| Objective: To test the consistency of artificial intelligence neural network algorithm model and rehabilitation physicians’ assessment of ICF-RS function classification of stroke inpatients. Methods: After national standardized ICF-RS training, rehabilitation physicians in the same ward evaluated 3 stroke inpatients. They were familiar with the evaluation process and mastered the evaluation criteria. Subsequently, each rehabilitation physician routinely used ICF-RS to evaluate 106 hospitalized stroke patients in their daily diagnosis and treatment work, and graded the three dimensions of body, activity and participation and the overall function, and compared with the functional level constructed by artificial intelligence neural network algorithm model. After data collection, Kendall’s coefficient of concordance was used to evaluate the consistency of the limited values of various items in three patients, and weighted kappa coefficient and intra group correlation coefficient (ICC) were used to evaluate the consistency between the prediction of artificial intelligence algorithm model and the functional classification of rehabilitation physicians’ clinical evaluation. Results: Kendall’s coefficient of concordance of ICF-RS categories of the three patients were 0.765, 0.849, and 0.874 respectively (all P<0.01), and the consistency was improved with the increase of patients. The weighted Kappa for the artificial intelligence algorithm model and rehabilitation physicians in determining functional classification in physical function, activity, participation, and overall functional grading of 106 inpatients with stroke were 0.718, 0.737, 0.750, and 0.825 respectively (all P<0.01), and the ICC coefficients were 0.789, 0.801, 0.806, and 0.862 respectively (all P<0.01). Conclusion: The consistency of ICF-RS assessment of hospitalized stroke patients by rehabilitation physicians is good, and the consistency of artificial intelligence algorithm model and clinical assessment of physicians is high. |
