Ethical Machine Learning in Healthcare

Causal fairness assessment of treatment allocation with electronic health records

OBJECTIVE

Healthcare continues to grapple with the persistent issue of treatment disparities, sparking concerns regarding the equitable allocation of treatments in clinical practice. While various fairness metrics have emerged to assess fairness in decision-making processes, a growing focus has been on causality-based fairness concepts due to their capacity to mitigate confounding effects and reason about bias. However, the application of causal fairness notions in evaluating the fairness of clinical decision-making with electronic health record (EHR) data remains an understudied domain. This study aims to address the methodological gap in assessing causal fairness of treatment allocation with electronic health records data. In addition, we investigate the impact of social determinants of health on the assessment of causal fairness of treatment allocation.

METHODS

We propose a causal fairness algorithm to assess fairness in clinical decision-making. Our algorithm accounts for the heterogeneity of patient populations and identifies potential unfairness in treatment allocation by conditioning on patients who have the same likelihood to benefit from the treatment. We apply this framework to a patient cohort with coronary artery disease derived from an EHR database to evaluate the fairness of treatment decisions.

RESULTS

Our analysis reveals notable disparities in coronary artery bypass grafting (CABG) allocation among different patient groups. Women were found to be 4.4%-7.7% less likely to receive CABG than men in two out of four treatment response strata. Similarly, Black or African American patients were 5.4%-8.7% less likely to receive CABG than others in three out of four response strata. These results were similar when social determinants of health (insurance and area deprivation index) were dropped from the algorithm. These findings highlight the presence of disparities in treatment allocation among similar patients, suggesting potential unfairness in the clinical decision-making process.

CONCLUSION

This study introduces a novel approach for assessing the fairness of treatment allocation in healthcare. By incorporating responses to treatment into fairness framework, our method explores the potential of quantifying fairness from a causal perspective using EHR data. Our research advances the methodological development of fairness assessment in healthcare and highlight the importance of causality in determining treatment fairness.

Artificial Intelligence in Cardiovascular Care-Part 2: Applications: JACC Review Topic of the Week

Recent artificial intelligence (AI) advancements in cardiovascular care offer potential enhancements in effective diagnosis, treatment, and outcomes. More than 600 U.S. Food and Drug Administration-approved clinical AI algorithms now exist, with 10% focusing on cardiovascular applications, highlighting the growing opportunities for AI to augment care. This review discusses the latest advancements in the field of AI, with a particular focus on the utilization of multimodal inputs and the field of generative AI. Further discussions in this review involve an approach to understanding the larger context in which AI-augmented care may exist, and include a discussion of the need for rigorous evaluation, appropriate infrastructure for deployment, ethics and equity assessments, regulatory oversight, and viable business cases for deployment. Embracing this rapidly evolving technology while setting an appropriately high evaluation benchmark with careful and patient-centered implementation will be crucial for cardiology to leverage AI to enhance patient care and the provider experience.

Enhancing psychiatric rehabilitation outcomes through a multimodal multitask learning model based on BERT and TabNet: An approach for personalized treatment and improved decision-making

Evaluating the rehabilitation status of individuals with serious mental illnesses (SMI) necessitates a comprehensive analysis of multimodal data, including unstructured text records and structured diagnostic data. However, progress in the effective assessment of rehabilitation status remains limited. Our study develops a deep learning model integrating Bidirectional Encoder Representations from Transformers (BERT) and TabNet through a late fusion strategy to enhance rehabilitation prediction, including referral risk, dangerous behaviors, self-awareness, and medication adherence, in patients with SMI. BERT processes unstructured textual data, such as doctor's notes, whereas TabNet manages structured diagnostic information. The model's interpretability function serves to assist healthcare professionals in understanding the model's predictive decisions, improving patient care. Our model exhibited excellent predictive performance for all four tasks, with an accuracy exceeding 0.78 and an area under the curve of 0.70. In addition, a series of tests proved the model's robustness, fairness, and interpretability. This study combines multimodal and multitask learning strategies into a model and applies it to rehabilitation assessment tasks, offering a promising new tool that can be seamlessly integrated with the clinical workflow to support the provision of optimized patient care.

The Risks and Challenges of Artificial Intelligence in Endocrinology

Artificial intelligence (AI) holds the promise of addressing many of the numerous challenges healthcare faces, which include a growing burden of illness, an increase in chronic health conditions and disabilities due to aging and epidemiological changes, higher demand for health services, overworked and burned-out clinicians, greater societal expectations, and rising health expenditures. While technological advancements in processing power, memory, storage, and the abundance of data have empowered computers to handle increasingly complex tasks with remarkable success, AI introduces a variety of meaningful risks and challenges. Among these are issues related to accuracy and reliability, bias and equity, errors and accountability, transparency, misuse, and privacy of data. As AI systems continue to rapidly integrate into healthcare settings, it is crucial to recognize the inherent risks they bring. These risks demand careful consideration to ensure the responsible and safe deployment of AI in healthcare.

Variation in monitoring: Glucose measurement in the ICU as a case study to preempt spurious correlations

OBJECTIVE

Health inequities can be influenced by demographic factors such as race and ethnicity, proficiency in English, and biological sex. Disparities may manifest as differential likelihood of testing which correlates directly with the likelihood of an intervention to address an abnormal finding. Our retrospective observational study evaluated the presence of variation in glucose measurements in the Intensive Care Unit (ICU).

METHODS

Using the MIMIC-IV database (2008-2019), a single-center, academic referral hospital in Boston (USA), we identified adult patients meeting sepsis-3 criteria. Exclusion criteria were diabetic ketoacidosis, ICU length of stay under 1 day, and unknown race or ethnicity. We performed a logistic regression analysis to assess differential likelihoods of glucose measurements on day 1. A negative binomial regression was fitted to assess the frequency of subsequent glucose readings. Analyses were adjusted for relevant clinical confounders, and performed across three disparity proxy axes: race and ethnicity, sex, and English proficiency.

RESULTS

We studied 24,927 patients, of which 19.5% represented racial and ethnic minority groups, 42.4% were female, and 9.8% had limited English proficiency. No significant differences were found for glucose measurement on day 1 in the ICU. This pattern was consistent irrespective of the axis of analysis, i.e. race and ethnicity, sex, or English proficiency. Conversely, subsequent measurement frequency revealed potential disparities. Specifically, males (incidence rate ratio (IRR) 1.06, 95% confidence interval (CI) 1.01 - 1.21), patients who identify themselves as Hispanic (IRR 1.11, 95% CI 1.01 - 1.21), or Black (IRR 1.06, 95% CI 1.01 - 1.12), and patients being English proficient (IRR 1.08, 95% CI 1.01 - 1.15) had higher chances of subsequent glucose readings.

CONCLUSION

We found disparities in ICU glucose measurements among patients with sepsis, albeit the magnitude was small. Variation in disease monitoring is a source of data bias that may lead to spurious correlations when modeling health data.

The Importance of Conducting Practice-oriented Research with Underserved Populations

There has been a growing emphasis on dissemination of empirically supported treatments. Dissemination, however, should not be restricted to treatment. It can and, in the spirit of the scientific-practitioner model, should also involve research. Because it focuses on the investigation of clinical routine as it takes place in local settings and because it can involve the collaboration of several stakeholders, practice-oriented research (POR) can be viewed as an optimal research method to be disseminated. POR has the potential of addressing particularly relevant gaps of knowledge and action when implemented in regions of the world that have limited resources for or experiences with empirical research, and/or in clinical settings that are serving clinical populations who are not typically receiving optimal mental care services - specifically, individuals in rural and inner cities that have limited economic and social resources. The establishment and maintenance of POR in such regions and/or settings, however, come with specific obstacles and challenges. Integrating the experiences acquired from research conducted in various continents (Africa, Europe, Latin America, and North America), the goal of this paper is to describe some of these challenges, strategies that have been implemented to address them, as well as new possible directions to facilitate the creation and growth of POR. It also describes how these challenges and ways to deal with them can provide helpful lessons for already existing POR infrastructures.

Current perspectives on the use of artificial intelligence in critical patient safety

Intensive Care Units (ICUs) have undergone enhancements in patient safety, and artificial intelligence (AI) emerges as a disruptive technology offering novel opportunities. While the published evidence is limited and presents methodological issues, certain areas show promise, such as decision support systems, detection of adverse events, and prescription error identification. The application of AI in safety may pursue predictive or diagnostic objectives. Implementing AI-based systems necessitates procedures to ensure secure assistance, addressing challenges including trust in such systems, biases, data quality, scalability, and ethical and confidentiality considerations. The development and application of AI demand thorough testing, encompassing retrospective data assessments, real-time validation with prospective cohorts, and efficacy demonstration in clinical trials. Algorithmic transparency and explainability are essential, with active involvement of clinical professionals being crucial in the implementation process.

Evaluating Algorithmic Bias in 30-Day Hospital Readmission Models: Retrospective Analysis

BACKGROUND

The adoption of predictive algorithms in health care comes with the potential for algorithmic bias, which could exacerbate existing disparities. Fairness metrics have been proposed to measure algorithmic bias, but their application to real-world tasks is limited.

OBJECTIVE

This study aims to evaluate the algorithmic bias associated with the application of common 30-day hospital readmission models and assess the usefulness and interpretability of selected fairness metrics.

METHODS

We used 10.6 million adult inpatient discharges from Maryland and Florida from 2016 to 2019 in this retrospective study. Models predicting 30-day hospital readmissions were evaluated: LACE Index, modified HOSPITAL score, and modified Centers for Medicare & Medicaid Services (CMS) readmission measure, which were applied as-is (using existing coefficients) and retrained (recalibrated with 50% of the data). Predictive performances and bias measures were evaluated for all, between Black and White populations, and between low- and other-income groups. Bias measures included the parity of false negative rate (FNR), false positive rate (FPR), 0-1 loss, and generalized entropy index. Racial bias represented by FNR and FPR differences was stratified to explore shifts in algorithmic bias in different populations.

RESULTS

The retrained CMS model demonstrated the best predictive performance (area under the curve: 0.74 in Maryland and 0.68-0.70 in Florida), and the modified HOSPITAL score demonstrated the best calibration (Brier score: 0.16-0.19 in Maryland and 0.19-0.21 in Florida). Calibration was better in White (compared to Black) populations and other-income (compared to low-income) groups, and the area under the curve was higher or similar in the Black (compared to White) populations. The retrained CMS and modified HOSPITAL score had the lowest racial and income bias in Maryland. In Florida, both of these models overall had the lowest income bias and the modified HOSPITAL score showed the lowest racial bias. In both states, the White and higher-income populations showed a higher FNR, while the Black and low-income populations resulted in a higher FPR and a higher 0-1 loss. When stratified by hospital and population composition, these models demonstrated heterogeneous algorithmic bias in different contexts and populations.

CONCLUSIONS

Caution must be taken when interpreting fairness measures' face value. A higher FNR or FPR could potentially reflect missed opportunities or wasted resources, but these measures could also reflect health care use patterns and gaps in care. Simply relying on the statistical notions of bias could obscure or underplay the causes of health disparity. The imperfect health data, analytic frameworks, and the underlying health systems must be carefully considered. Fairness measures can serve as a useful routine assessment to detect disparate model performances but are insufficient to inform mechanisms or policy changes. However, such an assessment is an important first step toward data-driven improvement to address existing health disparities.

Inherent Bias in Electronic Health Records: A Scoping Review of Sources of Bias

Objectives

1.1Biases inherent in electronic health records (EHRs), and therefore in medical artificial intelligence (AI) models may significantly exacerbate health inequities and challenge the adoption of ethical and responsible AI in healthcare. Biases arise from multiple sources, some of which are not as documented in the literature. Biases are encoded in how the data has been collected and labeled, by implicit and unconscious biases of clinicians, or by the tools used for data processing. These biases and their encoding in healthcare records undermine the reliability of such data and bias clinical judgments and medical outcomes. Moreover, when healthcare records are used to build data-driven solutions, the biases are further exacerbated, resulting in systems that perpetuate biases and induce healthcare disparities. This literature scoping review aims to categorize the main sources of biases inherent in EHRs.

Methods

1.2We queried PubMed and Web of Science on January 19th, 2023, for peer-reviewed sources in English, published between 2016 and 2023, using the PRISMA approach to stepwise scoping of the literature. To select the papers that empirically analyze bias in EHR, from the initial yield of 430 papers, 27 duplicates were removed, and 403 studies were screened for eligibility. 196 articles were removed after the title and abstract screening, and 96 articles were excluded after the full-text review resulting in a final selection of 116 articles.

Results

1.3Systematic categorizations of diverse sources of bias are scarce in the literature, while the effects of separate studies are often convoluted and methodologically contestable. Our categorization of published empirical evidence identified the six main sources of bias: a) bias arising from past clinical trials; b) data-related biases arising from missing, incomplete information or poor labeling of data; human-related bias induced by c) implicit clinician bias, d) referral and admission bias; e) diagnosis or risk disparities bias and finally, (f) biases in machinery and algorithms.

Conclusions

1.4Machine learning and data-driven solutions can potentially transform healthcare delivery, but not without limitations. The core inputs in the systems (data and human factors) currently contain several sources of bias that are poorly documented and analyzed for remedies. The current evidence heavily focuses on data-related biases, while other sources are less often analyzed or anecdotal. However, these different sources of biases add to one another exponentially. Therefore, to understand the issues holistically we need to explore these diverse sources of bias. While racial biases in EHR have been often documented, other sources of biases have been less frequently investigated and documented (e.g. gender-related biases, sexual orientation discrimination, socially induced biases, and implicit, often unconscious, human-related cognitive biases). Moreover, some existing studies lack causal evidence, illustrating the different prevalences of disease across groups, which does not per se prove the causality. Our review shows that data-, human- and machine biases are prevalent in healthcare and they significantly impact healthcare outcomes and judgments and exacerbate disparities and differential treatment. Understanding how diverse biases affect AI systems and recommendations is critical. We suggest that researchers and medical personnel should develop safeguards and adopt data-driven solutions with a "bias-in-mind" approach. More empirical evidence is needed to tease out the effects of different sources of bias on health outcomes.

Moral Engagement and Disengagement in Health Care AI Development

BACKGROUND

Machine learning (ML) is utilized increasingly in health care, and can pose harms to patients, clinicians, health systems, and the public. In response, regulators have proposed an approach that would shift more responsibility to ML developers for mitigating potential harms. To be effective, this approach requires ML developers to recognize, accept, and act on responsibility for mitigating harms. However, little is known regarding the perspectives of developers themselves regarding their obligations to mitigate harms.

METHODS

We conducted 40 semi-structured interviews with developers of ML predictive analytics applications for health care in the United States.

RESULTS

Participants varied widely in their perspectives on personal responsibility and included examples of both moral engagement and disengagement, albeit in a variety of forms. While most (70%) of participants made a statement indicative of moral engagement, most of these statements reflected an awareness of moral issues, while only a subset of these included additional elements of engagement such as recognizing responsibility, alignment with personal values, addressing conflicts of interests, and opportunities for action. Further, we identified eight distinct categories of moral disengagement reflecting efforts to minimize potential harms or deflect personal responsibility for preventing or mitigating harms.

CONCLUSIONS

These findings suggest possible facilitators and barriers to the development of ethical ML that could act by encouraging moral engagement or discouraging moral disengagement. Regulatory approaches that depend on the ability of ML developers to recognize, accept, and act on responsibility for mitigating harms might have limited success without education and guidance for ML developers about the extent of their responsibilities and how to implement them.

Application of machine learning for lung cancer survival prognostication-A systematic review and meta-analysis

Introduction

Machine learning (ML) techniques have gained increasing attention in the field of healthcare, including predicting outcomes in patients with lung cancer. ML has the potential to enhance prognostication in lung cancer patients and improve clinical decision-making. In this systematic review and meta-analysis, we aimed to evaluate the performance of ML models compared to logistic regression (LR) models in predicting overall survival in patients with lung cancer.

Methods

We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement. A comprehensive search was conducted in Medline, Embase, and Cochrane databases using a predefined search query. Two independent reviewers screened abstracts and conflicts were resolved by a third reviewer. Inclusion and exclusion criteria were applied to select eligible studies. Risk of bias assessment was performed using predefined criteria. Data extraction was conducted using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS) checklist. Meta-analytic analysis was performed to compare the discriminative ability of ML and LR models.

Results

The literature search resulted in 3,635 studies, and 12 studies with a total of 211,068 patients were included in the analysis. Six studies reported confidence intervals and were included in the meta-analysis. The performance of ML models varied across studies, with C-statistics ranging from 0.60 to 0.85. The pooled analysis showed that ML models had higher discriminative ability compared to LR models, with a weighted average C-statistic of 0.78 for ML models compared to 0.70 for LR models.

Conclusion

Machine learning models show promise in predicting overall survival in patients with lung cancer, with superior discriminative ability compared to logistic regression models. However, further validation and standardization of ML models are needed before their widespread implementation in clinical practice. Future research should focus on addressing the limitations of the current literature, such as potential bias and heterogeneity among studies, to improve the accuracy and generalizability of ML models for predicting outcomes in patients with lung cancer. Further research and development of ML models in this field may lead to improved patient outcomes and personalized treatment strategies.

Machine learning for healthcare that matters: Reorienting from technical novelty to equitable impact

Despite significant technical advances in machine learning (ML) over the past several years, the tangible impact of this technology in healthcare has been limited. This is due not only to the particular complexities of healthcare, but also due to structural issues in the machine learning for healthcare (MLHC) community which broadly reward technical novelty over tangible, equitable impact. We structure our work as a healthcare-focused echo of the 2012 paper "Machine Learning that Matters", which highlighted such structural issues in the ML community at large, and offered a series of clearly defined "Impact Challenges" to which the field should orient itself. Drawing on the expertise of a diverse and international group of authors, we engage in a narrative review and examine issues in the research background environment, training processes, evaluation metrics, and deployment protocols which act to limit the real-world applicability of MLHC. Broadly, we seek to distinguish between machine learning ON healthcare data and machine learning FOR healthcare-the former of which sees healthcare as merely a source of interesting technical challenges, and the latter of which regards ML as a tool in service of meeting tangible clinical needs. We offer specific recommendations for a series of stakeholders in the field, from ML researchers and clinicians, to the institutions in which they work, and the governments which regulate their data access.

Drop the shortcuts: image augmentation improves fairness and decreases AI detection of race and other demographics from medical images

BACKGROUND

It has been shown that AI models can learn race on medical images, leading to algorithmic bias. Our aim in this study was to enhance the fairness of medical image models by eliminating bias related to race, age, and sex. We hypothesise models may be learning demographics via shortcut learning and combat this using image augmentation.

METHODS

This study included 44,953 patients who identified as Asian, Black, or White (mean age, 60.68 years ±18.21; 23,499 women) for a total of 194,359 chest X-rays (CXRs) from MIMIC-CXR database. The included CheXpert images comprised 45,095 patients (mean age 63.10 years ±18.14; 20,437 women) for a total of 134,300 CXRs were used for external validation. We also collected 1195 3D brain magnetic resonance imaging (MRI) data from the ADNI database, which included 273 participants with an average age of 76.97 years ±14.22, and 142 females. DL models were trained on either non-augmented or augmented images and assessed using disparity metrics. The features learned by the models were analysed using task transfer experiments and model visualisation techniques.

FINDINGS

In the detection of radiological findings, training a model using augmented CXR images was shown to reduce disparities in error rate among racial groups (-5.45%), age groups (-13.94%), and sex (-22.22%). For AD detection, the model trained with augmented MRI images was shown 53.11% and 31.01% reduction of disparities in error rate among age and sex groups, respectively. Image augmentation led to a reduction in the model's ability to identify demographic attributes and resulted in the model trained for clinical purposes incorporating fewer demographic features.

INTERPRETATION

The model trained using the augmented images was less likely to be influenced by demographic information in detecting image labels. These results demonstrate that the proposed augmentation scheme could enhance the fairness of interpretations by DL models when dealing with data from patients with different demographic backgrounds.

FUNDING

National Science and Technology Council (Taiwan), National Institutes of Health.

Systemic lupus in the era of machine learning medicine

Artificial intelligence and machine learning applications are emerging as transformative technologies in medicine. With greater access to a diverse range of big datasets, researchers are turning to these powerful techniques for data analysis. Machine learning can reveal patterns and interactions between variables in large and complex datasets more accurately and efficiently than traditional statistical methods. Machine learning approaches open new possibilities for studying SLE, a multifactorial, highly heterogeneous and complex disease. Here, we discuss how machine learning methods are rapidly being integrated into the field of SLE research. Recent reports have focused on building prediction models and/or identifying novel biomarkers using both supervised and unsupervised techniques for understanding disease pathogenesis, early diagnosis and prognosis of disease. In this review, we will provide an overview of machine learning techniques to discuss current gaps, challenges and opportunities for SLE studies. External validation of most prediction models is still needed before clinical adoption. Utilisation of deep learning models, access to alternative sources of health data and increased awareness of the ethics, governance and regulations surrounding the use of artificial intelligence in medicine will help propel this exciting field forward.

Migraine headache (MH) classification using machine learning methods with data augmentation

Migraine headache, a prevalent and intricate neurovascular disease, presents significant challenges in its clinical identification. Existing techniques that use subjective pain intensity measures are insufficiently accurate to make a reliable diagnosis. Even though headaches are a common condition with poor diagnostic specificity, they have a significant negative influence on the brain, body, and general human function. In this era of deeply intertwined health and technology, machine learning (ML) has emerged as a crucial force in transforming every aspect of healthcare, utilizing advanced facilities ML has shown groundbreaking achievements related to developing classification and automatic predictors. With this, deep learning models, in particular, have proven effective in solving complex problems spanning computer vision and data analytics. Consequently, the integration of ML in healthcare has become vital, especially in developing countries where limited medical resources and lack of awareness prevail, the urgent need to forecast and categorize migraines using artificial intelligence (AI) becomes even more crucial. By training these models on a publicly available dataset, with and without data augmentation. This study focuses on leveraging state-of-the-art ML algorithms, including support vector machine (SVM), K-nearest neighbors (KNN), random forest (RF), decision tree (DST), and deep neural networks (DNN), to predict and classify various types of migraines. The proposed models with data augmentations were trained to classify seven various types of migraine. The proposed models with data augmentations were trained to classify seven various types of migraine. The revealed results show that DNN, SVM, KNN, DST, and RF achieved an accuracy of 99.66%, 94.60%, 97.10%, 88.20%, and 98.50% respectively with data augmentation highlighting the transformative potential of AI in enhancing migraine diagnosis.

Neuroimaging data repositories and AI-driven healthcare-Global aspirations vs. ethical considerations in machine learning models of neurological disease

Neuroimaging data repositories are data-rich resources comprising brain imaging with clinical and biomarker data. The potential for such repositories to transform healthcare is tremendous, especially in their capacity to support machine learning (ML) and artificial intelligence (AI) tools. Current discussions about the generalizability of such tools in healthcare provoke concerns of risk of bias-ML models underperform in women and ethnic and racial minorities. The use of ML may exacerbate existing healthcare disparities or cause post-deployment harms. Do neuroimaging data repositories and their capacity to support ML/AI-driven clinical discoveries, have both the potential to accelerate innovative medicine and harden the gaps of social inequities in neuroscience-related healthcare? In this paper, we examined the ethical concerns of ML-driven modeling of global community neuroscience needs arising from the use of data amassed within neuroimaging data repositories. We explored this in two parts; firstly, in a theoretical experiment, we argued for a South East Asian-based repository to redress global imbalances. Within this context, we then considered the ethical framework toward the inclusion vs. exclusion of the migrant worker population, a group subject to healthcare inequities. Secondly, we created a model simulating the impact of global variations in the presentation of anosmia risks in COVID-19 toward altering brain structural findings; we then performed a mini AI ethics experiment. In this experiment, we interrogated an actual pilot dataset (n = 17; 8 non-anosmic (47%) vs. 9 anosmic (53%) using an ML clustering model. To create the COVID-19 simulation model, we bootstrapped to resample and amplify the dataset. This resulted in three hypothetical datasets: (i) matched (n = 68; 47% anosmic), (ii) predominant non-anosmic (n = 66; 73% disproportionate), and (iii) predominant anosmic (n = 66; 76% disproportionate). We found that the differing proportions of the same cohorts represented in each hypothetical dataset altered not only the relative importance of key features distinguishing between them but even the presence or absence of such features. The main objective of our mini experiment was to understand if ML/AI methodologies could be utilized toward modelling disproportionate datasets, in a manner we term "AI ethics." Further work is required to expand the approach proposed here into a reproducible strategy.

Improving Prediction of Survival for Extremely Premature Infants Born at 23 to 29 Weeks Gestational Age in the Neonatal Intensive Care Unit: Development and Evaluation of Machine Learning Models

BACKGROUND

Infants born at extremely preterm gestational ages are typically admitted to the neonatal intensive care unit (NICU) after initial resuscitation. The subsequent hospital course can be highly variable, and despite counseling aided by available risk calculators, there are significant challenges with shared decision-making regarding life support and transition to end-of-life care. Improving predictive models can help providers and families navigate these unique challenges.

OBJECTIVE

Machine learning methods have previously demonstrated added predictive value for determining intensive care unit outcomes, and their use allows consideration of a greater number of factors that potentially influence newborn outcomes, such as maternal characteristics. Machine learning-based models were analyzed for their ability to predict the survival of extremely preterm neonates at initial admission.

METHODS

Maternal and newborn information was extracted from the health records of infants born between 23 and 29 weeks of gestation in the Medical Information Mart for Intensive Care III (MIMIC-III) critical care database. Applicable machine learning models predicting survival during the initial NICU admission were developed and compared. The same type of model was also examined using only features that would be available prepartum for the purpose of survival prediction prior to an anticipated preterm birth. Features most correlated with the predicted outcome were determined when possible for each model.

RESULTS

Of included patients, 37 of 459 (8.1%) expired. The resulting random forest model showed higher predictive performance than the frequently used Score for Neonatal Acute Physiology With Perinatal Extension II (SNAPPE-II) NICU model when considering extremely preterm infants of very low birth weight. Several other machine learning models were found to have good performance but did not show a statistically significant difference from previously available models in this study. Feature importance varied by model, and those of greater importance included gestational age; birth weight; initial oxygenation level; elements of the APGAR (appearance, pulse, grimace, activity, and respiration) score; and amount of blood pressure support. Important prepartum features also included maternal age, steroid administration, and the presence of pregnancy complications.

CONCLUSIONS

Machine learning methods have the potential to provide robust prediction of survival in the context of extremely preterm births and allow for consideration of additional factors such as maternal clinical and socioeconomic information. Evaluation of larger, more diverse data sets may provide additional clarity on comparative performance.

Factors influencing clinician and patient interaction with machine learning-based risk prediction models: a systematic review

Machine learning (ML)-based risk prediction models hold the potential to support the health-care setting in several ways; however, use of such models is scarce. We aimed to review health-care professional (HCP) and patient perceptions of ML risk prediction models in published literature, to inform future risk prediction model development. Following database and citation searches, we identified 41 articles suitable for inclusion. Article quality varied with qualitative studies performing strongest. Overall, perceptions of ML risk prediction models were positive. HCPs and patients considered that models have the potential to add benefit in the health-care setting. However, reservations remain; for example, concerns regarding data quality for model development and fears of unintended consequences following ML model use. We identified that public views regarding these models might be more negative than HCPs and that concerns (eg, extra demands on workload) were not always borne out in practice. Conclusions are tempered by the low number of patient and public studies, the absence of participant ethnic diversity, and variation in article quality. We identified gaps in knowledge (particularly views from under-represented groups) and optimum methods for model explanation and alerts, which require future research.

Machine and deep learning models for accurate detection of ischemia and scar with myocardial blood flow positron emission tomography imaging

BACKGROUND

Quantification of myocardial blood flow (MBF) is used for the noninvasive diagnosis of patients with coronary artery disease (CAD). This study compared traditional statistics, machine learning, and deep learning techniques in their ability to diagnose disease using only the rest and stress MBF values.

METHODS

This study included 3245 rest and stress rubidium-82 positron emission tomography (PET) studies and matching diagnostic labels from perfusion reports. Standard logistic regression, lasso logistic regression, support vector machine, random forest, multilayer perceptron, and dense U-Net were compared for per-patient detection and per-vessel localization of scars and ischemia.

RESULTS

Receiver-operator characteristic area under the curve (AUC) of machine learning models was significantly higher than those of traditional statistics models for per-patient detection of disease (0.92-0.95 vs. 0.87) but not for per-vessel localization of ischemia or scar. Random forest showed the highest AUC = 0.95 among the different models compared. On the final hold-out set for generalizability, random forest showed an AUC of 0.92 for detection and 0.89 for localization of perfusion abnormalities.

CONCLUSIONS

For per-vessel localization, simple models trained on segmental data performed similarly to a convolutional neural network trained on polar-map data, highlighting the need to justify the use of complex predictive algorithms through comparison with simpler methods.

Arrhythmia classification for non-experts using infinite impulse response (IIR)-filter-based machine learning and deep learning models of the electrocardiogram

Arrhythmias are a leading cause of cardiovascular morbidity and mortality. Portable electrocardiogram (ECG) monitors have been used for decades to monitor patients with arrhythmias. These monitors provide real-time data on cardiac activity to identify irregular heartbeats. However, rhythm monitoring and wave detection, especially in the 12-lead ECG, make it difficult to interpret the ECG analysis by correlating it with the condition of the patient. Moreover, even experienced practitioners find ECG analysis challenging. All of this is due to the noise in ECG readings and the frequencies at which the noise occurs. The primary objective of this research is to remove noise and extract features from ECG signals using the proposed infinite impulse response (IIR) filter to improve ECG quality, which can be better understood by non-experts. For this purpose, this study used ECG signal data from the Massachusetts Institute of Technology Beth Israel Hospital (MIT-BIH) database. This allows the acquired data to be easily evaluated using machine learning (ML) and deep learning (DL) models and classified as rhythms. To achieve accurate results, we applied hyperparameter (HP)-tuning for ML classifiers and fine-tuning (FT) for DL models. This study also examined the categorization of arrhythmias using different filters and the changes in accuracy. As a result, when all models were evaluated, DenseNet-121 without FT achieved 99% accuracy, while FT showed better results with 99.97% accuracy.

Machine learning and deep learning-based approach in smart healthcare: Recent advances, applications, challenges and opportunities

In recent years, machine learning (ML) and deep learning (DL) have been the leading approaches to solving various challenges, such as disease predictions, drug discovery, medical image analysis, etc., in intelligent healthcare applications. Further, given the current progress in the fields of ML and DL, there exists the promising potential for both to provide support in the realm of healthcare. This study offered an exhaustive survey on ML and DL for the healthcare system, concentrating on vital state of the art features, integration benefits, applications, prospects and future guidelines. To conduct the research, we found the most prominent journal and conference databases using distinct keywords to discover scholarly consequences. First, we furnished the most current along with cutting-edge progress in ML-DL-based analysis in smart healthcare in a compendious manner. Next, we integrated the advancement of various services for ML and DL, including ML-healthcare, DL-healthcare, and ML-DL-healthcare. We then offered ML and DL-based applications in the healthcare industry. Eventually, we emphasized the research disputes and recommendations for further studies based on our observations.

Peer review of clinical and translational research manuscripts: Perspectives from statistical collaborators

Research articles in the clinical and translational science literature commonly use quantitative data to inform evaluation of interventions, learn about the etiology of disease, or develop methods for diagnostic testing or risk prediction of future events. The peer review process must evaluate the methodology used therein, including use of quantitative statistical methods. In this manuscript, we provide guidance for peer reviewers tasked with assessing quantitative methodology, intended to complement guidelines and recommendations that exist for manuscript authors. We describe components of clinical and translational science research manuscripts that require assessment including study design and hypothesis evaluation, sampling and data acquisition, interventions (for studies that include an intervention), measurement of data, statistical analysis methods, presentation of the study results, and interpretation of the study results. For each component, we describe what reviewers should look for and assess; how reviewers should provide helpful comments for fixable errors or omissions; and how reviewers should communicate uncorrectable and irreparable errors. We then discuss the critical concepts of transparency and acceptance/revision guidelines when communicating with responsible journal editors.

Artificial intelligence in the neonatal intensive care unit: the time is now

Artificial intelligence (AI) has the potential to revolutionize the neonatal intensive care unit (NICU) care by leveraging the large-scale, high-dimensional data that are generated by NICU patients. There is an emerging recognition that the confluence of technological progress, commercialization pathways, and rich data sets provides a unique opportunity for AI to make a lasting impact on the NICU. In this perspective article, we discuss four broad categories of AI applications in the NICU: imaging interpretation, prediction modeling of electronic health record data, integration of real-time monitoring data, and documentation and billing. By enhancing decision-making, streamlining processes, and improving patient outcomes, AI holds the potential to transform the quality of care for vulnerable newborns, making the excitement surrounding AI advancements well-founded and the potential for significant positive change stronger than ever before.

Quantifying and reducing inequity in average treatment effect estimation

BACKGROUND

Across studies of average treatment effects, some population subgroups consistently have lower representation than others which can lead to discrepancies in how well results generalize.

METHODS

We develop a framework for quantifying inequity due to systemic disparities in sample representation and a method for mitigation during data analysis. Assuming subgroup treatment effects are exchangeable, an unbiased sample average treatment effect estimator will have lower mean-squared error, on average across studies, for subgroups with less representation when treatment effects vary. We present a method for estimating average treatment effects in representation-adjusted samples which enables subgroups to optimally leverage information from the full sample rather than only their own subgroup's data. Two approaches for specifying representation adjustment are offered-one minimizes average mean-squared error for each subgroup separately and the other balances minimization of mean-squared error and equal representation. We conduct simulation studies to compare the performance of the proposed estimators to several subgroup-specific estimators.

RESULTS

We find that the proposed estimators generally provide lower mean squared error, particularly for smaller subgroups, relative to the other estimators. As a case study, we apply this method to a subgroup analysis from a published study.

CONCLUSIONS

We recommend the use of the proposed estimators to mitigate the impact of disparities in representation, though structural change is ultimately needed.

Ethical Considerations for Artificial Intelligence in Medical Imaging: Data Collection, Development, and Evaluation

The development of artificial intelligence (AI) within nuclear imaging involves several ethically fraught components at different stages of the machine learning pipeline, including during data collection, model training and validation, and clinical use. Drawing on the traditional principles of medical and research ethics, and highlighting the need to ensure health justice, the AI task force of the Society of Nuclear Medicine and Molecular Imaging has identified 4 major ethical risks: privacy of data subjects, data quality and model efficacy, fairness toward marginalized populations, and transparency of clinical performance. We provide preliminary recommendations to developers of AI-driven medical devices for mitigating the impact of these risks on patients and populations.

Call for algorithmic fairness to mitigate amplification of racial biases in artificial intelligence models used in orthodontics and craniofacial health

Machine Learning (ML), a subfield of Artificial Intelligence (AI), is being increasingly used in Orthodontics and craniofacial health for predicting clinical outcomes. Current ML/AI models are prone to accentuate racial disparities. The objective of this narrative review is to provide an overview of how AI/ML models perpetuate racial biases and how we can mitigate this situation. A narrative review of articles published in the medical literature on racial biases and the use of AI/ML models was undertaken. Current AI/ML models are built on homogenous clinical datasets that have a gross underrepresentation of historically disadvantages demographic groups, especially the ethno-racial minorities. The consequence of such AI/ML models is that they perform poorly when deployed on ethno-racial minorities thus further amplifying racial biases. Healthcare providers, policymakers, AI developers and all stakeholders should pay close attention to various steps in the pipeline of building AI/ML models and every effort must be made to establish algorithmic fairness to redress inequities.

Data gaps and opportunities for modeling cancer health equity

Population models of cancer reflect the overall US population by drawing on numerous existing data resources for parameter inputs and calibration targets. Models require data inputs that are appropriately representative, collected in a harmonized manner, have minimal missing or inaccurate values, and reflect adequate sample sizes. Data resource priorities for population modeling to support cancer health equity include increasing the availability of data that 1) arise from uninsured and underinsured individuals and those traditionally not included in health-care delivery studies, 2) reflect relevant exposures for groups historically and intentionally excluded across the full cancer control continuum, 3) disaggregate categories (race, ethnicity, socioeconomic status, gender, sexual orientation, etc.) and their intersections that conceal important variation in health outcomes, 4) identify specific populations of interest in clinical databases whose health outcomes have been understudied, 5) enhance health records through expanded data elements and linkage with other data types (eg, patient surveys, provider and/or facility level information, neighborhood data), 6) decrease missing and misclassified data from historically underrecognized populations, and 7) capture potential measures or effects of systemic racism and corresponding intervenable targets for change.

The value of standards for health datasets in artificial intelligence-based applications

Artificial intelligence as a medical device is increasingly being applied to healthcare for diagnosis, risk stratification and resource allocation. However, a growing body of evidence has highlighted the risk of algorithmic bias, which may perpetuate existing health inequity. This problem arises in part because of systemic inequalities in dataset curation, unequal opportunity to participate in research and inequalities of access. This study aims to explore existing standards, frameworks and best practices for ensuring adequate data diversity in health datasets. Exploring the body of existing literature and expert views is an important step towards the development of consensus-based guidelines. The study comprises two parts: a systematic review of existing standards, frameworks and best practices for healthcare datasets; and a survey and thematic analysis of stakeholder views of bias, health equity and best practices for artificial intelligence as a medical device. We found that the need for dataset diversity was well described in literature, and experts generally favored the development of a robust set of guidelines, but there were mixed views about how these could be implemented practically. The outputs of this study will be used to inform the development of standards for transparency of data diversity in health datasets (the STANDING Together initiative).

Sources of Disparities in Surveillance Mammography Performance and Risk-Guided Recommendations for Supplemental Breast Imaging: A Simulation Study

BACKGROUND

Surveillance mammography is recommended for all women with a history of breast cancer. Risk-guided surveillance incorporating advanced imaging modalities based on individual risk of a second cancer could improve cancer detection. However, personalized surveillance may also amplify disparities.

METHODS

In simulated populations using inputs from the Breast Cancer Surveillance Consortium (BCSC), we investigated race- and ethnicity-based disparities. Disparities were decomposed into those due to primary breast cancer and treatment characteristics, social determinants of health (SDOH) and differential error in second cancer ascertainment by modeling populations with or without variation across race and ethnicity in the distribution of these characteristics. We estimated effects of disparities on mammography performance and supplemental imaging recommendations stratified by race and ethnicity.

RESULTS

In simulated cohorts based on 65,446 BCSC surveillance mammograms, when only cancer characteristics varied by race and ethnicity, mammograms for Black women had lower sensitivity compared with the overall population (64.1% vs. 71.1%). Differences between Black women and the overall population were larger when both cancer characteristics and SDOH varied by race and ethnicity (53.8% vs. 71.1%). Basing supplemental imaging recommendations on high predicted second cancer risk resulted in less frequent recommendations for Hispanic (6.7%) and Asian/Pacific Islander women (6.4%) compared with the overall population (10.0%).

CONCLUSIONS

Variation in cancer characteristics and SDOH led to disparities in surveillance mammography performance and recommendations for supplemental imaging.

IMPACT

Risk-guided surveillance imaging may exacerbate disparities. Decision-makers should consider implications for equity in cancer outcomes resulting from implementing risk-guided screening programs. See related In the Spotlight, p. 1479.

A dataset of skin lesion images collected in Argentina for the evaluation of AI tools in this population

In recent years, numerous dermatological image databases have been published to make possible the development and validation of artificial intelligence-based technologies to support healthcare professionals in the diagnosis of skin diseases. However, the generation of these datasets confined to certain countries as well as the lack of demographic information accompanying the images, prevents having a real knowledge of in which populations these models could be used. Consequently, this hinders the translation of the models to the clinical setting. This has led the scientific community to encourage the detailed and transparent reporting of the databases used for artificial intelligence developments, as well as to promote the formation of genuinely international databases that can be representative of the world population. Through this work, we seek to provide details of the processing stages of the first public database of dermoscopy and clinical images created in a hospital in Argentina. The dataset comprises 1,616 images corresponding to 1,246 unique lesions collected from 623 patients.

Variation in monitoring: Glucose measurement in the ICU as a case study to preempt spurious correlations

Objective

Health inequities can be influenced by demographic factors such as race and ethnicity, proficiency in English, and biological sex. Disparities may manifest as differential likelihood of testing which correlates directly with the likelihood of an intervention to address an abnormal finding. Our retrospective observational study evaluated the presence of variation in glucose measurements in the Intensive Care Unit (ICU).

Methods

Using the MIMIC-IV database (2008-2019), a single-center, academic referral hospital in Boston (USA), we identified adult patients meeting sepsis-3 criteria. Exclusion criteria were diabetic ketoacidosis, ICU length of stay under 1 day, and unknown race or ethnicity. We performed a logistic regression analysis to assess differential likelihoods of glucose measurements on day 1. A negative binomial regression was fitted to assess the frequency of subsequent glucose readings. Analyses were adjusted for relevant clinical confounders, and performed across three disparity proxy axes: race and ethnicity, sex, and English proficiency.

Results

We studied 24,927 patients, of which 19.5% represented racial and ethnic minority groups, 42.4% were female, and 9.8% had limited English proficiency. No significant differences were found for glucose measurement on day 1 in the ICU. This pattern was consistent irrespective of the axis of analysis, i.e. race and ethnicity, sex, or English proficiency. Conversely, subsequent measurement frequency revealed potential disparities. Specifically, males (incidence rate ratio (IRR) 1.06, 95% confidence interval (CI) 1.01 - 1.21), patients who identify themselves as Hispanic (IRR 1.11, 95% CI 1.01 - 1.21), or Black (IRR 1.06, 95% CI 1.01 - 1.12), and patients being English proficient (IRR 1.08, 95% CI 1.01 - 1.15) had higher chances of subsequent glucose readings.

Conclusion

We found disparities in ICU glucose measurements among patients with sepsis, albeit the magnitude was small. Variation in disease monitoring is a source of data bias that may lead to spurious correlations when modeling health data.

Critical Bias in Critical Care Devices

Critical care data contain information about the most physiologically fragile patients in the hospital, who require a significant level of monitoring. However, medical devices used for patient monitoring suffer from measurement biases that have been largely underreported. This article explores sources of bias in commonly used clinical devices, including pulse oximeters, thermometers, and sphygmomanometers. Further, it provides a framework for mitigating these biases and key principles to achieve more equitable health care delivery.

Ethical Considerations for Artificial Intelligence in Medical Imaging: Deployment and Governance

The deployment of artificial intelligence (AI) has the potential to make nuclear medicine and medical imaging faster, cheaper, and both more effective and more accessible. This is possible, however, only if clinicians and patients feel that these AI medical devices (AIMDs) are trustworthy. Highlighting the need to ensure health justice by fairly distributing benefits and burdens while respecting individual patients' rights, the AI Task Force of the Society of Nuclear Medicine and Molecular Imaging has identified 4 major ethical risks that arise during the deployment of AIMD: autonomy of patients and clinicians, transparency of clinical performance and limitations, fairness toward marginalized populations, and accountability of physicians and developers. We provide preliminary recommendations for governing these ethical risks to realize the promise of AIMD for patients and populations.

A call for better validation of opioid overdose risk algorithms

Clinical decision support (CDS) systems powered by predictive models have the potential to improve the accuracy and efficiency of clinical decision-making. However, without sufficient validation, these systems have the potential to mislead clinicians and harm patients. This is especially true for CDS systems used by opioid prescribers and dispensers, where a flawed prediction can directly harm patients. To prevent these harms, regulators and researchers have proposed guidance for validating predictive models and CDS systems. However, this guidance is not universally followed and is not required by law. We call on CDS developers, deployers, and users to hold these systems to higher standards of clinical and technical validation. We provide a case study on two CDS systems deployed on a national scale in the United States for predicting a patient's risk of adverse opioid-related events: the Stratification Tool for Opioid Risk Mitigation (STORM), used by the Veterans Health Administration, and NarxCare, a commercial system.

Recommendations for the use of pediatric data in artificial intelligence and machine learning ACCEPT-AI

ACCEPT-AI is a framework of recommendations for the safe inclusion of pediatric data in artificial intelligence and machine learning (AI/ML) research. It has been built on fundamental ethical principles of pediatric and AI research and incorporates age, consent, assent, communication, equity, protection of data, and technological considerations. ACCEPT-AI has been designed to guide researchers, clinicians, regulators, and policymakers and can be utilized as an independent tool, or adjunctively to existing AI/ML guidelines.

Challenges of Applying Automated Polysomnography Scoring at Scale

Automatic polysomnography analysis can be leveraged to shorten scoring times, reduce associated costs, and ultimately improve the overall diagnosis of sleep disorders. Multiple and diverse strategies have been attempted for implementation of this technology at scale in the routine workflow of sleep centers. The field, however, is complex and presents unsolved challenges in a number of areas. Recent developments in computer science and artificial intelligence are nevertheless closing the gap. Technological advances are also opening new pathways for expanding our current understanding of the domain and its analysis.

Addressing the Challenge of Biomedical Data Inequality: An Artificial Intelligence Perspective

Artificial intelligence (AI) and other data-driven technologies hold great promise to transform healthcare and confer the predictive power essential to precision medicine. However, the existing biomedical data, which are a vital resource and foundation for developing medical AI models, do not reflect the diversity of the human population. The low representation in biomedical data has become a significant health risk for non-European populations, and the growing application of AI opens a new pathway for this health risk to manifest and amplify. Here we review the current status of biomedical data inequality and present a conceptual framework for understanding its impacts on machine learning. We also discuss the recent advances in algorithmic interventions for mitigating health disparities arising from biomedical data inequality. Finally, we briefly discuss the newly identified disparity in data quality among ethnic groups and its potential impacts on machine learning.

Large language models encode clinical knowledge

Large language models (LLMs) have demonstrated impressive capabilities, but the bar for clinical applications is high. Attempts to assess the clinical knowledge of models typically rely on automated evaluations based on limited benchmarks. Here, to address these limitations, we present MultiMedQA, a benchmark combining six existing medical question answering datasets spanning professional medicine, research and consumer queries and a new dataset of medical questions searched online, HealthSearchQA. We propose a human evaluation framework for model answers along multiple axes including factuality, comprehension, reasoning, possible harm and bias. In addition, we evaluate Pathways Language Model1 (PaLM, a 540-billion parameter LLM) and its instruction-tuned variant, Flan-PaLM2 on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM achieves state-of-the-art accuracy on every MultiMedQA multiple-choice dataset (MedQA3, MedMCQA4, PubMedQA5 and Measuring Massive Multitask Language Understanding (MMLU) clinical topics6), including 67.6% accuracy on MedQA (US Medical Licensing Exam-style questions), surpassing the prior state of the art by more than 17%. However, human evaluation reveals key gaps. To resolve this, we introduce instruction prompt tuning, a parameter-efficient approach for aligning LLMs to new domains using a few exemplars. The resulting model, Med-PaLM, performs encouragingly, but remains inferior to clinicians. We show that comprehension, knowledge recall and reasoning improve with model scale and instruction prompt tuning, suggesting the potential utility of LLMs in medicine. Our human evaluations reveal limitations of today's models, reinforcing the importance of both evaluation frameworks and method development in creating safe, helpful LLMs for clinical applications.

Advancing AI in healthcare: A comprehensive review of best practices

Artificial Intelligence (AI) and Machine Learning (ML) are powerful tools shaping the healthcare sector. This review considers twelve key aspects of AI in clinical practice: 1) Ethical AI; 2) Explainable AI; 3) Health Equity and Bias in AI; 4) Sponsorship Bias; 5) Data Privacy; 6) Genomics and Privacy; 7) Insufficient Sample Size and Self-Serving Bias; 8) Bridging the Gap Between Training Datasets and Real-World Scenarios; 9) Open Source and Collaborative Development; 10) Dataset Bias and Synthetic Data; 11) Measurement Bias; 12) Reproducibility in AI Research. These categories represent both the challenges and opportunities of AI implementation in healthcare. While AI holds significant potential for improving patient care, it also presents risks and challenges, such as ensuring privacy, combating bias, and maintaining transparency and ethics. The review underscores the necessity of developing comprehensive best practices for healthcare organizations and fostering a diverse dialogue involving data scientists, clinicians, patient advocates, ethicists, economists, and policymakers. We are at the precipice of significant transformation in healthcare powered by AI. By continuing to reassess and refine our approach, we can ensure that AI is implemented responsibly and ethically, maximizing its benefit to patient care and public health.