LAUSR

Proposed Definition of Necrotizing Enterocolitis May Be of Limited Value To the Editor Battersby et al1 report on the development and diagnostic performance of a new gestational age–specific case definition for necrotizing enterocolitis (NEC) and claim to compare its diagnostic performance with that of the Vermont Oxford Network (VON) definition. Unfortunately, they did not use the complete VON definition, leading to incorrect estimates of its performance. The VON definition provides 3 options for identifying infants with NEC: (1) inspection during surgery, (2) inspection at post mortem, or (3) using standardized clinical and radiographic criteria (VON Manual of Operations, http://www .vtoxford.org). The authors based their assessment only on the third component of the VON definition, ignoring that cases could also be identified at surgery or post mortem, thus substantially underestimating the sensitivity, specificity, positive predictive value, and area under the receiver operating curve and overestimating the misclassification rate of the VON definition. However, even had the authors used the correct VON definition, we have concerns about their methods that call into question the estimates of diagnostic test performance from their study population. First, large proportions of infants are excluded: 67.5% of infants admitted to participating units (8019 of 11 885) and 41.4% of infants with abdominal radiographs (2728 of 6594). These high proportions raise serious questions about the generalizability of the study findings. Among the exclusions were infants with probable NEC, missing diagnoses, and inconsistencies during cross validation. These exclusions may have resulted in cases with specific clinical or radiographic findings required for the diagnosis of NEC being preferentially omitted, and the reported estimates of sensitivity and specificity may thus not apply beyond the study population. Furthermore, in the study population, the prevalence of NEC was 23%, which is nearly 5 times higher than in other populations.2 Because positive predictive value depends on disease prevalence, the reported positive predictive value will not apply to populations with lower NEC rates. Second, the gold standard used for the diagnosis of NEC is poorly defined. For the 3435 infants who did not have surgery, NEC was determined when the clinician considered the diagnosis as “unequivocally yes or no.” The criteria for an “unequivocal” diagnosis are not stated, and data on interrater reliability are not provided. Without a reliable gold standard, the estimates of diagnostic performance presented are of limited value.