Elastic Net Model Research Articles

Predicting cytotoxicity of engineered nanoparticles using regularized regression models: an in silico approach

ABSTRACT The widespread application of engineered nanoparticles (NPs) in various industries has demonstrated their effectiveness over the years. However, modifications to NPs’ physicochemical properties can lead to toxicological effects. Therefore, understanding the toxicity behaviour of NPs is crucial. In this paper, regularized regression models, such as ridge, LASSO, and elastic net, were constructed to predict the cytotoxicity of various engineered NPs. The dataset utilized in this study was compiled from several journals published between 2010 and 2022. Data exploration revealed missing values, which were addressed through listwise deletion and kNN imputation, resulting in two complete datasets. The ridge, LASSO, and elastic net models achieved F1 scores ranging from 91.81% to 92.65% during internal validation and 92.89% to 93.63% during external validation on Dataset 1. On Dataset 2, the models attained F1 scores between 92.16% and 92.43% during internal validation and 92% and 92.6% during external validation. These results indicate that the developed models effectively generalize to unseen data and demonstrate high accuracy in classifying cytotoxicity levels. Furthermore, the cell type, material, cell source, cell tissue, synthesis method, and coat or functional group were identified as the most important descriptors by the three models across both datasets.

Identification and prediction of immune checkpoint inhibitors-related pneumonitis by machine learning.

Immune checkpoint inhibitor (ICI)-related pneumonitis (IRP) is a common and potentially fatal clinical adverse event. The identification and prediction of the risk of ICI-related IRP is a major clinical issue. The objective of this study was to apply a machine learning method to explore risk factors and establish a prediction model. We retrospectively analyzed 48 patients with IRP (IRP group) and 142 patients without IRP (control group) who were treated with ICIs. An Elastic Net model was constructed using a repeated k-fold cross-validation framework (repeat = 10; k = 3). The prediction models were validated internally and the final prediction model was built on the entire training set using hyperparameters with the best interval validation performance. The generalizability of the final prediction model was assessed by applying it to an independent test set. The overall performance, discrimination, and calibration of the prediction model were evaluated. Eleven predictors were included in the final predictive model: sindillizumab, number of ≥2 underlying diseases, history of lung diseases, tirelizumab, non-small cell lung cancer (NSCLC), percentage of CD4+ lymphocytes, body temperature, KPS score ≤70, hemoglobin, cancer stage IV, and history of antitumor therapy. The external validation of the risk prediction model on an independent test set of 37 patients and showed good discrimination and acceptable calibration ability: with AUC of 0.81 (95% CI 0.58-0.90), AP of 0.76, scaled Brier score of 0.31, and Spiegelhalter-z of -0.29 (P-value:0.77). We also designed an online IRP risk calculator for use in clinical practice. The prediction model of ICI-related IRP provides a tool for accurately predicting the occurrence of IRP in patients with cancer who received ICIs.

A multi-sensor fusion-based prognostic model for systems with partially observable failure modes

With the rapid development of sensor and communication technology, multi-sensor data is available to monitor the degradation of complex systems and predict the failure modes. However, two huge challenges remain to be resolved: (i) how to predict the failure modes with limited failure mode labeled systems to alleviate the heavy dependence on expert experience; (ii) how to effectively fuze the useful information from the multi-sensor data to achieve an accurate estimation of the degradation status automatically. To address these issues, we propose a novel semi-supervised prognostic model for the systems with partially observable failure modes, where only a small fraction of the systems in the training set are known for their failure modes. First, we develop a graph-based semi-supervised learning method to extract features characterizing the failure modes. Then, we input these features as well as the multi-sensor streams into an elastic net functional regression model to predict the residual useful lifetime. The proposed model is validated by extensive simulation studies and a case study of aircraft turbofan engines available from the NASA repository.

Identification of a Twelve-microRNA Signature with Prognostic Value in Stage II Microsatellite Stable Colon Cancer.

We aimed to identify and validate a set of miRNAs that could serve as a prognostic signature useful to determine the recurrence risk for patients with COAD. Small RNAs from tumors of 100 stage II, untreated, MSS colon cancer patients were sequenced for the discovery step. For this purpose, we built an miRNA score using an elastic net Cox regression model based on the disease-free survival status. Patients were grouped into high or low recurrence risk categories based on the median value of the score. We then validated these results in an independent sample of stage II microsatellite stable tumor tissues, with a hazard ratio of 3.24, (CI95% = 1.05-10.0) and a 10-year area under the receiver operating characteristic curve of 0.67. Functional analysis of the miRNAs present in the signature identified key pathways in cancer progression. In conclusion, the proposed signature of 12 miRNAs can contribute to improving the prediction of disease relapse in patients with stage II MSS colorectal cancer, and might be useful in deciding which patients may benefit from adjuvant chemotherapy.

Biomarker Discovery in Rare Malignancies: Development of a miRNA Signature for RDEB-cSCC

Machine learning has been proven to be a powerful tool in the identification of diagnostic tumor biomarkers but is often impeded in rare cancers due to small patient numbers. In patients suffering from recessive dystrophic epidermolysis bullosa (RDEB), early-in-life development of particularly aggressive cutaneous squamous-cell carcinomas (cSCCs) represents a major threat and timely detection is crucial to facilitate prompt tumor excision. As miRNAs have been shown to hold great potential as liquid biopsy markers, we characterized miRNA signatures derived from cultured primary cells specific for the potential detection of tumors in RDEB patients. To address the limitation in RDEB-sample accessibility, we analyzed the similarity of RDEB miRNA profiles with other tumor entities derived from the Cancer Genome Atlas (TCGA) repository. Due to the similarity in miRNA expression with RDEB-SCC, we used HN-SCC data to train a tumor prediction model. Three models with varying complexity using 33, 10 and 3 miRNAs were derived from the elastic net logistic regression model. The predictive performance of all three models was determined on an independent HN-SCC test dataset (AUC-ROC: 100%, 83% and 96%), as well as on cell-based RDEB miRNA-Seq data (AUC-ROC: 100%, 100% and 91%). In addition, the ability of the models to predict tumor samples based on RDEB exosomes (AUC-ROC: 100%, 93% and 100%) demonstrated the potential feasibility in a clinical setting. Our results support the feasibility of this approach to identify a diagnostic miRNA signature, by exploiting publicly available data and will lay the base for an improvement of early RDEB-SCC detection.

Estimation of horizontal running power using foot-worn inertial measurement units.

Feedback of power during running is a promising tool for training and determining pacing strategies. However, current power estimation methods show low validity and are not customized for running on different slopes. To address this issue, we developed three machine-learning models to estimate peak horizontal power for level, uphill, and downhill running using gait spatiotemporal parameters, accelerometer, and gyroscope signals extracted from foot-worn IMUs. The prediction was compared to reference horizontal power obtained during running on a treadmill with an embedded force plate. For each model, we trained an elastic net and a neural network and validated it with a dataset of 34 active adults across a range of speeds and slopes. For the uphill and level running, the concentric phase of the gait cycle was considered, and the neural network model led to the lowest error (median ± interquartile range) of 1.7% ± 12.5% and 3.2% ± 13.4%, respectively. The eccentric phase was considered relevant for downhill running, wherein the elastic net model provided the lowest error of 1.8% ± 14.1%. Results showed a similar performance across a range of different speed/slope running conditions. The findings highlighted the potential of using interpretable biomechanical features in machine learning models for the estimating horizontal power. The simplicity of the models makes them suitable for implementation on embedded systems with limited processing and energy storage capacity. The proposed method meets the requirements for applications needing accurate near real-time feedback and complements existing gait analysis algorithms based on foot-worn IMUs.

#2683 NEXT GENERATION SEQUENCING: FROM BANFF SCORE MOLECULAR SIGNATURES AND CLASSIFIERS TO PREDICTED HISTOLOGICAL ARCHETYPES OF KIDNEY BIOPSIES

Abstract Background and Aims To improve risk stratification in kidney transplantation, molecular diagnostic tools are increasingly investigated for their feasibility and gain in diagnostic power. Nonetheless, previously published studies focused on sequencing technologies and gene panels with limited vision of the transcriptome (pathogenesis-based transcripts [PBTs], Banff Human Organ Transplant [BHOT] panel). The EU-TRAIN consortium was built to discover new predictive and informative biomarkers for kidney transplant histology and rejection diagnoses, leveraging a next generation sequencing technology. Method EU-TRAIN (NCT03652402) is a prospective multicentric study including unselected kidney transplant cohorts from 11 centres from 4 countries (France, Spain, Germany, Switzerland). We performed a bulk RNA sequencing on the polyadenylated probes of 770 kidney biopsies (n = 540 kidney recipients) collected between 2018 and 2021. Differential gene expression analyses were computed to obtain a molecular signature for all Banff score lesions. We then derived three different feature selections by either i) training an ElasticNet model on all differentially-expressed genes (DEGs), or by taking the top 30 ii) overall DEGs or iii) the top 30 DEGs focusing on transcripts included in the BHOT gene panel. From them, we trained four machine learning (ML) classifiers through a 10-times repeated 3-fold cross-validation. Models’ performances were assessed on a hold-out test set accounting for 30% of the total samples. Finally, we derived prototypic histological profiles using an archetypal analysis on the samples’ predicted Banff score probabilities from the best classifier. Results The ElasticNet feature selection lowered the number of DEGs to be included from a range of [859; 10,839] to [52; 867]. These selections were composed of a mixture of PBTs [55.2%; 84.2%], BHOT genes [2.2%;13.5%] and new transcripts [12.4%; 37.2%]. Based on these findings, four ML classifiers (Naïve Bayes, Extreme Gradient Boosting, Linear Support Vector Machine and K-Nearest Neighbours) were trained on the three different feature selections and their performances in predicting Banff score lesions were compared using the precision-recall area under the curve (PRAUC). In all settings, the ElasticNet feature selection outperformed the other two methods with a minimal/maximal increase in PRAUC of 0.068 (t) / 0.726 (cg). Excluding the ah score, the best discriminations were obtained with the Linear Support Vector Machine with a PRAUC in the interval [0.708; 0.980] (t/ptc). Excluding cv and ah, all models calibrated properly (Hosmer and Lemeshow goodness of fit p-value > 0.05). The archetypal analysis resulted in 8 profiles: acute and chronic antibody-mediated rejection (presence of circulating donor-specific antibodies, C4d deposition, g, ptc and cg lesions), acute T-cell mediated rejection (i, t and ti lesions), chronic TCMR (i, t, ti, ci, ct and i-IFTA lesions), mixed rejection (g, ptc, i, t and ti lesions), vascular injuries (cv and ah), fibrosis (ci and ct lesions, older donors with history of hypertension), minimal fibrotic change (ci and ct) and minor changes (no lesions). Conclusion From new transcripts, we managed to develop models that predict accurately the Banff score lesions and 8 profiles were identified among these predictions. External validation and archetypes’ association with graft loss will be addressed in the future.

Multidimensional analysis of immune cells from COVID-19 patients identified cell subsets associated with the severity at hospital admission.

SARS-CoV-2 emerged as a new coronavirus causing COVID-19, and it has been responsible for more than 760 million cases and 6.8 million deaths worldwide until March 2023. Although infected individuals could be asymptomatic, other patients presented heterogeneity and a wide range of symptoms. Therefore, identifying those infected individuals and being able to classify them according to their expected severity could help target health efforts more effectively. Therefore, we wanted to develop a machine learning model to predict those who will develop severe disease at the moment of hospital admission. We recruited 75 individuals and analysed innate and adaptive immune system subsets by flow cytometry. Also, we collected clinical and biochemical information. The objective of the study was to leverage machine learning techniques to identify clinical features associated with disease severity progression. Additionally, the study sought to elucidate the specific cellular subsets involved in the disease following the onset of symptoms. Among the several machine learning models tested, we found that the Elastic Net model was the better to predict the severity score according to a modified WHO classification. This model was able to predict the severity score of 72 out of 75 individuals. Besides, all the machine learning models revealed that CD38+ Treg and CD16+ CD56neg HLA-DR+ NK cells were highly correlated with the severity. The Elastic Net model could stratify the uninfected individuals and the COVID-19 patients from asymptomatic to severe COVID-19 patients. On the other hand, these cellular subsets presented here could help to understand better the induction and progression of the symptoms in COVID-19 individuals.

Factors associated with heartworm preventative use in the golden retriever lifetime study.

Heartworm disease is preventable with use of heartworm preventatives, but the reported prevalence of heartworm preventative use in the United States is low, some estimates falling around 50% of dogs. However, there are very few estimates of prevalence and its associated factors. We aimed to estimate prevalence and evaluate factors, including vaccination status, demographics, lifestyle, physical conditions, medications and supplements, and environment and living conditions, for their association with heartworm preventative use in a large dataset from the Golden Retriever Lifetime Study (N = 2,998). Due to the large number of predictors evaluated, we built a bootstrapped elastic net logistic regression model, which is robust to overfitting and multicollinearity. Variables were evaluated by calculating covariate stability (>80%) and statistical significance (p<0.02). In our sample, the prevalence of heartworm use was 39.5%. In our elastic net model, receiving vaccinations (rabies, Bordetella, or any other vaccine), being located in the Southern U.S., being altered, having an infectious disease or ear/ nose/throat system disease diagnosis, being on heartworm preventatives in the past, currently being on tick preventative, having sun exposure in an area with concrete flooring, living in a house with more rooms with carpeted floors, and spending time on hardwood flooring inside were associated with greater odds of heartworm preventative use. Supplementation use and being in the top quartile of height were associated with lower odds of heartworm preventative use. The explanatory factors we identified can be used to improve client communication. In addition, target populations for educational interventions and outreach can be identified. Future studies can validate the findings in a more diverse population of dogs.

Wellbeing Forecasting in Postpartum Anemia Patients.

Postpartum anemia is a very common maternal health problem and remains a persistent public health issue globally. It negatively affects maternal mood and could lead to depression, increased fatigue, and decreased cognitive abilities. It can and should be treated by restoring iron stores. However, in most health systems, there is typically a six-week gap between birth and the follow-up postpartum visit. Risks of postpartum maternal complications are usually assessed shortly after birth by clinicians intuitively, taking into account psychosocial and physical factors, such as the presence of anemia and the type of iron supplementation. In this paper, we investigate the possibility of using machine-learning algorithms to more reliably forecast three parameters related to patient wellbeing, namely depression (measured by Edinburgh Postnatal Depression Scale-EPDS), overall tiredness, and physical tiredness (both measured by Multidimensional Fatigue Inventory-MFI). Data from 261 patients were used to train the forecasting models for each of the three parameters, and they outperformed the baseline models that always predicted the mean values of the training data. The mean average error of the elastic net regression model for predicting the EPDS score (with values ranging from 0 to 19) was 2.3 and outperformed the baseline, which already hints at the clinical usefulness of using such a model. We further investigated what features are the most important for this prediction, where the EDPS score and both tiredness indexes at birth turned out to be by far the most prominent prediction features. Our study indicates that the machine-learning model approach has the potential for use in clinical practice to predict the onset of depression and severe fatigue in anemic patients postpartum and potentially improve the detection and management of postpartum depression and fatigue.

Plasma amino acid signature for sarcopenic phenotypes in community-dwelling octogenarians: Results from the Kawasaki Aging Wellbeing Project

Sarcopenia is one of the primary risk factors for various adverse health events in later life. However, its pathophysiology in the very old population remains unclear. Hence, this study aimed to examine whether plasma free amino acids (PFAAs) correlate with major sarcopenic phenotypes (i.e., muscle mass, muscle strength, and physical performance) in community-dwelling adults aged 85–89 years living in Japan. Cross-sectional data from the Kawasaki Aging Well-being Project were used. We included 133 adults aged 85–89 years. In this study, fasting blood was collected to measure 20 plasma PFAAs. Measures for the three major sarcopenic phenotypes included appendicular lean mass assessed by multifrequency bioimpedance, isometric handgrip strength, and gait speed from a 5 m walk at a usual pace. Furthermore, we used phenotype–specific elastic net regression models adjusted for age centered at 85 years, sex, body mass index, education level, smoking status, and drinking habit to identify significant PFAAs for each sarcopenic phenotype. Higher histidine and lower alanine levels were associated with poor gait speed, but no PFAAs correlated with muscle strength or mass. In conclusion, PFAAs such as plasma histidine and alanine are novel blood biomarkers associated with physical performance in community-dwelling adults aged 85 years or older.

Personality Computing With Naturalistic Music Listening Behavior: Comparing Audio and Lyrics Preferences

It is a long-held belief in psychology and beyond that individuals’ music preferences reveal information about their personality traits. While initial evidence relates self-reported preferences for broad musical styles to the Big Five dimensions, little is known about day-to-day music listening behavior and the intrinsic attributes of melodies and lyrics that reflect these individual differences. The present study (N = 330) proposes a personality computing approach to fill these gaps with new insights from ecologically valid music listening records from smartphones. We quantified participants’ music preferences via audio and lyrics characteristics of their played songs through technical audio features from Spotify and textual attributes obtained via natural language processing. Using linear elastic net and non-linear random forest models, these behavioral variables served to predict Big Five personality on domain and facet levels. Out-of-sample prediction performances revealed that – on the domain level – Openness was most strongly related to music listening (r = .25), followed by Conscientiousness (r = .13), while several facets of the Big Five also showed small to medium effects. Hinting at the incremental value of audio and lyrics characteristics, both musical components were differentially informative for models predicting Openness and its facets, whereas lyrics preferences played the more important role for predictions of Conscientiousness dimensions. In doing so, the models’ most predictive variables displayed generally trait-congruent relationships between personality and music preferences. These findings contribute to the development of a cumulative theory on music listening in personality science and may be extended in numerous ways by future work leveraging the computational framework proposed here.

Performance of evolutionary optimized machine learning for modeling total organic carbon in core samples of shale gas fields

Rock samples' TOC content is the best indicator of the organic matter in source rocks. The origin rock samples’ analysis is used to calculate it manually by specialists. This method requires time and resources because it relies on samples from many well intervals in source rocks. Therefore, research has been done to aid this effort. Machine learning algorithms can estimate total organic carbon instead of well logs and stratigraphic studies. In light of these efforts, the current work present a study on automating the total organic carbon estimation using machine learning approaches improved by an evolutionary methodology to give the model flexibility and precision. Genetic algorithms, differential evolution, particle swarm optimization, grey wolf optimization, artificial bee colony, and evolution strategies were used to improve machine learning models to predict TOC. The six metaheuristics were integrated into four machine learning methods: extreme learning machine, elastic net linear model, linear support vector regression, and multivariate adaptive regression splines. Core samples from the YuDong-Nan shale gas field, located in the Sichuan basin, were used to evaluate the hybrid strategy. The findings show that combining machine learning models with an evolutionary algorithms in a hybrid fashion produce flexible models that accurately predict TOC. The results show that, independent of the metaheuristic used to guide the model selection, optimized extreme learning machines attained the best performance scores according to six metrics. Such hybrid models can be used in exploratory geological research, particularly for unconventional oil and gas resources.

Combinations of genes at the 16p11.2 and 22q11.2 CNVs contribute to neurobehavioral traits.

The 16p11.2 and 22q11.2 copy number variants (CNVs) are associated with neurobehavioral traits including autism spectrum disorder (ASD), schizophrenia, bipolar disorder, obesity, and intellectual disability. Identifying specific genes contributing to each disorder and dissecting the architecture of CNV-trait association has been difficult, inspiring hypotheses of more complex models, such as multiple genes acting together. Using multi-tissue data from the GTEx consortium, we generated pairwise expression imputation models for CNV genes and then applied these elastic net models to GWAS for: ASD, bipolar disorder, schizophrenia, BMI (obesity), and IQ (intellectual disability). We compared the variance in these five traits explained by gene pairs with the variance explained by single genes and by traditional interaction models. We also modeled polygene region-wide effects using summed predicted expression ranks across many genes to create a regionwide score. We found that in all CNV-trait pairs except for bipolar disorder at 22q11.2, pairwise effects explain more variance than single genes. Pairwise model superiority was specific to the CNV region for all 16p11.2 traits and ASD at 22q11.2. We identified novel individual genes over-represented in top pairs that did not show single-gene signal. We also found that BMI and IQ have significant regionwide association with both CNV regions. Overall, we observe that genetic architecture differs by trait and region, but 9/10 CNV-trait combinations demonstrate evidence for multigene contribution, and for most of these, the importance of combinatorial models appears unique to CNV regions. Our results suggest that mechanistic insights for CNV pathology may require combinational models.

Appraisal of Different Artificial Intelligence Techniques for the Prediction of Marble Strength

Rock strength, specifically the uniaxial compressive strength (UCS), is a critical parameter mostly used in the effective and sustainable design of tunnels and other engineering structures. This parameter is determined using direct and indirect methods. The direct methods involve acquiring an NX core sample and using sophisticated laboratory procedures to determine UCS. However, the direct methods are time-consuming, expensive, and can yield uncertain results due to the presence of any flaws or discontinuities in the core sample. Therefore, most researchers prefer indirect methods for predicting rock strength. In this study, UCS was predicted using seven different artificial intelligence techniques: Artificial Neural Networks (ANNs), XG Boost Algorithm, Random Forest (RF), Support Vector Machine (SVM), Elastic Net (EN), Lasso, and Ridge models. The input variables used for rock strength prediction were moisture content (MC), P-waves, and rebound number (R). Four performance indicators were used to assess the efficacy of the models: coefficient of determination (R2), Root Mean Square Error (RMSE), Mean Square Error (MSE), and Mean Absolute Error (MAE). The results show that the ANN model had the best performance indicators, with values of 0.9995, 0.2634, 0.0694, and 0.1642 for R2, RMSE, MSE, and MAE, respectively. However, the XG Boost algorithm model performance was also excellent and comparable to the ANN model. Therefore, these two models were proposed for predicting UCS effectively. The outcomes of this research provide a theoretical foundation for field professionals in predicting the strength parameters of rock for the effective and sustainable design of engineering structures

Integrating eQTL and GWAS data characterises established and identifies novel migraine risk loci

Migraine—a painful, throbbing headache disorder—is the most common complex brain disorder, yet its molecular mechanisms remain unclear. Genome-wide association studies (GWAS) have proven successful in identifying migraine risk loci; however, much work remains to identify the causal variants and genes. In this paper, we compared three transcriptome-wide association study (TWAS) imputation models—MASHR, elastic net, and SMultiXcan—to characterise established genome-wide significant (GWS) migraine GWAS risk loci, and to identify putative novel migraine risk gene loci. We compared the standard TWAS approach of analysing 49 GTEx tissues with Bonferroni correction for testing all genes present across all tissues (Bonferroni), to TWAS in five tissues estimated to be relevant to migraine, and TWAS with Bonferroni correction that took into account the correlation between eQTLs within each tissue (Bonferroni-matSpD). Elastic net models performed in all 49 GTEx tissues using Bonferroni-matSpD characterised the highest number of established migraine GWAS risk loci (n = 20) with GWS TWAS genes having colocalisation (PP4 > 0.5) with an eQTL. SMultiXcan in all 49 GTEx tissues identified the highest number of putative novel migraine risk genes (n = 28) with GWS differential expression at 20 non-GWS GWAS loci. Nine of these putative novel migraine risk genes were later found to be at and in linkage disequilibrium with true (GWS) migraine risk loci in a recent, more powerful migraine GWAS. Across all TWAS approaches, a total of 62 putative novel migraine risk genes were identified at 32 independent genomic loci. Of these 32 loci, 21 were true risk loci in the recent, more powerful migraine GWAS. Our results provide important guidance on the selection, use, and utility of imputation-based TWAS approaches to characterise established GWAS risk loci and identify novel risk gene loci.

Two sides of the same coin: distinct neuroanatomical patterns predict crystallized and fluid intelligence in adults.

Crystallized intelligence (Gc) and fluid intelligence (Gf) are regarded as distinct intelligence components that statistically correlate with each other. However, the distinct neuroanatomical signatures of Gc and Gf in adults remain contentious. Machine learning cross-validated elastic net regression models were performed on the Human Connectome Project Young Adult dataset (N = 1089) to characterize the neuroanatomical patterns of structural magnetic resonance imaging variables that are associated with Gc and Gf. The observed relationships were further examined by linear mixed-effects models. Finally, intraclass correlations were computed to examine the similarity of the neuroanatomical correlates between Gc and Gf. The results revealed distinct multi-region neuroanatomical patterns predicted Gc and Gf, respectively, which were robust in a held-out test set (R2 = 2.40, 1.97%, respectively). The relationship of these regions with Gc and Gf was further supported by the univariate linear mixed effects models. Besides that, Gc and Gf displayed poor neuroanatomical similarity. These findings provided evidence that distinct machine learning-derived neuroanatomical patterns could predict Gc and Gf in healthy adults, highlighting differential neuroanatomical signatures of different aspects of intelligence.

Using machine-learning methods to predict in-hospital mortality through the Elixhauser index: A Medicare data analysis.

Accurate in-hospital mortality prediction can reflect the prognosis of patients, help guide allocation of clinical resources, and help clinicians make the right care decisions. There are limitations to using traditional logistic regression models when assessing the model performance of comorbidity measures to predict in-hospital mortality. Meanwhile, the use of novel machine-learning methods is growing rapidly. In 2021, the Agency for Healthcare Research and Quality published new guidelines for using the Present-on-Admission (POA) indicator from the International Classification of Diseases, Tenth Revision, for coding comorbidities to predict in-hospital mortality from the Elixhauser's comorbidity measurement method. We compared the model performance of logistic regression, elastic net model, and artificial neural network (ANN) to predict in-hospital mortality from Elixhauser's measures under the updated POA guidelines. In this retrospective analysis, 1,810,106 adult Medicare inpatient admissions from six US states admitted after September 23, 2017, and discharged before April 11, 2019 were extracted from the Centers for Medicare and Medicaid Services data warehouse. The POA indicator was used to distinguish pre-existing comorbidities from complications that occurred during hospitalization. All models performed well (C-statistics >0.77). Elastic net method generated a parsimonious model, in which there were five fewer comorbidities selected to predict in-hospital mortality with similar predictive power compared to the logistic regression model. ANN had the highest C-statistics compared to the other two models (0.800 vs. 0.791 and 0.791). Elastic net model and AAN can be applied successfully to predict in-hospital mortality.

Predicting the severity of occupational accidents in the construction industry using standard and regularized logistic regression models

Occupational accidents in the construction industry occur frequently when compared with other industries. Construction accidents still cannot be prevented at the desired level. These accidents have been explored for prediction purposes using traditional statistical and machine-learning techniques. However, some challenges regarding imbalanced and multicollinearity problems present in the dataset are not considered while analyzing large and categorical variable-rich data. For this reason, this study aims to predict the severity of non-fatal construction accidents considering these challenges to obtain accurate results. In this study, standard binary logistic regression with Firth, Ridge, Lasso, and Elastic Net regularized models were utilized for the prediction of lost workdays in the construction industry. The data used were classified into five groups: victim, workplace, accident time, accident and sequence of events, and post-accident state-related variables. The results showed that Firth’s logistic model is the best-performing model and age, education, vocational education, size of the workplace, project type, working environment, accident month and year, general and specific activities, material agent of deviation, type of injury, and part of body injured are the most significant variables. This study by providing interpretable machine learning tools is the first attempt to use proposed models in the construction safety field.

18 F]-FDG PET radiomic model as prognostic biomarker in diffuse large B-cell lymphoma.

To evaluate the association between radiomic features (RFs) extracted from 18 F-FDG PET/CT (18 F-FDG-PET) with progression-free survival (PFS) and overall survival (OS) in diffuse large-B-cell lymphoma (DLBCL) patients eligible to first-line chemotherapy. DLBCL patients who underwent 18 F-FDG-PET prior to first-line chemotherapy were retrospectively analyzed. RFs were extracted from the lesion showing the highest uptake. A radiomic score to predict PFS and OS was obtained by multivariable Elastic Net Cox model. Radiomic univariate model, clinical and combined clinical-radiomic multivariable models to predict PFS and OS were obtained. 112 patients were analyzed. Median follow-up was 34.7months (Inter-Quartile Range (IQR) 11.3-66.3months) for PFS and 41.1 (IQR 18.4-68.9) for OS. Radiomic score resulted associated with PFS and OS (p<0.001), outperforming conventional PET parameters. C-index (95% CI) for PFS prediction were 0.67 (0.58-0.76), 0.81 (0.75-0.88) and 0.84 (0.77-0.91) for clinical, radiomic and combined clinical-radiomic model, respectively. C-index for OS were 0.77 (0.66-0.89), 0.84 (0.76-0.91) and 0.90 (0.81-0.98). In the Kaplan-Meier analysis (low-IPI vs. high-IPI), the radiomic score was significant predictor of PFS (p<0.001). The radiomic score was an independent prognostic biomarker of survival in DLBCL patients. The extraction of RFs from baseline 18 F-FDG-PET might be proposed in DLBCL to stratify high-risk versus low-risk patients of relapse after first-line therapy, especially in low-IPI patients.