2577 - Impact of AI-Assisted Auto-Contouring Implementation on Workflow and Clinician Burden: A Paired Pre-Post Evaluation in a High-Volume Public Cancer Center

Purpose/Objective(s): High-volume radiation oncology centers face contouring bottlenecks that strain physician bandwidth and delay treatment planning. Although automation may improve efficiency, real-world workflow impact remains insufficiently quantified. We evaluated operational and clinician-level effects of implementing an AI-assisted auto-contouring platform in a high-volume public cancer center in a middle-income country using a prospective paired pre-post design to assess workflow efficiency, clinician burden, and adoption.

Materials/Methods: Clinicians completed structured surveys before implementation assessing contouring time and habits, perceived contouring burden (1=not burdensome, 5=extremely burdensome), perceived work-life impact, confidence in automated workflows (1=not confident, 5=very confident), and anticipated barriers. Following structured training and integration of AI-assisted organs-at-risk (OAR) auto-contouring into planning workflows, clinicians completed post-implementation surveys. Paired analyses evaluated changes in contour review time, perceived burden, routinely evaluated OARs, work-hour metrics, confidence, and workflow impact using paired t-tests or Wilcoxon signed-rank tests in R.

Results: Twelve clinicians (80% response rate) completed paired surveys: 67% were attendings and 33% residents; 75% had =10 years in practice. Median OAR review time decreased from 21–30 min to 5–10 min per case (p<0.01), corresponding to a median estimated savings of 3.5hrs/week per clinician (IQR 2-5, p<0.01). Mean contouring burden decreased from 4.2 to 2.5 (p<0.01, Cohen’s d=0.92), with high or extreme burden ratings declining from 67% to 17%. Mean confidence in reviewing auto-contours increased from 3.1 to 4.5 (p<0.01). Routinely reviewed OARs per disease site increased (mean +3.2 structures, IQR 2-5; p=0.01). 92% reported improved contouring efficiency/quality, 83% reduced cognitive workload, 75% improved contour consistency, and 67% increased time for complex planning or peer review. Qualitative responses highlighted reduced mental fatigue, increased focus on complex decision-making, and increased physicist participation in contour quality assurance. Ongoing concerns post-implementation concerns centered on workstation availability (58%), machine capacity constraints (50%), and occasional need for manual edits (42%), rather than accuracy limitations.

Conclusion: Implementation of AI-assisted auto-contouring was associated with significant reductions in contouring burden, measurable time reallocation toward higher-value clinical activities, improved contour standardization, and strengthened multidisciplinary QA integration. While infrastructure constraints persist, these findings suggest that the true impact of automation extends beyond contour accuracy, reflecting system-level workflow redesign and meaningful improvements in clinician experience in high-volume, resource-limited settings.