Super Learner Research Articles

Parametric and nonparametric propensity score estimation in multilevel observational studies.

There has been growing interest in using nonparametric machine learning approaches for propensity score estimation in order to foster robustness against misspecification of the propensity score model. However, the vast majority of studies focused on single-level data settings, and research on nonparametric propensity score estimation in clustered data settings is scarce. In this article, we extend existing research by describing a general algorithm for incorporating random effects into a machine learning model, which we implemented for generalized boosted modeling (GBM). In a simulation study, we investigated the performance of logistic regression, GBM, and Bayesian additive regression trees for inverse probability of treatment weighting (IPW) when the data are clustered, the treatment exposure mechanism is nonlinear, and unmeasured cluster-level confounding is present. For each approach, we compared fixed and random effects propensity score models to single-level models and evaluated their use in both marginal and clustered IPW. We additionally investigated the performance of the standard Super Learner and the balance Super Learner. The results showed that when there was no unmeasured confounding, logistic regression resulted in moderate bias in both marginal and clustered IPW, whereas the nonparametric approaches were unbiased. In presence of cluster-level confounding, fixed and random effects models greatly reduced bias compared to single-level models in marginal IPW, with fixed effects GBM and fixed effects logistic regression performing best. Finally, clustered IPW was overall preferable to marginal IPW and the balance Super Learner outperformed the standard Super Learner, though neither worked as well as their best candidate model.

Merging machine learning and geostatistical approaches for spatial modeling of geoenergy resources

Geostatistics is the most commonly used probabilistic approach for modeling earth systems, including quality parameters of various geoenergy resources. In geostatistics, estimates, either on a point or block support, are generated as a spatially-weighted average of surrounding samples. The optimal weights are determined through the stationary variogram model which accounts for the spatial structure of the samples. Recently, efficient modeling workflows using various machine learning algorithms (MLAs) have been expanded to the spatial context for modeling geological heterogeneity. The flexible use of MLAs as a spatial estimation tool stems mainly from the fact that unlike kriging, they do not require any variogram, nor do they depend strongly on a prior stationarity assumption (i.e., second order stationarity). This study evaluates the performance of two MLAs (ensemble super learner and elliptical radial basis neural network), ordinary kriging, and hybrid spatial modeling approaches using ordinary intrinsic collocated cokriging. The aforementioned modeling techniques are compared for estimating resources for four coal variables (wash yield, ash yield, calorific value and thickness) as an example. The results suggest that MLAs, when implemented alone, do not outperform ordinary kriging, but the estimation accuracy of the final model, measured by the root mean squared error tends to subtly improve (1.7% for wash yield, 6.98% for ash yield, 4.94% for calorific value and 0.36% for seam thickness) when MLAs and geostatistical algorithms are merged through the hybrid spatial modeling approaches.

Super learner approach to predict total organic carbon using stacking machine learning models based on well logs

Determining the total organic carbon (TOC) content is essential information for risk assessment in oil exploration, as it is a parameter used for the characterization of hydrocarbon-generating rocks, considering that intervals rich in organic matter are the basic requirements for oil and gas accumulation. However, the determination of TOC can be costly, demanding destructive tests in samples from the source rock, expensive laboratory machinery, and specialized personnel. In this context, one notes the necessity of the computational methods to bypass those problems and that machine learning models emerge as an option. One approach to integrating machine learning methods improves performance and, consequently, the prediction quality is stacking models. This paper presents a super learner strategy, based on stacking approaches, as a surrogate model for TOC modeling. The super-learner has three levels in this structure containing different types of learners (machine learning methods), where two stack models from the first two levels. The following machine learning models were used in the building of super learner the K-Neighbors Nearest (KNN), Linear Regression (LR), Multi-layer Perceptron Neural Network (MLP), Random Forest (RF), Ridge Regression (RR), and Support Vector Regression (SVR). The proposed model was compared with standalone machine learning models and other canonical stacking models. The resulting super learner stacking model attained the best average performance for the TOC modeling (R = 0.897, R2 = 0.80, RMSE = 1.16, MAE = 0.93, and MAPE = 28.30%). The proposed approach produces an alternative data-driven efficient model for TOC prediction, resulting in reliable automated technology to assist oil and gas well management and decision-making.

Sensor-based indoor air temperature prediction using deep ensemble machine learning: An Australian urban environment case study

Accurate prediction of indoor temperature is critical for climate change adaptation and occupant health. The aim of this study is to investigate an improved deep ensemble machine learning framework (DEML), by adjusting the model architecture with several machine learning (ML) and deep learning (DL) approaches to forecast the sensor-based indoor temperature in the Australian urban environment. We collected ambient station-based temperatures, satellite-based outdoor climate characteristics, and low-cost sensor-based indoor environmental metrics from 96 devices from August 2019 to November 2022, and established DEML with a rolling windows approach to assess the prediction stability over time. The DEML model was compared with several benchmark models, including Random Forest (RF), Support Vector Machine (SVM), eXtreme Gradient Boosting (XGboost), Long-short term memory (LSTM), and Super Learner model (SL). A total of 13,715 days [median: 341 days; IQR (the interquartile range): 221–977 days] of low-cost sensor-based indoor temperature were included in 25 commercial and residential buildings across eight cities. The prediction performance of DEML was superior to the other five benchmark models in most of the sensors [coefficients of determination (R2) of 0.861–0.990 and root mean square error (RMSE) of 0.125–0.886 °C], followed by RF and SL algorithms. DEML consistently achieved high accuracy across different climate zones, seasons, and building types, which could be used as a crucial tool for optimizing energy use, maintaining occupant comfort and health, and adapting to the impacts of climate change.

Super learner ensemble model: A novel approach for predicting monthly copper price in future

Companies and governments dependent on copper mining need to be able to predict copper prices in order to make important decisions. Despite the nonlinear and nonstationary nature of copper prices, their periods may vary as they fluctuate due to potential growth, cyclical fluctuations and errors. A trend-cycle refers to the combination of trend and cyclical components. Trend-cycles are characterized by different characteristics, which are crucial to making predictions. Therefore, this study focuses on developing and proposing a novel model based on an ensemble machine learning technique to predict monthly copper prices in the future by using different soft computing methods, including multi-layer perception (MLP) neural network, support vector regression (SVR), and extreme gradient boosting (XGBoost). The monthly copper price dataset from August 2001 to August 2021 was gathered for this aim based on the 14 effective parameters. These parameters were selected based on the suggestions of previous research. The main novelty of this study is the development most accurate model to predict monthly copper prices using Cubist algorithm-based super learner model as the new predictive system. The results indicated that the proposed super learner models outperformed of MLP, SVR, and XGBoost models based on the determination coefficient (R-squared), value account for (VAF), root mean square of errors (RMSE), Accuracy (Acc) and Mean Absolute Relative Error (MARE). A comprehensive comparison of different artificial intelligence models demonstrates that MLP neural network is the best model for predicting monthly copper prices. However, the standalone models involving MLP, SVR, and XGBoost, presented higher error with an RMSE in the interval of [278.3826 - 502.6946], and MARE in the interval of [0.056 - 0.1277]. Hence, Cubist-based super learner can be employed as a reliable system to predict monthly copper prices in the future. Besides, this study presents a rational mathematical model based on gene expression programming (GEP) for copper price prediction in future and for use by other researchers. Noteworthy, the final step of the study was sensitivity analysis conducting, which the results revealed that “lead price” and “euro to USD” parameters have respectively the lowest and highest impact on the monthly copper price.

Super ensemble based streamflow simulation using multi-source remote sensing and ground gauged rainfall data fusion

Traditional data-driven streamflow predictions usually apply a single model with inconsistent performance in different variability conditions. These days model ensembles or merging the benefits of different models without losing the general character of the data are becoming a trend in hydrology. This study compared three super ensemble learners with eight base models. Twelve years of monthly rolled daily time series data in three river catchments of Ethiopia (Borkena watershed: Awash river basin), (Gummera watershed: Abay river basin), and (Sore watershed: Baro Akobo river basin) is used for single-step daily streamflow simulation using previous thirty-day input timesteps. Five input scenarios are applied: three vegetation indices, three remote sensing-based precipitation products, ground-gauged rainfall, all fused inputs, and selected inputs with the Recursive Feature Elimination (RFE) algorithm. The time series is then divided into training and testing datasets with a ratio of 80:20. The performance of the proposed models was evaluated using the Root Mean Squared Error (RMSE), coefficient of determination (R2), Mean Absolute Error (MAE), and Median Absolute Error (MEDAE). Finally, the result is presented with the corresponding five input scenarios. The catchment's and input scenario's average performance indicated that the three super ensemble learners outperformed the eight base models with relatively stable performance. The top-ranked WASE model exceeded the linear regression baseline by 13.3%. XGB, CNN-GRU, and LSTM proved the highest performance of the base models. This study also revealed that LSTM's key downside is its performance drop in the absence of feature selection criteria. In comparison, XGB showed its superior performance after controlling redundant inputs internally. Moreover, this study uniquely highlights the potential of remote sensing-based vegetation indices in the science of data-driven streamflow modelling for non-gauged catchments with no meteorological time series.

Association analysis between symptomology and herpesvirus IgG antibody concentrations in myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and multiple sclerosis

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and multiple sclerosis (MS) are two complex and multifactorial diseases whose patients experience persistent fatigue, cognitive impairment, among other shared symptoms. The onset of these diseases has also been linked to acute herpesvirus infections or their reactivations. In this work, we re-analyzed a previously-described dataset related to IgG antibody responses to 6 herpesviruses (CMV – cytomegalovirus; EBV – Epstein-Barr virus; HHV6 – human herpesvirus-6; HSV1 and HSV2 – herpes simplex virus-1 and -2, respectively; VZV – varicella-zoster virus) from the United Kingdom ME/CFS biobank. The primary goal was to report the underlying symptomology and its association with herpesvirus IgG antibodies using data from 4 disease-trigger-based subgroups of ME/CFS patients (n = 222) and patients with MS (n = 46). The secondary objective was to assess whether serological data could distinguish ME/CFS and its subgroup from MS using a SuperLearner (SL) algorithm. There was evidence for a significant negative association between temporary eye insight disturbance and CMV antibody concentrations and for a significant positive association between bladder problems and EBV antibody concentrations in the MS group. In the ME/CFS or its subgroups, the most significant antibody-symptom association was obtained for increasing HSV1 antibody concentration and brain fog, a finding in line with a negative impact of HSV1 exposure on cognitive outcomes in both healthy and disease conditions. There was also evidence for a higher number of significant antibody-symptom associations in the MS group than in the ME/CFS group. When we combined all the serological data in an SL algorithm, we could distinguish three ME/CFS subgroups (unknown disease trigger, non-infection trigger, and an infection disease trigger confirmed in the lab at the time of the event) from the MS group. However, we could not find the same for the remaining ME/CFS group (related to an unconfirmed infection disease). In conclusion, IgG antibody data explains more the symptomology of MS patients than the one of ME/CFS patients. Given the fluctuating nature of symptoms in ME/CFS patients, the clinical implication of these findings remains to be determined with a longitudinal study. This study is likely to ascertain the robustness of the associations during natural disease course.

P-083 Associations of sperm epigenetic aging with semen parameters among men from the U.S. general population

Abstract Study question Is sperm epigenetic aging (SEA) associated with semen parameters among men in the general population? Summary answer While SEA was not associated with general semen parameters, advanced SEA was positively associated with sperm head defects such as length, perimeter, and pyriforms. What is known already We have previously shown that advanced SEA was strongly associated with longer time-to-pregnancy among couples in the general population. Study design, size, duration A population-based prospective cohort study of couples discontinuing contraception to become pregnant recruited from 16 US counties from 2005 to 2009. Participants/materials, setting, methods Sperm DNA methylation from 379 semen samples were assessed via Illumina EPIC Array and SEA was estimated using Super Learner, an ensemble machine learning algorithm. Linear regression models were employed to examine the associations between semen parameters and SEA adjusting for male age and current smoking. Main results and the role of chance None of the general semen characteristics such as count, concentration, motility or morphology were associated with SEA. However, several sperm head parameters were positively associated with SEA including length (β = 3.6, 95% confidence internal (CI): 1.01 - 6.23; p = 0.007); perimeter (β = 4.04, 95% CI: 0.1 – 0.05; p = 0.045) and pyriforms (β = 0.29, 95% CI: 0.1-0.49; p = 0.003). SEA was also inversely related to sperm elongation factor (β = -2.9, 95% CI: -4.8 - -1.1; p = 0.002). Limitations, reasons for caution This prospective cohort study consisted primarily of Caucasian men and women and thus large diverse cohorts are necessary to confirm the associations between SEA and sperm head defects in other races/ethnicities. Wider implications of the findings These data suggest that advanced sperm epigenetic aging may be related to improper sperm head condensation during spermatogenesis. Trial registration number N/A

Ceiling Effect of the Combined Norwegian and Danish Knee Ligament Registers Limits Anterior Cruciate Ligament Reconstruction Outcome Prediction

Background: Clinical tools based on machine learning analysis now exist for outcome prediction after primary anterior cruciate ligament reconstruction (ACLR). Relying partly on data volume, the general principle is that more data may lead to improved model accuracy. Purpose/Hypothesis: The purpose was to apply machine learning to a combined data set from the Norwegian and Danish knee ligament registers (NKLR and DKRR, respectively), with the aim of producing an algorithm that can predict revision surgery with improved accuracy relative to a previously published model developed using only the NKLR. The hypothesis was that the additional patient data would result in an algorithm that is more accurate. Study Design: Cohort study; Level of evidence, 3. Methods: Machine learning analysis was performed on combined data from the NKLR and DKRR. The primary outcome was the probability of revision ACLR within 1, 2, and 5 years. Data were split randomly into training sets (75%) and test sets (25%). There were 4 machine learning models examined: Cox lasso, random survival forest, gradient boosting, and super learner. Concordance and calibration were calculated for all 4 models. Results: The data set included 62,955 patients in which 5% underwent a revision surgical procedure with a mean follow-up of 7.6 ± 4.5 years. The 3 nonparametric models (random survival forest, gradient boosting, and super learner) performed best, demonstrating moderate concordance (0.67 [95% CI, 0.64-0.70]), and were well calibrated at 1 and 2 years. Model performance was similar to that of the previously published model (NKLR-only model: concordance, 0.67-0.69; well calibrated). Conclusion: Machine learning analysis of the combined NKLR and DKRR enabled prediction of the revision ACLR risk with moderate accuracy. However, the resulting algorithms were less user-friendly and did not demonstrate superior accuracy in comparison with the previously developed model based on patients from the NKLR alone, despite the analysis of nearly 63,000 patients. This ceiling effect suggests that simply adding more patients to current national knee ligament registers is unlikely to improve predictive capability and may prompt future changes to increase variable inclusion.

Modeling of microbial fuel cell power generation using machine learning-based super learner algorithms

Electricity generation from microbial fuel cells (MFCs) is a potential environment-friendly technology. This study provides Bayesian Algorithm (BA) based Support Vector Regression (SVR) and Boosted Regression Tree (BRT) as prospective super learner modeling tools (BA-SVR, BA-BRT) for predictions of electricity production from MFCs. The membrane thickness, external resistance, and anode area were considered independent variables, while power generation was taken as a response variable. The key novelties of this study include (i) hybridization of BA with SVR and BRT (separately) for forecasting power generation from fuel cells for the first time, (ii) performance comparison of the developed models (BA-SVR and BA-BRT) with the existing Response Surface Methodology (RSM) based on the coefficient of determination (R2), relative error (RE), mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), and computing efficiency, and the (iii) analysis of the models’ robustness by utilizing Gaussian white noise. Based on the performance indicators, the proposed super leaner models showed excellent performance compared to the existing M.J. Salar-García et al. RSM model. The BA-SVR model provided the lowest errors (MAE of 2.94, RSME of 7.2926, MAPE of 13.8341) with the highest R2 of 0.9981, compared to the BA-BRT and RSM models. The proposed BA-SVR model showed superior performance to the RSM and BA-BRT models in predicting the MFCs’ power generation, with a performance improvement of more than 90% regarding MAPE, as an example. The future prediction and high robustness of the proposed super learner model would ensure quick estimation for maximization of electricity generation that may lead to reducing massive lab trials and saving resources.

Predicting Seasonal Influenza Hospitalizations Using an Ensemble Super Learner: A Simulation Study.

Accurate forecasts can inform response to outbreaks. Most efforts in influenza forecasting have focused on predicting influenza-like activity, with fewer on influenza-related hospitalizations. We conducted a simulation study to evaluate a super learner's predictions of 3 seasonal measures of influenza hospitalizations in the United States: peak hospitalization rate, peak hospitalization week, and cumulative hospitalization rate. We trained an ensemble machine learning algorithm on 15,000 simulated hospitalization curves and generated weekly predictions. We compared the performance of the ensemble (weighted combination of predictions from multiple prediction algorithms), the best-performing individual prediction algorithm, and a naive prediction (median of a simulated outcome distribution). Ensemble predictions performed similarly to the naive predictions early in the season but consistently improved as the season progressed for all prediction targets. The best-performing prediction algorithm in each week typically had similar predictive accuracy compared with the ensemble, but the specific prediction algorithm selected varied by week. An ensemble super learner improved predictions of influenza-related hospitalizations, relative to a naive prediction. Future work should examine the super learner's performance using additional empirical data on influenza-related predictors (e.g., influenza-like illness). The algorithm should also be tailored to produce prospective probabilistic forecasts of selected prediction targets.

Predicting Miscarriage and Stillbirth Using Weighted Ensemble Machine Learning [ID: 1338167

INTRODUCTION: Prediction of miscarriage and stillbirth remains a clinical challenge. Prior efforts to use machine learning tools have not used an ensemble weighted machine learning approach, called “super learner,” which offers the opportunity to improve prediction performance by aggregating the outputs of constituent machine learning models and preferentially weighting the highest-performing models. METHODS: Data were obtained from hospital-wide electronic health records from a large academic institution. The sample comprised 13,337 patients who delivered between 2008 and 2019, 6,932 of whom experienced a miscarriage or stillbirth. Phecodes for ICD-9-CM and ICD-10-CM were used to define miscarriage and stillbirth and create comorbidity categories. The constituent models of the super learner were XGBoost, random forest, a regularized generalized linear model (both lasso and ridge regressions), and a support vector machine. The objective of this study was to develop a super learner algorithm to predict miscarriage and stillbirth. RESULTS: The super learner model predicted miscarriage and stillbirth classification with an area under the receiver operating characteristic curve of 0.94 and an accuracy of 92%. It used two models: random forest, weighted at 73%, and SVM, weighted at 27%. The most highly weighted predictors were amniotic cavity abnormalities, pelvic soft tissue abnormalities, and preeclampsia. CONCLUSION: A super learner performs comparably to other models in the literature. External validation is warranted. The promising results suggest that a super learner model can be used as a clinical decision support tool, supplementing clinical judgement in predicting miscarriage and stillbirth.

Modeling of capacitance for carbon-based supercapacitors using Super Learner algorithm

Due to some specifications such as high capacitance and power density, electrostatic double-layer capacitors (EDLCs) are more noticeable than other supercapacitors. Some physical and chemical properties, surface functional groups, and testing conditions affect the efficiency and in particular the capacitance of EDLCs with carbon-based electrodes. In this study, four machine learning models, including Super Learner (SL), Extremely Randomized Trees (Extra trees), Extreme learning machine (ELM), and Multivariate adaptive regression splines (MARS) were implemented to predict the EDLCs' capacitance based on different impressive properties. A large dataset was assigned to the 121 different carbonaceous electrodes collected under various conditions, including 13 physical and chemical properties: voltage window (V), specific surface area (SSA) and SSA of micropore, pore volume (PV), and micropore volume, the ratio of D-band and G-band (Id/Ig) and doping elements (inputs parameters). The results indicated that the SL model with the R2 values of 0.9781, 0.9717, and 0.9768 for training, testing, and total dataset, respectively, and the RSME value of 18.099 was the most accurate model in comparison with the others. Indeed, the sensitivity analysis results exhibited that SSA with the relevance factor of 0.323 is the most important feature in the capacitance of carbon-based electrodes, while the presence of boron, sulfur, fluorine, phosphorus doping elements and pore size can be ignored.

Disaggregating Latino nativity in equity research using electronic health records.

To develop and validate prediction models for inference of Latino nativity to advance health equity research. This study used electronic health records (EHRs) from 19,985 Latino children with self-reported country of birth seeking care from January 1, 2012 to December 31, 2018 at 456 community health centers (CHCs) across 15 states along with census-tract geocoded neighborhood composition and surname data. We constructed and evaluated the performance of prediction models within a broad machine learning framework (Super Learner) for the estimation of Latino nativity. Outcomes included binary indicators denoting nativity (US vs. foreign-born) and Latino country of birth (Mexican, Cuban, Guatemalan). The performance of these models was compared using the area under the receiver operating characteristics curve (AUC) from an externally withheld patient sample. Census surname lists, census neighborhood composition, and Forebears administrative data were linked to EHR data. Of the 19,985 Latino patients, 10.7% reported a non-US country of birth (5.1% Mexican, 4.7% Guatemalan, 0.8% Cuban). Overall, prediction models for nativity showed outstanding performance with external validation (US-born vs. foreign: AUC = 0.90; Mexican vs. non-Mexican: AUC = 0.89; Guatemalan vs. non-Guatemalan: AUC = 0.95; Cuban vs. non-Cuban: AUC = 0.99). Among challenges facing health equity researchers in health services is the absence of methods for data disaggregation, and the specific ability to determine Latino country of birth (nativity) to inform disparities. Recent interest in more robust health equity research has called attention to the importance of data disaggregation. In a multistate network of CHCs using multilevel inputs from EHR data linked to surname and community data, we developed and validated novel prediction models for the use of available EHR data to infer Latino nativity for health disparities research in primary care and health services research, which is a significant potential methodologic advance in studying this population.

A real-time green construction optimization strategy for engineering vessels considering fuel consumption and productivity: A case study on a cutter suction dredger

Energy saving in the process of engineering construction is an indispensable measure to achieve industrial low carbon emissions. To reduce fuel consumption in the construction process of engineering vessels, a real-time green construction optimization strategy for engineering vessels considering fuel consumption and productivity is proposed. A cutter suction dredger (CSD) is taken as a case study. Based on its working principle, a series of potential monitoring features are determined. In addition, details of data collection are presented. Then, the original data obtained are cleaned based on human experience. The Hilbert Schmidt independence criterion least absolute shrinkage and selection operator (HSIC Lasso) was used to filter out irrelevant and redundant factors to achieve feature dimensionality reduction, followed by data standardization based on the robust scaler function. Five machine learning models, including gradient boosting regression tree (GBRT), extreme gradient boosting (XGBoost), random forest regression (RFR), support vector regression (SVR) and super learner based on stacked strategy (SLBSS), are used to train and test offline data on the basis. The Taguchi test method is used to optimize the hyperparameters of these models. In addition, an online data collection, processing and analysis framework is proposed. The online data collected in real time are used to predict the fuel consumption level based on this framework. Finally, feature importance is analyzed. Aiming at the minimum fuel consumption per unit productivity (FCpup), specific and feasible optimization suggestions are made for CSD construction. The results show that machine learning models can predict the fuel consumption level effectively. The first seven important features should be used for prediction. The productivity, lubricating oil pressure of the pump gearbox in the cabin and slurry density are the three most important factors. Improving the parameters related to productivity and reducing the sailing speed, reamer motor voltage and washing pressure of reamer bearings can improve fuel efficiency, which is beneficial to energy savings and emission reduction. Engineering construction shows that this strategy can reduce the fuel consumption of the CSD by approximately 20%.

What if we wait? Using synthetic waiting lists to estimate treatment effects in routine outcome data

Objective: Due to the lack of randomization, pre–post routine outcome data precludes causal conclusions. We propose the “synthetic waiting list” (SWL) control group to overcome this limitation. Method: First, a step-by-step introduction illustrates this novel approach. Then, this approach is demonstrated using an empirical example with data from an outpatient cognitive–behavioral therapy (CBT) clinic (N = 139). We trained an ensemble machine learning model (“Super Learner”) on a data set of patients waiting for treatment (N = 311) to make counterfactual predictions of symptom change during this hypothetical period. Results: The between-group treatment effect was estimated to be d = 0.42. Of the patients who received CBT, 43.88% achieved reliable and clinically significant change, while this probability was estimated to be 14.54% in the SWL group. Counterfactual estimates suggest a clear net benefit of psychotherapy for 41% of patients. In 32%, the benefit was unclear, and 27% would have improved similarly without receiving CBT. Conclusions: The SWL is a viable new approach that provides between-group outcome estimates similar to those reported in the literature comparing psychotherapy with high-intensity control interventions. It holds the potential to mitigate common limitations of routine outcome data analysis.

Methodological progress note: Machine learning methods in healthcare research.

Methodological progress note: Machine learning methods in healthcare research.

Practical considerations for specifying a super learner.

Common tasks encountered in epidemiology, including disease incidence estimation and causal inference, rely on predictive modelling. Constructing a predictive model can be thought of as learning a prediction function (a function that takes as input covariate data and outputs a predicted value). Many strategies for learning prediction functions from data (learners) are available, from parametric regressions to machine learning algorithms. It can be challenging to choose a learner, as it is impossible to know in advance which one is the most suitable for a particular dataset and prediction task. The super learner (SL) is an algorithm that alleviates concerns over selecting the one 'right' learner by providing the freedom to consider many, such as those recommended by collaborators, used in related research or specified by subject-matter experts. Also known as stacking, SL is an entirely prespecified and flexible approach for predictive modelling. To ensure the SL is well specified for learning the desired prediction function, the analyst does need to make a few important choices. In this educational article, we provide step-by-step guidelines for making these decisions, walking the reader through each of them and providing intuition along the way. In doing so, we aim to empower the analyst to tailor the SL specification to their prediction task, thereby ensuring their SL performs as well as possible. A flowchart provides a concise, easy-to-follow summary of key suggestions and heuristics, based on our accumulated experience and guided by SL optimality theory.

AI-Based Automated Lipomatous Tumor Segmentation in MR Images: Ensemble Solution to Heterogeneous Data

Deep learning (DL) has been proposed to automate image segmentation and provide accuracy, consistency, and efficiency. Accurate segmentation of lipomatous tumors (LTs) is critical for correct tumor radiomics analysis and localization. The major challenge of this task is data heterogeneity, including tumor morphological characteristics and multicenter scanning protocols. To mitigate the issue, we aimed to develop a DL-based Super Learner (SL) ensemble framework with different data correction and normalization methods. Pathologically proven LTs on pre-operative T1-weighted/proton-density MR images of 185 patients were manually segmented. The LTs were categorized by tumor locations as distal upper limb (DUL), distal lower limb (DLL), proximal upper limb (PUL), proximal lower limb (PLL), or Trunk (T) and grouped by 80%/9%/11% for training, validation and testing. Six configurations of correction/normalization were applied to data for fivefold-cross-validation trainings, resulting in 30 base learners (BLs). A SL was obtained from the BLs by optimizing SL weights. The performance was evaluated by dice-similarity-coefficient (DSC), sensitivity, specificity, and Hausdorff distance (HD95). For predictions of the BLs, the average DSC, sensitivity, and specificity from the testing data were 0.72 pm 0.16, 0.73 pm 0.168, and 0.99 pm 0.012, respectively, while for SL predictions were 0.80 pm 0.184, 0.78 pm 0.193, and 1.00 pm 0.010. The average HD95 of the BLs were 11.5 (DUL), 23.2 (DLL), 25.9 (PUL), 32.1 (PLL), and 47.9 (T) mm, whereas of SL were 1.7, 8.4, 15.9, 2.2, and 36.6 mm, respectively. The proposed method could improve the segmentation accuracy and mitigate the performance instability and data heterogeneity aiding the differential diagnosis of LTs in real clinical situations.

Association of statin use with outcomes of patients admitted with COVID-19: an analysis of electronic health records using superlearner

ImportanceStatin use prior to hospitalization for Coronavirus Disease 2019 (COVID-19) is hypothesized to improve inpatient outcomes including mortality, but prior findings from large observational studies have been inconsistent, due in part to confounding. Recent advances in statistics, including incorporation of machine learning techniques into augmented inverse probability weighting with targeted maximum likelihood estimation, address baseline covariate imbalance while maximizing statistical efficiency.ObjectiveTo estimate the association of antecedent statin use with progression to severe inpatient outcomes among patients admitted for COVD-19.Design, setting and participantsWe retrospectively analyzed electronic health records (EHR) from individuals ≥ 40-years-old who were admitted between March 2020 and September 2022 for ≥ 24 h and tested positive for SARS-CoV-2 infection in the 30 days before to 7 days after admission.ExposureAntecedent statin use—statin prescription ≥ 30 days prior to COVID-19 admission.Main outcomeComposite end point of in-hospital death, intubation, and intensive care unit (ICU) admission.ResultsOf 15,524 eligible COVID-19 patients, 4412 (20%) were antecedent statin users. Compared with non-users, statin users were older (72.9 (SD: 12.6) versus 65.6 (SD: 14.5) years) and more likely to be male (54% vs. 51%), White (76% vs. 71%), and have ≥ 1 medical comorbidity (99% vs. 86%). Unadjusted analysis demonstrated that a lower proportion of antecedent users experienced the composite outcome (14.8% vs 19.3%), ICU admission (13.9% vs 18.3%), intubation (5.1% vs 8.3%) and inpatient deaths (4.4% vs 5.2%) compared with non-users. Risk differences adjusted for labs and demographics were estimated using augmented inverse probability weighting with targeted maximum likelihood estimation using Super Learner. Statin users still had lower rates of the composite outcome (adjusted risk difference: − 3.4%; 95% CI: − 4.6% to − 2.1%), ICU admissions (− 3.3%; − 4.5% to − 2.1%), and intubation (− 1.9%; − 2.8% to − 1.0%) but comparable inpatient deaths (0.6%; − 1.3% to 0.1%).Conclusions and relevanceAfter controlling for confounding using doubly robust methods, antecedent statin use was associated with minimally lower risk of severe COVID-19-related outcomes, ICU admission and intubation, however, we were not able to corroborate a statin-associated mortality benefit.