Schizophrenia risk is associated with both genetic and environmental risk factors. Furthermore, cognitive abnormalities are established core characteristics of schizophrenia. We aim to assess whether a classification approach encompassing risk factors, cognition, and their associations can discriminate patients with schizophrenia (SCZs) from healthy control subjects (HCs). We hypothesized that cognition would demonstrate greater HC-SCZ classification accuracy and that combined gene–environment stratification would improve the discrimination performance of cognition. Genome-wide association study–based genetic, environmental, and neurocognitive classifiers were trained to separate 337 HCs from 103 SCZs using support vector classification and repeated nested cross-validation. We validated classifiers on independent datasets using within-diagnostic (SCZ) and cross-diagnostic (clinically isolated syndrome for multiple sclerosis, another condition with cognitive abnormalities) approaches. Then, we tested whether gene–environment multivariate stratification modulated the discrimination performance of the cognitive classifier in iterative subsamples. The cognitive classifier discriminated SCZs from HCs with a balanced accuracy (BAC) of 88.7%, followed by environmental (BAC = 65.1%) and genetic (BAC = 55.5%) classifiers. Similar classification performance was measured in the within-diagnosis validation sample (HC-SCZ BACs, cognition = 70.5%; environment = 65.8%; genetics = 49.9%). The cognitive classifier was relatively specific to schizophrenia (HC–clinically isolated syndrome for multiple sclerosis BAC = 56.7%). Combined gene–environment stratification allowed cognitive features to classify HCs from SCZs with 89.4% BAC. Consistent with cognitive deficits being core features of the phenotype of SCZs, our results suggest that cognitive features alone bear the greatest amount of information for classification of SCZs. Consistent with genes and environment being risk factors, gene–environment stratification modulates HC-SCZ classification performance of cognition, perhaps providing another target for refining early identification and intervention strategies.

中文翻译：

---

根据遗传和环境风险分层的认知缺陷模式可靠地将精神分裂症患者与健康对照受试者进行分类

精神分裂症风险与遗传和环境风险因素有关。此外，认知异常是精神分裂症的核心特征。我们的目的是评估包含危险因素、认知及其关联的分类方法是否可以区分精神分裂症患者 (SCZ) 和健康对照受试者 (HC)。我们假设认知将表现出更高的 HC-SCZ 分类准确性，并且结合基因-环境分层将提高认知的辨别性能。基于全基因组关联研究的遗传、环境和神经认知分类器经过训练，可使用支持向量分类和重复嵌套交叉验证从 103 个 SCZ 中分离出 337 个 HC。我们使用诊断内（SCZ）和交叉诊断（多发性硬化症的临床孤立综合征，另一种认知异常的疾病）方法在独立数据集上验证了分类器。然后，我们测试了基因-环境多元分层是否调节了迭代子样本中认知分类器的辨别性能。认知分类器以 88.7% 的平衡准确度 (BAC) 区分 SCZ 和 HC，其次是环境分类器 (BAC = 65.1%) 和遗传分类器 (BAC = 55.5%)。在诊断内验证样本中测量了类似的分类性能（HC-SCZ BAC，认知 = 70.5%；环境 = 65.8%；遗传学 = 49.9%）。认知分类器对精神分裂症相对具体（HC——多发性硬化症临床孤立综合征 BAC = 56.7%）。基因-环境组合分层允许认知特征对来自 SCZ 的 HC 进行分类，BAC 为 89.4%。与认知缺陷是 SCZ 表型的核心特征相一致，我们的结果表明，认知特征本身就包含了 SCZ 分类的最大信息量。与基因和环境作为危险因素一致，基因-环境分层调节认知的 HC-SCZ 分类表现，或许为完善早期识别和干预策略提供另一个目标。