Heart rate variability monitoring for the detection of delayed cerebral ischemia after aneurysmal subarachnoid hemorrhage

1 INTRODUCTION

Delayed cerebral ischemia (DCI) is a strong determinant of poor neurological outcome as well as high mortality and morbidity after aneurysmal subarachnoid hemorrhage (aSAH). DCI develops in up to 30% of all patients with aSAH and usually occurs 4-14 days after the ictus, where patients with a more severe aSAH are at a higher risk.^{1, 2} DCI has been suggested to have a multifactorial background, including microvascular dysfunction, coagulation activation, neuroinflammation, and cortical spreading depolarization.¹

While early detection of DCI symptoms is critical for treatment initiation, timely DCI diagnosis is challenging because of limitations in monitoring capabilities. The most powerful and accurate diagnostic tool for DCI detection is still repeated neurological examination.³ The value, however, is limited in patients with impaired consciousness or sedation.^{2, 3}

Ideally, a monitoring tool for the early detection of incipient cerebral ischemia should provide continuous bedside information, irrespective of the patients’ neurological state.³ This would provide an early warning system that could alert clinicians to imminent brain ischemia, allowing the initiation of interventions before cerebral infarction develops.

Brain injury affects the autonomic nervous system (ANS) and changes in activity and sympathovagal balance could be detected as changes in heart rate variability (HRV). Thus, HRV, as a physiological biomarker, is also interesting in patients with aSAH. HRV is defined as the variability in the length of the RR interval in the ECG and provides an indirect measurement of the ANS.^4-6 Changes in the balance between the sympathetic and the parasympathetic nervous systems detected with HRV have been reported as predictors of poor outcomes in terms of mortality and risk of complications after traumatic brain injury⁷ and after stroke,^{8, 9} including aSAH.^10-14 However, the potential of HRV changes as an indicator of incipient cerebral ischemia has not been thoroughly evaluated.

The primary aim of this study was to investigate the possible potential of HRV monitoring for the detection of incipient cerebral ischemia. Secondary aims were to analyze if there are differences in HRV parameters between patients developing DCI and those who did not, whether HRV at arrival could predict DCI and whether differences in HRV in the acute phase were related to outcome. We hypothesized that significant HRV changes precede the development of DCI, that DCI could be predicted by early registration of HRV, and that differences would be related to outcome.

2 METHODS

This observational prospective study adheres to the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines.¹⁵

3 RESULTS

During the study period, 105 patients with aSAH were admitted to the NICU. For logistic reasons and lack of staff, it was not possible to include patients during holiday seasons. Thus, only 76 patients were screened for eligibility, of which four met any exclusion criteria, four died shortly after admission and were not included because of futile prognosis, and four did not admit consent. Hence, 64 patients were included in the study. Nine patients were excluded because of incomplete HRV recordings; thus, data from 55 patients were analyzed. (Figure 1). Median age was 58 (IQR 49-66, range 27-78) years, and 73% were women. Fifteen patients developed DCI and eight of those could be assessed for clinical deterioration where time of onset of DCI could be specified. One-year mortality was 9% (5/55). Patient characteristics are presented in Table 2.

3.1 Patterns of HRV over time

For the whole patient cohort, STDRR, RMSSD, and nHF significantly decreased over time, whereas nLF and LF/HF ratio significantly increased over time (Table 3). Patients with DCI had a significantly higher LF/HF ratio over time compared to patients without DCI, P = .012 (Table 4). Other HRV variables did not significantly differ over time between patients with and those without DCI (Table 4 and Figure 2).

TABLE 3. Temporal changes in HRV parameters over time for the whole study population

	Intercept	Beta	95% CI for beta	P value
STDRR	31.8	−0.17	−0.33 to −0.005	.044
RMSSD	29.1	−0.31	−0.50 to −0.14	.001
nLF	47.0	0.29	0.12 to 0.46	.001
nHF	52.9	−0.29	−0.46 to −0.12	.001
Total power	1363	−13.5	−28.0 to 1.21	.072
LF/HF	0.93	0.07	0.04 to 0.09	<.001

Note

• A linear mixed model was used for the evaluation of HRV variables over time.
• Abbreviations: CI, confidence interval; LF/HF, ratio of low-frequency power/high-frequency power; nHF, high-frequency power in normalized units; nLF, low-frequency power in normalized units; RMSSD, root mean square of the successive differences between adjacent RR intervals; STDRR, standard deviation of all RR intervals.

TABLE 4. Temporal changes in HRV parameters over time in patients with DCI vs. without DCI

	Intercept	Beta	95% CI for Beta	P value
STDRR	28.1	0.21	−0.14 to 0.56	.232
RMSSD	26.3	0.27	−0.11 to 0.65	.166
nLF	44.4	−0.15	−0.52 to 0.21	.409
nHF	55.4	0.14	−0.22 to 0.51	.435
Total power	1125	11.4	−20.1 to 42.9	.477
LF/HF	0.42	−0.07	−0.12 to −0.01	.012

Note

• A linear mixed model was used for the evaluation of HRV variables over time. Patients without DCI are classified as the reference group. Data presented are the temporal changes in HRV in DCI patients compared to those in patients without DCI.
• Abbreviations: DCI, delayed cerebral ischemia; CI, confidence interval; DCI, delayed cerebral ischemia; nHF, high-frequency power in normalized units; nLF, low-frequency power in normalized units; LF/HF, ratio of low-frequency power/high-frequency power; RMSSD, root mean square of the successive differences between adjacent RR intervals; STDRR, standard deviation of all RR intervals.

4 DISCUSSION

In this prospective observational cohort study on DCI after aSAH, HRV was analyzed intermittently (three times/24 hours).We found no significant changes in HRV preceding the development of DCI in the eight patients who could be clinically assessed. HRV in the first 48 hours after admission could not predict the development of DCI in the acute phase. However, HRV temporal pattern, which was analyzed as LF/HF ratio, increased more during the study period in patients developing DCI, possibly indicating increased sympathetic tone. Furthermore, attenuated HRV in the first 48 hours after admission correlated significantly with worse outcome, in terms of one-year mortality.

While overall mortality after aSAH has decreased over the past decades and the proportion of survivors returning to independent life is approximately 55%, the single factor with the greatest influence on outcome is the development of DCI.² Thus, to further improve patient outcomes, a key factor would be to reduce the incidence of DCI. This would require improved monitoring that enables early detection of imminent cerebral ischemia, which could in turn intensify treatment efforts and possibly prevent cerebral infarction. Commonly used diagnostic approaches, such as TCD ultrasonography and different imaging techniques, are insufficient for the detection of incipient ischemia.^{1, 3, 22}

In the search of a more suitable monitoring technique, analysis of HRV, which reflects the balance between the sympathetic and parasympathetic nervous systems, has gained interest.

Dysfunction of the autonomic nervous system, with a preponderance of sympathetic activity, has been observed after acute ischemic stroke and has been found to be associated with poor neurological outcome and increased mortality.^{23, 24}

HRV monitoring as a clinical application is relatively new despite its long existence. Studies performed indicate that HRV may be used for detection of DCI, but substantially more research is required before HRV may be used clinically.¹³

Reduced HRV, as an indication of an increase in sympathetic activity, has been reported in patients with cerebral infarction.²⁵

Changes in HRV have been correlated with poor prognosis following different types of stroke and traumatic brain injury.^{7-9, 26, 27} Studies in patients with SAH have shown that HRV analysis early after admission could predict poor outcome at hospital discharge,^{10, 11} which is consistent with our finding that attenuated HRV in the first 48 hours correlated with 1-year mortality. Moreover, changes in HRV have also been correlated with the development of complications in patients with SAH. Su et al reported a difference in HRV patterns (ie, LF/HF slope) between patients who developed different complications within a week after SAH and those who did not.¹⁴ These findings correspond to our results showing that patients developing DCI had a more pronounced increase in LF/HF ratio over the recorded period, indicating a shift towards increased sympathetic tone.

In a retrospective case-control study Schmidt et al found that HRV could predict occurrence of complications defined as DCI or infection 24 hours before onset.²⁸ The study design is however not comparable to ours where we focused only on clinically assessable DCI and the 24 hours of HRV data preceding DCI development. With offline analysis of short-term, 5-minute periods of HRV data, we could not predict DCI 24 hours before the event.

An advantage of HRV is that this biomarker could be continuously and non-invasively recorded and is independent of the patient's neurological status. Continuous monitoring implies that huge amount of data will be generated. In our study, we only managed to analyze 1% of the collected HRV data. To be able to analyze all available data, competence in handling large data sets is vital. Another challenge using HRV is that it could be affected by numerous patient- and situation-specific factors. HRV could be influenced by age, sex, weight, and comorbidities as well as body position, temperature, sedation, ventilatory settings, volemic status, multiple drugs affecting the ANS, and nursing maneuvers.^{5, 29} It is a challenge to sort out the important signals from the noise and influence of ICU-related activities.

In recent publications, the use of computer-based analysis of larger dynamic data sets on the individual level has been suggested to improve prediction precision. Roederer et al combined automated data from electronic medical records with physiological data, which could improve DCI risk stratification.²⁹ Megjhani et al managed to predict DCI better, using machine learning for the analysis of high-frequency sampled individual physiological data during the first 4 days after the SAH, compared to gold standard predictive models.³⁰ Artificial neural networks^{31, 32} have also been used to improve prediction models based on patient data on admission. However, prediction models lack accuracy and precision when applied to individual patients. The question is whether artificial intelligence using machine learning applied on continuously collected physiological data, including HRV, combined with data from other easily available noninvasive monitoring tools could not only better predict but also foresee the development of cerebral ischemia. Computational analysis of HRV for the early detection of clinical events rather than for prediction was used in a small study, the results indicate that HRV could be useful for the early detection of complications.¹²

The strength of this study is its prospective design, where HRV, neurological examination, and physiological parameters were recorded continuously during 10 days. Furthermore, the studied patient cohort is well representative of the SAH population in terms of age, gender and incidence of DCI. However, the size of the study population is a limitation. In nine patients HRV equipment failed and these patients had to be excluded. Although 15 of the 55 patients analyzed developed clinical and/or radiological signs of DCI, only eight patients could be timely assessed for deterioration attributed to DCI so that HRV data could be coupled to the time of symptom onset. Due to the limited patient number finally included for analysis the results should be interpreted with caution.

The study design where we only analyzed HRV in 5-minute periods three times a day makes it possible that we could have missed the presence of signals or patterns in the rest of the data that could indicate cerebral ischemia. Using software that could identify patterns in continuously collected individual data may generate different results. Future studies should concentrate on continuously monitored HRV data with a focus on intra-individual changes.

A next step could be to analyze HRV data continuously and to correlate HRV patterns to sedation, pharmacological agents and external disturbances, and by doing so eventually different results could be revealed.

In this study a special HRV monitor equipment was used providing the HRV parameters via a computer-based software. For the future the best and easiest way of monitoring HRV would probably be from calculations made from the ordinary ECG signal, available in all patients. However, it will require development of software that can transfer the ECG signal for HRV calculations.

5 CONCLUSIONS

Intermittently analyzed HRV could not detect upcoming DCI in patients with aSAH. HRV in the first 48 hours after admission could not predict the development of DCI but was associated with one-year mortality. A more pronounced increase in LF/HF ratio over the study period was noted in DCI patients consistent with increased sympathetic tone. These results indicate that the HRV monitoring may have potential in the detection of DCI after aSAH using different methods of analysis. In future studies analyzing HRV data, techniques for big data management should be considered.

CONFLICTS OF INTEREST

The authors declare that they have no conflicts of interest.

AUTHOR CONTRIBUTIONS

Study design: PL, JO, HOH, SN Data acquisition: SBW, PL, JO, JL, HOH Data analysis and interpretation of results: SBW, PL, JO, LJ, HOH, LB, SN Manuscript drafting: SBW, PL, HOH Manuscript revision and editing: JO, JL, LB, SN Manuscript approval: SBW, PL, JO, SN, LB, JL, HOH