Characteristics and limitations of national antimicrobial surveillance according to sales and claims data

Purpose Antimicrobial use (AMU) is estimated at the national level by using sales data (S-AMU) or insurance claims data (C-AMU). However, these data might be biased by generic drugs that are not sold through wholesalers (direct sales) and therefore not recorded in sales databases, or by claims that are not submitted electronically and therefore not stored in claims databases. We evaluated these effects by comparing S-AMU and C-AMU to ascertain the characteristics and limitations of each kind of data. We also evaluated the interchangeability of these data by assessing their relationship. Methods We calculated monthly defined daily doses per 1,000 inhabitants per day (DID) using sales and claims data from 2013 to 2017. To assess the effects of non-electronic claim submissions on C-AMU, we evaluated trends in the S-AMU/C-AMU ratio (SCR). To assess the effects of direct sales of S-AMU, we divided AMU into generic and branded drugs and evaluated each SCR in terms of oral versus parenteral drugs. To assess the relationship between S-AMU and C-AMU, we created a linear regression and evaluated its coefficient. Results Median annual SCRs from 2013 to 2017 were 1.046, 0.993, 0.980, 0.987, and 0.967, respectively. SCRs dropped from 2013 to 2015, and then stabilized. Differences in SCRs between branded and generic drugs were significant for oral drugs (0.820 vs 1.079) but not parenteral drugs (1.200 vs 1.165), suggesting that direct sales of oral generic drugs were omitted in S-AMU. Coefficients of DID between S-AMU and C-AMU were high (generic, 0.90; branded, 0.84) in oral drugs but relatively low (generic, 0.32; branded, 0.52) in parenteral drugs. Conclusions The omission of direct sales information and non-electronically submitted claims have influenced S-AMU and C-AMU information, respectively. However, these data were well-correlated, and it is considered that both kinds of data are useful depending on the situation.


Masaki Tanabe
Norio Ohmagari Response to Reviewers: Response to Reviewers We have made major revisions to the entire manuscript based on the reviewers' advice and suggestions. We are grateful for the opportunity to improve our work, and look forward to the reviewers' evaluation of the new structure and theories included in the manuscript. Our point-by-point responses to the comments are provided below: Reviewer 1 This is a very interesting exploration of differences between two comprehensive, population-based methods to observe antibiotic consumption in Japan. The authors are to be commended for performing such analyses in order to understand the available data better and in order to be able to deal with limitations. It is an example for researchers in other nations to do likewise.
Thank you for your time and effort in reviewing our work, and for your positive evaluation of our manuscript. We have made major revisions in accordance with your comments and suggestions.
1.The differences found are all by all minimal, which could incite the authors to conclude that Japan has 2 pretty reliable sources to monitor antibiotic consumption. The article should clearly state that this is true for consumption in primary care (and not hospital care). It this would be incorrect it should be stated, and it should also be indicated why it is not possible to make the distinction between primary care and hospital care dispensing and reimbursement.
Thank you for the advice. S-AMU and C-AMU include both primary care data and hospital data. However, S-AMU does not contain information about where an antibiotic is used, and therefore does not allow us to distinguish between primary care and hospital care dispensing/reimbursement. As advised, we have added the following information to the Data sources subsection of the Methods.
Page 7 Lines 106-109 However, S-AMU data do not include information on the purchasing parties (i.e., names and types of medical facilities) or patients, and therefore do not allow distinctions to be made between primary care and hospital care dispensing and reimbursement.
2.Some differences were observed but not statistically evaluated. Some explanation why this did not happen might be added (too small differences, complex reversing time lines).
As advised, we have added more statistical evaluations to our manuscript after consulting with a statistician.
Page 14 Lines 206-212 Effects on parenteral AMU For parenteral antimicrobials, both branded and generic drugs exhibited similar trends in S-AMU and C-AMU (Fig 2A). The use of generic drugs increased while the use of branded drugs decreased over the study period. Furthermore, the scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001) (Fig 2B). For both branded and generic drugs, S-AMU was higher than C-AMU throughout the study period.
Page 15 Lines 223-232 Effects on oral AMU The DID values of oral antimicrobials were more than 10 times higher than those of parenteral antimicrobials in both S-AMU and C-AMU. As with the parenteral antimicrobials, both branded and generic drugs in oral antimicrobials exhibited similar trends in S-AMU and C-AMU ( Fig 3A). Again, the use of generic drugs increased while the use of branded drugs decreased over the study period. Throughout this period, S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-AMU for generic drugs. The scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .996, P < .001,) and C-AMU (r = .997, P < .001) (Fig 3B).
3.The exploration of the small differences leads to hypotheses on potential sources of bias. Most of these hypotheses are plausible. For OTC sales of antibiotics in Japan, the authors states that it is illegal. That does not mean that it does not happens, even in Japan. Some indication about the actual practices in Japan, and some references to the international literature on this should be advisable. It seems plausible that this might have an influence on the observed difference in sales and claims of cheap oral antibiotics (where the remark on the legal status of OTC antibiotics was placed).
Thank you for suggesting this important point. OTC drugs are strictly regulated in Japan, and OTC antibiotics are, for all intents and purposes, not available (excluding some topical preparations). We would be hard pressed to think of any situation where a pharmacy would risk losing their retailing license and face criminal charges for illegally selling antibiotics that are easily and cheaply available with a physician's prescription. In addition, physician consultations/prescriptions are highly affordable under Japan's universal healthcare system. We have searched the international literature on the plausibility of OTC AMU in Japan, but could not find any relevant papers. In addition to the strict legislation on the sale of antibiotics, we think the main reason for this being such a non-issue is simply that the barriers to legally obtaining antibiotics are too low for illegal sales to be worth pursuing.
Although a very small number of personal trades/sales of antimicrobials through websites may occur, such AMU would not be reflected in sales or claims data. We have added these explanations to the limitations as follows: Page 20 Lines 317-324 First, the analysis provides insight into S-AMU and C-AMU in Japan, but the characteristics and limitations of these data sources may not be directly applicable to those of other countries. For example, other countries may have a non-negligible proportion of antimicrobial sales without prescriptions [23,24], which can lead to underestimations of C-AMU. In contrast, the inclusion of over-the-counter antimicrobial sales is not an issue in Japan due to strict regulations and easy access to prescribed antimicrobials. In this way, our findings should be evaluated in the context of Japan's healthcare system and AMU patterns.
We have also cited the following additional references. The topic of the paper is of utmost importance when it comes to monitoring AMR. However, I'm under the impression that the work has been hastily prepared. The paper in my view lacks clarity and the methods does not appear sound, unless the authors are dealing with data that lack granularity. In general, thru my read, I'm left wanting more. The sources of my disappointment are as follows:

23.Batista AD, A Rodrigues D, Figueiras
Thank you for your time and effort in reviewing our work, and we appreciate your advice and comments. The manuscript has undergone a major restructuring.
We estimated C-AMU using data obtained from the National Database of Health Insurance Claims and Specific Health Checkups of Japan (NDB), which is managed by the Ministry of Health, Labour and Welfare (MHLW). The NDB contains national-level health insurance claims data for healthcare encounters covered by insurance, and these data can be used for research purposes following the submission and approval of an application to the government [12]. Our dataset included data on the dates and types of antimicrobials prescribed to patients for all insurance-covered healthcare encounters. In Japan, all residents are required to enroll in health insurance, which entitles them to receive insurance-covered healthcare at any medical facility throughout the country. Enrollees pay monthly premiums to their insurers, and also pay a fixed proportion (10-30% depending on age and income) of the medical charges at the point of care. Healthcare providers send claims to the applicable insurers to be reimbursed for the remaining charges. Because insurance-covered care accounts for the majority of medical treatments provided in Japan, the NDB represents a near-comprehensive database of all treatments performed throughout the country. However, the NDB does not include claims data from patients with fully publicly funded healthcare (e.g., patients with intractable diseases, atomic bomb survivors, patients on welfare, patients with tuberculosis, and patients with human immunodeficiency virus infections) and patients who personally pay for all of their medical expenses (e.g., foreign travelers and cosmetic surgery patients). The detailed numbers of these patients are not published by the government. Furthermore, the NDB only includes data from claims that have been submitted electronically, and the lack of paper-based claims reduces its coverage of healthcare encounters. Under Japan's healthcare system, claims for medical care, dental care, and drug dispensing (by pharmacies) are handled separately. From 2011, electronically submitted claims for hospital-based medical care and drug dispensing claims have accounted for over 99.9% of total claims [13]. For clinic-based medical care, the proportions of electronically submitted claims increased from 91.0% in 2011 to 99.9% in 2017 [13]. For dental care, the proportions of electronically submitted claims have increased dramatically in recent years (31.5%, 46.4%, 55.7%, 69.5%, and 96.0% in 2011, 2012, 2013, 2014, and 2015, respectively) [13].
2.Methods look at first glance proper including potential time-series correlations but with limited granular data and its understanding, the authors are left with lots of hypotheses with some sense of triangulation of information/data We agreed with your opinion, and have revised our study's conceptual framework and development. Previously, we had, as you pointed out, aimed to explore the biases of S-AMU and C-AMU from various aspects. But we have changed this construction: We now consider the effects of the rapid shift in paper-based claims to electronic claims in dental care as a characteristic of C-AMU, and the effects of insufficient data coverage as a characteristic of S-AMU (which is not expected to greatly affect the representativeness of the AMU situation). This simplified the study, and allowed us to focus on the correlation between S-AMU and C-AMU. The specific changes are as follows: Page 14, Lines 209-211 Furthermore, the scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001).
3.Emphasis of the authors for the accuracy of AMU data while I believe estimating consistency of the data is much more important: I believe that the verification or validation of the consistency of the reporting from one year to another is more essential. Please recall we mean to use AMU trends as trends to be compared with AMR trends.
Thank you for the important suggestion. As advised, we have examined the longitudinal consistency of S-AMU compared to C-AMU. The specific changes are as follows: Page 10, Lines 159-162 To evaluate the longitudinal consistency of S-AMU, we compared the annual S-AMU with C-AMU between 2013 and 2017. In addition, we compared the use of branded and generic drugs in S-AMU and C-AMU to ascertain the effect of the government's initiative to promote the shift to generic drugs.
4.Indeed, comparison between C-AMU and S-AMU is interesting (if you have access to data, which is not the case here). Both systems could be used to monitor AMR trends as opposed to disqualifying C-AMU but is not capturing comprehensively the data.
Thank you for the suggestion. We have examined the utility of S-AMU through a comparison with C-AMU in the revised manuscript. The specific changes are as follows: Page 18, Lines 275-276 These findings indicate that S-AMU has utility as an alternative to C-AMU for monitoring the temporal trends in national AMU.
5.Perceived some confusion re definitions between accuracy versus precision and between accuracy and biases; e.g. unclear meaning of inadequate accuracy? What does incomplete coverage mean? and moreover incomplete coverage does not mean necessary biases. E.g. incomplete data can be precise if they are representative of a situation.
Thank you for the comment. As mentioned in our response to Comment 3, we have focused the study on comparing S-AMU and C-AMU. In the revised manuscript, we consider the effects of the rapid shift in paper-based claims to electronic claims in dental care as a characteristic of C-AMU, and the effects of insufficient data coverage as a characteristic of S-AMU.
Specific comments: 6.Line 26: abstract will benefit from a reference to the reason for AMU surveillance: AMR surveillance i.e. why these AMU estimates are important?
We have clarified the importance of developing rigorous AMU surveillance systems as follows.
Page 3, Lines 26-32 Purpose Countries require rigorous antimicrobial use (AMU) surveillance systems to monitor prescription trends and evaluate the effects of antimicrobial resistance countermeasures. AMU can be estimated using sales data (S-AMU) or insurance claims data (C-AMU), but their differences are not well characterized. We comparatively examined Japan's national S-AMU and C-AMU estimates across a 5year period, and explored the potential causes of their differences. 7.Line 34: unclear -as written -why the lack of paper-based claims is a limitation Thank you for the comment. The lack of paper-based claims potentially reduces the coverage of C-AMU. However, this may only have an effect on dental care (as reimbursements in the other types of care are predominantly managed by electronic claims). We have clarified this in the Data sources subsection of the Methods.
8.Line 36: the authors could explain why analyzing direct sales to medical facilities is important; and why medical facilities will only buy generics -how about private hospitals?
As we have set one of the study's aims to clarify the effect of a government policy that encouraged the transition from branded drugs to generic drugs, the discussion about direct sales has become more important when compared to the previous manuscript. The description has been revised in the Discussion. In Japan, the government determines the drug prices for all administered/prescribed drugs under the national health insurance system. Therefore, patients are charged the same price for a specific drug regardless of where the drug was prescribed (public or private institutes).
9.Line 37: unclear about the importance of comparison between generic and brand drugs?
We added the importance of comparing the trends between branded and generic drugs. This was to clarify the effect of a government policy that encouraged the transition from branded drugs to generic drugs.
Page 6, Lines 82-85 In addition, longitudinal consistency is a crucial aspect of a national AMU surveillance system, but antimicrobial sales may be affected by changes to healthcare policies. For example, the Japanese government has encouraged the shift from branded drugs to generic drugs from 2013 onward [10, 11].
10.Line 38: unclear significance of paper-based and electronic claims. In short, without any understanding on how it works in Japan, I'm having a hard time processing the rationality of the methods We apologize for the inadequate explanation. We have addressed the significance of paper-based and electronic claims in the additional explanation of C-AMU, as described in our response to Comment 1.
11.Line 41 -44: the authors share DID values and trends. How do we know that these values are statistically significantly different between each other? Because S-AMU and C-AMU are created from different data sources, it is difficult to interpret any statistical difference. Instead, we evaluated the correlation between S-AMU and C-AMU. Please refer to our response to Comment 2.
12.Line 46: interesting that DID of brand drugs are higher than generic drugs. Reasons are not explained, and values are not shown The DID values of each group (parenteral/oral, branded/generic, and S-AMU/C-AMU) are shown in Table 1. The higher DID for branded drugs may be due to pharmaceutical companies promoting their sale over generic drugs to healthcare providers, or a large proportion of patients and doctors who choose branded drugs based on perceptions of higher effectiveness and safety. We have added a description of the government's promotion of the transition from branded to generic drugs.
Page 6, Lines 84-85 For example, the Japanese government has encouraged the shift from branded drugs to generic drugs from 2013 onward [10, 11].
Page 18, Line 277-280 The initially high DID for branded drugs may have been due to their promotion by pharmaceutical companies and/or large proportions of patients and physicians who preferentially chose branded drugs based on perceptions of higher effectiveness and safety.

13.Line 47: what is a dispensing claim
A dispensing claim is an insurance claim for drug dispensing by pharmacies. We have clarified the language as follows.
Page 9, Lines 135-138 Under Japan's healthcare system, claims for medical care, dental care, and drug dispensing (by pharmacies) are handled separately. From 2011, electronically submitted claims for hospital-based medical care and drug dispensing claims have accounted for over 99.9% of total claims [13].
14.Line 51: unclear about the relationship between "AMU is underestimated" and "lack of direct sales"; do the authors mean the data are missing or not analyzable/accessible?
The sentence has been removed from the revised manuscript.
15.Line 51: Conclusion of abstract is weak: lots of results displayed but no indication in the conclusion to make sense of them. E.g. last sentence is a kind of catchall phrase We have revised the conclusion. It is difficult to meaningfully explain the results for branded/generic drugs and parenteral/oral drugs given the strict word limits of the abstract. Therefore, we have focused on the differences between S-AMU and C-AMU, and concluded that S-AMU may have utility in national AMU surveillance in Japan.
Page 4, Lines 51-56 Our study provides insight into the characteristics of S-AMU and C-AMU surveillance in Japan. Their differences may have been influenced by the omission of direct sales information in S-AMU and the lack of electronically submitted claims for dental care in C-AMU. However, these factors did not appear to differentially affect the temporal trends. The differences stabilized after 2015, suggesting that S-AMU is a viable surrogate indicator of Japan's national AMU. 16.Line 73: the authors argue that previous data did not account for time-series correlations. However, it appears the present study does not show any time-series correlation analysis either, does it?
We have added an analysis of the correlation of S-AMU and C-AMU according to year. Please refer to our response to Comment 2.
17.Line 84: this paragraph on data sources should be expanded to insert much more information on Japan's healthcare systems, detailed description of what is available in terms of variables and time periods of C-AMU and S-AMU. E.g. Private vs public hospitals; vet antibiotics versus antibiotics for humans; only one private company supplying Japan?
We have added more detailed explanations of S-AMU and C-AMU in our response to Comment 1. Furthermore, as advised, we have addressed veterinary AMU in the Discussion with newly cited reports.
Page 20, Lines 310-314 It is also important to adopt a "One Health" approach to controlling antimicrobial resistance, which includes monitoring AMU not only in humans, but also in animals [19]. The Japanese government's "Nippon AMR One Health Report" described the use of S-AMU to calculate the aggregate AMU in both humans and animals [20].
18.Line 105: meaning of inadequate accuracy? Is it important that the data is inaccurate if on can demonstrate that the data is precise and more importantly consistent overtime? If the authors claim inaccuracy -then what is the extent of the inaccuracy?
We have made major revisions to the manuscript, and the sentences have been deleted.
19.Line 106: just so that I'm clear: S-AMU data pertain to 99% of total sales but direct sales data to medical facilities are missing or not broken down or categorized as such in the dataset?
We apologize for the inadequate explanation. S-AMU data included 99% of wholesale data.
20.Line 117: would be interesting to mention how different these data are different from that of Europe in terms of available variables We searched for lists of available variables of S-AMU in European countries, and contacted IQVIA about these variables. Unfortunately, we could not obtain any lists that could support such a comparison.

21.
Line 120: what is the extent of the AMU among tourists and for cosmetic procedures? Does it matter if we consider the inaccuracy is consistent over time?
Recall that trends matter more for AMU and AMR surveillance purposes We agree with your comments. We have included a study aim to examine the longitudinal consistency between S-AMU and C-AMU, and addressed this lack of data from C-AMU in the limitations.
Page 20, Lines 324-327 Second, information on publicly funded and self-pay patients are not available in the C-AMU. However, this may not greatly affect the longitudinal trends in C-AMU as the number of such patients is likely to be consistent over time. Thank you for your suggestion. IQVIA is unable to provide more granular data due to confidentiality agreements with the wholesalers.
26.Line 175-176: the main result in my view is the consistency of differences over many of these factors (parental vs per os, generic/branded). Shouldn't the authors emphasize this more in the results?
Thank you for your suggestions. We have revised our study framework, and emphasize the consistency of the differences between generic vs branded drugs in both oral and parenteral drugs. Please refer to our response to Comment 2.
27.Line 181 -182: 2 sentences that say the same thing We have revised the sentence as follows: Page 15, Lines 228-230 Throughout this period, S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-AMU for generic drugs.
28.Line 185: figure 2 seems to show interesting trends over time but no analysis?
We added Pearson's correlation test to these results. Please refer to our response to Comment 2.
29.Line 198: only one seems to realize that paper-based claims are not submitted into the system.
We agree with your opinion. We have revised the discussion of paper-based claims such that it is now considered a characteristic of C-AMU. In general, pharmaceutical companies sell their products to medical facilities through wholesalers to reduce the burden of sales. However, some companies may sell their products directly to medical facilities to increase their profit margins that would otherwise be lost to the wholesalers.
32.Line 230: the paper becomes much more interesting if there is some attempt to estimate the extent of the biases. Unfortunately, we don't have this in this paper Although we agree with this point, we were unable to quantify the extent of these biases with our data. Nevertheless, the study has shifted from speculating about these possible biases, and focused on the differences between C-AMU and S-AMU.

33.Line 245: I disagree with authors; C-AMU is valuable if consistent overtime
We have changed the discussion according to your advice, and regarded C-AMU as the standard method for evaluating the national AMU. Please refer to our response to Comment 4.
34.Line 247: this information could be shared earlier in a potential section on detailed description of AMU data Thank you for the comment. However, the need for human-animal aggregate AMU is, although important, not central to our study aims. We have therefore kept this issue in the Discussion.
35.Line 260: why not doing the analysis of unadjusted DDD in this paper as this paper contains limited information and results The reason for using population-adjusted DDD is written in the limitations. Adjusted DID is a more frequently used metric of national AMU than unadjusted DDD, with major international organizations (e.g., WHO and ECDC) using this unit to evaluate the AMU of countries. If readers would like to evaluate unadjusted DDD, they can easily obtain the figures by multiplying DID with the population of Japan, as described in Reference #15 (Statistics of Japan The study did not involve any interventions in human subjects. All IQVIA data and NDB data were fully anonymized before being received by the authors. This study was approved by the institutional review board of the National Center for Global Health and Medicine (Approval Number: NCGM-G-002505-00).
information entered here is included in the Methods section of the manuscript. If the data are held or will be held in a public repository, include URLs, accession numbers or DOIs. If this information will only be available after acceptance, indicate this by ticking the box below. For example: All XXX files are available from the XXX database (accession number(s) XXX, XXX.).
• If the data are all contained within the manuscript and/or Supporting Information files, enter the following: All relevant data are within the manuscript and its Supporting Information files.
• If neither of these applies but you are able to provide details of access elsewhere, with or without limitations, please do so. For example: Data cannot be shared publicly because of [XXX]. Data are available from the XXX Institutional Data Access / Ethics Committee (contact via XXX) for researchers who meet the criteria for access to confidential data.
• Researchers should apply directly to IQVIA Japan (https://www.iqvia.com/jajp/locations/japan) for access to the study's antimicrobial drug sales data. The use of the criteria described in the Methods section will enable other researchers to identify the same data obtained for this study. The authors confirm that they did not have any special privileges to access the data that other researchers would not have. For the insurance claims data, the Ministry of Health, Labour and Welfare of Japan has placed strict legal restrictions on the release or sharing of these data. As a result, these data are not publicly available. But researchers can apply for such data as we used to the Ministry of Health, Labour and Welfare of Japan if passed qualification examination (phone number: +81-50-5546-9167).
The data underlying the results presented in the study are available from (include the name of the third party and contact information or URL). This text is appropriate if the data are owned by a third party and authors do not have permission to share the data.

• * typeset
Additional data availability information: Tick here if the URLs/accession numbers/DOIs will be available only after acceptance of the manuscript for publication so that we can ensure their inclusion before publication. Countries require rigorous antimicrobial use (AMU) surveillance systems to monitor 27 prescription trends and evaluate the effects of antimicrobial resistance countermeasures. 28 AMU can be estimated using sales data (S-AMU) or insurance claims data (C-AMU), 29 but their differences are not well characterized. We comparatively examined Japan's 30 national S-AMU and C-AMU estimates across a 5-year period, and explored the 31 potential causes of their differences. 32

Methods 33
Using wholesalers' drug sales data and national insurance claims data, we calculated the 34 annual S-AMU and C-AMU throughout Japan from 2013 to 2017. AMU was calculated 35 as defined daily doses/1,000 inhabitants/day (DID). We compared annual S-AMU and 36 C-AMU estimates, and evaluated their longitudinal consistency over the study period. 37 In addition, we examined S-AMU and C-AMU according to branded/generic and 38 parenteral/oral classifications, and estimated their correlations using Pearson's 39 correlation coefficient. parenteral drugs, S-AMU was consistently higher than C-AMU for both branded and 47 generic drugs. Among oral drugs, C-AMU was higher than S-AMU for generic drugs, 48 but not branded drugs. 49

Conclusions 50
Our study provides insight into the characteristics of S-AMU and C-AMU surveillance 51 in Japan. Their differences may have been influenced by the omission of direct sales 52 information in S-AMU and the lack of electronically submitted claims for dental care in 53 C-AMU. However, these factors did not appear to differentially affect the temporal 54 trends. The differences stabilized after 2015, suggesting that S-AMU is a viable 55 surrogate indicator of Japan's national AMU. 56 Introduction 57 Antimicrobial resistance has rapidly become a global public health and 58 economic issue, and an extensive reduction in antimicrobial use (AMU) is needed to 59 suppress the spread of resistant pathogens [1]. In order to monitor current usage patterns 60 and support the implementation of effective antimicrobial stewardship programs, 61 countries must first establish rigorous national AMU surveillance systems [2]. 62 AMU can be estimated using antimicrobial drug sales data (S-AMU) or 63 insurance claims data (C-AMU). While C-AMU incorporates patient characteristics 64 (e.g., age, sex, and recorded diagnoses), the lack of this information in S-AMU means 65 that the appropriateness of prescriptions cannot be evaluated In Japan, researchers may acquire national-level health insurance claims data 71 for C-AMU if their applications are approved by the government. However, this process 72 can take a long time as the government prioritizes the protection of patient data. As a 73 result, we have used S-AMU to first publish a preliminary report of annual AMU in 74 75 Several studies have also reported on Japan's national AMU trends using sales data [6-76 8]. However, the differences between S-AMU and C-AMU have not been well 77 characterized in Japan or elsewhere. To the best of our knowledge, only one study has 78 examined the differences between S-AMU and C-AMU [9]. That study found a 79 significant correlation between S-AMU and C-AMU for antimicrobial consumption 80 based on Anatomical Therapeutic Chemical (ATC) classifications, but did not evaluate 81 the cause of the differences. In addition, longitudinal consistency is a crucial aspect of a 82 national AMU surveillance system, but antimicrobial sales may be affected by changes 83 to healthcare policies. For example, the Japanese government has encouraged the shift 84 from branded drugs to generic drugs from 2013 onward [10,11]. 85 This study aimed to comparatively examine Japan's national S-AMU and C-86 AMU across a 5-year period, and explore the potential causes of their differences.  95 We estimated S-AMU using commercial data purchased from IQVIA Japan, a 96 private data firm that curates databases of medical drug sales and distribution 97 throughout the country. In Japan, medical drugs are generally sold by pharmaceutical 98 companies to medical facilities via wholesalers. IQVIA Japan collects sales data from 99 these wholesalers in order to construct databases. Although detailed information has not 100 been published, the company states that it collects data from almost all domestic 101 wholesalers, and that the data encompass more than 99% (monetary value) of all 102 wholesale drug sales in Japan (this coverage was confirmed through personal 103 communication with IQVIA Japan). Our dataset contained wholesalers' sales data of all 104 antimicrobials, as well as information on oral/parenteral classification and 105 branded/generic classification. However, S-AMU data do not include information on 106 the purchasing parties (i.e., names and types of medical facilities) or patients, and 107 therefore do not allow distinctions to be made between primary care and hospital care 108 dispensing and reimbursement. 109 110 C-AMU 111 We estimated C-AMU using data obtained from the National Database of Health 112

S-AMU
Insurance Claims and Specific Health Checkups of Japan (NDB), which is managed by 113 the Ministry of Health, Labour and Welfare (MHLW). The NDB contains national-level 114 health insurance claims data for healthcare encounters covered by insurance, and these 115 data can be used for research purposes following the submission and approval of an 116 application to the government [12]. Our dataset included data on the dates and types of 117 antimicrobials prescribed to patients for all insurance-covered healthcare encounters. 118 In Japan, all residents are required to enroll in health insurance, which entitles 119 them to receive insurance-covered healthcare at any medical facility throughout the 120 country. Enrollees pay monthly premiums to their insurers, and also pay a fixed 121 proportion (10-30% depending on age and income) of the medical charges at the point 122 of care. Healthcare providers send claims to the applicable insurers to be reimbursed for 123 the remaining charges. Because insurance-covered care accounts for the majority of 124 medical treatments provided in Japan, the NDB represents a near-comprehensive 125 database of all treatments performed throughout the country. 126 However, the NDB does not include claims data from patients with fully 127 publicly funded healthcare (e.g., patients with intractable diseases, atomic bomb 128 survivors, patients on welfare, patients with tuberculosis, and patients with human 129 immunodeficiency virus infections) and patients who personally pay for all of their 130 medical expenses (e.g., foreign travelers and cosmetic surgery patients). The detailed 131 numbers of these patients are not published by the government. 132 Furthermore, the NDB only includes data from claims that have been submitted 133 electronically, and the lack of paper-based claims reduces its coverage of healthcare 134 encounters. Under Japan's healthcare system, claims for medical care, dental care, and  For parenteral antimicrobials, both branded and generic drugs exhibited 207 similar trends in S-AMU and C-AMU (Fig 2A). The use of generic drugs increased 208 while the use of branded drugs decreased over the study period. Furthermore, the scatter 209 plots of branded and generic drugs demonstrated significant correlations between S-210 AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001) (Fig 2B). For both branded 211 and generic drugs, S-AMU was higher than C-AMU throughout the study period. The DID values of oral antimicrobials were more than 10 times higher than 224 those of parenteral antimicrobials in both S-AMU and C-AMU. As with the parenteral 225 antimicrobials, both branded and generic drugs in oral antimicrobials exhibited similar 226 trends in S-AMU and C-AMU (Fig 3A). Again, the use of generic drugs increased 227 while the use of branded drugs decreased over the study period. Throughout this period, 228 S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-229 AMU for generic drugs. The scatter plots of branded and generic drugs demonstrated 230 significant correlations between S-AMU (r = .996, P < .001,) and C-AMU (r = .997, P 231 < .001) (Fig 3B). and 0.78-0.95 for C-AMU. S-AMU was consistently higher than C-AMU for extended-249 spectrum penicillins. In contrast, S-AMU was lower than C-AMU from 2015 to 2017 250 for fluoroquinolones, which was similar to the trend for total AMU (Fig 1). It was interesting to find that C-AMU was higher than S-AMU from 2015 to 281 2017. S-AMU should be inherently higher than C-AMU because the former includes 282 dead stock and expired medications that are sold to medical facilities but not 283 administered to patients [9]. We posit that this disparity is influenced by the omission of 284 direct sales from pharmaceutical companies to medical facilities in S-AMU. In general, 285 pharmaceutical companies sell their products to medical facilities through wholesalers 286 to reduce the burden of sales. However, some companies may sell their products 287 directly to medical facilities to increase their profit margins that would otherwise be lost 288 to the wholesalers. Our analysis found that S-AMU was only higher than C-AMU for 289 generic oral antimicrobials. Pharmaceutical companies frequently sell generic drugs 290 directly to medical facilities [18], which may explain this finding. On the other hand, 291 the benefits of direct sales for pharmaceutical companies lie in the exemption of 292 commission charges [18], which are diminished for low-cost drugs. This corroborates 293 our observations that C-AMU was higher than S-AMU for high-cost antimicrobials 294 (fluoroquinolones), and that this pattern was reversed for low-cost antimicrobials 295 (extended-spectrum penicillins). Direct sales to consumers are illegal in Japan, but 296 direct sales to hospitals and clinics are not prohibited. Although we did not find any 297 studies that have addressed these direct sales in other countries, analysts should be 298 aware of their potential effects on national S-AMU surveillance. 299 Our analysis detected significant correlations between S-AMU and C-AMU for 300 all combinations of parenteral/oral drugs and branded/generic drugs. These findings 301 support the utility of S-AMU (despite the lack of direct sales) in monitoring national 302 AMU trends in Japan. This also supports our strategy to initially publish a preliminary 303 AMU report using S-AMU followed by a more comprehensive report using C-AMU 304 (which includes more detailed information such as stratifications by patient age and sex) 305 [5]. 306 Finally, these findings also underscore a need to characterize and understand the 307 limitations of the various data sources for AMU surveillance in order to accurately 308 interpret any findings, and that the appropriate surveillance method should be chosen 309 based on the designated purpose. It is also important to adopt a "One Health" approach 310 to controlling antimicrobial resistance, which includes monitoring AMU not only in 311 humans, but also in animals [19]. The Japanese government's "Nippon AMR One 312 Health Report" described the use of S-AMU to calculate the aggregate AMU in both 313 humans and animals [20]. As sales data are used in drug utilization analyses across 314 various epidemiological fields [21,22], the findings of this study may be useful not only 315 for researchers of infectious diseases, but also for other conditions. 316 Our study has several limitations. First, the analysis provides insight into S-317 AMU and C-AMU in Japan, but the characteristics and limitations of these data sources 318 may not be directly applicable to those of other countries. Countries require rigorous antimicrobial use (AMU) surveillance systems to monitor 27 prescription trends and evaluate the effects of antimicrobial resistance countermeasures. 28 AMU can be estimated using sales data (S-AMU) or insurance claims data (C-AMU), 29 but their differences are not well characterized. We comparatively examined Japan's 30 national S-AMU and C-AMU estimates across a 5-year period, and explored the 31 potential causes of their differences. 32

Methods 33
Using wholesalers' drug sales data and national insurance claims data, we calculated the 34 annual S-AMU and C-AMU throughout Japan from 2013 to 2017. AMU was calculated 35 as defined daily doses/1,000 inhabitants/day (DID). We compared annual S-AMU and 36 C-AMU estimates, and evaluated their longitudinal consistency over the study period. 37 In addition, we examined S-AMU and C-AMU according to branded/generic and 38 parenteral/oral classifications, and estimated their correlations using Pearson's 39 correlation coefficient. parenteral drugs, S-AMU was consistently higher than C-AMU for both branded and 47 generic drugs. Among oral drugs, C-AMU was higher than S-AMU for generic drugs, 48 but not branded drugs. 49

Conclusions 50
Our study provides insight into the characteristics of S-AMU and C-AMU surveillance 51 in Japan. Their differences may have been influenced by the omission of direct sales 52 information in S-AMU and the lack of electronically submitted claims for dental care in 53 C-AMU. However, these factors did not appear to differentially affect the temporal 54 trends. The differences stabilized after 2015, suggesting that S-AMU is a viable 55 surrogate indicator of Japan's national AMU. 56 Introduction 57 Antimicrobial resistance has rapidly become a global public health and 58 economic issue, and an extensive reduction in antimicrobial use (AMU) is needed to 59 suppress the spread of resistant pathogens [1]. In order to monitor current usage patterns 60 and support the implementation of effective antimicrobial stewardship programs, 61 countries must first establish rigorous national AMU surveillance systems [2]. 62 AMU can be estimated using antimicrobial drug sales data (S-AMU) or 63 insurance claims data (C-AMU). While C-AMU incorporates patient characteristics 64 (e.g., age, sex, and recorded diagnoses), the lack of this information in S-AMU means 65 that the appropriateness of prescriptions cannot be evaluated Several studies have also reported on Japan's national AMU trends using sales data [6-76 8]. However, the differences between S-AMU and C-AMU have not been well 77 characterized in Japan or elsewhere. To the best of our knowledge, only one study has 78 examined the differences between S-AMU and C-AMU [9]. That study found a 79 significant correlation between S-AMU and C-AMU for antimicrobial consumption 80 based on Anatomical Therapeutic Chemical (ATC) classifications, but did not evaluate 81 the cause of the differences. In addition, longitudinal consistency is a crucial aspect of a 82 national AMU surveillance system, but antimicrobial sales may be affected by changes 83 to healthcare policies. For example, the Japanese government has encouraged the shift 84 from branded drugs to generic drugs from 2013 onward [10,11]. 85 This study aimed to comparatively examine Japan's national S-AMU and C-86 AMU across a 5-year period, and explore the potential causes of their differences.  For parenteral antimicrobials, both branded and generic drugs exhibited 207 similar trends in S-AMU and C-AMU (Fig 2A). The use of generic drugs increased 208 while the use of branded drugs decreased over the study period. Furthermore, the scatter 209 plots of branded and generic drugs demonstrated significant correlations between S-210 AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001) (Fig 2B). For both branded 211 and generic drugs, S-AMU was higher than C-AMU throughout the study period. The DID values of oral antimicrobials were more than 10 times higher than 224 those of parenteral antimicrobials in both S-AMU and C-AMU. As with the parenteral 225 antimicrobials, both branded and generic drugs in oral antimicrobials exhibited similar 226 trends in S-AMU and C-AMU (Fig 3A). Again, the use of generic drugs increased 227 while the use of branded drugs decreased over the study period. Throughout this period, 228 S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-229 AMU for generic drugs. The scatter plots of branded and generic drugs demonstrated 230 significant correlations between S-AMU (r = .996, P < .001,) and C-AMU (r = .997, P 231 < .001) (Fig 3B). and 0.78-0.95 for C-AMU. S-AMU was consistently higher than C-AMU for extended-249 spectrum penicillins. In contrast, S-AMU was lower than C-AMU from 2015 to 2017 250 for fluoroquinolones, which was similar to the trend for total AMU (Fig 1). It was interesting to find that C-AMU was higher than S-AMU from 2015 to 281 2017. S-AMU should be inherently higher than C-AMU because the former includes 282 dead stock and expired medications that are sold to medical facilities but not 283 administered to patients [9]. We posit that this disparity is influenced by the omission of 284 direct sales from pharmaceutical companies to medical facilities in S-AMU. In general, 285 pharmaceutical companies sell their products to medical facilities through wholesalers 286 to reduce the burden of sales. However, some companies may sell their products 287 directly to medical facilities to increase their profit margins that would otherwise be lost 288 to the wholesalers. Our analysis found that S-AMU was only higher than C-AMU for 289 generic oral antimicrobials. Pharmaceutical companies frequently sell generic drugs 290 directly to medical facilities [18], which may explain this finding. On the other hand, 291 the benefits of direct sales for pharmaceutical companies lie in the exemption of 292 commission charges [18], which are diminished for low-cost drugs. This corroborates 293 our observations that C-AMU was higher than S-AMU for high-cost antimicrobials 294 (fluoroquinolones), and that this pattern was reversed for low-cost antimicrobials 295 (extended-spectrum penicillins). Direct sales to consumers are illegal in Japan, but 296 direct sales to hospitals and clinics are not prohibited. Although we did not find any 297 studies that have addressed these direct sales in other countries, analysts should be 298 aware of their potential effects on national S-AMU surveillance. 299 Our analysis detected significant correlations between S-AMU and C-AMU for 300 all combinations of parenteral/oral drugs and branded/generic drugs. These findings 301 support the utility of S-AMU (despite the lack of direct sales) in monitoring national 302 AMU trends in Japan. This also supports our strategy to initially publish a preliminary 303 AMU report using S-AMU followed by a more comprehensive report using C-AMU 304 (which includes more detailed information such as stratifications by patient age and sex) 305 [5]. 306 Finally, these findings also underscore a need to characterize and understand the 307 limitations of the various data sources for AMU surveillance in order to accurately 308 interpret any findings, and that the appropriate surveillance method should be chosen 309 based on the designated purpose. It is also important to adopt a "One Health" approach 310 to controlling antimicrobial resistance, which includes monitoring AMU not only in 311 humans, but also in animals [19]. The Japanese government's "Nippon AMR One 312 Health Report" described the use of S-AMU to calculate the aggregate AMU in both 313 humans and animals [20]. As sales data are used in drug utilization analyses across 314 various epidemiological fields [21,22], the findings of this study may be useful not only 315 for researchers of infectious diseases, but also for other conditions. 316 Our study has several limitations. First, the analysis provides insight into S-317 AMU and C-AMU in Japan, but the characteristics and limitations of these data sources 318 may not be directly applicable to those of other countries. Thank you for pointing this out. We have reformatted the manuscript to meet PLOS ONE's style requirements.
2. In your ethics statement in the Methods section and in the online submission form, please provide additional information about the data used in your retrospective study. Specifically, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent.
As advised, we have added the following statements to both the manuscript and the online submission system.

Page 11 Lines 177-180
The study did not involve any interventions in human subjects. All IQVIA data and NDB data were fully anonymized before being received by the authors. This study was approved by the institutional review board of the National Center for Global Health and Medicine (Approval Number: NCGM-G-002505-00).
3. Please describe how another researcher could obtain the data from IQVIA Japan used in this study.
As instructed, we have included the following in the online submission form: Researchers should apply directly to IQVIA Japan (https://www.iqvia.com/jajp/locations/japan) for access to the study's antimicrobial drug sales data. The use of the criteria described in the Methods section will enable other researchers to identify the same data obtained for this study. The authors confirm that they did not have any special privileges to access the data that other researchers would not have.
For the insurance claims data, the Ministry of Health, Labour and Welfare of Japan has placed strict legal restrictions on the release or sharing of these data. As a result, these data are not publicly available.

Response to Reviewers
We have made major revisions to the entire manuscript based on the reviewers' advice and suggestions. We are grateful for the opportunity to improve our work, and look forward to the reviewers' evaluation of the new structure and theories included in the manuscript. Our point-by-point responses to the comments are provided below:

Reviewer 1
This is a very interesting exploration of differences between two comprehensive, population-based methods to observe antibiotic consumption in Japan. The authors are to be commended for performing such analyses in order to understand the available data better and in order to be able to deal with limitations. It is an example for researchers in other nations to do likewise.
Thank you for your time and effort in reviewing our work, and for your positive evaluation of our manuscript. We have made major revisions in accordance with your comments and suggestions.
1. The differences found are all by all minimal, which could incite the authors to conclude that Japan has 2 pretty reliable sources to monitor antibiotic consumption.
The article should clearly state that this is true for consumption in primary care (and not hospital care). It this would be incorrect it should be stated, and it should also be indicated why it is not possible to make the distinction between primary care and hospital care dispensing and reimbursement.
Thank you for the advice. S-AMU and C-AMU include both primary care data and hospital data. However, S-AMU does not contain information about where an antibiotic is used, and therefore does not allow us to distinguish between primary care and hospital care dispensing/reimbursement. As advised, we have added the following information to the Data sources subsection of the Methods.

Page 7 Lines 106-109
However, S-AMU data do not include information on the purchasing parties (i.e., names and types of medical facilities) or patients, and therefore do not allow distinctions to be made between primary care and hospital care dispensing and reimbursement.
2. Some differences were observed but not statistically evaluated. Some explanation why this did not happen might be added (too small differences, complex reversing time lines).
As advised, we have added more statistical evaluations to our manuscript after consulting with a statistician.

Page 14 Lines 206-212
Effects on parenteral AMU For parenteral antimicrobials, both branded and generic drugs exhibited similar trends in S-AMU and C-AMU (Fig 2A). The use of generic drugs increased while the use of branded drugs decreased over the study period. Furthermore, the scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001) (Fig 2B). For both branded and generic drugs, S-AMU was higher than C-AMU throughout the study period.

Effects on oral AMU
The DID values of oral antimicrobials were more than 10 times higher than those of parenteral antimicrobials in both S-AMU and C-AMU. As with the parenteral antimicrobials, both branded and generic drugs in oral antimicrobials exhibited similar trends in S-AMU and C-AMU ( Fig 3A). Again, the use of generic drugs increased while the use of branded drugs decreased over the study period.
Throughout this period, S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-AMU for generic drugs. The scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .996, P < .001,) and C-AMU (r = .997, P < .001) (Fig 3B).
3. The exploration of the small differences leads to hypotheses on potential sources of bias. Most of these hypotheses are plausible. For OTC sales of antibiotics in Japan, the authors states that it is illegal. That does not mean that it does not happens, even in Japan. Some indication about the actual practices in Japan, and some references to the international literature on this should be advisable. It seems plausible that this might have an influence on the observed difference in sales and claims of cheap oral antibiotics (where the remark on the legal status of OTC antibiotics was placed).
Thank you for suggesting this important point. OTC drugs are strictly regulated in Japan, and OTC antibiotics are, for all intents and purposes, not available (excluding some topical preparations). We would be hard pressed to think of any situation where a pharmacy would risk losing their retailing license and face criminal charges for illegally selling antibiotics that are easily and cheaply available with a physician's prescription. In addition, physician consultations/prescriptions are highly affordable under Japan's universal healthcare system. We have searched the international literature on the plausibility of OTC AMU in Japan, but could not find any relevant papers. In addition to the strict legislation on the sale of antibiotics, we think the main reason for this being such a non-issue is simply that the barriers to legally obtaining antibiotics are too low for illegal sales to be worth pursuing.
Although a very small number of personal trades/sales of antimicrobials through websites may occur, such AMU would not be reflected in sales or claims data. We have added these explanations to the limitations as follows: Page 20 Lines 317-324 First, the analysis provides insight into S-AMU and C-AMU in Japan, but the characteristics and limitations of these data sources may not be directly applicable to those of other countries. For example, other countries may have a non-negligible proportion of antimicrobial sales without prescriptions [23,24], which can lead to underestimations of C-AMU. In contrast, the inclusion of over-the-counter antimicrobial sales is not an issue in Japan due to strict regulations and easy access to prescribed antimicrobials. In this way, our findings should be evaluated in the context of Japan's healthcare system and AMU patterns.
We have also cited the following additional references.

Reviewer 2
The topic of the paper is of utmost importance when it comes to monitoring AMR.
However, I'm under the impression that the work has been hastily prepared. The paper in my view lacks clarity and the methods does not appear sound, unless the authors are dealing with data that lack granularity. In general, thru my read, I'm left wanting more.
The sources of my disappointment are as follows: Thank you for your time and effort in reviewing our work, and we appreciate your advice and comments. The manuscript has undergone a major restructuring. We estimated S-AMU using commercial data purchased from IQVIA Japan, a private data firm that curates databases of medical drug sales and distribution throughout the country. In Japan, medical drugs are generally sold by pharmaceutical companies to medical facilities via wholesalers. IQVIA Japan collects sales data from these wholesalers in order to construct databases. Although detailed information has not been published, the company states that it collects data from almost all domestic wholesalers, and that the data encompass more than 99% (monetary value) of all wholesale drug sales in Japan (this coverage was confirmed through personal communication with IQVIA Japan). Our dataset contained wholesalers' sales data of all antimicrobials, as well as information on oral/parenteral classification and branded/generic classification. However, S-AMU data do not include information on the purchasing parties (i.e., names and types of medical facilities) or patients, and therefore do not allow distinctions to be made between primary care and hospital care dispensing and reimbursement.

C-AMU
Page 8, Lines 112-142 We estimated C-AMU using data obtained from the National Database of Health Insurance Claims and Specific Health Checkups of Japan (NDB), which is managed by the Ministry of Health, Labour and Welfare (MHLW). The NDB contains national-level health insurance claims data for healthcare encounters covered by insurance, and these data can be used for research purposes following the submission and approval of an application to the government [12]. Our dataset included data on the dates and types of antimicrobials prescribed to patients for all insurance-covered healthcare encounters.
In Japan, all residents are required to enroll in health insurance, which entitles them to receive insurance-covered healthcare at any medical facility throughout the country. Enrollees pay monthly premiums to their insurers, and also pay a fixed proportion (10-30% depending on age and income) of the medical charges at the point of care. Healthcare providers send claims to the applicable insurers to be reimbursed for the remaining charges. Because insurance-covered care accounts for the majority of medical treatments provided in Japan, the NDB represents a nearcomprehensive database of all treatments performed throughout the country.
However, the NDB does not include claims data from patients with fully publicly funded healthcare (e.g., patients with intractable diseases, atomic bomb survivors, patients on welfare, patients with tuberculosis, and patients with human immunodeficiency virus infections) and patients who personally pay for all of their medical expenses (e.g., foreign travelers and cosmetic surgery patients). The detailed numbers of these patients are not published by the government.
Furthermore, the NDB only includes data from claims that have been submitted electronically, and the lack of paper-based claims reduces its coverage of healthcare encounters. Under Japan's healthcare system, claims for medical care, dental care, and drug dispensing (by pharmacies) are handled separately. From 2011, electronically submitted claims for hospital-based medical care and drug dispensing claims have accounted for over 99.9% of total claims [13]. For clinic-based medical care, the proportions of electronically submitted claims increased from 91.0% in 2011 to 99.9% in 2017 [13]. For dental care, the proportions of electronically 2. Methods look at first glance proper including potential time-series correlations but with limited granular data and its understanding, the authors are left with lots of hypotheses with some sense of triangulation of information/data We agreed with your opinion, and have revised our study's conceptual framework and development. Previously, we had, as you pointed out, aimed to explore the biases of S-AMU and C-AMU from various aspects. But we have changed this construction: We now consider the effects of the rapid shift in paper-based claims to electronic claims in dental care as a characteristic of C-AMU, and the effects of insufficient data coverage as a characteristic of S-AMU (which is not expected to greatly affect the representativeness of the AMU situation). This simplified the study, and allowed us to focus on the correlation between S-AMU and C-AMU. The specific changes are as follows: Page 14, Lines 209-211 Furthermore, the scatter plots of branded and generic drugs demonstrated significant correlations between S-AMU (r = .100, P < .001) and C-AMU (r = .998, P < .001).
3. Emphasis of the authors for the accuracy of AMU data while I believe estimating consistency of the data is much more important: I believe that the verification or validation of the consistency of the reporting from one year to another is more essential. Please recall we mean to use AMU trends as trends to be compared with AMR trends.
Thank you for the important suggestion. As advised, we have examined the longitudinal consistency of S-AMU compared to C-AMU. The specific changes are as follows: Page 10, Lines 159-162 To evaluate the longitudinal consistency of S-AMU, we compared the annual S-AMU with C-AMU between 2013 and 2017. In addition, we compared the use of branded and generic drugs in S-AMU and C-AMU to ascertain the effect of the government's initiative to promote the shift to generic drugs.
4. Indeed, comparison between C-AMU and S-AMU is interesting (if you have access to data, which is not the case here). Both systems could be used to monitor AMR trends as opposed to disqualifying C-AMU but is not capturing comprehensively the data.
Thank you for the suggestion. We have examined the utility of S-AMU through a comparison with C-AMU in the revised manuscript. The specific changes are as follows: Page 18, Lines 275-276 These findings indicate that S-AMU has utility as an alternative to C-AMU for monitoring the temporal trends in national AMU.
5. Perceived some confusion re definitions between accuracy versus precision and between accuracy and biases; e.g. unclear meaning of inadequate accuracy? What does incomplete coverage mean? and moreover incomplete coverage does not mean necessary biases. E.g. incomplete data can be precise if they are representative of a situation.
Thank you for the comment. As mentioned in our response to Comment 3, we have focused the study on comparing S-AMU and C-AMU. In the revised manuscript, we consider the effects of the rapid shift in paper-based claims to electronic claims in dental care as a characteristic of C-AMU, and the effects of insufficient data coverage as a characteristic of S-AMU.
Specific comments: 6. Line 26: abstract will benefit from a reference to the reason for AMU surveillance: AMR surveillance i.e. why these AMU estimates are important?
We have clarified the importance of developing rigorous AMU surveillance systems as follows.
Page 3, Lines 26-32 Purpose Countries require rigorous antimicrobial use (AMU) surveillance systems to monitor prescription trends and evaluate the effects of antimicrobial resistance countermeasures. AMU can be estimated using sales data (S-AMU) or insurance claims data (C-AMU), but their differences are not well characterized. We comparatively examined Japan's national S-AMU and C-AMU estimates across a 5- year period, and explored the potential causes of their differences.
7. Line 34: unclear -as written -why the lack of paper-based claims is a limitation Thank you for the comment. The lack of paper-based claims potentially reduces the coverage of C-AMU. However, this may only have an effect on dental care (as reimbursements in the other types of care are predominantly managed by electronic claims). We have clarified this in the Data sources subsection of the Methods.
8. Line 36: the authors could explain why analyzing direct sales to medical facilities is important; and why medical facilities will only buy generics -how about private hospitals?
As we have set one of the study's aims to clarify the effect of a government policy that encouraged the transition from branded drugs to generic drugs, the discussion about direct sales has become more important when compared to the previous manuscript. The description has been revised in the Discussion. In Japan, the government determines the drug prices for all administered/prescribed drugs under the national health insurance system. Therefore, patients are charged the same price for a specific drug regardless of where the drug was prescribed (public or private institutes).
9. Line 37: unclear about the importance of comparison between generic and brand drugs?
We added the importance of comparing the trends between branded and generic drugs.
This was to clarify the effect of a government policy that encouraged the transition from branded drugs to generic drugs.
Page 6, Lines 82-85 In addition, longitudinal consistency is a crucial aspect of a national AMU surveillance system, but antimicrobial sales may be affected by changes to healthcare policies. For example, the Japanese government has encouraged the shift from branded drugs to generic drugs from 2013 onward [10, 11].
10. Line 38: unclear significance of paper-based and electronic claims. In short, without any understanding on how it works in Japan, I'm having a hard time processing the rationality of the methods We apologize for the inadequate explanation. We have addressed the significance of paper-based and electronic claims in the additional explanation of C-AMU, as described in our response to Comment 1.
11. Line 41 -44: the authors share DID values and trends. How do we know that these values are statistically significantly different between each other?
Because S-AMU and C-AMU are created from different data sources, it is difficult to interpret any statistical difference. Instead, we evaluated the correlation between S-AMU and C-AMU. Please refer to our response to Comment 2.
12. Line 46: interesting that DID of brand drugs are higher than generic drugs. Reasons are not explained, and values are not shown The DID values of each group (parenteral/oral, branded/generic, and S-AMU/C-AMU) are shown in Table 1. The higher DID for branded drugs may be due to pharmaceutical companies promoting their sale over generic drugs to healthcare providers, or a large proportion of patients and doctors who choose branded drugs based on perceptions of higher effectiveness and safety. We have added a description of the government's promotion of the transition from branded to generic drugs.
Page 6, Lines 84-85 For example, the Japanese government has encouraged the shift from branded drugs to generic drugs from 2013 onward [10, 11].
Page 18, Line 277-280 The initially high DID for branded drugs may have been due to their promotion by pharmaceutical companies and/or large proportions of patients and physicians who preferentially chose branded drugs based on perceptions of higher effectiveness and safety.
correlations. However, it appears the present study does not show any time-series correlation analysis either, does it?
We have added an analysis of the correlation of S-AMU and C-AMU according to year. Please refer to our response to Comment 2.
17. Line 84: this paragraph on data sources should be expanded to insert much more information on Japan's healthcare systems, detailed description of what is available in terms of variables and time periods of C-AMU and S-AMU. E.g. Private vs public hospitals; vet antibiotics versus antibiotics for humans; only one private company supplying Japan?
We have added more detailed explanations of S-AMU and C-AMU in our response to Comment 1. Furthermore, as advised, we have addressed veterinary AMU in the Discussion with newly cited reports.
Page 20, Lines 310-314 It is also important to adopt a "One Health" approach to controlling antimicrobial resistance, which includes monitoring AMU not only in humans, but also in animals [19]. The Japanese government's "Nippon AMR One Health Report" described the use of S-AMU to calculate the aggregate AMU in both humans and animals [20].
18. Line 105: meaning of inadequate accuracy? Is it important that the data is inaccurate if on can demonstrate that the data is precise and more importantly consistent overtime? If the authors claim inaccuracy -then what is the extent of the inaccuracy?
We have made major revisions to the manuscript, and the sentences have been deleted.
19. Line 106: just so that I'm clear: S-AMU data pertain to 99% of total sales but direct sales data to medical facilities are missing or not broken down or categorized as such in the dataset?
We apologize for the inadequate explanation. S-AMU data included 99% of wholesale data.
20. Line 117: would be interesting to mention how different these data are different from that of Europe in terms of available variables We searched for lists of available variables of S-AMU in European countries, and contacted IQVIA about these variables. Unfortunately, we could not obtain any lists that could support such a comparison.

21.
Line 120: what is the extent of the AMU among tourists and for cosmetic procedures?
Does it matter if we consider the inaccuracy is consistent over time? Recall that trends matter more for AMU and AMR surveillance purposes We agree with your comments. We have included a study aim to examine the longitudinal consistency between S-AMU and C-AMU, and addressed this lack of data from C-AMU in the limitations.
Page 20, Lines 324-327 Second, information on publicly funded and self-pay patients are not available in the C-AMU. However, this may not greatly affect the longitudinal trends in C-AMU as the number of such patients is likely to be consistent over time. We have added a more detailed explanation of the IQVIA data. The purchased data only contained sales of drugs from wholesalers to buyers (mainly medical facilities).
IQVIA has restrictions prohibiting the public sharing of the data. However, researchers are free to purchase similar data directly from IQVIA. Thank you for your suggestion. IQVIA is unable to provide more granular data due to confidentiality agreements with the wholesalers.
26. Line 175-176: the main result in my view is the consistency of differences over many of these factors (parental vs per os, generic/branded). Shouldn't the authors emphasize this more in the results?
Thank you for your suggestions. We have revised our study framework, and emphasize the consistency of the differences between generic vs branded drugs in both oral and parenteral drugs. Please refer to our response to Comment 2.
27. Line 181 -182: 2 sentences that say the same thing We have revised the sentence as follows: Page 15, Lines 228-230 Throughout this period, S-AMU was higher than C-AMU for branded drugs, but C-AMU was higher than S-AMU for generic drugs.
28. Line 185: figure 2 seems to show interesting trends over time but no analysis?
We added Pearson's correlation test to these results. Please refer to our response to Comment 2.
29. Line 198: only one seems to realize that paper-based claims are not submitted into the system.
We agree with your opinion. We have revised the discussion of paper-based claims such that it is now considered a characteristic of C-AMU. In general, pharmaceutical companies sell their products to medical facilities through wholesalers to reduce the burden of sales. However, some companies may sell their products directly to medical facilities to increase their profit margins that would otherwise be lost to the wholesalers.
32. Line 230: the paper becomes much more interesting if there is some attempt to estimate the extent of the biases. Unfortunately, we don't have this in this paper Although we agree with this point, we were unable to quantify the extent of these biases with our data. Nevertheless, the study has shifted from speculating about these possible biases, and focused on the differences between C-AMU and S-AMU.
33. Line 245: I disagree with authors; C-AMU is valuable if consistent overtime We have changed the discussion according to your advice, and regarded C-AMU as the standard method for evaluating the national AMU. Please refer to our response to Comment 4.
34. Line 247: this information could be shared earlier in a potential section on detailed description of AMU data Thank you for the comment. However, the need for human-animal aggregate AMU is, although important, not central to our study aims. We have therefore kept this issue in the Discussion.
35. Line 260: why not doing the analysis of unadjusted DDD in this paper as this paper contains limited information and results The reason for using population-adjusted DDD is written in the limitations. Adjusted DID is a more frequently used metric of national AMU than unadjusted DDD, with major international organizations (e.g., WHO and ECDC) using this unit to evaluate the AMU of countries. If readers would like to evaluate unadjusted DDD, they can easily obtain the figures by multiplying DID with the population of Japan, as described in Reference #15 (Statistics of Japan).
Page 21, Lines 327-332 Finally, we evaluated AMU using DDD values adjusted to the national population.
Because longitudinal evaluations that use this metric are affected by population changes, unadjusted DDD estimates may represent a more suitable metric to evaluate AMU transitions. Nevertheless, national AMU evaluations frequently use adjusted DID as an indicator, which allows for more practical and intuitive interpretations than unadjusted values.