Effect of the duration of the capecitabine regimen following colon cancer surgery in an elderly population: a retrospective cohort study

Patient data were collected from the electronic medical record system of Guizhou Province Cancer Hospital and the Affiliated Hospital of Guizhou Medical University. We included high-risk stage II and stage III patients who were aged ≥ 70 years and underwent complete mesocolic excision with lymph node dissection (Fig. 1). After surgery, the patients were recommended to receive capecitabine adjuvant chemotherapy (1000 mg/m² twice daily for 14 days followed by a 7-day rest period). The local ethics committee of Guizhou Province Cancer Hospital approved this study (2020-FZ 00,513).

This retrospective cohort study aimed to assess the association between postoperative capecitabine duration and CSS and DFS in elderly colon cancer patients. From August 2013 to September 2019, a total of 1217 surgery-treated colon cancer participants were retrospectively reviewed. The variables used to establish the multivariate adjusted models included (1) continuous variables: age, carcinoma antigen 199 (CA199, obtained before surgery), carcinoembryonic antigen (CEA, obtained before surgery), CCI (calculated using data regarding complications) [10], and (2) categorical variables: sex, T stage, N stage, histological type, and MMR status.

Follow-up for enrolled patients through outpatient visits was applied once every 3 months for the first 2 years, every 6 months for the next 3 years, and every 1 year thereafter. A complete physical examination, thoraco-abdominal CT, pelvic CT, serum CEA and CA199 assessment, and an annual colonoscopy were conducted at each follow-up. The last date of follow-up was September 1, 2020. The follow-up information was obtained by the second and third authors from the electronic medical record system of Guizhou Province Cancer Hospital and the Affiliated Hospital of Guizhou Medical University.

Continuous variables are described as the mean ± standard, and categorical variables are described as frequency or percentage. Student’s t test (normal distribution), Mann–Whitney U test (skewed distribution), or χ² test (categorical variables) was adopted as appropriate to investigate differences in the characteristics of the enrolled patients among the various postoperative chemotherapy duration groups. A univariate Cox proportional hazard model and a multivariate Cox proportional hazard model were applied to assess variables that may correlate with the risk of death. Three models were constructed: model 1, not adjusted for any covariate; model 2, adjusted for gender and age; and model 3, adjusted for the covariates presented in Table 1. Stratified analyses were performed by Cox proportional hazard models. A Cox proportional hazards regression model with cubic spline functions and smooth curve fitting was applied to explore the nonlinear relationship between postoperative chemotherapy duration and CSS. If nonlinearity was detected, then the inflection point was calculated by the recursive algorithm, and a two-piecewise Cox proportional hazard model was established on both sides of the inflection point. Finally, which model was more suitable for fitting the association between the target-independent variable and the outcome variable was determined by the log-likelihood ratio test.

We conducted a series of sensitivity analyses to guarantee the robustness of the data analysis. First, due to the limitations of the Cox proportional hazards model in addressing nonlinearity, a generalized additive model was used to adjust the covariates; we then compared its effect size with the fully adjusted model. Second, we used different covariates as the stratification variables to observe the trend of HR among the different subgroups and calculated the P for trend. All data analysis was performed with the statistical software packages R (http://​www.​R-project.​org, The R Foundation) and EmpowerStats (http://​www.​empowerstats.​com, X&Y Solutions, Inc., Boston, MA, USA). P values < 0.05 (two-sided) were considered statistically significant.

From August 2013 to September 2019, a total of 1217 surgery-treated colon cancer patients from Guizhou Province Cancer Hospital or the Affiliated Hospital of Guizhou Medical University were retrospectively reviewed, and 257 elderly patients with stage III and high-risk stage II disease were enrolled in the final analysis. Postoperative chemotherapy with a capecitabine-alone regimen was administered to 114 (114/257, 44.36%) patients, and 143 patients only received surgery. A total of 43 patients died of colon cancer (43/257, 16.73%) during follow-up, and 8 (8/257, 3.11%) were lost to follow-up. The median follow-up time was 30.4 months. Table 1 describes the baseline characteristics of the enrolled patients across the categories of chemotherapy duration. No statistically significant difference was found in the measures except for age, MMR status, and CCI among the different chemotherapy duration groups. Compared to patients receiving postoperative chemotherapy ≤ 12 weeks or > 12 weeks, those who did not receive chemotherapy were older. Moreover, patients without postoperative chemotherapy had higher rates of CCI and MMR deficiency (dMMR) incidents.

We evaluated the median CSS by the Kaplan–Meier method with a log-rank test. As shown in Fig. 2, the median CSS was 25.4 months (95% confidence interval (CI), 1.0–81.1) for the group without adjuvant chemotherapy, 27.4 months (95% CI, 1.0–67.9) for the group with chemotherapy ≤ 12 weeks and 35.5 months (95% CI, 11.2–84.2) for the group with chemotherapy > 12 weeks. The differences among the groups were significant (P = 0.003).

The results of the univariate analysis are shown in Table 2. Patients with postoperative capecitabine ≤ 12 weeks or > 12 weeks were associated with a better DFS and CSS than those treated with surgery alone. N2, stage III, and elevated CA199 were associated with poor DFS and CSS.

As reported in Tables 3 and 4, three models were used to analyze the independent association of postoperative capecitabine duration on CSS and DFS. In the unadjusted model shown in Table 3, 0.90 (HR) demonstrated that the risk of cancer-specific death was reduced by 10% as the capecitabine duration increased by 1 week. In the adjusted model 1, after adjusting for sex and age, the negative relationship was still robust. In the adjusted model 2, we adjusted for sex, age, stage, CEA (elevated, normal), CA199 (elevated, normal), CCI (≤ 4, > 4), histological type, and MMR status as confounding factors. After adjusting for confounding factors, every additional week of capecitabine duration was associated with an 11% decrease in the risk of death. As shown in Table 4, each additional week of capecitabine correlated with a 10% decrease in the risk of recurrence. Then, postoperative capecitabine duration was transformed from a continuous variable to a categorical variable for the sensitivity analysis. The results of postoperative capecitabine duration as a categorical variable were consistent with the results when the capecitabine duration was a continuous variable.

For the sensitivity analysis, we first set up model 2, which adjusted for sex, age, stage, CEA (elevated, normal), CA199 (elevated, normal), CCI (≤ 4, > 4), histological type, and MMR status. The results were consistent with those of the crude model and model 1. We then used different covariates as the stratification variables to observe the trend of HR among the different subgroups (Fig. 3). The core results were consistent with our initial cohort. The tests for interactions were not statistically significant for age, sex, stage, histological type, CEA, CA199, or MMR status (P > 0.05).

In our study, the nonlinearity of postoperative capecitabine duration on CSS was observed in stage III (Fig. 4a, P for nonlinearity = 0.003) but not in high-risk stage II (P for nonlinearity = 0.24). After adjusting for sex, age, T stage, CA199, CEA, histological type, and CCI and MMR status, the smooth curve analysis and the Cox proportional hazards regression model with cubic spline functions indicated that the relationship between the postoperative capecitabine duration and CSS was still nonlinear and presented an L-shape in stage III elderly patients (Fig. 4b, P for nonlinearity = 0.01). Both the Cox proportional hazard model and the two-piecewise Cox proportional hazard model were applied to fit the association. We selected the best fit model according to P of the log-likelihood ratio test.

Because its P for the log-likelihood ratio test was < 0.05, we selected the two-piecewise Cox proportional hazard model for fitting the association between postoperative capecitabine duration and CSS in stage III elderly patients. The inflection point was calculated as 16 by using the recursive algorithm and two-piecewise Cox proportional hazard model (Table 5). The HR for death was 0.79 with a postoperative capecitabine duration ≤ 16 weeks and 1.34 with a duration > 16 weeks. However, in high-risk stage II patients, we did not observe this nonlinear association.