Background: Classification trees (CTs) are widely used machine learning algorithms with growing applications in clinical research, especially for risk stratification. Their ability to generate interpretable decision rules makes them attractive to healthcare professionals. This review provides an accessible yet rigorous overview of CT methodology for clinicians, highlighting their utility through a case study addressing the “obesity paradox” in critically ill patients. Methods: We describe key methodological aspects of CTs, including model development, pruning, validation, and classification types (simple, ensemble, and hybrid). Using data from the ENPIC (Evaluation of Practical Nutrition Practices in the Critical Care Patient) study, which assessed artificial nutrition in ICU (intensive care unit) patients, we applied various CT approaches—CART (classification and regression trees), CHAID (chi-square automatic interaction detection), and XGBoost (extreme gradient boosting)—and compared them with logistic regression. SHAP (SHapley Additive exPlanation) values were used to interpret ensemble models. Results: CTs allowed for identification of optimal cut-off points in continuous variables and revealed complex, non-linear interactions among predictors. Although the obesity paradox was not confirmed in the full cohort, CTs uncovered a specific subgroup in which obesity was associated with reduced mortality. The ensemble model (XGBoost) achieved the best predictive performance (highest area under the ROC curve), though at the expense of interpretability. Conclusions: CTs are valuable tools in clinical epidemiology, complementing traditional models by uncovering hidden patterns and enhancing risk stratification. While ensemble models offer superior predictive accuracy, their complexity necessitates interpretability techniques such as SHAP. CT-based approaches can guide personalized medicine but require cautious interpretation and external validation.
The importance of artificial nutritional therapy is underrecognized, typically being considered an adjunctive rather than a primary therapy. We aimed to evaluate the influence of nutritional therapy on mortality in critically ill patients. This multicenter prospective observational study included adult patients needing artificial nutritional therapy for >48 h if they stayed in one of 38 participating intensive care units for ≥72 h between April and July 2018. Demographic data, comorbidities, diagnoses, nutritional status and therapy (type and details for ≤14 days), and outcomes were registered in a database. Confounders such as disease severity, patient type (e.g., medical, surgical or trauma), and type and duration of nutritional therapy were also included in a multivariate analysis, and hazard ratios (HRs) and 95% confidence intervals (95%CIs) were reported. We included 639 patients among whom 448 (70.1%) and 191 (29.9%) received enteral and parenteral nutrition, respectively. Mortality was 25.6%, with non-survivors having the following characteristics: older age; more comorbidities; higher Sequential Organ Failure Assessment (SOFA) scores (6.6 ± 3.3 vs 8.4 ± 3.7; P < 0.001); greater nutritional risk (Nutrition Risk in the Critically Ill [NUTRIC] score: 3.8 ± 2.1 vs 5.2 ± 1.7; P < 0.001); more vasopressor requirements (70.4% vs 83.5%; P=0.001); and more renal replacement therapy (12.2% vs 23.2%; P=0.001). Multivariate analysis showed that older age (HR: 1.023; 95% CI: 1.008-1.038; P=0.003), higher SOFA score (HR: 1.096; 95% CI: 1.036-1.160; P=0.001), higher NUTRIC score (HR: 1.136; 95% CI: 1.025-1.259; P=0.015), requiring parenteral nutrition after starting enteral nutrition (HR: 2.368; 95% CI: 1.168-4.798; P=0.017), and a higher mean Kcal/Kg/day intake (HR: 1.057; 95% CI: 1.015-1.101; P=0.008) were associated with mortality. By contrast, a higher mean protein intake protected against mortality (HR: 0.507; 95% CI: 0.263-0.977; P=0.042). Old age, higher organ failure scores, and greater nutritional risk appear to be associated with higher mortality. Patients who need parenteral nutrition after starting enteral nutrition may represent a high-risk subgroup for mortality due to illness severity and problems receiving appropriate nutritional therapy. Mean calorie and protein delivery also appeared to influence outcomes. ClinicaTrials.gov NCT: 03634943.
Dysphagia can be a consequence of prolonged hospitalization in intensive care units (ICUs) due to severe SARS-CoV-2 pneumonia. This study aims at Identifying the risk factors for dysphagia in ICU patients with COVID-19 pneumonia requiring invasive mechanical ventilation, and at determining the frequency of postextubation dysphagia in this population. Observational, descriptive, retrospective, cohort study of SARS-CoV-2 pneumonia patients admitted into the ICUs from March to May 2020. The Modified Viscosity Volume Swallowing Test (mV-VST) was used to screening for dysphagia during the first 48 h of extubation in patients requiring mechanical ventilation. Descriptive statistics, univariate and multivariate analyses were conducted. A logistic regression was applied to construct a predictive model of dysphagia. A total of 232 patients were admitted into the ICUs (age [median 60.5 years (95% CI: 58.5 to 61.9)]; male [74.1% (95% CI: 68.1 to 79.4)]; APACHE II score [median 17.7 (95% CI: 13.3 to 23.2)]; length of mechanical ventilation [median 14 days (95% CI: 11 to 16)]; prone position [79% (95% CI: 72.1 to 84.6)]; respiratory infection [34.5% (95% CI: 28.6 to 40.9)], renal failure [38.5% (95% CI: 30 to 50)])). 72% (167) of patients required intubation; 65.9% (110) survived; and in 84.5% (93) the mV-VST was performed. Postextubation dysphagia was diagnosed in 26.9% (25) of patients. APACHE II, prone position, length of ICU and hospital stay, length of mechanical ventilation, tracheostomy, respiratory infection and kidney failure developed during admission were significantly associated (p < 0.05) with dysphagia. Dysphagia was independently explained by the APACHE II score (OR: 1.1; 95% CI: 1.01 to 1.3; p = 0.04) and tracheostomy (OR: 10.2; 95% CI: 3.2 to 32.1) p < 0.001). The predictive model forecasted dysphagia with a good ROC curve (AUC: 0.8; 95% CI: 0.7 to 0.9). Dysphagia affects almost one-third of patients with SARS-COV-2 pneumonia requiring intubation in the ICU. The risk of developing dysphagia increases with prolonged mechanical ventilation, tracheostomy, and poorer prognosis on admission (worst APACHE II score).
Aspiration and dysphagia are frequent in critically ill patients, and evidence of the validity of bedside screening tests is lacking. This study evaluated the modified Volume-Viscosity Swallow Test (mV-VST) as a screening tool for aspiration and dysphagia in intensive care unit patients. An observational, prospective longitudinal cohort single-center study included patients older than 18 years old, on mechanical ventilation for at least 48 h, conscious and cooperative. Patients had been admitted in intensive care between March 2016 and August 2019 at a university hospital in Spain. Data from the mV-VST and the flexible endoscopic evaluation of swallowing (FEES) test in extubated and tracheostomized patients were collected; the ROC curve was obtained for each group, and the sensitivity (Se), specificity (Sp), positive (pPV) and negative (nPV) predictive values of mV-VST were calculated and compared with the FEES results. We calculated percentages and 95% confidence intervals (CI) for qualitative variables and means or medians for quantitative variables according to the Shapiro-Wilk test. A univariate analysis identified dysphagia risk factors in each group. The study included 87 patients: 44 extubated and 43 tracheostomized with similar age, body mass index, Sequential Organ Failure Assessment, Charlson comorbidity index, type and reason for admission. Aspiration with FEES was significantly higher in extubated patients than in tracheostomized patients, 43.2% vs. 23.2%, respectively, p = 0.04. With the mV-VST, aspiration was detected in 54.5% of extubated patients and in 39.5% of tracheostomized patients. In the extubated group, the Se of mV-VST to detect aspiration was 89.5%, Sp was 72%, and nPV was 90%. In the tracheostomized group, Se was 100%, Sp was 78.8%, and nPV was 100%. The ROC curve showed that mV-VST similarly identifies aspiration in extubated and tracheostomized patients. Dysphagia and aspiration are frequent amongst patients in intensive care after mechanical ventilation. The mV-VST is a valid screening tool to detect aspiration and dysphagia in extubated and tracheostomized patients.
Adequate medical nutrition therapy for critically ill patients has an impact on their prognoses. However, it requires an individualized approach that takes into account the activity (phases of metabolic stress) and particularity of these patients. We propose a comprehensive strategy considering the patients’ nutritional status and the set of modifiable circumstances in these patients, in order to optimize/support nutritional efficiency: (1) A detailed anamnesis and an adequate initial nutritional assessment must be performed in order to implement medical nutrition therapy that is in line with the needs and characteristics of each patient. Furthermore, risks associated with refeeding syndrome, nutritrauma or gastrointestinal dysfunction must be considered and prevented. (2) A safe transition between nutrition therapy routes and between health care units will greatly contribute to recovery. The main objective is to preserve lean mass in critically ill patients, considering metabolic factors, adequate protein intake and muscle stimulation. (3) Continuous monitoring is required for the successful implementation of any health strategy. We lack precise tools for calculating nutritional efficiency in critically ill patients, therefore thorough monitoring of the process is essential. (4) The medical nutrition approach in critically ill patients is multidisciplinary and requires the participation of the entire team involved. A comprehensive strategy such as this can make a significant difference in the functional recovery of critically ill patients, but leaders must be identified to promote training, evaluation, analysis and feedback as essential components of its implementation, and to coordinate this process with the recognition of hospital management.
Background: Classification trees (CTs) are widely used machine learning algorithms with growing applications in clinical research, especially for risk stratification. Their ability to generate interpretable decision rules makes them attractive to healthcare professionals. This review provides an accessible yet rigorous overview of CT methodology for clinicians, highlighting their utility through a case study addressing the "obesity paradox" in critically ill patients. Methods: We describe key methodological aspects of CTs, including model development, pruning, validation, and classification types (simple, ensemble, and hybrid). Using data from the ENPIC study, which assessed artificial nutrition in ICU patients, we applied various CT approaches—CART, CHAID, and XGBoost—and compared them with logistic regression. SHAP values were used to interpret ensemble models. Results: CTs allowed for identification of optimal cut-off points in continuous variables and revealed complex, non-linear interactions among predictors. Although the obesity paradox was not confirmed in the full cohort, CTs uncovered a specific subgroup in which obesity was associated with reduced mortality. The ensemble model (XGBoost) achieved the best predictive performance (highest AUC), though at the expense of interpretability. Conclusions: CTs are valuable tools in clinical epidemiology, complementing traditional models by uncovering hidden patterns and enhancing risk stratification. While ensemble models offer superior predictive accuracy, their complexity necessitates interpretability techniques such as SHAP. CT-based approaches can guide personalized medicine but require cautious interpretation and external validation.