.9-161.9) minutes at baseline and 150.9 (144.1-157.8) minutes after deployment (P < 0.001). Outpatient disposition from the ED increased from 29.2% to 46.9%, whereas 7-day secondary hospital admission was unchanged (5.2% vs. 6.1%). ePNa was used by ED clinicians in 67% of eligible patients. Conclusions: ePNa deployment was associated with improved processes of care and lower mortality. Clinical trial registered with www.clinicaltrials.gov (NCT03358342).

Intelligent Clinical Decision Support.

Early recognition of pathologic cardiorespiratory stress and forecasting cardiorespiratory decompensation in the critically ill is difficult even in highly monitored patients in the Intensive Care Unit (ICU). Instability can be intuitively defined as the overt manifestation of the failure of the host to adequately respond to cardiorespiratory stress. The enormous volume of patient data available in ICU environments, both of high-frequency numeric and waveform data accessible from bedside monitors, plus Electronic Health Record (EHR) data, presents a platform ripe for Artificial Intelligence (AI) approaches for the detection and forecasting of instability, and data-driven intelligent clinical decision support (CDS). Building unbiased, reliable, and usable AI-based systems across health care sites is rapidly becoming a high priority, specifically as these systems relate to diagnostics, forecasting, and bedside clinical decision support. The ICU environment is particularly well-positioned to demonstrate the value of AI in saving lives. The goal is to create AI models embedded in a real-time CDS for forecasting and mitigation of critical instability in ICU patients of sufficient readiness to be deployed at the bedside. Such a system must leverage multi-source patient data, machine learning, systems engineering, and human action expertise, the latter being key to successful CDS implementation in the clinical workflow and evaluation of bias. We present one approach to create an operationally relevant AI-based forecasting CDS system.

Pharmaceutical algorithms set in a real time clinical decision support targeting high-alert medications applied to pharmaceutical analysis

BackgroundPharmaceutical analysis of the prescription has to prop up the quality of patients' medication management in a context of medication’s risk acculturation. But this activity remains highly variable. Medication-related clinical decision support may succeed in reducing adverse drug events and healthcare costs. PurposeThis study aims to present AVICENNE as a real time medication-related clinical decision support (rt-CDS) applied to pharmaceutical analysis and its ability to detect Drug related problems (DRP) consecutively resolved by pharmacists.Basic proceduresA Medication-related rt-CDS is created by integrating the software PharmaClass® (Keenturtle), 5 health data streams on the patient and Pharmaceutical algorithms (PA). PA are created by modeling the pharmaceutical experiment about DRP and the thread of their criticality. They are partially encoded as computerized rules in Pharmaclass® allowing alerts’ issue.An observational prospective study is conducted during 9-months among 1000 beds in 2 health facilities. The first step is to identify alerts as DRP; their resolution follows with clear guidelines worked out for the pharmaceutical analysis. A basis on predictive positive values (PPV) of the PA is being built today helping to know the performance of DRP detection and resolution.Main findings71 PA are encoded as rules into Pharmaclass®: 40 targeted serious adverse drug events. 1508 alerts are analyzed by pharmacists. Among them 921 DRPs were characterized and 540 pharmaceutical interventions transmitted of which 219 were accepted by prescribers. Three PPV are defined depending on software, pharmacist and patient.Principal conclusionClinical pharmacy societies should host, share and update a national corpus of PA and exploit its educational interest.