The Risk Factors Analysis And Occurrence Forecasting Model Study For Retinopathy Of Prematurity

BackgroundRetinopathy of prematurity(ROP)is a proliferative retinal vascular disease affecting the retina of premature and low weight infants,and is one of the leading causes of childhood visual disability or blindness.The rate of blindness that caused by ROP can be reduced effectively if prompt screening and treatment were conducted.The screening guidelines at present are mainly based on gestational age and birth weight,too many babies were induced to screen,and the screening efficiency is low.Fundus examination is relatively invasive.Overwhelming examnations not only cost time,increase workload and financial burden,but also can hurt premature infants.Facing an increasing survival rates for premature infants and the shortage of specialized ophthalmologists,how to estimate the occurrence of ROP,to reduce screening workload on the premise of no misdiagnosis,are problems that need to be solved.To explore a non-invasive ROP occurrence forecasting method that integrates multiple risk factors will help to solve these problemsPurpose1.To analyze the potential risk factors of ROP,to find the best combination of predictive factors to predict the occurrence of ROP.2.To set up a non-invasive,low-cost ROP occurrence forecasting model,in order to recognize high risk infants earlier,make the screening more effective and help the prevention of ROP.3.To forecast disease by using artificial neural network,is an exploration of combining artificial intelligence with ophthalmology.MethodsA retrospective cohort of 591 preterm infants who were admitted at neonatal intensive care unit and accepted ROP screening in Zhujiang Hospital of Southern Medical University between September 2014 and September 2018 was studied.Infants who were born ?32 weeks of gestational age or weighing ?2000g were included.And the exclusion criteria were infants with severe congenital or chromosomal anomalies,infants who died before eye examination,infants with other severe eye diseases,and infants with incomplete data.The state of retinopathy was found in ROP screening records.Risk factors were collected by searching electronic medical records of inpatientsIn the first part,IBM SPSS 20 was used.Risk factors for ROP were assessed using univariate and multivariate analyses.After considering our results and some references,we chose the risk factors that can be used as ROP predictors.In the second part,MATLAB was used to establish BP neural network in order to forecast the occurrence of ROP.Then,BP neural network was improved by genetic algorithm,and its forecasting ability was enhanced.Results1.After considering the results of univariate and multivariate analyses and clinical knowledge,we ultimately chose 12 factors as predictors to predict the occurrence of ROP,including gestational age,birth weight,Caesarean delivery,5min Apgar score,bronchopulmonary dysplasia,respiratory distress syndrome,intracranial hemorrhage,neonatal hypoxic-ischemic encephalopathy,mechanical ventilation,blood transfusion,neonatal necrotizing enterocolitis and patent ductus arteriosus.2.For the logistic regression model,the sensitivity is 53.1%,specificity is 87.9%,accuracy is 76.6%,AUC is 0.855.For the BP neural network model,the sensitivity is 84.2%,specificity is 88.7%,accuracy is 86.8%,AUC is 0.857.For the GA-BP model,the sensitivity is 92.1%,specificity is 88.7%,accuracy is 90.1%,AUC is 0.908.ConclusionsBP neural network can perform better than traditional statistic method in dealing with non-linear data and its forecasting ability of ROP occurrence is better than logistic regression in this study.Moreover,the forecasting ability of BP neural network can be improved by genetic algorithm.Innovation1.Analyzing the risk factors of ROP carefully from many aspects such as pregnancy and delivery,neonatal common diseases,interventions and laboratory tests results after consulting most recent references.2.Using artificial neural network to predict the occurrence of ROP,not only provided a new forecasting model,but also was an exploration of combining artificial intelligence with ophthalmology.This study provided a new idea of integrating traditional clinical practice with big data and artificial intelligence.