Automated Decision Support Tools Research Articles

A qualitative interview study to determine barriers and facilitators of implementing automated decision support tools for genomic data access

Data access committees (DAC) gatekeep access to secured genomic and related health datasets yet are challenged to keep pace with the rising volume and complexity of data generation. Automated decision support (ADS) systems have been shown to support consistency, compliance, and coordination of data access review decisions. However, we lack understanding of how DAC members perceive the value add of ADS, if any, on the quality and effectiveness of their reviews. In this qualitative study, we report findings from 13 semi-structured interviews with DAC members from around the world to identify relevant barriers and facilitators to implementing ADS for genomic data access management. Participants generally supported pilot studies that test ADS performance, for example in cataloging data types, verifying user credentials and tagging datasets for use terms. Concerns related to over-automation, lack of human oversight, low prioritization, and misalignment with institutional missions tempered enthusiasm for ADS among the DAC members we engaged. Tensions for change in institutional settings within which DACs operated was a powerful motivator for why DAC members considered the implementation of ADS into their access workflows, as well as perceptions of the relative advantage of ADS over the status quo. Future research is needed to build the evidence base around the comparative effectiveness and decisional outcomes of institutions that do/not use ADS into their workflows.

Version [1.1.0]- [DLDiagnosis: A mobile and web application for diseases classification using Deep Learning]

This paper presents version 1.1.0 of the DLDiagnosis software, which serves as an automated decision support tool for disease detection and classification through imaging analysis using deep learning. The latest update of DLDiagnosis includes significant enhancements, expanding the system to manage a total of eight different diseases. Three new diseases have been added: keratoconus, breast cancer, and pneumonia. This update focuses on refining user interface elements, identifying gaps in code logic, and improving mobile and web interface elements. Additionally, various functionalities have been improved, including reduced diagnostic latency, faster results, enhanced responsiveness, and a more consistent and intuitive diagnostic experience.

Implementing an Automated Prediction Model to Improve Prescribing of HIV Preexposure Prophylaxis

Summary Antiretroviral preexposure prophylaxis (PrEP) is nearly 100% effective at decreasing HIV acquisition but is underused in priority populations. Primary care clinicians need tools to help them identify persons likely to benefit from PrEP use and prescribe it when appropriate. The researchers developed and validated an automated decision support tool with interactive alerts in the electronic health record to increase PrEP discussions and prescribing in primary care. They piloted the tool at three federally qualified health centers and assessed feasibility, acceptance by clinicians, and preliminary impact on PrEP care. Of 33,803 patients who visited the pilot clinics from July 2022 through January 2023, providers received PrEP alerts at the point of care for 2.2% of patients, demonstrating feasibility. Although numbers of PrEP prescriptions remained low, the proportion of all patients with new PrEP prescriptions was 4.5 times higher at pilot clinics compared with matched control clinics (0.09% vs. 0.02%). Implementation of the decision support tool was associated with a statistically nonsignificant 5.5% increase in HIV tests per 100 patients. In qualitative interviews, providers said the tool facilitated PrEP discussions with patients, particularly for those patients who would not have initiated discussions because of stigma. The researchers found that acceptance, use, and impact of machine-learning models for PrEP depends on collaborating with and building trust among providers, including blending a data-driven approach to identifying patients at increased risk for HIV acquisition with providers’ traditional decision-making framework. These approaches could be useful for health care organizations seeking to implement automated prediction models across all areas of medicine.

Slope displacement detection in construction: An automated management algorithm for disaster prevention

In construction sites, embankment slope failure has led to catastrophic events, necessitating advanced monitoring for disaster prevention. This study presents a novel Slope Displacement Inspection and Management Algorithm (SDIMA) that utilizes Global Navigation Satellite System (GNSS) data to detect significant slope displacements automatically. SDIMA analyzes GNSS data to discern displacements in various directions (North-South, East-West, Vertical) and separate actual change points from measurement errors by employing a robust mathematical model called Displacement Detector. It successfully detected change points with displacements greater than two times the samples' standard deviation, and thresholds were set for displacements ≥ 25 mm. A comparative analysis showed promising efficiency in North-South detection, while certain limitations were identified. The proposed SDIMA outperforms existing models, achieving a Mean Absolute Error (MAE) of 3.157 mm, Root Mean Square Error (RMSE) of 4.754 mm, Mean Absolute Scaled Error (MASE) of 0.294, and an average processing time of 0.578 min, making it a highly efficient and accurate expert system. The innovation of SDIMA lies in its ability to provide an automated decision support tool for site managers, laying the foundation for a low-cost anomaly detection system for slope failure disaster prevention. The algorithm's proficiency and its potential to integrate with existing practices herald a significant advancement in construction management and safety.

Автоматизация сравнительного анализа компетенций специалистов в управлении персоналом

The article is devoted to the discussion of ways to create tools for comparative analysis of the professional competencies of specialists to automate the recruitment processes of a modern enterprise, which are supposed to be developed based on methods for analyzing the semantic proximity of generated texts based on natural language processing technologies. The paper describes the need to create solutions designed for a comparative assessment of the professional competencies of an employee when building a unified recruitment management system covering all inter-ested participants in the labor market and substantiates the possibility of using collections of texts created by a specialist for a comparative analysis of his capabilities. The scientific novelty of the work lies in the description of the aggregation procedures of consolidated collections of generated texts characterizing the competence of a specialist, which are based on the classification proposed by the author. The presented materials are of practical importance for the creation of automated decision support tools by HR departments of companies in terms of implementing procedures for objective assessments of the competencies of job seekers.

Extraction of CD&R Work Phases from Eye-Tracking and Simulator Logs: A Topic Modelling Approach

Automation in Air Traffic Control (ATC) is gaining an increasing interest. Possible relevant applications are in automated decision support tools leveraging the performance of the Air Traffic Controller (ATCO) when performing tasks such as Conflict Detection and Resolution (CD&R). Another important area of application is in ATCOs’ training by aiding instructors to assess the trainees’ strategies. From this perspective, models that capture the cognitive processes and reveal ATCOs’ work strategies need to be built. In this work, we investigated a novel approach based on topic modelling to learn controllers’ work patterns from temporal event sequences obtained by merging eye movement data with data from simulation logs. A comparison of the work phases exhibited by the topic models and the Conflict Life Cycle (CLC) reference model, derived from post-simulation interviews with the ATCOs, indicated that there was a correspondence between the phases captured by the proposed method and the CLC framework. Another contribution of this work is a method to assess similarities between ATCOs’ work strategies. A first proof-of-concept application targeting the CD&R task is also presented.

Towards Safe Mechanical Ventilation Treatment Using Deep Offline Reinforcement Learning

Mechanical ventilation is a key form of life support for patients with pulmonary impairment. Healthcare workers are required to continuously adjust ventilator settings for each patient, a challenging and time consuming task. Hence, it would be beneficial to develop an automated decision support tool to optimize ventilation treatment. We present DeepVent, a Conservative Q-Learning (CQL) based offline Deep Reinforcement Learning (DRL) agent that learns to predict the optimal ventilator parameters for a patient to promote 90 day survival. We design a clinically relevant intermediate reward that encourages continuous improvement of the patient vitals as well as addresses the challenge of sparse reward in RL. We find that DeepVent recommends ventilation parameters within safe ranges, as outlined in recent clinical trials. The CQL algorithm offers additional safety by mitigating the overestimation of the value estimates of out-of-distribution states/actions. We evaluate our agent using Fitted Q Evaluation (FQE) and demonstrate that it outperforms physicians from the MIMIC-III dataset.

Explainable AI identifies diagnostic cells of genetic AML subtypes.

Explainable AI is deemed essential for clinical applications as it allows rationalizing model predictions, helping to build trust between clinicians and automated decision support tools. We developed an inherently explainable AI model for the classification of acute myeloid leukemia subtypes from blood smears and found that high-attention cells identified by the model coincide with those labeled as diagnostically relevant by human experts. Based on over 80,000 single white blood cell images from digitized blood smears of 129 patients diagnosed with one of four WHO-defined genetic AML subtypes and 60 healthy controls, we trained SCEMILA, a single-cell based explainable multiple instance learning algorithm. SCEMILA could perfectly discriminate between AML patients and healthy controls and detected the APL subtype with an F1 score of 0.86±0.05 (mean±s.d., 5-fold cross-validation). Analyzing a novel multi-attention module, we confirmed that our algorithm focused with high concordance on the same AML-specific cells as human experts do. Applied to classify single cells, it is able to highlight subtype specific cells and deconvolve the composition of a patient's blood smear without the need of single-cell annotation of the training data. Our large AML genetic subtype dataset is publicly available, and an interactive online tool facilitates the exploration of data and predictions. SCEMILA enables a comparison of algorithmic and expert decision criteria and can present a detailed analysis of individual patient data, paving the way to deploy AI in the routine diagnostics for identifying hematopoietic neoplasms.

HADA: An automated tool for hardware dimensioning of AI applications

In recent years, the uptake of Artificial Intelligence (AI) in industry is increasing. For many AI techniques, like Deep Learning, optimization, planning, etc., computational and storage requirements are significant. The problem of determining what is the right hardware (HW on premise or on the cloud) architecture and its dimensioning for AI algorithms is still crucial. Searching for the optimal solution is often challenging, as it is not trivial to anticipate the behavior of an algorithm on diverse architectures. This is especially true if the AI application must respect quality-of-service constraints or budgets. In this scenario, having an automated decision support tool to match algorithms, user constraints and HW resources would be a great advantage for companies and practitioners working with AI applications.In this paper, we tackle this challenge with an approach that relies on the Empirical Model Learning paradigm, based on the integration of Machine Learning (ML) models into an optimization problem. The key idea is to integrate domain knowledge held by experts with data-driven models that learn the relationships between HW requirements and AI algorithm performances. In particular, the approach starts with benchmarking multiple AI algorithms on different HW resources, generating data used to train ML models; then, optimization is used to find the best HW configuration that respects user-defined constraints (e.g., budget, time, solution quality).In the experimental evaluation we validate our approach on a complex problem, namely online algorithms for energy systems, an area characterized by uncertainty and tight HW and real-time constraints. Results show the effectiveness of our approach and its flexibility: We can train the ML models only once and reuse them in the optimization model to tackle a variety of problems, determined by different data instances and user-defined constraints.

A Bayesian belief data mining approach applied to rice and shrimp aquaculture.

In many parts of the world, conditions for small scale agriculture are worsening, creating challenges in achieving consistent yields. The use of automated decision support tools, such as Bayesian Belief Networks (BBNs), can assist producers to respond to these factors. This paper describes a decision support system developed to assist farmers on the Mekong Delta, Vietnam, who grow both rice and shrimp crops in the same pond, based on an existing BBN. The BBN was previously developed in collaboration with local farmers and extension officers to represent their collective perceptions and understanding of their farming system and the risks to production that they face. This BBN can be used to provide insight into the probable consequences of farming decisions, given prevailing environmental conditions, however, it does not provide direct guidance on the optimal decision given those decisions. In this paper, the BBN is analysed using a novel, temporally-inspired data mining approach to systematically determine the agricultural decisions that farmers perceive as optimal at distinct periods in the growing and harvesting cycle, given the prevailing agricultural conditions. Using a novel form of data mining that combines with visual analytics, the results of this analysis allow the farmer to input the environmental conditions in a given growing period. They then receive recommendations that represent the collective view of the expert knowledge encoded in the BBN allowing them to maximise the probability of successful crops. Encoding the results of the data mining/inspection approach into the mobile Decision Support System helps farmers access explicit recommendations from the collective local farming community as to the optimal farming decisions, given the prevailing environmental conditions.

Abstract No. 25 ▪ FEATURED ABSTRACT Use of machine learning to predict the risk of pneumothorax requiring chest tube placement after lung biopsy

Pneumothorax is the leading complication after image-guided lung nodule biopsy (15-25%) with 4-6% resulting in a pneumothorax requiring chest tube placement . Risk factors associated with lung biopsy induced pneumothorax have been elucidated; however, currently no model exists to prospectively identify patients most at risk for pneumothorax necessitating chest tube placement. Development of a machine learning algorithm to predict patients at high risk for pneumothorax and chest tube placement may facilitate appropriate peri-procedural planning and optimize patient care. Retrospective chart review was performed on patients who underwent lung nodule biopsy between January 2018 and July 2019. From this data, we selected 58 lung biopsy procedures that resulted in pneumothorax requiring chest tube placement and then another 51 procedures without chest tube placement (convenience sample). For each procedure, we collected 12 clinical features that would be available prior to biopsy: age, gender, smoking history, prior asthma or COPD diagnosis, cancer history, pulmonary function tests, biopsy needle gauge, lesion size, lesion location, lesion character (solid or not), patient position during biopsy, and pleura to lesion distance. The data was divided 70/30 into training and testing sets. We explored several conventional machine-learned binary classifiers with the clinical data as input features and need for chest tube placement as the output. Summary statistics of the data reveal that several of the features have unique means between the chest tube placement positive and negative groups. For example, the set requiring tube placement versus not had a mean lesion depth of 30 ± 19 mm versus 13 ± 14 mm (P< 0.005). That said, there was no single pathognomonic imaging or clinical feature. On the independent test set, use of decision tree, logistic regression, and Naïve Bayes achieved accuracies of identifying tube placement of 0.76, 0.74, and 0.76 and had AUC’s of 0.83, 0.76, 0.78. The learned features of high importance included lesion character, COPD status, lesion depth, and patient age. A coarse decision tree requiring only 3 inputs achieved comparable performance as other methods. We identified an interpretable decision tree to predict chest tube placement post biopsy with an accuracy of 76% and AUC of 0.83. This work suggests that it may be possible to predict the risk of pneumothorax requiring chest tube placement based on an automated decision support tool reliant on easily available clinical and imaging features.

Intelligent skin cancer detection applying autoencoder, MobileNetV2 and spiking neural networks

Abstract Melanocytes are skin cells that give color to the skin and form melanin color pigments. The unbalanced division and proliferation of these cells result in skin cancer. The early diagnosis and proper treatment of skin cancer are so important. In this scope, a novel model that relies upon the autoencoder, spiking, and convolutional neural networks is proposed to ensure a useful decision support tool in this study. The experiments were carried out on an open-access dataset called the ISIC skin cancer consisting of 1800 being and 1497 malignant tumor images. In the proposed approach, the dataset is reconstructed using the autoencoder model. The original dataset and structured dataset were trained and classified by the MobileNetV2 model that consists of residual blocks, and the spiking networks. The classification success rate of the study was 95.27%. As a result, it was seen that the autoencoder model and spiking networks contributed to enhancing the performance of the MobileNetV2 model. Thanks to the proposed model, a novel fully automated decision support tool with high sensitivity was ensured for skin cancer detection.

Computational Barthel Index: an automated tool for assessing and predicting activities of daily living among nursing home patients

BackgroundAssessment of functional ability, including activities of daily living (ADLs), is a manual process completed by skilled health professionals. In the presented research, an automated decision support tool, the Computational Barthel Index Tool (CBIT), was constructed that can automatically assess and predict probabilities of current and future ADLs based on patients’ medical history.MethodsThe data used to construct the tool include the demographic information, inpatient and outpatient diagnosis codes, and reported disabilities of 181,213 residents of the Department of Veterans Affairs’ (VA) Community Living Centers. Supervised machine learning methods were applied to construct the CBIT. Temporal information about times from the first and the most recent occurrence of diagnoses was encoded. Ten-fold cross-validation was used to tune hyperparameters, and independent test sets were used to evaluate models using AUC, accuracy, recall and precision. Random forest achieved the best model quality. Models were calibrated using isotonic regression.ResultsThe unabridged version of CBIT uses 578 patient characteristics and achieved average AUC of 0.94 (0.93–0.95), accuracy of 0.90 (0.89–0.91), precision of 0.91 (0.89–0.92), and recall of 0.90 (0.84–0.95) when re-evaluating patients. CBIT is also capable of predicting ADLs up to one year ahead, with accuracy decreasing over time, giving average AUC of 0.77 (0.73–0.79), accuracy of 0.73 (0.69–0.80), precision of 0.74 (0.66–0.81), and recall of 0.69 (0.34–0.96). A simplified version of CBIT with 50 top patient characteristics reached performance that does not significantly differ from full CBIT.ConclusionDischarge planners, disability application reviewers and clinicians evaluating comparative effectiveness of treatments can use CBIT to assess and predict information on functional status of patients.

����� ��������� � ������� �������� ��������� ����� ���� ������������ �������.

In conditions of constant displays of aggression in various fields of activity, which is one of the most acute problems of our time, corresponding preventive measures are becoming increasingly important. In its turn, it requires the improvement of the methodological support for decision-making to ensure security and counter various kinds of attacks. At the present stage, the remaining slowness of identification of incidents and determining the scale of their devel-opment is a characteristic feature of security management. Methodological complexity of the procedures for assessing security risks and corresponding calculations is stipulated by the lack of appropriate methods, software and technologies on the market, which complicates the devel-opment of automated decision support tools. This paper offers a universal method based on a formalized model of the relationship between objects (resources), aimed at them threats and expected risks. This model is based on the �threat � object� binary scheme expressed in a bipar-tite graph and provides for the division of the graph into subgraphs. There is built another (third) set that corresponds to the safety model with full overlap and protection mechanisms as well. Simultaneous utilization of the elements of ontological descriptions increases the level of the method concreteness and a clearer understanding of the environment state. A matrix ap-proach, which is closely related to graph models, is proposed for computations. The article pro-vides some schemes of technological implementation of the obtained solutions based on office applications and designing a corresponding information and analytical system. The proposed method can be applied in automated systems in many fields of activity.

Predicting Unplanned Readmissions Following a Hip or Knee Arthroplasty: Retrospective Observational Study.

BackgroundTotal joint replacements are high-volume and high-cost procedures that should be monitored for cost and quality control. Models that can identify patients at high risk of readmission might help reduce costs by suggesting who should be enrolled in preventive care programs. Previous models for risk prediction have relied on structured data of patients rather than clinical notes in electronic health records (EHRs). The former approach requires manual feature extraction by domain experts, which may limit the applicability of these models.ObjectiveThis study aims to develop and evaluate a machine learning model for predicting the risk of 30-day readmission following knee and hip arthroplasty procedures. The input data for these models come from raw EHRs. We empirically demonstrate that unstructured free-text notes contain a reasonably predictive signal for this task.MethodsWe performed a retrospective analysis of data from 7174 patients at Partners Healthcare collected between 2006 and 2016. These data were split into train, validation, and test sets. These data sets were used to build, validate, and test models to predict unplanned readmission within 30 days of hospital discharge. The proposed models made predictions on the basis of clinical notes, obviating the need for performing manual feature extraction by domain and machine learning experts. The notes that served as model inputs were written by physicians, nurses, pathologists, and others who diagnose and treat patients and may have their own predictions, even if these are not recorded.ResultsThe proposed models output readmission risk scores (propensities) for each patient. The best models (as selected on a development set) yielded an area under the receiver operating characteristic curve of 0.846 (95% CI 82.75-87.11) for hip and 0.822 (95% CI 80.94-86.22) for knee surgery, indicating reasonable discriminative ability.ConclusionsMachine learning models can predict which patients are at a high risk of readmission within 30 days following hip and knee arthroplasty procedures on the basis of notes in EHRs with reasonable discriminative power. Following further validation and empirical demonstration that the models realize predictive performance above that which clinical judgment may provide, such models may be used to build an automated decision support tool to help caretakers identify at-risk patients.

Outsourcing glucose management to AI

An automated decision support tool could expand access to intensive insulin therapy for patients with type 1 diabetes.

Insulin dose optimization using an automated artificial intelligence-based decision support system in youths with type 1 diabetes.

Despite the increasing adoption of insulin pumps and continuous glucose monitoring devices, most people with type 1 diabetes do not achieve their glycemic goals1. This could be related to a lack of expertise or inadequate time for clinicians to analyze complex sensor-augmented pump data. We tested whether frequent insulin dose adjustments guided by an automated artificial intelligence-based decision support system (AI-DSS) is as effective and safe as those guided by physicians in controlling glucose levels. ADVICE4U was a six-month, multicenter, multinational, parallel, randomized controlled, non-inferiority trial in 108 participants with type 1 diabetes, aged 10-21 years and using insulin pump therapy (ClinicalTrials.gov no. NCT03003806). Participants were randomized 1:1 to receive remote insulin dose adjustment every three weeks guided by either an AI-DSS, (AI-DSS arm, n = 54) or by physicians (physician arm, n = 54). The results for the primary efficacy measure-the percentage of time spent within the target glucose range (70-180 mg dl-1 (3.9-10.0 mmol l-1))-in the AI-DSS arm were statistically non-inferior to those in the physician arm (50.2 ± 11.1% versus 51.6 ± 11.3%, respectively, P < 1 × 10-7). The percentage of readings below 54 mg dl-1 (<3.0 mmol l-1) within the AI-DSS arm was statistically non-inferior to that in the physician arm (1.3 ± 1.4% versus 1.0 ± 0.9%, respectively, P < 0.0001). Three severe adverse events related to diabetes (two severe hypoglycemia, one diabetic ketoacidosis) were reported in the physician arm and none in the AI-DSS arm. In conclusion, use of an automated decision support tool for optimizing insulin pump settings was non-inferior to intensive insulin titration provided by physicians from specialized academic diabetes centers.

The Roles Of Assisters And Automated Decision Support Tools In Consumers' Marketplace Choices: Room For Improvement.

Assisters provide in-person and phone-based support to help consumers narrow their plan options on the Affordable Care Act's health insurance Marketplaces. We elicited the perspectives of a national sample of thirty-two assisters from ten states on consumer plan selection and available Marketplace decision support tools (for example, total cost estimators and provider network look-up tools). Assisters identified several shortcomings that limited their use of decision support tools, such as nonspecific cost estimates and inaccurate provider network data. Assisters instead provided individualized cost estimates, called provider offices to verify network coverage, and found innovative strategies to help consumers access care affordably under their chosen plan. Two priorities emerged for optimizing consumers' Marketplace insurance selection process: improve the quality of data used in decision support tools and invest in assister programs. Assister strategies should be a benchmark for improving decision support tools, with lessons to be learned for future tool development.

Simplifying Caregiver Resources in Eldercare: Identifying the Support Needs of Caregiving Employees

Background: The majority of long-term care provided to older adults and persons with disabilities is provided by unpaid family caregivers and friends. Employers have a stake in long-term care services as well since 60% of caregivers are employed outside the home, 49% have gone in late, left early, or taken time off during the day to deal with caregiving issues, and 15% have taken a prolonged leave of absence. Additionally, 87% of employed caregivers make telephone calls for caregiving from work. Presentism, the state of being on-the-job, but because of caregiving not fully functioning, can greatly reduce individual productivity. Caregivers emphasize the need for information to assist in their tasks with the major barriers to utilization of caregiver resources identified as lack of awareness of available services, challenges accessing community services, and understanding eligibility criteria. Context: Roobrik is an online decision support tool which uses branching tree questions about caregiver and dependent characteristics to help identify caregiver needs, matching these to local community services. The Council on Aging of Middle Tennessee’s (COA) Directory of Services provides a comprehensive online community resource guide. The human resources department of four Nashville-based employers previously supportive of COA’s work were contacted and agreed to provide information to employees experiencing caregiver concerns. Volunteer employed caregivers willing to utilize the Roobrik Tool and the COA directory and contribute to a focus group discussion were invited to participate. Methods: Phase I. Focus group responses from 19 participants were recorded by an experienced journalist and primary themes categorized by the focus group leader according to questions posed to the participants. Phase II. Secondary themes were developed by expert consensus following review of individual responses by three investigators guided by categories of the 2015 National Alliance for Caregiving’s National Caregiver Survey. Results: Caregivers spent less than 15 minutes to research and determine the most appropriate local care options. Half of caregivers contacted later planned to act on these recommendations. Based on focus group questions, primary themes were identified: 1) utilization of automated decision support tools, 2) how decision support tools help caregivers understand the caregiver role, 3) ways to improve decision support tool function, and 4) recommendations for improvement of decision support resources. Based on the content of specific caregiver responses, these secondary themes were identified: 1) information needs-medical, legal, quality of life, and the processes involved in obtaining services, 2) needs for specific community support services, 3) need for informal and professional emotional support to reduce caregiver stress, and 4) ways the workplace can support employed caregivers. Workplace policies including flexible time, caregiver support groups, and working remotely may be beneficial to caregivers as well as the workplace. Conclusion: Our experience implementing a demonstration of the Roobrik tool coupled with the online COA Directory of Services suggests these may be useful and accessible caregiver resources that simplify care navigation. Some caregiver needs are beyond the capability of online decision support and may require personal guidance.

Decision Support Tool Improves Real and Perceived Anesthesiology Resident Relief Equity.

The Accreditation Council of Graduate Medical Education requires monitoring of resident clinical and educational hours but does not require tracking daily work patterns or duty hour equity. Lack of such monitoring may allow for inequity that affects resident morale. No defined system for resident relief of weekday operating room (OR) clinical duties existed at our institution, leaving on-call residents to independently decide daily relief order. We developed an automated decision support tool (DST) to improve equitable decision making for clinical relief and assessed its impact on real and perceived relief equity. The DST sent a daily e-mail to the senior resident responsible for relief decisions. It contained a prioritized relief list of noncall residents who worked in the OR beyond 5 PM the prior clinical day. We assessed actual relief equity using the number of times a resident worked in the OR past 5:30 PM on 2 consecutive weekdays as our outcome, adjusting for the mean number of open ORs each day between 5:00 PM and 6:59 PM in our main OR areas. We analyzed 14 months of data before implementation and 16 months of data after implementation. We assessed perceived relief equity before and after implementation using a questionnaire. After implementing the DST, the percentage of residents held 2 consecutive weekdays over the total of resident days worked decreased from 1.33% to 0.43%. The percentage of residents held beyond 5:30 PM on any given day decreased from 18.09% to 12.64%. Segmented regression analysis indicated that implementation of the DST was associated with a reduction in biweekly time series of residents kept late 2 days in a row, independent of the mean number of ORs in use. Surveyed residents reported the DST aided their ability to make equitable relief decisions (pre 60% versus post 94%; P = .0003). Eighty-five percent of residents strongly agreed that a prioritized relief list based on prior day work hours after 5 PM aided their decision making. After implementation, residents reported fewer instances of working past 5 PM within the past month (P < .005). A DST systematizing the relief process for anesthesiology residents was associated with a lower frequency of residents working beyond 5:30 PM in the OR on 2 consecutive days. The DST improved the perceived ability to make equitable relief decisions by on-call senior residents and residents being relieved. Success with this tool allows for broader applications in resident education, enabling enhanced monitoring of resident experiences and support for OR assignment decisions.