Nocturnal peripheral vasoconstriction predicts the frequency of severe acute pain episodes in children with sickle cell disease

Abstract

The basic model of SCD physiology states that vaso‐occlusion occurs when hemoglobin S‐containing red blood cells (RBC) undergo sickling before they escape the capillary into a larger vessel. We have shown that mental stress, pain and cold, and events reported by patients to trigger SCD vaso‐occlusive crisis (VOC), cause rapid and significant decrease in blood flow, reducing the likelihood that RBC could transit the microvasculature before sickling occurs. However, the critical link between decrease in microvascular blood flow and the incidence of future sickle VOC has never been established experimentally in humans. Using data from centrally adjudicated, overnight polysomnograms (PSG), previously collected in a prospective multi‐center cohort sleep study, we analyzed the beat‐to‐beat amplitudes of vasoconstriction reported by the fingertip photoplethysmogram in 212 children and adolescents with SCD and developed an algorithm that detects vasoconstriction events and quantifies the magnitude (Mvasoc), duration, and frequency of vasoconstriction that reflect the individual's inherent peripheral vasoreactivity. The propensity to vasoconstrict, quantified by median Mvasoc, predicted the incidence rate of post‐PSG severe acute vaso‐occlusive pain events (P = .006) after accounting for age and hemoglobin. Indices of sleep‐disordered breathing contributed to median Mvasoc but did not predict future pain rate. Median Mvasoc was not associated with vaso‐occlusive pain events that occurred prior to each PSG. These results show that SCD individuals with high inherent propensity to vasoconstrict have more frequent severe acute pain events. Our empirical findings are consistent with the fundamental SCD hypothesis that decreased microvascular flow promotes microvascular occlusion.

Identifying Elevated Risk for Future Pain Crises in Sickle-Cell Disease Using Photoplethysmogram Patterns Measured During Sleep: A Machine Learning Approach

“…As in our previous study (18), our analyses were conducted using the same database derived from the multi-center Sleep and Asthma Cohort (SAC) study of children and adolescents with SCD (14). Data from 252 SCD subjects were considered for analysis: 240 subjects were homozygous for sickle cell hemoglobin [HbSS], and 12 were compound heterozygous for sickle β zero thalassemia [HbSβ0].…”

“…Respiratory influences in both signals were removed using a low pass filter with cutoff frequency of 0.18 Hz. Vasoconstriction events were detected using the signal processing algorithm detailed in Chalacheva et al (18). After detecting the vasoconstriction events, the resampled (2 Hz) PPGa and RRI signals around the events were selected such that each segment was 180 s in total duration.…”

“…Examples of 20-min segments selected from the overnight beat-to-beat PPGa and RRI time-series of 2 subjects (A and B) are displayed in Figure 5. The red vertical lines in all 4 plots represent the times at which significant vasoconstrictions, as detected by the algorithm of Chalacheva et al (18), were considered to have begun. In both cases displayed, each red line is accompanied by a noticeable drop in PPGa.…”

“…In SCD subjects, such events likely translate to recurring episodes in which microvascular blood flow is reduced, thus increasing the likelihood of triggering VOC. In a recently published analysis (18) of the same database of subjects in Willen et al's. ( 14) study, we identified a significant association between the median magnitude of vasoconstriction (M vasoc ), inferred from the finger photoplethysmogram (PPG) during sleep, and the rate of subsequent VOC.…”

“…While our previous study (18) was aimed at identifying the key physiological factors linking vasoconstriction to VOC pain frequency, the present work focuses on developing algorithms to improve the ability to predict future likelihood of VOC. This is achieved by employing an approach that uses a two-level stacking machine learning (ML) model to recognize features extracted from the complex dynamic patterns of the PPG and heart rate signals that occur in conjunction with episodic nocturnal vasoconstrictions.…”