This retrospective study included all consecutive patients with histologically confirmed high-risk stage II or III CC and determined dMMR tumors who received radical surgical resection from May 2011 to July 2019 at The Sixth Affiliated Hospital of Sun Yat-Sen University. Patients with rectal cancer, incomplete curative resection (R1 or R2 resection), stage I and stage II without any high-risk factors, or adjuvant chemotherapy with fluoropyrimidine alone, CAPOX regimen (capecitabine and oxaliplatin), or duration less than 3 months were excluded. High-risk stage II CC was defined by pathologic stage T4, vascular invasion, lymphatic infiltration, perineural invasion, initial bowel obstruction, tumor perforation, or fewer than 12 excised lymph nodes (15). This study was approved by the Institutional Review Boards of The Sixth Affiliated Hospital of Sun Yat-Sen University.

Formalin-fixed, paraffin-embedded tumors were stained for MLH1, MSH2, MSH6, and PMS2 proteins. Mismatch repair protein loss is defined as the absence of nuclear staining in neoplastic cells but positive nuclear staining in lymphocytes and normal adjacent colonic epithelium (16). Primary monoclonal antibodies against MLH1 (clone M1, Ventana, prediluted), MSH2 (clone G219-1129, Ventana, prediluted), MSH6 (clone 44, Ventana, prediluted), and PMS2 (clone EPR3947, Ventana, prediluted) were applied.

DNA was extracted from formalin-fixed paraffin-embedded (FFPE) tumor tissues. Comparative analysis of normal colon and tumor DNA using the five consensus monomorphic mononucleotide markers (BAT-25, BAT-26, NR-21, NR-24, NR-27) obtained through polymerase chain reaction (PCR)-based assay was adopted to assess the microsatellite instability (MSI). Specimens with at least 2 unstable markers were scored as highly unstable, while those with fewer than 2 unstable markers were stable (17).

MMR status was tested though the analysis of MMR protein expression by immunohistochemistry (IHC), and to be further confirmed by PCR-based MSI testing when the IHC result was undetermined. Deficient MMR phenotype tumors were defined as exhibiting the loss of expression of 1 or more MMR protein by IHC or high-level tumor DNA MSI by PCR. Tumors with discordant results between MMR protein expression and DNA MSI testing were not included in the study.

Under adequate quality-control procedures, mutation analysis was performed at the Molecular Diagnostic Laboratory of the Sixth Affiliated Hospital of Sun Yat-sen University. Genomic DNA was extracted from FFPE samples of surgery with an EZgene Tissue gDNA miniprep kit (Cat no: GD2211, Biomiga, China). KRAS (exons 2, 3, 4), NRAS (exons 2, 3, 4), BRAF (exon 15, V600E mutations), and PIK3CA (exon 9 and 20) were evaluated by bidirectional sequencing using ABI Prism 3 500 DX genetic Analyzer (Applied Biosystems, Foster City, CA).

All patients in this study underwent curative surgical resection, followed by either adjuvant chemotherapy with mFOLFOX6 regimen (oxaliplatin 85 mg/m², leucovorin 400 mg/m², fluorouracil 400 mg/m², and fluorouracil 2400 mg/m² by 48 hours continuous intravenous infusion) for 3 to 6 months or observation only. Follow-up routines consisted of physical examination, serum carcinoembryonic antigen (CEA) assay, and computed tomography scan (chest/abdominal/pelvic) every 3 to 6 months in the first 3 years and every 6 months in the following 2 years. The data were updated in December 2019.

We performed propensity score matching to reduce imbalances in baseline characteristics between patients who received 3 months of mFOLFOX6 or surgery alone and those who received 6 months of mFOLFOX6. A multivariable logistic regression model was constructed to generate propensity scores. We selected covariates for inclusion in the propensity model based on factors presumed to be associated with the patient’s survival outcomes. The following baseline data were included in the model: age at diagnosis, pathologic T stage, pathologic N stage, initial bowel obstruction, vascular invasion and/or lymphatic infiltration, and perineural invasion. Patients who received 3 months of mFOLFOX6 and surgery alone were matched to those who received 6 months of mFOLFOX6 in a 1:1 ratio respectively, according to a greedy nearest-neighbor matching algorithm with no replacement. A caliper width equal to 0.2 of the standard deviation was utilized as the logit of the propensity score. We compared baseline characteristics between the propensity score-matched group using standardized differences. A standardized difference of less than 0.1 can be regarded as negligible imbalance between groups (18).

Categorical variables were compared by dint of the Chi-square test or Fisher’s exact test. DFS was defined as the time from surgery to the first event of local or metastatic recurrence, second primary cancer, or death from any cause. DFS curves were estimated via the Kaplan-Meier method and were compared by means of COX proportional hazards regression model with hazard ratios (HR), 95% confidence intervals (CI), and P values for candidate prognostic factors. Variables with P values of 0.05 or less in univariate analysis or clinically relevant were eligible for the multivariate analyses. Two-sided P values of less than 0.05 were designated as statistically significant. Apart from propensity score matching, which was implemented in R, version 3.3.2 (R Foundation), using the package Matching, all statistical analyses were performed with the 22 version of SPSS software (SPSS Inc. Chicago, IL, USA).

A total of 242 patients with high-risk stage II and III dMMR CC were identified. A complete consolidated standards of reporting trials (CONSORT) diagram depicting the selection process is outlined on Figure 1 . The median age at diagnosis was 55 (range, 22 to 88), with 64.0% of the patients being men. All patients were tested for MMR status by IHC, and 38 cases were also confirmed by PCR-MSI testing. Among 242 patients, 139 (57.4%) had lost MLH1 and PMS2 protein expression, 60 (24.8%) with MSH2 and MSH6 expression, 14 (5.8%) with MSH6 expression, and 29 (12.0%) with PMS2 expression. Additionally, 176 patients had complete data for KRAS, NRAS, BRAF, and PI3KCA status.

Overall, 153 patients received adjuvant chemotherapy with mFOLOFX6 after surgery, which consisted of 6-month therapy (27.3%, n = 66; median cycles [range] = 12 [9-12]) or 3-month therapy (36.0%, n = 87; median cycles [range] = 6 [4-7]), and 89 patients (36.7%) were treated with surgery alone. Baseline patients and tumor characteristics between treatment durations were presented in Table 1 . Patients in the 6-month therapy group were more likely to be younger (< 65 years: 95.5%, 85.1%, 55.1%), with lower proportions of initial bowel obstruction (27.3%, 33.3%, 47.2%), but more of them proceeded to stage III (74.2%, 57.5%, 42.7%) with more positive lymph nodes examined (N1: 53.0%, 42.5%, 36.0%; N2: 21.2%, 14.9%, 6.7%) than those in the 3-month therapy and surgery alone group.

For the overall cohort at the time of data cutoff, the median follow-up was 21.9 months. There were 17 DFS events in patients with high-risk stage II and 29 in stage III disease that led to a 3-year DFS rate of 79.5% (95%CI, 70.1% to 88.9%) and 73.4% (95%CI, 64.2% to 82.6%) respectively.

The 3-year DFS rate was 72.8% (95%CI, 61.0% to 84.6%) for patients who received 3 months of mFOLFOX6 therapy and 86.1% (95%CI, 77.1% to 95.1%) for patients who received 6 months of therapy, along with an estimated multivariate HR of 2.78 (95CI%, 1.18 to 6.47; P= 0.019). For the patients treated with surgery alone, the 3-year DFS rate was 72.4% (95%CI, 60.8% to 84.0%). The multivariate HR for DFS compared with 6-month therapy was 2.30 (95%CI, 0.99 to 5.38; P=0.054) ( Figure 2 , Table 2 , and Supplementary Table S1 ).

Subgroup analysis demonstrated 3-year DFS rate for patients with stage III was 70.8% in the 3-month therapy group and 86.2% in the 6-month therapy group (HR=2.81, 95%CI, 1.03 to 7.67; P=0.044). In high-risk stage II subgroup analysis, no significant effect of adjuvant chemotherapy duration on DFS was observed in the 3-month therapy group compared with the 6-month therapy group ( Figure 3 and Supplementary Table S1 ).

At 1:1 propensity score matching, 51 patients who received 3-month therapy and 35 patients treated with surgery alone were matched to the patients who received 6-month therapy respectively. As shown in Table 3 , after propensity score matching, standardized differences for all included covariates among patients who received 6-month, 3-month therapy, and surgery alone were all less than 0.1, indicating a well-balanced covariate distribution after matching.

After matching, a significant difference in DFS in favor of the 6-month therapy group compared to the 3-month therapy group was observed. The 3-year DFS rate was 88.7% (95%CI, 79.3% to 98.1%) and 68.7% (95%CI, 52.2% to 85.2%) respectively, plus an estimated multivariate HR of 4.35 (95CI%, 1.46 to 13.00; P= 0.008). In subgroup analysis, we identified a benefit on DFS of the 6-month adjuvant chemotherapy compared with the 3-month therapy for stage III patients (3-year DFS rate: 90.3% vs. 64.5%; HR=5.88, 95%CI, 1.44 to 24.04; P=0.014) ( Figure 4 ). In contrast, there was still no significant difference in DFS between the two groups for high-risk stage II patients ( Supplementary Table S2 ). Marginally significant difference in DFS between the 6-month therapy group and the surgery alone group was observed in the multivariable analysis (HR=3.02, 95%CI, 1.00 to 9.07; P=0.049). The 3-year DFS rate was 84.6% (95%CI, 72.1% to 97.1%) and 65.4% (95%CI, 46.6% to 84.2%) respectively ( Figure 4 and Supplementary Table S3 ).