A new study has revealed potential predictors of treatment-resistant schizophrenia (TRS) in early-stage patients. The research, conducted on 1,400 patients with first-episode psychosis (FEP) over a 12-17 year follow-up period, aimed to identify individuals at higher risk of developing TR in order to provide personalized interventions and reduce delays in treatment initiation.
TRS affects approximately 15-40% of patients with schizophrenia and is associated with higher healthcare costs and poorer functional outcomes. Clozapine, an antipsychotic medication, is considered the most effective treatment for TRS. However, there are often significant delays in prescribing clozapine, with patients undergoing multiple antipsychotic trials before its initiation. Identifying patients at higher risk of TR development could help reduce these delays.
The study utilized an automated machine learning (autoML) approach to develop four probabilistic classification models based on baseline and longitudinal clinical information. These models aimed to predict the future development of TR in patients with FEP. Over the follow-up period, 191 FEP patients (13.7%) were prescribed clozapine, indicating TR development. The autoML models achieved an area under the receiver operating characteristic curve ranging from 0.676 to 0.774, demonstrating their predictive capabilities.
Various factors were identified as important predictors of TR development, including the diagnosis of schizophrenia, age of onset, symptoms variability, relapse, and the use of antipsychotics and anticholinergic medications. These features were incorporated into a risk calculator, known as TRipCal, to estimate individual risks of TR development in FEP patients. The development of TRipCal supports the ongoing research into data-driven clinical tools that can assist in personalized interventions and reduce delays in clozapine initiation.
Previous studies have investigated potential predictors of TRS, with early age of onset being the most consistent predictor. However, most studies had limited follow-up periods and did not consider treatment outcomes or clinical characteristics during the early stages of the illness. The use of advanced machine learning models, such as autoML, in this study improves prediction performance and provides valuable insights for personalized interventions in the early course of the illness.
The current study focused on clozapine initiation as a proxy for TR status, as clozapine is recommended for TRS patients in many regions. The findings of this study contribute to the understanding of TR development and may help healthcare professionals identify patients at higher risk, particularly in the early stages of the illness. By facilitating personalized and targeted interventions, the aim is to prevent or postpone the development of TR and improve outcomes for patients with FEP.
In conclusion, the new study utilizing autoML has highlighted potential predictors of treatment-resistant schizophrenia in early-stage patients. The risk calculator, TRipCal, developed through this research, offers valuable insights and support for personalized interventions in the early course of the illness. By identifying individuals at higher risk of TR development, healthcare professionals can intervene more effectively and reduce delays in initiating appropriate treatment. With further advancements in data-driven clinical tools, the hope is to continually improve outcomes for patients with schizophrenia.
