Predicting Mortality in Tanzanian Children with Sepsis Using Point-of-Care Biomarkers

Résumé fourni par la source

Overview: The primary project is a prospective, observational cohort study of children aged 28 days to 14 years with sepsis at Muhimbili National Hospital, the national referral hospital, in Dar es Salaam, Tanzania (July 2022–November 2024) to determine the risk factors, morbidity, and mortality related to pediatric sepsis in this setting. Children aged 28 days to 14 years who presented with sepsis were included in our study. Sepsis was defined as fever (>38°C or history of fever) with ≥1 World Health Organization danger sign: breathing difficulty, altered mental status, impaired perfusion, or risk of dehydration. Children were excluded for cardiac arrest on arrival, weight <4 kg, acute trauma, non-infectious fever (e.g., rheumatologic disease), isolated seizures in known epilepsy, malignancy, or congenital heart disease. We screened all pediatric patients presenting to the Emergency Department during the study period using consecutive sampling. Parents or guardians of the potential participants were approached by trained research personnel for consent prior to study enrollment. The primary outcome of interest was in-hospital mortality, defined as death occurring at any point during hospitalization. The primary predictors were point-of-care biomarker concentrations (procalcitonin, C-reactive protein, ferritin, lactate). Clinical severity was measured using the Lambaréné Organ Dysfunction Score, which includes weakness, breathing difficulty, and altered mental status. Nutritional status was measured by mid-upper arm circumference or weight-for-height z score, in accordance with World Health Organization standards. The submitted repository dataset includes all enrolled participants with de-identified clinical, biomarker, treatment, and outcome data. Objectives: We aimed to develop a context-specific, reproducible model using point-of-care biomarkers and clinical signs that could predict risk of mortality in children with sepsis in resource-limited settings. Data Collection Methods: Once enrolled in the study, participant clinical data, outcomes, and interventions (e.g., antibiotics, mechanical ventilation, blood product transfusions) were extracted from medical records by study staff and entered into REDCap (version 7.2.2), a secure, electronic database. Venous blood was collected in EDTA tubes from all participants within four hours of arrival using sterile technique. Point-of-care tests included rapid diagnostic tests (RDTs) for malaria (SD Bioline-Pf, Abbott Laboratories, Abbott Park, IL, USA) and HIV (Alere Determine HIV-1/2, Abbott Laboratories, Abbott Park, IL, USA), as well as hemoglobin (iSTAT, Abbott Laboratories, Abbott Park, IL, USA) and glucose (GlucoPlus, GlucoPlus Inc., Montreal, Canada). Patients with positive malaria RDT were considered positive. Patients with a positive HIV RDT received subsequent confirmatory testing. Prior to this study, lactate, hemoglobin, and blood glucose were available as point-of-care tests at MNH. Lactate was measured from a whole blood venous sample using an iSTAT system (Abbott Laboratories, Abbott Park, IL, USA). Whole blood was then separated, and plasma was used to measure procalcitonin, C-reactive protein, and ferritin by fluorescence immunoassay using the i-Chroma II system, provided by our study to facilitate real-time measurement (BodiTech Med Inc., Chuncheon-si, Gangwon-do, Republic of Korea). Blood cultures were obtained on all patients using BD BACTEC™ Peds Plus/F vials and processed in a research lab setting using a BD BACTEC FX40 system (Becton Dickinson, Franklin Lakes, New Jersey, USA). All testing followed local protocols and manufacturer instructions. Data Processing Methods: Key variables included in the submitted dataset: Key domains include demographics, anthropometrics and nutritional status, presenting symptoms and World Health Organization danger signs, vital signs, HIV and malaria status, point-of-care laboratory measurements (lactate, ferritin, C-reactive protein, procalcitonin, hemoglobin, glucose), blood culture results, therapeutic interventions (including antibiotics, transfusion, and mechanical ventilation), and hospital outcomes, including mortality. We used descriptive statistics to summarize patient characteristics and biomarker levels by survival status and reported continuous variables as medians with interquartile ranges, comparing them using the Wilcoxon rank-sum test. Categorical variables were expressed as frequencies and proportions and compared using Fisher’s exact test. Two-tailed p values <0.05 indicated statistical significance. Biomarkers were analyzed as continuous variables and tested individually and in combination for their ability to predict mortality. Candidate variables were selected a priori based on prior associations with pediatric mortality in resource-limited settings. These included age, sex, malaria status, HIV status, hemoglobin, blood glucose, vaccination status, hypoxemia, nutritional status, weakness, breathing difficulty, and altered mental status. To optimize a multivariable model using both biomarkers and clinical characteristics, a least absolute shrinkage and selection operator (LASSO) logistic regression was used to select variables to construct a predictive model for mortality. The LASSO was constructed using 10-fold cross-validation to optimize the lambda value (penalty parameter), ensuring the most informative variables were selected without overfitting the model. Individual and multivariable model performance was evaluated using area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (Cis). To optimize model parsimony, variables contributing incrementally to discrimination (<0.005 AUC) were sequentially removed. Odds ratios, 95% CIs, and standardized beta coefficients were reported. Model classification performance was evaluated at varying probability thresholds using sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy. All data cleaning, variable coding, and statistical analyses were conducted using Stata/SE 18.0 for Mac (Apple Silicon; StataCorp LLC, College Station, TX; Revision 17 Feb 2026) or SAS 9.4 (SAS Institute, Cary, NC). Stata scripts require one user-contributed package: table1 (available from the SSC archive; install via ssc install table1). All other commands (stset, stcox, sts graph, logit, roctab, roccomp, lincom, estat phtest) are included in the base Stata/SE 18 installation and require no additional packages. Ethics Declaration: Guardians provided written informed consent; children assented when developmentally and cognitively able; and participants received routine clinical care as indicated, including intravenous fluids, empiric antibiotics, and antimalarial therapy. Children living with Human Immunodeficiency Virus (HIV) continued their antiretroviral medications. The study was approved by the institutional review boards at Muhimbili University of Health and Allied Sciences (DA.282/298/01.C/374), the Tanzanian National Institute for Medical Research (NIMR) (NIMR/HQ/R.8a/Vol. IX/3576), and the University of California, San Francisco (19-27627). Permission to publish was granted by NIMR. The study followed STROBE reporting guidelines. Funding Sources: Research effort to create this publication was supported by the National Institute of Allergy and Infectious Diseases (K23AI144029 [TBK]) of the National Institutes of Health (NIH); the University of California Global Health Institute GloCal/Fogarty Fellowship [D43TW009343] and the University of California, San Francisco T32 Fellowship [5T32HD049303] awarded to Abigail Sorensen.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.