Twenty-six studies met the inclusion criteria (9,436 pregnant women). Seven studies included women with pain or bleeding and an inconclusive ultrasound; most were of high quality. All seven included participants representative of the population of interest and contained adequate details about the index and reference standard tests; none blinded interpreters of the reference standard. Nineteen studies included women with pain or bleeding alone; these were considered of intermediate quality, primarily due to poor reporting making the potential for bias uncertain.
Women with pain, bleeding and an inconclusive ultrasound (2,379 participants): When predicting a non-viable pregnancy, sensitivity ranged from 55% to 91% and specificity from 91% to 100%. The pooled estimate (five studies) of sensitivity was 74.6% (95% CI 50.6 to 89.4), specificity was 98.4% (95% CI 90.9 to 99.7), Positive likelihood ratio was 45 (95% CI 7.1 to 289), and negative likelihood ratio was 0.26 (95% CI 0.12 to 0.57) with cut-off values of 3.2 to 6ng/ml. The progesterone test was a very poor predictor of ectopic pregnancy; sensitivity ranged from 11% to 90% and specificity from 26% to 60%.
Women with pain or bleeding alone (7,057 participants): When predicting a non-viable pregnancy, sensitivity ranged from 34% to 100% and specificity from 32% to 100%. When predicting a non-viable pregnancy (nine studies; 4,689 participants), pooled sensitivity was 66.5% (95% CI 53.6% to 77.4%), specificity was 96.3% (95% CI 91.1% to 98.5%), positive likelihood ratio was was 18 (95% CI 7.2 to 45), and negative likelihood ratio was 0.35 (95% CI 0.24 to 0.50) when the cut-off for a positive test was 10ng/ml. This threshold had a higher specificity than the 15 and 20ng/ml thresholds.
Results for a range of cut-offs and other sensitivity analyses were reported.