Comments on Kothalawala et al.

Abstract

To the Editor, Systematic reviews of the growing literature on preschool asthma prediction can help determine whether the available evidence supports the use of existing prognostic models to inform childhood asthma prevention and clinical management. The most salient prognostic review questions should address not only the accuracy of predictions derived from a prediction model but more importantly how the results affect patient management and outcomes.

Decision tree‐based rules outperform risk scores for childhood asthma prognosis

“…Risk and prognostic factors are assumed to have time-invariant effects on asthma risk over time ignoring the intricate interplay of biologic pathways that result in the changing risk of asthma incidence and recurrence. 2 Yet, for example, previous longitudinal studies have shown that not all children who suffer from early persistent wheezing in childhood go on to develop asthma at 2, 7, and 15 years of age. 6 In this study, we use the data from the Canadian Asthma Primary Prevention Study (CAPPS) to develop and compare the prognostic performance of conditional inference tree-based decision rules with a revised PARS for the prediction of childhood asthma at 7 years.…”

“…These regression models are limited to a small number of variables (for model parsimony) to aid asthma prognosis. Although such models are useful and often necessary in association analyses, this is not necessarily the case in prediction analyses, where the focus is on the outcome instead of the predictors 2 . Moreover, the predictor variables in both the API and PARS are conceptualized as single time point assessments that do not mimic the way a healthcare provider may approach prognosis of a patient with a series of questions guiding the clinician, with subsequent questions based on the answer to the prior.…”

“…There are no widely accepted prognostic tools for childhood asthma 1,2 ; this is in part due to the multifactorial nature of mechanisms and risk factors that contribute to asthma development, which also may vary over time 3 . Consequently, clinicians often rely on risk scores such as the Asthma Predictive Index (sensitivity = 28%; specificity = 96%) 4 and recently the Pediatric Asthma Risk Score (PARS: sensitivity = 68%, specificity = 77%), 5 which have modest prognostic accuracy (area under the curve ≤80%).…”

“…Moreover, the predictor variables in both the API and PARS are conceptualized as single time point assessments that do not mimic the way a healthcare provider may approach prognosis of a patient with a series of questions guiding the clinician, with subsequent questions based on the answer to the prior. Risk and prognostic factors are assumed to have time‐invariant effects on asthma risk over time ignoring the intricate interplay of biologic pathways that result in the changing risk of asthma incidence and recurrence 2 . Yet, for example, previous longitudinal studies have shown that not all children who suffer from early persistent wheezing in childhood go on to develop asthma at 2, 7, and 15 years of age 6 …”

Reply to Owora et al.

“…To the Editor, We would like to thank Owora et al 1 for their interest in our systematic review of childhood asthma prediction models and welcome their subsequent analyses and recommendations for future research in this field.…”