Letter to the Editor: integrating Bayesian network-based personalized prediction models to enhance manual acupuncture outcomes for chronic spontaneous urticaria

Dear Editor, We read with great interest Yu et al’s systematic review showing that manual acupuncture confers multi-dimensional benefits for chronic spontaneous urticaria (CSU)[1]. The meta-analysis presented by the authors on 615 participants provides some valuable evidence on disease activity, quality of life, sleep quality, and psychological outcomes. We provide additional viewpoints from personalized prediction models that could improve clinical translation. This letter was prepared consistently with the TITAN 2025 guidelines[2] and that it was free from the assistance of any artificial intelligence tools in writing, data analysis, or figure generation. The significant treatment variations, as seen in subgroup analysis of disease duration and use of conventional drugs, suggest differential treatment responses. Based on our work in creating Bayesian network models for clinical assessments, we suggest a similar approach would help me with acupuncture for CSU patients.[3] Bayesian networks predict the individual responses of clinical variables using probabilities. A prediction model incorporating patient characteristics (demographics), diseases (baseline UAS7, disease duration), psychological comorbidities (HAMA/HAMD scores), and treatment specifications may help identify patients more likely to achieve MCID thresholds, thus improving patient selection. The authors state the mean improvements in UAS7, DLQI, and PSQI did not meet MCID thresholds. Although mean differences can put a dark cloud on a massive inter-individual variability. Our prediction research for dental diagnosis shows that while average accuracy might not appear high, the results are good for patients. Using further analysis of the responder analysis calculation checks the percentage of responders who achieved improvement at least a minute clinical important difference (MCID). If 40%–50% of people achieve MCID improvements, that’s a meaningful benefit for a large group. The progressive improvement in UAS7 in 2-week (2.47 points) and 4-week follow-ups (5.88 points) indicate dynamic improvements over time. We recommend that ongoing surveillance be scheduled on a weekly basis to identify early responders and late responders Models of machine learning could anticipate long-term results based on early trajectories. This allows for modification of treatment and early identification of non-responders. The meta-analysis shows benefits in five outcome domains, but not in relation to each other. Through our research synthesis of parameters into composite indices improved predictive accuracy. We propose a Composite Response Index with disease activity as primary, quality of life, and psychological well-being as secondary and sleep quality as tertiary with weights based on clinical importance. Bayesian networks could capture dependencies among dimensions – whether symptom reductions follow sleep improvements. Predictive risk stratification may improve safety for subcutaneous hemorrhage risk. We recommend pre-treatment risk scoring that considers coagulation parameters and anticoagulant use, which would allow protocols to be individualized, and risk estimates personalized for informed consent. All studies were conducted in Chinese populations. We see an opportunity for comparative effectiveness research, rather than a limitation. Our countries’ work shows that differences require country-specific models. We propose global partnerships to create Bayesian prediction models for CSU tailored to specific populations. To sum up, it was shown by Yu et al that acupuncture is effective for CSU. The next important step is the integration of personalized prediction models. Approaches based on Bayesian networks may identify optimal treatment candidates. While responder analysis captures individual differences that means obscure differences. Dynamic monitoring with weekly UAS7 tracking enables early. Who responded, plus Composite Response Indices that combine multi-source information. Domains provide superior predictive accuracy. This population-specific approach merges optimal pick identification with constant observation and the real adjustments toward an actual precision acupuncture therapy.