The ultrasonographic estimation of thyroid size has been advocated as being more precise than palpation to diagnose goitre. However, ultrasound also requires technical proficiency. This study was conducted among Saharawi refugees, where goitre is highly prevalent. The objectives were to assess the overall data quality of ultrasound measurements of thyroid volume (Tvol), including the intra- and inter-observer agreement, under field conditions, and to describe some of the practical challenges encountered.
In 2007 a cross-sectional study of 419 children (6-14 years old) and 405 women (15-45 years old) was performed on a population of Saharawi refugees with prevalent goitre, who reside in the Algerian desert. Tvol was measured by two trained fieldworkers using portable ultrasound equipment (examiner 1 measured 406 individuals, and examiner 2, 418 individuals). Intra- and inter-observer agreement was estimated in 12 children selected from the study population but not part of the main study. In the main study, an observer error was found in one examiner whose ultrasound images were corrected by linear regression after printing and remeasuring a sample of 272 images.
The intra-observer agreement in Tvol was higher in examiner 1, with an intraclass correlation coefficient (ICC) of 0.97 (95% CI: 0.91, 0.99) compared to 0.86 (95% CI: 0.60, 0.96) in examiner 2. The ICC for inter-observer agreement in Tvol was 0.38 (95% CI: -0.20, 0.77). Linear regression coefficients indicated a significant scaling bias in the original measurements of the AP and ML diameter and a systematic underestimation of Tvol (a product of AP, ML, CC and a constant). The agreement between re-measured and original Tvol measured by ICC (95% CI) was 0.76 (0.71, 0.81). The agreement between re-measured and corrected Tvol measured by ICC (95% CI) was 0.97 (0.96, 0.97).
An important challenge when using ultrasound to assess thyroid volume under field conditions is to recruit and train qualified personnel to perform the measurements. Methodological studies are important to assess data quality and can facilitate statistical corrections and improved estimates.