Citations

Abstract Generative artificial intelligence (AI) is rapidly populating medical records with synthetic or partially AI-generated content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability of medical data at population scale. By analysing more than 800,000 synthetic data points across clinical text generation, vision–language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architectures. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence. Models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies that can be easily integrated into existing clinical workflows, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

Abstract Generative artificial intelligence (AI) is rapidly populating medical records with synthetic or partially AI-generated content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability of medical data at population scale. By analysing more than 800,000 synthetic data points across clinical text generation, vision–language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architectures. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence. Models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies that can be easily integrated into existing clinical workflows, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

Background: Chronic nonspecific low back pain (CNLBP) is a prevalent, disabling condition worldwide. Tuina, a traditional Chinese manual therapy, has shown potential benefits, but evidence remains limited. Objective: This study evaluated the effectiveness and safety of Tuina therapy in managing CNLBP in a U.S. population. Methods: A prospective, single-blind, randomized controlled trial was conducted at Mayo Clinic Rochester campus (February 2020-April 2024). 204 participants with CNLBP were randomized into 3 groups: Tuina (N = 68), physical therapy (N = 68), or a combination of both (N = 68). Each group received 6 treatments over 2 months. The primary outcome was pain intensity (VAS); secondary outcomes included disability (ODI), quality of life (SF-36), and adverse events. Additionally, participants completed a satisfaction survey at the end of the trial. Research Electronic Data Capture (REDCap) was used for survey data collection. Data was assessed by analysis of variance (ANOVA) for the multiple repeated measurements, Kruskal-Wallis tests, while Chi-square tests, and Bonferroni correction in SAS software. Results: 185 subjects completed the study. Significant decreases in VAS, ODI, and SF-36 Overall Physical scores were found from baseline, to post treatment and follow-up in all groups ( P < .05), while there were no significant differences in VAS, ODI, or SF-36 outcomes between the Tuina group and the other 2 groups. There were 4 adverse events observed in the trial, 1 in the Physical therapy group and 3 in the Mixed group. Only 1 transient low back pain after Tuina therapy was observed in the Mixed group but did not require treatment. The high percentage of patients who received the Tuina program was extremely helpful (n = 94/126, 77.0%). The willingness to recommend was significantly higher for Tuina compared to physical therapy (Mixed group [N = 59/62, 95.2%], Tuina group [N = 57/64, 91.9%], and physical therapy group [N = 31/59, 59.6%] ( P < .05). Conclusion: Tuina therapy showed comparable benefits to physical therapy and combined treatment of both in improving pain, function, and quality of life in patients with CNLBP. With high patient satisfaction and a favorable safety profile, Tuina appears to be a safe and acceptable non-invasive treatment option. Further studies are needed to validate these findings in the future.

Purpose of review To review role of artificial intelligence in medicine. Recent findings Artificial intelligence is continuing to revolutionize access, diagnosis, personalization of medicine, and treatment in healthcare. As a matter of fact, artificial intelligence contributed to the research that resulted in 2024 Nobel Prizes in physics, chemistry, and economics. We are only at the tip of the iceberg in utilizing the abilities of artificial intelligence in medicine to improve accuracy of diagnoses and to enhance patient outcomes. Artificial intelligence has allowed better image analysis, prediction of progression of disease, personalized treatment plans, incorporations of genomics, and improved efficiency in care and follow-up utilizing home monitoring. In ocular health diagnosis and treatment of diabetic retinopathy, macular degeneration, glaucoma, corneal infections, and ectasia are only a few examples of how the power of artificial intelligence has been harnessed. Even though there are still challenges that need more work in the areas of patient privacy, Health Insurance Portability and Accountability Act (HIPAA) compliance, reliability, and development of regulatory frameworks, artificial intelligence has revolutionized and will continue to revolutionize medicine. Summary Artificial intelligence is enhancing medical diagnosis and treatment, as well as access and prevention. Ocular imaging, visual outcome, optics, intraocular pressure, and data points will continue to see growth it the field of artificial intelligence.

Background: Tuina has gained widespread attention and utilization for the management of chronic nonspecific low back pain (CNLBP). However, evidence-based guidance for choosing accurate and appropriate outcome measures of Tuina is lacking. The objective of this study is to systematically synthesize the existing outcome measures reported in randomized controlled trials (RCTs) evaluating Tuina therapy in CNLBP. Methods: Electronic literature searches were conducted in multiple English and Chinese databases from their inception to May 2022. RCTs were included if they involved clinical outcome measures in Tuina treatment for patients with CNLBP. Outcome instruments for each study were extracted and analyzed. Evidence from included studies were assessed using the Cochrane risk-of-bias tool. Results: Of the 735 identified articles, 17 articles with 1628 participants were included. Measurement domains in these RCTs were mainly reported in terms of pain (94%) and physical activity (71%), followed by safety (41%), Chinese medicine outcome (35%), and quality of life (12%). Moreover, several limitations with existing outcomes were reported, including lack of emphasis on the evaluation of quality of life, inadequate safety monitoring, as well as insufficient and vague Chinese medicine outcome measures. All trials were deemed to be of poor methodological quality. Conclusion: Pain and physical disability were the most frequently studied outcome domains in CNLBP treated by Tuina therapy. More rigorous and high-quality trials with appropriately selected outcome measures are needed in the future.

Objective: Chronic nonspecific low back pain (CNLBP) is a serious medical and social problem resulting in functional decline and decreased work ability. Tuina, a form of manual therapy, has been sparsely used to treat patients with CNLBP. To systematically assess the efficacy and safety of Tuina for patients with CNLBP. Methods: Multiple English and Chinese literature databases were searched until September 2022 for randomized controlled trials (RCTs) of Tuina in the treatment of CNLBP. The methodological quality was assessed using the Cochrane Collaboration’s tool, and certainty of the evidence was determined with the online Grading of Recommendations, Assessment, Development and Evaluation tool. Results: Fifteen RCTs with 1390 patients were included. Tuina demonstrated a significant effect on pain (SMD: −0.82; 95% CI −1.12 to −0.53; P < .001; I 2 = 81%) and physical function (SMD: −0.91; 95% CI −1.55 to −0.27; P = .005; I 2 = 90%) when compared to control. However, Tuina resulted in no significant improvement for quality of life (QoL) (SMD: 0.58; 95% CI −0.04 to 1.21; P = .07; I 2 = 73%;) compared to control. The Grading of Recommendations, Assessment, Development and Evaluation evidence quality was determined to be low level for pain relief, physical function, and QoL measurements. Only six studies reported adverse events; none were serious. Conclusion: Tuina might be an effective and safe strategy for treating CNLBP in terms of pain and physical function, but not for QoL. The study results should be interpreted with caution for their low-level evidence. More multicenter, large-scale RCTs with a rigorous design are required to further confirm our findings.

Abstract Background: Chronic non-specific low back pain (CNLBP) is a common complaint about medical care and carries a heavy social burden. The efficacy of Tuina (TN) or physiotherapy (PT) for CNLBP has been evaluated in previous systematic reviews. However, there is no high-quality evidence to support the efficacy of Tuina. Therefore, this study aims to conduct a large-scale, multicenter, high-quality clinical trial to provide evidence for Tuina to treat CNLBP. Methods: This is a multicenter, assessor-, and analyst-blinded, randomized controlled trial with 3 parallel arms: TN, PT, and TN combined with PT (Tuina combined with physiotherapy) group. Six hundred twelve eligible CNLBP patients will be randomly assigned to the groups in a 1:1:1 ratio in 3 centers. The TN intervention includes 9-step routine techniques, while the PT intervention includes a physiotherapy treatment plan based on a patient's symptoms. The interventions for both groups will last for 30 minutes and will be carried out for 6 sessions in 8 weeks. The primary outcome will be the visual analog scale pain score. And the secondary outcomes will include the Oswestry Disability Index, spinal range of motion, 36-item short-form health survey. Safety evaluation will be recorded during the whole study. All data in this randomized controlled trial will be analyzed by SAS 9.4. Discussion: The results of this trial will provide evidence to evaluate the efficacy of Tuina's value as a treatment for CNLBP. Trial registration: Chinese Clinical Trial Registry (ChiCTR2000040288, November 27, 2020).