AI to Revolutionize Testing and Treatment of Lyme Disease

S. are diagnosed with Lyme disease — a number expected to rise as climate change broadens the regions where ticks thrive. While the disease is highly treatable with antibiotics if detected early, the current standard test, known as two-tier serology, correctly identifies early Lyme in only about 30% of cases. This low accuracy represents a critical gap, as more than half of those not diagnosed or treated within the first few weeks of infection go on to suffer long-term issues like fatigue, arthritis, and neurocognitive impairments. Now, a new artificial intelligence (AI)-enhanced blood test unveiled at ADLM 2025 offers earlier and more reliable detection, potentially leading to significantly better patient outcomes. Another study presented at the show illustrates how generative AI tools can support adolescents in accessing helpful medical information. Together, these developments demonstrate how AI, when responsibly applied, can have a meaningful impact on health outcomes in clinical laboratory medicine.

ACES Diagnostics (New Orleans, LA, USA) has introduced a new test that utilizes AI to significantly enhance Lyme disease detection. With sensitivity and specificity both exceeding 90%, the test ensures that 9 out of 10 patients receive accurate results, thereby improving chances of timely treatment and reducing the risk of chronic complications. The test builds on immunological research conducted in rhesus macaques — animals whose immune response closely resembles that of humans when infected with Lyme-causing bacteria. Based on this, the researchers developed a single-panel test that screens for 10 different proteins (antigens), streamlining diagnosis compared to the traditional method that may require up to four separate tests. The ACES team further validated the test using blood samples from 123 Lyme-positive and 197 Lyme-negative individuals, integrating machine learning to recognize distinct immune response patterns. The final algorithm demonstrated robust performance across all stages of the disease, detecting over 90% of early-stage cases — a sharp contrast to the 27% accuracy rate seen with the standard two-tier approach. Cost-effective and compatible with widely available laboratory systems, the test is anticipated to enter the commercial market by late 2026.

In the second study presented at ADLM 2025, researchers examined how Medicine-GPT — a specialized ChatGPT-based model developed by physicians — can aid teenagers in finding reliable health information. Adolescents, known for their early adoption of technology, often turn to the internet for answers to sensitive health concerns. The study, conducted by Weill Cornell Medicine (New York, NY, USA), analyzed over 100 diagnostic and lab-related queries posted by users aged 10–19 on Reddit’s “Ask Doctors” forum, selecting the top-ranked posts based on user engagement. The team then compared the performance of Medicine-GPT with that of the general ChatGPT-4 model. While both models maintained full factual accuracy, Medicine-GPT outshone its predecessor in several categories: 66.6% for completeness, 60% for reasoning, and 46.6% for helpfulness — compared to 20%, 33.3%, and 23.3%, respectively, for ChatGPT-4. On clarity, Medicine-GPT also scored higher (80%) than ChatGPT-4 (70%). However, the study also identified a concern: these AI tools can sometimes unintentionally overwhelm users, particularly adolescents, by including rare or fatal diagnoses that may heighten anxiety.

“This highlights the need for future AI tools to not only be medically accurate, but also context-aware, user-sensitive, and aligned with how clinicians communicate,” said Austin Jin, a high school research intern at Weill Cornell Medicine. “Rather than discouraging use, providers can guide adolescents on how to use these tools responsibly, emphasizing that AI … should never replace professional medical advice or personalized evaluation.”

Related Links:
ACES Diagnostics
Weill Cornell Medicine