Predicting Different Psychological Response Subtypes Of Symptom Network After Trauma

Background:More than 80% of people may experience at least one traumatic life event at some stage in their life.Among all kinds of traumatic events,accidental trauma（also called accidental traumatic event）is the most common one category.Although most people can overcome the stress response caused by these traumatic events within a certain period of time.However,there are still some people who suffer from trauma-related mental disorders after experiencing unexpected traumatic events.The term "cohort" in modern epidemiology refers to "tracing the incidence of a certain disease,cause of death,or other outcome in a group of people with similar characteristics.From the perspective of clinical research,if there is evidence suggested there is a relationship between exposures and outcomes.The court study is suitable.At present,there are 27 cohorts include more than 10,0000 people in China,most of which focus on chronic diseases.There is still lack of databases on trauma-related mental disorders cohort studies.It is necessary to build a prospective trauma-related mental disorder cohort to explore the short-term/long-term outcome of accidental trauma and more effective treatment strategies.Unexpected traumatic events can increase the risk various mental disorders such as post-traumatic stress disorder,depression and anxiety among survivors.Existing studies on trauma-related mental disorders include two or three mental disorders,but they tend to focus on the prevalence or severity of the disease,while ignoring its troubles at the level of symptoms.For some survivors,they may not meet the diagnostic criteria for mental disorders,but in fact they also suffer from a variety of symptoms.Thus,the criteria of “yes” or “no” may ignore the heterogeneity of symptoms.The unsupervised machine learning methods that have emerged in recent years allow us to classify individuals with similar symptoms into one category and classify patients at the symptomatic level.Taking into account the interrelationship between symptoms,only classification is certainly not enough.Psychopathological network theory pointed out that the development and maintenance of mental disorders are related to the strength of the association between symptoms.Namely,the higher the correlation strength,the greater the risk of development and maintenance of mental disorders.In contrast,those symptoms that are strongly associated with other symptoms may be more important for the development and maintenance of mental disorders.From the perspective of clinical practical,it is necessary to explore the coexistence of posttraumatic stress disorder symptoms—depression symptoms—anxiety symptoms via network analysis.Previous studies mainly focused on the prediction/risk factors of the occurrence and outcome of trauma-related mental disorders,in order to help health workers,start the prevention and crisis intervention of survivors.We can summarize the current predictions/risk factors into four dimensions:（1）Socio-demographic information:such as gender,education level and ethnicity,etc.;（2）Pre-traumatic prediction/risk factors: such as negative events,childhood trauma,etc.;（3）Prediction/risk factors in the peri-traumatic period: such as the severity of the injury,the emotion of fear,etc.;（4）The prediction/risk factors after the trauma: social support,etc.At present,most of studies use sample size and a single dimension to explore the prediction/risk factor at group level.It is very necessary to integrate the above four dimensions to carry out a large sample and multi-dimensional cohort study to explore it.Evidences suggested that there is an important bidirectional association between stress and gut microbes.Stress can influence the composition of gut microbes or change the living habits of gut microbes.Animal experiments have found that both psychological and physiological stress can cause changes in the composition of intestinal microbes.Conversely,changes in intestinal microbes will also affect stressrelated behaviors.Mice suffering from chronic stress have significantly reduced anxiety-like behaviors after intervention with probiotics.At present,most studies on stress and gut microbes are animal experiments,and most of the studies only discuss the gut microbial characteristics of patients with post-traumatic stress disorder,depression or anxiety,without considering the symptoms between the three coexistence mode.Therefore,it is necessary to carry out population-based research to explore the differences in gut microbial characteristics among different subtypes of post-traumatic stress disorder,depression and anxiety symptoms.At the same time,exploring the differences between different subtypes of gut microbes will also help us understand the pathogenesis of trauma-related mental disorders and find more appropriate treatments.Objective:The overall research purpose of this paper is to explore the predicting different psychological response subtypes of symptom network and the structural characteristics after trauma.We achieve it through the following four chapters.In order to show more intuitively,we use text and pattern diagrams to illustrate.（1）Establish a dynamic follow-up cohort of trauma population in a large tertiary hospital;（2）Exploring the subtype of posttraumatic stress disorder symptoms—depression symptoms—anxiety symptoms.Further,exploring the network structure of different subtypes;（3）According to the classification results in Chapter 2,integrate 4 dimensions of information（socio-demographic information,pre-traumatic factors,peri-traumatic factors and post-traumatic factors）to predict the outcome.Calculate the evaluation indicators of the model,select the model with the best performance in predicting the outcome event,and rank the importance of multiple predictive factors（feature importance）;（4）According to the classification results in Chapter 2,16 S r RNA high-throughput sequencing was used to explore the structural characteristics of intestinal microbes between different subtypes.Further we used the PICRUSt to predict the function of different subtypes.Methods:This thesis uses a variety of methods to achieve the overall research purpose.Chapter 1: The study is based on patients who have experienced actual or lifethreatening experiences in the trauma center of a large tertiary hospital in the first 6 months,and established a dynamic trauma cohort.Evaluate enrolled patients and collect their biological samples（blood and feces）over multiple time periods（baseline,1 month,3 months,6 months,and 12 months）.Based on the information and data integration technology,using the hybrid architecture of Browser/Server（B/S）and Client/Server（C/S）,the data of various dimensions are aggregated and stored,and combined with third-party databases（such as hospital inspection systems and provinces）.Medical Insurance Bureau)docking to establish a trauma-related mental disorder cohort database.Chapter 2: Taking the enrolled patients in the trauma-related mental disorder cohort database who completed the baseline survey and the 1-month follow-up visit to the hospital as the research object,and limited the time of the traumatic event to 24 hours before admission.All data calculations are performed in R-Studio 4.0.First,the unsupervised machine learning method based on K-Means is used to classify the coexistence pattern of posttraumatic stress disorder symptoms—depression symptoms—anxiety symptoms.After determining the subtypes,we further use the method of psychopathological network analysis to explore the node attributes and network structure attributes of each subtype of psychopathological network.Chapter 3: The research object of this chapter is the same as that of Chapter 2.All data calculations are carried out in R-Studio 4.0.Using three different methods（Logistics regression analysis,random forest and XGBoost）to integrate four dimensions of information to predict the different subtypes that calculated in Chapter 2.Then,according to the area under the curve,the method with the best performance among the predicted subtypes is selected for subsequent calculations.The subsequent calculations mainly involve the visualization of confusion matrix,calculation specificity,sensitivity,and ranking of the importance of predictive factors.Chapter 4: Stool samples are derived from the trauma-related mental disorder cohort database established in Chapter 1.The bacterial DNA is extracted from the samples,and then subjected to PCR amplification and 2% agarose gel electrophoresis analysis.Finally on the lllumina Nova Seq platform The V3-V4 region of the 16 S r RNA gene was sequenced.Use the biological information analysis method to analyze the abovementioned sequencing data,screen the different bacterial genera between different subtypes,and further use PICRUSt2 to predict the function of the different bacterial genera.Results: Chapter 1: Established a trauma-related mental disorder cohort database management system.As of January 4,2021,715 patients were enrolled at baseline and 470 patients had completed follow-up in January.Overcome some of the problems in the previous cohort database,such as eliminating data "islands",realizing data sharing among multiple systems,improving the baseline response rate（>95%）and follow-up rate（the follow-up rate in January is about 80% or more）,established a cloud platform that is convenient for clinical scientific research personnel to use.Chapter 2: As of January 4,2021,715 patients were enrolled in the baseline group.470 patients completed the follow-up at one month time point.And we include 400 patients who meet our inclusion and exclusion criteria for this study.We found that there are three subtypes of coexistence patterns of posttraumatic stress disorder symptoms—depression symptoms—anxiety symptoms.325 people（81.3%）were divided into Cluster2,61 people（15.2%）were divided into Cluster1,and 14 people（3.5%）were divided into Cluster3.The meaning of the potential subtypes corresponding to Cluster is unknown to the algorithm in advance and needs to be provided by the researcher according to the actual clinical significance.Further descriptive calculations found that Cluster2 mainly included individuals with asymptomatic or mild symptoms,Cluster1 mainly included individuals with moderate symptoms,and Cluster3 mainly included individuals with more severe symptoms.In the following psychopathological network analysis,we merge the individuals that are divided into Cluster1 and Cluster3 based on data-driven,collectively referred to as the high risk subgroup（HRS）,and Cluster2 is referred to as the low risk subgroup（low risk subgroup,LRS）.Psychopathological network analysis found that PCL₇（avoidance of relevant places or people）,PCL₂0（light sleep）,PHQ₃（difficulty falling asleep）,PHQ₉（suicidal ideation）,GAD₂（unstoppable or controlled tension）in the high risk subgroup（HRS）The central attribute（Strength）of the GAD₆（easy to be impatient）node is higher.In the low risk subgroup（LRS）,the central attributes（Strength）of the PCL₄（feeling pain when mentioning traumatic events）and PCL₁（involuntary recall of traumatic events）nodes are higher.PCL₁1（increased negative feelings）,PHQ₂（low mood）,GAD₁（nervous and irritable）and GAD₇（feeling something terrible happened）The central attribute（Strength）of the nodes is in the high risk subgroup（HRS）and the low risk subgroup（LRS）Both are higher.In the high risk subgroup（HRS）,PCL-20（light sleep）,PHQ-3（difficulty falling asleep）and PHQ-4（fatigue）formed a closed loop.Through the network comparison test,the global connectivity of the psychopathological network in the high risk subgroup（HRS）is 6.64,and the global connectivity of the psychopathological network in the low risk subgroup（LRS）is 5.45,the difference in clustering coefficient is statistically significant（p<0.05）.The node stability coefficient（CS-coefficient）of the low risk subgroup（LRS）is 0.29,and the node stability coefficient（CS-coefficient）of the high risk subgroup（HRS）is 0.26.The node stability of the two groups is higher than 0.25.In an acceptable range.Chapter 3: In order to increase the clinical usability,in Chapter 2,the data-driven individuals divided into Cluster1 and Cluster3 are merged together,collectively referred to as the high risk subgroup（HRS）.This part found that the logistic regression analysis,random forest and XGBoost predicted that the area under the curve（AUC）of the high risk subgroup（HRS）and the low risk subgroup（LRS）were greater than 60%.Among them,XGBoost has the best predictive performance（AUC=77.6%）.Therefore,we further calculating the secondary index of AUC shows that the specificity of the XGBoost model is 85.59%,the sensitivity is 77.78%,and the accuracy is 85.0%.According to the requirements in the previous literature,we choose the top 5 factors in the prediction of feature importance（feature importance）,so as to ensure the practicality of the model and the simplicity of the model.Use the XGBoost model to predict the high risk subgroup（HRS）and the low risk subgroup（LRS）.The top five contributing variables are post-traumatic daytime dysfunction,pre-traumatic sleep disturbance,peri-traumatic diastolic blood pressure,and peri-traumatic period blood glucose and gender.Chapter 4: According to the high risk subgroup（HRS）and low risk subgroup（LRS）classified in Chapter 2,we randomly select a certain percentage of patients’ stool samples in the two groups.Our results found that the alpha diversity and beta diversity between the two groups were not statistically different.There are 6 different bacterial genera between the two groups,and the proportion of Bacteroides fragilis and Megasphaera in the low risk subgroup（LRS）increased.The proportion of Pyramidobacter_piscolens,Kluyvera,Klebsiella and Bacteroides in the high risk subgroup（HRS）increased.Further using PICRUSt2 to predict the KEGG metabolic pathways of the different bacterial genus between the two groups,it was found that the LRS group showed increased activity in six pathways,namely,the metabolism and reduction of glycine,serine and threonine metabolism.Meiosis-yeast,phenylpropanoid biosynthesis,biosynthesis and biodegradation of secondary metabolites,cyanoamino acid metabolism,cofactor and vitamin metabolism（metabolism）of cofactors and vitamins).The HRS group showed increased activity in two pathways,namely alanine,aspartate and glutamate metabolism（alanine,aspartate and glutamate metabolism）and flagellar assembly.Conclusion: 1)The establishment of this cohort database is helpful for researchers to explore the occurrence,development and maintenance of trauma-related mental disorders from different perspectives;2)Survivors of accidental trauma have three patterns of posttraumatic stress disorder symptoms—depression symptoms—anxiety symptoms at one month time point.The mild symptom group（low risk subgroup,LRS）accounted for the largest proportion,followed by the moderate symptom group,and the severe symptom group the least.Individuals in the moderate symptom group and severe symptom group are clinically concerned,so they are classified as high risk subgroup（HRS）;3)The psychopathological network of low risk subgroup（LRS）and high risk subgroup is different.The former has a higher clustering coefficient.The central symptoms of the HRS group and the LRS are different.Namely,the importance of symptoms in different subtypes is different;4)XGBoost has the best performance in predicting the high risk subgroup（HRS）and the low risk subgroup（LRS）.The top five contributing variables are post-traumatic daytime dysfunction,pre-traumatic sleep disturbance,peri-traumatic diastolic blood pressure and blood glucose,and gender;5)There are 6 different microbiota between different subgroups,resulting in the ecological imbalance of "good bacteria" and "harmful bacteria" in the high-risk group（HRS）.PICRUSt2 predicted that the KEGG metabolic pathway between two subgroups found that the high risk subgroup（HRS）had increased activity in the pathway related to tryptophan synthesis.