A Case-crossover Study On The Relationship Between Climatic Factors And Upper Gastrointestinal Bleeding In Fuzhou And Establishment Of Back Propagation Neural Network Model

Objective: To analyze the distribution of onset of upper gastrointestinal bleeding(UGIB)and the relationship with climatic factors in Fuzhou,and to establish a back propagation(BP)neural network prediction model based on local data.Methods: UGIB cases diagnosed by endoscopy from January 1st,2015 to December31 st,2017 from the First Affiliated Hospital of Fujian Medical University and Fuzhou General Hospital were collected.The meteorological data during the same period was also acquired.The onset of UGIB among season,month and solar terms were compared.A case-crossover design with solar term stratification was employed to investigate the influence of climatic factors on the acute onset of UGIB.Lag effect was investigated from the same day(lag 0)to three days(lag 3)after exposure.BP neural network toolbox of MATLAB software was used for prediction model establishment.The meteorological data and hospitalization data from 2015 to 2016 were set as training group,while these in 2017 were set as simulation group.Results: A total of 1624 patients were enrolled,of which 78.63% were male.There was a statistically significant difference in the occurrence of UGIB among different seasons(χ~2=18.054,P<0.05),months(χ~2=29.576,P<0.05),and solar terms(χ~2=164.487,P<0.05).UGIB occurred more frequently in winter than in summer,while the peak and valley were found in Great Cold and Beginning of Autumn,respectively.Spearman analysis showed that UGIB was negatively correlated with mean average temperature(r=-0.106,P<0.05),daily maximum(r=-0.104,P<0.05)and minimum temperature(r=-0.109,P<0.05),respectively;while it was positively correlated with average pressure(r=0.127,P<0.05),maximum(r=0.126,P<0.05)and minimum pressure(r=0.125,P<0.05),respectively.The case cross design of solar terms stratification showed that the onset of UGIB was mainly related to temperature and pressure,and the significant results were mostly distributed from Frost’s Descent to Greater Snow.Lag effect was found on the meteorological factors related UGIB onset,while the most of lag time were at lag 0 or lag 1.The lag time of the same meteorological factors in different solar terms were found different,and so did that in various meteorological factors in the same solar term.The training results of BP neural network model showed MSE=3.897,R=0.983.The prediction model showed MSE=63.007,R=0.404,and the average error rate was 7.36%(t=0.170,P=0.866).Conclusion: The onset of UGIB varied in different seasons,months and solar terms,while it was negatively correlated with the mean and extreme values of temperature,and positively correlated with the mean and extreme values of pressure,respectively.Lag effect existed on the influence of meteorological factors on the onset of UGIB,and the lag time varied among the meteorological factors and solar terms.