用机器学习方法解构精神病性障碍中的认知损害
这项跨6个中心的横断面研究纳入3370名参与者,使用极端梯度提升回归和嵌套交叉验证构建认知综合评分预测模型。模型预测认知评分的样本外皮尔逊相关系数为r=0.72(SD=0.03)。精神分裂症、分裂情感性障碍和伴精神病性症状的双相I型障碍患者的认知综合评分均显著低于对照组,且诊断和药物之外的因素解释了大部分认知损害。
要点
- 精神分裂症(z=-1.4)、分裂情感性障碍(z=-1.2)和伴精神病性症状的双相I型障碍(z=-0.5)患者认知综合评分均显著低于对照组。
- 诊断和药物之外的因素分别解释精神分裂症z=-0.73、分裂情感性障碍z=-0.64、伴精神病性症状的双相I型障碍z=-0.13的认知损害。
- 诊断本身解释的认知损害比例较小(精神分裂症z=-0.29、分裂情感性障碍z=-0.15、伴精神病性症状的双相I型障碍z=-0.13),抗精神病药使用在各诊断组中解释的损害相似(精神分裂症z=-0.37、分裂情感性障碍z=-0.33、伴精神病性症状的双相I型障碍z=-0.26)。
核心结论
- 本研究发现跨诊断因素解释了精神病性障碍中认知功能方差的 meaningful 部分,精神病性障碍中相当比例的认知损害可能反映了与一般人群认知功能相关的因素。
- 作者认为,在考虑干预时,采用不依赖诊断、以症状为目标的方法可能是合适的。
研究设计与样本
- 这项横断面研究使用双相-精神分裂症中间表型网络研究1和2的数据,覆盖6个中心。
- 参与者包括健康对照、精神分裂症患者、分裂情感性障碍患者、伴精神病性症状的双相I型障碍患者及患者亲属。
- 使用极端梯度提升回归和嵌套交叉验证训练认知综合评分预测模型,并使用Shapley加法解释值检验暴露与认知功能的关系。
主要发现
- 共纳入3370名参与者:840名健康对照、709名精神分裂症患者、541名分裂情感性障碍患者、457名伴精神病性症状的双相I型障碍患者、823名患者亲属。
- 平均(SD)年龄为37.9(13.3)岁;女性1887名(56%),男性1483名(44%)。
- 模型预测认知评分的样本外皮尔逊相关系数为r=0.72(SD=0.03)。
- 精神分裂症(z=-1.4)、分裂情感性障碍(z=-1.2)和伴精神病性症状的双相I型障碍(z=-0.5)患者的认知综合评分均显著低于对照组。
- 诊断和药物之外的因素分别解释精神分裂症z=-0.73、分裂情感性障碍z=-0.64、伴精神病性症状的双相I型障碍z=-0.13的认知损害;诊断分别解释z=-0.29、z=-0.15、z=-0.13;抗精神病药使用分别解释z=-0.37、z=-0.33、z=-0.26。
关键数值与推荐
- 认知评估使用精神分裂症简明认知评估量表。
- 暴露因素包括年龄、性别、种族和民族、童年逆境、教育、父母教育、父母社会经济地位、父母生育年龄、物质使用、抗精神病药剂量和诊断。
临床怎么用
- 作者认为,在考虑干预时,采用不依赖诊断、以症状为目标的方法可能是合适的。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
IMPORTANCE: Cognitive functioning is associated with various factors, such as age, sex, education, and childhood adversity, and is impaired in people with psychosis. In addition to specific effects of the disorder, cognitive impairments may reflect a greater exposure to general risk factors for poor cognition. OBJECTIVE: To determine the extent that impairments in cognition in psychosis reflect risk factor exposures. DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional study examined the relationship between exposures and cognitive function using data from the Bipolar-Schizophrenia Network on Intermediate Phenotypes studies 1 and 2 across 6 sites. Participants included healthy controls; patients with schizophrenia, schizoaffective disorder, or bipolar I disorder with psychosis; and relatives of patients. Predictive modeling was performed using extreme gradient boosting regression to train a composite cognitive score prediction model with nested cross-validation. Shapley additive explanations values were used to examine the relationship between exposures and cognitive function. EXPOSURE: Exposures were chosen based on associations with cognition previously identified: age, sex, race and ethnicity, childhood adversity, education, parental education, parental socioeconomic status, parental age at birth, substance use, antipsychotic dose, and diagnosis. MAIN OUTCOMES AND MEASURES: Cognition was assessed using the Brief Assessment of Cognition in Schizophrenia. RESULTS: A total of 3370 participants were included: 840 healthy controls, 709 patients with schizophrenia, 541 with schizoaffective disorder, 457 with bipolar I disorder with psychosis, and 823 relatives of patients. The mean (SD) age was 37.9 (13.3) years; 1887 were female (56%) and 1483 male (44%). The model predicted cognitive scores with high accuracy: out-of-sample Pearson correlation between predicted and observed cognitive composite score was r = 0.72 (SD = 0.03). Individuals with schizophrenia (z = -1.4), schizoaffective disorder (z = -1.2), and bipolar I disorder with psychosis (z = -0.5) all had significantly worse cognitive composite scores than controls. Factors other than diagnosis and medication accounted for much of this impairment (schizophrenia z = -0.73, schizoaffective disorder z = -0.64, bipolar I disorder with psychosis z = -0.13). Diagnosis accounted for a lesser proportion of this deficit (schizophrenia z = -0.29, schizoaffective disorder z = -0.15, bipolar I disorder with psychosis z = -0.13), and antipsychotic use accounted for a similar deficit across diagnostic groups (schizophrenia z = -0.37, schizoaffective disorder z = -0.33, bipolar I disorder with psychosis z = -0.26). CONCLUSIONS AND RELEVANCE: This study found that transdiagnostic factors accounted for a meaningful share of the variance in cognitive functioning in psychosis. A significant proportion of the cognitive impairment in psychosis may reflect factors relevant to cognitive functioning in the general population. When considering interventions, a diagnosis-agnostic, symptom-targeted approach may therefore be appropriate.
引用
- McCutcheon RA, Keefe RSE, McGuire PM, et al. Deconstructing Cognitive Impairment in Psychosis With a Machine Learning Approach. JAMA Psychiatry. 2025. PMID: 39382875. doi:10.1001/jamapsychiatry.2024.3062
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。