Kangli Chen

TRACING THE CORRELATION BETWEEN HRV AND MORTALITY. Kangli Chen¹, Phyllis K Stein,² Peter P. Domitrovich.²Department of Biology, Washington University, St. Louis, MO.¹ Department of Cardiology, Washington University School of Medicine, St. Louis, MO.²

Objectives: Low heart rate variability (HRV) is related to increased mortality both in cardiac patients and middle-aged or older adults. HRV is affected by respiratory sinus arrhythmia (RSA), the increase in heart rate during inhalation and decline during exhalation. Better RSA should mean better HRV. However, when rapid changes in heart rate are due to abnormal heart rate patterns, HRV spuriously appears higher than it actually is, resulting in a weaker relationship between low HRV and mortality. Also more abnormal HRV is associated with mortality. Hence, differentiation between normal and abnormal heart rate patterns is important.

Methods: Subjects were from the Cardiovascular Health Study, an NIH-funded population study of risk factors for cardiovascular disease and stroke in adults ≥ 65 years old. Subjects had previously-scanned Holter recordings. There were 5 experiments using annotated beat-to-beat files extracted from these recordings. All were based on symbolic dynamics, a novel non-linear HRV method. The procedure involved taking the difference between consecutive normal interbeat intervals (IBI). All increases in the IBI were categorized as one and all decreased or unchanged values as zero. Differences were assembled in groups of eight forming 8-bit binary patterns. Although there were 256 possible combinations, they were reduced to 17 categories of ApEn values, indicators of the degree of organization of the 8-bit series. In experiment 1 and 2, all steps were performed manually using EXCEL. Ten-minute segments were examined from 2 subjects, one with and one without abnormal HRV, and distributions visualized graphically (1) and categorized by ApEn (2). In experiment 3, N=289 beat files were examined using an automated program written in C. Each hourly segment had previously been categorized as normal, borderline or abnormal based on visual criteria and non-linear HRV measures. An hourly abnormality score was developed based on the regression coefficients of the binary ApEn groups that best separated normal and abnormal hours. This model was developed on the first half of the dataset and validated on the second half. In experiment 4, 20-min segments were analyzed a priori to classify them as normal, borderline and abnormal in 15 participants and a new regression model developed. In experiment 5, a regression model for mortality during up to 14-year follow up was developed among the 1270 participants with Holter recordings.

Results: In both the 20-min and hourly analyses, positive predictive accuracy for segments classified as normal was ≥ 90%. But, predictive accuracy for abnormal segments was 50 - 70%. Also, the hourly experiment had better results than the 20-min experiment. We discovered that there was more than one kind of abnormal rhythm, i.e., the expected beat-to-beat randomness, but also an abnormal rhythm that could be described as a short-cycle irregularity. The program only picked up the beat-to-beat randomness. In the last experiment a regression model based on only 2 binary ApEn groups was a strong predictor of mortality, indeed stronger than the previously used standard non-linear HRV measure, the short-term fractal scaling exponent.

Conclusions: Binary symbolic dynamics is a powerful tool for HRV analysis and may provide a powerful risk marker for mortality in the elderly.