Always look at the bright side of drugs?

I remember a research seminar at our department many years ago. Data from a cohort of patients with metastatic breast cancer treated with high-dose chemotherapy and autologous bone marrow transplantation were presented. An unprecedented long-term survival of more than 30% was observed. There was a feeling that we had witnessed a breakthrough in the treatment of metastatic breast cancer. However, there were some people who were more skeptical and started randomised trials that showed the new treatment concept to be no better than standard treatment [1]. The unprecedented data were reasonably due to patient selection.

Almost every speaker at cancer conference plenary sessions discussing drug treatment of metastatic colorectal cancer (mCRC) display in their introduction a slide showing the evolvement of median overall survival (OS) times as new drugs have been introduced to treat this disease. Before chemotherapy the median OS was 6–8 months, increasing to 12 months with use of 5-FU, to 16–20 months with use of 5-FU combined with irinotean or oxaliplatin and to 24–30 months by also adding new targeted drugs. Unexplained, however, is the fact that the pivotal randomised clinical trials that investigated the benefit of these treatment steps do not allow for a conclusion that drugs prolong median OS with up to 18–24 months as indicated by this standard slide. Furthermore, improvements seem to apply mainly for subgroups of patients [2,3]. The improvement illustrated is thus reasonably explained also by other factors that have changed over time and that contribute to the better outcome observed, e.g. patient selection, surveillance strategies, intensity in cancer treatment, better general care and stage migration.

The pivotal randomised clinical trial investigating the benefit of adding the VEGF-antibody bevacizumab to oxaliplatin-based chemotherapy in mCRC showed no effect in OS and only a very modest improvement in progression-free survival (PFS) [4]. However, subgroup analysis indicated that patients with prolonged use of bevacizumab seemed to benefit more. A benefit from prolonged use of bevacizumab was also indicated by the observational cohort BRiTE study showing an impressive median OS of 32 months in patients continuing bevacizumab beyond first line treatment progression [5]. However, the randomised trial that subsequently investigated this strategy showed a very modest median OS benefit of 1.4 months [6]. Again, the promising subgroup and observational data were reasonably due to a selection effect.

Recently I listened to a presentation of data from the new EPOC trial showing, unexpectedly, that the addition of the EGFR antibody cetuximab to Folfox in the neoadjuvant treatment of resectable liver metastatic CRC significantly shortened PFS compared with chemotherapy alone [7]. The presenter was very rapid to display a slide listing a number of potential confounders and factors considered to have introduced bias to the results, which, therefore, implicitly, should be disregarded [8]. I strongly doubt that such a list would have been displayed if the results had pointed to a benefit from the addition of the antibody.

I suggest that examples like these are kept in mind when conclusions are to be drawn from the two cohort studies addressing the potential benefit of bevacizumab in combination with standard chemotherapy in first line treatment of mCRC published in this issue of Acta Oncologica.

Stein et al. included 1777 patients at 261 sites in Germany in their prospective observational cohort trial [9]. The patient cohort had good prognostic characteristics with respect to age, performance status and number of metastatic sites. In total 82% of the patients had bevacizumab combined with 5-FU plus oxaliplatin or irinotecan and the median treatment duration was seven months. Following induction, patients could have de-escalated ‘maintenance’ treatment with 5-FU plus bevacizumab or the antibody alone. The tumour response rates ranged 52–65% and median PFS was approximately 10.5 months in the oxaliplatin and irinotecan combination groups and slightly shorter, 9.2 months, in patients having bevacizumab with 5-FU only. Median OS ranged 25–27 months and was approximately 29 months in patients on the de-escalation strategies. No data were provided on further line treatments and no unexpected safety problems were observed. The authors cautiously conclude that the efficacy and safety profile of bevacizumab plus chemotherapy in first line treatment of mCRC is consistent with that observed previously in similar cohort studies and in phase III trials.

Hammerman et al. more directly addressed the role of bevacizumab added to combination chemotherapy by comparing retrospective outcome registry data from two cohorts of mCRC patients treated within Israel's largest healthcare organisation; 687 patients treated with combination chemotherapy 2000–2004 when bevacizumab was not available and 1052 patients treated 2006–2009 with chemotherapy plus the antibody [10]. Key baseline prognostic data were not available but sex, age and comorbidity score were similar in the two cohorts. However, they differed with respect to type of first line chemotherapy and use of second line therapy and EGFR inhibitor in latter line therapy, with the bevacizumab cohort being more intensively treated. Median OS was 15 months in the chemotherapy alone cohort compared with 23 months in the cohort also being treated with bevacizumab. PFS was also longer, 14 compared with 10 months. Following acknowledgement of the limitations of the study, the authors conclude that these ‘real-life’ data support benefit from the addition of bevacizumab to first line treatment of mCRC.

My conclusion is that the studies show that a median OS of two years or more is possible to achieve outside prospective randomised clinical trials in cohorts of comparatively young mCRC patients selected for being tolerant to intensive therapy including bevacizumab. However, in contrast to Stein et al. and Hammerman et al., in my mind the data presented do not allow for any conclusions regarding the contribution of bevacizumab to this favourable outcome. The reasons for my more conservative view are, as indicated by the examples above, study design issues related to patient selection and changes over time in diagnostic and treatment strategies in addition to the use of bevacizumab.

As long as new cancer drugs cannot be expected to provide more than small incremental improvements in OS, ‘real-life’ cohort data just do not allow for conclusions on the efficacy of these drugs. Unfortunately, the detection of the small benefit from new drugs in intermediately drug sensitive solid tumours like mCRC requires the randomised trial design that removes the major sources of bias. This does not contradict my conclusion that ‘real-life’ cohort data on cancer drug treatment are justified to publish to reflect time-trends in treatment strategies used, the associated changes in outcome and, not the least, to identify safety issues not possible to detect even in the largest prospective randomised trials.

So why is it that data on the effect of cancer drugs from clinical trials and patient cohorts are so frequently over-interpreted to boost the effects observed [11], and why is it that preclinical cancer research findings that should provide the basis for development of new more active drugs are frequently not possible to reproduce and use for this development [12]? Well, my general and very simplistic view is that scientists cope with the strains of life in the same way as everyone else and to be overly optimistic is clearly a major coping strategy; we tend to see light also when there are mostly shades of grey.

Declaration of interest: The author reports no conflicts of interest. The author alone is responsible for the content and writing of the paper.