RESUMO

Breast cancer is a leading cause of mortality among women, with recurrence prediction remaining a significant challenge. In this context, artificial intelligence application and its resources can serve as a powerful tool in analyzing large amounts of data and predicting cancer recurrence, potentially enabling personalized medical treatment and improving the patient's quality of life. Thus, the systematic review examines the role of AI in predicting breast cancer recurrence using clinical data, imaging data, and combined datasets. Support Vector Machine (SVM) and Neural Networks, especially when applied to combined data, demonstrate strong potential in improving prediction accuracy. SVMs are effective with high-dimensional clinical data, while Neural Networks in genetic and molecular analysis. Despite these advancements, limitations such as dataset diversity, sample size, and evaluation standardization persist, emphasizing the need for further research. AI integration in recurrence prediction offers promising prospects for personalized care but requires rigorous validation for safe clinical application.

RESUMO

Genomic selection (GS) is a predictive methodology that is revolutionizing plant and animal breeding. However, the practical application of the GS methodology is challenging since a successful implementation requires a good identification of the best lines. For this reason, some approaches have been proposed to be able to select the top (or bottom) lines with more Precision. Despite the varying popularity of methods, with some being notably more efficient than others, this paper delves into the fundamentals of these techniques. We used five models/methods: (1) RC, known as the Bayesian Best Linear Unbiased Predictor (GBLUP); (2) R, which is like RC but uses a threshold; (3) RO, Regression Optimum, that leverages the RC model in its training process to fine-tune the threshold; (4) B, Threshold Bayesian Probit Binary model (TGBLUP) with a threshold of 0.5 to classify the cultivars as top or non-top; (5) BO is the TGBLUP but the threshold used is an optimal probability threshold that guarantees similar Sensitivity and Specificity. We also present a benchmark comparison of existing approaches for selecting the top (or bottom) performers, utilizing five real datasets for comprehensive analysis. For methods that necessitate a rigorous tuning process, we suggest a streamlined tuning approach that significantly decreases implementation time without notably compromising performance. Our analysis revealed that the regression optimal (RO) method outperformed other models across the five real datasets, achieving superior results in terms of the F1 score. Specifically, RO was more effective than models R, B, RC, and BO by 60.87, 42.37, 17.63, and 9.62%, respectively. When looking at the Kappa coefficient, the RO model was better than models B, BO, R, and RC by 37.46, 36.21, 52.18, and 3.95%, respectively. In terms of Sensitivity, the RO model outperformed models B, R, and RC by 145.74, 250.41, and 86.20, respectively. The second-best model was the model BO. It is important to point out that in the first stage, the BO and RO approaches train a classification and regression model, respectively, to classify the lines as the top (bottom) or not the top (not the bottom). However, both the BO and RO approaches optimize a threshold in the second stage to perform the classification of the lines that minimize the difference between the Sensitivity and Specificity. The BO and RO methods are superior for the selection of the top (or bottom) lines. For this reason, we encourage breeders to adopt these approaches to increase genetic gain in plant breeding programs.

RESUMO

Consistency of pituitary macroadenomas is a key determinant in surgical outcomes, with non-soft consistency linked to more complications and incomplete resections. This study aimed to develop a machine learning model to predict the consistency of pituitary macroadenomas to improve surgical planning and outcomes. A retrospective study of patients with pituitary macroadenomas was conducted. Data included brain magnetic resonance imaging findings (diameter and apparent diffusion coefficient), patient demographics (age and sex), and tumor consistency. Seventy patients were evaluated, 59 with soft consistency and 11 with non-soft consistency. The support vector machine (SVM) was the best model with ROC AUC score of 83.3% [95% CI 65.8, 97.6], AP AUC of 69.8% [95% CI 41.3, 91.1], sensitivity of 73.1% [95% CI 44.4, 100], specificity of 89.8% [95% CI 82, 96.7], F1 score of 0.63 [95% CI 0.36, 0.83], and Matthews correlation coefficient score of 0.57 [95% CI 0.29, 0.79]. These findings indicate a significant improvement over random classification, as confirmed by a permutation test (p < 0.05). Additionally, the model had a 67.4% probability of outperforming the second-best model in cross-validation, as determined through Bayesian analysis, and demonstrated statistical significance (p < 0.05) compared to non-ensemble models. Using explainability heuristics, both 2D and 3D probability maps highlighted areas with a higher probability of non-soft consistency. The attributes most influential in the correct classification by our best model were male sex and age ≤ 42.25 years. Despite some limitations, the SVM model showed promise in predicting tumor consistency, which could aid in surgical planning. To address concerns about generalizability, we have created an open-access repository to promote future external validation studies and collaboration with other research centers, with the goal of enhancing model prediction through transfer learning.

RESUMO

To evaluate and compare the effectiveness of prediction models for Argentine squid Illex argentinus trawling grounds in the Southwest Atlantic high seas based on vessel position and fishing log data, this study used AIS datasets and fishing log datasets from fishing seasons spanning 2019-2024 (December to June each year). Using a spatial resolution of 0.1° × 0.1° and a monthly temporal resolution, we constructed two datasets-one based on vessel positions and the other on fishing logs. Fishing ground levels were defined according to the density of fishing locations, and combined with oceanographic data (sea surface temperature, 50 m water temperature, sea surface salinity, sea surface height, and mixed layer depth). A CNN-Attention deep learning model was applied to each dataset to develop Illex argentinus trawling ground prediction models. Model accuracy was then compared and potential causes for differences were analyzed. Results showed that the vessel position-based model had a higher accuracy (Accuracy = 0.813) and lower loss rate (Loss = 0.407) than the fishing log-based model (Accuracy = 0.727, Loss = 0.513). The vessel-based model achieved a prediction accuracy of 0.763 on the 2024 test set, while the fishing log-based model reached an accuracy of 0.712, slightly lower than the former, indicating the high accuracy and unique advantages of the vessel position-based model in predicting fishing grounds. Using CPUE from fishing logs as a reference, we found that the vessel position-based model performed well from January to April, whereas the CPUE-based model consistently maintained good accuracy across all months. The 2024 fishing season predictions indicated the formation of primary fishing grounds as early as January 2023, initially near the 46° S line of the Argentine Exclusive Economic Zone, with grounds shifting southeastward from March onward and reaching around 42° S by May and June. This study confirms the reliability of vessel position data in identifying fishing ground information and levels, with higher accuracy in some months compared to the fishing log-based model, thereby reducing the data lag associated with fishing logs, which are typically available a year later. Additionally, national-level fishing log data are often confidential, limiting the ability to fully consider fishing activities across the entire fishing ground region, a limitation effectively addressed by AIS vessel position data. While vessel data reflects daily catch volumes across vessels without distinguishing CPUE by species, log data provide a detailed daily CPUE breakdown by species (e.g., Illex argentinus). This distinction resulted in lower accuracy for vessel-based predictions in December 2023 and May-June 2024, suggesting the need to incorporate fishing log data for more precise assessments of fishing ground levels or resource abundance during those months. Given the near-real-time nature of vessel position data, fishing ground dynamics can be monitored in near real time. The successful development of vessel position-based prediction models aids enterprises in reducing fuel and time costs associated with indiscriminate squid searches, enhancing trawling efficiency. Additionally, such models support quota management in global fisheries by optimizing resource use, reducing fishing time, and consequently lowering carbon emissions and environmental impact, while promoting marine environmental protection in the Southwest Atlantic high seas.

RESUMO

COVID-19 is no longer a global health emergency, but it remains challenging to predict its prognosis. Objective: To develop and validate an instrument to predict COVID-19 progression for critically ill hospitalized patients in a Brazilian population. Methodology: Observational study with retrospective follow-up. Participants were consecutively enrolled for treatment in non-critical units between January 1, 2021, to February 28, 2022. They were included if they were adults, with a positive RT-PCR result, history of exposure, or clinical or radiological image findings compatible with COVID-19. The outcome was characterized as either transfer to critical care or death. Predictors such as demographic, clinical, comorbidities, laboratory, and imaging data were collected at hospitalization. A logistic model with lasso or elastic net regularization, a random forest classification model, and a random forest regression model were developed and validated to estimate the risk of disease progression. Results: Out of 301 individuals, the outcome was 41.8 %. The majority of the patients in the study lacked a COVID-19 vaccination. Diabetes mellitus and systemic arterial hypertension were the most common comorbidities. After model development and cross-validation, the Random Forest regression was considered the best approach, and the following eight predictors were retained: D-dimer, Urea, Charlson comorbidity index, pulse oximetry, respiratory frequency, Lactic Dehydrogenase, RDW, and Radiologic RALE score. The model's bias-corrected intercept and slope were - 0.0004 and 1.079 respectively, the average prediction error was 0.028. The ROC AUC curve was 0.795, and the variance explained was 0.289. Conclusion: The prognostic model was considered good enough to be recommended for clinical use in patients during hospitalization (https://pedrobrasil.shinyapps.io/INDWELL/). The clinical benefit and the performance in different scenarios are yet to be known.

RESUMO

Influenza A and B viruses represent significant global health threats, contributing substantially to morbidity and mortality rates. However, a comprehensive understanding of the molecular epidemiology of these viruses in Brazil, a continental-size country and a crucial hub for the entry, circulation, and dissemination of influenza viruses within South America, still needs to be improved. This study addresses this gap by consolidating data and samples across all Brazilian macroregions, as part of the Center for Viral Surveillance and Serological Assessment project, together with an extensive number of other Brazilian sequences provided by a public database during the epidemic seasons spanning 2021-23. Phylogenetic analysis of the hemagglutinin segment of influenza A/H1N1pdm09, A/H3N2, and influenza B/Victoria-lineage viruses revealed that in 2021 and in the first semester of 2022, the A/H3N2 2a.3 strain was the predominant circulating strain. Subsequently, the A/H3N2 2b became the prevalent strain until October, when it was substituted by A/H1N1pdm09 5a.2a and 5a.2a.1 lineages. This scenario was maintained during the year of 2023. B/Victoria emerged and circulated at low levels between December 2021 and September 2022 and then became coprevalent with A/H1N1pdm09 5a.2a and 5a.2a.1 lineages. The comparison between the vaccine strain A/Darwin/9/2021 and circulating viruses revealed shared mutations to aspartic acid at residues 186 and 225 across all A/H3N2 lineages from 2021 to 2023, altering the charge in the receptor-binding domain. For A/H1N1pdm09, the 2022 consensus of 5a.2a.1 and the vaccine strain A/Victoria/2570/2019 showed 14 amino acid substitutions. Key residues H180, D187, K219, R223, E224, and T133 are involved in hydrogen interactions with sialic acids, while N130, K142, and D222 may contribute to distance interactions based on docking analyses. Importantly, distinct influenza A lineage frequency patterns were observed across Brazil's macroregions, underscoring the regional variations in virus circulation. This study characterizes influenza A and B viruses circulating in Brazil, providing insights into their circulation patterns and dynamics across Brazilian macroregions. These findings hold significant implications for public health interventions, informing strategies to mitigate transmission risks, optimize vaccination efforts, and enhance outbreak control measures.

RESUMO

BACKGROUND AND OBJECTIVE: The aim of this retrospective observational case-control study was to examine the significance of different renal Doppler marker variations within the initial 24-hour period as potential predictors of Acute Kidney Injury (AKI) in patients with sepsis. METHODS: A total of 198 sepsis patients were enrolled and categorized into two groups: the AKI group (n = 136) and the non-AKI group (n = 62). Three renal Doppler indices, Renal Resistive Index (RRI), Power Doppler Ultrasound (PDU) score and Renal Venous Stasis Index (RVSI), were measured within 6h (T0) and at 24h (T1) after ICU admission. RESULTS: The AKI group had more hypertension patients than the non-AKI group (p = 0.047). The cases of the AKI group showed higher levels of CRP (p = 0.001), PCT (p < 0.001), lactate (p < 0.001), AST (p = 0.003), ALT (p = 0.049), total bilirubin (p = 0.034), BNP (p = 0.019) and cTnI (p = 0.012). The RRI at T1 was significantly higher in the AKI group (p = 0.037). AKI group exhibited a lower incidence of reduced RRI at T1 compared with non-AKI group (p < 0.001). After controlling for age, sex, and BMI through partial correlation analysis, the results indicated significant associations between SA-AKI and CVP (r = -0.473), SOFA score (r = 0.425), lactate (r = 0.378), and RRI reduction (r = -0.344) in sepsis patients. The multivariate logistic regression analysis showed that variables including CVP, SOFA score, CRP, lactate, VIS, and RRI not reduced following 24h of ICU treatment were predictive indicators for early detection of SA-AKI in sepsis patients. CONCLUSION: CVP, SOFA score, CRP, lactate, VIS, and RRI not reduction following 24h of ICU treatment can be utilized as predictive indicators for early detection of SA-AKI in sepsis patients.

RESUMO

Non-native species can be major drivers of ecosystem alteration, especially through changes in trophic interactions. Successful non-native species have been predicted to have greater resource use efficiency relative to trophically analogous native species (the Resource Consumption Hypothesis), but rigorous evidence remains equivocal. Here, we tested this proposition quantitatively in a global meta-analysis of comparative functional response studies. We calculated the log response ratio of paired non-native and native species functional responses, using attack rate and maximum consumption rate parameters as response variables. Explanatory variables were consumer taxonomic group and functional feeding group, habitat, native assemblage latitude, and non-native species taxonomic distinctiveness. Maximum consumption rates for non-native species were 70% higher, on average, than those of their native counterparts; attack rates also tended to be higher, but not significantly so. The magnitude of maximum consumption rate effect sizes varied with consumer taxonomic group and functional feeding group, being highest in favour of non-natives for molluscs and herbivores. Consumption rate differences between non-native and native species tended to be greater for freshwater taxa, perhaps reflecting sensitivity of insular freshwater food webs to novel consumers; this pattern needs to be explored further as additional data are obtained from terrestrial and marine ecosystems. In general, our results support the Resource Consumption Hypothesis, which can partly explain how successful non-native species can reduce native resource populations and restructure food webs.

RESUMO

With growing evidence that genomic selection (GS) improves genetic gains in plant breeding, it is timely to review the key factors that improve its efficiency. In this feature review, we focus on the statistical machine learning (ML) methods and software that are democratizing GS methodology. We outline the principles of genomic-enabled prediction and discuss how statistical ML tools enhance GS efficiency with big data. Additionally, we examine various statistical ML tools developed in recent years for predicting traits across continuous, binary, categorical, and count phenotypes. We highlight the unique advantages of deep learning (DL) models used in genomic prediction (GP). Finally, we review software developed to democratize the use of GP models and recent data management tools that support the adoption of GS methodology.

RESUMO

BACKGROUND: Extreme hemodynamic changes, especially intraoperative hypotension (IOH), are common and often prolonged during Liver Transplant (LT) surgery and during initial hours of recovery. Hypotension Prediction Index (HPI) software is one of the tools which can help in proactive hemodynamic management. The accuracy of the advanced hemodynamic parameters such as Cardiac Output (CO) and Systemic Vascular Resistance (SVR) obtained from HPI software and prediction performance of the HPI in LT surgery remains unknown. METHODS: This was a retrospective observational study conducted in a tertiary academic center with a large liver transplant program. We enrolled 23 adult LT patients who received both Pulmonary Artery Catheter (PAC) and HPI software monitoring. Primarily, we evaluated agreement between PAC and HPI software measured CO and SVR. A priori, we defined a relative difference of less than 20% between measurements as an adequate agreement for a pair of measurements and estimated the Lin's Concordance Correlation Coefficient and Bland-Altman Limits of Agreement (LOA). Clinically acceptable LOA was defined as ± 1 L.min-1 for CO and ± 200 dynes s.cm-5 for SVR. Secondary outcome was the ability of the HPI to predict future hypotension, defined as Mean Arterial Pressure (MAP) less than 65 mmHg lasting at least one minute. We estimated sensitivity, positive predictive value, and time from alert to hypotensive events for HPI software. RESULTS: Overall, 125 pairs of CO and 122 pairs of SVR records were obtained from 23 patients. Based on our predefined criteria, only 42% (95% CI 30%, 55%) of CO records and 53% (95% CI 28%, 72%) of SVR records from HPI software were considered to agree with those from PAC. Across all patients, there were a total of 1860 HPI alerts (HPI ≥ 85) and 642 hypotensive events (MAP < 65 mmHg). Out of the 642 hypotensive events, 618 events were predicted by HPI alert with sensitivity of 0.96 (95% CI: 0.95). Many times, the HPI value remained above alert level and was followed by multiple hypotensive events. Thus, to evaluate PPV and time to hypotension metric, we considered only the first HPI alert followed by a hypotensive event ("true alerts"). The "true alert" was the first alert when there were several alerts before a hypotension. There were 614 "true alerts" and the PPV for HPI was 0.33 (95% CI 0.31, 0.35). The median time from HPI alert to hypotension was 3.3 [Q1, Q3: 1, 9.3] mins. CONCLUSION: There was poor agreement between the pulmonary artery catheter and HPI software calculated advanced hemodynamic parameters (CO and SVR), in the patients undergoing LT surgery. HPI software had high sensitivity but poor specificity for hypotension prediction, resulting in a high burden of false alarms.

RESUMO

The Min system is a key spatial regulator of cell division in rod-shaped bacteria and the first FtsZ-negative modulator to be recognized. Nevertheless, despite extensive genetic and in vitro studies, the molecular mechanism used by MinC to inhibit Z-ring formation remains incompletely understood. The crystallization of FtsZ in complex with other negative regulators such as SulA and MciZ has provided important structural information to corroborate in vitro experiments and establish the mechanism of Z-ring antagonism by these modulators. However, MinC and FtsZ have so far eluded co-crystallization, probably because their complex is too unstable to be crystallized. To gain structural insight into the mechanism of action of MinC, we determined the solution structure of the N-terminal domain of Bacillus subtilis MinC, and through NMR titration experiments and mutagenesis identified the binding interfaces involved in the MinCN-FtsZ interaction. By using our experimental results as restraints in docking, we also constructed a molecular model for the FtsZ:MinCN complex and validated it by molecular dynamics. The model shows that MinCN binding overlaps with the FtsZ polymerization interface on the C-terminal globular subdomain of FtsZ and, thus, provides a structural basis for MinCN inhibition of FtsZ filament formation. Given that the C-terminal polymerization interface of FtsZ corresponds to the plus end of FtsZ filaments, we propose that capping is the main mechanism employed by MinC to antagonize FtsZ polymerization.

RESUMO

Background/Objectives: Hospitalization among older adults is a growing challenge in Mexico due to the high prevalence of chronic diseases and limited public healthcare resources. This study aims to develop a predictive model for hospitalization using longitudinal data from the Mexican Health and Aging Study (MHAS) using the random forest (RF) algorithm. Methods: An RF-based machine learning model was designed and evaluated under different data partition strategies (ST) with and without variable interaction. Variable importance was assessed based on the mean decrease in impurity and permutation importance, enhancing our understanding of predictors of hospitalization. The model's robustness was ensured through modified nested cross-validation, with evaluation metrics including sensitivity, specificity, and the kappa coefficient. Results: The model with ST2, incorporating interaction and a 20% test proportion, achieved the best balance between sensitivity (0.7215, standard error ± 0.0038), and specificity (0.4935, standard error ± 0.0039). Variable importance analysis revealed that functional limitations (e.g., abvd3, 31.1% importance), age (12.75%), and history of cerebrovascular accidents (12.4%) were the strongest predictors. Socioeconomic factors, including education level (12.08%), also emerged as critical predictors, highlighting the model's ability to capture complex interactions between health and socioeconomic variables. Conclusions: The integration of variable importance analysis enhances the interpretability of the RF model, providing novel insights into the predictors of hospitalization in older adults. These findings underscore the potential for clinical applications, including anticipating hospital demand and optimizing resource allocation. Future research will focus on integrating subgroup analyses for comorbidities and advanced techniques for handling missing data to further improve predictive accuracy.

RESUMO

PURPOSE: Oral mucositis (OM) reflects a complex interplay of several risk factors. Machine learning (ML) is a promising frontier in science, capable of processing dense information. This study aims to assess the performance of ML in predicting OM risk in patients undergoing head and neck radiotherapy. METHODS: Clinical data were collected from 157 patients with oral and oropharyngeal squamous cell carcinoma submitted to radiotherapy. Grade 2 OM or higher was considered (NCI). Two dataset versions were used; in the first version, all data were considered, and in the second version, a feature selection was added. Age, smoking status, surgery, radiotherapy prescription dose, treatment modality, histopathological differentiation, tumor stage, presence of oral cancer lesion, and tumor location were selected as key features. The training process used a fivefold cross-validation strategy with 10 repetitions. A total of 4 algorithms and 3 scaling methods were trained (12 models), without using data augmentation. RESULTS: A comparative assessment was performed. Accuracy greater than 55% was considered. No relevant results were achieved with the first version, closest performance was Decision Trees with 52% of accuracy, 42% of sensitivity, and 60% of specificity. For the second version, relevant results were achieved, K-Nearest Neighbors outperformed with 64% accuracy, 58% sensitivity, and 68% specificity. CONCLUSION: ML demonstrated promising results in OM risk prediction. Model improvement was observed after feature selection. Best result was achieved with the KNN model. This is the first study to test ML for OM risk prediction using clinical data.

RESUMO

BACKGROUND: There has been debate over whether the existing World Health Organization (WHO) criteria accurately represent the severity of maternal near misses. OBJECTIVE: This study assessed the diagnostic accuracy of two WHO clinical and laboratory organ dysfunction markers for determining the best cutoff values in a Latin American setting. METHODS: A prospective multicenter cohort study was conducted in five Latin American countries. Patients with severe maternal complications were followed up from admission to discharge. Organ dysfunction was determined using clinical and laboratory data, and participants were classified according to severe maternal outcomes. This study compares the diagnostic criteria of Latin American Centre for Perinatology, Network for Adverse Maternal Outcomes (CLAP/NAMO) to WHO standards. RESULTS: Of the 698 women studied, 15.2% had severe maternal outcomes. Most measured variables showed significant differences between individuals with and without severe outcomes (all P-values <0.05). Alternative cutoff values suggested by CLAP/NAMOs include pH ≤7.40, lactate ≥2.3 mmol/L, respiratory rate ≥ 24 bpm, oxygen saturation ≤ 96%, PaO2/FiO2 ≤ 342 mmHg, platelet count ≤189 × 109 × mm3, serum creatinine ≥0.8 mg/dL, and total bilirubin ≥0.67 mg/dL. No significant differences were found when comparing the diagnostic performance of the CLAP/NAMO criteria to that of the WHO standards. CONCLUSION: The CLAP/NAMO values were comparable to the WHO maternal near-miss criteria, indicating that the WHO standards might not be superior in this population. These findings suggest that maternal near-miss thresholds can be adapted regionally, improving the identification and management of severe maternal complications in Latin America.

RESUMO

BACKGROUND: ML predictive models have shown their capability to improve risk prediction and assist medical decision-making, nevertheless, there is a lack of accuracy systems to early identify future rapid CKD progressors in Colombia and even in South America. OBJECTIVE: The purpose of this study was to develop a series of interpretable machine learning models that predict GFR at 6-months, 9-months, and 12-months. STUDY DESIGN AND SETTING: Over 29,000 CKD patients stage 1 to 3b (estimated GFR, <60 mL/min/1.73 m2) with an average of 3-year follow-up data were included. We used the machine learning extreme gradient boosting (XGBoost) to build three models to predict the next eGFR. Models were internally and externally validated. In addition, we included SHapley Additive exPlanation (SHAP) values to offer interpretable global and local prediction models. RESULTS: All models showed a good performance in development and external validation. However, the 6-months XGBoost prediction model showed the best performance in internal (MAE average = 6.07; RSME = 78.87), and in external validation (MAE average = 6.45, RSME = 18.94). The top 3 most influential features that pushed the predicted eGFR value to lower values were the interpolated values for eGFR and creatinine, and eGFR at baseline. CONCLUSION: In the current study we have developed and validated machine learning models to predict the next eGFR value at different intervals. Furthermore, we attempted to approach the need for prediction explanation by offering transparent predictions.

RESUMO

ABSTRACT Objective: To evaluate the predictive validity and cut-off point of heart rate and blood pressure on heart rate variability (HRV) changes in children with and without obesity. Methods: This study included 125 children, of whom 41 were normal weight and 84 were obese. Anthropometry, blood pressure, heart rate, and HRV were measured using an electronic scale and stadiometer, a sphygmomanometer, and HRV monitor. In addition, the receiver operating characteristic (ROC) curve was obtained by statistical analysis of the data. Results: Heart rate proved to be a good predictor for changes in the square root of the mean of the square of the differences between consecutive NN intervals (RMSSD) parameter in students of both sexes for the normal-weight group (ROC 0.89; 95%CI 0.77-1.00) and obesity (ROC 0.90; 95%CI 0.83-0.97). In addition, the heart rate cut-off point for alterations in the RMSSD parameter for normal-weight boys was 93 bpm (sensitivity 100.00% and specificity 87.50%) and for boys with obesity, the established cut-off point was 91 bpm (sensitivity 94.74% and specificity 63.64%). Heart rate also proved to be a good predictor considering low-frequency/high-frequency ratio (LF/HF) and standard deviation of long-term continuous NN intervals /standard deviation of the instantaneous variability of continuous NN intervals in the Poincaré graph ratio (SD2/SD1). Systolic and diastolic blood pressures were good predictors in more specific stratifications and, therefore, can be used in some cases. Conclusions: The predictive validity of heart rate was shown to be at a good level, with high sensitivity and acceptable specificity for the cut-off points according to the different analyses stratified by gender and nutritional status. In this sense, health professionals will be able to use heart rate to estimate cardiovascular risk in children of different sexes and nutritional status.

RESUMO Objetivo: Avaliar a validade preditiva e o ponto de corte da frequência cardíaca e da pressão arterial nas alterações da variabilidade da frequência cardíaca (VFC) em crianças com e sem obesidade. Métodos: Foram incluídas 125 crianças neste estudo, sendo 41 com peso normal e 84 com obesidade. Antropometria, pressão arterial, frequência cardíaca e VFC foram medidas por meio de balança eletrônica e estadiômetro, esfigmomanômetro e monitor de VFC. Além disso, a curva característica de operação do receptor (ROC) foi obtida pela análise estatística dos dados. Resultados: A frequência cardíaca mostrou-se um bom preditor de alterações no parâmetro de raiz quadrada da média do quadrado das diferenças entre os intervalos NN consecutivos (RMSSD) em escolares de ambos os sexos para o grupo de peso normal (ROC 0,89; IC95% 0,77-1,00) e obesidade (ROC 0,90; IC95% 0,83-0,97). Além disso, o ponto de corte da frequência cardíaca para alterações no parâmetro RMSSD para meninos com peso normal foi de 93 bpm (sensibilidade 100,00% e especificidade 87,50%), e para meninos com obesidade, o ponto de corte estabelecido foi de 91 bpm (sensibilidade 94,74% e especificidade 63,64%). A frequência cardíaca também se mostrou um bom preditor considerando os índices da relação baixa frequência/alta frequência (LF/HF) e razão desvio padrão de intervalos NN contínuos de longo prazo/desvio padrão da variabilidade instantânea de intervalos NN contínuos na relação gráfica de Poincaré (SD2/SD1). As pressões arteriais sistólicas e diastólicas foram bons preditores em estratificações mais específicas e, portanto, podem ser utilizadas em alguns casos. Conclusões: A validade preditiva da frequência cardíaca mostrou-se em bom nível, com alta sensibilidade e especificidade aceitável para os pontos de corte, conforme as diferentes análises estratificadas por sexo e estado nutricional. Desta forma, os profissionais de saúde poderão utilizar a frequência cardíaca para estimar o risco cardiovascular em crianças de diferentes sexos e estado nutricional.

RESUMO

Background: Pediatric trauma is a major global health concern, accounting for a substantial proportion of deaths and disease burden from age 5 onwards. Effective triage and management are essential in pediatric trauma care, and prediction models such as the Trauma Injury Severity Score (TRISS) play a crucial role in estimating survival probability and guiding quality improvement. However, TRISS does not account for age-specific factors in pediatric populations, limiting its applicability to younger patients. This study aimed to modify TRISS to account for age for children (Peds-TRISS) and to evaluate its performance relative to the original TRISS. We also assessed survival outcomes to explore the model's potential utility across various clinical settings. These efforts align with quality improvement initiatives to reduce preventable mortality and supporting sustainable development goals. Methods: This retrospective cohort study included patients under 18 years of age who were treated at a hospital in Colombia between 2011 and 2019. New coefficients for TRISS covariates were calculated using logistic regression, with age treated as a continuous variable. Model performance was evaluated based on discrimination (C statistic) and calibration, comparing Peds-TRISS with the original TRISS. Internal validation was conducted using bootstrap resampling. Survival outcomes were assessed using the M and Z statistics, which are commonly used for international trauma outcome comparisons. Results: The study included 1,013 pediatric patients with a median age of 12 years (IQR 5-15), of whom 73% were male. The leading causes of injury were traffic accidents (31.1%), falls (28.8%), and assaults (28.7%). The overall mortality rate was 5.7%. The Peds-TRISS model demonstrated good calibration (HL = 9.7, p = 0.3) and discrimination (C statistic = 0.98, 95% CI 0.97-0.99), with no statistically significant difference in the ROC curve comparison with the original TRISS. Internal validation demonstrated strong performance of Peds-TRISS. The M and Z statistics were 0.93 and 0, respectively, indicating no significant differences between expected and observed survival rates. Conclusions: Most fatalities occurred among adolescents and were due to intentional injuries. The Peds-TRISS model showed a partial improvement in performance compared to the original TRISS, with superior results in terms of calibration, although not in discrimination. These findings highlight the potential of model customization for specific populations. Prospective, multicenter studies are recommended to further validate the model's utility across diverse settings.

RESUMO

Background Predicting potential overcrowding is a significant tool in efficient emergency department (ED) management. Our aim was to develop and validate overcrowding predictive models using accessible and high quality information. Methods Retrospective cohort study of consecutive days in the Hospital Italiano de Buenos Aires ED from june 2016 to may 2018. We estimated hourly NEDOCS score for the entire period, and defined the outcome as Sustained Critical ED Overcrowding (EDOC) equal to occurrence of 8 or more hours with a NEDOCS score ≥ 180. We generated 3 logistic regression predictive models with different related outcomes: beginning, ending or occurrence of Sustained Critical EDOC. We estimated calibration and discrimination as internal (random validation group and bootstrapping) and external validation (different period and different ED). Results The main model included both the beginning and occurrence of NEDOCS, including weather variables, variables related to NEDOCS itself and patient flow variables. The second model considered only the beginning of Sustained Critical EDOC and included variables related to NEDOCS. The last model considered the end of Sustained Critical EDOC and it included variables related to NEDOCS, weather, bed occupancy and management. Discrimination for the main model had an area under the receiveroperator curve of 0.997 (95% CI 0.994 - 1) in the validation group. Calibration for the model was very high on internal validation and acceptable on external validation. Conclusion The Sustained Critical EDOC predictive model includes variables that are easily obtained and can be used for effective resource management in situations of overcrowding.

RESUMO

The extent to which college admissions test scores can forecast college grade point average (GPA) is often evaluated in predictive validity studies using regression analyses. A problem in college admissions processes is that we observe test scores for all the applicants; however, we cannot observe the GPA of applicants who were not selected. The standard solution to tackle this problem has relied upon strong assumptions to identify the exact value of the regression function in the presence of missing data. In this paper, we present an alternative approach based on the theory of partial identifiability that considers a variety of milder assumptions to learn about the regression function. Using a university admissions dataset we illustrate how results can vary as a function of the assumptions that one is willing to make about the selection process.

RESUMO

Traditional testing methods in pharmaceutical development can be time-consuming and costly, but in silico evaluation tools can offer a solution. Our in-house Active-IT system, a Ligand-Based Virtual Screening (LBVS) tool, was developed to predict the biological and pharmacological activities of small organic molecules. It includes four independent modules for generating molecular descriptors (3D-Pharma), machine learning modeling (ExCVBA), a database of bioactivity models, and a prediction module. Activity data collected from the PubChem BioAssay database was used for modelling SVM and Naïve Bayes machine learning methods. Models have been constructed using a recursive stratified partition method and validated through an activity randomization (Y-random) process. Over 3500 bioassays were modeled, each comprising 30 SVM and 30 Naïve Bayes models and 60 randomized models. Bioassays with low performance or discrimination between regular and randomized were discarded. Using the Active-IT system we have evaluated three bioactive compounds of Ayahuasca tea. The predictions were thoroughly validated using known targets described in several public databases. The external validation results are noteworthy, with 16 of 33 (48.5% with p-value.