Real-world use of patient-centered artificial intelligence (AI) in oncology outside business hours: Filling the care gap at scale.

Abstract

727 Background: Cancer patient symptoms, treatment questions, and emotional distress occur at all hours. A gap in cancer supportive care is both patient self management and care team availability after business hours. Patients use AI to address this gap, but current evidence suggests non-directed patient use is suboptimal (JCO 44, 2026 (suppl16, abs 9000). Live care is delivered during business hours and achieves only 78% call completion and requires extra work (Jernigan, ONF 2020 doi:10.1188/20.ONF.E44-E54). Around the clock coverage is not feasible for most community practices. We examined conversations with a patient-centered, prompt-engineered, conversational AI to help close this care gap. Methods: Retrospective real-world analysis of de-identified user conversations with HopeAI, an oncology-focused conversational AI from Manta Cares, from 11/27/25 to 6/2/26 (6.2 months). Registered users self-identified as Patient (P) or Care Partner (CP); analyses were restricted to P and CP. Off-hours was defined as any time outside Mon-Fri 9 AM-5 PM. A 10-theme codebook grouped into 5 domains was applied to conversation text via keyword/regex coding (non-exclusive); themes counted when patient-raised or sustained across ≥2 HopeAI messages. Symptom mentions in patient text were tallied separately. Results: 3,055 P+CP conversations from 1,119 users (P=1,047; CP=72). Cancer types: breast 85% (ER/PR+ 54%, HER2+ 10%, metastatic 7%, TNBC 6%); lung 10%; other 4%. Among age-disclosing users (n=916), 45.4% were ≥60 and 19.4% were ≥70 years old. Mean 7.9 messages/conversation (median 4); 44% of users returned for ≥2 conversations. 69% of P+CP conversations occurred outside business hours (weekends 26%; late night 11 PM-5 AM 16%); 44% crossed ≥2 thematic domains (Table). Top symptoms in user text: pain (8.4%), fatigue (3.3%), anxiety (2.8%), nausea (1.4%), lymphedema/swelling (1.2%), neuropathy (1.0%), depression (0.9%), and diarrhea (0.9%). HopeAI included an explicit care-team referral in 58% of conversations; 42% advised self-management or were informational. Conclusions: Patients and care partners, including those over 60 years of age, use conversational AI outside business hours for a variety of cancer-related concerns. Almost half return for repeat use. Conversations include clinically high-stakes content: symptoms, drug effects, treatment choices and financial concerns. Our deployed AI addresses these needs in real time and appropriately refers back to the care team. Systematic monitoring of patient-AI interactions identifies modifiable supportive-care gaps. Top conversation themes (% of P+CP conversations). Theme % Treatment Decisions 75 Staging/Diagnosis 74 Appointment prep 59 Side Effects 53 Surgery 44 Chemotherapy 41 Hormone Therapy 38 Radiation 33 Financial/Insurance 20 Clinical Trials 11