Re: Adverse outcomes following trimethoprim for urinary tract infection in patients over 65: a cohort study. Manuscript ID: BMJ.2017.

Rubin Minhas BMJ Associate Editor BMJ, BMA House, Tavistock Square, London, WC1H 9JP Non-Communicable Disease Epidemiology Direct line: +44 (0)20 7927 2922 Email: kathryn.mansfield@lshtm.ac.uk Monday 18 th September 2017 Dear Dr Minhas, Re: Adverse outcomes following trimethoprim for urinary tract infection in patients over 65: a cohort study. Manuscript ID: BMJ.2017.040349 Thank you for your interest in our paper and for giving us the opportunity to respond to reviewers comments. We have made a of changes in response to these points which we feel have substantially improved the. We address the points they raise on the following pages. We look forward to hearing from you again. Yours sincerely, Kathryn Mansfield Faculty of Epidemiology and Population Health E: kathryn.mansfield@lshtm.ac.uk

Manuscript committee meeting comments Ed C1 Question well defined, adequately addressed. No major issues. Thank you. Ed C2 Some clinicians more familiar with combination drugs internationally. We agree, please see our response to Reviewer 3, Comment 2. Ed C3 Same comments regarding combination drugs. Please see our response to Reviewer 1, Comment 3 and Reviewer 3, Comment 2 on combination antibiotics. Ed C4 Ciprofloxacin may be used for complicated UTI's We agree, we have added some additional discussion regarding this. Please see our response to Reviewer 1, Comment 3. Ed C5 Selection bias may invalidate the findings, authors should comment Please see our response to Reviewer 1, Comment 7. Reviewer 1: Suzanne Geerlings R1 C1 This is a nice study, which adds new information to what is already known published literature, since most studies are performed in patients using trimethoprimsulfamethoxazole instead of trimethoprim. We can discuss whether this work is of importance to general readers, because trimethoprim is mostly prescribed by general practitioners to treat uncomplicated urinary tract infections. The Research Question is clearly defined and appropriately answered. Thank you 2

R1 C2 The overall design of study is adequate. However, as mentioned introduction that co-trimoxazole results into rapid renal function deterioration needs more nuance, since an increase in creatinine is not the same as renal function decline, because trimethoprim reduces tubular secretion of creatinine. Furthermore, it is not already described introduction (only Clinical implications paragraph of the Discussion) what the current guidelines recommendations for treatment of uncomplicated urinary tract infections UK are, since I do not think that with the recent resistance percentages for the common causative uropathogens cefalexin and amoxicillin are still recommended. Thank you for suggesting this. We agree with your point about the important difference between rise in creatinine with trimethoprim, and AKI However, we feel that this nuance is better left to the discussion where we do address it. We have changed the wording of the introduction to make it clear that we are merely stating that a link between co-trimoxazole and AKI has been made previously. We have also included additional text Introduction to outline the current UK guidelines for treatment of uncomplicated urinary tract infections and how these relate to our comparator antibiotics. Page 4 To limit confounding by antibiotic indication we restricted our analysis to patients with antibiotic prescription for the same indication (UTI), and examined the risk of adverse outcomes in patients prescribed trimethoprim and four comparison antibiotics (amoxicillin, cefalexin, ciprofloxacin and nitrofurantoin). However, even when treatment is restricted to the same indication, different classes of antibiotic are prescribed in slightly different clinical scenarios. For simple UTIs in adults, current UK guidelines recommend nitrofurantoin as first-line treatment (except for patients with poor renal function) and trimethoprim as a first-line alternative where there is low risk of microbial resistance.[11] Ciprofloxacin and cefalexin are not currently recommended treatments for simple UTIs (although ciprofloxacin is a first-line option for pyelonephritis) but are used as determined by resistance to first-line antibiotics. In addition, ciprofloxacin and cefalexin were used in practice as treatment for simple UTIs during the years covered by this study.[12] We aimed to address the impact of these different usage patterns on the outcomes through adjustment for a range of comorbidities, and sensitivity analyses. 3

R1 C3 R1 C4 The Participants and Methods used are adequately described and their conditions defined, but I have a few additional questions: In the Exposure paragraph the authors mention that they exclude prescriptions for co trimoxazole and coamoxiclav as UK these choices suggest more severe or atypical infections. But is this not also the case for ciprofloxacin? Concerning the outcomes: Why did the authors not describe the outcome of the treatment of the UTI itself, since a worse outcome might have had influence on the development of the endpoints of the study We agree that this is the case, although as now outlined introduction, antibiotic prescribing for UTI has historically not followed guidelines with particular concern about overuse of quinolones. It is possible that the association of ciprofloxacin with AKI was due to treatment of more severe infection. However, quinolones have previously been associated with AKI and we have shown this again, despite no association with mortality (as might be expected if there was a strong link to use for more severe infections). Nonetheless we agree that we need to make it clearer that the association may be due to confounding and have added this to the discussion of the ciprofloxacin result. We agree that a worse outcome of UTI, such as the development of pyelonephritis or sepsis, would have influenced the development of all three endpoints we studied (death, AKI, or hyperkalaemia). We also agree that including a descriptive account of these intermediate outcomes might have been an interesting means of exploring potential mechanisms for the associations we observed. However, we felt that this was beyond the scope of this paper. Page 11 Our finding of an association of ciprofloxacin with AKI is similar to another populationbased study that showed that users of fluoroquinolones had a 2.18-fold (95% CI 1.74 2.73) higher rate of AKI.[28] However, ciprofloxacin is recommended for treatment of pyelonephritis, so an association with AKI may be due to confounding by severity of infection. R1 C5 Concerning the covariates: Baseline renal function was defined using the most recent biochemical test results, which limit in period before the UTI was defined? Our paper investigated the effect of trimethoprim on severe outcomes that have been implicated in association with co-trimoxazole in previous studies. Since intermediate outcomes of UTI would be on the causal pathway between antibiotic choice and the three outcomes we studied we felt that adjusting for them in an analysis would be inappropriate. We identified the most recent renal function measure recorded any time before UTI diagnosis. We have added some text to the Methods Section to clarify this. Page 6 Baseline renal function was defined using the most recent biochemical test results recorded in primary care at any time prior to each antibiotic-treated UTI. 4

R1 C6 The Results answer the research question, are credible and well presented, but I have a few additional questions: The authors mention paragraph about the study population that the majority of cases of hyperkalaemia was identified from primary care biochemical test results, but I think that in most patients with a UTI potassium is not measured. We agree, most patients with a simple UTI are unlikely to have a test for potassium during or soon after the infection. It is likely that most blood tests would have been undertaken in patients having routine blood tests (e.g. for monitoring of diabetes) or those who were most unwell. However, UTIs treated with trimethoprim were associated with lower comorbidities, suggesting differentially lower levels of detection in patients treated with this drug, and biasing the association of trimethoprim with hyperkalaemia to the null (and suggesting that our findings of increased odds of hyperkalaemia in those on trimethoprim compared to amoxicillin may be an underestimate). This would not be the case if GPs were testing patients treated with trimethoprim for hyperkalaemia but, since this has never been recommended, we do not believe this to have been the case. As there is already discussion Strengths and weaknesses Section that bias in testing may alter the results in either direction we have not made further changes at this stage. 5

R1 C7 R1 C8 R1 C9 How do the authors know that this was not a selection of patients, which could have influence the results (and conclusions), since it could also be possible that clinicians were more cautious to prescribe trimethoprim se patients, which could have underestimated the effect. On the other hand, sensitivity analysis the authors mention that after repeating the main analysis excluding patients without known baseline renal function minimal differences were found results, but the data are not shown. This is a very important analysis and could have influence in both directions, since the patients without a known baseline renal function could be a more healthy group, which could have decrease the found effect, or the awareness of the clinicians in patients with known renal function disturbances could lead to an underestimation of the effect. Can the authors show these data? Discussion: The authors describe a calendar period between 1997 2015, the results in Table S1 make clear that the incidence of AKI and hyperkalemia increases in this period. The authors do not discuss this point ir discussion and conclusion. Can they explase findings? We agree, and did mention this Limitations Section, although we think it unlikely that this would have influenced prescribing behaviour earlier on in the cohort period. We have made this potential source of bias clearer. As requested, we have included these sensitivity analyses in Supplementary Table S2. They make no difference to the results and we think that this is because in UK primary care testing is focussed on patients with risk factors for CKD. We have previously shown that the proportion of unmeasured CKD in CPRD is small and similar to that in a nationallyrepresentative survey (Health Survey for England). 1 We have added some additional text to the Discussion to comment on changes over time. Please also see our response to Reviewer 2, Comment 3. Page 10 Table S2 (Appendix) As trimethoprim was less frequently prescribed for patients with urological pathology, this would likely have led to underestimating the odds of adverse outcomes, particularly AKI, for trimethoprim compared to the true result. Similarly, clinicians may have been cautious in prescribing trimethoprim to those at highest risk of AKI and hyperkalaemia, again leading to an underestimation of the true risk of adverse outcomes, particularly for those taking RAS blockers. However, the strongest evidence of adverse outcomes in association with trimethoprim use for those taking RAS blockers was only published towards the end of the duration of this cohort.[2,20] 6

R1 C10 R1 C11 R1 C12 R1 C13 R1 C14 Results in Context: It is known from clinical practice that clinicians not think that AKI is only the result of the sulphonamide part of the antibiotic as is written by the authors. Interpretation and conclusions are derived from the data, but the authors can only conclude that this is true for patients with a known renal function or they must first show data of the patients without known baseline renal function. Also it is not mentioned that only adults aged 65 years and over were studied. References are up to date and relevant and most important recent study about only trimethoprim [ref 6] and these side effects, is also added. The Abstract/summary/key messages and What this paper adds accurately reflect what the paper says. The research design is a cohort study, but I did not find a relevant checklist (PRISMA, STARD etc). We agree that this is not clear. We have therefore edited the Results in Context section to clarify. In response to your previous comment (Comment 8), we have included the results of analyses excluding individuals with unknown baseline renal function in the appendix to address any concerns about the approach taken main analysis (i.e. including those with unknown renal function). We have also added some additional text to the Conclusions Section to clarify that are results are limited to the over 65s. Thank you. Thank you. We can only apologise for this. We did complete a STROBE-RECORD checklist for this study, but for some reason it was not uploaded with our first submission. We have uploaded the checklist with this submission. Page 11 Page 12 RECORD extension to STROBE checklist included in submission. This is an important distinction as the sulphonamide antibiotics (including sulfamethoxazole) have been long recognised to be associated with a substantial risk of acute renal impairment, which could have been assumed to be causal.[27] Our results demonstrate that trimethoprim is associated with greater risk of AKI and hyperkalaemia compared to other antibiotic treatment for UTI, among the general population over 65 years of age and not just those treated with RAS blockade. However, this is not associated with an increased risk of death. 7

Reviewer 2: Bruce Guthrie R2 general Thank you for asking me to review this paper which examines the risk of acute kidney injury, hyperkalaemia and total mortality period immediately following antimicrobial prescription for urinary tract infection, focusing on trimethoprim. The paper follows analyses in other datasets of risks associated with co trimoxazole and case control studies of the risks of trimethoprim exposure and mortality. The paper clearly lays out some of the problems of the previous literature (notably the risk of confounding by indication). The analysis is carefully done including an impressive of sensitivity analyses including a propensity score matched method which considerably increases confidence conclusions. My comments are all relatively minor, although the first is more important in making the findings more useful to clinicians. Thank you. 8

R2 C1 I thought it very helpful that the authors quantify the absolute risk of using trimethoprim rather than amoxicillin (2 3 additional cases of AKI or hyperkalaemia per 1000 prescriptions), but I think it important that the abstract and the short paper version also include this. However, it would be more valuable if a wider range of absolute risks were presented. I recognise that this isn t straightforward, but the implications of the pretty stable odds ratios they estimate will vary considerably. For example, although the OR is stable main analysis and in all but one of the sensitivity analyses (of which more later), the authors themselves point out that the absolute baseline risk will be higher in people taking RAS blockers (OR 2.22 for hyperkalaemia) and RAS blockers and potassium sparing diuretics (OR 2.08 for AKI and 6.17 for hyperkalaemia) (table S2). So the citing of a single whole population absolute risk increase but a pretty stable OR for trimethoprim exposure may imply that the absolute risk in individuals is pretty stable. Clearly there are too many subgroups to take this to extremes but one third of UTI episodes are in people on RAS blockers so this is definitely a non trivial subgroup, and one already highlighted discussion (obviously will be many fewer on RAS + KSD, although many of them will also have heart failure which is again a significant risk factor for AKI OR 1.65). Similarly, although the estimated OR for sudden death is similar to the previous case control studies (but not statistically significant), it would also be interesting to know what the estimated absolute risk, although I accept the authors may not think this is appropriate to calculate (I would guess it is pretty small). Some or all of those categories are suggested by the near final line: other prescribing options and monitoring of renal function and potassium levels should be considered for patients at high risk due to renal impairment, RAS blockade with potassium sparing diuretics and diabetes. Although I would add RAS alone and cardiac failure. Thank you, this is a very important consideration. We have calculated absolute risks within drug-related subgroups and present these Abstract, and the Discussion. We have also edited the Discussion where we present absolute risks to clarify that the crude absolute risk figures we present refer to antibiotictreated UTI episodes rather than patients. However, our strict definition of UTI may limit direct comparability of absolute rates from this study to the general population. 9 Page 2 Page 12 Abstract: Our results suggest that, for 1,000 antibiotictreated UTIs among people over 65, treatment with trimethoprim instead of amoxicillin would result in one to two additional cases of hyperkalaemia and two admissions with AKI, regardless of RAS blockade. However, for people taking RAS blockers AND spironolactone treatment with trimethoprim instead of amoxicillin would results in eighteen additional cases of hyperkalaemia and eleven admissions with AKI. Discussion: Given the different rates of outcomes within subgroups of the population, it is useful to consider absolute and relative risk together by calculating the excess of events attributable to trimethoprim treatment, although our very strict definition of a UTI means that these estimates should be treated with caution. As an example, our results suggest that for 1,000 antibiotic-treated UTI episodes in those over 65 years of age not taking RAS blockade, treatment with trimethoprim, instead of amoxicillin, would result in one additional case of hyperkalaemia and two of AKI. Among those taking either an ACE Inhibitor/ARB or spironolactone the figures would be very similar; two additional cases of both hyperkalaemia and AKI. However, treatment with both an ACE Inhibitor/ARB and spironolactone would result in eighteen additional cases of hyperkalaemia and eleven of AKI. A small increased absolute risk of a rare outcome (such as general population) from trimethoprim may be acceptable when set against a need for multiple treatment

R2 C2 Linked to this, there is always a balance between benefit and risk, and the reference group of amoxicillin isn t necessarily that helpful here as it is pretty rarely used to treat UTI currently (or ever study data). Cephalexin for example, is safe from the perspective of the outcomes examined, but actively discouraged by guidelines because of concerns about C diff and antimicrobial resistance in an antimicrobial group which is commonly used to treat severe infections in hospital. That kind of discussion might be more for an editorial though as goes well beyond the scope of the paper, but is a problem in real world decision making. We agree, but chose this study design in order to make our results directly comparable to previous research where amoxicillin was the comparator. Odds of adverse events compared to other antibiotics is possible through the data summarised in Figure 2. We absolutely agree that the overall translation of risks and benefits from observational studies into clinical care is complex. We hope that by analysing a range of antibiotics, and presenting the data as clearly as possible in a clinically useful way, that our results will be helpful in developing future guidance. options for patients with allergy to other antibiotics or bacterial resistance patterns. While AKI and hyperkalaemia may result in avoidable morbidity and hospital admission, it is reassuring that we have not identified an increased risk of death, suggesting that there is appropriate response to these outcomes. However, other prescribing options and monitoring of renal function and potassium levels should be considered for patients at high-risk of AKI and hyperkalaemia such as those with renal impairment, RAS blockade with potassium sparing diuretics, and diabetes. 10

R2 C3 R2 C4 R2 C5 In the limitations, can the authors comment on the very large increase in AKI and hyperkalaemia ascertainment over time (the OR for AKI later period vs the earliest is 15.2 and for hyperkalaemia 3.7)? Although they discuss that knowledge of the association might alter testing behaviour, they don t comment on this. It seems unlikely that the increase is due to that knowledge since OR rise steadily over the entire period, but if it s not an ascertainment problem then what is it? And if it is an ascertainment problem then could it cause bias particularly because antimicrobial choices have also changed over time. I agree with the authors that their design better accounts for confounding by indication, and I think that claim in strengths in limitations is consistent with the rather higher OR of adverse outcomes seen sensitivity analysis for Antibiotic exposure defined as antibiotic prescription for any indication as used in many previous studies. It is very helpful indeed for the authors to provide the complete set of codes uses as an appendix. Clearly this is only for the enthusiast but is essential for transparency and replication. One very minor comment is that the drug tables in that supplementary material are not well formatted. I think they are aligned centrally vertically, so not always easy to tell which word in drug names for example belongs to which row. So changing the alignment or putting the lines between rows in would make them easier to read. We have inserted a paragraph and relevant references to the Discussion section to address this, although in-depth discussion seems beyond the scope of the paper. Increasing familiarity and clinical focus on AKI has led to markedly increased coding of this diagnosis over time, particularly the dramatic rise since 2009. We also feel that it is likely that the much smaller rise odds of hyperkalaemia is due to increased testing (driven partly by QOF) of routine blood tests in people at high risk of hyperkalaemia diabetics and those with CKD as well as direct linkage between pathology systems and GP records. Thank you. Thank you, we have now archived all our code lists on our institutional server so that they can be downloaded if necessary (rather than include all code lists as supplementary material). We have included a DOI link to the material paper. Page 11 Page 5 This cohort covers the years 1997-2015, and we have shown a progressive increase odds of developing hyperkalaemia and AKI after a UTI in more recent years. The increase in hyperkalaemia may be due to an increased rate of blood testing in primary care (particularly among groups at risk of high potassium levels, such as diabetics and those with CKD) or improved automatic recording of test results in GP records. The marked increase in AKI over time as defined by HES coding is well established and likely to be predominantly related to increased clinical focus and the adoption of consensus definitions defined by changes in creatinine levels.[33,34] By adjusting for calendar period we aimed to minimise the influence of these secular trends on our results. All morbidity code lists are available in to download (https://doi.org/10.17037/data.229), and were either developed for use in other studies, or were developed in a consensus procedure by two authors with clinical experience NHS. 11

R2 C6 Table 1 has a ** by renal function with no footnote that I can see. Reviewer 3: Louie Mar A. Gangcuangco R3 - general R3 C1a R3 C1b R3 C2 R3 C3a This is a clinically relevant that addresses the question whether trimethoprim is truly associated with increased 1. death 2. hyperkalemia and 3. acute kidney injury among older patients with urinary tract infection. The strengths of this is the large sample size of the cohort and the relatively long study duration. Below are some of my comments/inquiries to further improve the : 1. ABSTRACT: 1a. Participants section, clarify "the primary care diagnosis of UTI." Although it has been later mentioned that patients with pyelonephritis were excluded, it would be good to clarify this Abstract. 1b. Results: The last sentence of the Results section needs to be re written for clarity. It must be emphasized that the odds of death was not statistically significant, both general population and among patients on RAS blockade. 2. "What this Paper Adds Section" 2a. The phrase "a component of co trimoxazole 'that is OFTEN used alone as an antibiotic" must be rephrased as "a component of co trimoxazole that MAY be used alone." This is because in some parts of the world (such as the Philippines), it is extremely rare for clinicians to prescribed trimethoprim alone without sulfamethoxazole. Although citations have been provided, please include a clarifying statement current how a database that includes only 7% of the UK population is "largely representative of the UK population." Thank you for noting this, this was a legacy from a previous version of the table, we have deleted the asterisks appropriately. Thank you. We agree, our definition of UTI was not straightforward. Given its complexity, we felt that it may be misleading to present a summary Abstract and have therefore elected to only include a detailed UTI definition main. Please also see our response to you later comment (R3 C5b). We believe we have reported our results correctly in keeping with the STROBE checklist. Statistical significance could be inferred from the confidence intervals. We have edited the What this paper adds Section as suggested. In response to Reviewer 1, Comment 3, we have also added some additional text regarding prescribing guidelines for UTI UK, which may also help to address your concerns. We have added some additional text to the Methods Section to clarify that the CPRD population is representative in terms of age, sex, ethnicity, and body mass index. 12 Table 1 Page 3 Page 4 Whether risks are similar with trimethoprim, a component of co-trimoxazole that is used alone as an antibiotic in some countries, is not known. Included patients are largely representative of the UK population in terms of age, sex, ethnicity, and body mass index.[13-15]

R3 C3b Under Exposures section: the primary care comorbidity code only included "uncomplicated UTI." However, it was later stated that patients with renal calculi or structural defects were also included. Were pregnant patients excluded analysis? Is there data on recent urethral catheterization? Please clarify these inclusion/exclusion criteria of the patients, as they are technically patients with complicated UTI. We have used morbidity codes for uncomplicated UTIs to identify UTI (for example, excluding pyelonephritis ), and we have also identified individuals with a history (within 12 months) of renal calculi, urological malignancies, or structural anomalies to use as a covariate. Our study did not define uncomplicated UTI by excluding those with a past history of renal or urological disease, as we could not be sure that these problems remained current at the time of UTI diagnosis (for example renal calculi may have been successfully treated) and felt that adjusting for a past history was a more appropriate course of action in this situation. We did not specifically exclude pregnant women from the study as the study population were over 65 and we assumed that any pregnancies in this group would be unlikely and therefore have minimal impact on results. We have also not used data on recent urethral catheterisation as a potential exclusion criterion as, in the UK, these procedures are often undertaken by community nurses and therefore may not be adequately captured by primary care electronic health records. 13

R3 C4 What is the proportion of patients with AKI and hyperkalemia who were exposed to trimethoprim? Although this has been shown in Table 1, it would be helpful to spell this out Results section. The crude proportion of outcomes occurring following each antibiotic may be misleading without appropriate adjustment. The information is included in table 1 for descriptive purposes and we suggest this is sufficient. R3 C5a First sentence said "In older patients, regardless of "other medications" treatment with trimethoprim for UTI is associated with a 72% increase in odds of AKI" please change "other medications" to more specifically mention "regardless of RAS and/or potassium sparing diuretics." By taking into account the differences between patients prescribed each antibiotic, the adjusted odds ratios represent a more useful summary estimate of the association between the antibiotics and the outcomes. We have therefore not included these proportions in our results. We have added some text in parentheses as suggested as a caveat to this statement. We have also edited a similar statement further on the Discussion. Page 9 Our results show that in older patients general population, regardless of use of RAS inhibitors or potassium-sparing diuretics, treatment with trimethoprim for UTI is associated with a 72% increase odds of AKI and a greater than doubling of the odds of hyperkalaemia compared to amoxicillin. R3 C5b Paragraph 4: "limiting the study to "simple UTIs" the use of the phrase "simple" UTI is misleading, as some patients with anatomic abnormalities (such as calculi and prostatic hypertrophy) were included. We have edited this statement to clarify that we limited the study to simple UTIs as defined by morbidity coding (but that this may have included some individuals with a history of more complex urological pathology). Page 11 Page 10 Our results suggest that, regardless of other drug use (RAS blockers or potassium-sparing diuretics), trimethoprim (but not ciprofloxacin), is associated with a more than doubling of the odds of hyperkalaemia. However, there are some important limitations. While we attempted to capture only simple UTIs (defined using primary care morbidity coding, but not excluding those with a history of more complex urological pathology) in our main analysis we may have included patients with complex pathology, or other infections. 14

R3 C5c R3 C6a R3 C6b R3 C6c Were there patients who were initially treated with one antibiotic and then later on switched to another drug class because of persistence of symptoms and/or urine culture & susceptibility results (for example, patients initially on amoxicillin and later switched to trimethoprim)? How were these patients handled analysis? Is the proportion of patients with diabetes statistically significantly higher TMP group compared to the amoxicillin group? Instead of presenting Figure 2, I believe that Supplementary Table 1 (S1) should be part of the main, as Table S1 presents more information on the other significant variables (such as age, co morbidities, etc.). Table S1. It is interesting to see a trend in increasing odds ratio for AKI as the calendar period progresses. Is there any explanation why the adjusted odds ratio of calendar year 2012 2015 is 15.22 relative to 1997 2000? In the main analysis we included individuals who had a previous antibiotic prescription with previous 28 days. It is likely that in many of these the antibiotic was of a different class. However, in a sensitivity analysis (Table S2, Analysis 3) we excluded individuals with a prescription for one of the study antibiotics with previous 28 days and this made minimal difference to our results. While we very much appreciate your comments, we disagree on the potential value of adding p-values, particularly to simple descriptive statistics. A large of papers have demonstrated the limited value of p-values in this situation (e.g. Altman DG, et al. BMJ 1983, 2 Browner WS, et al. JAMA 1987, 3 Goodman SN, et al. Ann Int Med 1999, 4 Sterne JAC et al. 2001 BMJ, 5 Greenland S et al. EJE 2016 6 ). We would therefore like to avoid terms such as significant or non-significant as we feel it puts too much emphasis on the, arguably arbitrary, threshold value of p<0.05 to guide or interpret results. We agree that Table S1 shows many interesting results and is key to understanding the full results. However, we feel that it is too cumbersome as it stands to include main and is unlikely to meet the editorial requirements. Further, we feel that Figure 2 well illustrates the important findings of the study. However, we are happy to discuss this if the editors disagree. Please see our response to Reviewer 2, Comment 3. 15

R3 C5d Having "absent" renal function (no baseline renal function measurement) is associated with higher odds for AKI. Do you suggest that patients >65 with clinical diagnosis of UTI need Creatinine measurement prior to antimicrobial therapy? The absent renal function category probably includes a large of individuals with normal renal function (see Comment 8 in response to Reviewer 1) as well as an unknown of people with abnormal renal function who have limited interactions with their primary health care provider. Therefore, it is not possible to draw any meaningful conclusions from this result. However, we agree that an implication of the overall findings of our paper is that clinicians should know the baseline renal function and potassium level for a patient when they consider prescribing trimethoprim. References 1 Iwagami M, Tomlinson LA, Mansfield KE, et al. Validity of estimated prevalence of decreased kidney function and renal replacement therapy from primary care electronic health records compared with national survey and registry data United Kingdom. Nephrol Dial Transplant 2017; : 1 9. 2 Altman DG, Gore SM, Gardner MJ, Pocock SJ. Statistical guidelines for contributors to medical journals. Br Med J (Clin Res Ed) 1983; 286: 1489 93. 3 Browner WS, Newman TB. Are All Significant P Values Created Equal? J Am Med Assoc 1987; 257: 2459 63. 4 Goodman SN. Toward Evidence-Based Medical Statistics. 1: The P Value Fallacy. Ann Intern Med 1999; 130: 995 1004. 5 Sterne J, Davey Smith G. Sifting the evidence-what s wrong with significance tests? BMJ 2001; 322: 226 31. 6 Greenland S, Senn SJ, Rothman KJ, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur J Epidemiol 2016; 31: 337 50. 16