Work

1272 Diagnosing early-onset neonatal sepsis in low-income and middle-income countries: development of a multivariable prediction model from routine clinical data

Year: 2021

Type: conference-abstract

Abstract: Background Neonatal sepsis causes 15% of global neonatal deaths. Non-specific clinical features and unavailability of laboratory tests in many low-income and middle-income countries make it difficult to diagnose. Clinical prediction models could support timely, accurate diagnosis of neonatal sepsis and rationalise antibiotic usage. The NeoTree app is an Android-based quality improvement platform embedded into clinical practice at two neonatal units in Malawi and Zimbabwe. It combines evidence-based clinical decision support, education on neonatal care and immediate digital data capture. Objectives We aimed to develop a clinical prediction model to diagnose neonatal sepsis with routine clinical data, without the need for laboratory tests. Methods We analysed secondary data from an ongoing quality improvement project at the neonatal unit of Sally Mugabe Central Hospital, Zimbabwe, collected via the NeoTree app. We included neonates admitted from 1 February 2019 to 31 March 2020 aged <72 hours, born at ≥32+0 weeks with birth weight ≥1500 grams. We developed a clinical prediction model with multivariable logistic regression for the outcome of early-onset neonatal sepsis (EOS, <72 hours) as defined by the treating consultant neonatologist. Candidate predictors were identified from expert opinion and literature review. Missing data were imputed using multivariate imputation with chained equations. We first fitted a ‘full’ main effects model containing all candidate predictors. We then examined several variations on the full model and selected the model favoured by the Bayesian information criterion. Performance was evaluated in the derivation cohort with the area under the receiver operating characteristic (ROC) curve (AUC) and sensitivity, specificity, predictive values and likelihood ratios at various thresholds of predicted probability. Results We included 2627 neonates with mean gestational age 38.0 ± 2.5 weeks and mean birth weight 2889 ± 703 grams. EOS was diagnosed in 297 neonates (11.3%). The final model included eight predictors: maternal fever, foul-smelling liquor, prolonged rupture of membranes, neonatal temperature, respiratory rate, activity, chest retractions and grunting. ROC analysis gave AUC = 0.735 (95% confidence interval 0.699–0.771). For a sensitivity of 95 (92–97)%, corresponding specificity = 15 (13–16)%, positive predictive value = 12 (11–14)%, negative predictive value = 96 (93–98)%, positive likelihood ratio = 1.12, and negative likelihood ratio = 0.32. Conclusions Our clinical prediction model for EOS achieved high sensitivity but with only modest specificity in the derivation cohort. This suggests our model may be best suited to ruling out a diagnosis of EOS, which may support healthcare workers’ decisions to withhold antibiotics and prevent unnecessary antibiotic exposure in non-septic neonates. Future work will validate and refine this model before incorporating it into the NeoTree app for clinical use.

Source: Abstracts

DOI: 10.1136/archdischild-2021-rcpch.523