Abstract / Summary
Abstract Background Anastomotic leakage after colorectal resection is often diagnosed late and after discharge. C-reactive protein (CRP) is widely used for early detection, but published thresholds derive largely from open or mixed cohorts, and whether the postoperative trajectory adds to a single absolute value is unsettled. We evaluated the diagnostic accuracy of postoperative CRP for anastomotic leakage after elective laparoscopic colorectal resection, compared absolute values with trajectory metrics, and examined performance by stoma status. Methods Retrospective analysis of a prospectively maintained single-centre database of consecutive adults undergoing elective laparoscopic colorectal resection with anastomosis between January 2017 and December 2025, reported according to STARD. The primary outcome was clinically manifest anastomotic leakage of any grade within 30 postoperative days, graded with a modified ISREC system. The predictive cohort excluded conversions, multivisceral resections, metastatic disease and diverting stomas. CRP on postoperative days (POD) 2–4 was evaluated by ROC analysis; imaging followed clinical suspicion, not protocol. Results Among 367 patients, 347 formed the descriptive cohort (50 leaks, 14.4%) and 260 the predictive cohort (28 leaks, 10.8%). POD4 CRP showed an AUC of 0.930 (95% CI 0.831–0.973); the Youden-derived threshold of 137.3 mg/L gave 92.9% sensitivity (77.4–98.0), 87.3% specificity (82.1–91.1), 98.9% negative predictive value (96.2–99.7) and 49.1% positive predictive value (36.1–62.1), with similar discrimination for invasive or surgical leakage. Trajectory metrics did not improve on absolute POD4 CRP (DeLong p = 0.25 and 0.02); discrimination was attenuated in diverted patients and both false negatives were diagnosed on day 13. Conclusions POD4 CRP showed promise as a risk-stratification marker in non-diverted patients. The cohort-derived threshold should not be a standalone diagnostic or discharge criterion. The positive predictive value was low and a reassuring value did not exclude delayed leakage. External validation is required before implementation.