Kim Annink

87 Cerebral ultrasound scoring system for neonates with HIE an MRI. We speculate that these would have shown severe MRI abnormalities, thereby improving the overall association between CUS and MRI. The performance of the model was good in both cohorts for both the cut-off value of ≥3 which can potentially be used if the number of false negatives should be as low as possible (i.e. for decisions about additional future neuroprotective strategies) and for higher cut-off values that can be preferable if the number of false positives should be as low as possible (i.e. for considering redirection of care in combination with neurophysiology and clinical features if MRI is not possible). The interrater variability was moderate between all observers implying that the interrater agreement among different hospitals and observers would also be moderate. In a prospective cohort, in which observers can be trained and the quality of CUS can be guaranteed, interrater variability can be further improved. Our study has several limitations. The first limitation is the retrospective design; no images were routinely taken based on a standard scan protocol. This resulted in poor quality of the images due to non-optimal settings in some cases, the absence of certain anatomical structures on the available images, the absence of CUS on certain days and the absence of follow-up data. As a consequence, a relatively large number of infants had to be excluded. Secondly, neonatal death due to redirection of care was defined as an adverse outcome; the decision was based on a combination of clinical findings, neurophysiology and neuroimaging findings. Even so, we cannot exclude that this leads to some bias because it is not certain that these infants would have experienced problems later in life. However, in cohort I 17 of the 26 infants that died had a postmortem examination and in all infants CUS findings were confirmed. Histopathology showed more extensive damage than CUS, which has also been described for MRI (34). Thirdly, because of the relatively small sample size and the low incidence of adverse motor or cognitive outcome, our study was not powered enough to perform sub- analyses for the different outcome parameters. Finally, the distinction between the categories ‘normal to mild’ and ‘moderate’ hyperechogenicity remains subjective. Severe hyperechogenicity is easier to distinguish from the other categories. However, this is a reflection of clinical practice and with this scoring system we finally have a method that supports the clinician in their daily routine. Furthermore, this scoring 4