Subset Of Variables Research Articles

Multiple Imputation for Systematically Missing Partner Variables in Survey Data

Systematically missing information on secondary respondents is a frequent problem in multiactor surveys. Budget and time constraints often prevent all variables collected for primary respondents (e.g., anchors) from being collected for secondary respondents (e.g., partners). Thus, a subset of variables are systematically missing for secondary respondents. This can severely limit the analysis potential of multiactor data, ruling out all research questions that would require (the same) information on primary and secondary respondents. The problem of systematically missing data is also present in other settings, for example, after changes in measurement instruments in repeated surveys or in ex post survey harmonization if one or more surveys did not include a specific variable. In these cases, using multiple imputation (MI) techniques to impute the missing variables is a common approach. The authors explore whether MI can be used when data on secondary respondents are systematically missing. Results from simulation studies show that imputation under the assumption of conditional independence for primary and secondary respondents variables leads to a strong bias toward zero in the estimated partial correlation between primary and secondary respondents. However, external data in the form of bridging studies can be used to estimate the partial correlation between the observed variable for the primary and the unobserved variable for the secondary respondent, leading to estimates with less bias after MI.

Classification of Chinese hickory with different aging times using two-dimensional correlation spectral (2DCOS) images combined with transfer learning

The Chinese hickory (Carya cathayensis Sarg.), highly valued in eastern China for its culinary appeal and nutritional value, is prone to rancidity during storage due to its high content of polyunsaturated fatty acids. This research introduces an innovative method for categorizing Chinese hickory nuts according to their aging durations by integrating two-dimensional correlation spectroscopy (2DCOS) with transfer learning, using hyperspectral imaging technology (HSI). The study captured hyperspectral images of 1,200 hickory kernels at various aging stages, extracting spectral data within the 900–––1670 nm range. Using this spectral data, synchronous, asynchronous, and integrative 2DCOS images were generated. The analytical framework comprised five spectral preprocessing methods:first derivative (FD), second derivative (SD), multiplicative scatter correction (MSC), standard normal variate (SNV), and detrend (DET); four wavelength selection techniques:successive projections algorithm (SPA), iteratively retaining information variables (IRIV), variable combination population analysis (VCPA), and iteratively variable subset optimization (IVSO); five transfer learning models:DarkNet19, DarkNet53, ResNet50, InceptionV3 and Xception; along with two traditional approaches: support vector machine (SVM) and linear discriminant analysis (LDA). This thorough strategy was designed to accurately classify the nuts on the basis of their aging time. The findings revealed that synchronous 2DCOS significantly outperformed asynchronous and integrative forms in classifying aged hickory kernels. The FD-IVSO-2DCOS-DarkNet19 model, employing 292 (62.80 %) wavelength variables, reached an impressive accuracy of 98.75 % in the test set, surpassing the SD-SPA-2DCOS-DarkNet53 model, which used only 40 (8.6 %) wavelength variables and achieved an accuracy of 97.92 %. These models excelled over other 2DCOS approaches that utilized the full wavelength range and significantly outdid the conventional one-dimensional spectral LDA and SVM models, regardless of wavelength scope. The study underscores that synchronous 2DCOS images, when combined with suitable transfer learning techniques, provide robust discriminative power, suggesting their potential utility in the classification aged hickory nuts.

Robust Bayesian causal estimation for causal inference in medical diagnosis

Causal effect estimation is a critical task in statistical learning that aims to find the causal effect on subjects by identifying causal links between a number of predictor (or, explanatory) variables and the outcome of a treatment. In a regressional framework, we assign a treatment and outcome model to estimate the average causal effect. Additionally, for high dimensional regression problems, variable selection methods are also used to find a subset of predictor variables that maximises the predictive performance of the underlying model for better estimation of the causal effect. In this paper, we propose a different approach. We focus on the variable selection aspects of high dimensional causal estimation problem. We suggest a cautious Bayesian group LASSO (least absolute shrinkage and selection operator) framework for variable selection using prior sensitivity analysis. We argue that in some cases, abstaining from selecting (or, rejecting) a predictor is beneficial and we should gather more information to obtain a more decisive result. We also show that for problems with very limited information, expert elicited variable selection can give us a more stable causal effect estimation as it avoids overfitting. Lastly, we carry a comparative study with synthetic dataset and show the applicability of our method in real-life situations.

The reliability of functional and systemic markers of muscle damage in response to a flywheel squat protocol.

To characterize the magnitude, timescale, and reliability of changes in functional and systemic outcome markers following moderate (MIR) and high (HIR) isoinertial resistance flywheel squat protocols (FSP). Twenty-four resistance-trained males completed two exercise trials (ET1 & ET2) separated by 32days. Functional and systemic markers were assessed at pre-exercise (PRE), immediately post-exercise (IP), and 24 (24H), 48 (48H), and 72 (72H) hours post-exercise. Three-way group x trial x time repeated measures ANOVAs were conducted to compare all dependent variables between groups (MIR &HIR) and experimental trials across time. Test-retest reliability between trials was assessed using intraclass correlation coefficients (ICC). At IP, both groups exhibited significantly decreased active range of motion, perceived recovery status, squat maximal voluntary isometric contraction force, and vertical jump performance, along with significantly increased muscle thickness and echo intensity (ultrasound), muscle soreness, and creatine kinase when compared to PRE. Most outcomes remained perturbed at 24H and 48H, especially in the HIR group. By 72H, only a subset of variables remained significantly changed from PRE. No significant attenuation of outcomes between trials were observed and test-retest reliability between trials was excellent for the FSP and moderate to excellent for most outcomes in both groups. Our findings indicate that the FSP is a robust and repeatable exercise stimulus capable of eliciting significant exercise-induced muscle damage and reliable subsequent perturbations to functional and systemic markers of muscle damage. Our findings also support the use of crossover designs in future EIMD research designs with resistance-trained men.

Variability and uncertainty of variance components in fixed-effects and mixed-effects ground-motion models

It has become common in the development of ground-motion models (GMMs), especially using mixed-effects regression with crossed random effects, to calculate standard deviations, referred to as variance components, from sample statistics of the residuals (i.e. random effects and within-group errors) rather than using the variance components reported by a mixed-effects regression program, calculated using the same algorithms used in the mixed-effects regression, or estimated using Bayesian inference. This practice leads to underestimating the standard deviations because it does not account for the uncertainty (i.e. standard errors) associated with fitting the random effects and within-group errors during the regression analysis. In this study, we used a series of ground-motion models for Fourier amplitude spectra developed using mixed-effects regression of an Next Generation Attenuation (NGA)-West2 database to show that residual-based standard deviations can be significantly smaller than regression-based standard deviations. These differences are exacerbated for those variance components with the fewest number of observations or when the residuals are partitioned (e.g. by earthquake magnitude). Using residual-based variance components not only results in smaller standard deviations but can lead to biased inferences when attempting to compare the efficacy of a model based solely on a comparison of the values of the variance components or their related mean square errors. It can also lead to biased fixed-effects coefficients if the variance components are derived from the residuals of a mixed-effects regression rather than estimated during a regression. We show that variance components can be decomposed into various random-effect grouping factors and partitioned into subsets of predictor variables, such as magnitude, from the total residuals of a non-Bayesian mixed-effects regression program using Bayesian inference. This process does not require the development of an entire GMM using Bayesian mixed-effects regression.

SPARTA: Interpretable functional classification of microbiomes and detection of hidden cumulative effects.

The composition of the gut microbiota is a known factor in various diseases and has proven to be a strong basis for automatic classification of disease state. A need for a better understanding of microbiota data on the functional scale has since been voiced, as it would enhance these approaches' biological interpretability. In this paper, we have developed a computational pipeline for integrating the functional annotation of the gut microbiota into an automatic classification process and facilitating downstream interpretation of its results. The process takes as input taxonomic composition data, which can be built from 16S or whole genome sequencing, and links each component to its functional annotations through interrogation of the UniProt database. A functional profile of the gut microbiota is built from this basis. Both profiles, microbial and functional, are used to train Random Forest classifiers to discern unhealthy from control samples. SPARTA ensures full reproducibility and exploration of inherent variability by extending state-of-the-art methods in three dimensions: increased number of trained random forests, selection of important variables with an iterative process, repetition of full selection process from different seeds. This process shows that the translation of the microbiota into functional profiles gives non-significantly different performances when compared to microbial profiles on 5 of 6 datasets. This approach's main contribution however stems from its interpretability rather than its performance: through repetition, it also outputs a robust subset of discriminant variables. These selections were shown to be more consistent than those obtained by a state-of-the-art method, and their contents were validated through a manual bibliographic research. The interconnections between selected taxa and functional annotations were also analyzed and revealed that important annotations emerge from the cumulated influence of non-selected taxa.

A short term multistep forecasting model for photovoltaic generation using deep learning model

Developed countries have substantial investments in renewable energy currently, particularly Photovoltaics (PV), for achieving net-zero emissions. But PV generation is highly volatile and hence achieving supply-demand balance is challenging. Robust forecasting models will help PV integration and penetration into the grid, making sure that there is an adequate supply to match the demand, ensuring reliability and stability of power systems. In this paper, a deep learning model is developed for PV generation multistep forecasting using a small subset of weather variables with a 15-minute resolution, with very low computation time. The forecasts very closely align with the actual generation, with a Normalized Mean Absolute Error (nMAE) of 0.04, much less than 1 kWh in terms of error in forecast generation. Direct and multioutput forecasting are combined here. Comparisons with Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) show performance improvement, by ∼15% compared to LSTM and ∼17% compared to GRU in terms of average nMAE. The model can be used in urban environments for short term forecasting. Also, if an accurate forecast is available, PV asset owners can plan their generation better when they export power back into the grid, make better bids in the energy markets, increase their revenues and eventually increase the share of renewables in the energy market.

Advanced Monthly Rainfall Forecasting in Southern Lazio: Integrating Climatic Indices, Classical or Deep Neural Networks, and a Feature Selection Algorithm

Accurately predicting monthly precipitation is crucial for managing water resources, particularly in regions with complex climates like Southern Lazio, Italy. This study addresses the challenges of selecting relevant climate parameters and optimizing model complexity for precise forecasting. Four models, based on Bi-directional Long Short-Term Memory (BI-LSTM) and Radial Basis Function (RBF) neural networks, were compared, focusing on their ability to predict monthly precipitation using diverse inputs such as mean temperature, relative humidity, sea level pressure, radiation, precipitation change, and 12 climate indices. One significant challenge was determining the most impactful subset of input variables from a large dataset to enhance model performance without overfitting. To address this, the Particle Swarm Optimization (PSO) feature selection algorithm was employed, significantly improving model accuracy by identifying the optimal subset of input variables. The comparative analysis revealed that PSO-based models consistently outperformed those using all input variables, with the PSO-RBF model achieving a Kling-Gupta Efficiency (KGE) value of up to 0.80 during testing. Notably, both PSO-BI-LSTM and PSO-RBF models demonstrated strong performance in predicting precipitation peaks, achieving KGE values of 0.74. The study highlights the efficacy of the RBF model combined with PSO, which optimizes fewer parameters compared to the more complex BI-LSTM network while reducing input variables. This approach is not only less computationally intensive but also easier to implement, particularly in regions with limited instrumentation. These findings suggest that the PSO-RBF model is a reliable and efficient tool for monthly precipitation prediction in challenging environments.

Crude oil price forecasting with multivariate selection, machine learning, and a nonlinear combination strategy

Crude oil price forecasting has been one of the research hotspots in the field of energy economics, which plays a crucial role in energy supply and economic development. However, numerous influencing factors bring serious challenges to crude oil price forecasting, and existing research has room for further improvement in terms of an integrated research roadmap that combines impact factor analysis with predictive modelling. This study aims to examine the impact of financial market factors on the crude oil market and to propose a nonlinear combined forecasting framework based on common variables. Four types of daily exogenous financial market variables are introduced: commodity prices, exchange rates, stock market indices, and macroeconomic indicators for ten indicators. First, various variable selection methods generate different variable subsets, providing more diversity and reliability. Next, common variables in the subset of variables are selected as key features for subsequent models. Then, four models predict crude oil prices using common features as inputs and obtain the prediction results for each model. Finally, the nonlinear mechanism of the deep learning technology is introduced to combine above single prediction results. Experimental results reveal that commodity and foreign exchange factors in financial markets are critical determinants of crude oil market volatility over the long term, as observed in experiments conducted on the West Texas Intermediate and Brent oil price datasets. The proposed model demonstrates strong performance regarding average absolute percentage error, recorded at 2.9962% and 2.4314%, respectively, indicating high forecasting accuracy and robustness. This forecasting framework offers an effective methodology for predicting crude oil prices and enhances understanding the crude oil market.

Identifying ecological factors mediating the spread of three invasive mosquito species: citizen science informed prediction

AbstractDue to their potential role in pathogen transmission, invasive mosquitoes pose considerable threats to human and animal health. Several studies have identified the most important ecological drivers mediating the establishment and spread of key mosquito species (e.g., Aedes aegypti, and Ae. albopictus), and made predictions for future distribution. We evaluated the effect of an exhaustive list of environmental predictors on the distribution of three invasive species in Hungary (Ae. albopictus, Ae. japonicus, and Ae. koreicus) by using the same standards for data collection based on citizen science observations. Current distribution maps of these species were generated from a 5-year survey, then were compared with various predictor maps reflecting climate, habitat type, food supply, traffic, and interspecific competition by using a boosted regression trees approach that resulted in a subset of variables with the strongest impact. The best predictor sets were used to predict the probability of occurrence of the focal species for the whole country, and these predictions based on citizen science were evaluated against the results of an independent recent field surveillance. We uncovered species-specific patterns and found that different predictor sets were selected for the three different species, and only predictions for Ae. albopictus could be validated with direct trapping data. Therefore, citizen science informed distribution maps can be used to identify ecological predictors that determine the spread of invasive mosquitoes, and to estimate risk based on the predicted distribution in the case of Ae. albopictus.

Arbitrary Random Variables and Wiman’s Inequality

We study the class of random entire functions given by power series, in which the coefficients are formed as the product of an arbitrary sequence of complex numbers and two sequences of random variables. One of them is the Rademacher sequence, and the other is an arbitrary complex-valued sequence from the class of sequences of random variables, determined by a certain restriction on the growth of absolute moments of a fixed degree from the maximum of the module of each finite subset of random variables. In the paper we prove sharp Wiman–Valiron’s type inequality for such random entire functions, which for given p∈(0;1) holds with a probability p outside some set of finite logarithmic measure. We also considered another class of random entire functions given by power series with coefficients, which, as above, are pairwise products of the elements of an arbitrary sequence of complex numbers and a sequence of complex-valued random variables described above. In this case, similar new statements about not improvable inequalities are also obtained.

A novel classification of dyslipidaemia through the analysis of five million lipid profiles: an unsupervised machine learning approach

Abstract Background Dyslipidemia encompasses a wide range of lipoprotein disorders categorised through two classifications (Fredrickson-Levy [FL] or Sniderman).(1,2) However, both classifications are criticised for relying on incomplete knowledge of lipoprotein metabolism, especially with the emergence of novel treatment options and variations in individual treatment responses.(3) Clustering, an unsupervised machine learning (ML) algorithm that can process a wide range of variables, has the potential to unmask patient groups with distinct molecular profiles and unique therapeutic targets that can inform more effective prevention strategies for cardiovascular disease (CVD).(4) Aim We aimed to use unsupervised ML algorithms to discover intrinsic dyslipidaemia categories from lipoprotein measurements, recognise the necessary components of lipid panels for classification, and analyse the similarities between the newly formed clusters, FL and Sniderman classifications. Methods Lipid profiles of 5,080,248 patients were obtained from the ‘Very Large Database of Lipids’ database. This yielded up to 78 blood components per patient, including at least 31 lipoprotein variables. The analysis involved unsupervised K-means clustering with optimised values for K and the subset of variables, determined in an unsupervised manner using a suitable measure of complexity. We then interpreted our clusters using probabilistic decision trees to provide compact and interpretable representations. Finally, we compared the clusters with Sniderman and FL categories. Results In a completely unsupervised fashion, we identified 14 clusters that could be matched to Sniderman categories. The confusion matrix showed total agreement of 76% (see Figure 1, left panel), relative Cohen’s kappa of 0.78 (the relative version captures accuracy on categories containing smaller numbers of patient profiles) and an accuracy of 96% on the small Type III class. Similar results were observed when matching to FL types. We accurately represented our clusters using probabilistic decision trees of small depth (see Figure 2). We discovered that the data had low intrinsic dimension and a manifold-like structure in which the different clusters could be illustrated (see Figure 1, right panel). Specifically, only 3 variables were needed to obtain our classification: apolipoprotein b, total cholesterol and triglycerides. Conclusion We showed that completely unsupervised ML techniques can uncover dyslipidaemia categories in lipoprotein profiles from a large patient population. The categories largely align with existing classifications based on prior knowledge of lipoprotein metabolism. Furthermore, few lipoprotein variables were required for categorisation (low-dimension data), which could aid in determining which lipoproteins should be measured in a clinical setting. Further analysis of the differences between ML clusters and traditional classifications is needed, which may enhance CVD risk management.

Prevalence and prediction of masked uncontrolled hypertension in patients recently hospitalised for an acute coronary syndrome

Abstract Background Masked uncontrolled hypertension (MUCH) is defined as normal office blood pressure (BP) but elevated out-of-office BP in patients on antihypertensive treatment. MUCH is associated with similar cardiovascular risk as sustained hypertension and is clinically challenging since this often remains undetected without out-of-office BP measurements. Purpose To study the prevalence of MUCH following an acute coronary syndrome and to utilise machine learning methods to develop prediction models for identifying MUCH. Methods Ambulatory 24-h BP measurement (ABPM) was performed in 100 patients attending the cardiology outpatient clinic at a Swedish university hospital following an acute coronary syndrome, as part of a study which screened for co-morbidities. From the SWEDEHEART registry, 106 variables during the hospital care period and subsequent follow-up visit (after 6-10 weeks) were pre-processed including filtering on 5 or more missing values and zero or near-zero variance. Variable importance for the prediction of MUCH (office BP < 140/90 mm Hg at ABPM start but mean 24-h BP ≥ 130/80 mm Hg) was assessed using the Boruta and least absolute shrinkage and selection operator (LASSO) machine learning algorithms. Subsequently, logistic regression, LASSO and random forest models using different variable subsets were evaluated by receiver operating characteristic area under the curve (AUC) in repeated cross-validation. Results Age was 62.0±8.4 years, 74% male, 54% had NSTEMI and 46% STEMI. The follow-up visit and ABPM were performed at median 7 and 11 weeks, respectively, after hospital discharge. Among 90 patients with complete data and ABPM recordings, 31 had mean 24-h BP above target levels, of which 18 were identified with MUCH. Patients with MUCH had lower eGFR (68±11 vs 76±12 ml/min/1.73m², P=0.009), and more often a history of hypertension (89 vs 53%, P=0.005) and diabetes mellitus (44 vs 11%, P=0.003). In total, 65 variables where eligible for machine learning after filtering. Boruta and LASSO identified pulse pressure at the follow-up visit, serum creatinine, diabetes mellitus and history of hypertension as important predictors. Random forest, logistic regression and LASSO showed mean AUC 0.826, 0.822, and 822, respectively, in cross validation using these predictors. Conclusions In this small trial, one in five had MUCH at follow-up after an acute coronary syndrome, which highlights the importance of out-of-office BP measurements. The easily accessible measures of pulse pressure at the follow-up visit, serum creatinine, diabetes mellitus and history of hypertension were identified as important predictors of MUCH. A simple prediction model may be used as a clinical decision support tool after appropriate external validation.

What can be learnt with wide convolutional neural networks?*

Abstract Understanding how convolutional neural networks (CNNs) can efficiently learn high-dimensional functions remains a fundamental challenge. A popular belief is that these models harness the local and hierarchical structure of natural data such as images. Yet, we lack a quantitative understanding of how such structure affects performance, for example the rate of decay of the generalisation error with the number of training samples. In this paper, we study infinitely wide deep CNNs in the kernel regime. First, we show that the spectrum of the corresponding kernel inherits the hierarchical structure of the network, and we characterise its asymptotics. Then, we use this result together with generalisation bounds to prove that deep CNNs adapt to the spatial scale of the target function. In particular, we find that if the target function depends on low-dimensional subsets of adjacent input variables then the decay of the error is controlled by the effective dimensionality of these subsets. Conversely, if the target function depends on the full set of input variables then the error decay is controlled by the input dimension. We conclude by computing the generalisation error of a deep CNN trained on the output of another deep CNN with randomly initialised parameters. Interestingly, we find that, despite their hierarchical structure, the functions generated by infinitely wide deep CNNs are too rich to be efficiently learnable in high dimensions.

Determining the Identity of Nucleotides and the Energy of Binding of tRNAs to Their Aminoacyl-tRNA Synthetases Using a Simple Logistic Model

This study showed that the predictor in logistic regression can be applied to estimating the Gibbs free energy of tRNAs’ recognition of and binding to their aminoacyl-tRNA synthetases. Then, 24 linear logistic regression models predicting different classes of tRNAs loaded with a corresponding amino acid were trained in a machine learning classification method, reducing the misclassification error to zero. The models were based on minimal subsets of Boolean explanatory variables describing the favorite presence of nucleotides or nucleosides localized in the different parts of the tRNA. In 90% of cases, they agree with the components of the consensus strand in a class of tRNAs loaded by a given amino acid. According to the proposed theoretical model, the values of the free energy for the entry of the recognition state in the process of tRNA charging were obtained, and the inputs from identity nucleotides and the tRNA strand backbone were distinguished. Almost all the resulting models indicated leading anticodon tandems defining the first and second positions of the anticodon (positions 35 and 36 of the tRNA strand) and the small sets (up to six positions) of the other nucleotides as the natural identity nucleotides most influential in the free energy balance. The magnitude of their input to this energy depends on the position in the strand, favoring positions −1, 35, and 36. The role of position 34 is relatively smaller. These identity attributes may not always be fully arranged in a real single adaptor molecule but were comprehensively present in a given tRNA class. A detailed analysis of the resulting models showed that the absolute value of the energy of binding the tandem 35–36 decreases with the number of identity positions, as well as with the decreasing number of possible hydrogen bonds. On the other hand, in these conditions, the absolute value of the energy of binding of other identity nucleotides increases. All the models indicate that the nucleotide-independent energy of the repulsion tRNA backbone decreases with the number of identity nucleotides. It was also shown that the total free energy change in entering the recognition state increases with the amino acid mass, making this process less spontaneous, which may have an evolutionary reference.

Programmable MCMC with Soundly Composed Guide Programs

Probabilistic programming languages (PPLs) provide language support for expressing flexible probabilistic models and solving Bayesian inference problems. PPLs with programmable inference make it possible for users to obtain improved results by customizing inference engines using guide programs that are tailored to a corresponding model program. However, errors in guide programs can compromise the statistical soundness of the inference. This article introduces a novel coroutine-based framework for verifying the correctness of user-written guide programs for a broad class of Markov chain Monte Carlo (MCMC) inference algorithms. Our approach rests on a novel type system for describing communication protocols between a model program and a sequence of guides that each update only a subset of random variables. We prove that, by translating guide types to context-free processes with finite norms, it is possible to check structural type equality between models and guides in polynomial time. This connection gives rise to an efficient type-inference algorithm for probabilistic programs with flexible constructs such as general recursion and branching. We also contribute a coverage-checking algorithm that verifies the support of sequentially composed guide programs agrees with that of the model program, which is a key soundness condition for MCMC inference with multiple guides. Evaluations on diverse benchmarks show that our type-inference and coverage-checking algorithms efficiently infer types and detect sound and unsound guides for programs that existing static analyses cannot handle.

On the Elimination of Fast Variables from the Langevin Equation.

In a multivariable system, there are usually a number of relaxation times. When some of the relaxation times are shorter than others, the corresponding variables will decay to their equilibrium value faster than the others. After the fast variables have decayed, the system can be described with a smaller number of variables. From the theory of nonequilibrium thermodynamics, as formulated by Onsager, we know that the coefficients in the linear flux-force relations satisfy the so-called Onsager symmetry relations. The question we will address in this paper is how to eliminate the fast variables in such a way that the coefficients in the reduced description for the slow variables still satisfy the Onsager relations. As the proof that Onsager gave of the symmetry relations does not depend on the choice of the variables, it is equally valid for the subset of slow variables. Elimination procedures that lead to symmetry breaking are possible, but do not consider systems that satisfy the laws of nonequilibrium thermodynamics.

Empirical Monte Carlo evidence on estimation of timing-of-events models

This article builds on the Empirical Monte Carlo simulation approach to study the estimation of Timing-of-Events (ToE) models. We exploit rich Swedish data of unemployed job seekers with information on participation in a training program to simulate placebo treatment durations. We first use these simulations to examine which covariates are key confounders to be included in dynamic selection models for training participation. The joint inclusion of specific short-term employment history indicators (notably, the share of time spent in employment), together with baseline socio-economic characteristics, regional and inflow timing information, is important to deal with selection bias. Next, we omit subsets of explanatory variables and estimate ToE models with discrete distributions for the ensuing systematic unobserved heterogeneity. In many cases, the ToE approach provides accurate effect estimates, especially if time-varying variation in the unemployment rate of the local labor market is taken into account. However, assuming too many or too few support points for unobserved heterogeneity may lead to large biases. Information criteria, in particular those penalizing parameter abundance, are useful to select the number of support points. A comparison with other duration models shows that a Stratified Cox model performs well with abundant multiple spells but less well when multiple spells are uncommon. The standard Cox regression model performs poorly in all configurations as it is unable to account for unobserved heterogeneity.

Cardiac Involvement in LAMA2-Related Muscular Dystrophy and SELENON-Related Congenital Myopathy: A Case Series.

LAMA2-related muscular dystrophy (LAMA2-MD) and SELENON-related myopathy (SELENON-RM) are two rare neuromuscular diseases characterized by proximal and axial muscle weakness, scoliosis, spinal rigidity, low bone quality and respiratory impairment. Cardiac involvement has previously been described in retrospective studies and case reports, but large case series and prospective studies in unselected cohorts are lacking. The objective of this study is to conduct prevalence estimations, perform cardiac phenotyping, and provide recommendations for clinical care. In this case series including two time points, we conducted comprehensive assessments with electrocardiography (ECG) and transthoracic echocardiography (TTE). ECGs were systematically assessed for a large subset of variables. TTE included left and right ventricular ejection fraction (LVEF/RVEF) and left ventricular global longitudinal strain (GLS), the latter being a more early and sensitive marker of left ventricular dysfunction. 21 LAMA2-MD (M = 5; 20±14 years) and 10 SELENON-RM patients (M = 7; 18±12 years) were included. In most patients, QRS fragmentation and Q waves, markers of heterogeneous ventricular activation, were present both at baseline and at follow-up. GLS was abnormal (age specific in children, > -18% in adults) in 33% of LAMA2-MD and 43% of SELENON-RM patients at baseline. Reduced LVEF (<52% in males, <54% in females and <55% in pediatric population) was observed in three LAMA2-MD patients at baseline and in none of the SELENON-RM patients. GLS and LVEF did not change between baseline and follow-up. RVEF was normal in all patients. ECG abnormalities and abnormal GLS are prevalent in LAMA2-MD and SELENON-RM, yet abnormal LVEF was only seen in LAMA2-MD patients. One LAMA2-MD patient had a clinically relevant deterioration in LVEF during 1.5-year follow-up. We advise routine screening of all patients with LAMA2-MD or SELENON-RM with ECG and echocardiography at diagnosis, minimally every two years from second decade of life and if new cardiac signs arise.

Sparsesurv: a Python package for fitting sparse survival models via knowledge distillation.

Sparse survival models are statistical models that select a subset of predictor variables while modeling the time until an event occurs, which can subsequently help interpretability and transportability. The subset of important features is often obtained with regularized models, such as the Cox Proportional Hazards model with Lasso regularization, which limit the number of non-zero coefficients. However, such models can be sensitive to the choice of regularization hyperparameter. In this work, we develop a software package and demonstrate how knowledge distillation, a powerful technique in machine learning that aims to transfer knowledge from a complex teacher model to a simpler student model, can be leveraged to learn sparse survival models while mitigating this challenge. For this purpose, we present sparsesurv, a Python package that contains a set of teacher-student model pairs, including the semi-parametric accelerated failure time and the extended hazards models as teachers, which currently do not have Python implementations. It also contains in-house survival function estimators, removing the need for external packages. Sparsesurv is validated against R-based Elastic Net regularized linear Cox proportional hazards models as implemented in the commonly used glmnet package. Our results reveal that knowledge distillation-based approaches achieve competitive discriminative performance relative to glmnet across the regularization path while making the choice of the regularization hyperparameter significantly easier. All of these features, combined with a sklearn-like API, make sparsesurv an easy-to-use Python package that enables survival analysis for high-dimensional datasets through fitting sparse survival models via knowledge distillation. sparsesurv is freely available under a BSD 3 license on GitHub (https://github.com/BoevaLab/sparsesurv) and The Python Package Index (PyPi) (https://pypi.org/project/sparsesurv/).