This meta-analysis, conducted by Monopoli et al., explored the usage of machine learning to predict the treatment outcomes of affective and non-affective psychotic disorders. 51 studies on machine learning use, from the MEDLINE, Web of Science, and PsycINFO databases, were incorporated for meta-analysis, in which model accuracy was evaluated. From the results, it was found that machine learning models were able to achieve a high accuracy (0.80), with the most statistically significant features being, from most significant to least: EEG data (accuracy = 0.88), multimodal/combination factors (0.85), fMRI data (0.81), sMRI data (0.79), and clinical and demographic information (0.72). Additionally, machine learning models exhibited a higher accuracy predicting outcomes to combined treatment plans (accuracy = 0.85) than purely pharmacological treatment plans (accuracy = 0.79). In addition to these findings, Monopoli et al. also analyzed study quality through direct examination of sources for bias and through the Prediction Model Risk of Bias Assessment Tool (PROBAST). From this, they found that 44 sources were at a high risk of bias, which could be explained through methodological reasons, such as small sample sizes, which decrease the studies’ overall generalizability.