Photo Credit: Monsitj
The following is a summary of “Prediction of cognitive impairment using higher order item response theory and machine learning models,” published in March 2024, issue of Psychiatry by Yao et al.
Early identification of cognitive impairment (CI) offers numerous advantages for patients and their families, including uncovering reversible or treatable causes, offering treatment opportunities to alleviate symptoms, and addressing safety concerns.
Researchers conducted a retrospective study to integrate all available data into a machine-learning model to enhance the prediction of CI in older adults.
They involved 86 elderly individuals and employed two validated iPad-based measures, picture sequence memory (PSM) and dimensional change card sort test (DCCS), from the NIH Toolbox ®. The effectiveness of various machine learning models, including traditional classifiers and neural networks, in predicting CI was compared. The performance of each model was evaluated using a 100-fold bootstrap replication process.
The results showed that the random forest model emerged as the most effective in predicting CI, achieving accuracy (0.902), precision (0.803), recall (0.758), F1 score (0.742), and specificity (0.951). This model incorporated a composite score derived from a 2-parameter higher-order item response theory (HOIRT) model, combining information from DCCS and PSM assessments.
Investigators concluded that more than just relying on fixed composite score cutoffs were needed for accurate CI prediction. Instead, it incorporates machine learning models that utilize HOIRT-derived scores and relevant features.
Source: frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2023.1297952/full