A Study Of High-risk Autism Spectrum Disorder Based On Still-face Paradigm

?Objective?To compare and analyze the characteristics of social behaviors in toddlers with high-risk autism spectrum disorder based on still-face paradigm(SFP),and then to explore the correlation between the characteristics of social behaviors and the severity of ASD symptoms.The machine learning of random forest was used to build an early diagnosis predictive model based on different features of social behaviors during the mother-infant interaction or/and the still face.?Methods?High-risk ASD toddlers were consecutive sampling from outpatient clinic of Child Mental Health Research Center of the Affiliated Brain Hospital of Nanjing Medicine University,and typical development participants were recruited from community since December 2017 to December 2018.Study 1: The Study of HR-ASD Early Social Behavior Based on Still-face ParadigmForty-five high-risk ASD and forty-three typical developmental toddlers,aged from 8 months to 23 months,were recruited in the study.The SFP videotapes were coded to measure the frequencies and duration of social behaviors,including non-social smiling,protest,eye contact,social smiling and social positive engagement.Gesell Development Scale was used to assess the developmental level of HR-ASD and TD.Communication and Symbolic Behavior Scales Developmental Profile(CSBS-DP),Childhood Autism Rating Scale(CARS)and Autism Behavior Checklist(ABC)were used to evaluate severity of clinical symptoms of HR-ASD.Independent-sample t test was used to compare the differences of social behaviors between HR-ASD and TD.And Pearson correlation analysis was used to analyze the correlation between the social behaviors of HR-ASD and age,the developmental level and the severity of ASD symptoms.Study 2: The Establishment of Early Diagnosis Predictive Model Based on Random ForestHR-ASD were divided into N-SD who did not meet clinical criteria for an ASD diagnosis and ASD who had ASD symptomatology after following up to 2 years old and re-diagnosis.Principal-components analysis was used to derive main components and Logistic regression analysis was used to select effective main components.Finally,Spyder was used to run Python programing language and considering the significant statistics as behavioral features after preliminary analysis,and then random forest was used to build an early diagnosis predictive model.?Results?Study 1: The Study of HR-ASD Early Social Behavior Based on Still-face Paradigm(1)During the period of mother-infant interaction,HR-ASD had significantly shorter duration and lower frequencies of eye contact,social smiling and social positive engagement compared with TD(t=-4.93,-6.17,-3.54,-2.90,-9.56,-8.34;all P<0.05).During the period of still face,HR-ASD had significantly shorter duration and lower frequencies of eye contact and social positive engagement,and significantly shorter duration of social smiling compared with TD(t=-4.94,-5.34,-4.49,-6.16;all P<0.05).(2)According to Pearson correlation analysis,the duration of eye contact had significantly positive influence on symbolic score and total score in HR-ASD group during the period of still face(r=0.32,0.30;all P<0.05).The duration of social positive engagement had significantly positive influence on total score(r= r=0.30;P<0.05).The frequencies of eye contact had significantly positive influence on social score,speech score and total score(r=0.35,0.33,0.36;all P<0.05).Study 2: The Establishment of Early Diagnosis Predictive Model Based on Random Forest(1)Forty-five HR-ASD were divided into 5 N-ASD and 40 ASD after following up and re-diagnosis.(2)According to principal-components analysis and logistic regression analysis,during the still face,main component 1 could distinguish between ASD and TD(OR=9.73,95%CI: 3.79~25.00,P<0.05).Main component 1 loaded highly on the frequencies and duration of eye contact and social positive engagement.(3)Considering the duration and frequencies of eye contact,social smiling and social positive engagement during the mother-infant interaction and that of eye contact and social positive engagement during the still face as behavioral features,the accuracy of early diagnosis predictive model was 80.72% during the mother-infant interaction and 83.13% during the still face based on random forest.?Conclusion?(1)SFP could effectively distinguish between ASD and TD toddlers less than 2 years age.Especially,eye contact,social smiling and social positive engagement had a better discrimination and had correlation with the severity of ASD core symptoms.(2)The follow-up study showed that ASD could be diagnosed at 2 years old according to the social behaviors performed in SFP before 2 years old.And based on social behavior performed during the still face,early diagnosis predictive model had a higher accuracy,which demonstrated that the social behaviors performed in SFP before 2 years old had effectively predictive power in ASD diagnosis.