The impact of imputation quality on machine learning classifiers for datasets with missing values

Prediction of inhibitor development in previously untreated and minimally treated children with severe and moderately severe hemophilia A using a machine-learning network

BACKGROUND

Prediction of inhibitor development in patients with hemophilia A (HA) remains a challenge.

OBJECTIVES

To construct a predictive model for inhibitor development in HA using a network of clinical variables and biomarkers based on the individual similarity network.

METHODS

Previously untreated and minimally treated children with severe/moderately severe HA, participants of the HEMFIL Cohort Study, were followed up until reaching 75 exposure days (EDs) without inhibitor (INH-) or upon inhibitor development (INH+). Clinical data and biological samples were collected before the start of factor (F)VIII replacement (T0). A predictive model (HemfilNET) was built to compare the networks and potential global topological differences between INH- and INH+ at T0, considering the network robustness. For validation, the "leave-one-out" cross-validation technique was employed. Accuracy, precision, recall, and F1-score were used as evaluation metrics for the machine-learning model.

RESULTS

We included 95 children with HA (CHA), of whom 31 (33%) developed inhibitors. The algorithm, featuring 37 variables, identified distinct patterns of networks at T0 for INH+ and INH-. The accuracy of the model was 74.2% for CHA INH+ and 98.4% for INH-. By focusing the analysis on CHA with high-risk F8 mutations for inhibitor development, the accuracy in identifying CHA INH+ increased to 82.1%.

CONCLUSION

Our machine-learning algorithm demonstrated an overall accuracy of 90.5% for predicting inhibitor development in CHA, which further improved when restricting the analysis to CHA with a high-risk F8 genotype. However, our model requires validation in other cohorts. Yet, missing data for some variables hindered more precise predictions.

Machine Learning from Veno-Venous Extracorporeal Membrane Oxygenation Identifies Factors Associated with Neurological Outcomes

BACKGROUND

Neurological complications are common in patients receiving veno-venous extracorporeal membrane oxygenation (VV-ECMO) support. We used machine learning (ML) algorithms to identify predictors for neurological outcomes for these patients.

METHODS

All demographic, clinical, and circuit-related variables were extracted for adults with VV-ECMO support at a tertiary care center from 2016 to 2022. The primary outcome was good neurological outcome (GNO) at discharge defined as a modified Rankin Scale of 0-3.

RESULTS

Of 99 total VV-ECMO patients (median age = 48 years; 65% male), 37% had a GNO. The best performing ML model achieved an area under the receiver operating characteristic curve of 0.87. Feature importance analysis identified down-trending gas/sweep/blender flow, FiO2, and pump speed as the most salient features for predicting GNO.

CONCLUSION

Utilizing pre- as well as post-initiation variables, ML identified on-ECMO physiologic and pulmonary conditions that best predicted neurological outcomes.

Biophysical cartography of the native and human-engineered antibody landscapes quantifies the plasticity of antibody developability

Designing effective monoclonal antibody (mAb) therapeutics faces a multi-parameter optimization challenge known as "developability", which reflects an antibody's ability to progress through development stages based on its physicochemical properties. While natural antibodies may provide valuable guidance for mAb selection, we lack a comprehensive understanding of natural developability parameter (DP) plasticity (redundancy, predictability, sensitivity) and how the DP landscapes of human-engineered and natural antibodies relate to one another. These gaps hinder fundamental developability profile cartography. To chart natural and engineered DP landscapes, we computed 40 sequence- and 46 structure-based DPs of over two million native and human-engineered single-chain antibody sequences. We find lower redundancy among structure-based compared to sequence-based DPs. Sequence DP sensitivity to single amino acid substitutions varied by antibody region and DP, and structure DP values varied across the conformational ensemble of antibody structures. We show that sequence DPs are more predictable than structure-based ones across different machine-learning tasks and embeddings, indicating a constrained sequence-based design space. Human-engineered antibodies localize within the developability and sequence landscapes of natural antibodies, suggesting that human-engineered antibodies explore mere subspaces of the natural one. Our work quantifies the plasticity of antibody developability, providing a fundamental resource for multi-parameter therapeutic mAb design.

Recent methodological advances in federated learning for healthcare

For healthcare datasets, it is often impossible to combine data samples from multiple sites due to ethical, privacy, or logistical concerns. Federated learning allows for the utilization of powerful machine learning algorithms without requiring the pooling of data. Healthcare data have many simultaneous challenges, such as highly siloed data, class imbalance, missing data, distribution shifts, and non-standardized variables, that require new methodologies to address. Federated learning adds significant methodological complexity to conventional centralized machine learning, requiring distributed optimization, communication between nodes, aggregation of models, and redistribution of models. In this systematic review, we consider all papers on Scopus published between January 2015 and February 2023 that describe new federated learning methodologies for addressing challenges with healthcare data. We reviewed 89 papers meeting these criteria. Significant systemic issues were identified throughout the literature, compromising many methodologies reviewed. We give detailed recommendations to help improve methodology development for federated learning in healthcare.

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.

Characterizing advanced heart failure risk and hemodynamic phenotypes using interpretable machine learning

BACKGROUND

Although previous risk models exist for advanced heart failure with reduced ejection fraction (HFrEF), few integrate invasive hemodynamics or support missing data. This study developed and validated a heart failure (HF) hemodynamic risk and phenotyping score for HFrEF, using Machine Learning (ML).

METHODS

Prior to modeling, patients in training and validation HF cohorts were assigned to 1 of 5 risk categories based on the composite endpoint of death, left ventricular assist device (LVAD) implantation or transplantation (DeLvTx), and rehospitalization in 6 months of follow-up using unsupervised clustering. The goal of our novel interpretable ML modeling approach, which is robust to missing data, was to predict this risk category (1, 2, 3, 4, or 5) using either invasive hemodynamics alone or a rich and inclusive feature set that included noninvasive hemodynamics (all features). The models were trained using the ESCAPE trial and validated using 4 advanced HF patient cohorts collected from previous trials, then compared with traditional ML models. Prediction accuracy for each of these 5 categories was determined separately for each risk category to generate 5 areas under the curve (AUCs, or C-statistics) for belonging to risk category 1, 2, 3, 4, or 5, respectively.

RESULTS

Across all outcomes, our models performed well for predicting the risk category for each patient. Accuracies of 5 separate models predicting a patient's risk category ranged from 0.896 +/- 0.074 to 0.969 +/- 0.081 for the invasive hemodynamics feature set and 0.858 +/- 0.067 to 0.997 +/- 0.070 for the all features feature set.

CONCLUSION

Novel interpretable ML models predicted risk categories with a high degree of accuracy. This approach offers a new paradigm for risk stratification that differs from prediction of a binary outcome. Prospective clinical evaluation of this approach is indicated to determine utility for selecting the best treatment approach for patients based on risk and prognosis.