Between March 9, 2005, and May 9, 2008, 2445 patients were randomly allocated to treatment groups at 111 centres in the UK and Ireland, with 1630 patients assigned to the comparison of continuous versus intermittent treatment. shows their baseline characteristics, which were well balanced between the trial arms. shows the progress of patients through the stages of the trial. At the time of analysis (database locked Sept 2, 2009), the median duration of follow-up among surviving patients assigned to arm A was 20·9 months (IQR 16·1–28·3) and to arm C was 21·8 months (16·2–29·5). 28 (7%) of 383 surviving patients had no data returned for 12 months and were deemed lost to follow-up.
35 (2%) patients did not start trial therapy because of clinical deterioration after randomisation, patient choice, or subsequent ineligibility after randomisation. Two-thirds of all patients randomly assigned to treatment groups (536 in arm A, 533 in arm C) received capecitabine-based chemotherapy, and 34% (279 in arm A, 282 in arm C) received fluorouracil-based therapy. No major differences were identified by choice of chemotherapy for any of the outcome measures of this comparison. In the per-protocol population, the geometric mean overall time on treatment in arm C was 5·2 months (95% CI 5·0–5·4) versus 7·5 months (7·3–7·8) in arm A (p<0·0001). Although total dose delivered of chemotherapeutic agents was greater in arm A than in arm C (p<0·0001), dose intensity was greater in arm C during on-treatment periods (p<0·0001). In the intermittent treatment group, 325 (64%) of 511 potentially eligible patients restarted a second 12-week course of chemotherapy after a chemotherapy-free interval.
1247 (77%) deaths had occurred in the ITT population at the time of analysis. Median survival in the ITT population was 15·8 months (IQR 9·4–26·1) in arm A and 14·4 months (8·0–24·7) in arm C, and in the per-protocol population was 19·6 months (13·0–28·1) in arm A and 18·0 months (12·1–29·3) in arm C. The HR point estimates were 1·084 (80% CI 1·008–1·165) in the ITT population and 1·087 (0·986–1·198) in the per-protocol population; the upper limits were higher than the predefined non-inferiority boundary (). Kaplan-Meier overall survival in the ITT population for arm A versus arm C was 28·7% versus 26·5% at 2 years and 13·0% versus 11·2% at 3 years.
Kaplan-Meier curves for overall survival in (A) the ITT population and (B) the per-protocol population, and strategy-failure-free survival in (C) the ITT population and (D) the per-protocol population
1166 (94%) of 1247 deaths in the ITT population were due to colorectal cancer. 22 (2%) deaths were reported to be treatment-related, 52 (4%) were from other causes, and the cause of death for the remaining seven (<1%) is currently unknown. 70 (6%) deaths occurred within 60 days of randomisation, with no difference between the two groups.
Median strategy-failure-free survival in the ITT population was 8·4 months (IQR 5·3–11·8) in arm A and 7·4 months (5·1–12·1) in arm C, and in the per-protocol population was 9·2 months (7·0–12·9) in arm A and 8·7 months (5·9–13·4) in arm C. From the Kaplan-Meier plots (), intermittent therapy seemed to result in a greater loss of patients (in terms of strategy) early on, especially around 6–9 months, but at 2 years there was little difference in number of patients remaining on strategy: 2-year survival in the ITT analysis was 7·4% (n=28) in the continuous treatment arm and 7·6% (n=35) in the intermittent treatment arm, and in the per-protocol analysis was 7·7% on continuous (n=18) and 7·6% on intermittent (n=24) treatment.
Among patients in the per-protocol population allocated intermittent treatment who commenced a chemotherapy-free interval (n=511), the median time to progression was 12·9 weeks (3·0 months; IQR 10·9–24·0 weeks). Among those who later restarted trial therapy (n=325), the median overall length of the chemotherapy-free interval before progression was 16·0 weeks (3·7 months; IQR 13·7–26·0 weeks). The median number of chemotherapy-free intervals was two (range one to six).
Protocol treatment in both treatment groups was identical up to 12 weeks (the time of the first scheduled radiological disease assessment). 192 (12%) patients in the ITT population had progressive disease at 12 weeks (). Best overall response (by RECIST criteria) did not differ between treatment groups, suggesting that most patients achieve best response within the first 12 weeks of therapy. Among the 268 patients allocated intermittent treatment who restarted a second 12-week course of chemotherapy after a chemotherapy-free interval and had a second response assessment 12 weeks later, 87 (32%) had a partial response, 101 (38%) had stable disease, and 80 (30%) had progressive disease.
Response at 12 weeks and best response in arms A and C and response after rechallenge (arm C only)
Haematological toxic effects and hand–foot syndrome were more common on continuous treatment than on intermittent treatment, whereas nausea and vomiting occurred more often in patients receiving intermittent treatment. Grade 3 or worse peripheral neuropathy and hand–foot syndrome were more frequent during continuous treatment (webappendix p 4
There were fewer eligible patients in arm C (79%, n=647) than in arm A (89%, n=724; p<0·0001), since more patients in arm C either remained on trial therapy (6% vs 1% in arm A, n=46 vs nine, p<0·0001) or died (14% vs 9% in arm A, n=115 vs 74, p=0·0015). Among patients judged eligible to receive second-line therapy, this treatment was given to significantly fewer patients on intermittent therapy (52%, n=336) than on continuous therapy (62%, n=448; p=0·00020).
Quality of life in the per-protocol population at baseline, 12 weeks, and 24 weeks was assessable in 279 (60%) patients on continuous treatment and 282 (55%) patients on intermittent treatment. Less than two-thirds of patients complied with the quality-of-life component of COIN. No clear evidence of differences in baseline characteristics was identified between those patients who completed quality-of-life questionnaires and all patients randomly allocated to treatment groups, but the risk of such differences cannot be eliminated. After adjustment for previous measurements, there were significant benefits from intermittent therapy at 24 weeks for fatigue, dry or sore mouth, problems eating and drinking, difficulty handling small objects, interference with daily activities, nausea or vomiting, appetite loss, constipation, and diarrhoea (; p<0·05 for all). Intermittent therapy also had benefits in terms of role functioning (p=0·015) and social functioning (p=0·016; ). By contrast, the only significant factor suggesting detriment from intermittent therapy was the symptom scale of pain (p=0·00029; ). At 12 weeks, there was a non-significant trend towards an adverse effect of intermittent therapy on emotional functioning (p=0·086), but this result was not evident at 24 weeks (p=0·90; ).
Quality of life at 12 and 24 weeks
We undertook a post-hoc exploratory analysis of the effects of protocol adherence on overall survival. Clinicians were classified according to the proportion of their patients assigned to intermittent therapy in the per-protocol population who restarted treatment after a chemotherapy-free interval. A cutoff of 60% was chosen because this proportion formed a pragmatic and appropriate benchmark midway between the 40% and 80% of restarts reported in OPTIMOX-1 and OPTIMOX-2, respectively. For so-called adherent clinicians, there was no survival difference between arms A and C. But for non-adherent clinicians, there was an early survival advantage for arm A. However, this difference lessens over time and the overall HR is non-significant ().
Effects of adherence to protocol on overall survival within the per-protocol population
15 prespecified factors were entered into a prognostic model to predict overall survival in the per-protocol population, together with the post-hoc factor of restart adherence rate. The final model contained the following previously identified prognostic variables: previous adjuvant chemotherapy or surgery; alkaline phosphatase; body surface-area; platelet count; time from diagnosis to randomisation; resection of primary tumour; mutation present in any of KRAS
; and number of metastatic sites. In particular, each of the four variables identified by Köhne and colleagues13
were present with the exception of white blood cell count (in our data, platelet count was a stronger prognostic indicator), and the score defined by Köhne was strongly prognostic (p<0·0001 for trend). The 16 factors entered into the prognostic model were also explored for predictive value for overall survival in the per-protocol population (). The strongest predictive factor, and the only one with an interaction significant at the 5% level, was platelet count. A raised platelet count at baseline (≥400 000 per μL, recorded for 271 [28%] of 978 patients) predicts a significant survival detriment from intermittent chemotherapy (p=0·0027 for interaction; ). Raised baseline platelet count had an even greater predictive effect for a detriment from intermittent therapy on 24-week quality of life, particularly for functional scales (p<0·05 for all), for which a consistent trend was noted towards benefit from intermittent therapy for patients with normal platelet count and an adverse effect for patients with raised platelet count. Additionally, the symptom scales of fatigue, pain, dyspnoea, appetite loss, constipation, and global quality of life followed this trend (p<0·05 for all). Furthermore, patients on intermittent therapy were significantly more likely than were those on continuous therapy to report at 24 weeks that their treatment had not been worthwhile if their baseline platelet count was raised (p=0·048 for interaction; webappendix p 7
). All these results were independent of known prognostic factors. Baseline platelet count and progression between 12 and 24 weeks were not associated in this dataset (p=0·41). Other factors with interactions significant at the 10% level were presence of metastases in the liver only (p=0·066) and tumoral KRAS
Subgroup analyses of overall survival within the per-protocol population
Kaplan-Meier curves of overall survival within the per-protocol population, by baseline platelet subgroup