Dear Editor,

We would like to thank you for the opportunity to re-submit our manuscript entitled “β-blockers and mortality after myocardial

infarction in patients without heart failure: multicentre prospective cohort study” for publication in BMJ.

We wish to thank the reviewers for the attention they have brought to our manuscript and for useful comments that have been

made to us. As requested we herewith provide a detailed point-to-point list of responses to the reviewers as well as changes

made to the manuscript.

We hope we have taken into account the remarks and hope to make the expected answers.

We look forward to the comments of the reviewers and to your decision.

Sincerely,

Etienne PUYMIRAT, MD, PhD and Nicolas Danchin, MD

Hôpital Européen Georges Pompidou

**Report from The BMJ’s manuscript committee meeting** These comments are an attempt to summarise the discussions at the manuscript meeting. They are not an exact transcript. Members of the committee were: Wim Weber (Chair), Tim Cole (Statistics advisor), Jessamy Bagenal, Elizabeth Loder, José Merino, Rubin Minhas, Amy Price, Tiago Villanueva. Decision: Put points Detailed comments from the meeting: We thought your paper addresses an interesting and important research question. Apart from the reviewers' comments, we had the following minor queries: Multivariable should be used instead of multivariate. We have corrected the different occurrences of “multivariate” into “multivariable”. Please describe propensity scoring better. How are patient numbers determined (e.g. 502 for population 1)? We have provided additional details on the way propensity score matching is performed. Within each population, a propensity score defining the probability for each patient of receiving (or not) beta-blockers is calculated using a non-parsimonious logistic regression analysis. Each patient therefore has a given score, reflecting his/her probability to get beta-blockers, according to his/her characteristics. Then, a matching procedure based on the propensity score is performed to pair one (or more) patient having actually received beta-blockers with one patient without beta-blocker. The procedure stops when no further matching is possible (i.e. when a patient without beta-blocker has not patient with beta-blockers having a similar propensity score). You might want to improve y-scales for Kaplan-Meier plots. We have redrawn the Kaplan-Meier plots, using a y scale from 60 to 100%.

First, please revise your paper to respond to all of the comments by the reviewers. Their reports are available at the end of this letter, below. In your response please provide, point by point, your replies to the comments made by the reviewers and the editors, explaining how you have dealt with them in the paper. Comments from Reviewers Reviewer: 1 Recommendation: Comments: Very relevant to patients. Long term use of B blockers should be evaluated for efficacy and it appears this paper is sufficiently powered, and has the right time limitations, to shed some light on that determination. Certainly valuable for a patient to be able to have an informed conversation with their provider to determine whether ongoing treatment is necessary or desired. A patient would be most likely to use the conclusions (or abstract - which appears thorough and accurate as well) alone however, as I think the discussion is too technical to be of practical use. Very thorough however, and an important finding to many. Medication and other follow ups were patient reported. PROs are a good sign of patient inclusion, but may they limit the validity and utility of the information? Overall, a good sign that patient reported information is included and their process seemed to involve patients and their feedback at regular intervals. Very nice to see. The authors seemed to be well aware of the study's limitations, and disclosed them. Impressive work - I have no real critiques to make. We thank the reviewer for her positive appreciation of the paper. We have provided additional details on the way follow-up data were obtained and reviewed. Additional Questions: Please enter your name: Dana E. Connors Job Title: Project Manager Institution: AcademyHealth Reviewer: 2 Recommendation: Comments: Puymirat et al. conducted a very interesting analysis of the FAST-MI registry specifically investigating the interaction between beta-blocker use at the different time points and all cause mortality. They demonstrated a clear benefit from an early use of beta-blockers on 30 days

mortality. This beneficial effect diminished at 1 year, and completely disappeared at 5 years in those patients still on beta-blockers at 12 months post-MI. This paper is well written and deals with an interest research field which has been debated lately: namely the need of prolonged beta-blocker therapy in the pPCI era in patients without significant impairment of LV function. The analysis is very well done and the conclusions seem to be supported by the data provided. The manuscript comes from a very experienced and renowned group in the field. I have only some minor comments: 1) Multivessel disease is not infrequent in STEMI patients (up to 50% of the cases). Revascularization within the first month of the non-IRA has been shown to have an impact on mortality. Did the authors account for this factor in their analysis? If not, can they comment on this?

We took into account all revascularisation procedures performed during the initial hospital

admission in the analysis. Three percent of the patients had a revascularisation procedure after

discharge and within a month of AMI (1.3% in those without beta-blockers at discharge, and 3.4% of

those with beta-blockers). Following the reviewer’s suggestion, we added post-discharge

revascularisation to the model and the results were virtually unchanged (HR 1.24, 0.68-2.23, P=0.48).

2) Could there be any difference in the beneficial effect of beta-blockers in patients presenting with anterior versus non-anterior MI? We checked the relationship between beta-blocker treatment and survival in STEMI patients, and there was no interaction with location of MI (anterior vs non-anterior). Likewise, adding location of MI in the multivariate analysis did not change its results (location of infarct was one of the first variables deleted from the stepwise analysis).. Additional Questions: Please enter your name: Emanuele Barbato Job Title: Prof. Institution: Cardiovascular Research Center Aalst, Belgium Reimbursement for attending a symposium?: No Reviewer: 3 Recommendation: Comments: The manuscript by Puymirat et al describes the outcome of early, one-year, and longer term beta blocker therapy after myocardial infarction in patients without heart failure. The data derive from

a nationwide French registry on ST- and non-ST-elevation MI (FAST-MI). The effect of beta blocker therapy on short-, mid-, and long-term survival after myocardial infarction has become a very significant question, as more recent data in the reperfusion era have reignited the question whether beta blockers reduce mortality post-MI, particularly in lower risk subgroups. The authors have therefore focused their analysis on patients without heart failure. The authors have indeed identified a very low-risk population as manifested by the low mortalities noted at the various time points in this group of patients. Given these low mortalities, though, it is important to consider whether the study is underpowered to evaluate the effect of beta blockers. Moreover, the authors should be more clear that the one-year data do not support a beneficial effect of beta blockers in this study. Other issues to consider are: 1) There are important issues related to the inclusion criteria that need to be clarified. The authors note that consent was required to participate in the study. They do not indicate the number of patients who consented and number who have refused consent. this could be a potential bias.

The number of refusals was less than 2%.

2) While the authors focused only on patients without heart failure, identifying a low-risk group that potentially does not benefit from beta blocker therapy, inclusion of the patients with the heart failure in this report would provide a great comparator to establish that patients at high risk are the ones that benefit from beta blocker therapy and not the ones at low risk (if this is the case). Inclusion of patients with heart failure or low ejection fraction (defined here as ≤40%) was beyond the scope of the current analysis, as we did not want to focus on the group of patients in whom the use of beta-blockers is undisputed. As suggested by the reviewer, however, we did a specific analysis on the (smaller) group of patients who had a history of heart failure, or low ejection fraction. In the multivariable analysis in such patients, treated with beta-blockers at discharge, the hazard ratio for 5-year mortality in those who had stopped beta-blockers at one year was 2.11 (1.16-3.83), P=0.014. We have added a sentence providing this additional result in the discussion section. 3) The population needs to be better defined in Table 1. The authors indicate PCI in the first 48 hours, but this is not necessarily acute PCI or thrombolytic therapy. This categorization is important. The number of patients undergoing coronary artery bypass surgery should also be indicated. Data on CK and troponin should be provided. LVEF data should be provided. The use of moderate to high dose beta blocker therapy should be indicated, particularly given the OBTAIN registry (JACC 2015) results showing differential effects for low and high dose beta blocker therapy. The outcome based on the use of moderate to high doses versus lower doses should be reported. We thank the reviewer for giving us the opportunity to provide additional details on our population. We have added the type of reperfusion therapy in STEMI patients, coronary bypass graft during the initial hospital stay, and data on maximal CK and LVEF. The OBTAIN registry is an important contribution in the field, and we have added it to the references. However, OBTAIN actually showed no significant difference between patients receiving

so-called “optimal” vs “suboptimal” doses of beta-blockers at discharge (very low doses being associated with slightly, although not significantly better 2-year outcomes). Nonetheless, we have added the information on beta-blocker doses at discharge in Table 1. As stated in the results section, (and in concordance with the OBTAIN results), beta-blocker dose was not associated with survival in multivariable analyses. 4) The authors should not confound the statistical analysis showing no effect of beta blockers on one-year mortality after multivariable and propensity score adjustment to indicate that there was a benefit. We agree that our results do not document a benefit of beta-blockers at one year, as the analyses showed no statistically significant difference, in spite of HRs <1.00. 5) It appears that total mortality was the primary outcome listed in the methods (rates of patient loss to follow-up for vital status was 0.3% at one year). No information is provided regarding adjudication for cardiovascular or sudden cardiac death. It is not clear how this was ascertained and whether the data are reliable. As such, these data should be removed from the manuscript. Furthermore, with no difference in total mortality, the relevance of these sub-categorizations is minimal. We have provided additional information on how follow-up data were acquired and adjudicated. We think that, because non-cardiovascular mortality is not uncommon in a post-MI population, particularly in the long-term, and because beta-blockers are unlikely to influence non-cardiovascular mortality, these results provide valuable additional information. We are ready to remove these data, however, if the editor wishes so. 6) The authors claim that the confounders are likely to influence the outcomes in a similar way. This does not seem to be justified. We agree with the reviewer’s comment. As mentioned confounders may be present in all observational cohorts. Our point was that there was a good chance that unrecorded confounders would have been the same whatever the type of treatment studied (beta-blockers or statins). We have rephrased the sentence: “In addition, unrecorded confounders would have likely be similar for statins and β-blockers”. Additional Questions: Please enter your name: jg Job Title: md Institution: um Reviewer: 4 Recommendation:

Comments: General points: The authors have attempted to answer an important clinical question but have used the wrong study design. Observational data is poorly equipped to assess the prognostic impact of treatments that have considerable prescription bias - in this case, bb are given to younger, less comorbid patients who have a much more intensive intervention admission (52% early PCI vs 39%). In this regard, early bb is merely an indicator of other prognostic factors, so no surprise that bb is associated with less 30 day mortality (no analysis needed!!). The authors need to carefully rewrite the tone, to avoid making causality statements (without randomization, there are many other known, masked and unknown confounders), not to mention physician bias in 'choosing' which patients go on therapy. As we demonstrated in the BMJ recently for digoxin (BMJ. 2015;351:h4451), observational data should not be used to determine treatment outcomes. In addition, our study highlighted that statistical adjustment (including propensity matching) cannot replace randomization - there are systematic differences in patients who do not receive bb in this cohort. We do agree with the reviewer that observational studies cannot replace randomised trials. However, their results are complementary, providing additional information to those of randomized trials. RCTs are not without limitations , the most common being often stringent patient selection, making questionable the representativeness of RCTs populations in routine practice . The example of digoxin is typical of this: for instance, in the DIG trial (a trial described as a “large simple trial”, in order to better reflect everyday life) patients were 63 years of age on average, i.e. more than 10 years younger than the average age of patients with atrial fibrillation seen in “real life”. For beta-blocker post-MI trials, the situation is pretty much the same, and most of the trials included patients much younger than those currently seen in everyday practice. Therefore, RCTs will provide answers in specific populations that may actually not reflect the way medications are used in everyday practice. We therefore consider that analysing observational data constitutes a valuable counterpoint to RCT findings. The limitations of such analyses, however, have been made clear in the Limitations section of the discussion, and we pretend in no way to provide a final answer to the question. Contemporary RCTs of beta-blockers in post-MI patients are currently organised, but their results won’t be known for many years at best. Abstract: "Thirty-day mortality was lower with early β-blockers" is incorrect. See above comments about ascertaining causality with observational data. In actual fact, you have found that those patients who were given bb by their clinician had a lower 30 day mortality than those patients not given or unable to tolerate bb. This mistake is again made further on, where you discuss whether persistence of bb "decreases" 5-year mortality. We have reworded the sentence on early mortality. We maintain that persistence of beta-blockers at one year was not associated with decreased mortality: this is purely factual. We have, however, changed the word “decreased” to “lower” to avoid any confusion. Introduction:

The first sentence should be better qualified; e.g. benefit shown in patients with heart failure and reduced ejection fraction in sinus rhythm (not just 'HF', which includes HFpEF and also AF patients - see Lancet 2014;384(9961):2235-43). We have reworded the first sentence to take into account the reviewer’s comment. Methods: It is unclear why only patients admitted to ICU were included in this study - surely that restricts the population receiving bb and reduces external validity? Virtually all patients hospitalised for AMI in France are admitted to ICUs. Only a few, extremely elderly patients, may get admitted to internal medicine or geriatric wards. The overall figures for patients hospitalised for AMI in France are similar when making an extrapolation from FAST-MI and when using the national hospital database (Tuppin et al. Arch Cardiovasc Dis 2009), giving support to the fact that FAST-MI is quite representative of the AMI population (hospitalised) in France. How robust follow-up for events was, needs to be discussed and referenced in the methods. Similarly, what was the reproducibility of data collection? The methods used for collecting and adjudicating follow-up events have now been described in more detail (see above). Was this tested, particularly for dosage and time periods on/off bb? What, if any, of this analysis was prospectively laid out or registered? In particular, you have excluded certain patients (e.g. Killip 1) and the reader needs to be made aware whether this was post-hoc or planned. Killip class I patients were included. We planned to determine the association of beta-blocker use and clinical outcomes at the different time-points a clinician may decide to use them. We therefore predetermined the 3 populations of the present analysis: the first population corresponds to the initial choice the clinician has when taking care of a patient with AMI, who has no history of heart failure and no clinical signs of heart failure when examined on admission. The second population corresponds to medication prescription at discharge: for this population, we also excluded patients who developed heart failure at any time during their hospital stay, and those with low ejection fraction. Finally, population 3 includes the subset of population 2 receiving beta-blockers at hospital discharge, and alive at one year, to address the issue of prescription maintenance in the long term. The exclusion of patients with a history of heart failure, Killip class >1 during hospital stay, or low LVEF was therefore planned when the current study was designed, in order to address specific clinical questions, as stated in the introduction: is beta-blocker treatment associated with improved survival in AMI patients without clinical heart failure and preserved left ventricular function and is the association with survival similar at different stages of the disease? Statistics: "cumulative hazard functions for each covariable were computed to assess proportionality" - in what way? were residuals not tested? What about interactions between variables in the models - how were these handled? Were adjustments made for multiple comparisons? A model inclusion value of p=0.05 is small and may have missed some important multivariate factors. There is a long list of various adjusted variables - in what way were the models and analyses pre-specified?

The text in this section is far too long and needs to be more concisely written (the whole PS section can be a method appendix). Was an analysis performed using a time-varying measure of beta-blocker use? We have tried to shorten this chapter by listing the different variables used as an appendix. As the editor requested more, and not fewer details on the propensity score analysis, however, we chose to leave the propensity score section in the main text. Should the editor prefer, however, we are quite ready to move it as an appendix as well. We did not make adjustments for multiple comparisons, as our main analysis was on all-cause mortality, and the other analyses were considered supportive. Likewise, analyses of the 3 different populations were considered different clinical questions in different groups of patients, and therefore we did not think statistical adjustment would be needed. We re-run the analyses using higher p values for inclusion and exclusion and the results were exactly similar (e.g. for the 5-year analysis, HR 1.19, 0.65-2.18). As stated, the variables used for the different analyses were pre-specified. Results: I would have preferred a much more simple analysis that looks at long-term outcomes only - showing no difference at 5-yrs bb vs not (irrespective of different treatment and risk) is a powerful statement that is currently lost. You could remove the statin section - this is distracting for the main analysis of beta-blockers (different prescription biases apply. See comments about causality. We postulated that beta-blockers might have differential effects depending on the timing of their administration from the acute episode of MI, and we thought that the main clinical question for beta-blockers in AMI patients was to determine whether they would still be useful after the initial months following the acute event. This was the reason for studying the association between their prescription and survival at different time points. We think the statin section is useful to show the difference in results with 2 classes of recommended medications, using a similar methodology. As shown here, there were striking differences between the 2 classes of medications and their association with long-term survival. Limitations: These major limitations of observational data are acknowledged, but do not appear to have impacted the way the authors have written their results. They are likely to regret the sentence "unlikely that major confounders were not recorded"; In fact the true benefit of randomization is to equalise both known and unknown cofounders and effect modifiers. The last sentence of the limitations does not take into account that not only is patient bias important, but also bias by physicians. We fully agree with the reviewer, and do not dispute the superiority of RCTs to balance known and unknown confounders. However, as discussed above, RCTs are not free of limitations. We think that the present study addresses a relevant question for physicians in charge of patients, for which final answers from contemporary randomised trials are not available. Pending the results of such contemporary RCTs on beta-blockers after AMI, our study is an endeavour to provide insight on their usefulness in the current era, granting all of the limitations described.

Tables: Too many tables with too many rows - recommend reducing to relevant variables. Tables that fit on one page are readable. Reviewer 3 actually requested more data on Tables 1 and 2. We feel that, in observational studies, providing a very detailed description of the populations is warranted, so that the reader can fully appreciate the differences between groups. Obviously, however, we would be ready to provide some or part of the Tables in the online material, should the editor prefer. Study flowchart: The flowchart is an essential component of the main manuscript and should not be relegated to the accessory data. The flowchart is incomplete - it should have clear details on those lost to follow-up or withdrawn from bb use at each stage with relevant reasons. Consider using a standard study flowchart model. We now propose to include the flow chart in the body of the text, as figure 1. Unfortunately, we do not have the reasons for treatment prescription/interruption. Additional Questions: Please enter your name: Dipak Kotecha Job Title: Cardiologist/Clinician Scientist Institution: Univ of Birmingham Reviewer: 5 Recommendation: Comments: Thank you for the opportunity to read this very well-written manuscript on a contemporary and important topic. The abstract was presented at ESC conference in 2014 and received much attention. We thank the reviewer for her positive appreciation of the manuscript. I have a few questions and comments: Why did authors use only FAST-MI 2005 data? There are data available from FAST-MI 2010 too at this point? The N in this paper I believe is a limitation since observational studies often require rather larger samples to make sure power is adequate to adjust for differences in characteristics. In that context, PS matching resulted in rather low N, so a larger sample size would be warranted perhaps? The central question of the study was the relationship of long-term outcomes with persistence of beta-blocker treatment. For the FAST-MI 2010 registry, only 3-year follow-up is available at present, which would mean a 2-year follow-up from the first year after the acute event. We thought this 2-year duration too short to actually assess the potential effect of treatment discontinuation in the long-term. In fact, all FAST-MI surveys have been carried at the very end of the corresponding calendar years; the method we use for collecting follow-up data takes considerable time as it is done

in several consecutive steps with several weeks/months between each step (city death registries, contact with physicians, repeated contact with patients and their family when no answer is obtained after first contact); after getting information, the events are reviewed by the critical events committee and further information is sought for relevant cardiovascular events or deaths before the final adjudication. We do agree with the reviewer, however, that a larger population would be appreciated, and we intend to do combined analyses, once 5-year follow-up will be available in the FAST-MI 2010 cohort (expected end of 2017/beginning of 2018). Surveillance: How often (i.e., which intervals) was this undertaken? How were endpoints during follow-up assessed? Validity? We have provided additional details on the methodology of follow-up (see above and methods in the revised manuscript). Follow-up was organised at 6 months, 12 months and every year thereafter. Beta-blockers during the first 48 hours after admission: Please specify when t=0 was defined for the survival analysis at first when you present this analysis. An immortal time bias will inventible be introduced if time is started at admission (pending that these analyses are not time-dependent), so think the correct thing is to start the time after 48 hours, excluding mortality cases within the first 48 hours. Moreover, did you consider to exclude patients with low EF, cariogenic shock, and low BP at time of admission because these patients may not be eligible for beta-blocker therapy? Those variables were not included in the multivariable model either, was that due to lack of that information? It appears a major limitation and makes the result a bit difficult to interpret. For population 1, we considered only patients without history of heart failure and without signs of heart failure on admission, in order to address the following clinical question: should physicians prescribe beta-blockers in patients without history or signs of heart failure and presenting with AMI. We did not exclude patients on the basis of LVEF for this first population as assessment of LV function is often made at a later stage. Patients in cardiogenic shock (and all those with Killip class >1) were excluded from this (and subsequent) populations. The point on immortal time bias is well taken; actually, we are confronted with a rather difficult choice: censoring patients who died during the first 48 hours might lessen the ability to demonstrate a benefit (or harm) of any given treatment (i.e. if a treatment given very early reduced or increased mortality during the first 48 hours), while including these patients is prone to immortal time bias. Therefore, we feel that there are pros and cons to both methods. However, we had done (but not reported) both types of analyses (with and without censoring patients dying early); in the analysis censoring patients who had died within 48 hours of admission, the adjusted HR for 30-day was 0.49 (0.26-0.92) for those receiving beta-blockers. These results were concordant with those of the analysis including patients who had died early. The KM curves after excluding patients dying early are as follows:

We have now added these results. Methods: Why backward selection? Were authors challenged on power? What did a model with all variables included show? The results were highly concordant whatever the type of analysis used (backward, or all possible). For instance, for the 5-year analysis, the HRs were 1.19 (0.65-2.18) for the backward analysis, and 1.22 (0.66-2.25) for the all possible variables analysis. We chose a backward procedure as the set of variables included in the analysis had already been chosen ad hoc on the basis of their potential physiological relevance. We did not do a formal sample size calculation, and the statistical power of our analysis can be questioned. We have added this limitation to the corresponding section of the discussion. Statistics: For the long-term analysis of population 3, assessment of beta-blocker status was done at 3 years. Did authors restart the analysis at this time-point to avoid immortal-time bias? I think defining the time when follow-up was started for all analyses could be clearer described to ensure readers it did not introduce immortal-time. For the main analysis, treatment at 3 years was not considered as a covariable and follow-up was started at one year. For the sensitivity analysis taking into account treatment at 3 years, only patients alive at 3 years were included. Information added in the results section. p-values < 0.05 as statistically significant: were they one or two-sided? P values were two-sided (added in the Methods section). Discussion: Overall it is excellent and particularly so the limitations section, but the authors may want to revise the initial statement with early beta-blockers, since I think removing deaths within 48 hours is the correct way of performing this analysis (which made analyses statistically insignificant). In that context, was there any data on when people were reperfused and if they were completely revascularized and did that influence the association of beta-blockers with mortality (following the hypothesis that beta-blockers may only be beneficial among people not promptly revascularized)?

We thank the reviewer for her comments. As stated above, the analysis censoring patients dying during the first 48 hours yielded concordant results, also statistically significant. Therefore, we preferred to keep the initial statement of the discussion as such. We performed an additional subgroup analysis on STEMI patients with or without reperfusion therapy. For one-year death (adjusted on propensity score for beta-blockers at discharge): In those with reperfusion HR 0.60 (0.13-2.75) In those without reperfusion therapy: HR 0.66 (0.19-2.22) For death at 5 years, in patients alive at one year: In those with reperfusion HR 0.76 (0.12-4.79)(adjusted on propensity score for beta-blockers at one year) In those without reperfusion therapy: only 22 patients with no beta-blockers at one year, and no death in this subset; 22 deaths in the 196 patients on continued beta-blockers at one year. Because of the small size of the subgroups, and the very wide confidence intervals, we did not add these analyses in the text. It does not seem however, that the benefit of beta-blockers was greater in STEMI patients without early reperfusion therapy. Additional Questions: Please enter your name: Charlotte Andersson Job Title: Post-doc in CV epidemiology and resident in cardiology Institution: Glostrup Hospital, Denmark Reviewer: 6 Recommendation: Comments: This is a very interesting study and the paper is well performed. Puymirat et al. used data from the nationwide French registry of ST- and non-ST-elevation MI, including 2,679 consecutive AMI patients without heart failure or left ventricular dysfunction treated in one of the 223 centres in October 2005. The statistical methods are well reported and well done. Some minor comments/suggestions for improvement of the manuscript: • Table 1: risk factors, n (%): family history of what? (CVD?) • eFigure1: this is a very helpful figure when reading the paper to understand the three populations studied. Therefore, I would suggest including this figure in the main paper (as Fig1) and make one figure (as Fig2) by combining Fig1, Fig2, and Fig3. We thank the reviewer for her positive comments. Table 1: family history corresponded to Family history of premature cardiovascular disease. Added. We thank the reviewer for the suggestion on eFigure 1, which we now propose to include in the main text as Figure 1, and we have combined Figures 1 to 3 into a single one.

Additional Questions: Please enter your name: Dr. Ute Amann Job Title: Research Associate Institution: MONICA/KORA Myocardial Infarction Registry, Central Hospital of Augsburg Reviewer: 7 Recommendation: Comments: Review of the manuscript : β-blockers and mortality after myocardial infarction in patients without heart failure: multicentre prospective cohort study. Journal : BMJ Impact Factor : 19.967 Background of the present manuscript. This paper is an analysis performed on a cohort of 2,679 Acute MI patients hospitalized in 223 Coronary Care Units participating in the French program the FAST-MI registry . This is a snapshot of patient care in the French Health Care system, performed during one-month at the end of the year 2005 (in absence of a national registry). For information, this program is conducted every five years, and the results of 2015 are pending. The subject is an important and contemporary subject: the safety and efficacy of beta-blockers therapy after myocardial infarction. Indeed, the benefit of long-term treatment of beta-blockers in post myocardial infarction was introduced in 1960 from clinical trials conducted before the advent of modern treatment of reperfusion (thrombolysis and angioplasty) and advances in drug therapy. Initially the European recommendations (European Society of Cardiology, ESC) assigned a Class I recommendation in the first year of treatment for the use of BB in post MI or post SCA. But in the absence of modern large-scale randomized trials, these data older than 30 years have resulted in an indication of a treatment often prescribe to life in post IDM patients. In addition, the contemporary treatment of myocardial infarction and the findings of some studies make questionable the risk / benefit ratio of treatment on chronic BB. Indeed the modern treatment of MI by reperfusion led to a dramatic decrease in the past 15 years, of the 1-year mortality from 15% to 5% with a decrease in myocardial injury and rates of heart failure. In addition several post hoc analyzes of randomized trials or registry (REACH) suggest that there is no benefit in prescribing BB after a year since the event curves remain parallel after the initial period. If there is a profit, it is driven by the patient risk group of patients with chronic heart failure. Considering the lack of reliable data, the latest recommendations from the ESC 2014 (STEMI) and 2015 (NSTEMI) no longer considers the prescription beta blockers should be systematic after a myocardial infarction. The only current recommendation is a class IIa B recommendations in 2014

for the continuation of BB after the initial hospitalization for the treatment of myocardial. In the recommendations of 2015 it is written that the usefulness of BB in patients with MI uncomplicated by heart failure is unknown in view of the absence of randomized controlled trial. Besides the lack of efficacy after the initial period, several drawbacks accompanied by a BB long-term treatment: • Many side effects can significantly impact the quality of life of patients such as dizziness, excessive fatigue, blurred vision, cold hands and cold feet, trouble sleeping (insomnia), lack of sexual desire (loss of libido), depression and male erection dysfunction. • In addition the drug interactions and potential complications of overdose, such as conduction disorders, can lead to fainting and implanting a pacemaker especially when patients develop kidney failure, are a concern in long-term use of these drugs especially in the elderly (the median age of patients at the time of infarction being about 60 years). Our group is conducting a randomized multicenter, national study evaluating open the interruption of BB in patients who had a myocardial infarction uncomplicated back at least 6 months in 3700 patients. While our study (the ABYSS trial, NCT pending) hasn’t started yet registry data on the subject are warranted. General Comments: The main outcome measures in this manuscript were : - 30 days mortality in relation with early use of β-blockers (≤48 hours of admission) - one year mortality in relation with hospital discharge prescription - and 5 years mortality in relation with one-year β-blockers prescription The rates of patients lost to follow-up for vital status were 0.3% at one year, and 4.1% at five years which is rather good. Authors found a strong benefit of BB prescription in the hospital phase on early mortality (30 days) (OR 0.46) however they found a non significant trend toward lower mortality at one year (OR 0.77) and a non significant trend toward an increase in mortality at 5 year (1.28). They conclude that early β-blocker use was associated with reduced 30-day mortality in AMI patients, while discontinuation of β-blockers at one year is safe and question the utility of prolonged β-blocker treatment after AMI in patients without heart failure or left ventricular dysfunction. This is a great registry analysis adding value to the debate of the benefit or the absence of benefit of prolonged BB therapy in MI and authors should be congratulated and this paper deserves publication in the BMJ however several remarks need to be addressed. We thank the reviewer for his encouraging comments. Comments 1) the statins results should not be stated in the abstract. And its added value in the manuscript is debatable, I would withdraw such results from the manuscript. Because observational data linking treatment with outcomes is subject to (potentially important) bias, we think the statin analysis, which was performed using a methodology similar to that we used for beta-blockers and led to extremely different results, reinforces the main finding, i.e. a lack of

significant relationship between discontinuation of beta-blockers during the first year, and late mortality. 2) More data could be displayed in the abstract such as the rate of BB prescription within the first 48 hours, at discharge and one year. We followed the reviewer’s comment by adding the data in the abstract. 3) The abstract should state what is the percentage of the AMI of the FAST-MI registry where considered to be without heart failure or EF <40%. The % has been added as asked. 4) The results for the one-year mortality and 5-year mortality could lack of power. Was a sample Size was calculated before to see if there is sufficient power to draw conclusion on it ? This need to be discussed in the discussion and added in the limitation paragraph. We agree with the reviewer’s comment. We did not do a formal sample size calculation, and added this limitation to the corresponding section of the discussion. 5) Is there any data on Heart rate control with or without BB? Any data on rehospitalization and Heart failure? Any data on side effects? This also need to be discussed in the papers and stated in the limitations. We previously reported an analysis on heart rate in the FAST MI registry (Seronde et al. Am J Med 2014), showing that there was no interaction between beta-blocker and heart rate, for the prediction of 5-year survival. Adding discharge heart rate to the covariables used to determine correlates of one-year mortality in population 2 yielded a similar HR (0.73, 0.43-1.26, P=0.26), compared to the HR without discharge heart rate in the model. The reason for discontinuation of beta-blockers (and in particular, the presence of side-effects) was not recorded. 6) Could the authors give the percentage of patients with the target dose within each population as we know that most of physician prescribe < than 25% of the targeted dose, probably to avoid side effects and increase tolerance, while maintaining the benefit We now have added the dose of beta-blockers prescribed at discharge in Table 1. 7) Title : « multicentre prospective cohort study “ should be deleted. We agree that “multicentre cohort study” can be deleted, should the editor prefer. 8) In patients without BB prescription, was they any prescription of Calcium Inhibitors ? ivabradin ? please comment. Calcium channel blockers were used at discharge in 42% of patients without beta-blockers, and 9% in those with beta-blockers. Nearly all calcium channel blockers used in patients with beta-blockers were dihydropyridines, while heart rate lowering CCBs were used in 26% of the patients without beta-blockers at discharge. CCB prescription at discharge was used as a covariable in the analyses. Ivabradine was not used in France in 2005.

Additional Questions: Please enter your name: Johanne SILVAIN Job Title: Associate Professor Institution: University Paris 6 (UPMC) - ACTION study group - APHP