Abstract
Background Vaso-occlusive crises (VOCs) in sickle cell disease (SCD) are recurrent, difficult-to-predict events that significantly impact health and quality of life (QoL). While research has advanced understanding of cohort-level biometric trends preceding VOCs, individual variation across sequential episodes and longitudinal changes over time remain poorly characterized due to limited availability of high-resolution real-world data. Analyzing these patterns could provide critical insights for personalized VOC prediction.
Aims This study aimed to analyze the stability and longitudinal evolution of biometric and patient-reported outcome (PRO) measures across repeated VOC episodes in SCD patients. Using real-world data from a digital health ecosystem, we sought to characterize individual-level patterns in metrics across time and VOC periods to support the development of personalized VOC predictive modeling.
Methods Data from 70 UK patients with high VOC burden were collected through a digital health ecosystem comprising FDA-cleared wearable devices and a PRO mobile app. Wearables tracked activity (steps, active time, elevation (equivalent floors climbed: 3 metres = 1 floor), intensity, distance), sleep (duration, depth, fragmentation), and heart rate, while the app captured PROs (EQ-5D, prodromal symptoms, VOC periods).
Patients were included if they reported ≥3 VOCs spanning ≥2 years. VOCs separated by ≥5 days were defined as distinct episodes. Percentage changes in metrics were calculated for pre-VOC, actual VOC, and post-VOC phases relative to the participant's first or previous VOC episode.
Cohort-level variation was assessed using median percentage changes and interquartile ranges (IQRs). Intra-subject variation was evaluated with Spearman correlations, and longitudinal drift over time was analyzed using Linear Mixed Models (LMMs) with user ID as a random effect and time from first VOC as a covariate. Sine and cosine transformations of the date accounted for seasonality effects.
Results Seventy participants (median age: 31 years, IQR: 18; 73% female; 69% HbSS genotype) with a median follow-up of 30.8 months (IQR: 12.4) were analyzed.
Wearable activity metrics showed significant inter-episode variability, with cohort-level median changes generally near 0% (-5% to 5%) but high IQRs (e.g., steps: 145%, active time: 122%). In contrast, PROs (e.g., EQ-5D: 33%, hydration: 32%), heart rate (29%), and calorie expenditure (28%) were more consistent.
Longitudinal analysis demonstrated significant time-related declines (p<0.05) in wearable metrics during VOCs, including steps, elevation, heart rate, calorie expenditure, and sleep parameters (deep, light, and total sleep duration; time to sleep/wake; and fragmentation). PROs showed modest but significant improvements over time, including psychological scores (p<0.001), EQ-5D (p=0.004), and reductions in pain scores (p<0.001).
Pre-VOC phases showed small but significant increases in soft activity (β=0.039, p=0.009), active time (β=0.017, p=0.047), deep sleep (β=0.026, p=0.002), and psychological scores (β=0.069, p<0.001), with decreases in heart rate (β=-0.015, p=0.005) and light sleep (β=-0.011, p=0.004). Post-VOC phases showed increases in active time (β=0.017, p<0.001) and psychological scores (β=0.033, p=0.033), alongside declines in heart rate (β=-0.029, p<0.001), deep sleep (β=-0.012, p=0.016), and light sleep (β=-0.014, p=0.016). While statistically significant, most effect sizes remained modest (β<0.04).
Seasonality analyses revealed no significant impacts, and longitudinal drift exhibited minimal variability across individuals, suggesting stable directional trends.
Conclusion Wearable activity metrics varied significantly across VOC episodes, while PRO measures remained relatively stable, underscoring the need for episode-specific predictive modeling approaches. Longitudinal analyses revealed small but significant changes over time, including modest improvements in QoL-linked scores (e.g., EQ-5D). These findings highlight the complementary roles of wearable metrics and PROs: wearable variability captures episode-specific dynamics, while PRO consistency provides reliable anchors for modeling.
Causality for longitudinal changes will require further exploration, particularly in light of increasing QoL-linked scores with sequential episodes over time, and whether factors such as subgroup or hospitalisation status impact these trends.
