Integrating Artificial Intelligence into International Classification of Functioning, Disability, and Health Coding: Effectiveness of a Mobile Application for Patient Questionnaires -Randomized controlled trial

Zhandos Kurban¹Didar Khassenov^2,3*Zhandos Burkitbaev⁴Sholpan Bulekbayeva⁵Azat Chinaliyev⁴Serik Bakhtiyar³Samat Saparbayev⁶Tokan Sultanaliyev⁷Ulzhalgas Zhunissova⁸Natalia Slivkina¹Elena Titskaya⁹Luis Arias¹⁰Dana Aldakuatova³Gulfairus Yessenbayeva⁷Zhanerke Ermakhan³

• ¹Department of Rehabilitation and Sports Medicine, NJSC “Astana Medical University”, Astana, Kazakhstan
• ²Department of Interventional Radiology, National Research Oncology and Transplantation Center, Nur-sultan, Kazakhstan
• ³Department of Public Health and Hygiene, NJSC “Astana Medical University”, Astana, Kazakhstan
• ⁴National Scientific Oncology Center, Astana, Kazakhstan
• ⁵National Scientific Center for the Development of the Social Protection Sector, Almaty, Kazakhstan
• ⁶"Al-Jami " LLC, Aktobe, Kazakhstan
• ⁷National Research Oncology Center LLP, Astana, Kazakhstan
• ⁸Department of biostatistics, bioinformatics and information technologies, NJSC “Astana Medical University”, Astana, Kazakhstan
• ⁹Tomsk Research Institute of Balneology and Physiotherapy of the Siberian Federal Research and Clinical Center of the Federal Medical and Biological Agency, Tomsk, Russia
• ¹⁰Department of Scientific Institute of Higher Education, Santa Cruz De La Sierra, Mexico

Background: Mobile health applications and artificial intelligence (AI) are increasingly utilized to streamline clinical workflows and support functional assessment. The International Classification of Functioning, Disability and Health (ICF) provides a standardized framework for evaluating patient functioning, yet AI-driven ICF mapping tools remain underexplored in routine clinical settings.Objective: This study aimed to evaluate the efficiency and accuracy of the MedQuest mobile applicationfeaturing integrated AI-based ICF mappingcompared to traditional paper-based assessment in hospitalized patients. Methods: A parallel-group randomized controlled trial was conducted in two medical centers in Astana, Kazakhstan. A total of 185 adult inpatients (≥18 years) were randomized to either a control group using paper questionnaires or an experimental group using the MedQuest app. Both groups completed identical standardized assessments (SF-12, IPAQ, VAS, Barthel Index, MRC scale). The co-primary outcomes were (1) total questionnaire completion time and (2) agreement between AIgenerated and clinician-generated ICF mappings, assessed using quadratic weighted kappa. Secondary outcomes included AI sensitivity/specificity, confusion matrix analysis, and physician usability ratings via the System Usability Scale (SUS). Results: The experimental group completed questionnaires significantly faster than the control group (median 18 vs. 28 minutes, p < 0.001). Agreement between AI-and clinician-generated ICF mappings was substantial (κ = 0.842), with 80.6% of qualifiers matching exactly. The AI demonstrated high sensitivity and specificity for common functional domains (e.g., codes 1-2), though performance decreased for rare qualifiers. The micro-averaged sensitivity and specificity were 0.806 and 0.952, respectively. Mean SUS score among physicians was 86.8, indicating excellent usability and acceptability.The MedQuest mobile application significantly improved workflow efficiency and demonstrated strong concordance between AI-and clinician-assigned ICF mappings. These findings support the feasibility of integrating AI-assisted tools into routine clinical documentation. A hybrid model, combining AI automation with clinician oversight, may enhance accuracy and reduce documentation burden in time-constrained healthcare environments.