Editorial

New Sepsis Criteria: A Change We Should Not Make

Metabolic syndrome (MetS) is a widely used index for finding people at risk for chronic diseases, including cardiovascular disease and diabetes. Early detection of MetS is especially important in prevention programs. Relying on previous studies that suggest machine learning methods as a valuable approach for diagnosing MetS, this study aimed to develop MetS prediction models based on support vector machine (SVM) algorithms, applying non-invasive and low-cost (NI&LC), and also dietary parameters.

This population-based research was conducted on a large dataset of 4596 participants within the framework of the Shahedieh cohort study. An Extremely Randomized Trees Classifier was used to select the most effective features among NI&LC and dietary data. The prediction models were developed based on SVM algorithms, and their performance was assessed by accuracy, sensitivity, specificity, positive prediction value, negative prediction value, f1-score, and receiver operating characteristic curve. MetS was diagnosed in 14% of men and 22% of women. Among NI&LC features, waist circumference, body mass index, waist-to-height ratio, waist-to-hip ratio, systolic blood pressure, and diastolic blood pressure were the most predictive variables. By using NI&LC features, models with 78.4% and 63.5% accuracy and 81.2% and 75.3% sensitivity were yielded for men and women, respectively. By incorporating NI&LC and dietary features, the accuracy of the model in women improved by 3.7%.

SVM algorithms had promising potential for early detection of MetS relying on NI&LC parameters. These models can be used in prevention programs, clinical practice, and personal applications.

The ‘quick Sepsis Related Organ Failure Assessment’ (qSOFA), ‘Systemic Inflammatory Response Syndrome’ (SIRS), and ‘National Early Warning Score’ 2 (NEWS2) scores are yet to be comparatively validated in ward-based cardiac surgical patients despite widespread routine use in clinical practice. We sought to assess the predictive validity of NEWS, SIRS, and qSOFA in identifying postoperative, ward-level cardiac surgical patients at risk of poor short-term mortality.

All adult patients who underwent cardiac surgery at a single tertiary center between November 2014 and October 2017 were identified. Data for bedside observations, hematological results, and microbiology requests were obtained from electronic health records. Survival data were acquired from a national registry. The primary outcome was the discriminatory ability, measured by the area under the receiver operating characteristic (AUROC), of each score for in-hospital mortality.

One thousand five hundred forty three (male n = 1101, 71%) patients were included. Overall in-hospital mortality was 2.4%. There was no significant difference in discriminatory ability of NEWS (AUROC 0.5060), SIRS (AUROC 0.4874), and qSOFA (AUROC 0.5139) for in-hospital mortality (P = 0.881). Sensitivity for this outcome was ubiquitously low (13.51-40.54%).

Current illness-severity scores show a low discriminatory ability for in-hospital mortality in ward-based cardiac surgical patients. Caution should be used in the application of these prognostic screening tools for early detection of poor outcomes in this population.

ICU mortality prediction is critical for board-certified physician to make efficient decision for seriously sick patients who stay in the ICU. Various techniques have been developed to overcome this problem based on the medical records. However, some physiological tests such as laboratory test, vital test are time consuming and need to handle by developing various machine learning (ML) models. In this paper, twelve different ML based classifiers are developed and from these classifiers, best three classifiers are selected and an ensemble learning classifier is proposed to predict the ICU patient's mortality. Three standard datasets Physionet challenge 2012, WiDS Datathon, 2020 and MIMIC-III are used for simulation and analysis as they are challenging. These datasets contain patient's data collected during the first 48 h, 24 h and 48 h respectively of patient's admission to ICU. The datasets have many features such as high dimension, huge missing values and highly imbalance which degrades the performance of the models. To handle these problems, k-NN imputation or mean imputation method is applied to fill the missing values and five variants of synthetic minority oversampling technique (SMOTE) such as BD-SMOTE, SVM-SMOTE, ADASYN, ENN-SMOTE and TL-SMOTE, are applied for resampling the data. To circumvent the curse of dimensionality problem, transform domain features are extracted using discrete wavelet transform (DWT). Then twelve different ML based classifiers are trained using wavelet coefficients as inputs and 5-fold cross validation over 10 independent experiments. Out of these twelve, (ENN-SMOTE + LGB) is performing best followed by (ENN-SMOTE + CatBoost) and (BD-SMOTE + XGB) in physionet challenge 2012 dataset. ENN-SMOTE + CatBoost is performing best followed by ENN-SMOTE + XGB and BD-SMOTE + RF models in WiDS Datathon, 2020. Similarly for MIMIC-III ENN-SMOTE + CatBoost is best performed model followed by ENN-SMOTE + XGB and ENN-SMOTE + RF. Finally, ensemble classifiers are proposed using these best performed models which resulted AUC of 92.66%, 94.44% and 98.06% for Physionet challenge 2012, WiDS Datathon, 2020 and MIMIC-III datasets respectively during testing phase. Further, the results of the proposed models are compared with the other reported models and seems to be outperformed all in terms of AUC score.

El diagnóstico y el tratamiento precoces de la sepsis sigue siendo un reto en los servicios de urgencias hospitalarios.

El objetivo es evaluar el rendimiento de la escala quick Sequential related Organ Failure Assessment (qSOFA) para la activación de un Código Sepsis en urgencias, respecto la aplicación de los criterios previos basados en los parámetros Systemic Inflammatory Response Syndrome (SIRS).

Se incluyeron los pacientes ≥ 18 años que consultaron en un servicio de urgencias de un hospital de tercer nivel durante dos períodos de tiempo (mayo-octubre de 2015) mediante criterios SIRS (grupo SIRS), y en los mismos meses de 2017 mediante criterios qSOFA (grupo qSOFA).

Se analizaron 125 episodios: 70 correspondieron al grupo SIRS y 55 al grupo qSOFA. Los grupos fueron comparables respecto a antecedentes, con mayor presencia de enfermedad neoplásica en el grupo qSOFA. No se evidenciaron diferencias en la forma de presentación, la actuación realizada en urgencias, los ingresos en unidades de críticos, el uso de vasopresores o la estancia media hospitalaria entre los dos grupos. Sí se objetivó una mayor tasa de mortalidad a 30 días en el grupo qSOFA (21,81% vs 4,3%; p = 0,003). En el estudio multivariado destaca la enfermedad neoplásica como factor independiente de mal pronóstico (OR: 5,36; p = 0,01), y la activación mediante criterios SIRS como factor de buen pronóstico (OR: 0,15; p = 0,021).

La activación del Código Sepsis mediante criterios qSOFA comporta una disminución en la detección, con mayor mortalidad asociada. Es necesario buscar otras herramientas complementarias para la detección precoz de la sepsis en urgencias.

The early diagnosis and treatment of sepsis remains a challenge in hospital emergency departments.

The objective was to evaluate the performance of the qSOFA (quick Sequential related Organ Failure Assessment) for activating a Sepsis Code in the emergency department, compared to applying previous criteria based on SIRS (Systemic Inflammatory Response Syndrome).

Patients ≥ 18 years were included who consulted an emergency department of a tertiary hospital over two periods of time (May-October 2015), using SIRS criteria (SIRS Group) and over the same months in 2017 using qSOFA criteria (qSOFA Group).

One hundred and five episodes were analysed: 70 corresponded to the SIRS Group and 55 to the qSOFA Group. The groups were comparable in terms of history, with a greater presence of neoplastic disease in the qSOFA Group. No differences were observed in the form of presentation, the action taken in the emergency department, admissions to critical care units, use of vasopressors or mean hospital stay between the two groups. A higher 30-day mortality rate was observed in the qSOFA group (21.81% vs. 4.3%; P = .003). In the multivariate study, neoplastic disease was noteworthy as an independent factor of poor prognosis (OR 5.36; P = .01), and activation using the SIRS criteria as a factor of good prognosis (OR .15; P = .021).

The activation of the Sepsis Code using qSOFA criteria results in a reduction in detection, with greater associated mortality. Other complementary tools need to be sought for the early detection of sepsis in the emergency department.

Limited data are available on the discriminatory capacity of quick sequential [sepsis-related] organ failure assessment (qSOFA) versus IDSA/ATS minor criteria for predicting mortality in patients with community-acquired pneumonia (CAP).

An observational prospective cohort study of 2116 patients with CAP was performed. Construct validity was determined using Cronbach α. Discrimination was assessed using the area under the receiver operating characteristic curve (AUROC) and net reclassification improvement (NRI).

Overall in-hospital mortality was 6.43%. Mortality was 25.96% for patients with a qSOFA score of 2 or higher versus 3.05% for those with a qSOFA score less than 2 (odds ratio for mortality 6.57, P < 0.0001), and 13.85% for patients with at least 3 minor criteria versus 2.03% for those with 2 or fewer minor criteria (odds ratio for mortality 2.27, P < 0.0001). qSOFA had a higher correlation with mortality than minor criteria, as well as higher internal consistency (Cronbach alpha 0.43 versus 0.14) and diagnostic values of individual elements (larger AUROCs and higher Youden's indices). qSOFA ≥2 was less sensitive but more specific for predicting mortality than ≥3 minor criteria (qSOFA sensitivity 59.6%, specificity 88.3% and positive likelihood ratio 5.11 versus ≥3 minor criteria sensitivity 80.1%, specificity 65.8% and positive likelihood ratio 2.34). The predictive validity of qSOFA was good for mortality (AUROC = 0.868), was statistically greater than minor criteria, was equal to pneumonia severity index, and was inferior compared with CURB-65 (AUROC, 0.824, 0.902, 0.919; NRI, 0.088, −0.068, −0.103; respectively).

The qSOFA predicted mortality in CAP better than IDSA/ATS minor criteria and worse than CURB-65 with robust elements and higher convergence. qSOFA as a bedside prompt might be positioned as a proxy for minor criteria and increase the recognition and thus merit more appropriate management of CAP patients likely to fare poorly, which might have implications for more accurate clinical triage decisions.

To investigate the prognostic effects of delayed administration of appropriate antimicrobial therapy (AAT) in older persons experiencing bacteremia with and without initial sepsis syndrome, respectively.

A 4-year multicenter cohort study.

Older people (≥65 years of age) with community-onset bacteremia in the emergency department (ED) of 3 participating hospitals.

Clinical data were retrospectively collected and causative microorganisms were prospectively collected for susceptibilities to determine the period of delayed AAT for each bacteremia episode. Sepsis was defined based on the Sepsis-3 criteria. A multivariable regression model was used to investigate the prognostic effects of delayed AAT, after adjusting independent determinants of 30-day mortality.

Of the total 2357 patients, their median (interquartile range) age was 78 (72–84) years and septic patients accounted for 48.4% (1140 patients) of the overall patients. Compared with nonseptic patients, septic individuals exhibited the shorter period of delayed AAT (median, 2.0 vs 2.5 hours; P < .001), longer hospitalization (median, 11 vs 9 days; P < .001), and higher crude mortality rates at 15 (28.9% vs 2.1%; P < .001) and 30 days (34.6% vs 4.0%; P < .001). In multivariable regression analyses, each hour of delayed AAT resulted in average increases in the 30-day crude mortality rates of 0.38% [adjusted odds ratio (AOR) 1.0038; P < .001), 0.42% (AOR 1.0042; P < .001), and 0.31% (AOR 1.0031; P = .04) among overall, septic, and nonseptic patients, respectively.

For older persons with community-onset bacteremia, irrespective of whether or not patients experiencing initial sepsis presentations, the prognostic impacts of delayed AAT have been evidenced. Notably, because of the longer period of delayed AAT in patients without fulfilling the Sepsis-3, adopting a stricter sepsis definition and/or early bacteremia predictor to avoid delayed AAT and unfavorable prognoses in patients with bacteremia is necessary.