Optimizing Multi-Call Memory in AI Healthcare Communication: A Mixed-Effects Analysis of Engagement and Satisfaction Outcomes (Preprint)

Abstract

<sec> <title>BACKGROUND</title> Multi-call memory capabilities in AI-powered health communication systems show promise for enhancing patient engagement, but their impact on engagement and patient satisfaction remains unclear. </sec> <sec> <title>OBJECTIVE</title> To evaluate the relationship between multi-call memory usage and key patient experience metrics including call duration and satisfaction scores in an AI-powered healthcare communication system. </sec> <sec> <title>METHODS</title> We conducted a retrospective analysis of 4415 AI care agent calls from 4189 patients using linear mixed-effects models to account for multiple calls per patient. The primary predictor was the number of memories used per call. Outcomes included call duration (minutes), Net Promoter Score (NPS), and patient satisfaction ratings. We analyzed the full dataset and relevant subsets (completed calls only, memory-using calls only) to assess robustness of findings. </sec> <sec> <title>RESULTS</title> Memory usage significantly increased call duration, with each additional memory extending conversations by 2.47 minutes (95% CI: 2.03, 2.91; p&lt;0.001). This effect was consistent across sensitivity analyses, though attenuated in completed calls only (+0.54 minutes per memory, p=0.004). Memory usage showed no significant association with patient satisfaction across any analysis. Given that a small subset of calls used any memories and satisfaction data were available only for completed calls, the study may have been underpowered to detect an association between memory use and NPS or satisfaction ratings. </sec> <sec> <title>CONCLUSIONS</title> Multi-call memory usage may significantly enhance behavioral engagement in AI healthcare conversations but may not impact patient satisfaction measures. The findings reveal a disconnect between engagement duration and patient-reported experience, suggesting that memory optimization strategies should focus on behavioral engagement metrics while considering factors beyond usage quantity for patient satisfaction. These results provide evidence-based guidance for healthcare organizations implementing memory-enabled AI communication systems. </sec>

Ähnliche Arbeiten