Detailed statistical analysis plan for the outcomes quality of life and mortality and cardiovascular outcomes of the randomised 2x3 factorial Copenhagen Insulin and Metformin Therapy (CIMT) trial

By

Thomas P Almdal chief physician1,7, Trine W Boesgaard physician3, Leif Breum chief physician4, Birthe Gade-Rasmussen chief physician1 , Elsebeth Duun chief physician2, Christian Gluud head of department5, Christoffer Hedetoft chief physician4, Bianca Hemmingsen physician5,6, Tonny Jensen chief physician7,Thure Krarup chief physician8 ,Louise Lundby-Christensen physician,3,5 9, Søren S Lund physician3, Sten Madsbad professor1, 13,Elisabeth R Mathiesen professor7,13, Oluf Pedersen professor3,12,13, Hans Perrild associate professor8, Michael Røder chief physician1,6, Simone B Sneppen chief physician2 , Ole Snorgaard chief physician1, Lise Tarnow professor 3,6, Birger Thorsteinsson professor6,13, Henrik Vestergaard associate professor10,12,13, Allan Vaag professor3,7,13, Jørn Wetterslev chief physician4, Niels Wiinberg physician11, The CIMT Trial Group, Janus C Jakobsen physician, statistician5

Affiliations

1 Department of Endocrinology, Hvidovre, Copenhagen University Hospital, Kettegaard Allé 30, 2650 Hvidovre, Denmark.

2 Department of medicine, Gentofte, Copenhagen University Hospital, Kildegaardsvej 28, 2900 Hellerup, Denmark.

3 Steno Diabetes Center, Niels Steensens Vej 2, 2820 Gentofte, Denmark.

4 Department of Medicine, University Hospital Køge, Lykkebækvej 1,4600 Køge, Denmark:

5 Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Blegdamsvej 9, 2100 Copenhagen, Denmark.

6 Department of Cardiology, Nephrology and Endocrinology, Nordsjællands University Hospital - Hillerød, Dyrehavevej 29, 3400 Hillerød, Denmark

7 Department of Endocrinology, Rigshospitalet, University of Copenhagen, Blegdamsvej 9, 2100 Copenhagen, Denmark

8 Department of Endocrinology, Bispebjerg, Copenhagen University Hospital, Bispebjerg Bakke 23, 2400 Copenhagen, Denmark

9 Department of Paediatrics, Hvidovre, Copenhagen University Hospital, Kettegaard Allé 30, 2650 Hvidovre, Denmark

10 Department of Endocrinology, Herlev, Copenhagen University Hospital, Herlev Ringvej 75, 2730 Herlev, Denmark

1

11 Department of Physiology and nuclear medicine, Bispebjerg Hospital, Copenhagen University Hospital, Bispebjerg Bakke 23, 2400 Copenhagen, Denmark

12 The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, University of Copenhagen

13 University of Copenhagen, Denmark

List of E-mail adresses

Thomas P Almdal - almdal@dadlnet.dk

Trine W Boesgaard - tweb@novonordisk.com

Leif Breum - lbru@regionsjaelland.dk

Birthe Gade-Rasmussen -Birte.Gade-Rasmussen@regionh.dk

Elsebeth Duun - Elsebeth.Duun@regionh.dk

Christian Gluud – cgluud@ctu.dk.dk

Christoffer Hedetoft – hedetoft@dadlnet.dk

Bianca Hemmingsen – biancahemmingsen@hotmail.com

Tonny Jensen – tonny.jensen@regionh.dk

Thure Krarup – Thure.Krarup@regionh.dk

Louise Lundby-Christensen – Louise.Lundby.Christensen

Søren S Lund – sslund@dadlnet.dk

Sten Madsbad – sten.madsbad@regionh.dk

Elisabeth R Mathiesen – elisabeth.reinhardt.mathiessen@regionh.dk

Oluf Pedersen – oluf@sund.ku.dk

Hans Perrild – hans.perrild@regionh.dk

Michael Røder – mir@dadlnet.dk

Simone B Sneppen – simone.sneppen@regionh.dk

Ole Snorgaard – ole.snorgaard@regionh.dk

Lise Tarnow – lise@vibesholm.dk

Birger Thorsteinsson – birger.thorsteinsson

Henrik Vestergaard – henrik.vestergaard@sund.ku.dk

2

Allan Vaag – allan.arthur.vaag@regionh.dk

Jørn Wetterslev – wetterslev@ctu.dk

Niels Wiinberg – niels.wiinberg@rgionh.dk

Janus C Jakobsen – jcj@ctu.dk

Correspondence to:

Thomas Almdal Dr. Med. Sci.

Department of Endocrinology,

Rigshospitalet

Copenhagen University Hospital,

Blegdamsvej 9,

2100 Copenhagen,

Denmark

E-mail: almdal@dadlnet.dk

3

Abstract

Background: The evidence on the effects of metformin and insulin in type 2 diabetes (T2D) on

quality of life, mortality, and cardiovascular outcomes is unclear.

Methods: The Copenhagen Insulin and Metformin Therapy (CIMT) trial is an investigator-initiated,

randomised, placebo-controlled trial with a 2x3 factorial design conducted at eight hospitals in

Denmark. Participants with type 2 diabetes were randomised to: metformin (n=206) versus placebo

(n=206); in combination with open-label biphasic insulin aspart one to three times daily (n=137)

versus insulin aspart three times daily in combination with insulin detemir once daily (n=138)

versus insulin detemir once daily (n=137).

The quality of life outcomes in the CIMT trial are assessed by the Short Form Health Survey (SF-

36), physical component score and mental component score, the Diabetes Medication Satisfaction

Questionnaire to be completed score and Insulin Treatment Satisfaction Questionnaire score.

These outcomes are assessed at randomisation and 18 months after. The outcomes of time to a

clinical event are: (1) The composite of time to first of the following: death, myocardial infarction,

stroke, peripheral amputation, coronary revascularization, or peripheral revascularisation; (2) The

composite of time to the first of the following: death, myocardial infarction, stroke, or peripheral

amputation; (3) time to cardiovascular death. The individual components of each of these

outcomes will be evaluated separately as well. These outcomes are assessed from start of trial in

May 2008 and until June 2014. This article describes our analytic strategies for these outcomes

before any assessment of the data.

We will report the results of these exploratory outcomes in two papers, one in relation to the

metformin versus placebo comparison and one in relation to the three insulin comparisons.

Trial registration: The protocol was approved by the Regional Committee on Biomedical

Research Ethics (H-D-2007-112), the Danish Medicines Agency, and registered within

ClinicalTrials.gov (NCT00657943, 8th of april 2008)

Key words: Type 2 diabetes, metformin, insulin, randomised clinical trial, clinical outcomes, quality

of life, and statistical analysis plan.

4

Update

Clinical trials ought to be conducted according to a pre-specified analysis plan in order to prevent

reporting bias and data driven analysis results (1-4). The Copenhagen Insulin and Metformin

Therapy (CIMT) trial is an investigator initiated multicentre, randomized, placebo-controlled

superiority trial with a 2x3 factorial design. The CIMT trial evaluated the effect of an 18-month

treatment with metformin versus placebo and the effect of three insulin analogue regimens on the

progression of mean carotid carotid intima media thickness (IMT) in patients with type 2 diabetes

(5). The CIMT trial was conducted from May 2008 to December 2012 and the trial results in

relation to changes in carotid IMT have been reported previously (6, 7). In the trial protocol (5), we

pre-specified to assess the time to a number of clinical outcomes as well as measures of quality of

life as secondary and additional outcomes.

The CIMT trial participant inclusion criteria were: type 2 diabetes; age >30 years; body-mass index

>25 kg/m2; HbA1c ≥7.5% (58 mmol/mol); and treatment with oral antidiabetic drugs for at least one

year and/or insulin treatment for at least three months (5). After a screening visit, participants were

centrally randomised at the Copenhagen Trial Unit according to the 2x3 factorial design.

Participants were randomised 1:1:1 to treatment with one of three insulin analogue regimens and

in a factorial way randomised 1:1 to treatment with metformin versus placebo. Randomisation was

stratified by age >65 years (yes/ no), insulin treatment at trial entry (yes /no), and treatment at

Steno Diabetes Center (yes/ no). The insulin treatment was open-labelled, whereas participants,

investigators, and medical staff were blinded to the metformin and placebo intervention (numbered

identical containers).

The protocol was approved by the Regional Committee on Biomedical Research Ethics (H-D-2007-

112) and the Danish Medicines Agency, registered within ClinicalTrials.gov (NCT00657943), and

5

conducted in accordance with The Declaration of Helsinki and guidelines for Good Clinical

Practice. Informed consent was obtained from each participant enrolled in the trial.

In this paper we describe the detailed statistical analyses of the patient-relevant clinical outcomes

in the CIMT trial.

Sample size and power considerations

Based on the sample size calculation in relation to the primary outcomes of the trial (5), we

planned to include a total of 950 patients in the trial. However, the trial was stopped at the end of

the financing grant after the inclusion of 412 participants. For this factorial trial, this means that

there is a comparison on 206 on metformin versus 206 participants on placebo and another

comparison of 3 groups on different insulin regimens each including 137 and 138 participants. For

simplicity, we have only included power calculations based on the metformin versus placebo (206

versus 206) comparison, but it must be noted that the power for the insulin comparisons will be

less because of the lower number of randomised participants in each intervention group. We have

used an alfa on 1% in the power estimations because this threshold for significance was chosen as

the multiplicity adjusted (due to multiple outcomes) threshold in the protocol (5). The following

power estimations are based on the 412 randomised patients:.

Short Form Health Survey (SF-36)

Using an acceptable risk of type I error of 1%, an anticipated mean SF-36 score in the placebo

group of 45 points, and a SD of 9.5 points (8), we estimate that we will have 95.4% power to

confirm or reject a 4 point increase or decrease on the SF-36 score. We will consider 4 points as

the minimal clinically relevant difference (9). The minimal detectable mean difference with 80%

power will be 3.21 points.

6

Diabetes Medication Satisfaction Questionnaire (Diab-Medsat)

Using an acceptable risk of type I error of 1%, an anticipated mean Diab Medsat score in the

control group of 60 , and a SD of 18 points in relation to the three domains burden, efficacy and

symptoms (9), we estimate that we will have 37 % power to confirm or reject a 4 point increase or

decrease in the DTSQ score. We will consider 4 points as the minimal clinically relevant difference

(10). The minimal detectable mean difference with 80% power will be 9.

Insulin Treatment Satisfaction Questionaire (ITSQ)

Using an acceptable risk of type I error of 1%, an anticipated mean ITSQ score in the control group

of 80 points, and a SD of 9 points (10), we estimate that we will have 97.2% power to confirm or

reject a 4 point increase or decrease on the ITSQ score. We will consider 4 points as the minimal

clinically relevant difference (10). The minimal detectable mean difference with 80% power will be

3.04 points.

Time to death, myocardial infarction, stroke, peripheral amputation, coronary

revascularisation, or peripheral revascularisation

We will report the number of outcomes for the composite as well as the individual components per

intervention group at both 18-month landmark follow up as well as at maximal follow-up. However,

due to our power problems and in order to reduce multiplicity we will only conduct analyses of the

latter data. Using an acceptable risk of type I error of 1% and an overall estimated 29% risk of

experiencing one or more of the following outcomes: death, non-fatal myocardial infarction, stroke,

peripheral amputation, coronary revascularization, or peripheral revascularisation in the placebo

7

group (12), we estimate that we will have 26.6% power to confirm or reject a hazard ratio reduction

of 30% on this composite outcome in the metformin group.

Time to the first of the following clinical outcomes death, myocardial infarction, stroke, or

peripheral amputation

We will report the number of outcomes for the composite per intervention group at both 18-month

landmark follow up as well as at maximal follow-up. However, due to our power problems and in

order to reduce multiplicity we will only conduct analyses of the latter data. Using an acceptable

risk of type I error of 1% and an overall estimated 12.5% risk of experiencing one or more of the

following outcomes: death, non-fatal myocardial infarction and stroke, and peripheral amputation in

the placebo group (12), we estimate that we will have 9.75% power to confirm or reject a hazard

ratio reduction of 30% on this composite outcome in the metformin group.

Time to cardiovascular death

We will report the number of outcomes per intervention group at both 18-month landmark follow up

as well as at maximal follow-up. However, due to our power problems and in order to reduce

multiplicity we will only conduct analyses of the latter data. Using an acceptable risk of type I error

of 1% and an overall estimated 3.8% risk of experiencing death from cardiovascular disease in the

placebo group (12), we estimate that we will have only 3.1% power to confirm or reject a hazard

ratio reduction of 30% on this outcome in the metformin group.

Stratification and design variables The randomisations were performed using three stratification variables: age >65 years (yes/ no),

insulin treatment at trial entry (yes/ no), and treatment at Steno Diabetes Center (yes/ no).

8

Other design variables were sex (male/ female), prior cardiovascular disease (yes/ no), statin

treatment at baseline (yes/ no), and glutamic acid decarboxylase antibodies (positive/ negative)

(5).

Outcomes

All outcomes will be considered as exploratory.

1. Short Form Health Survey (SF-36) physical component score.

2. Short Form Health Survey (SF-36) mental component score.

3. Diabetes Medication Satisfaction Questionnaire burden, efficacy and

symptoms scores.

4. Insulin Treatment Satisfaction Questionnaire (ITSQ) score.

5. The composite of time to death, myocardial infarction, stroke, peripheral

amputation, coronary revascularisation, or peripheral revascularisation.

6. The composite of time to death, myocardial infarction, stroke, or peripheral

amputation.

7. Time to cardiovascular death.

Outcomes 1 to 4 were evaluated only at the end of intervention after 18 months treatment. The

outcomes were all assessed by the participants who were blinded to metformin versus placebo but

not to the three insulin regimens.

The observation period in relation to outcomes 5 to 7 is from inclusion of the individual patients,

May 2008 to April 2011, and followed up to June 2014. The outcomes were assessed by an

adjudication committee blinded to all interventions ( 13, 14, 15, 16, 17).

Populations and handling of missing dataThe following populations will be analysed:

A modified intention-to-treat population. This population includes all patient, which passed the

initial investigations and were randomized – but excluding the following:

- randomised participants who did not fulfill the inclusion criteria;

9

- randomised participants fulfilling one or more exclusion criteria;

- randomised participants who did not receive any of the planned interventions.

A per protocol population. This population includes all those includes in the modified intention-to-

treat population, but excludes those characterised by the following major protocol violations:

- patients who did not attend more than 4 out of the planned 6 visits following randomisation.

We do not except any missing values for the time to a clinical outcomes as all deaths and

admissions to Danish hospitals are registered in the Civil Registry and the National Patient

Registry and were adjudicated for clinical events by the blinded adjudication committee (13, 14, 15,

16, 17). If missing data for the remaining outcomes or covariates represents less than 5%, a

complete participant analysis without multiple imputation of missing values will be conducted. If

more than 5% of data are missing, a blinded statistician will assess whether missing completely at

random may be ruled out based on a rational assessment of the pattern of missing data (18).

Little’s test will be used if there is doubt (19). If more than 5% of the data are missing and missing

completely at random is ruled out, multiple imputation using chained equations will be performed

by creating ten imputed data sets that will be analysed under the assumption that data are missing

at random (19, 20). We will use the following three stratification and four design variables during

the multiple imputation procedure to estimate the missing values, i.e., age >65 years (yes/ no),

insulin treatment at trial entry (yes/ no), and treatment at Steno Diabetes Center (yes/ no) as well

as sex (male/ female), prior cardiovascular disease (yes/ no), statin treatment at baseline (yes/ no),

and glutamic acid decarboxylase antibodies (positive/ negative).

In relation to quality of life measures, participants who died during the 18 month trial period will be

assigned the value zero. If multiple imputation is used, then the primary result of the trial will be

based on these data. To take account of the possibility that data may be ‘missing not at random’,

we will use a best-worst and worst-best case scenario as sensitivity analyses, which will assess

the potential impact of the missing data on the trial results (10). In the ’best-worst-case’ scenario it

is assumed that all patients lost to follow-up in the experimental group have had a beneficial

outcome; and all those with missing outcomes in the control group have had a harmful outcome.

Conversely, in the ’worst-best-case’ scenario, it is assumed that all patients who were lost to follow

up in the experimental group have had a harmful outcome; and that all those lost to follow-up in the

control group have had a beneficial outcome. When continuous outcomes are analysed, a

‘beneficial outcome’ will be defined as the group mean plus two SDs of the group mean, and a

10

‘harmful outcome’ will be defined as the group mean minus two SDs of the group mean for ‘best-

worst case’ imputation.

Statistical analysis

We will assess if the thresholds for statistical significance and clinical significance are crossed

using the five-point procedure as suggested by Jakobsen et al. (9). This procedure will include

adjustments of thresholds for significance according to the number of outcome comparisons (P-

value threshold 0.01) and number of randomised participants. We will use a Bayes factor threshold

for significance of 0.1 based on the a priory anticipated intervention effect (10).

Analysis of quality of life outcomes

Primary analysis: linear regression adjusted for the three stratification variables (age >65 years

(yes/ no), insulin treatment at trial entry (yes/ no), and treatment at Steno Diabetes Center (yes/

no) (5)

Secondary analysis: linear regression adjusted for stratification variables (see above) plus the

design variables (sex (male/ female), prior cardiovascular disease (yes/ no), statin treatment at

baseline (yes/ no), and glutamic acid decarboxylase antibodies (positive/ negative) (5).

All analyses will primarily include the modified intention-to-treat population. We will also perform

the analyses adjusted for the stratification variables including the per protocol population.

Assumptions behind the linear regression will be tested by ‘qq’ plots of the residuals and plots of

residuals against co-variates and fitted values.

Analysis of the time to a clinical outcome

Primary analysis: hazard ratios estimated by Cox regression analysis adjusted for the three

stratification variables (age >65 years (yes/ no), insulin treatment at trial entry (yes/ no), and

treatment at Steno Diabetes Center (yes/ no).

11

Secondary analysis: hazard ratios estimated by Cox regression analysis adjusted for stratification

variables (see paragraph above) and other design variables (sex (male/ female), prior

cardiovascular disease (yes/ no), statin treatment at baseline (yes/ no), and glutamic acid

decarboxylase antibodies (positive/ negative) (5).

The assumption of proportional hazards will be tested by (a) plots of Shoenfeld residuals against

each variable (continuous and categorical) and (b) against fitted values (22).

Possible interactions between treatment and covariates will be investigated by tests of interaction.

Characteristics of patients at baseline

We will present a description of baseline characteristics by intervention group. Discrete variables

will be summarised by frequencies and percentages. Percentages will be calculated according to

the number of patients where data are available. Where values are missing, the actual

denominator will be stated.

Continuous variables will be summarised using standard measures of central tendency and

dispersion using either mean + SD for data with normal distribution or median and interquartile

range for non-normally distributed data.

Outline of figures and tables

The first figure will be a consort flow chart of respectively metformin versus placebo or the three

insulin regimens.

The first table will describe the baselines characteristics as shown in Table 1.

The quality of life outcomes: The second to fifth figures will be whiskers and box plots (median

with interquartile ranges) of the four measures of quality of life comparing respectively metformin

versus placebo and the three insulin regimens compared.

12

The clinical outcomes: The sixth to eight figures will be Kaplan Meier curves for the clinical

outcomes 5, 6, and 7 in relation to metformin versus placebo respectively the three insulin

regimens compared.

The individual clinical outcomes will in relation to metformin versus placebo be reported as shown

in Table 2.

In the insulin comparison, one of the groups (A) will randomly be identified as the reference group

and compared with the two other groups (B, C) as shown in Table 3.

Deviations from the initial design and methods of the CIMT trial

Based on our sample size estimation for the primary outcome of the CIMT trial allowing for five

comparisons of the involved intervention groups we planned to include 950 patients (5). However,

eventually only 412 present were included, as the trial had to be stopped at the scheduled duration

of the trial and lack of additional financial support. In the present analysis plan we schedule to

analyse the quality of life measures and the clinical outcomes based on a strict factorial 2 x 3

design comparing the effects of metformin versus placebo and comparing the three different insulin

analogue regimens. We originally planned to analyse the clinical outcomes as secondary

outcomes and the quality of life data as exploratory outcomes. Given the low power in relation to

the clinical outcomes we consider all outcomes to be exploratory and that most emphasis should

be given to the quality of life data according to our result of the power calculations.

13

In the original protocol it was not specified how the patient-important quality of life data and clinical

outcomes would be analysed and presented. Given the fact that the trial is a 2 x 3 factorial design

and two sets of data are presented we anticipate to publish the data in two separate papers.

Discussion

In order to avoid risks of outcome reporting bias and data-driven results, this paper presents the

detailed statistical analysis plan for the CIMT trial analysing patient-relevant outcomes at a time

when the data have been gathered but not yet inspected or analysed. Contrary to our pre-specified

plans for the trial (5), we now plan to report clinical outcomes as exploratory outcomes. We will use

the five-step procedure proposed by Jakobsen et al. to assess whether the thresholds for statistical

and clinical significance are crossed or not (10). We will analyse data in accordance to the

modified intention-to-treat principle and, if necessary, use data sets generated by multiple

imputations, and a best-worst/worst-best case scenario to assess the potential impact of the

missing data on the results.

Strengths

This statistical analysis plan has a number of strengths. First of all, we want to make the reporting

of the patient-important outcomes from the CIMT trial transparent facilitating the use of these data

in systematic reviews and meta-analyses of the effect of metformin and the insulin analogues used

in the trial (23, 24, 25). The overall trial methodology has been pre-defined (5), we take the

multiplicity problem into account, use validated statistical methods, we systematically test for

underlying statistical assumptions, and we use multiple imputations and best-worst and worst-best

case scenarios to take account of incomplete outcome data bias. In order to minimize the risk of

multiplicity we restrict our analyses to the outcomes that were predefined (5). As an example, it

14

might also have been of interest to analyse a composite outcome of cardiovascular death and

other cardiovascular outcomes but we abstained from this due to the risks of multiplicity

Limitations

The statistical analysis plan has limitations. First, the effects of the interventions on the primary

unvalidated surrogate outcome in the main publications were neutral and has been accepted for

publication. Due to multiplicity caused by multiple outcome comparisons our results should be

interpreted as exploratory and thus hypothesis generating (10). Second, we plan to use Cox

regression in the analysis of the time to a clinical outcomes. We plan to use the data based on the

entire trial observation period, i.e., some participants (those who are randomised early in the trial

period) will be observed much longer than others. This may make the interpretation of the trial

results difficult with events occurring long time after the intervention was stopped maybe less likely

to have been caused by the intervention. Third, we are assessing a number of clinically relevant

outcome measures using survival analysis but we do not have sufficient power to reliable access

and conclude on these time to event outcomes. Therefore, all neutral as well as significant findings

should be interpreted cautiously. To try to control the risks of type I errors we will assess

significance using the Lan-DeMets monitoring boundaries for outcomes with power less than 80%

(10). Forth, the power estimations on the QOL outcomes show power estimations over 90%, which

result in a high risk of showing statistically significant results without any clinical relevance (10). We

have pre-specified a minimal clinically relevant difference to limit this risk. Finally, we use multiple

imputation to handle missing data on the remaining outcome measures and it cannot be ruled out

that missingness are not at random and multiple imputation may produce biased results and the

intervention effect estimate based on the multiple imputation might therefore also be biased.

Conclusion

15

This article describes the principles of the statistical analyses of a number of patient important

exploratory outcomes used in the CIMT trial in order to minimize risk of data-driven results and

outcome reporting bias.

List of abbreviations:

CIMT = The Copenhagen Insulin and Metformin Therapy

IMT = intima media thickness

SF-36 = Short Form Health Survey 36

SD = Standard Deviation

Diab Medsat = Diabetes Medication Satisfaction Questionnaire

ITSQ = Insulin Treatment Satisfaction Questionaire

Competing Interest. LLC, LT, TA, AV, OP, TB, BC, SL, TJ have reported shares in Novo Nordisk

A/S; SL owns shares in dynamically traded investment funds, which may own stocks from

pharmaceutical companies LLC, LT, TA, AV, TB, SL have reported former employment at Steno

Diabetes Center, which is a diabetes hospital and academic institution owned by Novo Nordisk; SL

is an employee at Boehringer Ingelheim, Germany. SL contribution was his alone and does not

necessarily reflect the official position of Boehringer Ingelheim. AV has received fees from Novo

Nordisk; TB is employed at Novo Nordisk A/S; BT is member of advisory board for Eli Lilly; OS has

received fees from AstraZeneca, Sanofi, MSD, Boehringer-Ingelheim, Eli Lilly, NovoNordisk; CH

has received fees from Bristol-Myers Squibb; LB has received fees from and attended advisory for

Novo Nordisk A/S; SM has served as a consultant or adviser to: Novartis Pharma, Novo Nordisk,

Merck Sharp & Dome, Sanofi-Aventis, AstraZeneca, Johnson & Johnson, Rosche, Mankind, Astra-

Zeneca, Boehringer-Ingelheim, Zeeland, E Lilly, Intarcia Therapeutics, Bristol-Meyer Squibb, has

received fee for speaking from Novo Nordisk, Merck, Sharp & Dome, Astra-Zeneca, Johnson and

16

Johnson, Rosche, Shering-Ploug, Sanofi-Aventis, Novartis Pharma, E Lilly, Bristol-Meyer Squibb,

Boehringer Ingelheim, and has received 2 research grants from Novo Nordisk. BH, BGR, JW, CG,

NW, MR, HV, ED, HP, TK, SS, EM have no conflicts of interests to declare.

Author Contributions. TA: Drafted the present manuscript and revised and approved the final version.

TB: Revised and approved the final version of the manuscript

LB: Revised and approved the final version of the manuscript

BG-R: Revised and approved the final version of the manuscript

ED: Revised and approved the final version of the manuscript

CG: Drafted the present manuscript and revised and approved the final version.

CH: Revised and approved the final version of the manuscript

BH: Revised and approved the final version of the manuscript

TJ: Revised and approved the final version of the manuscript

TK: Revised and approved the final version of the manuscript

LL-C: Revised and approved the final version of the manuscript

SSL: Revised and approved the final version of the manuscript

SM: Revised and approved the final version of the manuscript

ERM: Revised and approved the final version of the manuscript

OP: Revised and approved the final version of the manuscript

HP: Revised and approved the final version of the manuscript

MR: Revised and approved the final version of the manuscript

SBS: Revised and approved the final version of the manuscript

OS: Revised and approved the final version of the manuscript

LT: Revised and approved the final version of the manuscript

BT: Revised and approved the final version of the manuscript

HV: Revised and approved the final version of the manuscript

17

AV: Revised and approved the final version of the manuscript

JW: Drafted the present manuscript and revised and approved the final version.

NW: Revised and approved the final version of the manuscript

JCJ: Drafted the present manuscript and revised and approved the final version.

Funding. The investigators received an unrestricted grant from Novo Nordisk to enable conduct of

the trial. The trial and its main 2x3 factorial design was 100% initiated and conducted by the

investigators. Novo Nordisk was allowed to comment on the protocol, on protocol changes during

the trial, and on the manuscript prior to submission.

18

References

1. Dwan K, Gamble C, Williamson PR, Altman DG. Reporting of clinical trials: a review of research funders guidelines. Trials 2008; 9: 66 – 77.

2. Dwan K, Altman DG, Clarke M, Gamble C, Higgins JPT, et al. Evidence for the selective reporting of analyses and discrepancies in clinical trials: a systematic review of cohort studies of clinical trials. PLoS Med 2014;11(6): e1001666. doi:10.1371/ journal.pmed.1001666.

3. Ebrahim S, Sohani ZN, Montoya L, Agarwal A, Thorlund K, Mills EJ, Ioannidis JPA. Reanalyses of randomized clinical trial data. JAMA. 2014;312(10):1024-1032. doi:10.1001/ jama.2014.9646.

4. Ioannidis JPA. How to make more published research true. PLoS Med 2014;11(10): e1001747.doi:10.1371/journal.pmed.1001747

5. Lundby Christensen, L., Almdal, T., Boesgaard, et al. Study rationale and design of the CIMT trial: the Copenhagen Insulin and Metformin Therapy trial. Diabetes, Obesity & Metabolism, 2009;11(4), 315–22. doi:10.1111/j.1463-1326.2008.00959.

6. Lundby-Christensen L, The CIMT Trial Group. Effects of metformin versus placebo in combination with insulin analogues in patients with type 2 diabetes mellitus: The CIMT randomized clinical trial. BMJ-Open, accepted for publication 2015

7. Lundby-Christensen L, The CIMT Trial Group. Effect of three insulin regimens on carotid intima-media thickness in patients with type 2 diabetes: the randomised Copenhagen Insulin and Metformin Therapy (CIMT) trial. BMJ-Open, accepted for publication 2015

8. Donk M v d, Griffins S, Stellato R et al. Effects of early intensive multifactorial therapy compared with routine care on self-reported health status, general well-being, diabetes specific quality of life, and treatment satisfaction in screen detected type 2 diabetes mellitus patients (ADDITION-Europe): a cluster randomized trial. Diabetologia; 2013: 56: 2367-77.

9. Al-Aujan A, AL-Aqeel S, Al-Harbi A, Al-Abdulltif E. Patients satisfaction with diabetes medications in one hospital, Saudi Arabia. Patients Preferences and Adherence 2012: 6: 735 - 740

10. Jacobsen JC, Gluud C, Winkel P, Lange T, Wetterslev J. The thresholds for statistical and clinical significance – a five step procedure for evaluation of intervention effects in randomized clinical trials. BMC Med Res Methodol 2014; 14: 34 – 40.

11. Farmer AJ, Oke J, Stevens R, Holmann RR. Differences in insulin treatment satisfaction following randomized addition of biphasic, prandial or basal insulin to oral therapy in type 2 diabetes. DOM 2011; 13: 1136 – 41.

12. Hemmingsen, B., Lund, S. S., Gluud, C., Vaag, A., Almdal, T., Hemmingsen, C., & Wetterslev, J. Intensive glycaemic control for patients with type 2 diabetes: systematic review with meta-analysis and trial sequential analysis of randomised clinical trials. BMJ (Clinical Research Ed.), 2011;343(November), d6898. doi:10.1136/bmj.d6898.

13. Jespersen CM, Als-Nielsen B, Damgaard M, Hansen JF, Hansen S, Helø OH, Hildebrandt P, Hilden J, Jensen GB, Kastrup J, Kolmos HJ, Kjøller E, Lind I, Nielsen H, Petersen L, Gluud C; CLARICOR Trial Group Randomised placebo controlled multicentre trial to assess short term clarithromycin for patients with stable coronary heart disease: CLARICOR trial. BMJ. 2006;332(7532):22-7.

19

14. Gluud C, Als-Nielsen B, Damgaard M, Fischer Hansen J, Hansen S, Helø OH, Hildebrandt P, Hilden J, Jensen GB, Kastrup J, Kolmos HJ, Kjøller E, Lind I, Nielsen H, Petersen L, Jespersen CM; CLARICOR Trial Group. Clarithromycin for 2 weeks for stable coronary heart disease: 6-year follow-up of the CLARICOR randomized trial and updated meta-analysis of antibiotics for coronary heart disease. Cardiology. 2008;111(4):280-7. doi: 10.1159/000128994. Epub 2008 May 2.

15. Kjoller E, Hilden J, Winkel P, Frandsen NJ, Galatius S, Jensen G, Kastrup J, Hansen JF, Kolmos HJ, Jespersen CM, Hildebrandt P, Gluud C; CLARICOR Trial Group. Good interobserver agreement was attainable on outcome adjudication in patients with stable coronary heart disease. J Clin Epidemiol. 2012;65(4):444-53. doi: 0.1016/ j.jclinepi.2011.09.011. Epub 2012 Jan 17.

16. Kjøller E, Hilden J, Winkel P, Galatius S, Frandsen NJ, Jensen GB, Fischer Hansen J, Kastrup J, Jespersen CM, Hildebrandt P, Kolmos HJ, Gluud C; CLARICOR Trial Group. Agreement between public register and adjudication committee outcome in a cardiovascular randomized clinical trial.Am Heart J. 2014;168(2):197-204.e1-4. doi: 10.1016/j.ahj.2013.12.032. Epub 2014 May 4.

17. Winkel P, Hilden J, Hansen JF, Kastrup J, Kolmos HJ, Kjøller E, Jensen GB, Skoog M, Lindschou J, Gluud C; the CLARICOR trial group. Clarithromycin for stable coronary heart disease increases all-cause and cardiovascular mortality and cerebrovascular morbidity over 10years in the CLARICOR randomised, blinded clinical trial. Int J Cardiol. 20156;182C:459-465. doi: 10.1016/j.ijcard. 2015.01.020.

18. Mallinckrodt CH, Lin Q, Molensbergh M. A structured framework for assessing sensitivity to missing data assumptions in longitudinal clinical trials. Pharm Stat. 2013; 12: 1 – 6.

19. Little RJA. A test of missing completely at random multivariate data ith missing values. J Am Stat Assoc. 1988; 83: 1198.

20. Schafer JL Multiple imputations: a primer. Stat Methods Med Res. 1999; 8: 3 – 15.

21. Vickers AJ, Altman DG. Statistical notes: missing outcomes in randomised trials. BMJ 2013: 346: f3438.

22. Mario Cleves WG, Gutierrez RG, Marchenko YV. An Introduction to Survival Analysis Using Stata, Third Edition. Stata Press. 2010

23. Guyatt GH, Mills EJ, Elbourne D. In the era of systematic reviews, does the size of an individual trial still matter. PLoS Medicine 2008;5(1):e4.doi:10.1371/ journal.pmed.0050004

24. Borm GF, Lemmers O, Fransen J, Donders R. The evidence provided by a single trial is less reliable than its statistical analysis suggest. J Clin Epidemiol 2009; 62: 711 – 715

25. Inthout J, Ioannidis JPA, Borm GF. Obtaining evidence by a single well-powered trial or several modestly powered trials. SMMR 2012: 1 – 15.

20

Table 1. Baseline characteristics in the randomised comparison of metformin versus placebo in the CIMT trial.

  Metformin + insulin (n=206)

Placebo + insulin (n=206)

Age (years)No (%) maleWeight (kg)Body-mass index*Waist-hip ratioNo (%) smokersMedian (IQR) alcohol consumption (units/week)No (%) caucasiansDuration of type 2 diabetes (years)No (%) GAD65 antibodies ≥25U/ml HbA1c (%)HbA1c (mmol/mol)No (%) prior cardiovascular diseaseSystolic blood pressure (mmHg)Diastolic blood pressure (mmHg)Total cholesterol (mmol/L)LDL cholesterol (mmol/L)Median (IQR) VLDL cholesterol (mmol/L)HDL cholesterol (mmol/L)Median (IQR) triglycerides (mmol/L)Mean carotid IMT (mm)

MedicationNo (%) metformin‡ No (%) insulin‡No (%) sulphonylurea‡No (%) other antihyperglycaemic drug‡No (%) RAS blockadeNo (%) other antihypertensive drugNo (%) statinNo (%) aspirinGAD=glutamic acid decarboxylase; HbA1c=haemoglobin A1c; LDL=low-density lipoprotein; VLDL=very low-density lipoprotein; HDL=high-density lipoprotein; IMT=intima-media thickness; RAS=Renin angiotensin system.

* Body mass index is calculated as weight (kg) divided by height (m)2.

21

‡ All antihyperglycaemic drugs were terminated at randomisation.

Table 2. Clinical outcomes in the randomised comparison of metformin versus placebo in the CIMT trial.

Outcomes Number of outcomes per intervention group by 18

monthPlacebo Metformin

Hazard ratioWith 95% confidence interval

(P values)

Time to death, myocardial infarction, stroke, peripheral amputation, coronary revascularisation or peripheral revascularizationTime to death, myocardial infarction, stroke or peripheral amputationTime to cardiovascular deathAll-cause mortalityMyocardial infarctionCoronary revascularizationPercutaneous coronary intervention (PCI)StrokePeripheral revascularizationPeripheral amputation

22

Table 3. Clinical outcomes in the randomised comparison of metformin versus placebo in the CIMT trial.

Outcomes Number of outcomes per intervention group by 18 month

Insulin A Insulin B Insulin C

Hazard ratio with 95 % confidence intervals

Comparison of insulin regimen A versus B

Hazard ratio with 95 % confidence intervals

Comparison of insulin regimen A versus C

Time to death, myocardial infarction, stroke, peripheral amputation, coronary revascularisation or peripheral revascularisationTime to death, myocardial infarction, stroke or peripheral amputationTime to cardiovascular deathAll-cause mortalityMyocardial infarctionCoronary revascularisationPercutaneous coronary intervention (PCI)StrokePeripheral revascularisationPeripheral amputation

23