Publishing in the journal Diabetologia a group of researchers have compared gastric surgery with lifestyle and medical interventions for the treatment of type 2 diabetes. In the trial with 32 patients the authors report that the group which received surgery saw greater remission of diabetes. Roundup comments accompanied this analysis.
Title, Date of Publication & Journal
‘Gastric bypass surgery vs intensive lifestyle and medical intervention for type 2 diabetes: the CROSSROADS randomised controlled trial’ by David E. Cummings et al will be published in Diabetologia at 23:01 UK time on Thursday 17 March 2016.
Study’s main claims – and are they supported by the data
The study does provide some evidence that the Roux-en-Y gastric bypass (RYGB), a type of bariatric surgery, can be more effective than some intensive lifestyle and drug interventions at achieving remission of type 2 diabetes in mild to moderately obese patients, but this is a small study and there are some issues with the randomisation (see limitations section below).
Diabetes remission after one year (the primary outcome) was seen in 9 out of 15 patients who underwent RYGB versus 1 out of 17 without surgery; that is 60% versus 5.9%. Whilst this is a large effect size, smaller differences between the groups were observed in secondary outcomes.
The mean HbA1c values (glycated haemoglobin; a measure of blood sugar and diabetes status) were not actually statistically different between the groups at any point of the trial. The authors report that “The HbA1c decline over 1 year was only modestly more after RYGB than ILMI: from 7.7 ± 1.0% (60.7 mmol/mol) to 6.4±1.6% (46.4 mmol/mol) vs 7.3±0.9% (56.3 mmol/mol) to 6.9±1.3% (51.9 mmol/mol)”. Since diabetes remission was defined as HbA1c <6.0% and being off all diabetes medicines, the large difference seen in this primary outcome is mostly down to the fact that the surgical group used fewer medications at 1 year (mean 0.5) in comparison to the non-surgical group (mean 1.2). From the numbers it follows that 9 of 15 surgery patients ended the trial off medication with their HbA1c below 6.0%, but the mean HbA1c seen in the surgical group was 6.4% (SD of 1.6). These summary statistics tell us that most of the remaining 8 surgery patients actually had increased their HbA1c – a huge spread of outcomes that is not addressed in the study.
The authors conclude that “the longstanding practice of using strict BMI cut-offs as the primary criteria for selection for bariatric surgery among patients with type 2 diabetes” should be questioned. Neither baseline BMI nor weight loss were found to predict diabetes remission in the study. However the lack of statistical significance between BMI or weight and diabetes remission here could just reflect the small size of the study. The study was not specifically designed to detect these associations and was relatively small – 43 patients were randomized but only 32 were actually analysed. There is therefore not sufficient evidence to support the claim that it may no longer be appropriate to consider someone for bariatric surgery among patients with type 2 diabetes based primarily on just their body mass index.
The authors additionally conclude that the trials results confirm and extends findings from previous studies that RYGB surgery is safe with no greater risk for those with a BMI<35 kg/m2 compared to those with greater BMI. But there were only 5 surgery patients with a BMI<35 kg/m2 in the study. More data is therefore needed to confirm this claim.
Participants were randomised, which increases confidence in results – but see caveat below.
Subjects were well-matched at baseline in most reported characteristics – but see caveat below.
Although participants were randomised, the statistical analysis was flawed because not all randomised patients were included in the analysis. 8 out of 23 (35%) were not included in the surgery group for analysis and 3 out of 20 (15%) were not included in the non-surgery group, since they did not start either intervention. Having excluded 8 surgical and 3 non-surgical patients from analysis after randomisation the benefits of randomisation are essentially lost. Further unmeasured characteristics could consequently be imbalanced between the treatment groups, reducing the reliability of these results. No information is given about these 11 missing patients – if they differed in some important way to the included patients this could drastically change the results, especially if this was related to why they didn’t start the intervention.
Patients from one medical centre who did not have a preference for one treatment over the other were included and were randomised. However this calls into question the studies generalisability and led to a small sample size – only 43 of 1808 (2.4%) patients screened were included in the randomisation. Further only 32 (1.8%) were then included in the analysis.
The authors state that they did find significant differences in terms of sex, disease severity and hypoglycaemic medication use between the randomised patients and the overall population (no further details are provided), again questioning whether the results of the study might apply to the wider population.
The study indicates that sample size was based on prior related studies. However no further details are given as to how the sample size was calculated. It is therefore not possible to tell whether the study was actually of an adequate size. Usually a calculation is done to ensure the study has enough subjects to detect an important effect but it cannot be established that this was the case.
Participants were aware what treatment they were given, so it was not possible to blind treatment allocation. Steps were taken to ensure outcomes were determined by a blinded outcome assessor, but participants’ behaviour could have been affected by their knowledge of treatment group. Additionally, those giving drug advice were not blinded, which could have also led to bias. This is highly important since the primary outcome (diabetes remission) required patients to be off drug treatment.
The results of the study are highly dependent on the subject’s compliance with the exercise, dietary and medical care interventions in the non-surgery group. An alternative explanation for the observed result might be that patients in the non-surgery arm did not actually complete the scheduled exercise or diary interventions. A summary of compliance in the non-surgery group was not provided in the study report, so this possibility cannot be ruled out. The study reports that physical activity questionnaires were submitted during the trial but this data is not shown.
It should be noted that the patients in the surgery group were also given a post-operative behavioral treatment programme which consisted of diet counselling and exercise recommendations.
The study only lasted one year, which the authors acknowledge.
HbA1c = glycated haemoglobin, a blood measurement taken to monitor diabetes.
Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.