2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Sunday, December 13, 2026
06:00 p.m. - 08:00 p.m.
MCCNO - Hall F
Access, Representation, Population Health, Technology and Procedures, Imaging, Computational Biology, Emerging Technologies, Treatment Considerations, Combination Therapy, Drug Development, Supportive Care, Therapy Sequence, Biological Therapies, Gene Therapy, Non-Biological Therapies, Women’s Health, Adverse Events, Artificial Intelligence (AI), Biological Processes, Pathogenesis, Microbiome, Multi-Systemic Interactions, Clinical Practice (Health Services and Quality), Clinical Guidelines, Patient Safety, Quality of Care, Diseases, Genetic Disorders, Hemoglobinopathies, Sickle Cell Disease, Sickle Cell Trait, Thalassemia, Infectious Diseases, Viral, Bacterial, Research, Translational Research, Clinical Research, Clinical Trials, Epidemiology, Health Outcomes Research, Health Disparities Research, Patient-Reported Outcomes, Real-World Evidence, Registries, Study Population, Human, Adult, Elderly, Pediatric, Young Adult
PAPER

2944 Physiology based VOC risk prediction model to optimize clinical trial recruitment in sickle cell disease

Saranya Veluswamy, MD, Mingjing Chen, BS2, Michael CK Khoo, PhD3, Thomas D. Coates, MD4
1 Children's Hospital Los Angeles, Los Angeles, California, United States
2 University of Southern California, Los Angeles, California, United States
3 University of Southern California, Los Angeles, California, United States
4 Children's Hospital Los Angeles, Los Angeles, California, United States

If participants enrolled in a clinical trial experience few clinical events during the study, therapies have limited opportunity to demonstrate benefit. Development of therapies to reduce vaso-occlusive crises (VOC) in sickle cell disease (SCD) is particularly challenging because VOC frequency varies substantially within individuals over time. Thus, many participants enrolled in randomized trials using the current eligibility criterion (≥2 VOCs in the preceding year) experience few or no VOCs during the study, limiting their contribution to efficacy analyses while increasing sample size and costs. More effective identification of patients at highest short-term risk for recurrent VOCs could enrich clinical trials for informative events, improving trial efficiency and statistical power.

Autonomic nervous system-mediated peripheral vasoconstriction reduces microvascular blood flow and contributes to VOC onset in SCD. Vasoconstriction biomarkers derived from finger pulse oximetry pulse wave amplitude (PWA) signals recorded during a single overnight sleep study predict future VOC hospitalization frequency (Chalacheva, Am J Hematol. 2021). Building on these observations, we developed  a machine learning model to improve our previous prediction of future VOC risk (Ji,Front Digit Health. 2021).

Aims:

Compare the performance of a physiology-based VOC risk prediction model with historical VOC frequency alone to identify patients with SCD at high risk for future VOCs.

Methods:

Overnight sleep study recordings, clinical data, and VOC rates were analyzed for 251 pediatric patients enrolled in a multicenter SCD sleep study (Willen, Am J Hematol. 2018). The previously described vasoconstriction biomarker was combined with pulse wave amplitude and R-R interval (RRI) time-series to develop a neural network model incorporating linear and nonlinear dynamics of vasoconstriction and heart rate variability. This optimized model reduced the number of input variables from 20 to 7 while improving prediction performance. The model was evaluated in a subset of 50 patients with a baseline history of at least 1 VOC/year and compared with historical VOC frequency alone for predicting future VOC rates using thresholds of ≥1.5 and ≥ 2 VOCs/year.

Results:

Using a threshold of ≥1.5 VOCs/year, historical VOC frequency identified only 13 of 20 patients who subsequently experienced ≥1.5 VOCs/year (sensitivity 65%), with poor specificity (33%). The physiology-based model identified 17 of 20 high-risk patients (sensitivity 85%), improved specificity ( 50%), increased positive predictive value from 39% to 53%, negative predictive value from 59% to 83%. The overall accuracy increased from 46% to 64%. To illustrate the impact on clinical trial enrollment, we simulated a study designed to recruit 50 participants expected to experience  ≥1.5 VOCs/year.Use of historical VOC frequency alone would require enrollment of approximately 127 patients, whereas the physiology-based model would require only 94, representing a 26% reduction in recruitment while enriching the study population for clinically informative VOC events.

Using a threshold of ≥2 VOCs/year, historical VOC frequency identified only 9 of 17 patients who subsequently experienced ≥2 VOCs/year (sensitivity 53%) with a specificity of 64%. In contrast, the physiology-based model identified 14 of 17 high-risk patients (sensitivity 82%) with a specificity of 46%. Although overall classification accuracy was similar, the physiology-based model identified substantially more patients who subsequently experienced recurrent VOCs, reducing the number of high-risk patients missed during trial enrollment.

Conclusion:

A physiology-based VOC risk prediction model incorporating vasoconstriction biomarkers that can be derived from home-based pulse wave recordings outperformed historical VOC frequency alone for identifying patients at risk for future high VOC burden. By preferentially identifying patients most likely to experience recurrent VOCs during the study period, this approach has the potential to improve clinical trial enrichment, reduce recruitment burden, lower development costs, and increase the likelihood of demonstrating therapeutic efficacy.

Saranya Veluswamy, MD: NovoNordisk - Research Funding