Eight studies were included in the analysis: two were prospective and six retrospective; total sample size appeared to be 523 patients (range 35 to 157). QUADAS scores ranged from 8 to 12 out of 14. Three areas of concern were unclear disease progression bias, whether assessors were blinded to the results of the reference test, and the existence of uninterpretable results.
Pooled sensitivity of cyst fluid CEA level in predicting malignant pancreatic cysts was 63% (95% CI 56% to 70%; Ι²=78%); pooled specificity was also 63% (95% CI 58% to 69%; Ι²=89%). The DOR was 3.84 (95% CI 1.37 to 10.75; Ι²=77%) and the AUC was 0.70.
The post hoc analysis did not materially change the overall results and heterogeneity remained high.
In subgroup analysis including only patients with mucinous cysts, the DOR was 4.74 (95% CI 1.46 to 15.37; five studies). When only higher quality studies were included (QUADAS score ≥11), the DOR was 5.15 (95% CI 1.38 to 19.28; five studies) and heterogeneity was reported to remain high.
Sensitivity analysis revealed that type of pancreatic cyst and study quality did not have statistically significant effects on the overall DOR.