McCutcheon et. al. investigates the various factors underlying cognitive impairments across the broad psychosis spectrum, encompassing schizophrenia, schizoaffective disorder, and bipolar I disorder with psychosis. Investigators applied a nonlinear machine-learning model, XGBoost regression, on data from the Bipolar-Schizophrenia Network on Intermediate Phenotypes studies 1 and 2 to examine sociodemographic, clinical, and medication-related contributors to cognitive impairment in patients with psychotic disorders. The study included 3370 participants (840 controls, 709 with schizoaffective disorder, 457 with Bipolar I disorder with psychosis, and 823 relatives of patients). Key findings revealed that reduced education levels, childhood trauma, socioeconomic status, and antipsychotic medication exposure are strongly associated with lower cognitive scores. These associations were consistent across patient groups, with schizophrenia showing the highest impairment. Interestingly, relatives of individuals with psychotic disorders also showed cognitive impairments, though less pronounced. Excluding sociodemographic factors increased the weight of clinical predictors, indicating that cognitive issues are not solely illness-related. Moreover, medication dose had a nonlinear effect, suggesting complex interactions between treatment and cognition. Detecting these complex and nonlinear relationships was a strength of using a nonlinear machine learning algorithm when compared to traditional linear models. The authors propose that cognitive deficits in psychosis reflect not just the illness itself but a mix of shared environmental and genetic factors. These findings align with the cognitive reserve hypothesis, where lower cognitive resilience increases vulnerability to psychosis.