Uludag et al., in a study published in Schizophrenia Research, explored the potential of antipsychotic-based machine learning models for predicting tardive dyskinesia (TD) in patients with schizophrenia. The study employed algorithms such as random forest and support vector machines, and found that the random forest model performed best, achieving 88% accuracy in both the second-generation and first-generation antipsychotic groups. This finding suggests that machine learning approaches could help clinicians identify high-risk patients and thereby support individualized medication decisions.
However, the generalizability and clinical reliability of the model still require validation through more experiments. When the same optimized model was tested on an independent sample, sensitivity may drop dramatically, indicating that model performance may vary considerably across different populations or clinical settings. If applied clinically without sufficient external validation, this could lead to missed diagnoses or misdiagnosis. Therefore, similar approaches should be adopted and encouraged, but they must be accompanied by more prospective, multicenter experiments to confirm their robustness—ensuring that they do not produce wrong diagnoses when used to assist clinical decision-making.
