Theoretical re-analysis of two previously published datasets

Abstract

We read with interest the theoretical impact analysis of a screening tool for pediatric abusive head trauma,1Hymel K.P. Herman B.E. Narang S.K. Graf J.M. Frazier T.N. Stoiko M. et al.Potential impact of a validated screening tool for pediatric abusive head trauma.J Pediatr. 2015; 167: 1375-1381.e1Abstract Full Text Full Text PDF Scopus (17) Google Scholar the third in a series of articles regarding the tool.2Hymel K.P. Willson D.F. Boos S.C. Pullin D.A. Homa K. Lorenz D.J. et al.Derivation of a clinical prediction rule for pediatric abusive head trauma.Pediatr Crit Care Med. 2013; 14: 210-220Crossref PubMed Scopus (56) Google Scholar, 3Hymel K.P. Armijo-Garcia V. Foster R. Frazier T.N. Stoiko M. Christie L.M. et al.Validation of a clinical prediction rule for pediatric abusive head trauma.Pediatrics. 2014; 134: e1537-e1544Crossref PubMed Scopus (65) Google Scholar We are concerned about the inherent circular reasoning associated with the application of the authors' a priori definitional criteria for abusive head trauma as a reference standard against which to compare the performance of their tool. Items in the screening tool (eg, bruising of the ear, neck, or torso) appear in the a priori defining criteria (eg, skin bruising, abrasions, or lacerations in two or more distinct locations other than the knees, shins, or elbows), thus potentially introducing incorporation bias.4Worster A. Carpenter C. Incorporation bias in studies of diagnostic tests: how to avoid being biased about bias.CJEM. 2008; 10: 174-175PubMed Google Scholar No references are provided for the extensive list of extra-cranial injuries within their a priori definitional criteria “considered moderately or highly suspicious for abuse,” for example “dry contact burns”and “intra-abdominal injuries.” The authors estimate that 12/58 higher risk children not evaluated for abuse were abused. However, it is a logical fallacy to assume that these 12 children must therefore have had positive findings on their skeletal survey and/or ophthalmology exams. Probability estimates were calculated by implicitly fitting a saturated model, with one probability for each combination of features. These could have alternatively been estimated by fitting a simple logistic regression model. This would have improved estimation of the variance, by permitting information borrowing across categories. Potential Impact of a Validated Screening Tool for Pediatric Abusive Head TraumaThe Journal of PediatricsVol. 167Issue 6PreviewTo conduct a retrospective, theoretical comparison of actual pediatric intensive care unit (PICU) screening for abusive head trauma (AHT) vs AHT screening guided by a previously validated 4-variable clinical prediction rule (CPR) in datasets used by the Pediatric Brain Injury Research Network to derive and validate the CPR. Full-Text PDF ReplyThe Journal of PediatricsVol. 171PreviewWe thank the correspondents for their interest in, and close scrutiny of, our work. Maguire et al indicate concerns for circularity. The sole variable appearing both in the rule and in the a priori criteria is bruising. Of the 73 patients in our derivation/validation datasets (N = 500) who revealed “skin bruising, abrasions, or lacerations in 2 or more distinct locations other than the knees, shins, or elbows,” and “bruising of the ear(s), neck or torso,” 61 met other historical criteria for abusive head trauma (AHT). Full-Text PDF