Machine Learning Applied to Electronic Health Records: Identification of Chemotherapy Patients at High Risk for Preventable Emergency Department Visits and Hospital Admissions

Using machine learning models for predicting monthly iPTH levels in hemodialysis patients

BACKGROUND AND OBJECTIVE

Intact parathyroid hormone (iPTH), also known as active parathyroid hormone, is an important indicator commonly for monitoring secondary hyperparathyroidism (SHPT) in patients undergoing hemodialysis. The aim of this study was to use machine learning (ML) models to predict monthly iPTH levels in patients undergoing hemodialysis.

METHODS

We conducted a retrospective study on patients undergoing regular hemodialysis. Patients' blood examinations data was collected from Taiwan Society of Nephrology - Kidney Dialysis, Transplantation (TSN-KiDiT) registration system, and patients' medications data was collected from Pingtung Christian Hospital (PTCH), Taiwan. We used five different ML models to classify patients into three distinct categories based on their iPTH levels: iPTH < 150, iPTH ≥ 150 & iPTH < 600, and iPTH ≥ 600(pg/ml).

RESULTS

We ultimately included 1,351 patients in our study and processed the data in four different ways. These methods varied based on the duration of the data (either using data from just one month or continuously over three months) and the number of features used (either all 52 features or only 20 most important features identified by SHapley Additive exPlanations (SHAP) analysis). The XGBoost model, using data from a continuous three-month period and all available features, yielded the best Weighted AUROC (0.922).

CONCLUSIONS

ML is highly effective in predicting iPTH levels in hemodialysis patients, notably accurate for those with iPTH over 600 pg/ml. This method enables early identification of high-risk patients, reducing reliance on retrospective blood test assessments. Future research should focus on advancing explainable AI methods to foster clinicians' trust, and developing adaptable ML frameworks that could seamlessly integrate with existing healthcare systems.

Secondary Neutropenias

Neutrophils are a critical component of immunity, particularly against bacteria and other pathogens, but also in inflammation and tissue repair. As a consequence, individuals with neutropenia, defined by a reduction in absolute neutrophil counts, exhibit a strong propensity to severe infections that typically present with muted symptoms. Neutropenias encompass a heterogeneous set of disorders, comprising primary neutropenias, in which specific genes are mutated, and the more common secondary neutropenias, which have diverse non-genetic causes. These include hematological and other cancers, involving both direct effects of the cancer itself and indirect impacts via the chemotherapeutic, biological agents and cell-based approaches used for treatment. Other significant causes of secondary neutropenias are non-chemotherapeutic drugs, autoimmune and other immune diseases, infections and nutrient deficiencies. These collectively act by impacting neutrophil production in the bone marrow and/or destruction throughout the body. This review describes the biological and clinical manifestations of secondary neutropenias, detailing their underlying causes and management, with a discussion of alternative and emerging therapeutic approaches.

Can prediction models for hospital readmission be improved by incorporating patient-reported outcome measures? A systematic review and narrative synthesis

PURPOSE

To investigate the roles, challenges, and implications of using patient-reported outcome measures (PROMs) in predicting the risk of hospital readmissions.

METHODS

We systematically searched four bibliometric databases for peer-reviewed studies published in English between 1 January 2000 and 15 June 2023 and used validated PROMs to predict readmission risks for adult populations. Reported studies were analysed and narratively synthesised in accordance with the CHARMS and PRISMA guidelines.

RESULTS

Of the 2858 abstracts reviewed, 23 studies met predefined eligibility criteria, representing diverse geographic regions and medical specialties. Among those, 19 identified the positive contributions of PROMs in predicting readmission risks. Seven studies utilised generic PROMs exclusively, eleven used generic and condition-specific PROMs, while 5 focussed solely on condition-specific PROMs. Logistic regression was the most used modelling approach, with 13 studies aiming at predicting 30-day all-cause readmission risks. The c-statistic, ranging from 0.54 to 0.84, was reported in 22/23 studies as a measure of model discrimination. Nine studies reported model calibration in addition to c-statistic. Thirteen studies detailed their approaches to dealing with missing data.

CONCLUSION

Our study highlights the potential of PROMs to enhance predictive accuracy in readmission models, while acknowledging the diversity in data collection methods, readmission definitions, and model evaluation approaches. Recognizing that PROMs serve various purposes beyond readmission reduction, our study supports routine data collection and strategic integration of PROMs in healthcare practices to improve patient outcomes. To facilitate comparative analysis and broaden the use of PROMs in the prediction framework, it is imperative to consider the methodological aspects involved.

Digital Remote Monitoring Using an mHealth Solution for Survivors of Cancer: Protocol for a Pilot Observational Study

BACKGROUND

Healthy lifestyle interventions have a positive impact on multiple disease trajectories, including cancer-related outcomes. Specifically, appropriate habitual physical activity, adequate sleep, and a regular wholesome diet are of paramount importance for the wellness and supportive care of survivors of cancer. Mobile health (mHealth) apps have the potential to support novel tailored lifestyle interventions.

OBJECTIVE

This observational pilot study aims to assess the feasibility of mHealth multidimensional longitudinal monitoring in survivors of cancer. The primary objective is to test the compliance (user engagement) with the monitoring solution. Secondary objectives include recording clinically relevant subjective and objective measures collected through the digital solution.

METHODS

This is a monocentric pilot study taking place in Bangor, Wales, United Kingdom. We plan to enroll up to 100 adult survivors of cancer not receiving toxic anticancer treatment, who will provide self-reported behavioral data recorded via a dedicated app and validated questionnaires and objective data automatically collected by a paired smartwatch over 16 weeks. The participants will continue with their normal routine surveillance care for their cancer. The primary end point is feasibility (eg, mHealth monitoring acceptability). Composite secondary end points include clinically relevant patient-reported outcome measures (eg, the Edmonton Symptom Assessment System score) and objective physiological measures (eg, step counts). This trial received a favorable ethical review in May 2023 (Integrated Research Application System 301068).

RESULTS

This study is part of an array of pilots within a European Union funded project, entitled "GATEKEEPER," conducted at different sites across Europe and covering various chronic diseases. Study accrual is anticipated to commence in January 2024 and continue until June 2024. It is hypothesized that mHealth monitoring will be feasible in survivors of cancer; specifically, at least 50% (50/100) of the participants will engage with the app at least once a week in 8 of the 16 study weeks.

CONCLUSIONS

In a population with potentially complex clinical needs, this pilot study will test the feasibility of multidimensional remote monitoring of patient-reported outcomes and physiological parameters. Satisfactory compliance with the use of the app and smartwatch, whether confirmed or infirmed through this study, will be propaedeutic to the development of innovative mHealth interventions in survivors of cancer.

INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID)

PRR1-10.2196/52957.

Drivers of Emergency Department Use Among Oncology Patients in the Era of Novel Cancer Therapeutics: A Systematic Review

BACKGROUND

Patients diagnosed with cancer are frequent users of the emergency department (ED). While many visits are unavoidable, a significant portion may be potentially preventable ED visits (PPEDs). Cancer treatments have greatly advanced, whereby patients may present with unique toxicities from targeted therapies and are often living longer with advanced disease. Prior work focused on patients undergoing cytotoxic chemotherapy, and often excluded those on supportive care alone. Other contributors to ED visits in oncology, such as patient-level variables, are less well-established. Finally, prior studies focused on ED diagnoses to describe trends and did not evaluate PPEDs. An updated systematic review was completed to focus on PPEDs, novel cancer therapies, and patient-level variables, including those on supportive care alone.

METHODS

Three online databases were used. Included publications were in English, from 2012-2022, with sample sizes of ≥50, and reported predictors of ED presentation or ED diagnoses in oncology.

RESULTS

45 studies were included. Six studies highlighted PPEDs with variable definitions. Common reasons for ED visits included pain (66%) or chemotherapy toxicities (69.1%). PPEDs were most frequent amongst breast cancer patients (13.4%) or patients receiving cytotoxic chemotherapy (20%). Three manuscripts included immunotherapy agents, and only one focused on end-of-life patients.

CONCLUSION

This updated systematic review highlights variability in oncology ED visits during the last decade. There is limited work on the concept of PPEDs, patient-level variables and patients on supportive care alone. Overall, pain and chemotherapy toxicities remain key drivers of ED visits in cancer patients. Further work is needed in this realm.

Ensuring fair, safe, and interpretable artificial intelligence-based prediction tools in a real-world oncological setting

BACKGROUND

Cancer patients often experience treatment-related symptoms which, if uncontrolled, may require emergency department admission. We developed models identifying breast or genitourinary cancer patients at the risk of attending emergency department (ED) within 30-days and demonstrated the development, validation, and proactive approach to in-production monitoring of an artificial intelligence-based predictive model during a 3-month simulated deployment at a cancer hospital in the United States.

METHODS

We used routinely-collected electronic health record data to develop our predictive models. We evaluated models including a variational autoencoder k-nearest neighbors algorithm (VAE-kNN) and model behaviors with a sample containing 84,138 observations from 28,369 patients. We assessed the model during a 77-day production period exposure to live data using a proactively monitoring process with predefined metrics.

RESULTS

Performance of the VAE-kNN algorithm is exceptional (Area under the receiver-operating characteristics, AUC = 0.80) and remains stable across demographic and disease groups over the production period (AUC 0.74-0.82). We can detect issues in data feeds using our monitoring process to create immediate insights into future model performance.

CONCLUSIONS

Our algorithm demonstrates exceptional performance at predicting risk of 30-day ED visits. We confirm that model outputs are equitable and stable over time using a proactive monitoring approach.

A Bayesian approach to predictive uncertainty in chemotherapy patients at risk of acute care utilization

BACKGROUND

Machine learning (ML) predictions are becoming increasingly integrated into medical practice. One commonly used method, ℓ1-penalised logistic regression (LASSO), can estimate patient risk for disease outcomes but is limited by only providing point estimates. Instead, Bayesian logistic LASSO regression (BLLR) models provide distributions for risk predictions, giving clinicians a better understanding of predictive uncertainty, but they are not commonly implemented.

METHODS

This study evaluates the predictive performance of different BLLRs compared to standard logistic LASSO regression, using real-world, high-dimensional, structured electronic health record (EHR) data from cancer patients initiating chemotherapy at a comprehensive cancer centre. Multiple BLLR models were compared against a LASSO model using an 80-20 random split using 10-fold cross-validation to predict the risk of acute care utilization (ACU) after starting chemotherapy.

FINDINGS

This study included 8439 patients. The LASSO model predicted ACU with an area under the receiver operating characteristic curve (AUROC) of 0.806 (95% CI: 0.775-0.834). BLLR with a Horseshoe+ prior and a posterior approximated by Metropolis-Hastings sampling showed similar performance: 0.807 (95% CI: 0.780-0.834) and offers the advantage of uncertainty estimation for each prediction. In addition, BLLR could identify predictions too uncertain to be automatically classified. BLLR uncertainties were stratified by different patient subgroups, demonstrating that predictive uncertainties significantly differ across race, cancer type, and stage.

INTERPRETATION

BLLRs are a promising yet underutilised tool that increases explainability by providing risk estimates while offering a similar level of performance to standard LASSO-based models. Additionally, these models can identify patient subgroups with higher uncertainty, which can augment clinical decision-making.

FUNDING

This work was supported in part by the National Library Of Medicine of the National Institutes of Health under Award Number R01LM013362. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Reducing Avoidable Emergency Visits and Hospitalizations With Patient Risk-Based Prescriptive Analytics: A Quality Improvement Project at an Oncology Care Model Practice

PURPOSE

Cancer-related emergency department (ED) visits and hospitalizations that would have been appropriately managed in the outpatient setting are avoidable and detrimental to patients and health systems. This quality improvement (QI) project aimed to leverage patient risk-based prescriptive analytics at a community oncology practice to reduce avoidable acute care use (ACU).

METHODS

Using the Plan-Do-Study-Act (PDSA) methodology, we implemented the Jvion Care Optimization and Recommendation Enhancement augmented intelligence (AI) tool at an Oncology Care Model (OCM) practice, the Center for Cancer and Blood Disorders practice. We applied continuous machine learning to predict risk of preventable harm (avoidable ACU) and generated patient-specific recommendations that nurses implemented to avert it.

RESULTS

Patient-centric interventions included medication/dosage changes, laboratory tests/imaging, physical/occupational/psychologic therapy referral, palliative care/hospice referral, and surveillance/observation. Nurses contacted patients every 1-2 weeks after initial outreach to assess and maintain adherence to recommended interventions. Per 100 unique OCM patients, monthly ED visits dropped from 13.7 to 11.5 (18%), a sustained month-over-month improvement. Quarterly admissions dropped from 19.5 to 17.1 (13%), a sustained quarter-over-quarter improvement. Overall, the practice realized potential annual savings of $2.8 million US dollars (USD) on avoidable ACU.

CONCLUSION

The AI tool has enabled nurse case managers to identify and resolve critical clinical issues and reduce avoidable ACU. Effects on outcomes can be inferred from the reduction; targeting short-term interventions toward patients most at-risk translates to better long-term care and outcomes. QI projects involving predictive modeling of patient risk, prescriptive analytics, and nurse outreach may reduce ACU.

Revolutionizing Chronic Kidney Disease Management with Machine Learning and Artificial Intelligence

Chronic kidney disease (CKD) poses a significant public health challenge, affecting approximately 11% to 13% of the global population [...].

Predict, diagnose, and treat chronic kidney disease with machine learning: a systematic literature review

OBJECTIVES

In this systematic review we aimed at assessing how artificial intelligence (AI), including machine learning (ML) techniques have been deployed to predict, diagnose, and treat chronic kidney disease (CKD). We systematically reviewed the available evidence on these innovative techniques to improve CKD diagnosis and patient management.

METHODS

We included English language studies retrieved from PubMed. The review is therefore to be classified as a "rapid review", since it includes one database only, and has language restrictions; the novelty and importance of the issue make missing relevant papers unlikely. We extracted 16 variables, including: main aim, studied population, data source, sample size, problem type (regression, classification), predictors used, and performance metrics. We followed the Preferred Reporting Items for Systematic Reviews (PRISMA) approach; all main steps were done in duplicate.

RESULTS

From a total of 648 studies initially retrieved, 68 articles met the inclusion criteria. Models, as reported by authors, performed well, but the reported metrics were not homogeneous across articles and therefore direct comparison was not feasible. The most common aim was prediction of prognosis, followed by diagnosis of CKD. Algorithm generalizability, and testing on diverse populations was rarely taken into account. Furthermore, the clinical evaluation and validation of the models/algorithms was perused; only a fraction of the included studies, 6 out of 68, were performed in a clinical context.

CONCLUSIONS

Machine learning is a promising tool for the prediction of risk, diagnosis, and therapy management for CKD patients. Nonetheless, future work is needed to address the interpretability, generalizability, and fairness of the models to ensure the safe application of such technologies in routine clinical practice.

Clinical Informatics Approaches to Understand and Address Cancer Disparities

OBJECTIVES

Disparities in cancer incidence and outcomes across race, ethnicity, gender, socioeconomic status, and geography are well-documented, but their etiologies are often poorly understood and multifactorial. Clinical informatics can provide tools to better understand and address these disparities by enabling high-throughput analysis of multiple types of data. Here, we review recent efforts in clinical informatics to study and measure disparities in cancer.

METHODS

We carried out a narrative review of clinical informatics studies related to cancer disparities and bias published from 2018-2021, with a focus on domains such as real-world data (RWD) analysis, natural language processing (NLP), radiomics, genomics, proteomics, metabolomics, and metagenomics.

RESULTS

Clinical informatics studies that investigated cancer disparities across race, ethnicity, gender, and age were identified. Most cancer disparities work within clinical informatics used RWD analysis, NLP, radiomics, and genomics. Emerging applications of clinical informatics to understand cancer disparities, including proteomics, metabolomics, and metagenomics, were less well represented in the literature but are promising future research avenues. Algorithmic bias was identified as an important consideration when developing and implementing cancer clinical informatics techniques, and efforts to address this bias were reviewed.

CONCLUSIONS

In recent years, clinical informatics has been used to probe a range of data sources to understand cancer disparities across different populations. As informatics tools become integrated into clinical decision-making, attention will need to be paid to ensure that algorithmic bias does not amplify existing disparities. In our increasingly interconnected medical systems, clinical informatics is poised to untap the full potential of multi-platform health data to address cancer disparities.

Cancer-related emergency and urgent care: expanding the research agenda

Purpose of review

Cancer-related emergency department (ED) visits often result in higher hospital admission rates than non-cancer visits. It has been estimated many of these costly hospital admissions can be prevented, yet urgent care clinics and EDs lack cancer-specific care resources to support the needs of this complex population. Implementing effective approaches across different care settings and populations to minimize ED and urgent care visits improves oncologic complication management, and coordinating follow-up care will be particularly important as the population of cancer patients and survivors continues to increase. The National Cancer Institute (NCI) and the Office of Emergency Care (OECR) convened a workshop in December 2021, “Cancer-related Emergency and Urgent Care: Prevention, Management, and Care Coordination” to highlight progress, knowledge gaps, and research opportunities. This report describes the current landscape of cancer-related urgent and emergency care and includes research recommendations from workshop participants to decrease the risk of oncologic complications, improve their management, and enhance coordination of care.

Recent findings

Since 2014, NCI and OECR have collaborated to support research in cancer-related emergency care. Workshop participants recommended a number of promising research opportunities, as well as key considerations for designing and conducting research in this area. Opportunities included better characterizing unscheduled care services, identifying those at higher risk for such care, developing care delivery models to minimize unplanned events and enhance their care, recognizing cancer prevention and screening opportunities in the ED, improving management of specific cancer-related presentations, and conducting goals of care conversations.

Summary

Significant progress has been made over the past 7 years with the creation of the Comprehensive Oncologic Emergency Research Network, broad involvement of the emergency medicine and oncology communities, establishing a proof-of-concept observational study, and NCI and OECR’s efforts to support this area of research. However, critical gaps remain.

Introducing AI to the molecular tumor board: one direction toward the establishment of precision medicine using large-scale cancer clinical and biological information

Since U.S. President Barack Obama announced the Precision Medicine Initiative in his New Year's State of the Union address in 2015, the establishment of a precision medicine system has been emphasized worldwide, particularly in the field of oncology. With the advent of next-generation sequencers specifically, genome analysis technology has made remarkable progress, and there are active efforts to apply genome information to diagnosis and treatment. Generally, in the process of feeding back the results of next-generation sequencing analysis to patients, a molecular tumor board (MTB), consisting of experts in clinical oncology, genetic medicine, etc., is established to discuss the results. On the other hand, an MTB currently involves a large amount of work, with humans searching through vast databases and literature, selecting the best drug candidates, and manually confirming the status of available clinical trials. In addition, as personalized medicine advances, the burden on MTB members is expected to increase in the future. Under these circumstances, introducing cutting-edge artificial intelligence (AI) technology and information and communication technology to MTBs while reducing the burden on MTB members and building a platform that enables more accurate and personalized medical care would be of great benefit to patients. In this review, we introduced the latest status of elemental technologies that have potential for AI utilization in MTB, and discussed issues that may arise in the future as we progress with AI implementation.

A Comparison of Logistic Regression Against Machine Learning Algorithms for Gastric Cancer Risk Prediction Within Real-World Clinical Data Streams

PURPOSE

Noncardia gastric cancer (NCGC) is a leading cause of global cancer mortality, and is often diagnosed at advanced stages. Development of NCGC risk models within electronic health records (EHR) may allow for improved cancer prevention. There has been much recent interest in use of machine learning (ML) for cancer prediction, but few studies comparing ML with classical statistical models for NCGC risk prediction.

METHODS

We trained models using logistic regression (LR) and four commonly used ML algorithms to predict NCGC from age-/sex-matched controls in two EHR systems: Stanford University and the University of Washington (UW). The LR model contained well-established NCGC risk factors (intestinal metaplasia histology, prior Helicobacter pylori infection, race, ethnicity, nativity status, smoking history, anemia), whereas ML models agnostically selected variables from the EHR. Models were developed and internally validated in the Stanford data, and externally validated in the UW data. Hyperparameter tuning of models was achieved using cross-validation. Model performance was compared by accuracy, sensitivity, and specificity.

RESULTS

In internal validation, LR performed with comparable accuracy (0.732; 95% CI, 0.698 to 0.764), sensitivity (0.697; 95% CI, 0.647 to 0.744), and specificity (0.767; 95% CI, 0.720 to 0.809) to penalized lasso, support vector machine, K-nearest neighbor, and random forest models. In external validation, LR continued to demonstrate high accuracy, sensitivity, and specificity. Although K-nearest neighbor demonstrated higher accuracy and specificity, this was offset by significantly lower sensitivity. No ML model consistently outperformed LR across evaluation criteria.

CONCLUSION

Drawing data from two independent EHRs, we find LR on the basis of established risk factors demonstrated comparable performance to optimized ML algorithms. This study demonstrates that classical models built on robust, hand-chosen predictor variables may not be inferior to data-driven models for NCGC risk prediction.