Design: Two-cohort multicentre diagnostic accuracy study
Population: 614 patients after elective colon cancer resection with primary anastomosis
Primary outcome: Anastomotic leak (Clavien-Dindo ≥3) with day-2 CRP <100 mg/L
Effect: NPV 98.8% (test) and 99.0% (validation)

Investigators tested a day-2 C-reactive protein threshold in two cohorts of patients having elective colon cancer resection with primary anastomosis (401 test, 213 validation). Leak was defined as Clavien-Dindo grade 3 or higher.

Leaks occurred in 4.0% and 4.7%. A day-2 CRP below 100 mg/L was present in 60% and 46% of patients, and only 1.3% and 1.0% of them leaked, giving negative predictive values of 98.8% and 99.0%. Above 100 mg/L, leak rates were about 8%. In that higher-risk group, a day-3 CRP lower than day 2 identified further low-risk patients (NPV 98.9% after minimally invasive and 97.9% after open surgery). Using both steps, nearly 90% could be discharged by day 2 or 3 with a leak risk near 1%.

The usual rule uses day 3. Bringing the check a day earlier fits enhanced recovery pathways, which increasingly discharge before day 3.

A low CRP lowers the probability of leak; it does not replace clinical judgement, and patients need clear advice on returning with pain, fever or ileus.

Check CRP on day 2 after elective colon resection within enhanced recovery.
Below 100 mg/L with a well patient, consider discharge.
Above 100 mg/L, repeat on day 3 and look for a falling trend before discharge.
Give every discharged patient clear safety-net advice for late leaks.
This was colon, not rectal, resection; do not extend it to low pelvic anastomoses.

Why it matters

It removes the main biochemical reason to keep a recovering patient until day 3.

The statistics, in plain English

A negative predictive value of 99% means about 1 in 100 patients with a low day-2 CRP still had a leak. That risk is low but not zero, which is why safety-netting matters. The result held in an independent validation cohort, which makes it more trustworthy than a single-cohort finding.