Comments on “Development and Validation of a Temporal Progression-based Risk Assessment Tool for Respiratory Syncytial Virus Infection in Infants”

To the Editors Kawamoto et al1 describe the development and temporal validation of a temporal progression-based risk assessment tool for respiratory syncytial virus (RSV) infection in infants, culminating in a practical symptom-based score for clinical use. The focus on temporal symptom patterns and synchronicity is innovative and aligns well with current guidance on multivariable prediction models.2 We commend the authors for separating development and validation cohorts and for reporting both discrimination and calibration. Several methodologic issues, however, are important for interpretation. First, the derivation and validation cohorts comprise only infants who underwent RSV rapid antigen testing at a tertiary center and who had at least 1 predefined respiratory symptom.1 This 2-stage selection means the model estimates the probability of antigen-positive RSV among already selected, symptomatic, test-eligible infants, rather than among unselected infants in primary care or community settings. As emphasized by transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD), clarity about the target population is crucial for judging transportability and intended use.2 Second, the reference standard is a rapid antigen test, which has substantially lower sensitivity than reverse transcription polymerase chain reaction (RT-PCR). Systematic misclassification of true RSV cases as “negative” is likely, especially in milder or later-presenting illness. This can attenuate regression coefficients and distort calibration. TRIPOD highlights the need to define and justify the outcome and its measurement.2 Future work could consider more sensitive reference standards or dual testing in a subset to quantify misclassification and its impact. Third, while the number of RSV events in the development cohort is reasonably large, the model includes multiple temporal and clinical predictors, and the validation cohort contains only 36 RSV-positive infants. Recent work on minimum sample size for prediction models suggests that planning based on events per parameter and anticipated model complexity is needed to limit overfitting and imprecision in performance estimates.3 The small number of validation events, especially in subgroups, implies wide uncertainty around the reported area under the curve and risk-group probabilities. Moreover, handling of missing data is not clearly described; complete-case analysis is common but can induce bias and loss of precision, which both TRIPOD and methodological reviews identify as a frequent weakness in prediction studies.2,4 Fourth, the authors describe their temporal split and selected subgroup analyses as validation. This is an important step, but it remains a form of internal temporal validation within a single institution and case-mix. External validation, ideally in different centers and healthcare settings, is essential before recommending a model for broad clinical deployment.4 Finally, when converting the logistic model to a points-based score, the article does not fully describe how the score is linked back to absolute risk, or whether any recalibration was performed. Transparent presentation of the full model and its simplified form is needed to allow independent validation and proper assessment of calibration.2,4 The proposed RSV symptom index is a promising and clinically intuitive tool, but should presently be regarded as an internally validated model whose performance and calibration are uncertain outside the development setting. Prospective, multicenter validation with stronger reference standards, explicit missing-data handling, and adherence to contemporary reporting and sample-size guidance would further strengthen its potential clinical impact.