437 - Development and Validation of ‘Clinical Prediction Model’ for Early onset sepsis in Preterm Neonates ≤ 34 weeks of gestation.

Publication Number: 2425.437

Sushma Nangia, Lady Hardinge Medical College & Kalawati Saran Children's Hospital, New Delhi, Delhi, India; Veenashree Bhat, Lady Hardinge Medical College, New Delhi, Delhi, India; TAPAS BANDYOPADHYAY, VMMC and Safdarjung Hospital, New Delhi, Delhi, India; Pratima Anand, Lady hardinge medical college new delhi, New Delhi, Delhi, India

Background: Neonatal sepsis poses a diagnostic challenge due to nonspecific clinical signs with resultant inadvertent use of antibiotics using risk factor-based approach. Clinical sepsis prediction models bridge the diagnostic gap and help the clinician in making sound therapeutic decision.

Objective: To develop and validate 'Clinical Prediction Model' for antibiotic use in early onset sepsis in preterm neonates born with gestation age ≤ 34 weeks.

Design/Methods: In this prospective cohort study, all neonates born ≤ 34 weeks were assessed for all putative perinatal risk factors and clinical signs during first 72 hours of life. Multivariable logistic regression with backward elimination was applied after univariate logistic regression and then potential predictors were identified. Final variables of the model were chosen following multicollinearity check and variables were assigned weights. The model performance was evaluated using area under the receiver operating characteristic (AUROC) and calibration curves.

Results: Among the 1190 neonates with mean gestation of 32 weeks and mean birth weight of 1600 grams, 27 perinatal risk factors and 22 clinical signs were analyzed. Univariate analysis identified 7 risk factors and 7 clinical signs as predictors. Nine significant parameters on multivariable logistic regression were included in the final model. A composite score of ≥ 4 out of 17 indicated sepsis with 91% sensitivity and 92% specificity. The AUROC was 0.97 for development cohort and 0.96 for validation. Additionally, the straight calibration curve further indicated that the predicted probabilities accurately reflect the true likelihood of the event.

Conclusion(s): The study has developed the first of its kind, clinical based prediction model for preterm neonates, that has the potential to aid in antibiotic decision and support antibiotic stewardship.

Perinatal risk factors and clinical signs pooled in univariate analysis
Table illustrates details of Perinatal risk factors and Clinical signs pooled in univariate analysis for identification of potential predictors

Multivariable logistic regression of individual predictors for sepsis
Table shows multivariate analysis of potential predictors which have been selected after multivariate logistic regression with regression coefficient with weighted points of each predictors.

Final model for prediction of Early onset sepsis in neonates with AUROC
Figure shows final prediction model for early onset sepsis with total nine predictors which included 3 predictors and 6 clinical signs ,with total score of 17 ,and a score of 4 or more is diagnostic of sepsis with AUROC of 0.97.