At the recent EPA 2023: 31^st European Congress of Psychiatry, Professor Paz Garcia-Portilla discussed the evolution of schizophrenia staging from the early Clinical Global Impressions Scale (CGI-S) system to the recently empirical schizophrenia staging model developed by her team.¹ 

The utilization of staging in psychiatry could be dated back to 1976 when the CGI-S system was established.¹ This scale allows clinicians to classify patients according to the severity of their illnesses, using a scale from 1 (i.e., normal) to 7 (i.e., among the most severely ill).¹ The CGI-S was intended to be a simple and global measure that reflects a clinician's general impression of a patient's condition.¹ When evaluating the mental health status of a patient, the CGI-S requires clinicians to consider their total clinical experiences.¹ Clinicians frequently utilize this instrument as a heuristic to evaluate patients, then make treatment decisions.¹ 

Although in routine clinical settings, the CGI-S has the potential to be a useful tool for evaluating patient progress and improving adoption, it would be beneficial to have anchors that could be applied across different illnesses and to increase reliability by clarifying the anchor points used for scoring.¹ In a survey involving 24 clinical trial investigators on developing more widely applicable CGI-S scoring anchors in 2017, it was found that symptom severity was the most critical element in determining CGI-S scores, followed by patient functional status.¹ The importance of self-reported symptom scores, staff observations, and adverse effects (AEs) was diminished.¹ The relative importance of those factors did not vary among researchers based on their experiences or patient contact time.¹ 

However, these modifications were insufficient to address the severity of schizophrenia patients due to the need for an integration of additional variables, such as clinical data, biomarkers, and comorbid disorders.¹ In this regard, Prof. Garcia-Portilla and her team developed a new machine learning-based staging model for schizophrenia.¹ This new model incorporates information from 5 different domains, namely the clinical characteristics, number of hospitalizations over the patient's lifetime; the current severity of positive, negative, general, and depressive symptomatology; the scores on the speed of processing, visual learning, and social cognition; the blood peripheral markers of inflammation; and the global level of functionality.¹ 

In this study, two-thirds of 61 samples were utilized for training and one-third for validation by a genetic algorithm.¹ This procedure was repeated 10,000 times with 212 data sets and support vector machines.¹ It was determined that the concordance between the developed staging model and the gold standard (i.e., CGI-S) was 61.97 % [Standard deviation (SD)=5.17%].¹ However, the CGI-S accuracy of C1 vs. S1 and C5 vs. S5 is only 0% and 25%, respectively.¹ Prof. Garcia-Portilla believed that the low concordance at the more severe end of the range could be attributed to using only outpatients in the training model.¹ As a subsequent step, the researchers are currently investigating the specific characteristics of patients whose staging model and CGI-S classifications are inconsistent.¹ 

In conclusion, this new staging model provides a more comprehensive understanding of a patient's condition and hence can be used to develop customized treatment plans.¹ In addition, psychometrical evaluation prior to incorporation into routine clinical practice is essential.¹