Association Between Time to Initiation of Adjuvant Chemotherapy and Survival in Colorectal Cancer: Title and subTitle BreakA Systematic Review and Meta-analysis FREE

Colorectal cancer (CRC) is the third leading cause of cancer mortality in the Western world.¹ Quiz Ref IDWhile surgical resection remains the cornerstone of management for patients with stage I-III disease, a considerable proportion of patients will ultimately relapse and die from their disease.Quiz Ref ID Large randomized clinical trials of adjuvant chemotherapy (AC) after curative resection of CRC have consistently demonstrated improvement in survival, which dictates the current standard of care.² Adjuvant chemotherapy is routinely recommended after curative surgical resection of stage II-III rectal cancer, stage III (node-positive) colon cancer, and stage II (node-negative) colon cancer in which high-risk features are present.^{3 - 4} However, the optimal time from surgery to the start of chemotherapy in CRC is not known.

Because most clinical trials mandate that AC is to commence within 6 to 8 weeks after surgery, a routine clinical assumption is that chemotherapy should commence as soon as practical. Additionally, chemotherapy is often assumed by clinicians to have little or no adjuvant benefit beyond a 3-month delay. There is no direct evidence to support either of these assumptions. The timing to start of AC has not been subjected to a randomized controlled trial; nor is such a trial likely to be undertaken. Reasons for a delay in time to AC may relate to patient factors such as postoperative complications or comorbid conditions or health-system logistic factors such as delays in referral or wait times.

The question of time to AC is an important one. Timely access to AC is often cited and tracked as a quality indicator.⁵ Furthermore, beyond a certain time frame from surgery, such as the often quoted 12 weeks, it is uncertain whether the adjuvant benefit diminishes or is even lost entirely. To address this important gap in the literature, we undertook a formal systematic review of the literature and meta-analysis to identify studies that assessed the relationship between time to AC and survival in CRC.

Potentially relevant studies were identified through a structured literature search of MEDLINE (1975 to the end of January 2011) using the Medical Subject Headings adjuvant chemotherapy, colorectal neoplasms, drug administration schedules, time factors, survival rate, and survival analysis. These were combined with keyword searches to define a primary collection of the relevant literature. The EMBASE database, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials were searched but did not yield any additional relevant articles. Studies were selected for further analysis based on careful reading of the online abstracts. The reference lists from these studies were reviewed to identify studies not identified in the original search. Abstracts available from the online proceedings of the annual meetings of the American Society of Clinical Oncology and European Society for Medical Oncology (2007-2010) were searched for reports of newly completed relevant studies, which may not yet be included in MEDLINE. To reduce the effect of publication bias, both fully published articles and abstracts were eligible for inclusion. Only English-language studies were included.

Studies were required to meet the following inclusion criteria: all patients were treated with AC, a clearly defined measure of time from surgery to initiation of AC was documented, and the relationship between time interval from surgery to initiation of AC and patient outcomes in colorectal cancers was reported. Studies that used nonstandard forms of AC (eg, perioperative chemotherapy), used cross-trial comparison designs, or examined the effect of concurrent or sequencing of additional adjuvant therapies (eg, radiation or hormonal) were excluded.

To minimize the effect of confounding between comparison groups, we adapted a validity methodology used by our group previously⁶ ; only studies identified as “high validity” by the following criteria were included in the final study cohort: first, the relevant prognostic factors were adequately described between comparator groups of patients; and second, either the comparison groups were balanced for the relevant prognostic factors, or the reported results were adjusted for these prognostic factors. Two reviewers (J.J.B. and M.J.R.) independently assessed studies for inclusion in the final cohort and completed data abstraction into an electronic database that was first piloted using 4 studies; disagreements were resolved by consensus with a third author (C.M.B.).

The measure of effect in all studies was a hazard ratio (HR) for 2 survival measures, overall survival (OS) and disease-free survival (DFS). The eligible studies used disparate categorical representations of waiting time (WT). To provide a common representation for synthesis of the results of studies, we converted the WT effect size reported in each study to a regression coefficient (β) and standard error corresponding to a continuous representation per 4 weeks of WT. For the WT categories in each article, a central value of WT was assigned to each category. Taking into consideration the range of WTs reported in the article and usual clinical care, expert review by 2 medical oncologists (J.J.B. and C.M.B.) determined a central value.

Regression coefficients and standard errors for dichotomous and ordinal representations were converted as follows. For studies with only 2 WT groups, the β was calculated as log(HR)/([x _n − x₀]*3.92), and the corresponding SE was calculated as (log[upper CI]−log[lower CI])/([x _n − x₀]*3.92), where CI is confidence interval, x _n denotes exposure at group n level, and x₀ denotes exposure at reference group. If only a P value was provided, the SE was calculated as the “test-based” method: SE = (log[HR])/Z_p, where Z_p is the value of a unit-normal test (eg, Z_p = 1.96 if P = .05, 2-tailed test). For the studies with more than 2 groups, weighted linear regression of the log of the HR on exposure level was used to estimate the β with weights equal to the inverse of the variance of the HR estimates.⁷ The SE for β was then computed using the approach described in Greenland and Longnecker.⁸ The HR can be considered as a measure of the incidence rate ratio and the summary measures presented here can be interpreted as the incidence rate ratio for the outcome (eg, death or recurrence) with each 4 weeks of additional waiting for AC. These estimates are only relevant for the range of WTs covered in the underlying studies and assume a log linear relationship across WTs.

The adjusted regression coefficients from individual studies were combined using a fixed-effects model. The inverse variance (1/SE²) was used to weight individual studies. Analysis was done for OS and DFS. The homogeneity assumption in the meta-analysis was assessed by Cochran χ² statistic,⁷ and I² statistics were calculated for each result. The statistical analyses were performed using R package meta version 1.5-0 (R Foundation for Statistical Computing, Vienna, Austria). A 2-tailed P < .05 was considered statistically significant.

To detect publication bias, we examined the asymmetry of standard error–based funnel plots using the linear regression method suggested by Egger et al.⁹ In addition, the trim and fill approach was used to obtain an adjusted effect size that takes into account publication bias.^{10 - 11} The sensitivity of the analysis to the largest studies was examined by sequentially removing the largest studies and calculating a combined result from among the remaining studies.

The search strategy yielded 198 reports, of which 22 met the inclusion criteria (Figure 1). Twelve were excluded either because of lack of data or because they did not meet the high-validity criteria, leaving 10 and 6 eligible studies for OS and DFS analysis, respectively (Table 1 and Table 2).^{12 - 21} The studies were published between 2005 and the end of January 2011 as abstracts (n = 3) or full articles (n = 7). There were 15 410 patients with colon and rectal cancer, with a range of sample size from 179 to 6059. Six of the studies had two categorical time variables, 2 had three, and 2 had four categories. One study reported time-to-AC data as a secondary analysis within a randomized controlled trial of chemotherapy treatment, while 9 studies were registry or population based.

The HR results from individual studies listed in Table 2 are plotted in Figure 2A, which shows the HRs for categorical representations of WT in the 10 studies. For the 2 studies reporting an HR for a continuous WT representation,^{14 ,17} the points correspond to a 4-week interval in the range of WTs included in that study. The WTs covered by the studies ranged from 3 to 26 weeks. This figure serves to illustrate that all HRs were in the direction of increasing risk and that the HRs at different WTs were similar and therefore supports conversion of HRs from categories to an HR for a continuous WT representation.

For each study, a single HR corresponding to the relative increase in mortality risk with each additional 4 weeks of WT was extracted (Figure 2B). For studies using more than 2 categories of WT, the HR was estimated by the weighted HR per 4 weeks of each sequential WT category. For studies contrasting 2 WT categories, the line is the same as that presented in Figure 2A. The slopes of the lines in Figure 2B represent the HR from each study that was used in the meta-analysis.

Figure 3A presents the meta-analysis for OS, including HRs and 95% CIs for the 10 studies per 4 weeks of WT. The combined HR was 1.14 (95% CI, 1.10-1.17). The Cochran χ² test showed no evidence of heterogeneity (P = .30), justifying our use of the fixed-effects model.

The weight assigned to each study was influenced by the number of patients, event rate, and range of WT. The study by Cheung et al¹⁶ received the largest weight in this analysis, followed by that of Hershman et al¹⁸ and Lima et al¹⁹ ; after we sequentially removed these 3 largest studies and recalculated a combined HR estimate from the remaining studies, consistent and statistically significant results were maintained. The HR after removal of Cheung et al¹⁶ and Hershman et al¹⁸ was 1.15 (95% CI, 1.10-1.22), and then after removing Lima et al,¹⁹ the HR was 1.12 (95% CI, 1.05-1.21).

The funnel plot for the degree of asymmetry of the individual study results around the combined HR for OS is shown in Figure 4. The degree of asymmetry was not statistically significant by Egger method (P = .19). We used the trim and fill approach to adjust our estimate of effect size for potential asymmetry. The imputed estimate (HR, 1.13; 95% CI, 1.10-1.17) was similar to that in the main analysis, indicating that results are unlikely to be explained by publication bias.

The analyses were repeated for DFS (Figure 3B). Six of 10 studies reported the related end points of DFS,^{12 ,14} relapse-free survival,¹³ or cancer-specific survival^{16 ,18 - 19} and met the high-validity criteria. These 6 studies included 12 584 CRC patients. The combined HR was 1.14 (95% CI, 1.10-1.18). There was no evidence of heterogeneity (P = .51). The funnel plot gave a degree of asymmetry that was not statistically significant, and the trim and fill corrected HR estimate was similar to that in the main analysis (eFigure 1). Finally, a separate analysis specifically of the 3 studies that reported cancer-specific survival^{16 ,18 - 19} calculated an HR of 1.15 (95% CI, 1.10-1.19) similar to the combined estimate for OS (eFigure 2).

Adjuvant chemotherapy is well established as the standard of care for high-risk CRC. However, the published clinical trials on which this practice is based do not specifically inform about the timing of chemotherapy after surgery. Quiz Ref IDIn this report, our systematic review and meta-analysis indicates that relative OS decreases by 14% for every 4-week delay to initiation of AC. Our results are also consistent across DFS and cancer-specific survival analyses.

The available evidence that describes a relationship between time to AC and patient outcomes is limited mostly to retrospective analyses of patients with colorectal and breast cancers. The first report specifically addressing time to AC in CRC used Surveillance, Epidemiology, and End Results–Medicare data for 4382 patients with colon cancer, which found that a delay in time to AC beyond 3 months was associated with a significantly decreased cancer-specific survival and OS.¹⁸ Studies have based their time to AC analyses on a variety of different categorical WT comparisons.

Quiz Ref ID The effect of AC on survival is thought to be eradication of micrometastatic deposits in a proportion of patients who would otherwise be destined to have cancer recurrence. There is a substantial theoretical rationale to initiate AC promptly after curative surgery. Studies in animal models suggest that surgery may increase the numbers of circulating tumor cells and potentiate the growth of metastatic deposits. This increase in metastatic growth is thought to correlate with a reduction in angiogenesis inhibitors, such as angiostatin, following removal of the primary tumor.^{22 - 25} Surgery has also been shown to enhance production of oncogenic growth factors, such as transforming growth factor α, which can increase tumor growth.^{26 - 27} Furthermore, the classic mathematical model by Goldie and Coldman²⁸ predicts that the probability of mutations that lead to drug resistance increases over time, dependent on mutation rate and tumor size. Whether the more recent discovery of pluripotent colon cancer stem cells may also play a role in relapse following AC awaits further investigation.^{29 - 31}

Thus, preclinical tumor growth and kinetic models support a hypothesis that chemotherapy is most effective if initiated promptly when tumor burden is low. The combination of biological plausibility and an association between time to AC and decreased survival across published studies has lead to the widespread hypothesis that early initiation of AC is clinically important. Our meta-analysis has demonstrated an association between delay to treatment and adverse outcomes. In an attempt to infer causality, we note that the available evidence satisfies Hill's criteria.³² Indeed, evidence assessing levels of potential harm has a requirement for proof that is different from the evidence for treatments. With the knowledge that this time to AC–survival relationship will not be subjected to prospective assessment, we believe the level of evidence from this study provides sufficient evidence of causality.³³ Our findings therefore suggest that timing of AC plays a critical role in the management and outcomes of patients with CRC and that it would be prudent for clinicians and jurisdictions to avoid delays in access to chemotherapy.

Applying our findings to a patient who is ready to initiate AC 4 weeks after surgery but is delayed because of logistical rather than medical reasons, that same patient would have a 14% increased risk of mortality if treated at 8 weeks and 30% increased risk at 12 weeks. The following hypothetical example makes use of Adjuvant! Online³⁴ to illustrate the potential effect time to AC may have on patient outcome: a 65-year-old male patient in good general health, with T3N2 moderately differentiated colon cancer who is treated with fluorouracil-based chemotherapy has an estimated 5-year survival rate of 60%. If we assume this estimate depends on time to AC of 4 weeks (which is reasonable because Adjuvant! Online is based on clinical trials data, which have strict time-to-AC limits), then a delay to 8 weeks and 12 weeks would reduce 5-year survival to 54% and 48%, respectively. In perspective, the survival effect of prompt time to AC would be comparable with the magnitude of benefit seen with the addition of oxaliplatin to fluoropyrimidine chemotherapy in the adjuvant setting.^{35 - 36}

A second important aspect of time to AC that warrants elaboration based on our results relates to the commonly used (and arguably arbitrary) clinical parameter of a 3-month limit beyond which AC is often no longer recommended. Returning to the hypothetical patient, OS at 5 years without chemotherapy is approximately 45%; this increases to 60% with AC. Our results indicate survival of 48% if chemotherapy is administered at 12 weeks instead of 4 weeks, suggesting there may be some benefit to chemotherapy beyond a 12-week window. It is possible that a reasonable limit may be more in the order of 4 to 5 months.

A recent meta-analysis by Des Guetz et al³⁷ also investigated the effect of time to AC on mortality in patients with CRC. Consistent with our results, they observed a statistically significant increase in mortality risk among patients with delayed initiation of chemotherapy. The main analysis in the article by Des Guetz et al³⁷ focuses on a dichotomous WT of longer than 8 weeks vs 8 weeks or less. In contrast, we calculated an effect per each 4 weeks of delay, and as a result, we were able to use results over a broader range of WTs. We also included weighted estimates in our meta-analysis of the 10 studies included. Although there was considerable overlap in articles included, our study did not include 4 of the studies that were in the article by Des Guetz et al³⁷ (Berglund et al,³⁸ Gray et al,³⁹ André et al,⁴⁰ and Taal et al⁴¹ ) because these studies did not meet our inclusion criteria with respect to control for potential confounders. Our meta-analysis also included 3 studies not included in the article by Des Guetz et al³⁷ (Biagi et al,¹⁴ Lima et al,¹⁹ and Zeig-Owens et al²¹ ).

The relationship between WT and clinical outcomes for adjuvant radiotherapy has also been investigated. Preclinical models support a hypothesis that increased WT to adjuvant radiation may be detrimental. The only available evidence from the clinical literature is again indirect. In the only systematic review on this subject, Chen et al⁶ performed a meta-analysis on the effect of WT on local control rates for several cancers. The rate ratio of local recurrence was 1.14 per 4 weeks of delay (95% CI, 1.09-1.21), which was also consistent across specific cancers, including breast, head and neck, and sarcoma.

Although reasons for delays in the cancer system have not been well studied, many jurisdictions track and report on patient WTs. Our results indicate that at a population level, the effect of delays might be substantial. With approximately 140 000 new cases of CRC diagnosed in the United States in 2009,⁴² of which roughly 35% or 49 000 had stage III disease, the population at risk is sizeable. We leave it to readers to model the potential effect of delays within their jurisdictions.

Quiz Ref IDAlthough our analysis may overestimate or underestimate the effect of WT on survival, we believe the results of our study should inform policy for agencies that provide cancer services. The implication that time to AC may be as relevant to patient outcome as access to modern chemotherapy should encourage these agencies to streamline and coordinate the delivery of AC.

Our results need to be interpreted in context of the study's strengths and limitations. First, we acknowledge an inherent bias in the current analysis and in all time-to-AC analyses to date: the effect of a patient's postoperative performance status on time to AC. Poor patient performance status may independently prolong WT. On a theoretical level, postoperative stress may impair the immune system and stimulate acute phase cytokines, such as insulinlike growth factor binding protein 3 and matrix metalloproteinase 9, that may enhance cancer cell growth biology and kinetics.^{43 - 44}

Second, the completion rates for AC are not known, and this factor may not be balanced between the comparative groups. Third, our study analysis relies on the assumption of a log-linear relationship for the effect of WT on survival. Individual study results plotted in Figure 2 all trend upward over time, which supports this assumption, at least over the time periods of the reference studies. The assumption of linearity to this relationship may be problematic; for instance, it is unrealistic to extrapolate the 14% relative risk of death per 4-week delay indefinitely. It may not be suitable to extrapolate outside of the time period covered by those of the studies in our analysis. Our analysis does not allow us to comment on a potential effect of time to AC in the earliest few weeks of the postoperative period.

Fourth, our results are based on trials mainly from the era of fluoropyrimidine AC without oxaliplatin. Whether our results can be extrapolated to the current oxaliplatin era is not known. And fifth, our study relies on nonrandomized and retrospective data. However, it is unrealistic to expect that a randomized trial of time to AC will ever be done; rather, analyses such as ours are likely to provide the only evidence of such an effect. Indeed, it has been argued that the parameters applied to evidence that determines harm are not the same as evidence required to describe therapeutic benefit.⁴⁵ We therefore believe that the level of evidence provided by our analysis, combined with biologic plausibility, preclinical models, and clinical evidence, provides sufficient proof of a substantial negative relationship between prolonged time to AC and reduced patient outcomes.

In conclusion, our results demonstrate a significant adverse association between time to AC and survival in CRC, supporting a position that clinicians and jurisdictions need to optimize patient flow logistics to minimize time to AC. The results also suggest that the timing of chemotherapy may need to be more strictly controlled as a variable in future adjuvant trials. Furthermore, it is possible that AC beyond 12 weeks may still provide adjuvant benefit. Our results provide further validation of the intuitive concept of early time to AC. Physicians may need to more carefully consider timing when discussing AC with patients.