Machine Learning-based Clinical Decision Support Research Articles

The economics of deep and machine learning-based algorithms for COVID-19 prediction, detection, and diagnosis shaping the organizational management of hospitals

Research background: Deep and machine learning-based algorithms can assist in COVID-19 image-based medical diagnosis and symptom tracing, optimize intensive care unit admission, and use clinical data to determine patient prioritization and mortality risk, being pivotal in qualitative care provision, reducing medical errors, and increasing patient survival rates, thus diminishing the massive healthcare system burden in relation to severe COVID-19 inpatient stay duration, while increasing operational costs throughout the organizational management of hospitals. Data-driven financial and scenario-based contingency planning, predictive modelling tools, and risk pooling mechanisms should be deployed for additional medical equipment and unforeseen healthcare demand expenses. Purpose of the article: We show that deep and machine learning-based and clinical decision making systems can optimize patient survival likelihood and treatment outcomes with regard to susceptible, infected, and recovered individuals, performing accurate analyses by data modeling based on vital and clinical signs, surveillance data, and infection-related biomarkers, and furthering hospital facility optimization in terms of intensive care unit bed allocation. Methods: The review software systems employed for article screening and quality evaluation were: AMSTAR, AXIS, DistillerSR, Eppi-Reviewer, MMAT, PICO Portal, Rayyan, ROBIS, and SRDR. Findings & value added: Deep and machine learning-based clinical decision support tools can forecast COVID-19 spread, confirmed cases, and infection and mortality rates for data-driven appropriate treatment and resource allocations in effective therapeutic and diagnosis protocol development, by determining suitable measures and regulations and by using symptoms and comorbidities, vital signs, clinical and laboratory data and medical records across intensive care units, impacting the healthcare financing infrastructure. As a result of heightened use of personal protective equipment, hospital pharmacy and medication, outpatient treatment, and medical supplies, revenue loss and financial vulnerability occur, also due to expenses related to hiring additional staff and to critical resource expenditures. Hospital costs for COVID-19 medical care, screening, treatment capacity expansion, and personal protective equipment can lead to further financial losses while affecting COVID-19 frontline hospital workers and patients.

Prediction of Antibiotic Resistance in Patients With a Urinary Tract Infection: Algorithm Development and Validation.

The early prediction of antibiotic resistance in patients with a urinary tract infection (UTI) is important to guide appropriate antibiotic therapy selection. In this study, we aimed to predict antibiotic resistance in patients with a UTI. Additionally, we aimed to interpret the machine learning models we developed. The electronic medical records of patients who were admitted to Yongin Severance Hospital, South Korea were used. A total of 71 features extracted from patients' admission, diagnosis, prescription, and microbiology records were used for classification. UTI pathogens were classified as either sensitive or resistant to cephalosporin, piperacillin-tazobactam (TZP), carbapenem, trimethoprim-sulfamethoxazole (TMP-SMX), and fluoroquinolone. To analyze how each variable contributed to the machine learning model's predictions of antibiotic resistance, we used the Shapley Additive Explanations method. Finally, a prototype machine learning-based clinical decision support system was proposed to provide clinicians the resistance probabilities for each antibiotic. The data set included 3535, 737, 708, 1582, and 1365 samples for cephalosporin, TZP, TMP-SMX, fluoroquinolone, and carbapenem resistance prediction models, respectively. The area under the receiver operating characteristic curve values of the random forest models were 0.777 (95% CI 0.775-0.779), 0.864 (95% CI 0.862-0.867), 0.877 (95% CI 0.874-0.880), 0.881 (95% CI 0.879-0.882), and 0.884 (95% CI 0.884-0.885) in the training set and 0.638 (95% CI 0.635-0.642), 0.630 (95% CI 0.626-0.634), 0.665 (95% CI 0.659-0.671), 0.670 (95% CI 0.666-0.673), and 0.721 (95% CI 0.718-0.724) in the test set for predicting resistance to cephalosporin, TZP, carbapenem, TMP-SMX, and fluoroquinolone, respectively. The number of previous visits, first culture after admission, chronic lower respiratory diseases, administration of drugs before infection, and exposure time to these drugs were found to be important variables for predicting antibiotic resistance. The study results demonstrated the potential of machine learning to predict antibiotic resistance in patients with a UTI. Machine learning can assist clinicians in making decisions regarding the selection of appropriate antibiotic therapy in patients with a UTI.

Machine learning-based clinical decision support system for treatment recommendation and overall survival prediction of hepatocellular carcinoma: a multi-center study

The treatment decisions for patients with hepatocellular carcinoma are determined by a wide range of factors, and there is a significant difference between the recommendations of widely used staging systems and the actual initial treatment choices. Herein, we propose a machine learning-based clinical decision support system suitable for use in multi-center settings. We collected data from nine institutions in South Korea for training and validation datasets. The internal and external datasets included 935 and 1750 patients, respectively. We developed a model with 20 clinical variables consisting of two stages: the first stage which recommends initial treatment using an ensemble voting machine, and the second stage, which predicts post-treatment survival using a random survival forest algorithm. We derived the first and second treatment options from the results with the highest and the second-highest probabilities given by the ensemble model and predicted their post-treatment survival. When only the first treatment option was accepted, the mean accuracy of treatment recommendation in the internal and external datasets was 67.27% and 55.34%, respectively. The accuracy increased to 87.27% and 86.06%, respectively, when the second option was included as the correct answer. Harrell’s C index, integrated time-dependent AUC curve, and integrated Brier score of survival prediction in the internal and external datasets were 0.8381 and 0.7767, 91.89 and 86.48, 0.12, and 0.14, respectively. The proposed system can assist physicians by providing data-driven predictions for reference from other larger institutions or other physicians within the same institution when making treatment decisions.

Holistic Human-Serving Digitization of Health Care Needs Integrated Automated System-Level Assessment Tools.

Digital health tools, platforms, and artificial intelligence- or machine learning-based clinical decision support systems are increasingly part of health delivery approaches, with an ever-greater degree of system interaction. Critical to the successful deployment of these tools is their functional integration into existing clinical routines and workflows. This depends on system interoperability and on intuitive and safe user interface design. The importance of minimizing emergent workflow stress through human factors research and purposeful design for integration cannot be overstated. Usability of tools in practice is as important as algorithm quality. Regulatory and health technology assessment frameworks recognize the importance of these factors to a certain extent, but their focus remains mainly on the individual product rather than on emergent system and workflow effects. The measurement of performance and user experience has so far been performed in ad hoc, nonstandardized ways by individual actors using their own evaluation approaches. We propose that a standard framework for system-level and holistic evaluation could be built into interacting digital systems to enable systematic and standardized system-wide, multiproduct, postmarket surveillance and technology assessment. Such a system could be made available to developers through regulatory or assessment bodies as an application programming interface and could be a requirement for digital tool certification, just as interoperability is. This would enable health systems and tool developers to collect system-level data directly from real device use cases, enabling the controlled and safe delivery of systematic quality assessment or improvement studies suitable for the complexity and interconnectedness of clinical workflows using developing digital health technologies.

Prospective evaluation of a machine learning-based clinical decision support system (ViSIG) in reducing adverse outcomes for adult critically ill patients

IntroductionProspective clinical evaluations of decision support tools for the ICU are almost non-existent. We sought to test whether a novel clinical decision support tool called ViSIG, based on machine learning algorithms, could be delivered via a user interface to clinicians in real-time, and if this led to changes in ICU mortality, length of stay (LOS), readmission, and duration of mechanical ventilation (dMV). MethodsThe study took place in six adult ICUs at two hospitals in the U.S. There were two study phases: one where clinicians were blinded to ViSIG scores (ViS), followed by a phase where ViS were visible via a user interface. The patients' ViS were updated every 30 min. ICU mortality, ICULOS, readmission status, and dMV were assessed on each patient in both phases, along with possible severity of illness confounders. Changes in outcomes between phases were evaluated by multivariable models, adjusting for patients’ severity of illness. ResultsThere were 2256 admissions in the blinded phase, followed by 1890 admissions in the unblinded phase. ICU mortality decreased from the blinded to the unblinded phase, but severity-adjusted mortality was unchanged. Actual and severity-adjusted ICULOS declined significantly from the blinded to unblinded phase (0.38 days). Severity-adjusted readmissions were reduced by 35 %, as was severity-adjusted dMV (0.62 days). ConclusionMaking information from ViSIG visible to clinicians in real-time was statistically associated with severity-adjusted decreases in ICULOS, readmission, and dMV. Future studies are needed to determine whether the decreases in these outcomes are due to changes in treatment delivery resulting from clinician knowledge of patients’ ViSIG scores.

Machine learning-based clinical decision support for infection risk prediction.

Healthcare-associated infection (HAI) remains a significant risk for hospitalized patients and a challenging burden for the healthcare system. This study presents a clinical decision support tool that can be used in clinical workflows to proactively engage secondary assessments of pre-symptomatic and at-risk infection patients, thereby enabling earlier diagnosis and treatment. This study applies machine learning, specifically ensemble-based boosted decision trees, on large retrospective hospital datasets to develop an infection risk score that predicts infection before obvious symptoms present. We extracted a stratified machine learning dataset of 36,782 healthcare-associated infection patients. The model leveraged vital signs, laboratory measurements and demographics to predict HAI before clinical suspicion, defined as the order of a microbiology test or administration of antibiotics. Our best performing infection risk model achieves a cross-validated AUC of 0.88 at 1 h before clinical suspicion and maintains an AUC >0.85 for 48 h before suspicion by aggregating information across demographics and a set of 163 vital signs and laboratory measurements. A second model trained on a reduced feature space comprising demographics and the 36 most frequently measured vital signs and laboratory measurements can still achieve an AUC of 0.86 at 1 h before clinical suspicion. These results compare favorably against using temperature alone and clinical rules such as the quick sequential organ failure assessment (qSOFA) score. Along with the performance results, we also provide an analysis of model interpretability via feature importance rankings. The predictive model aggregates information from multiple physiological parameters such as vital signs and laboratory measurements to provide a continuous risk score of infection that can be deployed in hospitals to provide advance warning of patient deterioration.

EE435 Cost-Analysis of a Machine Learning-Based Clinical Decision Support System to Guide Resilience-Strengthening Intervention Decisions in Breast Cancer Treatment: The Bounce Study

EE435 Cost-Analysis of a Machine Learning-Based Clinical Decision Support System to Guide Resilience-Strengthening Intervention Decisions in Breast Cancer Treatment: The Bounce Study

Machine learning-based clinical decision support using laboratory data.

Artificial intelligence (AI) and machine learning (ML) are becoming vital in laboratory medicine and the broader context of healthcare. In this review article, we summarized the development of ML models and how they contribute to clinical laboratory workflow and improve patient outcomes. The process of ML model development involves data collection, data cleansing, feature engineering, model development, and optimization. These models, once finalized, are subjected to thorough performance assessments and validations. Recently, due to the complexity inherent in model development, automated ML tools were also introduced to streamline the process, enabling non-experts to create models. Clinical Decision Support Systems (CDSS) use ML techniques on large datasets to aid healthcare professionals in test result interpretation. They are revolutionizing laboratory medicine, enabling labs to work more efficiently with less human supervision across pre-analytical, analytical, and post-analytical phases. Despite contributions of the ML tools at all analytical phases, their integration presents challenges like potential model uncertainties, black-box algorithms, and deskilling of professionals. Additionally, acquiring diverse datasets is hard, and models' complexity can limit clinical use. In conclusion, ML-based CDSS in healthcare can greatly enhance clinical decision-making. However, successful adoption demands collaboration among professionals and stakeholders, utilizing hybrid intelligence, external validation, and performance assessments.

Implementing a Machine Learning Screening Tool for Malnutrition: Insights From Qualitative Research Applicable to Other Machine Learning-Based Clinical Decision Support Systems.

Machine learning (ML)-based clinical decision support systems (CDSS) are popular in clinical practice settings but are often criticized for being limited in usability, interpretability, and effectiveness. Evaluating the implementation of ML-based CDSS is critical to ensure CDSS is acceptable and useful to clinicians and helps them deliver high-quality health care. Malnutrition is a common and underdiagnosed condition among hospital patients, which can have serious adverse impacts. Early identification and treatment of malnutrition are important. This study aims to evaluate the implementation of an ML tool, Malnutrition Universal Screening Tool (MUST)-Plus, that predicts hospital patients at high risk for malnutrition and identify best implementation practices applicable to this and other ML-based CDSS. We conducted a qualitative postimplementation evaluation using in-depth interviews with registered dietitians (RDs) who use MUST-Plus output in their everyday work. After coding the data, we mapped emergent themes onto select domains of the nonadoption, abandonment, scale-up, spread, and sustainability (NASSS) framework. We interviewed 17 of the 24 RDs approached (71%), representing 37% of those who use MUST-Plus output. Several themes emerged: (1) enhancements to the tool were made to improve accuracy and usability; (2) MUST-Plus helped identify patients that would not otherwise be seen; perceived usefulness was highest in the original site; (3) perceived accuracy varied by respondent and site; (4) RDs valued autonomy in prioritizing patients; (5) depth of tool understanding varied by hospital and level; (6) MUST-Plus was integrated into workflows and electronic health records; and (7) RDs expressed a desire to eventually have 1 automated screener. Our findings suggest that continuous involvement of stakeholders at new sites given staff turnover is vital to ensure buy-in. Qualitative research can help identify the potential bias of ML tools and should be widely used to ensure health equity. Ongoing collaboration among CDSS developers, data scientists, and clinical providers may help refine CDSS for optimal use and improve the acceptability of CDSS in the clinical context.

Development of a machine learning-based clinical decision support system to predict clinical deterioration in patients visiting the emergency department

This study aimed to develop a machine learning-based clinical decision support system for emergency departments based on the decision-making framework of physicians. We extracted 27 fixed and 93 observation features using data on vital signs, mental status, laboratory results, and electrocardiograms during emergency department stay. Outcomes included intubation, admission to the intensive care unit, inotrope or vasopressor administration, and in-hospital cardiac arrest. eXtreme gradient boosting algorithm was used to learn and predict each outcome. Specificity, sensitivity, precision, F1 score, area under the receiver operating characteristic curve (AUROC), and area under the precision-recall curve were assessed. We analyzed 303,345 patients with 4,787,121 input data, resampled into 24,148,958 1 h-units. The models displayed a discriminative ability to predict outcomes (AUROC > 0.9), and the model with lagging 6 and leading 0 displayed the highest value. The AUROC curve of in-hospital cardiac arrest had the smallest change, with increased lagging for all outcomes. With inotropic use, intubation, and intensive care unit admission, the range of AUROC curve change with the leading 6 was the highest according to different amounts of previous information (lagging). In this study, a human-centered approach to emulate the clinical decision-making process of emergency physicians has been adopted to enhance the use of the system. Machine learning-based clinical decision support systems customized according to clinical situations can help improve the quality of care.

Development and validation of machine learning-based clinical decision support tool for identifying malnutrition in NICU patients

Hospitalized newborns have an increased risk of malnutrition and, especially preterm infants, often experience malnutrition-related extrauterine growth restriction (EUGR). The aim of this study was to predict the discharge weight and the presence of weight gain at discharge with machine learning (ML) algorithms. The demographic and clinical parameters were used to develop the models using fivefold cross-validation in the software-R with a neonatal nutritional screening tool (NNST). A total of 512 NICU patients were prospectively included in the study. Length of hospital stay (LOS), parenteral nutrition treatment (PN), postnatal age (PNA), surgery, and sodium were the most important variables in predicting the presence of weight gain at discharge with a random forest classification (AUROC:0.847). The AUROC of NNST-Plus, which was improved by adding LOS, PN, PNA, surgery, and sodium to NNST, increased by 16.5%. In addition, weight at admission, LOS, gestation-adjusted age at admission (> 40 weeks), sex, gestational age, birth weight, PNA, SGA, complications of labor and delivery, multiple birth, serum creatinine, and PN treatment were the most important variables in predicting discharge weight with an elastic net regression (R2 = 0.748). This is the first study on the early prediction of EUGR with promising clinical performance based on ML algorithms. It is estimated that the incidence of EUGR can be improved with the implementation of this ML-based web tool (http://www.softmed.hacettepe.edu.tr/NEO-DEER/) in clinical practice.

Bringing Machine Learning Systems into Clinical Practice: A Design Science Approach to Explainable Machine Learning-Based Clinical Decision Support Systems

Clinical decision support systems (CDSSs) based on machine learning (ML) hold great promise for improving medical care. Technically, such CDSSs are already feasible but physicians have been skeptical about their application. In particular, their opacity is a major concern, as it may lead physicians to overlook erroneous outputs from ML-based CDSSs, potentially causing serious consequences for patients. Research on explainable AI (XAI) offers methods with the potential to increase the explainability of black-box ML systems. This could significantly accelerate the application of MLbased CDSSs in medicine. However, XAI research to date has mainly been technically driven and largely neglects the needs of end users. To better engage the users of ML-based CDSSs, we applied a design science approach to develop a design for explainable ML-based CDSSs that incorporates insights from XAI literature while simultaneously addressing physicians’ needs. This design comprises five design principles that designers of ML-based CDSSs can apply to implement user-centered explanations, which are instantiated in a prototype of an explainable ML-based CDSS for lung nodule classification. We rooted the design principles and the derived prototype in a body of justificatory knowledge consisting of XAI literature, the concept of usability, and an online survey study involving 57 physicians. We refined the design principles and their instantiation by conducting walk-throughs with six radiologists. A final experiment with 45 radiologists demonstrated that our design resulted in physicians perceiving the ML-based CDSS as more explainable and usable in terms of the required cognitive effort than a system without explanations.

Machine Learning-Based Clinical Decision Support System for Automatic Diagnosis of COVID-19 based on Clinical Data

Introduction: Needless to say that correct and real-time detection and effective prognosis of the COVID-19 are necessary to deliver the best possible care for patients and, accordingly, diminish the pressure on the healthcare industries. Hence our paper aims to present an intelligent algorithm for selecting the best features from the dataset and developing Machine Learning(ML) based models to predict the COVID-19 and finally opted for the best-performing algorithm.
 Methods: In this developmental study, the clinical data of 1703 COVID-19 and non-COVID-19 patients Using a single-center registry from February 9, 2020, to December 20, 2020, were used. The Minimum Redundancy Maximum Relevance (mRMR) feature selection algorithm identified the most relevant variables. Then, chosen features feed into the several data mining methods, including K-Nearest Neighbors, AdaBoost Classifier, Decision Tree, HistGradient Boosting Classifier, and Support Vector Machine. A 10-fold cross-validation method and six performance evaluation metrics were used to evaluate and compare these implemented algorithms, and finally, the best model was implemented.
 Results: Out of the 34 included features, 11 variables were selected as the essential features. The results of using ML algorithms indicated that the best performance belongs to the AdaBoost classifier with mean accuracy = 92.9%, mean specificity = 89.3%, mean sensitivity = 94.2%, mean F-measure = 91.6 %, mean KAPA = 94.3% and mean ROC = 92.1 %.
 Conclusion: The empirical results reveal that the Adaboost model yielded higher performance than other classification models and developed our Clinical Decision Support Systems (CDSS) interface to discriminate positive COVID-19 from negative cases.

Should We Rely on AI to Help Avoid Bias in Patient Selection for Major Surgery?

Many regard iatrogenic injuries as consequences of diagnosis or intervention actions. But inaction-not offering indicated major surgery-can also result in iatrogenic injury. This article explores some surgeons' overestimations of operative risk based on patients' race and socioeconomic status as unduly influential in their decisions about whether to perform major cancer or cardiac surgery on some patients with appropriate clinical indications. This article also considers artificial intelligence and machine learning-based clinical decision support systems that might offer more accurate, individualized risk assessment that could make patient selection processes more equitable, thereby mitigating racial and ethnic inequity in cancer and cardiac disease.

Factors driving provider adoption of the TREWS machine learning-based early warning system and its effects on sepsis treatment timing.

Machine learning-based clinical decision support tools for sepsis create opportunities to identify at-risk patients and initiate treatments at early time points, which is critical for improving sepsis outcomes. In view of the increasing use of such systems, better understanding of how they are adopted and used by healthcare providers is needed. Here, we analyzed provider interactions with a sepsis early detection tool (Targeted Real-time Early Warning System), which was deployed at five hospitals over a 2-year period. Among 9,805 retrospectively identified sepsis cases, the early detection tool achieved high sensitivity (82% of sepsis cases were identified) and a high rate of adoption: 89% of all alerts by the system were evaluated by a physician or advanced practice provider and 38% of evaluated alerts were confirmed by a provider. Adjusting for patient presentation and severity, patients with sepsis whose alert was confirmed by a provider within 3 h had a 1.85-h (95% CI 1.66-2.00) reduction in median time to first antibiotic order compared to patients with sepsis whose alert was either dismissed, confirmed more than 3 h after the alert or never addressed in the system. Finally, we found that emergency department providers and providers who had previous interactions with an alert were more likely to interact with alerts, as well as to confirm alerts on retrospectively identified patients with sepsis. Beyond efforts to improve the performance of early warning systems, efforts to improve adoption are essential to their clinical impact and should focus on understanding providers' knowledge of, experience with and attitudes toward such systems.

Personalized antibiograms for machine learning driven antibiotic selection

BackgroundThe Centers for Disease Control and Prevention identify antibiotic prescribing stewardship as the most important action to combat increasing antibiotic resistance. Clinicians balance broad empiric antibiotic coverage vs. precision coverage targeting only the most likely pathogens. We investigate the utility of machine learning-based clinical decision support for antibiotic prescribing stewardship.MethodsIn this retrospective multi-site study, we developed machine learning models that predict antibiotic susceptibility patterns (personalized antibiograms) using electronic health record data of 8342 infections from Stanford emergency departments and 15,806 uncomplicated urinary tract infections from Massachusetts General Hospital and Brigham & Women’s Hospital in Boston. We assessed the trade-off between broad-spectrum and precise antibiotic prescribing using linear programming.ResultsWe find in Stanford data that personalized antibiograms reallocate clinician antibiotic selections with a coverage rate (fraction of infections covered by treatment) of 85.9%; similar to clinician performance (84.3% p = 0.11). In the Boston dataset, the personalized antibiograms coverage rate is 90.4%; a significant improvement over clinicians (88.1% p < 0.0001). Personalized antibiograms achieve similar coverage to the clinician benchmark with narrower antibiotics. With Stanford data, personalized antibiograms maintain clinician coverage rates while narrowing 69% of empiric vancomycin+piperacillin/tazobactam prescriptions to piperacillin/tazobactam. In the Boston dataset, personalized antibiograms maintain clinician coverage rates while narrowing 48% of ciprofloxacin to trimethoprim/sulfamethoxazole.ConclusionsPrecision empiric antibiotic prescribing with personalized antibiograms could improve patient safety and antibiotic stewardship by reducing unnecessary use of broad-spectrum antibiotics that breed a growing tide of resistant organisms.

Evaluation of domain generalization and adaptation on improving model robustness to temporal dataset shift in clinical medicine

Temporal dataset shift associated with changes in healthcare over time is a barrier to deploying machine learning-based clinical decision support systems. Algorithms that learn robust models by estimating invariant properties across time periods for domain generalization (DG) and unsupervised domain adaptation (UDA) might be suitable to proactively mitigate dataset shift. The objective was to characterize the impact of temporal dataset shift on clinical prediction models and benchmark DG and UDA algorithms on improving model robustness. In this cohort study, intensive care unit patients from the MIMIC-IV database were categorized by year groups (2008–2010, 2011–2013, 2014–2016 and 2017–2019). Tasks were predicting mortality, long length of stay, sepsis and invasive ventilation. Feedforward neural networks were used as prediction models. The baseline experiment trained models using empirical risk minimization (ERM) on 2008–2010 (ERM[08–10]) and evaluated them on subsequent year groups. DG experiment trained models using algorithms that estimated invariant properties using 2008–2016 and evaluated them on 2017–2019. UDA experiment leveraged unlabelled samples from 2017 to 2019 for unsupervised distribution matching. DG and UDA models were compared to ERM[08–16] models trained using 2008–2016. Main performance measures were area-under-the-receiver-operating-characteristic curve (AUROC), area-under-the-precision-recall curve and absolute calibration error. Threshold-based metrics including false-positives and false-negatives were used to assess the clinical impact of temporal dataset shift and its mitigation strategies. In the baseline experiments, dataset shift was most evident for sepsis prediction (maximum AUROC drop, 0.090; 95% confidence interval (CI), 0.080–0.101). Considering a scenario of 100 consecutively admitted patients showed that ERM[08–10] applied to 2017–2019 was associated with one additional false-negative among 11 patients with sepsis, when compared to the model applied to 2008–2010. When compared with ERM[08–16], DG and UDA experiments failed to produce more robust models (range of AUROC difference, − 0.003 to 0.050). In conclusion, DG and UDA failed to produce more robust models compared to ERM in the setting of temporal dataset shift. Alternate approaches are required to preserve model performance over time in clinical medicine.

An explainable machine learning-based clinical decision support system for prediction of gestational diabetes mellitus

Gestational Diabetes Mellitus (GDM), a common pregnancy complication associated with many maternal and neonatal consequences, is increased in mothers with overweight and obesity. Interventions initiated early in pregnancy can reduce the rate of GDM in these women, however, untargeted interventions can be costly and time-consuming. We have developed an explainable machine learning-based clinical decision support system (CDSS) to identify at-risk women in need of targeted pregnancy intervention. Maternal characteristics and blood biomarkers at baseline from the PEARS study were used. After appropriate data preparation, synthetic minority oversampling technique and feature selection, five machine learning algorithms were applied with five-fold cross-validated grid search optimising the balanced accuracy. Our models were explained with Shapley additive explanations to increase the trustworthiness and acceptability of the system. We developed multiple models for different use cases: theoretical (AUC-PR 0.485, AUC-ROC 0.792), GDM screening during a normal antenatal visit (AUC-PR 0.208, AUC-ROC 0.659), and remote GDM risk assessment (AUC-PR 0.199, AUC-ROC 0.656). Our models have been implemented as a web server that is publicly available for academic use. Our explainable CDSS demonstrates the potential to assist clinicians in screening at risk patients who may benefit from early pregnancy GDM prevention strategies.

What is morally at stake when using algorithms to make medical diagnoses? Expanding the discussion beyond risks and harms

In this paper, we examine the qualitative moral impact of machine learning-based clinical decision support systems in the process of medical diagnosis. To date, discussions about machine learning in this context have focused on problems that can be measured and assessed quantitatively, such as by estimating the extent of potential harm or calculating incurred risks. We maintain that such discussions neglect the qualitative moral impact of these technologies. Drawing on the philosophical approaches of technomoral change and technological mediation theory, which explore the interplay between technologies and morality, we present an analysis of concerns related to the adoption of machine learning-aided medical diagnosis. We analyze anticipated moral issues that machine learning systems pose for different stakeholders, such as bias and opacity in the way that models are trained to produce diagnoses, changes to how health care providers, patients, and developers understand their roles and professions, and challenges to existing forms of medical legislation. Albeit preliminary in nature, the insights offered by the technomoral change and the technological mediation approaches expand and enrich the current discussion about machine learning in diagnostic practices, bringing distinct and currently underexplored areas of concern to the forefront. These insights can contribute to a more encompassing and better informed decision-making process when adapting machine learning techniques to medical diagnosis, while acknowledging the interests of multiple stakeholders and the active role that technologies play in generating, perpetuating, and modifying ethical concerns in health care.

Enhancing the value to users of machine learning-based clinical decision support tools: A framework for iterative, collaborative development and implementation.

Health care organizations are integrating a variety of machine learning (ML)-based clinical decision support (CDS) tools into their operations, but practitioners lack clear guidance regarding how to implement these tools so that they assist end users in their work. We designed this study to identify how health care organizations can facilitate collaborative development of ML-based CDS tools to enhance their value for health care delivery in real-world settings. We utilized qualitative methods, including 37 interviews in a large, multispecialty health system that developed and implemented two operational ML-based CDS tools in two of its hospital sites. We performed thematic analyses to inform presentation of an explanatory framework and recommendations. We found that ML-based CDS tool development and implementation into clinical workflows proceeded in four phases: iterative solution coidentification, iterative coengagement, iterative coapplication, and iterative corefinement. Each phase is characterized by a collaborative back-and-forth process between the technology's developers and users, through which both users' activities and the technology itself are transformed. Health care organizations that anticipate iterative collaboration to be an integral aspect of their ML-based CDS tools' development and implementation process may have more success in deploying ML-based CDS tools that assist end users in their work than organizations that expect a traditional technology innovation process. Managers developing and implementing ML-based CDS tools should frame the work as a collaborative learning opportunity for both users and the technology itself and should solicit constructive feedback from users on potential changes to the technology, in addition to potential changes to user workflows, in an ongoing, iterative manner.