GENERAL COMMENTS

The question is clear, and the authors argue that data of more than a decade ago are appropriate for investigating the a priori hypothesis that multi-drug prescription influences optimal HF drug prescribing. However, in my opinion these data are outdated to answer the first question (that is, to describe multi-drug prescribing in the cardiovascular general practice population). It is to be expected that much has changed in 12-13 year in the general practice population, both regarding the population itself as well as regarding their treatment. This is also problematic for answering the second question (whether multi-drug prescribing influences optimal HF drug prescribing), especially since optimal HF drug prescribing has become more complex (as the authors themselves mention, the combination treatment of RAS/betablockers in HF has increased since the new guidelines became available in 2001). One can expect that the (negative) influence of multi-drug prescribing is larger when optimal HF drug treatment involves more drugs or more complex treatment regimes. I would strongly recommend to conduct the study using more recent data.

Other remarks:

1. The author mention this is a cross-sectional clinical data linkage study but in the methods it is not explained where the data came from (electronic medical records?) and how they were linked for this study.

2. In what way is this population (3 practices in North Staffordshire that are part of a research network) representative of a more general population (in the UK)?

3. Why was the reference group chosen of patients with at least 1 CVD drug prescription without a CVD diagnosis? What kind of population would this constitute (it is a substantial group!)? Are CVD diagnoses just lacking or are the drugs prescribed for non CVD diagnoses?

4. What was the underlying reason for categorising the multi-drug into four count categories: (i) 0 (i.e. no non-CVD drugs), (ii) 1 to 3, (iii) 4 to 6 and (iv) 7 or more? The choice of these levels may very well determine whether or not (significant) differences are seen between groups. I would prefer to see and compare the data first without categorisation.

5. Is it possible to conduct a more detailed algorithm of optimal therapy (see, for example, the GAI indices in the Komaja study) taking into account levels of HF?

6. To assess the impact of comorbidity on optimal HF drug prescribing, the chosen approach using adjustments on odds ratios comparing the HF failure group with a reference group is rather difficult to interpret. This may show that the influence of a multidrug regimen on prescribing recommended HF drugs is similar for HF and non HF patients (but that was not the question). I would prefer to have a guideline adherence measure within the HF population as primary outcome and assess the influence of the multidrug regimen on this outcome to answer the second question. This prevents that the multidrug effect is 'discounted' because it also occurs in the reference group.

7. The whole issue of comorbidity in HF (and thus multidrug therapy) and how this may affect not only HF treatment but also guideline recommendations is a hot topic at the moment. See for example, Page RL 2nd, Lindenfeld J. The comorbidity conundrum: a focus on the role of noncardiovascular chronic conditions in the heart failure patient. Curr Cardiol Rep. 2012 Jun;14(3):276-84. It would be good to address / discuss this using some of such recent publications.

8. The STROBE checklist only provides a Y for all items, where a reference to a page/line number –where appropriate- would be helpful. As far as I could see, some of the items were not (explicitly) addressed in the manuscript (see also some of the points above). The authors state: "This step wise adjustment was performed so that the influence of non-CVD multi-drug therapy on the observed associations could be identified." It is not clear how they actually do this. It seems that by entering first age, gender, depreviation, the 'influence' of multi-drug therapy could be underestimated. In the results it is stated: "Additional adjustment for non-CVD multi- drug therapy counts did not alter the associations between HF and any of the related drug groups." From the methods, it is not clear whether/how the authors planned to test that the associations were not altered.

GENERAL COMMENTS

After reading the point-by-point response, I got completely lost in the revised manuscript. The authors say they have changed several major issues, but I cannot see this reflected in the manuscript. As such, it appears that most of the comments have not been resolved in the manuscript itself.

General point: Selection of reference group As I understand now, there were 170 patients with HF and CVD prescription, 1246 patients with other CVD diagnoses and CVD prescriptions, and 1739 patients with CVD prescriptions without a CVD diagnosis. What is still not clear is whether these 1739 patients lack any linked diagnosis or have a non-CVD diagnosis linked to their CVD prescription (or both). The problem is that patients lacking any diagnosis are likely to include patients with HF, and therefore this group would be unfitting as reference group. The authors have not solved or clarified this problem adequately.

Moreover, the authors claim they have changed all analysis by excluding the group without a CVD label but -as far as I can see- this was not the case. They did NOT remove the other group from tables 1 – 3. Also, the supplemental table 1 does NOT present the three groups in comparison. Instead, it presents overlapping groups (reference n=1739, non-hf CVD n= 2985, HF n=170; that is 4894 patients instead of 3155 patients).

Point 1. Use of more contemporary data I agree –and already acknowledged- that historical data can allow for the testing of a priory hypothesis but my comment that the data are outdated to answer the first question has not been resolved. There was no a priory hypothesis for 'describing multidrug prescribing in the cardiovascular general practice population‘.

Point 2. Explaining the data linkage As I understand from the answer, drug prescription and diagnostic consultation data are both retrieved from electronic medical records as provided general practice network. There was not sentence added to this part to explain this better. The sentence: "This study used a large clinical electronic medical database linking diagnostic data to all prescribed drugs" still suggests that data were linked from different medical record sources (which is not the case, if I understand the answer correctly).

Point 4. Multidrug count categories The authors refer to a previous paper from one of the authors to defend their categorization. This other paper, however, used completely different cut-points for categorization (and also did not give any substantiation for this). N.b. Adding the mean without any information on distribution or (standard) deviation in Table 2 is not very helpful. I can imagine that a median with IQR might be more meaningful. However, given point 1, the value of Table 2 is limited.

Point 5/6. Optimal therapy algorithm It is a pity that no data on severity could be used and this is a limitation that could be better addressed.

Abstract was not changed at all.

Reference list contains duplicates.

Point 6/7. Adjusting for multidrug counts

To assess what the influence is of non-CVD multidrug use on an observed association, is a question of effect modification and not of covariate analysis. Adjusting on multidrug use in a linear model is not the correct approach to identify/assess an effect modifier.