Survival outcomes associated with completion of adjuvant oxaliplatin‐based chemotherapy for stage III colon cancer: A national population‐based study

The impact of cycle completion rates of oxaliplatin‐based adjuvant chemotherapy for stage III colon cancer in real‐world practice is unknown. We assessed its impact, and that of treatment modification, on 3‐year cancer‐specific mortality. Four thousand one hundred and forty‐seven patients with pathological stage III colon cancer undergoing major resection from 2014 to 2017 in the English National Health Service were included. Chemotherapy data came from linked national administrative datasets. Competing risk regression analysis for 3‐year cancer‐specific mortality was performed according to completion of <6, 6‐11, or 12 5‐fluoropyrimidine and oxaliplatin (FOLFOX) cycles, or <4, 4‐7, or 8 capecitabine and oxaliplatin (CAPOX) cycles, adjusted for patient, tumour and hospital‐level characteristics. Median age was 64 years. Thirty‐two per cent of patients had at least one comorbidity. Forty‐two per cent of patients had T4 disease, and 40% had N2 disease. Compared to completion of 12 FOLFOX cycles, cancer‐specific mortality was higher in patients completing <6 cycles [subdistribution hazard ratios (sHR) 2.17; 95% CI 1.56‐3.03] or 6‐11 cycles (sHR 1.40; 95% CI 1.09‐1.78) (P < .001). Compared to completion of 8 CAPOX cycles, cancer‐specific mortality was higher in patients completing <4 cycles (sHR 2.02; 95% CI 1.53‐2.67) or 4‐7 cycles (sHR 1.63; 95% CI 1.27‐2.10) (P < .001). Dose reduction and early oxaliplatin discontinuation did not impact mortality in patients completing all cycles. Completion of all cycles of chemotherapy was associated with improved cancer‐specific survival in real‐world practice. Poor prognostic factors may have affected findings, however, patients completing <50% of cycles had poor outcomes. Clinicians may wish to facilitate completion with treatment modification in those able to tolerate it.


| INTRODUCTION
Adjuvant chemotherapy after a planned curative surgical resection for stage III colon cancer is an established treatment. 1,2 However, many patients do not complete the planned duration of chemotherapy, even in clinical trials, and in real-world practice this proportion is even higher with non-completion rates reported as high as 45%. 3,4 The impact of not completing adjuvant oxaliplatin-based chemotherapy on patient outcomes in real-world practice is unknown. FOLFOX and CAPOX have been shown in RCTs to improve outcomes compared to fluoropyrimidines alone. [5][6][7] After publication of this data, standard practice involved 6 months of adjuvant chemotherapy (8 cycles of CAPOX or 12 cycles of FOLFOX). However, longterm morbidity, in particular cumulative neurotoxicity associated with oxaliplatin-based chemotherapy, is concerning. 5 The recent IDEA collaborative study sought to establish the impact of reducing treatment duration by comparing 6 months treatment to 3 months (4 cycles of CAPOX or 6 cycles of FOLFOX). 8,9 Although the study failed to demonstrate overall noninferiority of a reduced target of 3 months of oxaliplatin-based chemotherapy, subgroup analysis suggested that 3 months of CAPOX, particularly in patients with low-risk disease, may be as effective as 6 months with reduced toxicity. The study found that those prescribed FOLFOX, or with high-risk disease, may still benefit from longer target durations. [8][9][10] Evidence comparing the efficacy of different target durations of adjuvant chemotherapy comes from high quality, large RCTs. 5,11,12 RCTs, however, include highly selected patient populations under rigorously controlled conditions, generally underrepresenting elderly, frail and comorbid patients. One study showed that 59% of stage II or III colon cancer patients in a real-world setting would not be eligible for RCT inclusion. 13 Population-based studies, using data such as electronic healthcare records, are needed to assess the effectiveness of actual durations of adjuvant chemotherapy on outcomes in diverse non-selected populations under routine clinical conditions to complement trial findings. [14][15][16][17][18] To date, observational studies evaluating the impact of cycle completion rates for oxaliplatin-based adjuvant chemotherapy on survival for colon cancer have been limited by a lack of accountability for important confounders, and their small sample size (most have fewer than 500 patients). 19 In addition, previous studies have not evaluated the survival impacts of treatment modifications (eg, dose reductions) which aim to reduce toxicity and support completion of the target duration of therapy.
In this national population-based study using linked administrative datasets, we assessed the impact of the cycle completion rate of oxaliplatin-based adjuvant chemotherapy on cancer-specific survival for stage III colon cancer patients treated in the English NHS, accounting for important confounders in the largest observational study to date. In addition, the effects of treatment modification on cancerspecific survival, namely dose reduction and early discontinuation of oxaliplatin, were analysed.

| Study population
Our study used NBOCA data, 20 HES-APC 21  Previously established methods were used to ascertain adjuvant chemotherapy receipt, regimen, and number of recorded cycles making use of the information in both SACT and HES-APC. 26 Patients receiving oxaliplatin-based adjuvant chemotherapy according to either SACT or HES-APC were included in the analysis ( Figure 1).

| Study outcome and comparison groups
The primary outcome was colorectal cancer-specific death within 3 years from the date of the first cycle of adjuvant chemotherapy.
Date and underlying cause of death were obtained from linkage to official death records provided by the ONS. 27 The date of the latest available death record was 10th February 2020, at which point follow-up times were censored.
Separate subanalyses were undertaken to evaluate two common treatment modification strategies: dose reduction and early discontinuation of oxaliplatin, both stratified by regimen. For these analyses, only patients completing 12 cycles of FOLFOX or 8 cycles of CAPOX, and with linked SACT records were included (3375 patients). Dose reduction is a binary (yes/no) variable within SACT which refers to dose reduction of "any anti-cancer drug administered at any point in the regimen after commencement of the regimen". 22 Discontinuation of oxaliplatin was derived from drug-level information. Inclusion criteria for adjuvant chemotherapy use • First recorded cycle of chemotherapy administered within 4 months of NBOCA date of surgery • Gaps of more than 3 months between the administration of consecutive chemotherapy cycles is assumed to represent a switch to different line of treatment (ie, palliative) • Chemotherapy cycles up to 9 months after the first recorded chemotherapy cycle are considered as adjuvant chemotherapy to allow for potential delays • Patients commencing nonstandard ACT regimens are assumed to be receiving palliative treatment • Patients switching on to nonstandard ACT regimens partway through treatment are assumed to have switched to a different line of treatment (ie, palliative) • Chemotherapy administration dates within 5 days of each other are considered to be the same cycle

| Statistical analysis
Patient, tumour and hospital-level characteristics were compared using Ҳ 2 tests stratified by chemotherapy regimen alone, and then by regimen and level of completion. The proportion of patients with a dose reduction and the proportion of patients discontinuing oxaliplatin early were reported according to regimen type and level of completion.
As our study evaluates survival outcomes in relation to the completion of chemotherapy, starting the analysis from initiation of chemotherapy may introduce bias as patients who die are unable to receive further cycles of chemotherapy. To account for this, a landmark analysis was undertaken. 33 This involved the designation of a period of time, a priori, from a baseline date (initiation of chemotherapy) to the study start date (the landmark date) known as the exposure period. Patients who died during the exposure period (6 months after chemotherapy initiation) were excluded from the analysis. A sensitivity analysis was undertaken to include those who died within this 6-month period.
The crude 3-year cumulative incidence of cancer-specific death was calculated for each regimen, according to the level of completion, using a competing risks analysis in which other-cause death was the competing event. 34  The same methodology was used to calculate unadjusted and adjusted sHRs for the risk of 3-year cancer-specific death in just those patients completing 100% of cycles, according to whether or not dose reduction occurred. This was then repeated for early discontinuation of oxaliplatin.
Missing values for risk-adjustment variables were imputed with multiple imputation using chained equations, creating 10 datasets and using Rubin's rules to combine the sHRs across the datasets. 36

| FOLFOX
Fifty per cent of patients completed 12 cycles (100%) of FOLFOX, 37% completed 6-11 cycles (50%-92%), and 13% of patients completed <6 cycles (<50%) ( Table 1). Patients completing the least FOLFOX chemotherapy were more likely to be female (P < .001), have a history of cardiac (P = .012) or renal disease (P = .042), undergo emergency surgery (P = .035), and commence chemotherapy within 8 weeks of surgery (P = .025). There was also a suggestion that patients who were from more deprived areas were less likely to complete chemotherapy, although this was not statistically significant (P = .073).  The adjusted competing risk regression analysis showed that the risk of 3-year cancer-specific death in patients completing <6 cycles or 6-11 cycles of FOLFOX was up to twice as high (P < .001) as those completing 12 cycles (Tables 2 and S1).  (Figure 2b).

| CAPOX
After adjustment, the risk of 3-year cancer-specific death in those completing <4 cycles or 4-7 cycles was up to twice as high (P < .001) as those completing 8 cycles (Tables 2 and S2).

| FOLFOX
In the 747 patients completing all cycles of FOLFOX, 47% had a dose reduction and 60% discontinued oxaliplatin early (

| CAPOX
In the 941 patients completing all cycles of CAPOX who had linked SACT data, 48% had a dose reduction and 37% discontinued oxaliplatin early (Table S3). The adjusted risk of 3-year cancer-specific death in those receiving a reduced dose was similar to those receiving the full dose, although the confidence interval was wide (sHR 0.92; 95% CI 0.62-1.34; P = .651) ( Table 3). The adjusted risk of 3-year cancer-specific death in those discontinuing oxaliplatin early was similar to those completing it although, again, the confidence interval was wide (sHR 0.84; 95% CI 0.55-1.28; P = .414) ( Table 3).
The sensitivity analyses including those patients who died within 6 months of their first chemotherapy dose did not show any significant differences in results (not presented).

| DISCUSSION
This is the largest cohort study of real-world practice to date evaluating cancer-specific survival according to the cycle completion rates of oxaliplatin-based adjuvant chemotherapy in stage III colon cancer patients, and the first to assess the impact of treatment modification strategies.
T A B L E 3 Three-year cancer-specific death according to dose reduction and early discontinuation of oxaliplatin for those completing 100% of FOLFOX (12 cycles) or CAPOX (8 cycles) This suggests that completion of adjuvant chemotherapy with treatment modifications rather than early cessation may confer survival advantages.

| Strengths and limitations
The main limitation of our study is that, although we found that the completion of adjuvant chemotherapy is associated with improved cancer-specific survival, we cannot simply assume a causal relationship. The factors which make patients less likely to complete their chemotherapy, for example, age and comorbidity, can also make them less likely to survive. However, the use of cancer-specific survival reduces the impact that these factors have on survival differences.
Whilst we have adjusted for many confounders (eg, age, comorbidity, performance status), we were unable to account for other causes of chemotherapy discontinuation, for example, patient preference, psychosocial support, and health behaviours. Despite this, the effect sizes seen are large and unlikely to be fully explained by residual confounding.
Second, duration of follow-up was limited by the availability of SACT data from 2014 onwards and cancer recurrence data was not available. The implications of this were that we reported 3-year cancer-specific survival. However, given our early event rate and the survival differences observed within this shorter timeframe, longer follow-up is only expected to accentuate our findings. In addition, approximately 80% of recurrences occur within the first 3 years after major resection. 37 The strengths of our study include using a large, contemporary and highly representative cohort of patients, which includes all centres providing colon cancer treatment in the English NHS (UK) without exclusions, and 95% of eligible patients. 20 The patient and clinical characteristics of our study are comparable to other observational studies with regards to staging, performance status, surgical urgency, and time from surgery to adjuvant chemotherapy initiation. [38][39][40][41][42] We have also overcome the biases present in previous observational studies by performing extensive risk-adjustment for important confounders. 19 The study period did not include SCOT trial patients and preceded publication of the IDEA collaborative results, meaning treatment duration reflects toxicity or intolerance, rather than patient or clinician choice informed by these results. 8,43 A landmark analysis was used to exclude patients who died within 6 months of their first chemotherapy dose (n = 75, 2%). 44 This was intended to account for immortal time bias; patients who died during the time they should have received chemotherapy would have been unable to complete treatment. Finally, patients within our cohort were analysed by the recorded number of chemotherapy cycles in a validated national curated chemotherapy dataset 26 rather than, for example, insurance or claims data. This has the advantage of using known individual chemotherapy administration dates compared to, for example, estimating completion based on the duration between the first and last claims for chemotherapy without taking account of individual cycles. 45

| Completion and survival
Our adjuvant chemotherapy completion rates of approximately 50% for FOLFOX and CAPOX are comparable to those from previous observational studies. 3,4 They are also plausible compared to the completion rates of 59% within the SCOT trial for both regimens, given that adherence rates within trial settings are known to be higher. 43 Of interest, completion of the least FOLFOX was associated with being female. There has been ongoing debate as to whether an underlying difference in toxicities exists with 5-FU due to gender. 46  However, the findings of this review were limited by the majority of studies failing to address important confounders such as chemotherapy regimen, age, sex, tumour site, and stage. We have systematically addressed these potential confounders in the current study. 19 In addition, many of the studies used outdated data and small sample sizes with the largest study available in abstract format only. 50 The IDEA collaborative study used intention-to-treat analyses to assess efficacy of two different target durations of adjuvant chemotherapy, whereas our study sought to evaluate the impact of actual completion rates on survival in real-world practice. 8

| Treatment modification and survival
For patients completing 100% of target cycles, we found that dose reductions or early discontinuation of oxaliplatin was not associated with any difference in cancer-specific survival. This has been observed in other studies, which have demonstrated that reduced dose intensity may not negatively influence survival. 53,54 This may reflect the relative importance of the fluoropyrimdine component of treatment and the uncertain effect in patients aged 70 and above who constitute a quarter of our cohort. 55,56 We found that between 40%-60% of patients discontinued their oxaliplatin (depending on regimen), but continued on a single agent fluoropyrimidine for the rest of their cycles.
Further analysis of the impact of dose reductions or oxaliplatin discontinuation in low-and high-risk prognostic groups was limited by small numbers (reflected in the wide confidence intervals), and an inability to quantify exact dose reductions.

| Implications for policy and practice
Current recommendations advise 12 cycles of FOLFOX and suggest that 4 cycles of CAPOX can be used dependent on other risk factors, particularly staging. 57 Our findings suggest that in real world practice, once combination chemotherapy has been commenced, completion of at least 4 cycles of CAPOX or 6 cycles of FOLFOX confers a survival advantage over early discontinuation.
Unless toxicities are very severe, our data suggests that patients may benefit from attempting to complete adjuvant chemotherapy with treatment modifications. Improved strategies to support completion of chemotherapy might include prompt identification and management of chemotherapy-related adverse effects, clear clinician-patient communication and education, and provision of adequate support to overcome any physical or psychosocial barriers.

| CONCLUSION
Our study demonstrated that in real-world clinical practice only half

CONFLICT OF INTEREST
The authors have declared no conflicts of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of our study are available from the corresponding author upon reasonable request after permission of HQIP.

ETHICS STATEMENT
All patient data used are fully anonymised and are therefore exempt from United Kingdom National Research Ethics Committee approval.

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher's website.