Early identification of symptoms that can predict treatment-resistant schizophrenia (TRS) could help clinicians to avoid delays in clozapine therapy. This study aims to investigate symptom patterns that could predict TRS using a discovery/replication study design. First, we followed a cohort of inpatients with schizophrenia (n = 164) in which the most discriminative items at baseline of the Positive and Negative Syndrome Scale (PANSS) were determined using logistic regression with TRS status as an outcome. Using Receiver Operating Characteristic (ROC) curves, we tested the prediction performance of multiple combinations of the identified items. The same items' combination was tested in an independent replication sample of (n = 207) outpatients with schizophrenia. In the discovery sample, the best combination to predict TRS at the discharge was the sum of three baseline PANSS items - conceptual disorganization (P2), difficulty in abstract thinking (N5), and unusual thought content (G9). The P2 + N5 + G9 model yielded an area under the curve (AUC) of 0.881, a sensitivity of 77.8%, and a specificity of 83.3%. In the outpatient sample, the model P2 + N5 + G9 predictive accuracy for TRS was only in the range of "acceptable" with an AUC of 0.756 and sensitivity of 72.3% and a specificity of 74.4%. Overall, the P2 + N5 + G9 model corresponds to the construct of formal thought disorder composed of disorganized thinking, concrete thinking, and bizarre-idiosyncratic thinking. Pronounced levels of these symptoms are easily identifiable in clinical practice and may be a feasible strategy in TRS. Replicating in first-episode cohorts is desirable to understand the likely clinical utility.

中文翻译：

---

预测难治性精神分裂症的症状组合——现实世界临床实践的策略

早期识别可预测难治性精神分裂症 (TRS) 的症状可以帮助临床医生避免氯氮平治疗的延误。本研究旨在调查可以使用发现/复制研究设计预测 TRS 的症状模式。首先，我们跟踪了一组精神分裂症住院患者（n = 164），其中阳性和阴性综合征量表 (PANSS) 基线时最具辨别力的项目是使用逻辑回归将 TRS 状态作为结果确定的。使用接收者操作特征 (ROC) 曲线，我们测试了已识别项目的多种组合的预测性能。在（n = 207）精神分裂症门诊患者的独立复制样本中测试了相同项目的组合。在发现样本中，在出院时预测 TRS 的最佳组合是三个基线 PANSS 项目的总和 - 概念混乱 (P2)、抽象思维困难 (N5) 和异常思维内容 (G9)。P2 + N5 + G9 模型的曲线下面积 (AUC) 为 0.881，敏感性为 77.8%，特异性为 83.3%。在门诊样本中，模型P2+N5+G9对TRS的预测准确率仅在“可接受”范围内，AUC为0.756，敏感性为72.3%，特异性为74.4%。总体而言，P2 + N5 + G9 模型对应于由杂乱无章的思维、具体的思维和奇异的特质思维组成的形式思维障碍的结构。这些症状的显着水平在临床实践中很容易识别，并且可能是 TRS 中的可行策略。