Predictive Value of Power Spectral Measures of RR Variability Late after Myocardial Infarction
To determine whether power spectral measures of RR variability measured late after infarction predict death. Bigger et al. (49) studied the 331 patients in the Cardiac Arrhythmia Pilot Study who survived for one year, had a 24-hour ECG recording made after the CAPS drug was washed out, and were discharged on no antiarrhythmic drug therapy. The CHRONOS algorithms were used for this inestigation. Thirty deaths occurred during an average of 788 days of follow-up. Periodograms were calculated using 24-hour continuous ECG recordings and six power spectral measures of RR variability were calculated: ULF, VLF, LF, HF, and total power and the LF/HF ratio. Because of the increase in RR variability that takes place after myocardial infarction, the optimum outpoints for measures of RR variability in the one-year recordings were substantially higher than those previously determined for 24-hour ECG recordings made about two weeks after infarction. Each power spectral measure of RR variability had a strong and significant univariate association with 2.5-year all-cause mortality, the relative risks for these variables ranged from 2.5 to 5.6. The prediction of death from the measures of RR variability made one year after myocardial infarction was not improved when measures of RR variability made within a month of myocardial infarction were added to the survival model. After adjustment for age. New York Heart Association functional class, rales in the CCU, left ventricular ejection fraction, and ventricular arrhythmias, measures of RR interval variability still had a strong and significant independent association with all-cause mortality. It is clear from this study that RR variability, measured late after infarction, predicts death independent of other important post infarction risk predictors. Some patients recover to normal values of RR variability during the year after myocardial infarction. Others show very little recovery or actually decrease. The final recovery values for RR variability are the best predictors of subsequent events, i.e., knowledge of the recovery pattern does not improve prediction significantly.
Predictive Value of Power Spectral Measures of RR Variability in the Thrombolytic Era: " ATRAMI (Autonomic Tone and Reflexes after Myocardial infarction)
Based on the strong experimental evidence that baroreflex sensitivity predicted ischemic ventricular - fibrillation and equally encouraging evidence from small clinical studies in coronary heart disease (xx21.22), a large epidemiologic study was launched. The planning group (Executive Committee) for the study included Drs. MT LaRovere, JT Bigger Jr, FI Marcus, PJ Schwartz, and A Ìîëîòà. The primary objective of ATRAMI (Autonomic Tone and Reflexes after Myocardial infarction) was to provide a definitive assessment of the predictive value of baroreflex sensitivity after myocardial infarction. Other important objectives were to determine how the predictive value of baroreflex sensitivity compared with that of SDNN (the primary variable selected to represent RR variability), left ventricular ejection fraction, and VPC and how to combine these variables to increase predictive accuracy. Also, a pre-specified analysis was to compare the predictive accuracy of various measures of RR variability as measured by the CHRONOS algorithms. Finally, ATRAMI investigated the influence of treatments that were introduced since initial studies of RR variability (e.g., angiotensin converting enzyme inhibitors and thrombolytic drugs) on the predictive value of RR variability. Between May 1991 and February 1994,1248 patients were enrolled in ATRAMI in 25 centers in Europe, United States and Japan and were followed for a minimum of 12 months. Most of the patients were treated with thrombolytic drugs during the acute phase of myocardial infarction. Patients were enrolled between 6 and 28 days after myocardial infarction and all of them had baroreflex sensitivity measured by the phenylephrine method, a 24-hour continuous ECG recording for measurement of heart rate, RR variability, and ventricular arrhythmias, and measurement of left ventricular ejection fraction. The primary endpoint for the study was cardiac death or non-fatal cardiac arrest whichever came first.
The a priori hypotheses were that a baroreflex sensitivity <3.0 msec/mmHg would predict higher cardiac mortality and that it would add predictive power to that of SDNN <70 msec. The strength of association of baroreflex sensitivity and SDNN with cardiac death/cardiac arrest with and without adjusting for left ventricular ejection fraction and VPC was estimated using Cox regression models.
The average age of the 1248 patients was 57±10 years and 8770 of them were men. The ATRAMI sample was a relatively low-risk group, as indicated by its mean left ventricular ejection fraction of 0.49±0.12. The average baroreflex sensitivity for the group was 7.2±4.6 msec/mmHg and the average SDNN was 108±35 msec; 17% of the patients had SIO VPC per hour. Patients with depressed baroreflex sensitivity or SDNN were older, included more females, had more VPC, and a lower left ventricular ejection fraction. During a mean follow-up of 21±8 months, cardiac death/cardiac arrest occurred in 49 patients (3.970). Mortality was 970 among patients with baroreflex sensitivity <3 msec/mmHg compared to 2% in patients with baroreflex sensitivity *3.0 msec/mmHg (p<0.0001) definitively establishing the predictive value of this autonomic marker. Similarly, patients with SDNN <70 msec had a 1070 mortality compared to 270 in those with SDNN *70 msec (p<0.0001), thus confirming the predictive value of SDNN. Baroreflex sensitivity and SDNN had a similar strength of association with cardiac death/cardiac arrest. The relative risk of patients with baroreflex sensitivity values <3.0 msec/mmHg or SDNN values <70 msec remained strong even after adjusting for left ventricular ejection fraction and VPC.
The combination of depressed baroreflex sensitivity and SDNN was associated with a significantly greater rate of cardiac mortality/cardiac arrest than either alone. The I -year mortality was strikingly higher for the patients who were categorized as high risk by both markers (15% vs 1%, p<0.0001). Similarly, baroreflex sensitivity or SDNN combined with left ventricular ejection fraction improved predictive accuracy for cardiac mortality/cardiac arrest. When left ventricular ejection fraction <0.35 was combined with a SDNN <70 msec the relative risk was 7.4 (9570 confidence interval, 3.1-17.8) and when it was associated with a baroreflex sensitivity < 3.0 msec/mmHg the relative risk was 11.9 (95%CI, 5.1-27.4). This large prospective study validated the predictive value of SDNN calculated with CHRONOS algorithms. SDNN added significantly to the predictive value of baroreflex sensitivity and vice versa indicating that these two indicators of autonomic nervous system activity were not redundant. Moreover, the study indicated that measures of RR variability were excellent predictors of cardiac death or non-fatal cardiac arrest in low patients treated with modern therapies as they were in sicker patients who participated in earlier studies.
The ATRAMI investigators did an analysis to determine which power spectral measure of RR variability best predicted cardiac death or non-fatal cardiac arrest. The 24-hour power spectral density was computed with the CHRONOS algorithms and power was partitioned into four frequency bands: (1) power <0.04 Hz, which does not have a clear mechanistic physiological interpretation, (3) 0.04 to <0. 15 Hz, low frequency (LF) power, which reflects modulation of sympathetic or parasympathetic tone by baroreflex activity;17 and (4) 0.15 to 0.40 Hz, high frequency (HF) power, which reflects modulation of vagal tone, primarily by breathing.18,19 In addition, total power (power < 0.40 Hz) and the ratio of low to high frequency power (LF/HF ratio), a measure that has been used as an indicator of sympathovagal balance, were calculated.20 High values for the ratio suggest predominance of sympathetic nervous activity. For ease of communication and for eventual clinical use, each variable was dichotomized at about the 15th percentile (rounded to the nearest integer ending in 5 or 0) when estimating their association with mortality. Each frequency domain measure of RR variability had a statistically significant and a strong (hazard ratio > 3) univariate association with cardiac mortality/cardiac arrest. Mean NN interval, frequent VPCs, and ventricular tachycardia also had statistically significant and strong (hazard ratio > 2.5) univariate associations with cardiac mortality/cardiac arrest.
A step-up approach was used in a multivariate Cox regression model to evaluate the independent association of the three mutually exclusive frequency domain measures of RR variability with cardiac mortality/cardiac arrest. LF power was selected first and, after LF power entered the Cox regression model, the other power spectral measures (power < 0.04 Hz and HF power) did not significantly improve the prediction of outcome. Then the other Halter measures, mean NN interval, frequent VPCs, and ventricular tachycardia, were added to the model with LF power. In this model, LF power (relative risk = 2.77) and VT (relative risk = 2.20) were strong independent predictors of cardiac mortality. Mean NN interval was also independently associated with cardiac mortality/cardiac arrest and had a relative risk of 2.06. LF power also was a better univariate and multivariate predictor of cardiac mortality/cardiac arrest than left ventricular ejection fraction.
