Author s response to reviews Title: Application of the analytic hierarchy approach to the risk assessment of Zika virus disease transmission in Guangdong Province, China Authors: Xing Li (lixing.echo@foxmail.com) Tao Liu (gztt_2002@163.com) Lifeng Lin (13902245184@139.com) Tie Song (Tsong@cdcp.org.cn) Xiaolong Du (531335585@qq.com) Hualiang Lin (linhualiang2002@163.com) Jianpeng Xiao (jpengx@163.com) Jianfeng He (hjf@cdcp.org.cn) Liping Liu (gdcdcliu@qq.com) Guanghu Zhu (48735473@qq.com) Weilin Zeng (letitiazeng@foxmail.com) Lingchuan Guo (glcbzbs@126.com) Zheng Cao (845439543@qq.com) Wenjun Ma (mawj@gdiph.org.cn) Yonghui Zhang (zyh@cdcp.org.cn) Version: 1 Date: 25 Oct 2016 Author s response to reviews: Reviewer #1: Thanks you for allowing me to review this paper. It is well written but I have some concerns. Thank you for your insightful comments and suggestions. It s really helpful to improve the manuscript.
Major - Consider redrafting the title to indicate that the work describes a modeling approach Response: Thank you for your suggestion. The title has been updated as Application of the analytic hierarchy approach to the risk assessment of Zika virus disease transmission in Guangdong Province, China (page 1, line 1). - Introduction- identify in a few lines the role of modeling approaches in public health planning, including in cases of infectious diseases Response: Thank you for your suggestion. The analytic hierarchy approach (AHP) has been widely used in public health, including risk assessment of infectious diseases. As a semiquantitative method, AHP offers accurate assessment of risk factors and spatial/seasonal distribution for infectious diseases, which could help to provide precise prevention and control strategies. Related expression has been updated in the Introduction section (page 4, line 82-87). - 2. Materials 2.1 Study design- provide citations and several lines of text further describing the analytic hierarchy approach Response: Thank you for your comment. We followed standard AHP methodology by Saaty to conduct the analysis. First, indicators were selected by reference to citations and expert consultations. Then we invited experts to identify the index weight, subsequently generated a hierarchical structure and collected related data. After that, a series of pairwise comparisons were used to determine the relative importance of the criteria relative to ZIKV transmission, which were then combined into a numerical score using a weighting process that accounts for direct and indirect comparisons. The approach allows transparent judgments based on numerical scores. Finally, the optimal model of risk of ZIKV transmission based on this study was developed for future reference. Relevant expressions have been updated in the manuscript (materials and methods section, page 4, line 94-102). Reference: Saaty TL: A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology 1977, 15(3):234-281.
- 2.2 Risk assessment framework- provide additional citations and text providing support to use of additive model Response: Thank you for your comment. The additive model here stands for adding up the standardized indicators together. Additive model is the most commonly used model for equal weight index, and it s easy to understand. In our study, all the elements (natural risk, mosquitoborne risk, endemic risk, economic and social risk) are equal weight to each other. So we chose additive model. Similar method was describe by Tu et al. The procedure is as follows: first, indicators include the natural, mosquito-borne, endemic, economic and social elements for each county were standardized as values ranged from 0 to 1, then summed up, e.g. for county a, the risk index equals (natural risk) a+ (mosquito-borne risk) a+ (endemic risk) a+ (economic and social risk) a. The expression was updated in the revised manuscript (materials and methods section, page 5, line 106-107). References: Tu, C., et al. (2014). "Application of the analytic hierarchy process to a risk assessment of emerging infectious diseases in Shaoxing city in Southern China." Jpn J Infect Dis 67(6): 417-422. 2.5. Standardization and weight determination-provide citations and further rationalization for the approach you took Response: Thank you for your comments. Using the standardization approach, the data was rescaled into a common measurement scale that ranged between 0 and 1. It could remove the inconsistency of dimensions in different indicators, which made it more comparable. Similar standardization and weight determination approach was used by Zhu et al. The procedure is as follows: the indicator of every county/district was divided by the maximum value of this indicator in all counties/districts, which resulted in a ratio. The citation has been added in the materials and methods section, page 6, line 139. Reference: Zhu Q, Liu T, Lin H, Xiao J, Luo Y, Zeng W, Zeng S, Wei Y, Chu C, Baum S et al: The spatial distribution of health vulnerability to heat waves in Guangdong Province, China. Glob Health Action 2014, 7:25051.
- Results- how are you measuring uncertainly in your model? This is unclear. examples of this type of discussion are seen in the literature- (e.g. Cressie, Ecological Applications, Vol 19, Issue 3 pages 553) Response: Thank you for your suggestions. AHP uses a scale of numbers to make comparisons. Assigning pairwise comparisons usually involves uncertainty because of the inherent subjective nature of human judgments and the complexity involved in real-world decision-making problems. Ideally, the uncertainty of the data used in each calculated unit (county/district) should be taken into account. However, there are very few intensive tracer field experiments and literatures available in the study region for using computational method as Cressie quoted to evaluate the uncertainty of the results. Region-wide monitoring network of the risk factors should be established and improved. We have emphasized this weakness in the discussion section, page 12, line 270-275. But we tried our best to measure and reduce the uncertainty. First, the data unavailability may induce bias, which would introduce uncertainty, e.g. the entry-exit population and the airline transportation network data from endemic countries is not available. Therefore we exclude these indicators in the process of indicator selection. Second, we invited experts from multiple related disciplines (including epidemiology, infectious disease control, mosquito-borne disease, public health, geographic information, entry-exit inspection and quarantine fields) to avoid cognitive bias. Thirdly, consistency ratio (CR) was calculated to evaluate the consistency of the pairwise comparisons. All the pairwise comparisons were generally considered as consistent with CR values <0.10. Relative expressions are added in discussion section, page 11, line 263-272. - Discussion- where else has this modelling been used to describe vector-borne infectious disease- please describe further- consider discussing the work done with Dengue and the relevance to your project (e.g. Bhatt, Nature Vol 496, issue 7446) Response: Thank you for your suggestion. AHP has been used in risk assessment of emerging infectious diseases (EIDs) [1] including Dengue [2]. Tu et al. [1] assessed the likelihood of epidemic of EIDs by generating a 3-layer hierarchy with 14 evaluation factors. He et al. [2] used AHP to assess the risk of local transmission of Dengue caused by introduced cases. Zika is a kind of EIDs, and similar to Dengue, it is a mosquito-borne disease. Therefore we used AHP in line with these studies to assess the spatial and seasonal distribution of ZIKV transmission risk in Guangdong province, China.
I also would like to thank you for your reference. Bhatt et al. used a boosted regression tree (BRT) approach to establish a multivariate empirical relationship between the probability of occurrence of a dengue virus infection and the environmental conditions sampled, which is very helpful to deepen our study in the future. Relevant discussions are updated in discussion section in page 9, line 197-204. References: 1. Tu, C., et al. (2014). "Application of the analytic hierarchy process to a risk assessment of emerging infectious diseases in Shaoxing city in southern China." Jpn J Infect Dis 67(6): 417-422. 2. He, F., et al. (2015). "[Semi-quantitative risk assessment on local transmission of Dengue fever caused by introduced cases]." Zhejiang Da Xue Xue Bao Yi Xue Ban 44(6): 645-652. - Related to that- have you tried to use your approach to model the transmission of other flaviruses (eg Dengue) in China as a control approach? Response: Thank you for your comments. He et al. [1] have applied AHP to model the transmission of Dengue in Zhejiang province, China. In our model, we included Dengue transmission as a risk indicator. Although we did not directly use this method to assess the transmission risks of dengue in China, we would like to use this model in Guangdong province for the next step. Reference: [1] He, F., et al. (2015). "[Semi-quantitative risk assessment on local transmission of Dengue fever caused by introduced cases]." Zhejiang Da Xue Xue Bao Yi Xue Ban 44(6): 645-652. - Can you expand on the limitations section and discuss the literature regarding criticisms of analytic hierarchical approaches-how does your work deal with issues with these approaches discussed elsewhere? Response: Thank you for your suggestions. Although the general consensus agree that AHP is both technically valid and practically useful, it does have limitations. Firstly, there may be substantial disparity in understanding level of AHP among experts, it may be difficult to reach an
agreement through the approach. Therefore, we included a detailed introduction and guideline of AHP to all the experts. And if the judgement matrix can t pass the consistency test, the expert would be asked to recheck the logicality of his answers. Secondly, it is possible that criteria with large number of sub-criteria receive more weights, which needs to be considered. To solve this problem, we calculated the consistency ratio (CR). All the pairwise comparisons were generally considered as consistent with CR values <0.10, which represented the consistency of the pairwise comparisons. In addition, the uncertainty is a potential problem with all multi-criteria models, including AHP. To deal with this weakness, we excluded unavailable indicators in the process of indicator selection, and invited experts from multiple related disciplines (including epidemiology, infectious disease control, mosquito-borne disease, public health, geographic information, entry-exit inspection and quarantine fields) to avoid uncertainty. Despite these potential limits, the results provide validity for the general reliability of the overall assessment results. The limitations section has been expanded in discussion section, page 11-12, line 256-275. Reviewer #2: The study "Risk assessment of Zika virus disease transmission in the Guangdong Province, China" is a semi-quantitative risk assessment model of Zika virus infection in the Guangdong province in China. The investigators built a risk assessment framework using natural, mosquito born endemic, economic, and social elements and develop indicators of these dimensions. They calculated the risk index in spring summer and autumn and winter. They identified indicators based on published literature and expert consultations; they standardized the indicators, and a multidisciplinary panel were invited to determine the relative importance of all indicators. They identified 15 factors that could influence the risk of Zika virus transmission. They factor that received the largest weight was epidemic of Zika in Guangdong providence, followed by the influences of mosquito density and epidemic of Dengue virus. Then they mapped these factors in the Guangdong province during the seasons. The authors concluded that higher risk is in the Pearl River Delta and that risk is greater in summer and autumns. Comments and questions Has the risk assessment model of Zika virus infection been used for other mosquito borne diseases? And if yes, how has this model performed? Response: Thank you for your questions. AHP has been used for evaluating the risk of emerging infectious diseases (EIDs) including dengue [1-2]. Similar to Zika, Dengue belongs to the virus
family Flaviviridae and genus Flavivirus. They are both transmitted by the common mosquito vectors, namely Aedes species. Tu et al. [1] initially decomposed the EID risk into a hierarchy of more easily comprehended sub-problems, then invited experts to select suitable EIDs and optimal criteria and to subsequently to generate a hierarchical structure and collect data. The AHP methods were then used with five steps: (a) constructing the decision hierarchy for the EID risk, (b) collecting data via paired comparisons, (c) verifying the consistency of the material judgments, (d) applying the eigenvector method for weight computation, and (e) aggregating the weights to determine the EID risk ranking. He et al. [2] first reviewed risk indexes of local transmission of Dengue caused by introduced cases, then computed the weights of indexes by AHP and further used them to generate absolute risk values by multiplying indexes. Then comprehensive indexes were computed to describe relative risk by combining AHP and technique for order preference by similarity to an ideal solution (TOPSIS) methods. Both studies provided quantitative, directive results for decision makers. Relevant expressions have been updated in discussion section, page 9, line 197-200. References: 1. Tu C, Fang Y, Huang Z, Tan R: Application of the analytic hierarchy process to a risk assessment of emerging infectious diseases in Shaoxing city in southern China. Jpn J Infect Dis 2014, 67(6):417-422. 2. He, F., et al. (2015). "[Semi-quantitative risk assessment on local transmission of Dengue fever caused by introduced cases]." Zhejiang Da Xue Xue Bao Yi Xue Ban 44(6): 645-652. Is there any other similar model to assess this type of risk? Any comparison using this methodology? Response: Thank you for your comments. For ZIKV, Nah et al. [1] employed a mathematical survival analysis model to analyze the risk for 198 countries, while Tu et al. [2] conducted a qualitative and quantitative risk assessment to evaluate the risk of importation and autochthonous transmission in Mainland China. Unfortunately, we have not made the comparison yet. The reasons are as below: there is no local transmitted ZIKV case in Guangdong; and we have no access to airline transportation network data between counties of Guangdong and ZIKV epidemic countries. Since the lack of data, we
could neither apply mathematical survival analysis model (Nah et al. s method) nor conduct Tu et al. s risk assessment in our study. But we are now working hard on airline data collection; hope we could conduct the comparison soon. Relevant discussion has been updated in discussion section, page 10-11, line 243-253. References: 1. Nah K, Mizumoto K, Miyamatsu Y, Yasuda Y, Kinoshita R, Nishiura H: Estimating risks of importation and local transmission of Zika virus infection. PeerJ 2016, 4:e1904. 2. Wenxiao TU, Tao MA, Yu LI, Liu QY, Yin WW, Hong ZH, Qun LI, Daxin NI: Risk assessment on importation and autochthonous transmission of Zika virus disease in the mainland of China,2016. Chinese Science Bulletin 2016. Any plans for validating the model in regions of ongoing epidemic? Response: Thank you for your question. Now we are making great efforts to collect related data in ZIKV epidemic area. Future studies would be done to evaluate the validity of the AHP in regions of ongoing epidemic (e.g. Singapore or Thailand) for the next step.