Objective: To construct a diagnostic model of schizophrenia (SCZ) based on biomarkers such as serum neurotrophic factor. Methods: Patients of schizophrenia (SCZ group) and healthy controls (HC group) who were admitted to the First Affiliated Hospital of Zhengzhou University from January 2017 to December 2019 were prospectively selected. In the SCZ group, the mental symptoms were assessed by the positive and negative symptom scale (PANSS), cognitive function was assessed by the MATRICS consensus cognitive battery (MCCB), brain-derived neurotrophic factor (BDNF), glial cell derived neurotrophic factor (GDNF), fasting glucose (FGB) and fasting insulin (FINS) levels were detected, and insulin resistance (HOMA-IR) was calculated. The same methods were used to evaluate cognitive function, measure BDNF, GDNF, FGB and FINS levels, and calculate HOMA-IR in HC group. The indexes with statistically significant differences between the two groups were selected to be included in the model. The diagnostic model was constructed by machine learning and verified by cross-validation method, the receiver operating curve (ROC) was plotted, and the area under the curve (AUC), sensitivity and specificity were calculated. Results: (1) A total of 142 patients (70 males and 72 females) with schizophrenia were finally included, and aged (25±4) years. Meanwhile, 140 healthy controls (72 males and 68 females) were also enrolled, and aged (26±4) years. In SCZ group, scores in all areas of cognitive function were lower than those in HC group (all P<0.001), the levels of serum BDNF and GDNF [(6.7±1.8) ng/ml and (405±93) pg/ml] were also lower than those in HC group [(12.3±3.2) ng/ml and (574±139) pg/ml] (both P<0.001), but the levels of FINS and HOMA-IR [(8.4±0.8) μU/ml and 1.7±0.3] were higher than those in HC group [(6.7±0.9) μU/ml and 1.4±0.3] (both P<0.001). (2) Correlation analysis showed that the level of serum BDNF had a negative correlation with negative symptom scores and total scores (r=-0.31, P<0.001; r=-0.17, P=0.040), but had a positive correlation with attention/alertness (CPT-IP) T scores, working memory (WSM-Ⅲ) T scores and visual learning (BVMT) T scores in SCZ group (r=0.39, 0.37 and 0.29, all P<0.001). The level of serum GDNF also had a positive correlation with CPT-IP T scores, WSM-Ⅲ T scores and BVMT T scores (r=0.32, P<0.001; r=0.23, P=0.007; r=0.40, P<0.001). The values of HOMA-IR had a positive correlation with social cognition (MSCEIT) T scores in SCZ group (r=0.18, P=0.033). (3) AUC of the early diagnosis model constructed by combining BDNF, GDNF and HOMA-IR was 0.890 (95%CI: 0.832-0.940), the accuracy was 0.89, the sensitivity and specificity was 0.94 and 0.82, respectively. Conclusion: The final diagnostic model based on biomarkers of serum neurotrophic factor has good diagnostic efficiency for SCZ, but large-scale independent sample verification is still needed.
目的： 构建基于血清神经营养因子等生物标记物的精神分裂症（SCZ）诊断模型。 方法： 前瞻性选取2017年1月至2019年12月在郑州大学第一附属医院就诊的首发未服药SCZ患者（SCZ组）和同期的健康对照者（HC组），SCZ组分别采用阳性和阴性症状量表（PANSS）和认知功能成套测验（MCCB）评估精神症状和认知功能，并检测外周脑源性神经营养因子（BDNF）、胶质细胞源性神经营养因子（GDNF）、空腹血糖（FGB）和空腹胰岛素（FINS）水平，评估受试者胰岛素抵抗情况（HOMA-IR），HC组采用同样方法评估认知功能和测定BDNF、GDNF、FGB、FINS水平及评估HOMA-IR，选择两组间差异有统计学意义的指标纳入模型。通过机器学习构建诊断模型，用交叉验证方法进行验证，并绘制受试者工作特征（ROC）曲线，计算平均ROC曲线下面积（AUC）、灵敏度、特异度。 结果： （1）最终纳入SCZ患者142例，男70例，女72例，年龄（25±4）岁；HC 140名，男72名，女68名，年龄（26±4）岁。SCZ组认知功能各领域评分均低于HC组（均P<0.001），血清BDNF、GDNF水平［（6.7±1.8）ng/ml、（405±93）pg/ml］低于HC组［（12.3±3.2）ng/ml、（574±139）pg/ml］（均P<0.001），而FINS水平、HOMA-IR［（8.4±0.8）μU/ml、1.7±0.3］则高于HC组［（6.7±0.9）μU/ml、1.4±0.3］（均P<0.001）。（2）相关分析结果显示，SCZ患者的BDNF水平与PANSS中阴性症状分、总分呈负相关（r=-0.31、P<0.001，r=-0.17、P=0.040），与认知功能领域中的注意及警觉性（CPT-IP）T分、工作记忆（WSM-Ⅲ）T分、视觉学习（BVMT）T分呈正相关（r=0.39、0.37、0.29，均P<0.001）；GDNF水平与认知功能领域中的CPT-IP T分、WSM-Ⅲ T分、BVMT T分呈正相关（r=0.32、P<0.001，r=0.23、P=0.007，r=0.40、P<0.001）；HOMA-IR与认知功能领域中的社会认知（MSCEIT）T分呈正相关（r=0.18，P=0.033）。（3）将BDNF、GDNF和HOMA-IR组合构建的早期诊断模型的AUC为0.890（95%CI：0.832~0.940），准确度为0.89，灵敏度为0.94，特异度为0.82。 结论： 以血清神经营养因子为主的生物标记物建立的最终诊断模型对SCZ有较好的诊断效能，但仍需大规模独立样本验证。.