Derivation and Validation of a Novel Severity Scoring System for Pneumonia at Intensive Care Unit Admission

Development of a Prognostic Scoring System for Tracheostomized Patients Requiring Prolonged Ventilator Care: A Ten-Year Experience in a University-Affiliated Tertiary Hospital

Background and Objectives: This study aimed to assess the value of a novel prognostic model, based on clinical variables, comorbidities, and demographic characteristics, to predict long-term prognosis in patients who received mechanical ventilation (MV) for over 14 days and who underwent a tracheostomy during the first 14 days of MV. Materials and Methods: Data were obtained from 278 patients (66.2% male; median age: 71 years) who underwent a tracheostomy within the first 14 days of MV from February 2011 to February 2021. Factors predicting 1-year mortality after the initiation of MV were identified by binary logistic regression analysis. The resulting prognostic model, known as the tracheostomy-ProVent score, was computed by assigning points to variables based on their respective ß-coefficients. Results: The overall 1-year mortality rate was 64.7%. Six factors were identified as prognostic indicators: platelet count < 150 × 103/μL, PaO2/FiO2 < 200 mmHg, body mass index (BMI) < 23.0 kg/m2, albumin concentration < 2.8 g/dL on day 14 of MV, chronic cardiovascular diseases, and immunocompromised status at admission. The tracheostomy-ProVent score exhibited acceptable discrimination, with an area under the receiver operating characteristic curve (AUC) of 0.786 (95% confidence interval: 0.733-0.833, p < 0.001) and acceptable calibration (Hosmer-Lemeshow chi-square: 2.753, df: 8, p = 0.949). Based on the maximum Youden index, the cut-off value for predicting mortality was set at ≥2, with a sensitivity of 67.4% and a specificity of 76.3%. Conclusions: The tracheostomy-ProVent score is a good predictive tool for estimating 1-year mortality in tracheostomized patients undergoing MV for >14 days. This comprehensive model integrates clinical variables and comorbidities, enhancing the precision of long-term prognosis in these patients.

Ability of the modified NUTRIC score to predict mortality in patients requiring short-term versus prolonged acute mechanical ventilation: a retrospective cohort study

BACKGROUND

The modified NUTRIC (nutritional risk in the critically ill) score has been reported to predict clinical outcomes in critically ill patients. However, the applicability of this score may differ between patients undergoing short-term mechanical ventilation (STMV, < 96 h) and those undergoing prolonged acute mechanical ventilation (PAMV, ⩾96 h), as PAMV patients typically experience significantly higher morbidity and mortality.

OBJECTIVE

This study aimed to investigate the predictive ability of modified NUTRIC score for predicting 28-day mortality in patients receiving STMV and PAMV.

DESIGN

Retrospective single-center cohort study.

METHODS

We enrolled patients who received mechanical ventilation (MV) on the day of admission to the intensive care unit (ICU) from 1 December 2015 to 30 November 2020. Modified NUTRIC scores were calculated based on the clinical data of each patient at ICU admission.

RESULTS

The study population comprised 464 patients, including 319 (68.8%) men with a mean age of 69.7 years. Among these patients, 132 (28.4%) received STMV and 332 (71.6%) received PAMV. The overall 28-day mortality rate was 26.7%, which was significantly higher in STMV patients than in PAMV patients (37.9% versus 22.3%, p < 0.001). Evaluation of the predictive performance of the modified NUTRIC score for 28-day mortality revealed areas under the receiver operating characteristic curves of 0.672 [95% confidence interval (CI): 0.627-0.714] for total patients, 0.819 (95% CI, 0.742-0.880) for STMV patients, and 0.595 (95% CI, 0.540-0.648) for PAMV patients. The best overall cutoff value was 5 in total, STMV, and PAMV patients. This cutoff value was a significant predictor of 28-day mortality based on the Cox proportional hazard model for total [hazards ratio (HR): 2.681; 95% CI: 1.683-4.269] and STMV (HR: 5.725; 95% CI: 2.057-15.931) patients, but not for PAMV patients.

CONCLUSION

The modified NUTRIC score is more effective in predicting 28-day mortality in patients undergoing STMV than in those undergoing PAMV.

ExpertNet: A Deep Learning Approach to Combined Risk Modeling and Subtyping in Intensive Care Units

Risk models play a crucial role in disease prevention, particularly in intensive care units (ICUs). Diseases often have complex manifestations with heterogeneous subpopulations, or subtypes, that exhibit distinct clinical characteristics. Risk models that explicitly model subtypes have high predictive accuracy and facilitate subtype-specific personalization. Such models combine clustering and classification methods but do not effectively utilize the inferred subtypes in risk modeling. Their limitations include tendency to obtain degenerate clusters and cluster-specific data scarcity leading to insufficient training data for the corresponding classifier. In this article, we develop a new deep learning model for simultaneous clustering and classification, ExpertNet, with novel loss terms and network training strategies that address these limitations. The performance of ExpertNet is evaluated on the tasks of predicting risk of (i) sepsis and (ii) acute respiratory distress syndrome (ARDS), using two large electronic medical records datasets from ICUs. Our extensive experiments show that, in comparison to state-of-the-art baselines for combined clustering and classification, ExpertNet achieves superior accuracy in risk prediction for both ARDS and sepsis; and comparable clustering performance. Visual analysis of the clusters further demonstrates that the clusters obtained are clinically meaningful and a knowledge-distilled model shows significant differences in risk factors across the subtypes. By addressing technical challenges in training neural networks for simultaneous clustering and classification, ExpertNet lays the algorithmic foundation for the future development of subtype-aware risk models.

The prognostic value of rapid risk scores among patients with community-acquired pneumonia : A retrospective cohort study

BACKGROUND

Community-acquired pneumonia (CAP) is a frequent reason for emergency department (ED) presentations. Various risk scores have been validated in the management of CAP and are recommended for daily practice.

OBJECTIVE

The aim of the study was to evaluate the performance of the rapid risk scores (the rapid acute physiology score (RAPS), the rapid emergency medicine score (REMS), the Worthing physiological scoring system (WPS), CURB-65 and CRB-65) among patients with CAP.

METHODS

This retrospective cohort study was conducted in the ED of a tertiary hospital between 1 January 2019 and 31 December 2019. Patients aged ≥ 18 years and diagnosed with CAP were included. Patients who were transferred from another center or with missing records were excluded. Demographic information, vital signs, level of consciousness, laboratory results, and outcomes were recorded.

RESULTS

A total of 2057 patients were included in the final analysis. The 30-day mortality of the patients was 15.2% (n = 312). The WPS achieved the most successful results for all three outcomes, 30-day mortality, intensive care unit (ICU) admission and mechanical ventilation (MV) needs (area under the curve, AUC 0.810, 0.918, and 0.910, respectively; p < 0.001). In the prediction of mortality, RAPS, REMS, CURB-65, and CRB-65 had a moderate overall performance (AUC 0.648, 0.752, 0.778, and 0.739, respectively). In the prediction of ICU admission and MV needs, RAPS, REMS, CURB-65, and CRB-65 had moderate to good overall performance (AUC at ICU admission 0.793, 0.873, 0.829, and 0.810; AUC for MV needs 0.759, 0.892, 0.754, and 0.738, respectively). Advanced age, lower levels of mean arterial pressure and peripheral oxygen saturation, presence of active malignancy and cerebrovascular disease, and ICU admission were associated with mortality (p < 0.05).

CONCLUSION

The WPS outperformed other risk scores in patients with CAP and can be used safely. The CRB-65 can be used to discriminate critically ill patients with CAP due to its high specificity. The overall performances of the scores were satisfactory for all three outcomes.

Novel pneumonia score based on a machine learning model for predicting mortality in pneumonia patients on admission to the intensive care unit

BACKGROUND

Scores for predicting the long-term mortality of severe pneumonia are lacking. The purpose of this study is to use machine learning methods to develop new pneumonia scores to predict the 1-year mortality and hospital mortality of pneumonia patients on admission to the intensive care unit (ICU).

METHODS

The study population was screened from the MIMIC-IV and eICU databases. The main outcomes evaluated were 1-year mortality and hospital mortality in the MIMIC-IV database and hospital mortality in the eICU database. From the full data set, we separated patients diagnosed with community-acquired pneumonia (CAP) and ventilator-associated pneumonia (VAP) for subgroup analysis. We used common shallow machine learning algorithms, including logistic regression, decision tree, random forest, multilayer perceptron and XGBoost.

RESULTS

The full data set of the MIMIC-IV database contained 4697 patients, while that of the eICU database contained 13760 patients. We defined a new pneumonia score, the "Integrated CCI-APS", using a multivariate logistic regression model including six variables: metastatic solid tumor, Charlson Comorbidity Index, readmission, congestive heart failure, age, and Acute Physiology Score III. The area under the curve (AUC) and accuracy of the integrated CCI-APS were assessed in three data sets (full, CAP, and VAP) using both the test set derived from the MIMIC-IV database and the external validation set derived from the eICU database. The AUC value ranges in predicting 1-year and hospital mortality were 0.784-0.797 and 0.691-0.780, respectively, and the corresponding accuracy ranges were 0.723-0.725 and 0.641-0.718, respectively.

CONCLUSIONS

The main contribution of this study was a benchmark for using machine learning models to build pneumonia scores. Based on the idea of integrated learning, we propose a new integrated CCI-APS score for severe pneumonia. In the prediction of 1-year mortality and hospital mortality, our new pneumonia score outperformed the existing score.

Establishment of risk model for elderly CAP at different age stages: a single-center retrospective observational study

Community-acquired pneumonia (CAP) is one of the main reasons of mortality and morbidity in elderly population, causing substantial clinical and economic impacts. However, clinically available score systems have been shown to demonstrate poor prediction of mortality for patients aged over 65. Especially, no existing clinical model can predict morbidity and mortality for CAP patients among different age stages. Here, we aimed to understand the impact of age variable on the establishment of assessment model and explored prognostic factors and new biomarkers in predicting mortality. We retrospectively analyzed elderly patients with CAP in Minhang Hospital, Fudan University. We used univariate and multiple logistic regression analyses to study the prognostic factors of mortality in each age-based subgroup. The prediction accuracy of the prognostic factors was determined by the Receiver Operating Characteristic curves and the area under the curves. Combination models were established using several logistic regressions to save the predicted probabilities. Four factors with independently prognostic significance were shared among all the groups, namely Albumin, BUN, NLR and Pulse, using univariate analysis and multiple logistic regression analysis. Then we built a model with these 4 variables (as ABNP model) to predict the in-hospital mortality in all three groups. The AUC value of the ABNP model were 0.888 (95% CI 0.854-0.917, p < 0.000), 0.912 (95% CI 0.880-0.938, p < 0.000) and 0.872 (95% CI 0.833-0.905, p < 0.000) in group 1, 2 and 3, respectively. We established a predictive model for mortality based on an age variable -specific study of elderly patients with CAP, with higher AUC value than PSI, CURB-65 and qSOFA in predicting mortality in different age groups (66-75/ 76-85/ over 85 years).

Machine learning-based prediction of in-ICU mortality in pneumonia patients

Conventional severity-of-illness scoring systems have shown suboptimal performance for predicting in-intensive care unit (ICU) mortality in patients with severe pneumonia. This study aimed to develop and validate machine learning (ML) models for mortality prediction in patients with severe pneumonia. This retrospective study evaluated patients admitted to the ICU for severe pneumonia between January 2016 and December 2021. The predictive performance was analyzed by comparing the area under the receiver operating characteristic curve (AU-ROC) of ML models to that of conventional severity-of-illness scoring systems. Three ML models were evaluated: (1) logistic regression with L2 regularization, (2) gradient-boosted decision tree (LightGBM), and (3) multilayer perceptron (MLP). Among the 816 pneumonia patients included, 223 (27.3%) patients died. All ML models significantly outperformed the Simplified Acute Physiology Score II (AU-ROC: 0.650 [0.584-0.716] vs 0.820 [0.771-0.869] for logistic regression vs 0.827 [0.777-0.876] for LightGBM 0.838 [0.791-0.884] for MLP; P < 0.001). In the analysis for NRI, the LightGBM and MLP models showed superior reclassification compared with the logistic regression model in predicting in-ICU mortality in all length of stay in the ICU subgroups; all age subgroups; all subgroups with any APACHE II score, PaO₂/FiO₂ ratio < 200; all subgroups with or without history of respiratory disease; with or without history of CVA or dementia; treatment with mechanical ventilation, and use of inotropic agents. In conclusion, the ML models have excellent performance in predicting in-ICU mortality in patients with severe pneumonia. Moreover, this study highlights the potential advantages of selecting individual ML models for predicting in-ICU mortality in different subgroups.

Exploration of prognostic factors for prediction of mortality in elderly CAP population using a nomogram model

Background

The incidence and mortality rate of community-acquired pneumonia (CAP) in elderly patients were higher than the younger population. The assessment tools including CURB-65 and qSOFA have been applied in early detection of high-risk patients with CAP. However, several disadvantages exist to limit the efficiency of these tools for accurate assessment in elderly CAP. Therefore, we aimed to explore a more comprehensive tool to predict mortality in elderly CAP population by establishing a nomogram model.

Methods

We retrospectively analyzed elderly patients with CAP in Minhang Hospital, Fudan University. The least absolute shrinkage and selection operator (LASSO) logistic regression combined with multivariate analyses were used to select independent predictive factors and established nomogram models via R software. Calibration plots, decision curve analysis (DCA) and receiver operating characteristic curve (ROC) were generated to assess predictive performance.

Results

LASSO and multiple logistic regression analyses showed the age, pulse, NLR, albumin, BUN, and D-dimer were independent risk predictors. A nomogram model (NB-DAPA model) was established for predicting mortality of CAP in elderly patients. In both training and validation set, the area under the curve (AUC) of the NB-DAPA model showed superiority than CURB-65 and qSOFA. Meanwhile, DCA revealed that the predictive model had significant net benefits for most threshold probabilities.

Conclusion

Our established NB-DAPA nomogram model is a simple and accurate tool for predicting in-hospital mortality of CAP, adapted for patients aged 65 years and above. The predictive performance of the NB-DAPA model was better than PSI, CURB-65 and qSOFA.

Calibration and validation of the Pneumonia Shock Score in critically ill patients with SARS-CoV-2 infection, a multicenter prospective cohort study

Background

Prognostic tools developed to stratify critically ill patients in Intensive Care Units (ICUs), are critical to predict those with higher risk of mortality in the first hours of admission. This study aims to evaluate the performance of the pShock score in critically ill patients admitted to the ICU with SARS-CoV-2 infection.

Methods

Prospective observational analytical cohort study conducted between January 2020 and March 2021 in four general ICUs in Salvador, Brazil. Descriptive statistics were used to characterize the cohort and a logistic regression, followed by cross-validation, were performed to calibrate the score. A ROC curve analysis was used to assess accuracy of the models analyzed.

Results

Six hundred five adult ICU patients were included in the study. The median age was 63 (IQR: 49–74) years with a mortality rate of 33.2% (201 patients). The calibrated pShock-CoV score performed well in prediction of ICU mortality (AUC of 0.80 [95% Confidence Interval (CI): 0.77–0.83; p-value < 0.0001]).

Conclusions

The pShock-CoV score demonstrated robust discriminatory capacity and may assist in targeting scarce ICU resources during the COVID-19 pandemic to those critically ill patients most likely to benefit.

Development of a prognostic scoring system in patients with pneumonia requiring ventilator care for more than 4 days: a single-center observational study

BACKGROUND

The aim of the present study was to develop a prognostic model using demographic characteristics, comorbidities, and clinical variables measured on day 4 of mechanical ventilation (MV) for patients with prolonged acute mechanical ventilation (PAMV; MV for >96 hours).

METHODS

Data from 437 patients (70.9% male; median age, 68 years) were obtained over a period of 9 years. All patients were diagnosed with pneumonia. Binary logistic regression identified factors predicting mortality at 90 days after the start of MV. A PAMV prognosis score was calculating ß-coefficient values and assigning points to variables.

RESULTS

The overall 90-day mortality rate was 47.1%. Five factors (age ≥65 years, body mass index <18.5 kg/m2, hemato-oncologic diseases as comorbidities, requirement for vasopressors on day 4 of MV and requirement for neuromuscular blocking agents on day 4 of MV) were identified as prognostic indicators. Each factor was valued as +1 point, and used to develop a PAMV prognosis score. This score showed acceptable discrimination (area under the receiver operating characteristic curve of 0.695 for mortality, 95% confidence interval 0.650-0.738, p<0.001), and calibration (Hosmer-Lemeshow chi-square=6.331, with df 7 and p=0.502). The cutoff value for predicting mortality based on the maximum Youden index was ≤2 (sensitivity, 87.5%; specificity, 41.3%). For patients with PAMV scores ≤1, 2, 3 and ≥4, the 90-day mortality rates were 29.2%, 45.7%, 67.9%, and 90.9%, respectively (P<0.001).

CONCLUSIONS

Our study developed a PAMV prognosis score for predicting 90-day mortality. Further research is needed to validate the utility of this score.