Random forest algorithm-based prediction model for recurrence of schizophrenia patients
Objective To explore the influencing factors of recurrence in schizophrenia patients and build a prediction model of recurrence based on random forest algorithm.Methods 192 patients with schizophrenia admitted in hospital from March 2020 to June 2022 were retrospectively selected as the study subjects.According to whether the patients had relapsed schizophrenia within 1 year,they were divided into a relapse group(n=58)and a relapse-free group(n=134).The corresponding influencing factors were obtained through univariate factor analysis.The R language software was used to construct a predictive model for the relapse of schizophrenia within 1 year based on the random forest algorithm,and the receiver operating characteristic(ROC)curve was used to evaluate the predictive value of the model constructed by the random forest algorithm for the relapse.Results Univariate factor analysis showed that history of drinking,duration of illness,employment status,Social Support Rating Scale(SSRS)score,7-item Generalized Anxiety Disorder Scale(GAD-7)score,Pittsburgh Sleep Quality Index(PSQI)score,medication use,and family history of psychosis were associated with relapse of schizophrenia.Index(Pittsburgh Sleep Quality Index,PSQI)score,medication taking,and family history of psychosis were all associated with schizophrenia relapse,with statistically significant differences(P<0.05).The prediction model constructed using the random forest algorithm showed that the order of importance of the top 6 influencing factors was GAD-7 score(GiNi=16.64)>duration of illness(GiNi=15.09)>SSRS score(GiNi=14.96)>PSQI score(GiNi=14.00)>employment status(GiNi=3.94)>history of alcohol consumption(GiNi=3.72).The ROC results showed that the ROC curve of the random forest model had an Area Under Curve(AUC)of 0.840,and the optimal cut offvalue of 0.255 corresponded to a sensitivity of 82.81%and a specificity of 73.18%.Conclusion Relapse in schizophrenia patients is affected by GAD-7 score,disease duration,SSRS score,employment status,PSQI score,and drinking history,and the prediction model of relapse risk in schizophrenia patients constructed based on randomized forests had a better prediction efficacy,which can be further promoted and applied to validate the efficacy of the prediction model.
SchizophreniaRelapsePrediction modelRandom forestMedicationSocial supportSleep qualityAnxiety state
NETL
NSTL
万方数据
