Major depressive disorder (MDD) is among the most common mental disorders, and it features particularly severe social function impairment among depressive disorders (Lu et al., 2021). Drug therapy is the preferred and main MDD treatment method (Cipriani et al., 2018). Selective serotonin reuptake inhibitors (SSRIs) are often the preferred first-line antidepressant drugs and are thus used widely in clinical practice, but their therapeutic effect is limited and they are not effective for all patients with depression (Cipriani et al., 2009). In the Sequenced Treatment Alternatives to Relieve Depression study, in which the SSRI citalopram was used as a therapeutic agent, depression remission rates were only 28–33 % (Trivedi et al., 2006). The low efficacy of antidepressants leads to the use of prolonged treatment cycles, which increases the economic and living burdens on patients with depression (Greenberg et al., 2021) and reduces their medication adherence(Zheng, 2022). It is necessary to develop an efficient model for the early prediction of medication efficacy based on a plausible predictive hypothesis to reduce the length of MDD treatment cycles and increase the cure rate.

Despite the recent emergence of antidepressants with rapid onset mechanisms (Nanxin et al., 2010), SSRIs remain the main first-line drugs for MDD treatment. The traditional monoamine hypothesis explains their therapeutic mechanism by focusing on neurotransmitter imbalances (Willner et al., 2013), which posits that depression is caused by the insufficient activity of monoaminergic neurons, due especially to low levels of 5-hydroxytryptamine (5-HT) and norepinephrine in the brain (Schildkraut, 1965). However, this neurotransmitter-focused explanation of the therapeutic effects of antidepressants does not account for the lag between rapid increases in synaptic monoamine concentrations and the slower onset of clinical benefits, nor does it explain why antidepressants are ineffective for some patients with depressive disorders (Bell et al., 2001; Delgado et al., 1999). These observations suggest that changes in the concentrations of neurotransmitters, including 5-HT, are insufficient to elicit clinical effects. The main reason for the delayed onset of antidepressant effects may be the complexity of changes in neuroplasticity resulting from combinations of physiological and social influences. Thus, the consideration of a broad range of possible psychological, physiological, and social factors when attempting to predict antidepressant efficacy is essential.

In 2021, Godlewska and Harmer expanded and refined the cognitive neuropsychological model (Roiser et al., 2012), and proposed the cognitive neuropsychological (CNP) theory, which integrates physiological, social, and psychological factors, providing a theoretical framework for the understanding of the delayed onset of antidepressant drug effects (Godlewska and Harmer, 2021). Patients with MDD who take antidepressants undergo a series of physiological changes within a few hours, which form the basis for subsequent changes in neural activity and behavioral preferences. These changes often result in an improvement in negative bias in the early stages of treatment (1–2 weeks), but this improvement is not stable. Individuals must engage in social interactions to learn and strengthen their positive bias to achieve sustained improvement. This process takes 4–6 weeks, during which time the processing of different brain systems regains balance. Due to this complex process, several weeks are often required for antidepressants to exert therapeutic effects. Unlike previous biologically based theories, the CNP hypothesis incorporates numerous psychosocial factors that are relevant to the etiology of depression and drug efficacy, such as negative cognitive emotional processing bias and social support. Multiple studies conducted with diverse populations, including healthy volunteers (Harmer et al., 2011; Murphy et al., 2009), individuals with high neuroticism (Di Simplicio et al., 2014), depressed patients (Komulainen et al., 2018; Shiroma et al., 2014a), and those who have recovered from depression (Anderson et al., 2011), have supported the CNP hypothesis.

The core idea of the CNP theory is the relationship between emotional processing and depression, that is, negative processing bias is a key factor in the development and maintenance of depression. It suggests that antidepressant medications can facilitate improvements in this bias, leading to a reduction in depressive symptoms. In reality, the CNP theory does not explicitly delineate indicators reliably reflective of negative processing bias. Prior research has conventionally employed questionnaire methodologies and emotion recognition tasks to gauge negative processing bias. Commonly used questionnaires for measuring negative processing bias include the Dysfunctional Attitudes Scale (DAS) (Weissman and Beck, 1978), Automatic Thoughts Questionnaire (ATQ) (Hollon and Kendall, 1980), and Cognitive Bias Questionnaire (CBQ) (Krantz and Hammen, 1979). Previous research using the CBQ found that pre-existing negative cognition can independently predict the speed of improvement in depressive symptoms (Beevers et al., 2007). In 2011, Beck and colleagues proposed that negative cognitive schemas manifest in daily life as negative attention bias, negative processing bias, negative memory bias, and rumination (Disner et al., 2011a). The Negative Cognitive Processing Bias Questionnaire (NCPBQ) directly measures the four components mentioned above(Zhang, 2015). Therefore, utilizing this scale to measure negative processing bias can provide more detailed information. Additionally, researchers utilized emotion recognition tasks and consistently found that the improvement in early cognitive bias can predict the response and relief in antidepressant treatment (Shiroma et al., 2014a; Tranter et al., 2009). Overall, regardless of the assessment method, the findings support that improvements in negative processing bias may be a potential mechanism for alleviating depressive symptoms. Moreover, neuroticism is widely recognized as a crucial endophenotype reflecting emotional processing abnormalities in MDD (Webb et al., 2016), exhibiting a significant correlation with negative processing bias (Klamer et al., 2017). Higher neuroticism in MDD is more likely to achieve better outcomes from sertraline treatment compared to a placebo (Webb et al., 2019). Therefore, it is essential to incorporate neuroticism when investigating the relationship between negative processing bias and medication response.

Additionally, the CNP theory emphasizes the importance of social support and positive interactions in clinical improvement. As an integrated theory, the CNP theory includes more evidence of the association between etiologically related psychosocial factors and drug efficacy than the previous purely biological basis hypothesis, such as negative cognitive emotion processing bias and social support. Therefore, the prediction model based on the CNP hypothesis may achieve better results than the conventional physiological characteristics model, which also provides a new direction for the research of objective markers of depression efficacy. Previous research on the efficacy of antidepressant medications has often examined individual factors in isolation, using univariate analysis to assess the impact of individual factors. Moreover, these studies have primarily relied on cross-sectional comparisons between groups, lacking longitudinal validation, which greatly restricts the generalizability of research findings and makes it difficult to translate them into clinical practice.

In recent years, the rise of computational psychiatry has led an increasing number of researchers to attempt to incorporate multidimensional characteristics in the prediction of therapeutic efficacy (Adams et al., 2015). Among them, machine learning, a common research method, can better solve the problem of fusing multiple modal features, particularly in longitudinal prediction, with strong cross-time stability (Gao et al., 2018). However, most current studies of the application of machine-learning technology to the prediction of drug efficacy have had cross-sectional and classified designs. Research on the establishment of predictive models using strictly longitudinal cohort data is lacking. Having a strong theoretical hypothesis can be critical for developing predictive models for clinical treatment, rather than solely relying on data-driven machine learning. That's why this study's approach, which is based on the CNP theory, shows promise for building a clinical predictive model.

The original CNP theory did not propose physiological indicators for integrated predictive models of antidepressant efficacy, which may have impeded the development of such models. Physiological indicators should be selected for inclusion in the CNP theory based on previous studies of objective markers of depression. In recent years, objective markers of antidepressant drug response (molecular, electrophysiological, brain imaging, and clinical) have become a hot topic in depression research. Among these markers, resting-state electroencephalography (EEG) has outstanding advantages such as minimal equipment requirements, low cost, and ease of data collection, and can be used widely in primary hospitals. This method provides highly accurate data and enables the direct quantification of brain activity, making it one of the best for brain imaging–based research transformation and application (Michel and Murray, 2012). Alpha asymmetry and theta cordance are resting-state EEG indicators that show promising application prospects, with numerous findings highlighting their potential relationships to drug efficacy (Bares et al., 2010; Jaworska et al., 2012a). Thus, they served as key physiological indicators in the predictive model of antidepressant efficacy developed in this study.

In this study, a clinical predictive model based on the CNP theory and machine learning was developed to examine SSRI efficacy in the treatment of MDD. The model features a multimodal fusion of cognitive psychological characteristics, objective resting-state EEG markers, socioenvironmental factors, and clinical symptoms. We hypothesized that the model can effectively predict the response outcomes of patients with first-episode MDD receiving SSRIs for 8 weeks.