Re: Manuscript 2448175836820289 Dear Editor, In this previous response to reviewer and editorial comments, we have highlighted the comments which were already added in the previous revision of the manuscript in yellow, and the ones that are added now are highlighted in red font. Please note that many of the comments made by the Associate Editor were reinforcing the comments made by the two reviewers. Please let us know if anything further were needed. Sincerely, Rakhi Dandona Previous response to comments Dear Editor, Thank you for suggesting revision of our manuscript "Burden of out-of-pocket expenditure for road traffic injuries in urban India”. We appreciate that reviewers considered the study findings useful and important. The constructive comments of reviewers and editor are addressed as follows in the revised manuscript: Reviewer: Steffen Flessa a. I assume that many accidents (even severe ones) will never show up – in particular the poor. The same is true for fatal accidents. How do you handle that? Page 12, paragraph 1. This study was conducted in two leading public sector hospitals and three branches of a large private hospital which are among the major providers of emergency care in Hyderabad city. The study recruited RTI cases including fatal and non-fatal cases of varying severity which could be considered representative of those that show up at a hospital, and hence provide a wide ranging perspective on the costs associated with RTI. We also conducted follow-up on these cases to document costs and outcomes post discharge from the hospital in order to get a more comprehensive picture of costs associated with RTI. We agree that all RTI cases do not report at large hospitals either because such care is not needed or because it is not accessible due to various reasons. The data presented shows the variety of costs based on fatal or non-fatal outcome of RTI, and are generally representative of the RTI cases that report to large health facilities in an Indian city. We have not attempted to generalize the findings from this study to the population level because not all cases of RTI can be accounted for. b. Whenever you write “723 patients” you should also state the period (per year?) We have mentioned in the methods section (page 5, paragraph 1) that this study was conducted between November 2005 and June 2006, and the patients were followed on average for 6 1 months from the date of discharge/death. Stating this every time 723 patients are referred to in the manuscript would make the language cumbersome and therefore suggest that this may not be necessary. c. Your definition of “catastrophic health care expenditure” is based on annual household income. I know that some do it this way, but it is not completely convincing. What is included? Subsidy from government? Income from wealth? Non-monetary income, e.g. from subsistence farming? Page 6, last paragraph. We documented monthly household income for each RTI case, and annualized it to calculate catastrophic health care expenditure. The monthly household income included household income for all members of the household from all sources including salary/wages, income from rents, royalties on leased lands or properties, interests, and income from farming, livestock etc. While calculating the out-of-pocket expenditure, we have discounted for reimbursements or subsidies, if any, which the RTI patients were able to claim from the employer, insurance company or from the other party in the crash. The treatment in public sector hospitals is subsidized by the government. As our analysis is for out-of-pocket expenditure incurred by the household, it is not necessary to take this subsidy into account. d. You use too many abbreviations. Try to avoid as far as possible, as it makes reading very cumbersome. We understand this. As the terms are long and used very frequently, we have used abbreviations. We leave this decision to the Editor and would be happy to give full terms everywhere if this were considered necessary. e. You state and prove that you published the methodology elsewhere. Before this article is accepted you have to state clearly the added value of this article in comparison to the two former articles. The cost data presented in this study was collected as part of a larger RTI study which had many objectives. The previous papers referred to in this manuscript describe in detail the study area, methodology of recruiting the RTI cases, and data collection. When there is a variety of output from a particular research study, it is generally acceptable to refer to the relevant previous publications for details on methodology or results instead of repeating them. The findings in these previously published papers describe injury patterns in crashes involving motorized two-wheeled vehicles and auto-rickshaws. These papers do not present any data on RTI costs that are presented in this current manuscript. f. If you reference web pages (reference 19) you should always state the date of access. Date of access is now added. Reviewer: Ricardo Pérez-Núñez Major compulsory revisions: 2 1. At the end of paragraph 1 of results (page 8), it says that differences in median non-medical expenditure between the richest income quartile vs the poorest is not statistically significant, whereas in footnotes of table 1 says the opposite. Authors should clarify this. Thank you for pointing this out. The footnote of table 1 is correct and we have now corrected the text (now page 9, paragraph 1, last sentence). 2. It seems to be another mistake. Second paragraph of page 8 reads: “while the miscellaneous expenditure was significantly higher in the latter [private]”, however, this is inconsistent with table 2 which specifies that public=250 and private=50. Which one is correct? There was error in this text which we have now corrected (now page 9, paragraph 2, line 5). Minor essential revisions: 1. A more detailed justification should be provided about the decision to use 25% as a threshold to define catastrophic expenditures. Authors state that this threshold is reasonable in the urban context, but they do not specify why they think that is so. A wide range of thresholds have been used in the literature to define catastrophic expenditure ranging from 10% to 40% of the household expenditure. We have presented results using a threshold of 25% to define catastrophic expenditure, which is in the middle of the range used in the literature. We also attempted this analysis using a threshold of >20% and >30% of the annual household income, and the results of multiple logistic regression were similar. We now provide this justification in the discussion on page 13, lines 1-3. In addition, we examined the intensity of catastrophic payment with another approach as well using multiple classification analysis. 2. I think authors should discuss that their COPE figures could be underestimated since expenditures were estimated under a 6 month period (in most cases) while income was taken into consideration as an annual figure. Traditionally the estimation of these indicators consider the same periods of time for both numerator and denominator. On the other hand since individuals tend to underreport income, the estimated number of households incurring in COPE could also be overestimated. We agree that the COPE figures could be underestimated due to the reason stated. It is also possible that COPE figures could be over-estimated as under-reporting of income is a common phenomenon and a limitation for estimating more accurate expenditures. We now mentioned the following in discussion on page 13, paragraph 2. “It is possible that the catastrophic expenditure due to RTI is underestimated in these findings as it is likely that some patients would have continued to incur RTI expenditure beyond the follow-up period of this study. However, it is important to note that as these expenditure figures for RTI cases include costs 6 months beyond the crash they are more complete as compared with costs incurred only during the hospital admission/visit. On the other hand, since incomes are often underreported in household surveys, our catastrophic expenditures could be an overestimate for this reason.” 3. In this sense, it should be stated as limitation of the study not to have included the estimation of income throughout expenditure. Authors clarify that this decision was taken since no expenditure information was collected as part of their study. However, I believe it would be ideal to recognize this as a potential limitation of the study in the discussion section. There is a wide literature commenting the inconvenience of estimating “true” income through self-report figures. 3 We now state this as a limitation on page 13, last paragraph. Discretionary comments: 1. I think that more context on the organization of health services in Hyderabad (ie. The total number of hospitals, % that are private, selection criteria of the three hospitals that participated in this study), would help potential readers to understand better results documented as part of this effort. In our understanding, describing health services in Hyderabad seems beyond the scope of this manuscript. We acknowledge the need for the readers to know about selection of these hospitals, and hence have added reason for this on page 5, paragraph 1, lines 4-5. 2. Authors could also discuss the limitation of annualizing monthly household income to estimate catastrophic expenditures. To what extent, income documented for any particular month could not represent the monthly average of the individual’s income per year? Considerable literature is available using annualisation of income and expenditure data using 30-days (1 month) recall period, and hence this would be considered reasonable. 3. It would also be desirable to make explicit how the goodness of fit of the final models was evaluated. We have added these data in the footnotes of relevant tables. 4. I would recommend incorporating the ISS variable into the models in its original form (0-75) instead of categorizing it into less severe and severe categories. Otherwise, a greater detailed justification should be made on why authors used 4 as a threshold to classify severity. We categorised ISS variable based on the median score (4). This is now stated on page 8, line 2. We conducted the analysis using ISS as continuous variable, and the results were similar. 5. Authors should analyzed and provide information regarding whether those RTI cases that were not finally included (n=58) are different in terms of age, sex, etc. than those that were finally analyzed (n=723). Since they represent 7.42% of the total recruited cases, this might not be absolutely necessary, but I think it is desirable in terms of evaluating potential bias. In fact it was not clear for me whether the 58 excluded from the analysis included the 40 injured that died or if the 723 included those who died after arriving alive to the hospital. There was no significant difference in terms of age (chi-square test, p=0.72), sex (p=0.85), income (p=0.70), and type of road user (0.11) between the RTI cases that were included (723) and not included (58) in the analysis. We have now re-written the details of sample available for analysis to avoid confusion on page 7, paragraph 1, lines 1-4). Of the 781 RTI cases recruited for this study, 741 (94.9%) had arrived alive and remaining 40 cases were dead on arrival at the study hospitals. Of the 741 cases that had arrived alive, follow-up interview was completed for 723 (97.6%) cases. Among the 40 cases that had arrived dead, follow-up interview was completed for 34 (85%) cases. 4 Associate Editor's comments a. Reviewer 1 questions the validity of using only the sample of individuals who report to the hospital emergency departments. This study was conducted in two leading public sector hospitals and three branches of a large private hospital which are among the major providers of emergency care in Hyderabad city. The study recruited RTI cases including fatal and non-fatal cases of varying severity which could be considered representative of those that show up at a hospital, and hence provide a wide ranging perspective on the costs associated with RTI. We also conducted follow-up on these cases to document costs and outcomes post discharge from the hospital in order to get a more comprehensive picture of costs associated with RTI. We agree that all RTI cases do not report at large hospitals either because such care is not needed or because it is not accessible due to various reasons. The data presented shows the variety of costs based on fatal or non-fatal outcome of RTI, and are generally representative of the RTI cases that report to large health facilities in an Indian city. We have not attempted to generalize the findings from this study to the population level because not all cases of RTI can be accounted for. b. Please check the potential errors in the paper. Reviewer 2 highlights a couple on p8. These errors are now corrected. c. Another potential error relates to the Footnotes of table 2 which says that p-value associated to the Kruskal test for equality of medians in vehicle expenditure between quartiles of income is p<0.001. However as can be seen in the table, all quartiles of income have median=0. Thank you for pointing this out and we have examined this in detail. It is a statistical artifact with the Kruskal test showing significance despite a median value of zero for all sub-groups. Even though the vehicle and legal expenses were “0” for more than two-thirds of the observations in the sample (hence the median was “0”), there was difference in the means. Because the Kruskal test ranks all observations per group and then sums the ranks from the groups to compare with the expected rank sum, it is possible, although not very common, for groups to have different rank sums and yet have equal or nearly equal medians. We have now removed these values from the footnotes for subgroups with median value 0 to avoid potential misinterpretation. d. Both reviewers demand some more explanations around the definition of catastrophic expenditure. Concern of Reviewer 1 is addressed in response point c for Reviewer 1, which is as follows: We documented monthly household income for each RTI case, and annualized it to calculate catastrophic health care expenditure. The monthly household income included household income for all members of the household from all sources including salary/wages, income from rents, royalties on leased lands or properties, interests, and income from farming, livestock etc. While calculating the out-of-pocket expenditure, we have discounted for reimbursements or subsidies, if any, which the RTI patients were able to claim from the employer, insurance company or from the other party in crash. The treatment in public sector hospitals is subsidized by the government. As our analysis is for 5 out-of-pocket expenditure incurred by the household, it is not necessary to take this subsidy into account. Concern of Reviewer 2 is addressed in response points 1 and 2 under minor essential revisions for Reviewer 2, which is as follows: A wide range of thresholds have been used in the literature to define catastrophic expenditure ranging from 10% to 40% of the household expenditure. We have presented results using a threshold of 25% to define catastrophic expenditure, which is in the middle of the range used in the literature. We also attempted this analysis using a threshold of >20% and >30% of the annual household income, and the results of multiple logistic regression were similar. We now provide this justification in the discussion on page 13, lines 1-3. In addition, we examined the intensity of catastrophic payment with another approach as well using multiple classification analysis. We agree that the COPE figures could be underestimated due to the reason stated. It is also possible that COPE figures could be over-estimated as under-reporting of income is a common phenomenon and a limitation for estimating more accurate expenditures. We now mentioned the following in discussion on page 13, paragraph 2. “It is possible that the catastrophic expenditure due to RTI is underestimated in these findings as it is likely that some patients would have continued to incur RTI expenditure beyond the follow-up period of this study. However, it is important to note that as these expenditure figures for RTI cases include costs 6 months beyond the crash they are more complete as compared with costs incurred only during the hospital admission/visit. On the other hand, since incomes are often underreported in household surveys, our catastrophic expenditures could be an overestimate for this reason.” e. Explain clearly what this article adds, in particular on top of two previously published articles [references: 14-15] which draw on the same data as this paper has used. The cost data presented in this study was collected as part of a larger RTI study which had many objectives. The previous papers referred to in this manuscript describe in detail the study area, methodology of recruiting the RTI cases, and data collection. When there is a variety of output from a particular research study, it is generally acceptable to refer to the relevant previous publications for details on methodology or results instead of repeating them. The findings in these previously published papers describe injury patterns in crashes involving motorized two-wheeled vehicles and auto-rickshaws. These papers do not present any data on RTI costs that are presented in this current manuscript. f. Provide more details on the choice of your modelling strategy, whether you did any model diagnostics (justify if not) and report and comment to what extent your data fit the models. Explain how you handled the missing data and the huge variability in the reported expenditure. What implications these would have for your conclusions? Page 7, last paragraph. The selection of variables of interest was based on the previously published literature within the Indian context. Of all the possible variables that were explored, we selected those variables for the multivariable model that had significant association (p-value <0.25) with the 6 outcome variables. In addition, we examined the deviance of a fitted model; smaller the deviance (-2 Log likelihood), the closer the fitted value is to the saturated model. In our models, -2 Log likelihood is 598.1 for distress financing, and 583.6 and 773.9 for medical and total out of pocket of expenditure, respectively. As these are relatively small deviance, our models fit the available data reasonably well. Missing data was minimal in our study, and these cases were excluded from relevant analyses. g. Given the potential for a wide international readership, please consider presenting your results in USD only (the current footnote on exchange rate should stay if people also want to know INR figure). We have changes all figures to USD, and kept the INR to USD conversion in Table footnotes. h. Rewrite the conclusion to reflect main findings of the study and their implications. As it stands currently, it is a too general statement. We have rewritten the conclusion to make it more specific to the study findings. i. Copyediting: After reading through your manuscript, we feel that the quality of written English needs to be improved before the manuscript can be considered further. We have made edits in the language to address this point. j. Tables - Please note that we are unable to display vertical lines or text within tables, no display merged cells: please re-layout your table without these elements. Tables should be formatted using the Table tool in your word processor. Please ensure the table title is above the table and the legend is below the table. For more information, see the instructions for authors on the journal website. Tables have now been reformatted. Thank you for considering our manuscript. If any further clarifications were required, we would be pleased to respond. Yours sincerely, Rakhi Dandona 7