Poor 'real-life' negative predictive value of cross-sectional imaging in obstructive jaundice

Cross-sectional imaging remains the first-line test for obstructive jaundice despite high miss rates for pancreatobiliary tumours. Improvements in resolution and slice thickness of spiral computed tomography⁄magnetic resonance imaging⁄magnetic resonance cholangiopancreatography promised to increase accuracy.

To assess whether the post-test probability of neoplasm is truly altered by the presence or absence of a mass on computed tomography⁄magnetic resonance imaging in obstructive jaundice.

The institutional endoscopic ultrasound (EUS) database was retrospectively reviewed to stratify patients presenting to EUS over a two-year period for obstructive jaundice (suspicious for malignancy) according to their pre-EUS imaging results. The primary analysis involved the calculation of the positive predictive value and negative predictive value (NPV) of imaging with 95% binomial CIs. Test performance of EUS⁄fine-needle aspiration (FNA) was also calculated. Final diagnosis was determined by positive cytology⁄histology; negative EUS was supplemented by clinical follow-up.

The positive predictive value (n = 51) and NPV (n = 53) of pre-EUS imaging was 98% (95% CI 90% to 100%) and 9% (95% CI 3% to 21%), respectively (accuracy 53%), with post-test suspicion of malignancy similar between imaging-positive and -negative groups. EUS demonstrated a mass in 96% of imaging-positive cases versus 85% in imaging-negative cases (exact P = 0.09). Malignant or suspicious FNA cytology was obtained with EUS in 92% of the imaging-positive group, and 62% of the imaging-negative group (75% of subgroup with FNA) (P < 0.001).

Lack of a definite mass on pre-EUS imaging had low NPV, and was clearly not sufficiently accurate or reassuring in this clinical setting. In suspicious obstructive jaundice, EUS with FNA has a high diagnostic yield regardless of the findings of pre-EUS cross-sectional imaging and, as such, EUS may be a more reasonable first-line test in this high-suspicion setting.