Addressing the variation in adjuvant chemotherapy treatment for colorectal cancer: Can a regional intervention promote national change?

Analysis of routine population‐based data has previously shown that patterns of surgical treatment for colorectal cancer can vary widely, but there is limited evidence available to determine if such variation is also seen in the use of chemotherapy. This study quantified variation in adjuvant chemotherapy across both England using cancer registry data and in more detail across the representative Yorkshire and Humber regions. Individuals with Stages II and III colorectal cancer who underwent major resection from 2014 to 2015 were identified. Rates of chemotherapy were calculated from the Systemic Anticancer Treatment database using multilevel logistic regression. Additionally, questionnaires addressing different clinical scenarios were sent to regional oncologists to investigate the treatment preferences of clinicians. The national adjusted chemotherapy treatment rate ranged from 2% to 46% (Stage II cancers), 19% to 81% (Stage III cancers), 24% to 75% (patients aged <70 years) and 5% to 46% (patients aged ≥70 years). Regionally, the rates of treatment and the proportions of treated patients receiving combination chemotherapy varied by stage (Stage II 4%‐26% and 0%‐55%, Stage III 48%‐71% and 40%‐84%) and by age (<70 years 35%‐68% and 49%‐91%; ≥70 years 15%‐39% and 6%‐75%). Questionnaire responses showed significant variations in opinions for high‐risk Stage II patients with both deficient and proficient mismatch repair tumours and Stage IIIB patients aged ≥70 years. Following a review of the evidence, open discussion in our region has enabled a consensus agreement on an algorithm for colorectal cancer that is intended to reduce variation in practice.

≥70 years. Following a review of the evidence, open discussion in our region has enabled a consensus agreement on an algorithm for colorectal cancer that is intended to reduce variation in practice. European Society for Medical Oncology (ESMO) guidance. [7][8][9] More recently the duration of treatment has been further shortened to 3 months; the evidence strongest when using oxaliplatin and capecitabine (CAPOX) but the shorter course can still be considered for single agent capecitabine or oxaliplatin and 5FU (FOLFOX). 10,11 Again, ASCO and NCCN guidelines quickly adopted these changes as a new standard of care for all except patients with high-risk Stage III disease where 6 months is still recommended. More recently NICE have recommended 3 months for all patients when using CAPOX. 12 Beyond stage, benefits of adjuvant chemotherapy may differ by molecular phenotype, patient age and site of disease. Microsatelite instability is present in 15% of colorectal cancers and is a result of deficient mismatch repair (dMMR) caused by a germline mutation in an MMR gene or epigenetic inactivation of the MLH1 gene. 13 Although dMMR is a good prognostic factor, especially in Stage II disease, data on whether it is predictive treatment efficacy are mixed. 14,15 In terms of increasing age, data are mixed for single-agent fluoropyrimidine; the QUASAR trial suggested diminishing benefit whereas a pooled analysis of the ACCENT database showed benefit was preserved in patients over 70 years. 16,17 Data on oxaliplatin are more consistent with no benefit found for patients over 70 years. 18,19 Age-related variance in benefit may relate to both an ability to tolerate chemotherapy and differences in biology of disease with increasing age. 20 The evidence for a benefit of adjuvant chemotherapy for rectal cancer patients is inconclusive. Following a review of the evidence, ESMO concluded that the benefit of adjuvant 5FU following surgery alone was smaller than for colon cancer. Randomized trials have not shown benefit for adjuvant 5FU for patients who have received neoadjuvant radiotherapy and no consistent survival benefit has been shown for the addition of oxaliplatin to 5FU either during chemoradiotherapy and/or in the adjuvant setting. [21][22][23][24][25] Two studies exploring total neoadjuvant treatment recently reported improved 3-year disease-free survival or disease-related treatment failure, but data are too early to assess impact on survival. 26,27 International bodies have provided guidance in these areas. The NCCN and ESMO guidelines support testing and recommend observation alone for dMMR high-risk pT3 stage II disease. 8  Here, the authors delved into the population data to quantify the variation in adjuvant chemotherapy across England. They found a surprisingly high variation in chemotherapy rates for high-risk stage II patients, and they suggest this difference results in part from clinicians' differing opinions about the effectiveness of chemotherapy for these patients. By highlighting the amount of regional variation in treatment practice, they achieved a consensus agreement on an algorithm they developed to provide standardized guidance for the use of chemotherapy.
options are considered. This study aimed to quantify any variation in adjuvant chemotherapy treatment across England and to explore this variation in greater detail using the large representative region of Yorkshire. This evidence was then used to help develop a consensus guideline for implementation across the region's 16 MDTs in an effort to minimize variation and improve colorectal cancer outcomes.

| Study design
The study was designed as part of the Yorkshire Cancer Research Bowel Cancer Improvement Programme (YCR BCIP); a regional interventionbased programme that is aiming to significantly improve colorectal cancer outcomes across a large representative region (Yorkshire and the Humber). It aims to do this by quantifying variation in practice and engaging with regional MDTs to understand this and develop educa-  Figure S1). Variation in the use of adjuvant chemotherapy was investigated at a national level, using population-based data across all colorectal cancer MDTs treating patients in England. A more detailed account of the variation was then undertaken at a regional level, covering the MDTs participating in the YCR BCIP. Simultaneously in a separate qualitative study, the first of two rounds of questionnaires addressing the use of adjuvant chemotherapy for CRC were sent to regional oncologists.
A series of face-to-face and teleconference meetings for all oncologists from the YCR BCIP region were held to discuss the variance both in prescribing practice seen in regional and national data and the approaches taken by different MDTs seen from the questionnaire responses. Subsequently a further round of questionnaires was sent to assess if differences persisted and a further meeting to agree a final treatment algorithm, for use across the 16 MDTs in the region, with the aim to reduce treatment variation. Service. Through its data repository COloRECTal Repository (CORECT-R), the UK Colorectal Cancer Intelligence Hub provides linkage of these cancer registry data to a number of additional routine data sets across the English National Health Service, including hospital admission data and the systematic anticancer therapy (SACT) data set. 29 Submission of SACT data via electronic prescribing is mandatory for all NHS-funded providers in England and includes all cancer patients receiving systemic anticancer treatment. All patients who underwent a major resection and had not received neoadjuvant radiotherapy treatment were identified using previously described algorithms. 30,31 Patients were assigned a managing colorectal MDT using the hospital admission procedure closest to the patient's diagnosis date. If no procedure was found, the closest inpatient or outpatient appointment to the diagnosis date at a hospital with a colorectal MDT was used. The <1% of patients that could not be assigned an MDT were excluded from the study. In addition, 3% of all patients in the cancer registry data had an unknown stage of disease so it was not possible to ascertain if any of these were Stage II or III cancer and will have been excluded from these analyses. Two MDTs, including one within the YCR BCIP region, were found to have not submitted SACT data in the adjuvant setting at the time of data collection so were excluded from the analysis.

| Population-based data
Patients receiving chemotherapy treatment were identified in the SACT data set if their first regimen after resection was within 6 months and the primary treatment diagnosis within SACT was confirmed as C18-C20. Those receiving the combination regimens of CAPOX or FOLFOX and the single agent regimens of capecitabine or 5FU were classified as receiving adjuvant treatment. Patients receiving a regimen usually used for metastatic disease were assumed to have progressed to an advanced disease stage and excluded (3% of those receiving chemotherapy). Patients receiving nonstandard regimens, commonly used for treatment of different cancers, were assumed to have been coded incorrectly and excluded (<1% of those receiving chemotherapy).
Multilevel logistic regression models were used to assess factors associated with adjuvant chemotherapy treatment, treating MDT as a random effect. The binary-dependent variable was set as whether the patient received adjuvant chemotherapy or not. Analyses were stratified by age and tumour stage, with the following covariates consid-   Figure 1A) and from 19% to 81% for Stage III patients ( Figure 1B). When stratifying by age, the treatment rate ranged from 24% to 75% for patients aged <70 ( Figure 1C) and from 5% to 46% for patients aged ≥70 ( Figure 1D).

| Assessment of regional variation
The variation seen at national level was mirrored within the YCR BCIP region (n = 2375), with overall rates of adjuvant treatment similar to those outside the region (Supplementary Table S1).
Although a number of MDTs showed significantly outlying adjusted rates for both Stage II and III diseases, this was most prominent for Stage II (Figure 1). Variation was seen in the observed rates of patients receiving treatment (range: Stage II 4%-26%, Stage III 48%-71%). Combination therapy was mostly higher for Stage III patients (range: 40%-84%) but not for Stage II patients (range: 0%-55%) (Figure 2).

| Further investigation of regional variation
Responses to Round 1 (Stage II and III questionnaires) were received from oncologists at 15 of the 16 regional MDTs. Responses to Round 2 (Stage II, Stage III and rectal cancer after neoadjuvant radiotherapy questionnaires) were received from oncologists at 10, 10 and 13 of the 16 MDTs, respectively. All but one regional MDT partook in at least one of the questionnaires. MDT representation for each individual questionnaire can be found in Supplementary Table S2.

| Stage II disease
In Round 1 of the questionnaire (Figure 3), for patients displaying lowrisk features, the 15 oncologists were largely in agreement, with at least 11 (73%) selecting surveillance at the treatment option for all sets of patients <70 years. No oncologist reported chemotherapy as a treatment option for those low-risk feature patients aged ≥70.
There was less agreement for patients displaying high-risk features; for dMMR patients aged <70 years, clinicians were split between the use of combination chemotherapy and observation; for pMMR patients, clinicians were mostly split on whether they are single agent or combination. With increasing age, clinician's decision making became more conservative.

| Stage III disease
In Round 1 of the questionnaire (Figure 4), for all Stage III patients aged <70 years there was uniformity among the 15 oncologists for recommending combination chemotherapy for pMMR patients, bar one who recommended single agent treatment for those aged 60 to 69 years.
In Round 2 (Figure 4), oncologists generally displayed more agreement but were still split on single agent (7/11, 64% in IIIA and 6/11, 55% in IIIB) or combination treatment (4/11, 36% in IIIA and 5/11, 45% in IIIB) for pMMR patients aged ≥70. information, three (25%) reported that they used just only preoperative cross-sectional imaging and one (8%) only postoperative histological stage. Six (50%) oncologists reported they had a higher threshold for using adjuvant chemotherapy for this group than for patients with colon cancer and six also reported they did not consider single-agent FU chemotherapy worthwhile for this patient group. Six (50%) oncologists reported they predominantly restrict adjuvant chemotherapy to those under age 70. For patients with pathological complete response, three (25%) reported that they would consider adjuvant chemotherapy based on clinical stage at diagnosis.

| Treatment guideline
After the discussion among the region's oncologists at the consultation meetings in which the regional variation and previous published evidence were discussed, a treatment algorithm for Stage II and III colorectal patients was constructed and proposed as a guideline for regional MDTs at various stages during the study, before the finalised version was agreed upon ( Figure 5) We also did not include a specific age cut point. Sporadic dMMR occurs more commonly in the elderly and the algorithm recommends a more conservative approach in the absence of T4 or node positive disease for this molecular type. For those with more advanced T or N stage combination chemotherapy at full dose or observation is recommended on the understanding that comorbidity and performance status will be taken into account in accordance with NICE guidance.
Hence, an older patient is less likely to be treated given the higher rate of dMMR, comorbidity and frailty related lower performance sta- ing to treatment decisions such as comorbidities not recorded through the Charlson comorbidity index. However, the additional data collection in this study compensates somewhat for these omissions and emphasises the variation shown in national data sets. Patient choice is also likely to be factor here, but it was not possible to account for this in the data used in this study. Previous studies have demonstrated that an increased travel burden is associated with a decreased likelihood of receiving adjuvant chemotherapy. 37,38 It was not possible to assess this within our data, however, a systematic review in the United Kingdom concluded that while variation between healthcare boundaries were observed, other factors such as capacity and treatment policy were more influential than geographical factors. 39 The patient scenarios in the questionnaires used age as a patient group. Although in some cases this may be a suitable proxy for frailty and fitness to undergo chemotherapy treatment, discussions among the regional oncologists made it clear that a more appropriate measure would be ideal. The algorithm has used the relatively bland term "fit for chemotherapy" based on the expectation that clinicians will factor a patient's comorbidity, performance status and preference during decision making in keeping with NICE guidance. We recognize that ASCO have released guidance on assessment of older patients receiving chemotherapy. 40 The tools highlighted in this guidance document have been developed largely in patients with advanced disease and are not tumour site specific. [41][42][43] There is a need to develop a tool to help predict which patients are likely to tolerate and complete adjuvant chemotherapy for colorectal cancer. The linked datasets accessed through this work provides an opportunity to further explore this area.
The algorithm developed here provides general guidance to adjuvant treatment; however, it is not expected to provide a definite stratification of patients as it is not possible to take in account all patient characteristics at such a level. Care must be taken if extending the algorithm for use in other populations, for instance, this study used National Health Service data where access to treatment is good and therefore it was not necessary to take into account factors such as health insurance status and patient access.
In summary, this study has identified considerable variation in the management of adjuvant chemotherapy for colorectal cancer at regional and national levels. Bringing this information to the attention of clinicians through the YCR BCIP enabled a consensus agreement on a proposed algorithm to reduce treatment variation across a large representative region of England. Such a process could be replicated in other regions.