| With the astounding development of our economy and the accelerating pace of life,the mental pressure on people is increasing.Mental disorder is becoming the first global disease in this era.However,at present,the proportion of patients with mental disorders who go to the psychological clinic is only 8.9% in China.Moreover,the recognition rate of mental disorders with somatization symptoms in China is only between 10% and 21%.These obvious physical discomfort symptoms mislead the patients,make them repeatedly see a doctor in the general hospital but not get any clear diagnosis,resulting in a tortuous condition.Therefore,it is very necessary to include mental health screening in daily physical examinations.The Symptom Checklist 90(SCL-90)is the most widely used mental symptom rating scale in China and abroad,which is mainly suitable for adults with neurosis,adaptive disorder and light mental disorders.However,there are up to 90 questions in the version of SCL-90 currently used in China,which need to take 20 minutes to fill in.In the scene of the medical examination center,the compliance of the population in medical examination center is poor,so it is difficult to complete all of them.In recent years,with the development of machine learning algorithms,researches on the application of machine learning algorithms to the optimization of psychiatric scales have begun to appear one after another.In this paper,the machine learning algorithms are used to train 2,982 subject records from physical examination center to obtain the group predictive model to shorten questionnaires groups and the questionnaires predictive model to shorten questionnaires within a group.In order to find the optimal solution of the questionnaires groups model,five machine learning algorithms are used,and one questionnaires group in the scale is cyclically selected as the predicted questionnaires group,and the allowable error of prediction is used as the selection criterion of the optimization model.In the case of distinguishing the negative and positive of the pre-questionnaires group,with sensitivity and specificity higher than 80% as the standard,the forward feature selection method is used to obtain the questionnaires model in order to achieve the dynamic reduction of the questionnaires.In this way,a model was generated to shorten questionnaires groups and questionnaires of SCL-90.When the allowable error value of the group predictive model prediction is set to 0.5,7 of the 10 questionnaires groups can be deleted with the precision of 75.9% to 81.3%,and when the sensitivity and specificity of the questionnaires predictive model are both set to 80%,each questionnaires group can be reduced from 6 to 13 questions to 3 to 7 questions.The research results illustrate that the machine learning algorithm can be used to shorten questionnaires of SCL-90 with the similar accuracy,and the proposed dynamic shorten scale is more suitable for physical examination scenarios because the less time-consuming virtue due to the dynamic reduction of the questionnaires.Moreover,a revision method of SCL-90 norm is proposed in the paper.Compared with the other four SCL-90 samples in different periods,the samples in 2019 have great changes in each factor.With the traditional 2-point factor score as the threshold of the mental health screening of the physical examination population,the detection rate is between 35% and 51%,which is significantly different from the latest epidemiological findings.With proposed norm score as the threshold,the concentrated detection rate is between 10% and 20%,which is more applicable as a measurement standard.At present,the shortened scale has been used in many medical examination centers to provide intelligent psychological examination services for the medical examination population. |
