Aim & Relevance
The initiative resolves a critical “Diagnostic Gap” in Electrodiagnostic (EMG) laboratories, where highly specialized physician time is traditionally consumed by low-yield, normal referrals. Indiscriminate pooling creates a 40% “Normal” bottleneck, causing unfair, dangerous delays for patients with urgent pathology. By transitioning from a “First-Come, First-Served” model to an AI-driven “Risk-Based Priority” system, the initiative ensures patients with urgent neuromuscular pathology receive accelerated, equitable access to specialized care.
Unique Approach
Unlike standard models focused on mere accuracy—which is insufficient for uneven clinical populations—this initiative utilizes a “Risk Stratification Engine” optimized for triage, safety, and operational workflow. It uniquely translates medical knowledge into engineered logic, extracting 21 clinical interaction variables to identify hidden patient phenotypes. By defining success through Recall (the “Safety Net” ensuring pathology is not missed) and Precision (building efficiency and trust), it safely separates high-risk patients from routine screenings. This creates distinct operational lanes without requiring additional staffing.
Patient Description & Involvement
The target population comprises patients referred for neuromuscular symptoms, ranging from simple numbness or weakness to queried conditions like Carpal Tunnel Syndrome (CTS), Radiculopathy, or severe pathologies like ALS and Myasthenia. Patients benefit directly through optimized access matched to their clinical acuity. Those with acute, focal needs bypass the 40% diagnostic bottleneck, while younger, lower-risk patients receive efficient, specialized screening.
Methods
Leveraging 10 years of longitudinal clinical data (2014-2024), the AI employs an Engineered XGBoost machine learning model. Crucially, it relies strictly on pre-clinic referral data (e.g., age, sex, referring service) to prevent data leakage. Evaluating variables like biological risk (Age > 55 captures 69% of pathology variance) and clinical acuity, it automatically sorts patients into three tracks: • Track 1 (High Priority): A Consultant-Led Clinic for high-risk patients (e.g., rapid progression), predicting an 85% pathology rate. • Track 2 (Standard Priority): Standard diagnostic EMG for moderate-risk focal symptoms, predicting ~50% pathology. • Track 3 (Screening Lane): A Technology-Led lane for younger adults with generalized symptoms, predicting an ~80% normal rate.
Size & Scope
Developed at the KFSHRC-J Neurophysiology Laboratory in Jeddah, the model encompasses 6,205 total patients. Rigorous validation utilized 5-Fold Cross-Validation for stability and included a training set of 4,654 patients and a holdout test set of 1,552 patients.
Preliminary Results
The Engineered XGBoost model achieves an 85.5% Recall (Safety metric), ensuring pathological cases are reliably flagged for physician intervention. Operationally, redirecting normal studies to the tech-led lane removes 40% of wasted physician time. This strategic “Scrubbing” effect reclaims 150 physician hours annually and increases the clinic’s capacity to evaluate complex cases by 200 patients per year. Ultimately, this yields a 16% increase in total throughput and provides 26.6% more access to pathology-positive patients while utilizing existing clinical resources.