Introduction

Adjuvant chemotherapy following surgical resection of colon cancer is intended to eliminate micrometastatic disease and increase overall and disease-free survival [1]. Evidence from randomized controlled trials (RCTs) established that adjuvant chemotherapy increased survival among stage III colon cancer patients, [2,3,4] and thus adjuvant chemotherapy for stage III colon cancer is standard of care. In addition, evidence suggests that early initiation of adjuvant chemotherapy may improve survival among colon cancer patients, [5,6,7,8] and timing of adjuvant chemotherapy initiation is included in quality of care guidelines [9, 10]. Nevertheless, the evidence for timing of adjuvant chemotherapy was derived from observational studies, which may be sensitive to time-related biases [11, 12].

In contrast to RCTs, where the design ensures alignment of key elements such as treatment allocation, eligibility, and follow-up at a common starting point (i.e., time zero) to minimize time-related biases, this alignment requires additional methodologic considerations in observational studies [11, 12]. For example, prior studies [5,6,7,8] included comparisons of survival between patients who initiated adjuvant chemotherapy ≤8 weeks and patients who initiated adjuvant chemotherapy > 8 weeks following surgical resection. Such comparisons are problematic because patients must survive a specified duration before initiating treatment, and analyses that do not properly account for this duration will incur immortal time bias [11, 13,14,15,16,17]. Several studies have illustrated severe overestimation of treatment effects on survival in observational studies of cancer patients, which was attributable to immortal time bias [18,19,20]. Immortal time bias has not been addressed in the context of timing of adjuvant chemotherapy initiation for colon cancer survival and could have implications for current quality of care guidelines. Therefore, we aimed to estimate the effect of initiating adjuvant chemotherapy for stage III colon cancer within an interval of interest (8 or 12 weeks) following surgical resection on survival using a study design that helps align time zero and reduce immortal time bias.

Methods

Study population

We used institutional registry data from JPS Oncology and Infusion Center (JPS), an accredited Comprehensive Community Cancer Program. The center is part of an urban safety-net health system, which is the primary source of care for socioeconomically marginalized populations in Tarrant County, TX. Eligible patients were diagnosed with first primary stage III colon cancer between 2011 and 2017, aged 18–79 years at cancer diagnosis, received surgical resection with curative intent, and received at least part of the first course treatment at JPS. We excluded patients for whom adjuvant chemotherapy was contraindicated.

Variables

Our primary outcome of interest was 5-year overall survival (i.e., complement of all-cause mortality) and secondary outcome of interest was 5-year disease-free survival (i.e., complement of recurrence or mortality [21]). These outcomes were selected because adjuvant chemotherapy is intended to improve overall and disease-free survival [1]. In addition, the American Society of Clinical Oncology statement about clinically meaningful outcomes defines overall survival as the primary outcome of interest [22]. Our exposure (intervention) of interest was initiation of adjuvant chemotherapy within 8 or 12 weeks of surgical resection (i.e., initiators vs. non-initiators within 8 weeks in one analysis and initiators vs. non-initiators within 12 weeks in a separate analysis). We also extracted baseline information from the registry including age at diagnosis, sex, race/ethnicity, insurance coverage, marital status, comorbidities, body mass index, tumor grade, and surgical procedure.

Data analysis

We emulated a sequence of observational “trials,” [11, 23,24,25] where study eligibility criteria were applied and intervention status was defined within a sequence of trials based on 2-week intervals through 8 or 12 weeks from surgical resection. One exception was that the interval for the first trial was 4 weeks because no one initiated adjuvant chemotherapy within 2 weeks of surgical resection. Baseline (i.e., time zero) for the first trial was the date of surgical resection. For the first trial, patients were classified as initiators if adjuvant chemotherapy was initiated ≤4 weeks (i.e., 28 days) of surgical resection, and as non-initiators if adjuvant chemotherapy was not initiated ≤4 weeks. We applied the eligibility criteria and initiator definition for sequential 2-week intervals, where time zero for each trial was the beginning of each interval. Consequently, patients could have been eligible for up to three trials for evaluating initiation ≤8 weeks and five trials for evaluating initiation ≤12 weeks of surgical resection but were no longer eligible for subsequent trials if adjuvant chemotherapy was initiated, the patient died, had recurrence (for disease-free survival), or were lost to follow-up in a previous trial.

We subsequently pooled data from all three (for evaluating initiation ≤8 weeks of surgical resection) or five trials (for evaluating initiation ≤12 weeks of surgical resection), which allowed for reducing variance [11, 23,24,25]. We fit a logistic regression model to compute stabilized inverse probability of treatment weights (IPTW) [26] for adjuvant chemotherapy initiation. Stabilized IPTW were based on a minimal sufficient set of covariates to reduce confounding bias identified using the back-door criterion in a directed acyclic graph of dependency assumptions [27,28,29] between adjuvant chemotherapy initiation and mortality or recurrence. The minimal sufficient set of covariates included baseline measurements of age, sex (male or female), race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, or non-Hispanic other), insurance coverage (uninsured without JPS Connection [a hospital-based medical assistance program for eligible individuals without insurance], uninsured with JPS Connection, or commercial/public insurance), marital status (single/never married, married, or divorced/separated/widowed), comorbidities classified by the National Cancer Institute [30] (0 or > 0), body mass index (BMI; BMI < 25, 25 ≤ BMI < 30, or BMI ≥ 30), tumor grade (well/moderately differentiated or poorly differentiated/undifferentiated), and surgical procedure (partial colectomy/segmental resection or hemicolectomy/subtotal/total colectomy). The standardized mean differences for covariates between initiators and non-initiators of adjuvant chemotherapy did not suggest meaningful imbalance after weighting except for one category of marital status in the 8-week analysis (Supplementary Table S1 and S2) [31].

We adjusted Kaplan-Meier estimators using stabilized IPTW [23, 46]. Lastly, survival could be affected by adherence to adjuvant chemotherapy [47, 48], but addressing adherence would change the question of interest and require a different study design [49]. Our study was designed to address the effect of initiating adjuvant chemotherapy, which is the question of interest relevant to quality of care guidelines and the basis of prior studies.

Our findings differ from prior studies [50,51,52,53], in which point estimates suggested 25–55% lower mortality hazards for initiation of adjuvant chemotherapy ≤8 weeks after surgical resection compared with later initiation or no initiation among colon cancer patients. Time-related biases [11, 12, 14, 15, 54] from misaligned start of eligibility, time of treatment assignment, and start of follow-up are a key consideration for effect heterogeneity between our study and prior studies. For example, immortal time bias is a concern in prior studies [50,51,52,53, 55] because follow-up time was measured from surgical resection, but adjuvant chemotherapy was initiated after follow-up time began. The consequence is misclassified person-time, where the time between start of follow-up and treatment initiation is considered “immortal.” Several studies have reported substantial bias away from the null because of immortal time, [19, 20, 25, 56] and this bias can be more severe than unmeasured confounding [23, 46]. We used analytic methods to mitigate immortal time bias, [11, 57] which may partly explain why our point estimates are closer to the null than prior studies. In addition, prior studies [50,51,52] excluded patients who were eligible for but did not initiate adjuvant chemotherapy, which incurs selection bias [58]. Lastly, effect heterogeneity across studies may be related to clinical setting and population characteristics. For example, our study was conducted in a cancer center that provides care for socioeconomically marginalized populations. Our estimates may be closer to the null if adverse effects of social determinants of health override benefits of earlier treatment initiation [59, 60]. Consequently, our results may generalize to other safety-net settings but not necessarily academic cancer centers.

In summary, assuming no substantial effect of biases, our results suggest that initiating adjuvant chemotherapy ≤8 or ≤ 12 weeks of surgical resection for stage III colon cancer patients < 80 years may not be as beneficial as reported in prior studies. Nevertheless, our results were imprecise and require confirmation. Future studies that also address time-related biases and have larger samples or longer follow-up may provide greater precision. Alternatively, estimates from multiple similar studies may be combined in a meta-analysis to improve precision. Such evidence could be valuable considering that cancer care delivery organizations dedicate considerable resources to meet guidelines for timely care, but some guidelines may not be optimized for meaningful outcomes or for certain settings.