key: cord-329459-k55ldc3m authors: Zhang, Xuan; Ma, Shi-lei; Liu, Zhong-di; He, Juan title: Correlation Analysis of Rubella Incidence and Meteorological Variables Based on Chinese Medicine Theory of Yunqi date: 2018-11-22 journal: Chin J Integr Med DOI: 10.1007/s11655-018-3016-0 sha: doc_id: 329459 cord_uid: k55ldc3m OBJECTIVE: To analyze the correlations between the incidence of rubella and meteorological factors over the same period and previous periods including 1, 2, 3 and 4 year ago (defined according to Chinese medicine Yunqi theory of "pestilence occurring after 3 years") and establish the rubella-meteorological forecast models for Beijing area, China. METHODS: Data regarding the incidence of rubella between 1990 and 2004 from Beijing Center for Disease Control and Prevention, and the meteorological variables including daily average temperatures, daily average wind speeds, average precipitations, average relative humidity, average vapor pressures and average low cloud covers between 1986 and 2004 were collected from the Beijing Meteorological Observatory. Descriptive statistics and back-propagation artificial neural network for forecast model’s establishment were adopted for data analysis. RESULTS: The average temperature and relative humidity have a great contribution (100%) to the rubella morbidity. But the combination of other meteorological factors contributed to improve the accuracy of rubella-meteorological forecast models. The forecast accuracy could be improved by 76% through utilizing a combination of meteorological variables spanning from 3 years ago to the present rather than utilizing data from a single year or dating back to more earlier time than 3 years. CONCLUSIONS: There is a close relationship between the incidence of rubella and meteorological variables in current year and previous 3 years. This finding suggests that rubella prediction would benefit from consideration to previous climate changes. doctrine is guided by the overall concept that human and nature correspond and have provided new ideas on infectious diseases research. By the Yunqi theory "pestilence occurring after 3 years", the prevalence of infectious disease is not only related to weather factors within the current year but also associated with meteorological factors that occurred 3 years ago. Some scholars found that the prevalence of severe acute respiratory syndrome (SARS) in 2003 in northern China was associated with the climate changes in 2000, which was characteristic of high temperature, less precipitation and a large area of arid climate. (7) The practical value of "pestilence occurring after 3 years" received increasing more attention in recent years. (8, 9) Since the meteorological variables may be the early indicators of outbreaks and epidemics, our study focuses on the association of rubella and meteorological variables of current year and particularly previous years. Based on the Yunqi theory, a forecasting model was developed using the previous data on rubella and meteorological variables. The research involves no direct work with human subjects. The rubella case data were extracted from Beijing Center for Disease Control and Prevention. These reports were governmental documents that summarized the counts of patients who were diagnosed at health care facilities with a variety of diseases. All individual-level data were anonymous. Beijng, the capital city of China, 40°N and 116°E, was selected as the study area. Many previous studies on Yunqi theory have been conducted using climate data of Beijing area. (10, 11) According to the Yunqi theory, the climatic characteristics are distinct during a year, and a CM concept called "six qi" mode describe this seasonal feature using the representative climate factor. (12) In this study, the meteorological are collected from the Beijing Meteorological Observatory from 1986 to 2004, including daily average temperature, daily average wind speed, daily precipitation, daily average relative humidity, daily average vapor pressure and daily average low cloud cover. The rubella data was collected from Beijing Center for Disease Control and Prevention from 1990 to 2004. T h e s t a t i s t i c a l d a t a f o r r u b e l l a a n d t h e meteorological factors are calculated using the concept of 'six qi'. (13) (14) (15) (16) The incidence of rubella in Beijing for each phase of the 'six qi' was calculated to establish a database from 1990 to 2003. The data in 2004 could not be calculated due to a lack of data about diseases in 2005. Similarly, the average values of 6 meteorological variables in Beijing for each phase of the 'six qi' from 1986 to 2003 were also calculated. Firstly, a descriptive statistic for meteorological factors and rubella morbidity were made to show the general regularities of weather and rubella prevalence during the past years. Secondly, the correlation between meteorological factors and rubella morbidity were analyzed by back-propagation artificial neural networks, a non-linear dynamic systems that belong to the class of feed-forward neural networks. (17) (18) (19) (20) In this study, the back-propagation artificial neural networks methods were used to assess the importance of various meteorological factors of previous years and the current year on the outbreak of rubella. The meteorological factors with impact degrees (i.e. the standardized significance by model calculations) above 50% were selected as important variables related to the incidence of rubella. Thirdly, the rubellameteorological forecast models were established and assessed. Specifi cally, a total of 8 rubella-meteorological forecast models for the prediction of the prevalence of rubella, as presented in Table 1 , were established by back-propagation artificial neural networks, and the forecasting accuracy was compared. The 8 models were divided into two groups, including Group 1 (e.g. models 1-4) analyzed rubella morbidity and meteorological factors of a single year; and Group 2 (e.g. models 5-8) analyzed rubella morbidity and combined meteorological factors of previous years. All of the samples were divided into training samples and testing samples to prevent over fitting. An ideal network model could be achieved when the test samples are less than the training samples. Consequently, the data for training and forecasting were randomly selected at the default ratio of 7:3 that was the most commonly used in the statistical analysis. The criteria of successful model assessment are according to the relative error values of training and testing, which should be less than 1, and the model will be considered to be successfully created. (21) indicated in Figure 1 , the rubella morbidity showed an obvious seasonality with the highest prevalence in the 2nd qi, accounting for 57% of the average annual incidence of rubella. Comparatively, rubella was in a state of low prevalence during the 4th, 5th, 6th and 1st qi, which accounted for 1.7%, 0.8%, 3.8% and 11.0% of the average annual incidence of rubella, respectively. A summary of the meteorological data counted by "six qi" phase from 1987 to 2003 is presented in Table 2 . The peaks of average temperature, average precipitation, average vapor pressure and average low cloud cover appeared in the 3rd qi (i.e. June and July). In comparison, the low points of these variables appeared in the 6th qi (i.e., December and January of the next year) and the average low cloud cover has a minimum in the 1st qi (i.e., February and March). Additionally, the peak of average relative humidity appeared in the 4th qi (i.e., August and September) and low point appeared in the 1st qi. In contrast, the peak of average wind speed appeared in the 2nd qi (i.e., April and May) and low point occurred in the 4th qi. In Beijing area, 54,186 cases of rubella were reported from January 1990 to January 2004. As Table 3 , all 4 prediction models (Models 1-4) were successful established, while the prediction accuracy were lower than 30%. In addition, several meteorological factors from different year periods had significant influences on the incidence of rubella, including average temperature, average relative humidity and average precipitation from the current year, average relative humidity, average temperature and average low cloud cover from 1 year ago; average temperature, average relative humidity, average low cloud cover and average wind speed from 2 years ago; and average precipitation and average low cloud cover from 3 years ago. Of them, the largest contribution (100%) factors included average temperature from the current year and 2 years ago, average relative humidity from 1 year ago, and average precipitation from 3 years ago (Table 4 ). As presented in Table 3 , Models 5 to 8 (Group 2) were successful established and the prediction accuracy of these models were higher than those in Models 1 to 4 (Group 1). The highest forecast accuracy appeared in Model 7 (76%), which used the meteorological factors from current year to 3 years ago. However, when data from 4 years ago were added to create Model 8, the prediction accuracy was reduced. Thus, there was no need to consider the meteorological factors of earlier periods than 3 years ago. Additionally, there were many meteorological factors from different year combinations, which presented infl uences above 50% on the incidence of rubella. Among them, the largest contribution (100%) factors were average temperature and average relative humidity from the current year, and average wind speed from 2 years ago (Table 4 ). Based on the CM Yunqi theory, this retrospective study analyzed the correlations between the rubella morbidity and meteorological factors in Beijing area. By using back propagation artificial neural network, several rubella-meteorological forecasting models were established, which were characterized by the time spans ranging from the current year to 3 years ago, and the vital meteorological factors on the incidence of rubella were selected. The study focused on the impact of meteorological variables over the same period and previous periods on the incidence of rubella. The doctrine 'pestilence occurring after 3 years' was found in the Huangdi's Internal Classic which is the source of CM theory. It contributed noteworthy observations about the influence of abrupt climatic changes or unusual weather conditions from 1 to 3 years ago on the present incidence of various infectious diseases. Consequently, we found that the prediction accuracies of rubella-meteorological models in Group One, which used the data of rubella morbidity and meteorological factors of single year, were undesirable. However, the prediction accuracies of rubellameteorological models in Group 2, which used the data of rubella morbidity and combined meteorological factors of previous years, were obviously improved compared with Group 1. With the addition of more meteorological data coming from previous years, the prediction accuracy showed a rising trend, reaching the top accuracy (76%) when the meteorological data within previous 4 years were combined (current year + 1 year ago + 2 years ago + 3 years ago). Then a downward trend appear if more data of previous meteorological data (current year + 1 year ago + 2 years ago + 3 years ago + 4 years ago) were added. Thus, there is no need to use the data of earlier periods than 3 years, which prove the doctrine of Yunqi theory "pestilence occurring after 3 years". We also found that average temperature and average relative humidity, especially in the current year, were selected as vital meteorological factors because they made the largest contributions to the rubella-meteorological forecasting models, as indicated in the back propagation artificial neural network analysis. This suggests that the conditions of temperature and humidity in Beijing area are significantly related to the outbreak of rubella. However, if considering other meteorological factors in 1-3 years ago, the prediction results of rubellameteorological forecasting models would be greatly improved. Thus, the rubella prevention and control will benefi t from considering the local previous climate changes, which consistent with the findings of other studies in infectious diseases. (22) (23) (24) The innovations of this study are summarized as follows. Firstly, based on the CM theory of Yunqi, this study divided one year into 6 equal periods (4 solar terms per period) that are called the 'six qi' or 'six phases' to describe the seasonal variations in the rubella incidence and meteorological variables. Secondly, this study focused on the effects of previous changes of weather that occurred in the past 3 years on the present incidence of rubella, which could improve the timeliness of forecasts based on established early warning models. Thirdly, back propagation artificial neural network was introduced to select the most significant weather indicator and establish the rubella-meteorological predictive models. Fourthly, this study selected a good model by analyzing and comparing each of the models. This study also has some limitations. Although meteorological variables were used to identify possible causal factors for the incidence of rubella in Beijing area, the outbreak of rubella is a complex course affected by multiple factors. Some other important issues should be considered, such as biological, sociological, and economical factors. T a k e n t o g e t h e r , t h i s s t u d y p r o v i d e s a quantitative evidence that the incidence of rubella in Beijing area was associated with meteorological variables, such as temperature and relative humidity, in the corresponding and previous years. Moreover, the rubella-meteorological forecast model from the perspective of 3 years ago to the present was successfully established and presented a relatively high accuracy rate, which suggested the use of a combination of meteorological factors spanning from 3 years ago is superior to the use of data from a single. And there is no need to use the data of earlier periods. This information could be helpful in the early prediction of rubella outbreaks and build public health prevention and intervention strategies. Therefore, some guiding ideology on Yunqi theory regarding "pestilence occurring after 3 years" in CM may be beneficial to provide new ideas for the research of using climate to predict outbreaks of infectious diseases, such as rubella. Also, similar studies in other geographical areas and over longer time periods are needed to better understand the impact of weather variables on rubella. The authors declare no competing interests. Zhang X analyzed the data and wrote the manuscript. Ma SL revised the manuscript. Liu ZD collected and checked the data. He J conceived and designed the experiments. All authors have read and approved the contents of the fi nal version. Rubella and congenital rubella syndrome: global update Epidemiological analysis of rubella in Beijing Seasonal pattern of infectious disease: similarity and differences Unusual climatic conditions and infectious diseases: observations made by Hippocrates Vernacular solution of 'Plain Questions of the Huangdi's Canon of Medicine'. Beijing: People's Medical Publishing House Seasonality of infectious diseases Discussion of severe acute respiratory syndromes (SARS) from perspective of Yunqi theory Recognition of generations of doctors on reasons of epidemic disease Based on the theory of 'transformation of plague in three years' to explore the correlation between incidence of dysentery and climate changes s mistake in calculation model for annotating Yunqi (fi ve circuits and six qi) based on Beijing meteorological data for 60 years Analysis of climatic characteristic of dominant yun in Beijing for 60 years Analysis of climate changes in Beijing based on the six qi theory of Chinese medicine Analysis of climate characteristics of 24 solar terms in Beijing Correlation analysis for the attack of respiratory diseases and meteorological factors Correlation between attack of bacillary dysentery and six qi or meteorological factors in Beijing Coincidence between Yun-qi of the heavenly stems and earthly branches and actual climatic changes in Beijing for 60 years Exploration of the use of mathematical methods in the syndrome analysis research Beijing: People's Correlation analysis for the attack of bacillary dysentery and meteorological factors based on the Chinese medicine theory of Yunqi and the medical-meteorological forecast model Application of neural network in forecasting tuberculosis caused by meteorological factors Case pristine in data analysis and mining of IBM SPSS Relationship between meteorological factors and incidence of dysentery in Beijing area based on theory of 'pestilence occurring after three years Correlation between typhoid/paratyphoid and meteorological factors and establishment of forecast model based on theory of fi ve circuits and six qi Correlation between meteorological factors and mumps incidence in Beijing from 1990 to We would like to thank the Beijing Meteorological Observatory for collecting the meteorological data and the Beijing Center for Disease Control and Prevention for the data related to the incidence of rubella.