Frequency and cause of disagreements in diagnoses for fetuses referred for ventriculomegaly

Abstract

Purpose: To prospectively assess the frequency and cause of disagreements in diagnoses at ultrasonography (US) and magnetic resonance (MR) imaging for fetuses referred for ventriculomegaly (VM).

Materials and methods: One hundred ninety-five women, aged 18-44 years, with 200 fetal referrals for VM, were recruited in a prospective IRB-approved, HIPAA-compliant study. Written informed consent was obtained. US scans were prospectively interpreted by three obstetric radiologists and MR examinations were read by one obstetric radiologist and three pediatric neuroradiologists. Final diagnosis was reached by consensus (198 US, 198 MR, and 196 US-MR comparisons). Gestational age, ventricular size, types of disagreements, and reasons for disagreements were recorded. Interreader agreement was assessed with kappa statistics. Ventricular diameter, gestational age, and confidence scores were analyzed by using mixed-model analysis of variance, accounting for correlation within reader and fetus.

Results: There was prospective agreement on 118 (60%) of 198 US and 104 (53%) of 198 MR readings. Consensus was more likely when the final diagnosis was isolated VM (83 of 104, 80% at US; 82 of 109, 75% at MR) than when the final diagnosis included other anomalies as well (14 of 63, 22% at US; seven of 68, 10% at MR; P < .001). There was disagreement on 19 (10%) of 196 and 31 (16%) of 196 fetuses about the presence of VM at US and MR, respectively, and on 29 (15%) of 198 and 39 (20%) of 198 fetuses regarding the presence of major findings at US and MR, respectively. Reasons for discrepancies in reporting major findings included errors of observation, lack of real-time US scanning, lack of neuroradiology experience, as well as modality differences in helping depict abnormalities.

Conclusion: Of radiologists who read high-risk obstetric US and fetal MR images for VM, there is considerable variability in central nervous system diagnosis.

(c) RSNA, 2008.

Figures

Figure 1:

Study design. * = MR not performed in two fetuses. ** = US performed by one obstetric radiologist and independently read by two obstetric radiologists from second institution. † = Two fetal US with NTD findings were incomplete at MR and not included in US consensus. ‡ = MR consensus performed by three pediatric neuroradiologists used to assess need for consensus diagnosis by obstetric radiologist who read MR, diagnoses of neuroradiologists.

Figure 2a:

ACC with heterotopias at 35 weeks gestational age. (a) Transverse US shows colpocephaly with slitlike frontal horns and areas of increased echogenicity (arrows) lining ventricles. (b) Transverse T2-weighted MR image (echo spacing, 4.2 msec; echo time, 60 msec; echo train length, 72; one acquisition; section thickness, 4 mm; field of view, 26 × 30 cm; matrix, 192 × 256; sequence acquisition, 16 seconds; and section acquisition, 420 msec) shows similar findings with areas of low signal intensity projecting into ventricle (arrows). ACC was coded by all readers, but heterotopias were coded by only one of three US readers and all three neuororadiologists.

Figure 2b:

ACC with heterotopias at 35 weeks gestational age. (a) Transverse US shows colpocephaly with slitlike frontal horns and areas of increased echogenicity (arrows) lining ventricles. (b) Transverse T2-weighted MR image (echo spacing, 4.2 msec; echo time, 60 msec; echo train length, 72; one acquisition; section thickness, 4 mm; field of view, 26 × 30 cm; matrix, 192 × 256; sequence acquisition, 16 seconds; and section acquisition, 420 msec) shows similar findings with areas of low signal intensity projecting into ventricle (arrows). ACC was coded by all readers, but heterotopias were coded by only one of three US readers and all three neuororadiologists.

Figure 3:

Confidence of US and MR readers in diagnosis of fetal CNS abnormalities. Bars indicate mean ± standard error on five-point scale from very confident to not confident. P values from analysis of variance are adjusted for interreader and intersubject variability.