To determine if postoperative physical activity prevents or delays cancer recurrence, we estimated the confounder-adjusted hazard rates and HR by physical activity category with continuous time. 14 15 This method was previously applied to randomised trial data to describe the time course of benefit from fluoropyrimidine and oxaliplatin chemotherapy in stage II and III colon cancer. 16 17 We used data from a prospective cohort of patients with stage III colon cancer enrolled in a randomised multicentre trial of postoperative treatment that the National Cancer Institute sponsored. 18 We hypothesised that physical activity would prevent cancer recurrence ( online supplemental figure 1 ), translating into an overall survival benefit.
Graphical depictions of disease-free survival from randomised studies are often presented with Kaplan-Meier plots. 8 Although these plots offer information about the absolute rate of disease recurrence and death, they do not directly depict the risk of an event at a specific time. 9 A separation in Kaplan-Meier curves does not explicitly communicate if an advantage is achieved early and maintained longitudinally or achieved incrementally over time. 9 In non-randomised studies, the unadjusted Kaplan-Meier plot may be misleading due to confounding. 10 11 Confounders in non-randomised studies are often adjusted using Cox regression, which assumes that the hazard rate between two groups is proportional over time, and the effect is estimated as a single HR. 12 13 Consequently, the detailed time course of cancer recurrence by physical activity in colon cancer is unknown.
Physical activity after surgical resection for stage III colon cancer is associated with significantly longer disease-free survival. 1 2 However, the biological mechanisms by which physical activity improves disease-free survival remain incompletely understood. 3 4 Common hypotheses of biological mechanisms include eradicating circulating tumour cells and reducing stimuli within the host microenvironment that foster micro-metastatic growth, such as inflammation, hyperinsulinemia and immune suppression. 5 Physical activity affects all cells and tissues distinctly 6 7 ; therefore, physical activity may confer a disease-free survival benefit by preventing or merely delaying the time of cancer recurrence.
Data were collected by the Alliance Statistics and Data Management Center. Data quality was ensured by review of data by the Alliance Statistics and Data Management Center and by the study chairperson following Alliance policies. Data analysis was conducted by the Alliance Statistics and Data Management Center using SAS (V.9.4) and R (V.4.1.0) on a data set locked on 10 August 2020.
In supplementary analyses, we used flexible parametric proportional hazards survival models to quantify the association between physical activity and cancer recurrence. 28 Parametric survival models estimate absolute and relative effects while permitting flexibility in the baseline hazard function shape (3 knots at default 25th, 50th and 75th centiles of log-time). 29 30 The absolute risk of cancer recurrence is presented as risk differences with bootstrapped 95% CIs, and the relative risk is presented as a HR using all observed data in a time-to-event framework. 31 The number needed to treat (eg, the number of patients who would need to increase their postdiagnosis physical activity to prevent one cancer recurrence) was quantified as the inverse of the absolute risk difference. In sensitivity analysis, we modelled the endpoints without excluding patients with cancer recurrence or death within 60 days after completing the first physical activity assessment.
We used the method of Müller and Wang to plot the continuous-time and confounder-adjusted cause-specific hazard of recurrence from completion of the first physical activity assessment up to year 6 of follow-up, stratified by physical activity category (physically active vs physically inactive). 14 This method allows for the intuitive visualisation of the instantaneous risk of recurrence or death over time. Comparisons of physically active and physically inactive patients were calculated as continuous-time log-hazard ratios (with time starting at the first physical activity assessment) with 95% pointwise CIs estimated using the method of Zhang et al to allow for confounder adjustment, where a pointwise interval that excludes zero represents a statistically significant effect at that time. 15 These methods permit different risk patterns to be examined without modelling assumptions.
To test for differences in baseline patient characteristics by physical activity category, the χ 2 test was used for categorical variables, and the independent t-test was used for continuous variables. Physical activity was modelled using cumulative averaging, which quantifies the time-weighted average of all reported physical activity. 26 27 Follow-up started at the first physical activity assessment, and we updated physical activity based on the results of the second questionnaire that was weighted proportional to the linear function of time between the first and the second questionnaire and the time between the second questionnaire and the disease-free survival period. Repeatedly measured variables (eg, body mass index and dietary patterns) were included as time-varying confounders.
Confounders were selected using the modified disjunctive cause criterion. 24 Data for patient demographic factors, including age, sex, race and ethnicity, were self-reported. Clinical factors, including the extent of tumour invasion through the bowel wall (T-stage), the extent of lymph node metastases (N-stage), pathological risk group (low (T1, T2 or T3, N1) or high (T4, N2 or both)), tumour location, performance status and low-dose aspirin use, were obtained from a combination of physician assessment and the medical record. Smoking history was self-reported. Body mass index was abstracted from a combination of the electronic medical record and self-report. Diet was assessed using a 131-item food frequency questionnaire 25 ; prudent and western dietary patterns were defined using previously validated factor loadings in this population. 26 Body mass index and diet were updated when physical activity was reassessed. All statistical models included the previously described confounders.
The endpoints included disease-free survival, time to cancer recurrence and overall survival. Disease-free survival was defined as the time from completing the first physical activity assessment to the date of documented disease recurrence or death from any cause. Time to cancer recurrence was defined as the time from completion of the first physical activity assessment to the date of documented disease recurrence. Overall survival was defined as the time from completing the first physical activity assessment to the date of death from any cause. Patients were assessed for cancer recurrence by history, physical examination and carcinoembryonic antigen measures every 3 months following randomisation and subsequently every 6 months for 6 years or until recurrence, whichever came first. All patients had surveillance imaging of the chest, abdomen and pelvis every 6 months for at least 3 years and then yearly for 3 years, or until recurrence.
Physical activity was assessed midway through postoperative chemotherapy (4 months following surgical resection) and 6 months after completing postoperative chemotherapy (14 months after surgical resection). Patients reported their average weekly time spent on a range of recreational physical activities during the preceding 2 months using a validated questionnaire. 21 Each physical activity was assigned a metabolic equivalent (MET) energy expenditure according to standardised criteria. 22 We calculated the MET-hours per week (MET-h/wk) for each activity by multiplying the MET value by the patient’s reported number of hours of physical activity each week. We classified patients who reported ≥9 MET-h/wk as physically active (comparable with the energy expenditure of 150 min/wk of brisk walking, consistent with the current physical activity guidelines for cancer survivors 23 ) and patients who reported <9 MET-h/wk as physically inactive. Considering the potential for declining health to bias the physical activity assessment, we pre-specified that the patients who experienced disease recurrence or death within 60 days after completing the first physical activity assessment would be excluded from the analysis.
Patients were enrolled at community and academic centres across the National Cancer Trials Network in the USA and Canada. Eligible patients had surgically resected (margin negative) and histologically documented colonic adenocarcinoma. Patients were enrolled ≥21 days and ≤56 days after surgical resection. Tumours had at least one pathologically confirmed metastatic lymph node or N1c designation (tumour deposit(s) in the subserosa, mesentery, or non-peritonealised pericolic tissue without regional lymph node metastases). Patients were aged ≥18 years, with an Eastern Cooperative Oncology Group (ECOG) performance status ≤2. 18
The Cancer and Leukemia Group B (now part of the Alliance for Clinical Trials in Oncology) and Southwest Oncology Group (SWOG) trial 80 702 was a phase III double-blinded randomised study that used a 2×2 factorial design to test the primary hypothesis of the superiority of celecoxib compared with placebo and the secondary hypothesis of the non-inferiority of 3 months compared with 6 months of chemotherapy. The primary and secondary hypotheses have been published. 18–20 At the time of trial enrolment, patients were offered the option to participate in a nested cohort study of lifestyle factors, including completing standardised assessments at specified time intervals. The randomised trial and nested cohort study were designed in collaboration with the National Cancer Institute.
Sensitivity analyses that did not exclude patients with cancer recurrence or death within 60 days after completing the first physical activity assessment did not substantively change the previously described patterns. In supplementary analyses, the confounder-adjusted 5-year cumulative cancer recurrence rate was 20.4% in physically active patients and 31.5% in physically inactive patients (absolute risk difference: 11.1 percentage points, 95% CI 7.0 to 15.2, p<0.001; HR 0.65, 95% CI 0.49 to 0.82, p<0.001 ( online supplemental table 1 )). Comparing the model fit between the parametric Weibull and Cox regression demonstrated an improved fit using the parametric model ( online supplemental table 2 ).
Plots of the confounder-adjusted continuous-time estimate of the log-hazard ratios for (A) disease-free survival, (B) time to recurrence and (C) overall survival by physical activity group over the follow-up period, with 95% pointwise CIs. Values less than 0 indicate a benefit associated with physical activity. For example, a value of −0.35 on the log-hazard scale is a HR of 0.70, comparing physically active to physically inactive groups (e −0.35 = 0.70), a 30% relative reduction in the risk of an event.
The confounder-adjusted HRs with 95% pointwise CIs comparing physically active to physically inactive patients, plotted on the logarithmic scale, for the endpoints of disease-free survival, time to cancer recurrence and overall survival are plotted ( figure 2 ). Physical activity is associated with a statistically significant disease-free survival benefit for approximately the first postoperative year (estimated log-hazard ratio: −0.38; HR 0.68, 95% CI 0.51 to 0.92), then diminishes in magnitude during follow-up; this pattern is comparable for the endpoint of time to recurrence. Physical activity is associated with a statistically significant overall survival benefit for approximately the first three postoperative years, with the largest risk reduction occurring at postoperative year 2 (estimated log-hazard ratio: −1.16; HR 0.32, 95% CI 0.19 to 0.51).
Plots of the confounder-adjusted hazard rates by time from the first assessment of physical activity for (A) disease-free survival, (B) time to recurrence and (C) overall survival by physical activity group over the follow-up period. The hazard rate of physically inactive patients is plotted in red, and physically active patients are plotted in blue.
The confounder-adjusted hazard rates over time for the endpoints of disease-free survival, time to recurrence and overall survival by physical activity group are plotted ( figure 1 ). For both physically active and physically inactive patients, the hazard of disease recurrence peaks between 1 and 2 years postoperatively and declines gradually to year 5. For the disease-free survival and time to recurrence endpoints, the hazard in physically active patients is consistently lower than in physically inactive patients. For the overall survival endpoint, the hazard in physically active patients is initially lower until approximately year 4, when the risks converge, then the hazard in physically active patients begins to separate again. For no endpoint does the hazard rate in physically active patients ever exceed that of the physically inactive patients during the follow-up period.
Between June 2010 and November 2015, 1696 patients from 654 academic and community oncology centres in the USA and Canada were enrolled in the prospective nested cohort study. Among these patients, 795 (46.9%) were classified as physically active, and 901 (53.1%) were classified as physically inactive. Physically active patients were younger (59.8 vs 61.6 years; p<0.001), more likely to be male (62.6 vs 49.2%; p<0.001), more likely to be white (86.5 vs 76.6%; p<0.001), with better ECOG performance status (80.3 vs 64.6% with ECOG zero; p<0.001), a lower body mass index (27.5 vs 29.1 kg/m 2 ; p<0.001), more likely to be never smokers (52.3 vs 47.1%; p=0.029) and consume a prudent diet pattern (60.1 vs 41.2%; p<0.001) ( table 1 ). During a median follow-up of 5.9 years (IQR 4.9, 6.0), 457 patients experienced disease recurrence or death; 397 patients experienced cancer recurrence, and 281 patients died from any cause. The censoring proportions for disease-free survival, time to cancer recurrence and overall survival were 0.73, 0.77 and 0.83, respectively; the primary reason for censoring was the last disease evaluation date, as reported previously. 18
Discussion
In this nested cohort study of 1696 patients with stage III colon cancer enrolled in a randomised multicentre trial, the risk of cancer recurrence in physically active patients never exceeded that of physically inactive patients during follow-up. These results are consistent with the hypothesis that postoperative physical activity may prevent, as opposed to delay, cancer recurrence in some patients with stage III colon cancer. Based on these data, we speculate that postoperative physical activity improves disease-free survival by reducing the cancer recurrence rate within the first year of treatment, which translates into an overall survival benefit. These findings refine our understanding of how physical activity improves cancer survivorship in a manner that may be relevant to tumour biology and cancer care delivery.
The pleiotropic effects of physical activity may impede several mechanistic determinants of metastatic colonisation by circulating tumour cells.5 The proportion of circulating tumour cells in the bloodstream that undergo cell fragmentation, cell cycle arrest and cell death is positively related to the magnitude and duration of shear stress exposure.32 Physical activity causes significant increases in shear stress,33 and aerobic activity reduces circulating tumour cells in patients with stage I–III colon cancer.34 Physical activity improves immune surveillance (eg, NK cell activity)35 and may foster improved distant organ tissue defences against infiltrating tumour cells.36 Physical activity reduces inflammation and hyperinsulinemia in patients with colon cancer,37 38 and may limit the availability of niches that can support the metabolic demands for metastatic cell growth.39
We previously reported that physical activity was associated with improved chemotherapy relative dose intensity.2 Postoperative fluoropyrimidine and oxaliplatin chemotherapy in colon cancer eradicate residual cancer cells and micro-metastases, thereby curing some patients.16 17 In our prior study, randomised chemotherapy length (3 months vs 6 months) did not significantly modify the association between physical activity and disease-free survival.2 However, we cannot rule out the possibility that physical activity reduces cancer recurrence by improving chemotherapy adherence. In 2022, the National Cancer Institute funded a Bayesian randomised trial (U01-CA271279) to test the primary hypothesis that aerobic exercise improves chemotherapy relative dose intensity in patients with colon cancer.
In patients with metastatic colorectal cancer, physical activity during chemotherapy is associated with a significantly longer progression-free survival but not overall survival.40 Therefore, prior to this analysis, it was plausible that in stage III colon cancer, physical activity may simply delay cancer recurrence, for example, by inducing the cellular dormancy of disseminated tumour cells (single cancer cells that have survived infiltration into distant organs) or by inducing tumour mass dormancy.41 However, during a median 5.9-year follow-up, we found no evidence to support this hypothesis. Standard-of-care follow-up and surveillance protocols for recurrent cancer remain appropriate for physically active patients.42
Clinical implications Physical activity is safe for cancer survivors and recommended during chemotherapy.43 Our analysis indicates that the magnitude of benefit from physical activity on cancer recurrence is larger in the early postoperative period and attenuates with time. This time course may be relevant to patients who seek to understand the optimal time to begin physical activity to reduce their cancer recurrence risk. In addition, randomised trials demonstrate that physical activity during chemotherapy reduces cancer-related fatigue and improves physical functioning and health-related quality of life.44 The association between physical activity and the overall survival endpoint did appear to be a combination of prevention and delay. In patients with resected colon cancer, approximately 80% of deaths that occur within the first three to five postoperative years are attributed colon cancer and preceeded by tumour recurrence.45 However, beyond 5 years, cardiovascular events (eg, myocardial infarction, stroke) become the principal cause of death.46 Physical activity is associated with a lower risk of cardiovascular disease.47 We speculate that the hazard pattern between physical activity and the overall survival endpoint may illustrate the time when the principal causes of death transition from cancer related to cardiovascular related.