External validation of the discriminative validity of the ReSVinet score & development of simplified ReSVinet scores in secondary care

Abstract

Abstract Background There is no consensus on how to best quantify disease severity in infants with respiratory syncytial virus (RSV) and/or bronchiolitis; this lack of a sufficiently validated score complicates the provision of clinical care and, the evaluation of trials of therapeutics and vaccines. The ReSVinet score appears to be one of the most promising; however it is too time-consuming to be incorporated into routine clinical care. We aimed to develop and externally validate simplified versions of this score. Methods Data were used from a multinational (Netherlands, Spain & United Kingdom) multicentre case-control observational study of infants with RSV to develop simplified versions of the ReSVinet by conducting a grid search to determine the best combination of equally weighted parameters to maximise for the discriminative ability of the scores across a range of outcomes (hospitalisation, intensive care unit admission, ventilation requirement). Subsequently discriminative validity of the score for a range of secondary care outcomes was externally validated by conducting a secondary analysis of data collected in infants with respiratory infection from tertiary hospitals in Rwanda and Colombia. Results Three candidate simplified scores were identified using the development dataset; they were excellent (area under the receiver-operator characteristic curve [AUROC] >0.9) in the development dataset at discriminating for a range of outcomes, and their performance was not statistically significantly different to the original ReSVinet score despite having fewer parameters. In the external validation datasets, the simplified scores were moderate-excellent (AUROC 0.7-1) across a range of outcomes. In all outcomes, except for in a single dataset at predicting admission to the high dependency unit, they performed at least as well as the original ReSVinet score. Conclusions Three promising candidate simplified scores were developed; however further external validation work in larger datasets, ideally from resource-limited settings needs to be conducted before any recommendation regarding their use.