Quality control of ultrasound for fetal biometry: results from the INTERGROWTH-21(st) Project.

Objectives To assess a comprehensive package of ultrasound quality control in a large multicentre study of fetal growth – the Fetal Growth Longitudinal Study of the INTERGROWTH-21st Project. Methods We performed quality control (QC) measures on 20,313 ultrasound scan images taken prospectively from 4,321 fetuses at 14-41 weeks’ gestation in eight geographical locations. At the time of each ultrasound examination, three fetal biometric variables were measured in triplicate on separately generated images: head circumference (HC), abdominal circumference (AC) and femur length (FL). All measurements were taken in a blinded fashion. QC had two elements: 1) qualitative QC: visual assessment by sonographers at each study site of their images based on specific criteria with 10% of images being re-assessed at the Oxford-based Ultrasound Quality Unit (compared using an adjusted kappa statistic), and 2) quantitative QC: measurement data were assessed by (a) comparing the first, second and third measurement (intraobserver variability); (b) re-measurement of caliper replacement in 10% (interobserver variability), both by Bland-Altman plots, and (c) plotting frequency histograms of the SDs of triplicate measurements and assessing how many were above or below 2SDs of the expected distribution. The system allowed the sonographers’ performance to be regularly monitored. Results A high level of agreement between the self- and external scoring was demonstrated for all measurements (kappa = 0.99 [95% confidence interval: 0.98, 0.99] for HC, 0.98 [0.97, 0.99] for AC, and 0.96 [0.95, 0.98] for FL. Intraobserver variability (95% limits of agreement (LoA)) of ultrasound measures for HC, AC and FL were ±3%, ±6% and ±6%, respectively; the corresponding values for interobserver variability were ±4%, ±6% and ±6%. The SD distribution of triplicate measurements for all biometric variables showed excessive variability for three of 31 sonographers, allowing prompt identification and retraining. Conclusions Qualitative and quantitative QC monitoring was feasible and highly reproducible in a large multicentre research study, which facilitated the production of high-quality ultrasound images. We recommend that the QC system we developed is implemented in future research studies and clinical practice.