Beyond Top-Box Scores: Using Natural Language Processing of Press Ganey Survey Data to Differentiate Enterprise-Wide from Site-Specific Quality Improvement Targets

Purpose: In multi-site enterprise, uniformly high top-box (5/5) Press Ganey scores can obscure site specific performance preventing targeted intervention and implementation of best practices. We show that automated natural language processing (NLP) of patient comments reveals operational patterns not seen in aggregated quantitative scores.

Methods/Materials: We analyzed 10,447 surveys from 36 imaging centers in 5 regions (Oct 2022–Aug 2025). Quantitative ratings were uniformly high (enterprise mean 4.81/5.0; regional range 4.72–4.88). Using NLP sentiment classification, we evaluated 5,789 free-text comments across 8 domains (Staff Behavior, Wait Time, Check-in, Physical Environment, Communication, Comfort, Results Delivery, Scheduling). We calculated net sentiment (% positive minus % negative mentions) by site/region and used ANOVA to test performance gaps. Findings were categorized as universal issues, localized underperformance, or best-practice leaders.

Results: Despite comparable top-box rates (78–91% scoring 5/5), NLP separated three patterns. Check-in had negative sentiment across all regions (net −2.4% to −0.7%), consistent with an enterprise-wide process problem. Wait Time negativity concentrated in a single region (net +7.9% vs +20.6% top performer; p< 0.001), indicating a targeted operational opportunity. Staff Behavior was strongly positive overall (+64.5% net; 4,076 positive mentions), with one region significantly outperforming peers (+69.4% vs +65.2% next-best), enabling internal best-practice identification. Automated processing required < 5 minutes versus an estimated 40+ hours for manual review.

Conclusions: Press Ganey top-box scores showing similar satisfaction across all enterprise regions provides no actionable differentiation for quality improvement prioritization. When top-box scores cannot differentiate priorities, NLP converts Press Ganey narratives into scalable operational intelligence supporting enterprise wide fixes, targeted interventions, and spread of internal best practices. This study provides a practical framework for radiology business analytics, demonstrating how informatics tools reveal improvement priorities hidden within apparently uniform satisfaction data.