Predicting pregnancy-related pelvic girdle pain using machine learning

BACKGROUND

Pregnancy-related pelvic girdle pain (PPGP) is a common complication during gestation which negatively influences pregnant women's quality of life. There are numerous risk factors associated with PPGP, however, there is limited information about being able to predict the diagnosis of PPGP.

OBJECTIVE

To compare machine learning (ML) and traditional predictive modelling to predict the clinical diagnosis of PPGP.

METHODS

This study reanalysed data from 780 pregnant women attending a tertiary hospital. ML algorithms, including Logistic Regression (LR), Random Forest, Xtreme Gradient Boost (XGBoost), and K-Nearest Neighbors, were used to predict the clinical diagnosis of PPGP. Feature selection methods and cross-validation were employed to optimize model performance, with the Area Under the Receiver Operating Characteristic Curve (AUROC) as the primary outcome measure.

RESULTS

The ML models, particularly XGBoost and LR, demonstrated high levels of predictive accuracy (AUROC = 0.70). Key predictive factors were a history of low back pain/pelvic girdle pain (LBP/PGP) in previous pregnancies, family history, gestational age, and a longer duration of standing during the day. The history of LBP/PGP in previous pregnancies emerged as the most significant predictor.

CONCLUSIONS

This study highlighted the potential of ML models to enhance the ability to predict PPGP and offers a more accurate and comprehensive approach to identifying women at risk of PPGP. The integration of ML techniques into clinical practice could improve early identification and inform preventative and intervention strategies, potentially reducing the impact of PPGP on pregnant women.

Cross-Combination Analyses of Random Forest Feature Selection and Decision Tree Model for Predicting Intraoperative Hypothermia in Total Joint Arthroplasty

BACKGROUND

In total joint arthroplasty patients, intraoperative hypothermia (IOH) is associated with perioperative complications and an increased economic burden. Previous models have some limitations and mainly focus on regression modeling. Random forest (RF) algorithms and decision tree modeling are effective for eliminating irrelevant features and making predictions that aid in accelerating modeling and reducing application difficulty.

METHODS

We conducted this prospective observational study using convenience sampling and collected data from 327 total joint arthroplasty patients in a tertiary hospital from March 4, 2023, to September 11, 2023. Of those, 229 patients were assigned to the training and 98 to the testing sets. The Chi-square, Mann-Whitney U, and t-tests were used for baseline analyses. The feature variables selection used the RF algorithms, and the decision tree model was trained on 299 examples and validated on 98. The sensitivity, specificity, recall, F1 score, and area under the curve were used to test the model's performance.

RESULTS

The RF algorithms identified the preheating time, the volume of flushing fluids, the intraoperative infusion volume, the anesthesia time, the surgical time, and the core temperature after intubation as risk factors for IOH. The decision tree was grown to 5 levels with 9 terminal nodes. The overall incidence of IOH was 42.13%. The sensitivity, specificity, recall, F1 score, and area under the curve were 0.651, 0.907, 0.916, 0.761, and 0.810, respectively. The model indicated strong internal consistency and predictive ability.

CONCLUSIONS

The preheating time, the volume of flushing fluids, the intraoperative infusion volume, the anesthesia time, the surgical time, and the core temperature after intubation could accurately predict IOH in total joint arthroplasty patients. By monitoring these factors, the clinical staff could achieve early detection and intervention of IOH in total joint arthroplasty patients.

Comparison of machine learning and conventional statistical modeling for predicting readmission following acute heart failure hospitalization

INTRODUCTION

Developing accurate models for predicting the risk of 30-day readmission is a major healthcare interest. Evidence suggests that models developed using machine learning (ML) may have better discrimination than conventional statistical models (CSM), but the calibration of such models is unclear.

OBJECTIVES

To compare models developed using ML with those developed using CSM to predict 30-day readmission for cardiovascular and noncardiovascular causes in HF patients.

METHODS

We retrospectively enrolled 10,919 patients with HF (> 18 years) discharged alive from a hospital or emergency department (2004-2007) in Ontario, Canada. The study sample was randomly divided into training and validation sets in a 2:1 ratio. CSMs to predict 30-day readmission were developed using Fine-Gray subdistribution hazards regression (treating death as a competing risk), and the ML algorithm employed random survival forests for competing risks (RSF-CR). Models were evaluated in the validation set using both discrimination and calibration metrics.

RESULTS

In the validation sample of 3602 patients, RSF-CR (c-statistic=0.620) showed similar discrimination to the Fine-Gray competing risk model (c-statistic=0.621) for 30-day cardiovascular readmission. In contrast, for 30-day noncardiovascular readmission, the Fine-Gray model (c-statistic=0.641) slightly outperformed the RSF-CR model (c-statistic=0.632). For both outcomes, The Fine-Gray model displayed better calibration than RSF-CR using calibration plots of observed vs predicted risks across the deciles of predicted risk.

CONCLUSIONS

Fine-Gray models had similar discrimination but superior calibration to the RSF-CR model, highlighting the importance of reporting calibration metrics for ML-based prediction models. The discrimination was modest in all readmission prediction models regardless of the methods used.

Using machine learning to predict bacteremia in urgent care patients on the basis of triage data and laboratory results

BACKGROUND

Despite advancements in antimicrobial therapies, bacteremia remains a life-threatening condition. Appropriate antimicrobials must be promptly administered to ensure patient survival. However, diagnosing bacteremia based on blood cultures is time-consuming and not something emergency department (ED) personnel are routinely trained to do.

METHODS

This retrospective cohort study developed several machine learning (ML) models to predict bacteremia in adults initially presenting with fever or hypothermia, comprising logistic regression, random forest, extreme gradient boosting, support vector machine, k-nearest neighbor, multilayer perceptron, and ensemble models. Random oversampling and synthetic minority oversampling techniques were adopted to balance the dataset. The variables included demographic characteristics, comorbidities, immunocompromised status, clinical characteristics, subjective symptoms reported during ED triage, and laboratory data. The study outcome was an episode of bacteremia.

RESULTS

Of the 5063 patients with initial fever or hypothermia from whom blood cultures were obtained, 128 (2.5 %) were diagnosed with bacteremia. We combined 36 selected variables and 10 symptoms subjectively reported by patients into features for analysis in our models. The ensemble model outperformed other models, with an area under the receiver operating characteristic curve (AUROC) of 0.930 and an F1-score of 0.735. The AUROC of all models was higher than 0.80.

CONCLUSION

The ML models developed effectively predicted bacteremia among febrile or hypothermic patients in the ED, with all models demonstrating high AUROC values and rapid processing times. The findings suggest that ED clinicians can effectively utilize ML techniques to develop predictive models for addressing clinical challenges.

The usefulness of artificial intelligence in breast reconstruction: a systematic review

BACKGROUND

Artificial Intelligence (AI) offers an approach to predictive modeling. The model learns to determine specific patterns of undesirable outcomes in a dataset. Therefore, a decision-making algorithm can be built based on these patterns to prevent negative results. This systematic review aimed to evaluate the usefulness of AI in breast reconstruction.

METHODS

A systematic review was conducted in August 2022 following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. MEDLINE, EMBASE, SCOPUS, and Google Scholar online databases were queried to capture all publications studying the use of artificial intelligence in breast reconstruction.

RESULTS

A total of 23 studies were full text-screened after removing duplicates, and twelve articles fulfilled our inclusion criteria. The Machine Learning algorithms applied for neuropathic pain, lymphedema diagnosis, microvascular abdominal flap failure, donor site complications associated to muscle sparing Transverse Rectus Abdominis flap, surgical complications, financial toxicity, and patient-reported outcomes after breast surgery demonstrated that AI is a helpful tool to accurately predict patient results. In addition, one study used Computer Vision technology to assist in Deep Inferior Epigastric Perforator Artery detection for flap design, considerably reducing the preoperative time compared to manual identification.

CONCLUSIONS

In breast reconstruction, AI can help the surgeon by optimizing the perioperative patients' counseling to predict negative outcomes, allowing execution of timely interventions and reducing the postoperative burden, which leads to obtaining the most successful results and improving patient satisfaction.

Developing machine learning models to predict multi-class functional outcomes and death three months after stroke in Sweden

Globally, stroke is the third-leading cause of mortality and disability combined, and one of the costliest diseases in society. More accurate predictions of stroke outcomes can guide healthcare organizations in allocating appropriate resources to improve care and reduce both the economic and social burden of the disease. We aim to develop and evaluate the performance and explainability of three supervised machine learning models and the traditional multinomial logistic regression (mLR) in predicting functional dependence and death three months after stroke, using routinely-collected data. This prognostic study included adult patients, registered in the Swedish Stroke Registry (Riksstroke) from 2015 to 2020. Riksstroke contains information on stroke care and outcomes among patients treated in hospitals in Sweden. Prognostic factors (features) included demographic characteristics, pre-stroke functional status, cardiovascular risk factors, medications, acute care, stroke type, and severity. The outcome was measured using the modified Rankin Scale at three months after stroke (a scale of 0-2 indicates independent, 3-5 dependent, and 6 dead). Outcome prediction models included support vector machines, artificial neural networks (ANN), eXtreme Gradient Boosting (XGBoost), and mLR. The models were trained and evaluated on 75% and 25% of the dataset, respectively. Model predictions were explained using SHAP values. The study included 102,135 patients (85.8% ischemic stroke, 53.3% male, mean age 75.8 years, and median NIHSS of 3). All models demonstrated similar overall accuracy (69%-70%). The ANN and XGBoost models performed significantly better than the mLR in classifying dependence with F1-scores of 0.603 (95% CI; 0.594-0.611) and 0.577 (95% CI; 0.568-0.586), versus 0.544 (95% CI; 0.545-0.563) for the mLR model. The factors that contributed most to the predictions were expectedly similar in the models, based on clinical knowledge. Our ANN and XGBoost models showed a modest improvement in prediction performance and explainability compared to mLR using routinely-collected data. Their improved ability to predict functional dependence may be of particular importance for the planning and organization of acute stroke care and rehabilitation.

Predictive Modelling of Postpartum Haemorrhage Using Early Risk Factors: A Comparative Analysis of Statistical and Machine Learning Models

Postpartum haemorrhage (PPH) is a significant cause of maternal morbidity and mortality worldwide, particularly in low-resource settings. This study aimed to develop a predictive model for PPH using early risk factors and rank their importance in terms of predictive ability. The dataset was obtained from an observational case-control study in northern Rwanda. Various statistical models and machine learning techniques were evaluated, including logistic regression, logistic regression with elastic-net regularisation, Random Forests, Extremely Randomised Trees, and gradient-boosted trees with XGBoost. The Random Forest model, with an average sensitivity of 80.7%, specificity of 71.3%, and a misclassification rate of 12.19%, outperformed the other models, demonstrating its potential as a reliable tool for predicting PPH. The important predictors identified in this study were haemoglobin level during labour and maternal age. However, there were differences in PPH risk factor importance in different data partitions, highlighting the need for further investigation. These findings contribute to understanding PPH risk factors, highlight the importance of considering different data partitions and implementing cross-validation in predictive modelling, and emphasise the value of identifying the appropriate prediction model for the application. Effective PPH prediction models are essential for improving maternal health outcomes on a global scale. This study provides valuable insights for healthcare providers to develop predictive models for PPH to identify high-risk women and implement targeted interventions.

Interpretable ensemble imbalance learning strategies for the risk assessment of severe-low-level wind shear based on LiDAR and PIREPs

The occurrence of severe low-level wind shear (S-LLWS) events in the vicinity of airport runways poses a significant threat to flight safety and exacerbates a burgeoning problem in civil aviation. Identifying the risk factors that contribute to occurrences of S-LLWS can facilitate the improvement of aviation safety. Despite the significant influence of S-LLWS on aviation safety, its occurrence is relatively infrequent in comparison to non-SLLWS incidents. In this study, we develop an S-LLWS risk prediction model through the utilization of ensemble imbalance learning (EIL) strategies, namely, BalanceCascade, EasyEnsemble, and RUSBoost. The data for this study were obtained from PIREPs and LiDAR at Hong Kong International Airport. The analysis revealed that the BalanceCascade strategy outperforms EasyEnsemble and RUSBoost in terms of prediction performance. Afterward, the SHapley Additive exPlanations (SHAP) interpretation tool was used in conjunction with the BalanceCascade model for the risk assessment of various factors. The four most influential risk factors, according to the SHAP interpretation tool, were hourly temperature, runway 25LD, runway 25LA, and RWY (encounter location of LLWS). S-LLWS was likely to happen at Runway 25LD and Runway 25LA in temperatures ranging from low to moderate. Similarly, a high proportion of S-LLWS events occurred near the runway threshold, and a relatively small proportion occurred away from it. The EIL strategies in conjunction with the SHAP interpretation tool may accurately predict the S-LLWS without the need for data augmentation in the data pre-processing phase.

Blood Urea Nitrogen-to-Albumin Ratio May Predict Mortality in Patients with Traumatic Brain Injury from the MIMIC Database: A Retrospective Study

Traumatic brain injury (TBI), a major global health burden, disrupts the neurological system due to accidents and other incidents. While the Glasgow coma scale (GCS) gauges neurological function, it falls short as the sole predictor of overall mortality in TBI patients. This highlights the need for comprehensive outcome prediction, considering not just neurological but also systemic factors. Existing approaches relying on newly developed biomolecules face challenges in clinical implementation. Therefore, we investigated the potential of readily available clinical indicators, like the blood urea nitrogen-to-albumin ratio (BAR), for improved mortality prediction in TBI. In this study, we investigated the significance of the BAR in predicting all-cause mortality in TBI patients. In terms of research methodologies, we gave preference to machine learning methods due to their exceptional performance in clinical support in recent years. Initially, we obtained data on TBI patients from the Medical Information Mart for Intensive Care database. A total of 2602 patients were included, of whom 2260 survived and 342 died in hospital. Subsequently, we performed data cleaning and utilized machine learning techniques to develop prediction models. We employed a ten-fold cross-validation method to obtain models with enhanced accuracy and area under the curve (AUC) (Light Gradient Boost Classifier accuracy, 0.905 ± 0.016, and AUC, 0.888; Extreme Gradient Boost Classifier accuracy, 0.903 ± 0.016, and AUC, 0.895; Gradient Boost Classifier accuracy, 0.898 ± 0.021, and AUC, 0.872). Simultaneously, we derived the importance ranking of the variable BAR among the included variables (in Light Gradient Boost Classifier, the BAR ranked fourth; in Extreme Gradient Boost Classifier, the BAR ranked sixth; in Gradient Boost Classifier, the BAR ranked fifth). To further evaluate the clinical utility of BAR, we divided patients into three groups based on their BAR values: Group 1 (BAR < 4.9 mg/g), Group 2 (BAR ≥ 4.9 and ≤10.5 mg/g), and Group 3 (BAR ≥ 10.5 mg/g). This stratification revealed significant differences in mortality across all time points: in-hospital mortality (7.61% vs. 15.16% vs. 31.63%), as well as one-month (8.51% vs. 17.46% vs. 36.39%), three-month (9.55% vs. 20.14% vs. 41.84%), and one-year mortality (11.57% vs. 23.76% vs. 46.60%). Building on this observation, we employed the Cox proportional hazards regression model to assess the impact of BAR segmentation on survival. Compared to Group 1, Groups 2 and 3 had significantly higher hazard ratios (95% confidence interval (CI)) for one-month mortality: 1.77 (1.37-2.30) and 3.17 (2.17-4.62), respectively. To further underscore the clinical potential of BAR as a standalone measure, we compared its performance to established clinical scores, like sequential organ failure assessment (SOFA), GCS, and acute physiology score III(APS-III), using receiver operator characteristic curve (ROC) analysis. Notably, the AUC values (95%CI) of the BAR were 0.67 (0.64-0.70), 0.68 (0.65-0.70), and 0.68 (0.65-0.70) for one-month mortality, three-month mortality, and one-year mortality. The AUC value of the SOFA did not significantly differ from that of the BAR. In conclusion, the BAR is a highly influential factor in predicting mortality in TBI patients and should be given careful consideration in future TBI prediction research. The blood urea nitrogen-to-albumin ratio may predict mortality in TBI patients.

The State of Machine Learning in Outcomes Prediction of Transsphenoidal Surgery: A Systematic Review

The purpose of this analysis is to assess the use of machine learning (ML) algorithms in the prediction of postoperative outcomes, including complications, recurrence, and death in transsphenoidal surgery. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we systematically reviewed all papers that used at least one ML algorithm to predict outcomes after transsphenoidal surgery. We searched Scopus, PubMed, and Web of Science databases for studies published prior to May 12, 2021. We identified 13 studies enrolling 5,048 patients. We extracted the general characteristics of each study; the sensitivity, specificity, area under the curve (AUC) of the ML models developed as well as the features identified as important by the ML models. We identified 12 studies with 5,048 patients that included ML algorithms for adenomas, three with 1807 patients specifically for acromegaly, and five with 2105 patients specifically for Cushing's disease. Nearly all were single-institution studies. The studies used a heterogeneous mix of ML algorithms and features to build predictive models. All papers reported an AUC greater than 0.7, which indicates clinical utility. ML algorithms have the potential to predict postoperative outcomes of transsphenoidal surgery and can improve patient care. Ensemble algorithms and neural networks were often top performers when compared with other ML algorithms. Biochemical and preoperative features were most likely to be selected as important by ML models. Inexplicability remains a challenge, but algorithms such as local interpretable model-agnostic explanation or Shapley value can increase explainability of ML algorithms. Our analysis shows that ML algorithms have the potential to greatly assist surgeons in clinical decision making.

Machine learning as a tool to identify inpatients who are not at risk of adverse drug events in a large dataset of a tertiary care hospital in the USA

AIMS

Adverse drug events (ADEs) are a major threat to inpatients in the United States of America (USA). It is unknown how well machine learning (ML) is able to predict whether or not a patient will suffer from an ADE during hospital stay based on data available at hospital admission for emergency department patients of all ages (binary classification task). It is further unknown whether ML is able to outperform logistic regression (LR) in doing so, and which variables are the most important predictors.

METHODS

In this study, 5 ML models- namely a random forest, gradient boosting machine (GBM), ridge regression, least absolute shrinkage and selection operator (LASSO) regression, and elastic net regression-as well as a LR were trained and tested for the prediction of inpatient ADEs identified using ICD-10-CM codes based on comprehensive previous work in a diverse population. In total, 210 181 observations from patients who were admitted to a large tertiary care hospital after emergency department stay between 2011 and 2019 were included. The area under the receiver operating characteristics curve (AUC) and AUC-precision-recall (AUC-PR) were used as primary performance indicators.

RESULTS

Tree-based models performed best with respect to AUC and AUC-PR. The gradient boosting machine (GBM) reached an AUC of 0.747 (95% confidence interval (CI): 0.735 to 0.759) and an AUC-PR of 0.134 (95% CI: 0.131 to 0.137) on unforeseen test data, while the random forest reached an AUC of 0.743 (95% CI: 0.731 to 0.755) and an AUC-PR of 0.139 (95% CI: 0.135 to 0.142), respectively. ML statistically significantly outperformed LR both on AUC and AUC-PR. Nonetheless, overall, models did not differ much with respect to their performance. Most important predictors were admission type, temperature and chief complaint for the best performing model (GBM).

CONCLUSIONS

The study demonstrated a first application of ML to predict inpatient ADEs based on ICD-10-CM codes, and a comparison with LR. Future research should address concerns arising from low precision and related problems.

Development of postoperative delirium prediction models in patients undergoing cardiovascular surgery using machine learning algorithms

Associations between delirium and postoperative adverse events in cardiovascular surgery have been reported and the preoperative identification of high-risk patients of delirium is needed to implement focused interventions. We aimed to develop and validate machine learning models to predict post-cardiovascular surgery delirium. Patients aged ≥ 40 years who underwent cardiovascular surgery at a single hospital were prospectively enrolled. Preoperative and intraoperative factors were assessed. Each patient was evaluated for postoperative delirium 7 days after surgery. We developed machine learning models using the Bernoulli naive Bayes, Support vector machine, Random forest, Extra-trees, and XGBoost algorithms. Stratified fivefold cross-validation was performed for each developed model. Of the 87 patients, 24 (27.6%) developed postoperative delirium. Age, use of psychotropic drugs, cognitive function (Mini-Cog < 4), index of activities of daily living (Barthel Index < 100), history of stroke or cerebral hemorrhage, and eGFR (estimated glomerular filtration rate) < 60 were selected to develop delirium prediction models. The Extra-trees model had the best area under the receiver operating characteristic curve (0.76 [standard deviation 0.11]; sensitivity: 0.63; specificity: 0.78). XGBoost showed the highest sensitivity (AUROC, 0.75 [0.07]; sensitivity: 0.67; specificity: 0.79). Machine learning algorithms could predict post-cardiovascular delirium using preoperative data.Trial registration: UMIN-CTR (ID; UMIN000049390).

Predicting whether patients will achieve minimal clinically important differences following hip or knee arthroplasty

Aims

A substantial fraction of patients undergoing knee arthroplasty (KA) or hip arthroplasty (HA) do not achieve an improvement as high as the minimal clinically important difference (MCID), i.e. do not achieve a meaningful improvement. Using three patient-reported outcome measures (PROMs), our aim was: 1) to assess machine learning (ML), the simple pre-surgery PROM score, and logistic-regression (LR)-derived performance in their prediction of whether patients undergoing HA or KA achieve an improvement as high or higher than a calculated MCID; and 2) to test whether ML is able to outperform LR or pre-surgery PROM scores in predictive performance.

Methods

MCIDs were derived using the change difference method in a sample of 1,843 HA and 1,546 KA patients. An artificial neural network, a gradient boosting machine, least absolute shrinkage and selection operator (LASSO) regression, ridge regression, elastic net, random forest, LR, and pre-surgery PROM scores were applied to predict MCID for the following PROMs: EuroQol five-dimension, five-level questionnaire (EQ-5D-5L), EQ visual analogue scale (EQ-VAS), Hip disability and Osteoarthritis Outcome Score-Physical Function Short-form (HOOS-PS), and Knee injury and Osteoarthritis Outcome Score-Physical Function Short-form (KOOS-PS).

Results

Predictive performance of the best models per outcome ranged from 0.71 for HOOS-PS to 0.84 for EQ-VAS (HA sample). ML statistically significantly outperformed LR and pre-surgery PROM scores in two out of six cases.

Conclusion

MCIDs can be predicted with reasonable performance. ML was able to outperform traditional methods, although only in a minority of cases.

Performance of advanced machine learning algorithms overlogistic regression in predicting hospital readmissions: A meta-analysis

Objectives

Machine learning algorithms are being increasingly used for predicting hospital readmissions. This meta-analysis evaluated the performance of logistic regression (LR) and machine learning (ML) models for the prediction of 30-day hospital readmission among patients in the US.

Methods

Electronic databases (i.e., Medline, PubMed, and Embase) were searched from January 2015 to December 2019. Only studies in the English language were included. Two reviewers performed studies screening, quality appraisal, and data collection. The quality of the studies was assessed using the Quality in Prognosis Studies (QUIPS) tool. Model performance was evaluated using the Area Under the Curve (AUC). A random-effects meta-analysis was performed using STATA 16.

Results

Nine studies were included based on the selection criteria. The most common ML techniques were tree-based methods such as boosting and random forest. Most of the studies had a low risk of bias (8/9). The AUC was greater with ML to predict 30-day all-cause hospital readmission compared with LR [Mean Difference (MD): 0.03; 95% Confidence Interval (CI) 0.01-0.05]. Subgroup analyses found that deep-learning methods had a better performance compared with LR (MD 0.06; 95% CI, 0.04-0.09), followed by neural networks (MD: 0.03; 95% CI, 0.03-0.03), while the AUCs of the tree-based (MD: 0.02; 95% CI -0.00-0.04) and kernel-based (MD: 0.02; 95% CI 0.02 (-0.13-0.16) methods were no different compared to LR. More than half of the studies evaluated heart failure-related rehospitalization (N = 5). For the readmission prediction among heart failure patients, ML performed better compared with LR, with a mean difference in AUC of 0.04 (95% CI, 0.01-0.07). The leave-one-out sensitivity analysis confirmed the robustness of the findings.

Conclusion

Multiple ML methods were used to predict 30-day all-cause hospital readmission. Performance varied across the ML methods, with deep-learning methods showing the best performance over the LR.

Computer-aided screening of aspiration risks in dysphagia with wearable technology: a Systematic Review and meta-analysis on test accuracy

Aspiration caused by dysphagia is a prevalent problem that causes serious health consequences and even death. Traditional diagnostic instruments could induce pain, discomfort, nausea, and radiation exposure. The emergence of wearable technology with computer-aided screening might facilitate continuous or frequent assessments to prompt early and effective management. The objectives of this review are to summarize these systems to identify aspiration risks in dysphagic individuals and inquire about their accuracy. Two authors independently searched electronic databases, including CINAHL, Embase, IEEE Xplore^® Digital Library, PubMed, Scopus, and Web of Science (PROSPERO reference number: CRD42023408960). The risk of bias and applicability were assessed using QUADAS-2. Nine (n = 9) articles applied accelerometers and/or acoustic devices to identify aspiration risks in patients with neurodegenerative problems (e.g., dementia, Alzheimer's disease), neurogenic problems (e.g., stroke, brain injury), in addition to some children with congenital abnormalities, using videofluoroscopic swallowing study (VFSS) or fiberoptic endoscopic evaluation of swallowing (FEES) as the reference standard. All studies employed a traditional machine learning approach with a feature extraction process. Support vector machine (SVM) was the most famous machine learning model used. A meta-analysis was conducted to evaluate the classification accuracy and identify risky swallows. Nevertheless, we decided not to conclude the meta-analysis findings (pooled diagnostic odds ratio: 21.5, 95% CI, 2.7-173.6) because studies had unique methodological characteristics and major differences in the set of parameters/thresholds, in addition to the substantial heterogeneity and variations, with sensitivity levels ranging from 21.7% to 90.0% between studies. Small sample sizes could be a critical problem in existing studies (median = 34.5, range 18-449), especially for machine learning models. Only two out of the nine studies had an optimized model with sensitivity over 90%. There is a need to enlarge the sample size for better generalizability and optimize signal processing, segmentation, feature extraction, classifiers, and their combinations to improve the assessment performance. Systematic Review Registration: (https://www.crd.york.ac.uk/prospero/), identifier (CRD42023408960).

What We Are Missing: Using Machine Learning Models to Predict Vitamin C Deficiency in Patients with Metabolic and Bariatric Surgery

PURPOSE

Vitamin C (VC) is implicated in many physiological pathways. Vitamin C deficiency (VCD) can compromise the health of patients with metabolic and bariatric surgery (patients). As symptoms of VCD are elusive and data on VCD in patients is scarce, we aim to characterize patients with measured VC levels, investigate the association of VCD with other lab abnormalities, and create predictive models of VCD using machine learning (ML).

METHODS

A retrospective chart review of patients seen from 2017 to 2021 at a tertiary care center in Northeastern USA was conducted. A 1:4 case mix of patients with VC measured to a random sample of patients without VC measured was created for comparative purposes. ML models (BayesNet and random forest) were used to create predictive models and estimate the prevalence of VCD patients.

RESULTS

Of 5946 patients reviewed, 187 (3.1%) had VC measures, and 73 (39%) of these patients had VC<23 μmol/L(VCD. When comparing patients with VCD to patients without VCD, the ML algorithms identified a higher risk of VCD in patients deficient in vitamin B1, D, calcium, potassium, iron, and blood indices. ML models reached 70% accuracy. Applied to the testing sample, a "true" VCD prevalence of ~20% was predicted, among whom ~33% had scurvy levels (VC<11 μmol/L).

CONCLUSION

Our models suggest a much higher level of patients have VCD than is reflected in the literature. This indicates a high proportion of patients remain potentially undiagnosed for VCD and are thus at risk for postoperative morbidity and mortality.

Small molecule biomarker discovery: Proposed workflow for LC-MS-based clinical research projects

Mass spectrometry focusing on small endogenous molecules has become an integral part of biomarker discovery in the pursuit of an in-depth understanding of the pathophysiology of various diseases, ultimately enabling the application of personalized medicine. While LC-MS methods allow researchers to gather vast amounts of data from hundreds or thousands of samples, the successful execution of a study as part of clinical research also requires knowledge transfer with clinicians, involvement of data scientists, and interactions with various stakeholders. The initial planning phase of a clinical research project involves specifying the scope and design, and engaging relevant experts from different fields. Enrolling subjects and designing trials rely largely on the overall objective of the study and epidemiological considerations, while proper pre-analytical sample handling has immediate implications on the quality of analytical data. Subsequent LC-MS measurements may be conducted in a targeted, semi-targeted, or non-targeted manner, resulting in datasets of varying size and accuracy. Data processing further enhances the quality of data and is a prerequisite for in-silico analysis. Nowadays, the evaluation of such complex datasets relies on a mix of classical statistics and machine learning applications, in combination with other tools, such as pathway analysis and gene set enrichment. Finally, results must be validated before biomarkers can be used as prognostic or diagnostic decision-making tools. Throughout the study, quality control measures should be employed to enhance the reliability of data and increase confidence in the results. The aim of this graphical review is to provide an overview of the steps to be taken when conducting an LC-MS-based clinical research project to search for small molecule biomarkers.

Using Machine Learning to make nanomaterials sustainable

Sustainable development is a key challenge for contemporary human societies; failure to achieve sustainability could threaten human survival. In this review article, we illustrate how Machine Learning (ML) could support more sustainable development, covering the basics of data gathering through each step of the Environmental Risk Assessment (ERA). The literature provides several examples showing how ML can be employed in most steps of a typical ERA.A key observation is that there are currently no clear guidance for using such autonomous technologies in ERAs or which standards/checks are required. Steering thus seems to be the most important task for supporting the use of ML in the ERA of nano- and smart-materials. Resources should be devoted to developing a strategy for implementing ML in ERA with a strong emphasis on data foundations, methodologies, and the related sensitivities/uncertainties. We should recognise historical errors and biases (e.g., in data) to avoid embedding them during ML programming.

Using automated methods to detect safety problems with health information technology: a scoping review

OBJECTIVE

To summarize the research literature evaluating automated methods for early detection of safety problems with health information technology (HIT).

MATERIALS AND METHODS

We searched bibliographic databases including MEDLINE, ACM Digital, Embase, CINAHL Complete, PsycINFO, and Web of Science from January 2010 to June 2021 for studies evaluating the performance of automated methods to detect HIT problems. HIT problems were reviewed using an existing classification for safety concerns. Automated methods were categorized into rule-based, statistical, and machine learning methods, and their performance in detecting HIT problems was assessed. The review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta Analyses extension for Scoping Reviews statement.

RESULTS

Of the 45 studies identified, the majority (n = 27, 60%) focused on detecting use errors involving electronic health records and order entry systems. Machine learning (n = 22) and statistical modeling (n = 17) were the most common methods. Unsupervised learning was used to detect use errors in laboratory test results, prescriptions, and patient records while supervised learning was used to detect technical errors arising from hardware or software issues. Statistical modeling was used to detect use errors, unauthorized access, and clinical decision support system malfunctions while rule-based methods primarily focused on use errors.

CONCLUSIONS

A wide variety of rule-based, statistical, and machine learning methods have been applied to automate the detection of safety problems with HIT. Many opportunities remain to systematically study their application and effectiveness in real-world settings.

Machine Learning-Based Mortality Prediction Model for Critically Ill Cancer Patients Admitted to the Intensive Care Unit (CanICU)

BACKGROUND

Although cancer patients are increasingly admitted to the intensive care unit (ICU) for cancer- or treatment-related complications, improved mortality prediction remains a big challenge. This study describes a new ML-based mortality prediction model for critically ill cancer patients admitted to ICU.

PATIENTS AND METHODS

We developed CanICU, a machine learning-based 28-day mortality prediction model for adult cancer patients admitted to ICU from Medical Information Mart for Intensive Care (MIMIC) database in the USA (n = 766), Yonsei Cancer Center (YCC, n = 3571), and Samsung Medical Center in Korea (SMC, n = 2563) from 2 January 2008 to 31 December 2017. The accuracy of CanICU was measured using sensitivity, specificity, and area under the receiver operating curve (AUROC).

RESULTS

A total of 6900 patients were included, with a 28-day mortality of 10.2%/12.7%/36.6% and a 1-year mortality of 30.0%/36.6%/58.5% in the YCC, SMC, and MIMIC-III cohort. Nine clinical and laboratory factors were used to construct the classifier using a random forest machine-learning algorithm. CanICU had 96% sensitivity/73% specificity with the area under the receiver operating characteristic (AUROC) of 0.94 for 28-day, showing better performance than current prognostic models, including the Acute Physiology and Chronic Health Evaluation (APACHE) or Sequential Organ Failure Assessment (SOFA) score. Application of CanICU in two external data sets across the countries yielded 79-89% sensitivity, 58-59% specificity, and 0.75-0.78 AUROC for 28-day mortality. The CanICU score was also correlated with one-year mortality with 88-93% specificity.

CONCLUSION

CanICU offers improved performance for predicting mortality in critically ill cancer patients admitted to ICU. A user-friendly online implementation is available and should be valuable for better mortality risk stratification to allocate ICU care for cancer patients.

The application of machine learning to predict high-cost patients: A performance-comparison of different models using healthcare claims data

Our aim was to predict future high-cost patients with machine learning using healthcare claims data. We applied a random forest (RF), a gradient boosting machine (GBM), an artificial neural network (ANN) and a logistic regression (LR) to predict high-cost patients in the following year. Therefore, we exploited routinely collected sickness funds claims and cost data of the years 2016, 2017 and 2018. Various specifications of each algorithm were trained and cross-validated on training data (n = 20,984) with claims and cost data from 2016 and outcomes from 2017. The best performing specifications of each algorithm were selected based on validation dataset performance. For performance comparison, selected models were applied to unforeseen data with features of the year 2017 and outcomes of the year 2018 (n = 21,146). The RF was the best performing algorithm measured by the area under the receiver operating curve (AUC) with a value of 0.883 (95% confidence interval (CI): 0.872-0.893) on test data, followed by the GBM (AUC = 0.878; 95% CI: 0.867-0.889). The ANN (AUC = 0.846; 95% CI: 0.834-0.857) and LR (AUC = 0.839; 95% CI: 0.826-0.852) were significantly outperformed by the GBM and the RF. All ML algorithms and the LR performed ´good´ (i.e. 0.9 > AUC ≥ 0.8). We were able to develop machine learning models that predict high-cost patients with 'good' performance facilitating routinely collected sickness fund claims and cost data. We found that tree-based models performed best and outperformed the ANN and LR.

Predicting Bacteremia among Septic Patients Based on ED Information by Machine Learning Methods: A Comparative Study

Introduction: Bacteremia is a common but life-threatening infectious disease. However, a well-defined rule to assess patient risk of bacteremia and the urgency of blood culture is lacking. The aim of this study is to establish a predictive model for bacteremia in septic patients using available big data in the emergency department (ED) through logistic regression and other machine learning (ML) methods. Material and Methods: We conducted a retrospective cohort study at the ED of National Cheng Kung University Hospital in Taiwan from January 2015 to December 2019. ED adults (≥18 years old) with systemic inflammatory response syndrome and receiving blood cultures during the ED stay were included. Models I and II were established based on logistic regression, both of which were derived from support vector machine (SVM) and random forest (RF). Net reclassification index was used to determine which model was superior. Results: During the study period, 437,969 patients visited the study ED, and 40,395 patients were enrolled. Patients diagnosed with bacteremia accounted for 7.7% of the cohort. The area under the receiver operating curve (AUROC) in models I and II was 0.729 (95% CI, 0.718–0.740) and 0.731 (95% CI, 0.721–0.742), with Akaike information criterion (AIC) of 16,840 and 16,803, respectively. The performance of model II was superior to that of model I. The AUROC values of models III and IV in the validation dataset were 0.730 (95% CI, 0.713–0.747) and 0.705 (0.688–0.722), respectively. There is no statistical evidence to support that the performance of the model created with logistic regression is superior to those created by SVM and RF. Discussion: The advantage of the SVM or RF model is that the prediction model is more elastic and not limited to a linear relationship. The advantage of the LR model is that it is easy to explain the influence of the independent variable on the response variable. These models could help medical staff identify high-risk patients and prevent unnecessary antibiotic use. The performance of SVM and RF was not inferior to that of logistic regression. Conclusions: We established models that provide discrimination in predicting bacteremia among patients with sepsis. The reported results could inspire researchers to adopt ML in their development of prediction algorithms.

Machine Learning Methods in Health Economics and Outcomes Research-The PALISADE Checklist: A Good Practices Report of an ISPOR Task Force

Advances in machine learning (ML) and artificial intelligence offer tremendous potential benefits to patients. Predictive analytics using ML are already widely used in healthcare operations and care delivery, but how can ML be used for health economics and outcomes research (HEOR)? To answer this question, ISPOR established an emerging good practices task force for the application of ML in HEOR. The task force identified 5 methodological areas where ML could enhance HEOR: (1) cohort selection, identifying samples with greater specificity with respect to inclusion criteria; (2) identification of independent predictors and covariates of health outcomes; (3) predictive analytics of health outcomes, including those that are high cost or life threatening; (4) causal inference through methods, such as targeted maximum likelihood estimation or double-debiased estimation-helping to produce reliable evidence more quickly; and (5) application of ML to the development of economic models to reduce structural, parameter, and sampling uncertainty in cost-effectiveness analysis. Overall, ML facilitates HEOR through the meaningful and efficient analysis of big data. Nevertheless, a lack of transparency on how ML methods deliver solutions to feature selection and predictive analytics, especially in unsupervised circumstances, increases risk to providers and other decision makers in using ML results. To examine whether ML offers a useful and transparent solution to healthcare analytics, the task force developed the PALISADE Checklist. It is a guide for balancing the many potential applications of ML with the need for transparency in methods development and findings.

What do we really know about the drivers of undeclared work? An evaluation of the current state of affairs using machine learning

It is nowadays widely understood that undeclared work cannot be efficiently combated without a holistic view on the mechanisms underlying its existence. However, the question remains whether we possess all the pieces of the holistic puzzle. To fill the gap, in this paper, we test if the features so far known to affect the behaviour of taxpayers are sufficient to detect noncompliance with outstanding precision. This is done by training seven supervised machine learning models on the compilation of data from the 2019 Special Eurobarometer on undeclared work and relevant figures from other sources. The conducted analysis not only does attest to the completeness of our knowledge concerning the drivers of undeclared work but also paves the way for wide usage of artificial intelligence in monitoring and confronting this detrimental practice. The study, however, exposes the necessity of having at disposal considerably larger datasets compared to those currently available if successful real-world applications of machine learning are to be achieved in this field. Alongside the apparent theoretical contribution, this paper is thus also expected to be of particular importance for policymakers, whose efforts to tackle tax evasion will have to be expedited in the period after the COVID-19 pandemic.

Machine learning models for diabetes management in acute care using electronic medical records: A systematic review

BACKGROUND

Machine learning (ML) is a subset of Artificial Intelligence (AI) that is used to predict and potentially prevent adverse patient outcomes. There is increasing interest in the application of these models in digital hospitals to improve clinical decision-making and chronic disease management, particularly for patients with diabetes. The potential of ML models using electronic medical records (EMR) to improve the clinical care of hospitalised patients with diabetes is currently unknown.

OBJECTIVE

The aim was to systematically identify and critically review the published literature examining the development and validation of ML models using EMR data for improving the care of hospitalised adult patients with diabetes.

METHODS

The Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) guidelines were followed. Four databases were searched (Embase, PubMed, IEEE and Web of Science) for studies published between January 2010 to January 2022. The reference lists of the eligible articles were manually searched. Articles that examined adults and both developed and validated ML models using EMR data were included. Studies conducted in primary care and community care settings were excluded. Studies were independently screened and data was extracted using Covidence® systematic review software. For data extraction and critical appraisal, the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) was followed. Risk of bias was assessed using the Prediction model Risk Of Bias Assessment Tool (PROBAST). Quality of reporting was assessed by adherence to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guideline. The IJMEDI checklist was followed to assess quality of ML models and the reproducibility of their outcomes. The external validation methodology of the studies was appraised.

RESULTS

Of the 1317 studies screened, twelve met inclusion criteria. Eight studies developed ML models to predict disglycaemic episodes for hospitalized patients with diabetes, one study developed a ML model to predict total insulin dosage, two studies predicted risk of readmission, and one study improved the prediction of hospital readmission for inpatients with diabetes. All included studies were heterogeneous with regard to ML types, cohort, input predictors, sample size, performance and validation metrics and clinical outcomes. Two studies adhered to the TRIPOD guideline. The methodological reporting of all the studies was evaluated to be at high risk of bias. The quality of ML models in all studies was assessed as poor. Robust external validation was not performed on any of the studies. No models were implemented or evaluated in routine clinical care.

CONCLUSIONS

This review identified a limited number of ML models which were developed to improve inpatient management of diabetes. No ML models were implemented in real hospital settings. Future research needs to enhance the development, reporting and validation steps to enable ML models for integration into routine clinical care.

Comparing LASSO and random forest models for predicting neurological dysfunction among fluoroquinolone users

BACKGROUND

Fluoroquinolones are associated with central (CNS) and peripheral (PNS) nervous system symptoms, and predicting the risk of these outcomes may have important clinical implications. Both LASSO and random forest are appealing modeling methods, yet it is not clear which method performs better for clinical risk prediction.

PURPOSE

To compare models developed using LASSO versus random forest for predicting neurological dysfunction among fluoroquinolone users.

METHODS

We developed and validated risk prediction models using claims data from a commercially insured population. The study cohort included adults dispensed an oral fluoroquinolone, and outcomes were CNS and PNS dysfunction. Model predictors included demographic variables, comorbidities and medications known to be associated with neurological symptoms, and several healthcare utilization predictors. We assessed the accuracy and calibration of these models using measures including AUC, calibration curves, and Brier scores.

RESULTS

The underlying cohort contained 16 533 (1.18%) individuals with CNS dysfunction and 46 995 (3.34%) individuals with PNS dysfunction during 120 days of follow-up. For CNS dysfunction, LASSO had an AUC of 0.81 (95% CI: 0.80, 0.82), while random forest had an AUC of 0.80 (95% CI: 0.80, 0.81). For PNS dysfunction, LASSO had an AUC of 0.75 (95% CI: 0.74, 0.76) versus an AUC of 0.73 (95% CI: 0.73, 0.74) for random forest. Both LASSO models had better calibration, with Brier scores 0.17 (LASSO) versus 0.20 (random forest) for CNS dysfunction and 0.20 (LASSO) versus 0.25 (random forest) for PNS dysfunction.

CONCLUSIONS

LASSO outperformed random forest in predicting CNS and PNS dysfunction among fluoroquinolone users, and should be considered for modeling when the cohort is modest in size, when the number of model predictors is modest, and when predictors are primarily binary.

Can minimal clinically important differences in patient reported outcome measures be predicted by machine learning in patients with total knee or hip arthroplasty? A systematic review

OBJECTIVES

To systematically review studies using machine learning (ML) algorithms to predict whether patients undergoing total knee or total hip arthroplasty achieve an improvement as high or higher than the minimal clinically important differences (MCID) in patient reported outcome measures (PROMs) (classification problem).

METHODS

Studies were eligible to be included in the review if they collected PROMs both pre- and postintervention, reported the method of MCID calculation and applied ML. ML was defined as a family of models which automatically learn from data when selecting features, identifying nonlinear relations or interactions. Predictive performance must have been assessed using common metrics. Studies were searched on MEDLINE, PubMed Central, Web of Science Core Collection, Google Scholar and Cochrane Library. Study selection and risk of bias assessment (ROB) was conducted by two independent researchers.

RESULTS

517 studies were eligible for title and abstract screening. After screening title and abstract, 18 studies qualified for full-text screening. Finally, six studies were included. The most commonly applied ML algorithms were random forest and gradient boosting. Overall, eleven different ML algorithms have been applied in all papers. All studies reported at least fair predictive performance, with two reporting excellent performance. Sample size varied widely across studies, with 587 to 34,110 individuals observed. PROMs also varied widely across studies, with sixteen applied to TKA and six applied to THA. There was no single PROM utilized commonly in all studies. All studies calculated MCIDs for PROMs based on anchor-based or distribution-based methods or referred to literature which did so. Five studies reported variable importance for their models. Two studies were at high risk of bias.

DISCUSSION

No ML model was identified to perform best at the problem stated, nor can any PROM said to be best predictable. Reporting standards must be improved to reduce risk of bias and improve comparability to other studies.

Predicting Physician Consultations for Low Back Pain Using Claims Data and Population-Based Cohort Data-An Interpretable Machine Learning Approach

(1) Background: Predicting chronic low back pain (LBP) is of clinical and economic interest as LBP leads to disabilities and health service utilization. This study aims to build a competitive and interpretable prediction model; (2) Methods: We used clinical and claims data of 3837 participants of a population-based cohort study to predict future LBP consultations (ICD-10: M40.XX-M54.XX). Best subset selection (BSS) was applied in repeated random samples of training data (75% of data); scoring rules were used to identify the best subset of predictors. The rediction accuracy of BSS was compared to randomforest and support vector machines (SVM) in the validation data (25% of data); (3) Results: The best subset comprised 16 out of 32 predictors. Previous occurrence of LBP increased the odds for future LBP consultations (odds ratio (OR) 6.91 [5.05; 9.45]), while concomitant diseases reduced the odds (1 vs. 0, OR: 0.74 [0.57; 0.98], >1 vs. 0: 0.37 [0.21; 0.67]). The area-under-curve (AUC) of BSS was acceptable (0.78 [0.74; 0.82]) and comparable with SVM (0.78 [0.74; 0.82]) and randomforest (0.79 [0.75; 0.83]); (4) Conclusions: Regarding prediction accuracy, BSS has been considered competitive with established machine-learning approaches. Nonetheless, considerable misclassification is inherent and further refinements are required to improve predictions.

Social media mining in drug development-Fundamentals and use cases

The incorporation of patients' perspectives into drug discovery and development has become critically important from the viewpoint of accounting for modern-day business dynamics. There is a trend among patients to narrate their disease experiences on social media. The insights gained by analyzing the data pertaining to such social-media posts could be leveraged to support patient-centered drug development. Manual analysis of these data is nearly impossible, but artificial intelligence enables automated and cost-effective processing, also referred as social media mining (SMM). This paper discusses the fundamental SMM methods along with several relevant drug-development use cases.

Comparison of Multivariable Logistic Regression and Machine Learning Models for Predicting Bronchopulmonary Dysplasia or Death in Very Preterm Infants

Bronchopulmonary dysplasia (BPD) is the most prevalent and clinically significant complication of prematurity. Accurate identification of at-risk infants would enable ongoing intervention to improve outcomes. Although postnatal exposures are known to affect an infant's likelihood of developing BPD, most existing BPD prediction models do not allow risk to be evaluated at different time points, and/or are not suitable for use in ethno-diverse populations. A comprehensive approach to developing clinical prediction models avoids assumptions as to which method will yield the optimal results by testing multiple algorithms/models. We compared the performance of machine learning and logistic regression models in predicting BPD/death. Our main cohort included infants <33 weeks' gestational age (GA) admitted to a Canadian Neonatal Network site from 2016 to 2018 (n = 9,006) with all analyses repeated for the <29 weeks' GA subcohort (n = 4,246). Models were developed to predict, on days 1, 7, and 14 of admission to neonatal intensive care, the composite outcome of BPD/death prior to discharge. Ten-fold cross-validation and a 20% hold-out sample were used to measure area under the curve (AUC). Calibration intercepts and slopes were estimated by regressing the outcome on the log-odds of the predicted probabilities. The model AUCs ranged from 0.811 to 0.886. Model discrimination was lower in the <29 weeks' GA subcohort (AUCs 0.699-0.790). Several machine learning models had a suboptimal calibration intercept and/or slope (k-nearest neighbor, random forest, artificial neural network, stacking neural network ensemble). The top-performing algorithms will be used to develop multinomial models and an online risk estimator for predicting BPD severity and death that does not require information on ethnicity.

A Machine Learning Approach to Identify Predictors of Potentially Inappropriate Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) Use in Older Adults with Osteoarthritis

Evidence from some studies suggest that osteoarthritis (OA) patients are often prescribed non-steroidal anti-inflammatory drugs (NSAIDs) that are not in accordance with their cardiovascular (CV) or gastrointestinal (GI) risk profiles. However, no such study has been carried out in the United States. Therefore, we sought to examine the prevalence and predictors of potentially inappropriate NSAIDs use in older adults (age > 65) with OA using machine learning with real-world data from Optum De-identified Clinformatics^® Data Mart. We identified a retrospective cohort of eligible individuals using data from 2015 (baseline) and 2016 (follow-up). Potentially inappropriate NSAIDs use was identified using the type (COX-2 selective vs. non-selective) and length of NSAIDs use and an individual's CV and GI risk. Predictors of potentially inappropriate NSAIDs use were identified using eXtreme Gradient Boosting. Our study cohort comprised of 44,990 individuals (mean age 75.9 years). We found that 12.8% individuals had potentially inappropriate NSAIDs use, but the rate was disproportionately higher (44.5%) in individuals at low CV/high GI risk. Longer duration of NSAIDs use during baseline (AOR 1.02; 95% CI:1.02-1.02 for both non-selective and selective NSAIDs) was associated with a higher risk of potentially inappropriate NSAIDs use. Additionally, individuals with low CV/high GI (AOR 1.34; 95% CI:1.20-1.50) and high CV/low GI risk (AOR 1.61; 95% CI:1.34-1.93) were also more likely to have potentially inappropriate NSAIDs use. Heightened surveillance of older adults with OA requiring NSAIDs is warranted.

Role for machine learning in sex-specific prediction of successful electrical cardioversion in atrial fibrillation?

Objective

Electrical cardioversion is frequently performed to restore sinus rhythm in patients with persistent atrial fibrillation (AF). However, AF recurs in many patients and identifying the patients who benefit from electrical cardioversion is difficult. The objective was to develop sex-specific prediction models for successful electrical cardioversion and assess the potential of machine learning methods in comparison with traditional logistic regression.

Methods

In a retrospective cohort study, we examined several candidate predictors, including comorbidities, biochemistry, echocardiographic data, and medication. The outcome was successful cardioversion, defined as normal sinus rhythm immediately after the electrical cardioversion and no documented recurrence of AF within 3 months after. We used random forest and logistic regression models for sex-specific prediction.

Results

The cohort comprised 332 female and 790 male patients with persistent AF who underwent electrical cardioversion. Cardioversion was successful in 44.9% of the women and 49.9% of the men. The prediction errors of the models were high for both women (41.0% for machine learning and 48.8% for logistic regression) and men (46.0% for machine learning and 44.8% for logistic regression). Discrimination was modest for both machine learning (0.59 for women and 0.56 for men) and logistic regression models (0.60 for women and 0.59 for men), although the models were well calibrated.

Conclusions

Sex-specific machine learning and logistic regression models showed modest predictive performance for successful electrical cardioversion. Identifying patients who will benefit from cardioversion remains challenging in clinical practice. The high recurrence rate calls for thoroughly informed shared decision-making for electrical cardioversion.

Development of a Model for Predicting Early Discontinuation of Adjuvant Chemotherapy in Stage III Colon Cancer

PURPOSE

To develop a tool that can be used to predict early discontinuation of adjuvant chemotherapy among patients with stage III colon cancer.

PATIENTS AND METHODS

Through record linkage of Alberta administrative and tumor registry databases, we identified a cohort of individuals age ≥ 18 years who were diagnosed with stage III colon cancer and who received adjuvant chemotherapy in Alberta between 2004 and 2015. Early discontinuation was defined as receipt of < 5 months of a planned 6-month course of chemotherapy. By a systematic review of the literature and a survey of medical oncologists, the following candidate variables were identified: age (years), number of comorbidities (0, 1, ≥ 2), cancer stage (IIIC v IIIA-B), type of chemotherapy (fluorouracil, leucovorin, and oxaliplatin; capecitabine and oxaliplatin; or monotherapy), time from surgery to chemotherapy initiation (weeks), type of treatment facility (academic or community), and distance from home to treatment center (kilometers). Models developed using penalized logistic regression and the random forest algorithm were compared. Model performance was assessed using the C-statistic, Brier score, and a calibration plot. Internal validation was performed using the bootstrap method.

RESULTS

From an initial 3,115 patients identified, 1,378 were deemed eligible for inclusion. Of these patients, 474 patients (34.4%) failed to complete at least 5 months of chemotherapy. Although well calibrated, the penalized logistic regression model had poor discrimination (optimism-adjusted C-statistic, 0.63; 95% CI, 0.60 to 0.67). In contrast, the random forest model achieved adequate discrimination (optimism-adjusted C-statistic, 0.80; 95% CI, 0.79 to 0.82). Although the degree of calibration of the random forest was acceptable, it was slightly worse than that of the penalized logistic regression model.

CONCLUSION

Internal validation of our random forest model suggests that it may have clinical utility. Additional research regarding its external validation and clinical impact is needed.

Revisiting genome-wide association studies from statistical modelling to machine learning

Over the last decade, genome-wide association studies (GWAS) have discovered thousands of genetic variants underlying complex human diseases and agriculturally important traits. These findings have been utilized to dissect the biological basis of diseases, to develop new drugs, to advance precision medicine and to boost breeding. However, the potential of GWAS is still underexploited due to methodological limitations. Many challenges have emerged, including detecting epistasis and single-nucleotide polymorphisms (SNPs) with small effects and distinguishing causal variants from other SNPs associated through linkage disequilibrium. These issues have motivated advancements in GWAS analyses in two contrasting cultures-statistical modelling and machine learning. In this review, we systematically present the basic concepts and the benefits and limitations in both methods. We further discuss recent efforts to mitigate their weaknesses. Additionally, we summarize the state-of-the-art tools for detecting the missed signals, ultrarare mutations and gene-gene interactions and for prioritizing SNPs. Our work can offer both theoretical and practical guidelines for performing GWAS analyses and for developing further new robust methods to fully exploit the potential of GWAS.

Differences in the immune-inflammatory profiles of unipolar and bipolar depression

BACKGROUND

Major depressive disorder (MDD) and bipolar depression (BD) both share increased immune-inflammatory activation. However, there are unclear patterns of differences in peripheral immune profiles between them.

METHODS

We examined such differences in 245 MDD and 59 BD patients, recruited in the same center, who were in an acute depressive episode of moderate severity. Hierarchical binary logistic regression analyses and generalized linear models were used to compare levels of plasma biomarkers between groups and to predict dichotomous classification.

RESULTS

Interleukin (IL)-1β, tumor necrosis factor (TNF)-α, soluble TNF receptor (sTNFR)1, IL-12 and IL-10 were significantly higher in MDD than in BD, whereas IL-6, sTNFR2, IL-18, IL-33, ST2 (IL1R Like 1) and KLOTHO were significantly higher in BD than in MDD. Moreover, logistic regression analyses correctly classified BD and MDD patients with 98.1% accuracy, using a combination of IL-6, IL-8, ST2, sTNFR2 (directly associated with BD) and IL-12 and TNF-α (directly associated with MDD). Patients with MDD with melancholic features showed higher IL-1β levels than those without melancholia. The sTNFR1 / sTNFR2 ratio significantly predicted MDD and state and trait anxiety and negative affect. Results remained significant after covariate adjustment, including drug use.

LIMITATIONS

Cross-sectional study. Lack of control comparison group. Differences in exposure to medications among participants.

CONCLUSIONS

Differences in immune profiles between BD and MDD patients exist, especially for the compensatory immune-regulatory system (CIRS): increased IL-10 is the primary immune-regulatory mechanism in MDD, while increased sTNFR2 and KLOTHO are the primary regulatory mechanisms in BD.

Severity Index for Suspected Arbovirus (SISA): Machine learning for accurate prediction of hospitalization in subjects suspected of arboviral infection

BACKGROUND

Dengue, chikungunya, and Zika are arboviruses of major global health concern. Decisions regarding the clinical management of suspected arboviral infection are challenging in resource-limited settings, particularly when deciding on patient hospitalization. The objective of this study was to determine if hospitalization of individuals with suspected arboviral infections could be predicted using subject intake data.

METHODOLOGY/PRINCIPAL FINDINGS

Two prediction models were developed using data from a surveillance study in Machala, a city in southern coastal Ecuador with a high burden of arboviral infections. Data were obtained from subjects who presented at sentinel medical centers with suspected arboviral infection (November 2013 to September 2017). The first prediction model-called the Severity Index for Suspected Arbovirus (SISA)-used only demographic and symptom data. The second prediction model-called the Severity Index for Suspected Arbovirus with Laboratory (SISAL)-incorporated laboratory data. These models were selected by comparing the prediction ability of seven machine learning algorithms; the area under the receiver operating characteristic curve from the prediction of a test dataset was used to select the final algorithm for each model. After eliminating those with missing data, the SISA dataset had 534 subjects, and the SISAL dataset had 98 subjects. For SISA, the best prediction algorithm was the generalized boosting model, with an AUC of 0.91. For SISAL, the best prediction algorithm was the elastic net with an AUC of 0.94. A sensitivity analysis revealed that SISA and SISAL are not directly comparable to one another.

CONCLUSIONS/SIGNIFICANCE

Both SISA and SISAL were able to predict arbovirus hospitalization with a high degree of accuracy in our dataset. These algorithms will need to be tested and validated on new data from future patients. Machine learning is a powerful prediction tool and provides an excellent option for new management tools and clinical assessment of arboviral infection.

Suicide prediction models: a critical review of recent research with recommendations for the way forward

Suicide is a leading cause of death. A substantial proportion of the people who die by suicide come into contact with the health care system in the year before their death. This observation has resulted in the development of numerous suicide prediction tools to help target patients for preventive interventions. However, low sensitivity and low positive predictive value have led critics to argue that these tools have no clinical value. We review these tools and critiques here. We conclude that existing tools are suboptimal and that improvements, if they can be made, will require developers to work with more comprehensive predictor sets, staged screening designs, and advanced statistical analysis methods. We also conclude that although existing suicide prediction tools currently have little clinical value, and in some cases might do more harm than good, an even-handed assessment of the potential value of refined tools of this sort cannot currently be made because such an assessment would depend on evidence that currently does not exist about the effectiveness of preventive interventions. We argue that the only way to resolve this uncertainty is to link future efforts to develop or evaluate suicide prediction tools with concrete questions about specific clinical decisions aimed at reducing suicides and to evaluate the clinical value of these tools in terms of net benefit rather than sensitivity or positive predictive value. We also argue for a focus on the development of individualized treatment rules to help select the right suicide-focused treatments for the right patients at the right times. Challenges will exist in doing this because of the rarity of suicide even among patients considered high-risk, but we offer practical suggestions for how these challenges can be addressed.

Machine-Learning vs. Expert-Opinion Driven Logistic Regression Modelling for Predicting 30-Day Unplanned Rehospitalisation in Preterm Babies: A Prospective, Population-Based Study (EPIPAGE 2)

Introduction: Preterm babies are a vulnerable population that experience significant short and long-term morbidity. Rehospitalisations constitute an important, potentially modifiable adverse event in this population. Improving the ability of clinicians to identify those patients at the greatest risk of rehospitalisation has the potential to improve outcomes and reduce costs. Machine-learning algorithms can provide potentially advantageous methods of prediction compared to conventional approaches like logistic regression. Objective: To compare two machine-learning methods (least absolute shrinkage and selection operator (LASSO) and random forest) to expert-opinion driven logistic regression modelling for predicting unplanned rehospitalisation within 30 days in a large French cohort of preterm babies. Design, Setting and Participants: This study used data derived exclusively from the population-based prospective cohort study of French preterm babies, EPIPAGE 2. Only those babies discharged home alive and whose parents completed the 1-year survey were eligible for inclusion in our study. All predictive models used a binary outcome, denoting a baby's status for an unplanned rehospitalisation within 30 days of discharge. Predictors included those quantifying clinical, treatment, maternal and socio-demographic factors. The predictive abilities of models constructed using LASSO and random forest algorithms were compared with a traditional logistic regression model. The logistic regression model comprised 10 predictors, selected by expert clinicians, while the LASSO and random forest included 75 predictors. Performance measures were derived using 10-fold cross-validation. Performance was quantified using area under the receiver operator characteristic curve, sensitivity, specificity, Tjur's coefficient of determination and calibration measures. Results: The rate of 30-day unplanned rehospitalisation in the eligible population used to construct the models was 9.1% (95% CI 8.2-10.1) (350/3,841). The random forest model demonstrated both an improved AUROC (0.65; 95% CI 0.59-0.7; p = 0.03) and specificity vs. logistic regression (AUROC 0.57; 95% CI 0.51-0.62, p = 0.04). The LASSO performed similarly (AUROC 0.59; 95% CI 0.53-0.65; p = 0.68) to logistic regression. Conclusions: Compared to an expert-specified logistic regression model, random forest offered improved prediction of 30-day unplanned rehospitalisation in preterm babies. However, all models offered relatively low levels of predictive ability, regardless of modelling method.

A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models

OBJECTIVES

The objective of this study was to compare performance of logistic regression (LR) with machine learning (ML) for clinical prediction modeling in the literature.

STUDY DESIGN AND SETTING

We conducted a Medline literature search (1/2016 to 8/2017) and extracted comparisons between LR and ML models for binary outcomes.

RESULTS

We included 71 of 927 studies. The median sample size was 1,250 (range 72-3,994,872), with 19 predictors considered (range 5-563) and eight events per predictor (range 0.3-6,697). The most common ML methods were classification trees, random forests, artificial neural networks, and support vector machines. In 48 (68%) studies, we observed potential bias in the validation procedures. Sixty-four (90%) studies used the area under the receiver operating characteristic curve (AUC) to assess discrimination. Calibration was not addressed in 56 (79%) studies. We identified 282 comparisons between an LR and ML model (AUC range, 0.52-0.99). For 145 comparisons at low risk of bias, the difference in logit(AUC) between LR and ML was 0.00 (95% confidence interval, -0.18 to 0.18). For 137 comparisons at high risk of bias, logit(AUC) was 0.34 (0.20-0.47) higher for ML.

CONCLUSION

We found no evidence of superior performance of ML over LR. Improvements in methodology and reporting are needed for studies that compare modeling algorithms.

A Brief Review of Random Forests for Water Scientists and Practitioners and Their Recent History in Water Resources

Random forests (RF) is a supervised machine learning algorithm, which has recently started to gain prominence in water resources applications. However, existing applications are generally restricted to the implementation of Breiman’s original algorithm for regression and classification problems, while numerous developments could be also useful in solving diverse practical problems in the water sector. Here we popularize RF and their variants for the practicing water scientist, and discuss related concepts and techniques, which have received less attention from the water science and hydrologic communities. In doing so, we review RF applications in water resources, highlight the potential of the original algorithm and its variants, and assess the degree of RF exploitation in a diverse range of applications. Relevant implementations of random forests, as well as related concepts and techniques in the R programming language, are also covered.

A contemporary review of machine learning in otolaryngology-head and neck surgery

One of the key challenges with big data is leveraging the complex network of information to yield useful clinical insights. The confluence of massive amounts of health data and a desire to make inferences and insights on these data has produced a substantial amount of interest in machine-learning analytic methods. There has been a drastic increase in the otolaryngology literature volume describing novel applications of machine learning within the past 5 years. In this timely contemporary review, we provide an overview of popular machine-learning techniques, and review recent machine-learning applications in otolaryngology-head and neck surgery including neurotology, head and neck oncology, laryngology, and rhinology. Investigators have realized significant success in validated models with model sensitivities and specificities approaching 100%. Challenges remain in the implementation of machine-learning algorithms. This may be in part the unfamiliarity of these techniques to clinician leaders on the front lines of patient care. Spreading awareness and confidence in machine learning will follow with further validation and proof-of-value analyses that demonstrate model performance superiority over established methods. We are poised to see a greater influx of machine-learning applications to clinical problems in otolaryngology-head and neck surgery, and it is prudent for providers to understand the potential benefits and limitations of these technologies. Laryngoscope, 130:45-51, 2020.

Priority-Lasso: a simple hierarchical approach to the prediction of clinical outcome using multi-omics data

BACKGROUND

The inclusion of high-dimensional omics data in prediction models has become a well-studied topic in the last decades. Although most of these methods do not account for possibly different types of variables in the set of covariates available in the same dataset, there are many such scenarios where the variables can be structured in blocks of different types, e.g., clinical, transcriptomic, and methylation data. To date, there exist a few computationally intensive approaches that make use of block structures of this kind.

RESULTS

In this paper we present priority-Lasso, an intuitive and practical analysis strategy for building prediction models based on Lasso that takes such block structures into account. It requires the definition of a priority order of blocks of data. Lasso models are calculated successively for every block and the fitted values of every step are included as an offset in the fit of the next step. We apply priority-Lasso in different settings on an acute myeloid leukemia (AML) dataset consisting of clinical variables, cytogenetics, gene mutations and expression variables, and compare its performance on an independent validation dataset to the performance of standard Lasso models.

CONCLUSION

The results show that priority-Lasso is able to keep pace with Lasso in terms of prediction accuracy. Variables of blocks with higher priorities are favored over variables of blocks with lower priority, which results in easily usable and transportable models for clinical practice.

Random forest versus logistic regression: a large-scale benchmark experiment

BACKGROUND AND GOAL

The Random Forest (RF) algorithm for regression and classification has considerably gained popularity since its introduction in 2001. Meanwhile, it has grown to a standard classification approach competing with logistic regression in many innovation-friendly scientific fields.

RESULTS

In this context, we present a large scale benchmarking experiment based on 243 real datasets comparing the prediction performance of the original version of RF with default parameters and LR as binary classification tools. Most importantly, the design of our benchmark experiment is inspired from clinical trial methodology, thus avoiding common pitfalls and major sources of biases.

CONCLUSION

RF performed better than LR according to the considered accuracy measured in approximately 69% of the datasets. The mean difference between RF and LR was 0.029 (95%-CI =[0.022,0.038]) for the accuracy, 0.041 (95%-CI =[0.031,0.053]) for the Area Under the Curve, and - 0.027 (95%-CI =[-0.034,-0.021]) for the Brier score, all measures thus suggesting a significantly better performance of RF. As a side-result of our benchmarking experiment, we observed that the results were noticeably dependent on the inclusion criteria used to select the example datasets, thus emphasizing the importance of clear statements regarding this dataset selection process. We also stress that neutral studies similar to ours, based on a high number of datasets and carefully designed, will be necessary in the future to evaluate further variants, implementations or parameters of random forests which may yield improved accuracy compared to the original version with default values.

Extending Statistical Boosting

Background: Boosting algorithms to simultaneously estimate and select predictor effects in statistical models have gained substantial interest during the last decade.

Objectives: This review highlights recent methodological developments regarding boosting algorithms for statistical modelling especially focusing on topics relevant for biomedical research.

Methods: We suggest a unified framework for gradient boosting and likelihood-based boosting (statistical boosting) which have been addressed separately in the literature up to now.

Results: The methodological developments on statistical boosting during the last ten years can be grouped into three different lines of research: i) efforts to ensure variable selection leading to sparser models, ii) developments regarding different types of predictor effects and how to choose them, iii) approaches to extend the statistical boosting framework to new regression settings.

Conclusions: Statistical boosting algorithms have been adapted to carry out unbiased variable selection and automated model choice during the fitting process and can nowadays be applied in almost any regression setting in combination with a large amount of different types of predictor effects.