Fusar‐Poli et al's insightful paper¹ is a timely appraisal of the foundations of preventive psychiatry. It is a call to action for our field to mount an individual, societal and global response to improve the lives of people with and those at risk for mental disorders. The authors outline a series of ambitious next steps in preventive psychiatry. They seek to advance this goal by integrating universal and targeted frameworks and by advancing our epidemiological knowledge of the multifactorial causation of mental disorders. An additional important step is to use such data toward devel­oping stratified and personalized approaches. However, a major challenge in tackling these ambitious goals is the enormous heterogeneity of mental disorders, at symptomatic, pathophysiologic and etiologic levels. In this light, several strategies deserve consideration toward a successful move forward with Fusar‐Poli et al's suggested next steps.

Any effort at prevention should first clar­ify what we are planning to prevent. For this reason, an accurate and valid diagnosis is critically important. As the authors point out, caseness is difficult to determine in psychiatry, because the disorders are defined based on symptoms, not on biology. For this reason, psychiatric diagnostic systems currently lack validity². A biomarker‐based nosology is clearly a critical next step toward stratification of populations meaningfully separating more homogeneous en­tities.

In a biomarker‐driven effort to address the heterogeneity of psychotic disorders, investigators in the Bipolar‐Schizophre­nia Network on Intermediate Phenotypes (BSNIP) consortium recently used a K‐means clustering approach to parse alterations in cognition and electrophysiology (event‐related potentials and eye tracking) across the three major psychotic disorders: schizophrenia, schizoaffective disorder, and psychotic bipolar disorder.

Three distinct “biotypes” were identified which seemed orthogonal to the DSM‐based categories³. Biotype 1 is characterized by severe cognitive impairments, reduced neural response to salient stimuli, marked gray matter reductions, social function deficits, more frequent family history of psychosis, and prominent negative symptoms. Biotype 2 is marked by moderate cognitive and social impairments and gray matter reductions, and by enhanced neural reactivity. Biotype 3 shows few neurobiological differences from healthy controls. These ob­servations point to the possibility that bio­marker‐derived classifications may potentially better distinguish subtypes within the psychotic spectrum.

However, having a disease‐related biomarker is not sufficient for early identification and prevention purposes, unless the biomarker is demonstrated to be present at illness onset or even before overt clinical manifestations of the disorders. This points to the potential value of identifying premorbid biotypes. Interestingly, biotype 1 appears to identify the deficit syndrome, and premorbid adjustment and cognitive profile can distinguish the schizophrenia deficit subgroup with moderate accura­cy⁴. It is noteworthy that biotype 1 is associated with higher frequency of family history of psychosis compared to the other biotypes. It is also of interest that cognitive impairment and family history of psychosis⁵, as well as biomarkers characterizing biotype 1 such as decreased auditory P300 amplitudes⁶, are together strong predictors of risk for conversion to psychosis among individuals at clinical high risk.

A testable prediction, therefore, is whether biotype 1 psychosis may be preceded by a biotype 1‐like biomarker signature in the premorbid phase of the illness that is similar to the features seen later in this subtype. Likewise, it is possible that a biotype 1‐like biomarker profile may predict impaired functional outcome in early course psychosis patients. Identifying such premorbid bio‐signatures requires prospective longitudinal characterization in individuals at familial and clinical high‐risk, and those in the early course of a psychotic illness.

Neurobiological entities seem to cut a­cross psychiatric diagnostic categories. Con­sistent with this view, biotypes of depression⁷ and autism⁸ have been identified in studies examining the heterogeneity of these syndromes. Interestingly, similar to psychotic disorders, cognitive impairments may serve as valuable stratification markers in these populations as well.

It is useful to consider biomarker‐driven approaches in the light of the traditional (primary vs. secondary vs. tertiary) and the more recent (US Institute of Medicine and World Health Organization) models of prevention outlined by Fusar‐Poli et al. The identification of transdiagnostic premorbid biomarker signatures and biotypes may be of particular relevance to the field of selective prevention, though not for universal prevention. Biomarker‐driven prediction is an aspirational goal for primary selective prevention (e.g., preventing psychosis in individuals at familial high risk for psychosis), though more work is needed in this area. On the other hand, there is emerging evidence in the literature supporting the possibility of predicting psychosis for indicated secondary prevention in individuals at clinical high risk for psychosis⁶, and of predicting relapse and functional outcome for the purpose of tertiary prevention in patients in the early course of psychosis⁹.

The steady expansion of new knowledge of brain function, and of new approaches, such as imaging, genetics, proteomic and metabolomic technologies, offers the possibility for developing predictive biomarkers in the near future. However, the complex multifactorial determination of mental illnesses and the enormous amount of the available “omics” data make this goal challenging. As Fusar‐Poli et al rightly point out, advancing stratified approaches for prevention requires a multicausal, transdiagnostic, multifinal epidemiological knowledge at an individual level. Large multi‐site studies, carefully characterized populations, and sophisticated computational approaches, including machine learning, are needed to generate and harness such “big” data sets toward the development of actionable biomarkers for personalized medicine.

In summary, I agree with Fusar‐Poli et al's articulation of the need to urgently develop a blueprint for preventive strategies in psychiatry. First, a transdiagnostic view may be applicable not only to psychoses as outlined here, but to all of psychiatric disorders. Second, a neuroscience‐based categorization of distinct subtypes in these disorders, as opposed to symptom‐based categories, may improve our ability to predict outcome and treatment response. Finally, extending such a translational approach to clinical and familial high‐risk states and to early course clinical populations may help identify early predictors of illness and enable individually tailored preventive interventions.

Fusar-Poli等人的有见地的论文¹是对预防精神病学基础的及时评估。这是呼吁我们的领域采取行动，以个人，社会和全球的方式作出反应，以改善精神障碍患者和有精神障碍风险的人们的生活。作者概述了预防精神病学中的一系列宏伟的下一步。他们试图通过整合通用和针对性的框架并通过提高我们对精神障碍的多因果关系的流行病学知识来推进这一目标。另一个重要的步骤是将此类数据用于开发分层和个性化的方法。但是，解决这些宏伟目标的主要挑战是在症状，病理生理和病因水平上精神疾病的巨大异质性。有鉴于此，

预防方面的任何努力都应首先阐明我们计划预防的内容。因此，准确有效的诊断至关重要。正如作者所指出的，在精神病学中很难确定病因，因为这种疾病是根据症状而不是生物学来定义的。因此，精神病诊断系统目前缺乏有效性²。基于生物标志物的分类学显然是朝着有意义地分离更均质实体的人群分层迈出的关键一步。

为了解决精神病性疾病的异质性问题，由生物标记驱动的研究中，双相精神分裂症中型表型网络（BSNIP）的研究人员最近使用K-均值聚类方法来分析认知和电生理的变化（与事件有关的电位和眼睛）追踪）涵盖三种主要的精神病性疾病：精神分裂症，分裂情感障碍和精神病性双相情感障碍。

确定了三个不同的“生物型”，它们似乎与基于DSM的类别³正交。生物型1的特征是严重的认知障碍，对显性刺激的神经反应减少，灰质明显减少，社会功能缺陷，精神病家族史更加频繁以及明显的阴性症状。生物型2的特征是中等程度的认知和社交障碍以及灰质减少，以及神经反应性增强。生物型3与健康对照组相比几乎没有神经生物学差异。这些观察结果表明，生物标志物衍生的分类可能会更好地区分精神病谱系中的亚型。

但是，拥有疾病相关的生物标志物不足以进行早期识别和预防，除非已证明该生物标志物在疾病发作时或在疾病的明显临床表现之前就已存在。这表明了鉴定病前生物型的潜在价值。有趣的是，生物型1出现来识别缺陷综合征，和病前调整和认知轮廓可以区分患有中度ACCURA-CY精神分裂症赤字子群⁴。值得注意的是，与其他生物型相比，生物型1与精神病家族史的发生频率更高。认知障碍和精神病的家族史也很有趣⁵以及表征生物型1的生物标志物，例如听觉P300振幅降低⁶，都是临床高危人群中转化为精神病风险的有力预测指标。

因此，一个可检验的预测是，在疾病的发病前阶段，是否可能在生物型1精神病之前出现类似于生物型1的生物标志物签名，这一特征与该亚型后面的特征相似。同样，生物型1样生物标志物谱可能会预测早期病程性精神病患者的功能预后受损。识别这种病前生物特征需要对家族和临床高风险个体以及精神病早期患者进行前瞻性纵向表征。

Neurobiological entities seem to cut a­cross psychiatric diagnostic categories. Con­sistent with this view, biotypes of depression⁷ and autism⁸ have been identified in studies examining the heterogeneity of these syndromes. Interestingly, similar to psychotic disorders, cognitive impairments may serve as valuable stratification markers in these populations as well.

It is useful to consider biomarker‐driven approaches in the light of the traditional (primary vs. secondary vs. tertiary) and the more recent (US Institute of Medicine and World Health Organization) models of prevention outlined by Fusar‐Poli et al. The identification of transdiagnostic premorbid biomarker signatures and biotypes may be of particular relevance to the field of selective prevention, though not for universal prevention. Biomarker‐driven prediction is an aspirational goal for primary selective prevention (e.g., preventing psychosis in individuals at familial high risk for psychosis), though more work is needed in this area. On the other hand, there is emerging evidence in the literature supporting the possibility of predicting psychosis for indicated secondary prevention in individuals at clinical high risk for psychosis⁶, and of predicting relapse and functional outcome for the purpose of tertiary prevention in patients in the early course of psychosis⁹.

对脑功能的新知识以及诸如成像，遗传学，蛋白质组学和代谢组学技术等新方法的不断扩展，为在不久的将来开发预测性生物标志物提供了可能性。但是，对精神疾病的复杂多因素确定以及大量可用的“组学”数据使这一目标具有挑战性。正如Fusar-Poli等正确地指出的那样，推进分层的预防方法需要在个人层面上具有多因果关系，经过诊断，多方面的流行病学知识。为了生成和利用此类“大”数据集来开发个性化医学的可行生物标记物，需要进行大型的多站点研究，仔细表征的种群以及包括计算机学习在内的复杂的计算方法。

总而言之，我同意Fusar-Poli等人的观点，即有必要紧急制定精神病学预防策略的蓝图。首先，经诊断的观点不仅适用于此处概述的精神病，而且适用于所有精神病。其次，与基于症状的类别相反，这些疾病中不同亚型的基于神经科学的分类可能会提高我们预测结局和治疗反应的能力。最后，将这种转换方法扩展到临床和家族性高危状态以及早期临床人群可能有助于确定疾病的早期预测因素，并能够进行量身定制的预防性干预措施。