Objective: To evaluate the feasibility of AI-assisted screening for early Parkinson’s disease (PD) in Tunisian primary care.

Background: PD poses a growing challenge in Tunisia, with an estimated burden of 1.5 years lived with disability (YLD), 5.6 years of life lost (YLL), and 7.1 disability-adjusted life years (DALY) per 100,000 population [1]. However, diagnosis is often delayed due to a shortage of neurologists (2.3 per 100,000). AI tools using voice analysis, motor assessments, and facial recognition have shown promise in early PD detection [2], but their feasibility remains untested.

Method:

A two-phase mixed-methods feasibility study.

Phase 1: Retrospective Analysis of Clinical Pathways

Data: Retrospective patient records (2018–2023) from Tunisian neurology wards.

Outcomes: Time from GP consultation to diagnosis, referral rates, and urban/rural access disparities.

Analysis: Kaplan-Meier for diagnostic delays and logistic regression to identify delay predictors.

Phase 2: Feasibility Assessment of AI-Assisted Screening

AI-Model Selection: Review of validated PD screening models (voice analysis, motor tests) and their applicability to primary care, with performance estimated from literature if local validation is unavailable.

Stakeholder Input: Interviews with GPs and neurologists to assess barriers like internet access, training, and patient acceptability.

Simulation: ABM (NetLogo/Python) to simulate AI screening impact on diagnostic delays, factoring in AI sensitivity, GP adoption, and adherence.

Results: Primary outcome: Estimated reduction in diagnostic delay with AI-assisted screening compared to standard practice, based on retrospective data and simulation projections.

Secondary outcomes: GP adoption feasibility (measured by willingness to integrate AI, training completion rates, and perceived usefulness), patient referral efficiency, and rural vs. urban differences in screening effectiveness.

Sensitivity analyses: The impact of GP training level, AI model accuracy, and varying screening uptake rates on diagnostic performance.

Conclusion: This protocol provides a first step toward AI-integrated PD screening in Tunisian primary care. By assessing feasibility and barriers, it informs future AI-driven interventions in Tunisia and other LMICs.

References: 1. Global Health Estimates: Life expectancy and leading causes of death and disability [Internet]. [cited 2025 Mar 12]. Available from: https://www.who.int/data/gho/data/themes/mortality-and-global-health-estimates
2. Bourouhou, A., Jilbab, A., Nacir, C., & Hammouch, A. (2016). “Comparison of classification methods to detect Parkinson’s disease.” 2016 International Conference on Electrical and Information Technologies (ICEIT), 421-424. IEEE.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/feasibility-of-ai-assisted-screening-for-early-parkinsons-in-resource-limited-settings/