SEARCH

SEARCH BY CITATION

Keywords:

  • Epilepsy;
  • Meta-analysis;
  • Heart rate variability;
  • Systematic review;
  • SUDEP

Summary

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

Purpose:  Epilepsy is associated with near-fatal and fatal arrhythmias, and sudden unexpected death in epilepsy (SUDEP) is partly related to cardiac events. Dysfunction of the autonomous nervous system causes arrhythmias and, although previous studies have investigated the effects of epilepsy on the autonomic control of the heart, the results are still mixed regarding whether imbalance of sympathetic, vagal, or both systems is present in epilepsy, and also the importance of anticonvulsant treatment on the autonomic system. Therefore, we aimed to investigate epilepsy and its treatment impact on heart rate variability (HRV), assessed by sympathetic and parasympathetic activity expressed as low-frequency (LF) and high-frequency (HF) power spectrum, respectively.

Method:  We performed a systematic review from the first date available to July 2011 in Medline and other databases; key search terms were “epilepsy”; “anticonvulsants”; “heart rate variability”; “vagal”; and “autonomous nervous system.” Original studies that reported data and/or statistics of at least one HRV value were included, with data being extracted by two independent authors. We used a random-effects model with Hedges’s g as the measurement of effect size to perform two main meta-analyses comparing LF and HF HRV values in (1) epilepsy patients versus controls; (2) patients receiving versus not receiving treatment; and (3) well-controlled versus refractory patients. Secondary analyses assessed other time- and frequency-domain measurements (nonlinear methods were not analyzed due to lack of sufficient data sets). Quality assessment of each study was verified and also meta-analytic techniques to identify and control bias. Meta-regression for age and gender was performed.

Key Findings:  Initially, 366 references were identified. According to our eligibility criteria, 30 references (39 studies) were included in our analysis. Regarding HF, epilepsy patients presented lower values (g −0.69) than controls, with the 95% confidence interval (CI) ranging from −1.05 to −0.33. No significant differences were observed for LF (g −0.18; 95% CI −0.71 to 0.35). Patients receiving treatment presented HF values to those not receiving treatment (g −0.05; 95% CI −0.37 to 0.27), with a trend for having higher LF values (g 0.1; 95% CI −0.13 to 0.33), which was more pronounced in those receiving antiepileptic drugs (vs. vagus nerve stimulation). No differences were observed for well-controlled versus refractory patients, possibly due to the low number of studies. Regression for age and gender did not influence the results. Finally, secondary time-domain analyses also showed lower HRV and lower vagal activity in patients with epilepsy, as shown by the standard deviation of normal-to-normal interval (SDNN) and the root mean square of successive differences (RMSSD) indexes, respectively.

Significance:  We confirmed and extended the hypothesis of sympathovagal imbalance in epilepsy, as showed by lower HF, SDNN, and RMSSD values when compared to controls. In addition, there was a trend for higher LF values in patients receiving pharmacotherapy. As lower vagal (HF) and higher sympathetic (LF) tone are predictors of morbidity and mortality in cardiovascular samples, our findings highlight the importance of investigating autonomic function in patients with epilepsy in clinical practice. Assessing HRV might also be useful when planning therapeutic interventions, as some antiepileptic drugs can show hazardous effects in cardiac excitability, potentially leading to cardiac arrhythmia.

Sudden unexpected death in epilepsy (SUDEP) is the most prevalent epilepsy-related cause of death, being responsible for up to 17% of all deaths in epilepsy (Scorza et al., 2009). The etiology of SUDEP is unknown; one hypothesis is that it is related to cardiac mechanisms (Devinsky, 2004). In this framework, chronic, repeated activation of the autonomic nervous system leads to sympathovagal imbalance, ultimately triggering fatal arrhythmias during or between ictal events (Scorza et al., 2009; Jansen & Lagae, 2010). Hitherto, however, studies exploring the brain–heart connection have shown mixed results. For instance, recent comprehensive reviews (Jansen & Lagae, 2010; Sevcencu & Struijk, 2010) observed that epilepsy was not solely associated with vagal suppression (which is a risk factor for arrhythmias) but also with sympathetic activation, vagal activation, and sympathetic-vagal suppression. Along these lines, it should be acknowledged that some antiepileptic drugs (AEDs) influence cardiac excitability and conduction, potentially leading to cardiac arrhythmia, also contributing to impaired autonomic cardiac activity in these patients. Therefore, the pathophysiology of SUDEP and arrhythmias in epilepsy is still unclear.

Heart rate variability (HRV) is, indeed, an interesting proxy for autonomic activity on the heart. HRV is easily measured with standard electrocardiography (ECG) devices and appropriate software. In short, it involves registering successive beat-to-beat heart activity to assess instantaneous heart rate (ESC, 1996). Several measurements can be estimated through HRV analysis; of particular interest are those extracted from the power spectral density such as the high-frequency (HF) and low-frequency (LF) indexes, which are, respectively, related to vagal and sympathetic activity (Karemaker, 1999). Low HRV (i.e., low power in the HF band) has been associated with several conditions such as major depressive disorder (Kemp et al., 2010) and myocardial infarction (Buccelletti et al., 2009); on the contrary, physical exercise increases HRV (Sandercock et al., 2005; Nolan et al., 2008).

Notwithstanding, HRV changes have been inconsistently found in epilepsy studies (Jansen & Lagae, 2010; Sevcencu & Struijk, 2010). However, some methodologic issues can be responsible for these mixed findings, such as underpowered studies and the great number of methods for HRV measurement, which hinder definite conclusions. A suitable method for handling these issues is meta-analysis, which aggregates studies with different outcomes in a single-estimate measure of effect. Here, we aimed to perform a meta-analysis of studies that evaluated HRV in patients with epilepsy to investigate whether HRV is reduced in these patients and the impact of AEDs in HRV. We also investigated whether HRV is associated with refractoriness and vagus nerve stimulation (VNS). The study is important in addressing HRV—an index of vagal and sympathetic influence over the heart—measurements and changes in epilepsy and its treatment, which can contribute to understanding the physiologic mechanisms of sympathetic-parasympathetic dysfunction in epilepsy and of important conditions such as SUDEP.

Methods

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

This systematic review and meta-analysis was conducted according to PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) (Moher et al., 2009) and Cochrane Handbook guidelines (Higgins & Green, 2009).

Literature review

We performed the literature search in the following databases: PubMed/Medline, Embase, and Web of Science. We also browsed for additional references the Cochrane Central Register of Controlled Trials (CENTRAL) and the Connecting Research in Security to Practice/Research Portfolio Online Reporting Tool (CRISP/RePORT) National Institutes of Health (NIH) database to check for unpublished trials, and we examined reference lists in retrieved and review articles. Two authors independently searched from the first date available up to July 2011, using the following syntax: (“Epilepsy”[Mesh] OR “Anticonvulsants”[Mesh]) AND (“heart rate variability” OR “vagal” OR “autonomous nervous system”).

Selection criteria

We included references that used the following designs: (1) case–control (patients with epilepsy against matched controls); (2) quasi-controlled trials (i.e., without randomization and/or control group); (3) randomized, controlled trials; (4) long-term, follow-up studies; (5) drug-withdrawal studies. We also adopted the following criteria: (6) written in English and (7) studies that reported mean and standard deviation (SD) of at least one HRV measurement or provided data as to estimate these values (e.g., p level, confidence interval, interquartile range, and so forth). Of importance to note, we adopted the 1996 Consensus/Task Force of the European Society of Cardiology and North American Society of Pacing and Electrophysiology (ESC, 1996) when considering the standard methods of HRV measurement; particularly, LF was the power spectral density within the frequency ranges of 0.04–0.15 Hz and HF the power within the ranges of 0.15–0.4 Hz.

Exclusion criteria were designs such as series of cases, case reports, cross sectional studies without comparison groups; crossover designs (risk of effect bias); and also studies that aimed to validate and/or replicate HRV methodology (i.e., validity and reliability studies).

Finally, if one reference reported two data sets (e.g., quasi-controlled trial with a control group), then two studies were considered.

Data extraction

For each study, two authors using a structured form independently extracted data. The following variables were extracted: (1) mean, SD, and other statistics of HRV measurements for each group; (2) demographic, clinical, and treatment characteristics (e.g., number of patients, age, gender, type of treatment, type of epilepsy syndrome; refractoriness); (3) characteristics of measurement (time, power, or nonlinear analysis of HRV; period of time of ECG recording; ECG position—supine, seated, Holter, etc.; and so forth); and (4) study design.

Quality assessment

Because most studies were noncontrolled, we performed individual and comprehensive quality assessment for each study, looking for selection bias (we checked whether the matching of controls was adequate and whether the diagnostic methods for epilepsy patients were described), attrition bias (we looked for evidence of intention-to-treat analysis and selective reporting of outcomes), and measurement bias (by checking whether HRV analysis were blinded). We also used meta-analytic techniques for identifying biases across studies, such as the chi-square and I2 statistics that index between-study heterogeneity; sensitivity analysis that successively excludes individual articles to assess its particular influence on the net results; and funnel plot that compares standardized mean differences of each study against its standard error, suggesting publication bias when such distribution is asymmetrical.

Quantitative analysis

All analyses were performed using Stata 10.0 (StataCorp, College Station, TX, U.S.A.). First, a standardized effect size was determined for each study. Here, we used the Hedges’ g effect size, which is more adequate than the Cohen’s d for small sample sizes. We also used a random-effects (instead of a fixed effects) model considering that this is a more conservative method that takes into account that study heterogeneity can vary beyond chance, thus providing more generalizable results. Meta-regression was performed for age and gender using appropriate statistics (Knapp & Hartung, 2003). We regressed just one variable at a time.

Similarly to a previous HRV meta-analysis (Kemp et al., 2010), we considered several variables, for each of which an individual meta-analysis was performed:

  • 1
     SDNN—the standard deviation of NN intervals (SDNN), which reflects HRV itself.
  • 2
     RMSSD—the root mean square of successive differences, a time-domain variable that is relatively free of respiratory effects and thus is an index of vagal activity (Hill & Siebenbrock, 2009).
  • 3
     HF HRV—the main variable was HF power measured in ms2. When the study reported only HF normalized values, we calculated absolute HF power values using the following formula: HFn.u = [HF/Total Power − VLF] × 100 (ESC, 1996).
  • 4
     LF HRV—same as for HF HRV.
  • 5
     LF/HF ratio.
  • 6
     Nonlinear measures—this variable was composed by collapsed data from the following nonlinear methods: SD1 (SD of instantaneous beat-to-beat variability), SD2 (SD of continuous long-term variability), alpha (measurement of fractal property that is defined as the short-term correlation of RR interval data), and entropy (which estimates the amount of irregularity in the RR interval).

According to our main study question (to address sympathovagal dysfunction in epilepsy), changes in the HF and LF indexes were our primary outcomes. Additional outcomes considered the changes of other HRV measurements. Three comparisons were performed: epilepsy patients versus healthy subjects; patients receiving treatment versus patients without treatment (considering VNS and AEDs); and refractory versus well-controlled epilepsy patients. No analyses were performed regarding HRV changes in long-term follow-up and before/after ictus due to the low number of available studies – three (Galli, 2003; Suorsa et al., 2011) and two (Toth et al., 2010; Pradhan et al., 2011), respectively. Moreover, as to reduce the number of statistical tests, we did not perform meta-analysis of outcomes that presented three or less studies.

Results

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

Our initial search strategy yielded 366 records. Most were excluded after browsing the title and abstract, since they were reviews (74 studies); animal studies (40); trials with VNS (130) and epilepsy surgery (10); case reports, letters, and editorials (32); and other (e.g., studies assessing other neuropsychiatric disorders and studies that assessed heart rate but not its variability) (31). Therefore, 49 studies were further retrieved for full-text assessment. In this step, five studies were excluded because no HRV analysis was performed (Frysinger et al., 1993; Delamont et al., 1998; Banzett et al., 1999; Binks, 2001; Stamboulis et al., 2005); three because they did not perform standardized HRV measurements (Setty et al., 1998; Delamont et al., 1999; Zaatreh et al., 2003); 11 studies used other designs such as case reports, case series, cross sectional, crossover designs, and validity/reliability designs (Yoon et al., 1997; Toichi & Sengoku, 1998; Frei & Osorio, 2001; Pruvost et al., 2006; El-Sayed et al., 2007; DeGiorgio et al., 2008; Malarvili & Mesbah, 2008, 2009; DeGiorgio et al., 2010;Jeppesen et al., 2010; Kamal, 2010) and one study reported duplicated data (Persson et al., 2007). Therefore, 30 articles were included in qualitative synthesis. Finally, as several articles reported more than one study, 39 studies were considered for quantitative synthesis and meta-analysis (Fig. 1). The included studies had several designs: 19 were case–control; 11 were clinical trials (five with VNS and six with AEDs); four studies compared refractory versus nonrefractory patients; three were follow-up studies; and two compared HRV changes before and after ictus.

image

Figure 1.   Systematic review flow chart, showing the number of studies included at each step of the selection process and the reasons for it.

Download figure to PowerPoint

Quality assessment revealed that the risks of selection bias and attrition bias were low, as virtually all studies used age- and gender-matched controls, confirmed epilepsy diagnosis according to standardized criteria, and performed analyses in the whole sample. Nonetheless, only half of studies assembled samples with one specific type of epilepsy (e.g., temporal lobe epilepsy), whereas the other half included various types of epilepsy in their samples.

Regarding ECG analysis, only 8 (23.5%) of 34 included studies reported blinded assessment of ECG data prior to analysis. In addition, 12 studies did not report the length of ECG epoch duration, whereas the remaining studies used epochs varying from 2 min to 1 h (Table 1). In addition, 20 studies (59%) used 24 h-Holter ECG measurements. Only four of these studies reported different values for day versus night measures. Ten studies (30%) assessed HRV during daytime, from time periods ranging from 5 min to 1 h. Only four of these studies reported specific time periods (e.g., “from 8–10 a.m.” instead of during daytime). The remaining four studies assessed ECG during the sleep phases (Table 1). We have also described the length of the epoch used for analysis for each study in Table 1.

Table 1.   Summary of all data sets included for meta-analysis
Case–control
AuthorConditionHRV analysisPeriod/duration of ECGDay/nightECG analysisMean age (years)Female (%)PatientsControls
Tomson et al. (1998)JMET, P24 hNo5-min29N/A2121
Tomson et al. (1998)TLET, P24 hNo5-min37N/A2121
Yang et al. (2001)VariousT, P15 min (day)NoUnclear6303030
Ansakorpi et al. (2002)TLET, P, NL24 hNo60-min29303434
Ferri et al. (2002)VariousT, P12 h (sleep)No5-min11.5551111
Ansakorpi et al. (2004)TLET, P, NL24 hNo60-min32673972
Evrengül et al. (2005)GeneralizedT, P1 h (9 a.m.–12)No5-min2004343
Persson et al. (2006)TLET, P24 hNoUnclearN/A382121
Hattori et al. (2007)West SyndP10 min (sleep)NoUnclear<140159
Persson et al. (2007)VariousT, P24 hYes5-min33592222
Harnod et al. (2008)VariousT, P5 min (day)No288 s10.6503030
Hallioglu et al. (2008)VariousT, P5 min (8–10 a.m.)No5-min7.5501483
Assaf et al. (2008)VariousT5 minNoUnclear9.6393636
Hallioglu et al. (2008)VariousT, P5 min (8–10 a.m.)No5-min7.5357883
Harnod et al. (2009)FLET, P5 min (day)No5-min16.6442525
Surges et al. (2009a)SUDEPT, P24 hYesUnclear346077
Mativo et al. (2010)Idiopathic GTCGT, P24 hNoUnclear29.3252020
Delogu et al. (2011)Dravet SyndromeT, P24 hNoUnclear6.8552020
Yildiz et al. (2011)VariousT, P24 hNoUnclear27553732
With versus without treatment
AuthorDesign/treatHRV analysisPeriod/duration of ECGDay/nightECG analysisMean age (years)Female (%)WithWithout
Kamath et al. (1992)BAD/VNST, P45 min (day)No2-min342544
Kamath et al. (1992)BAD/VNST, P45 min (day)No2-min342544
Kennebäck et al. (1997)BAD/AEDT, P24 hNoUnclear66401010
Hennessy et al. (2001)BAD/AEDP2 h (sleep)NoUnclear34331212
Persson et al. (2003)BAD/AEDT, P24 hYes5-min43.4531515
Galli (2003)BAD/VNST, P24 hYes512 RRs47.64377
Lossius et al. (2007)BAD/AEDT, P24 hNo5-min38333939
Barone et al. (2007)BAD/VNST, P24 hNo5-min327588
Hallioglu et al. (2008)MC/AEDT, P5 min (8–10 a.m.)No5-min7.7377814
Cadeddu et al. (2010)BAD/VNST, P24 hOnly T60-min29401010
Delogu et al. (2011)MC/AEDT, P24 hNoUnclear6.4552020
Well-controlled versus refractory patients
AuthorConditionHRV analysisPeriod/duration of ECGDay/nightECG analysisMean age (years)Female (%)ControlledRefractory
  1. FLE, frontal lobe epilepsy; TLE, temporal lobe epilepsy; JME, juvenile myoclonic epilepsy; SUDEP, sudden unexpected death in epilepsy; GTCS, generalized tonic–clonic seizure; NL, nonlinear; MC, matched comparison; BAD, before-and-after design; AED, antiepileptic drug; VNS, vagus nerve stimulation; T, time analysis; P, power analysis.

  2. Period/duration of ECG shows the total amount of time of the ECG assessment and the time period in which it was assessed (when assessment was not performed by a 24 h-Holter measurement). Day/night shows whether the study reported HRV measures according to the time period. ECG analysis represents the time length of the epoch used in the analysis.

Ansakorpi et al. (2002)TLET, P, NL24 hNo60-min2930259
Mukherjee et al. (2009)VariousT, P5 min (9 a.m.–4 p.m.)No5-min20353031
Suorsa et al. (2011)TLET, P, NL24 hNo60-min32651818
Yildiz et al. (2011)VariousT, P24 hNoUnclear26621720

Main analysis

The 19 case–control studies enrolled 524 epilepsy patients and 620 healthy controls. Of them, 44% (SD 19%, median 44%, interquartile range 35–55%) were female and the mean age was 19.2 years (SD 11.9 years, median 18.3 years, interquartile range 7.5–29.3 years). Only four studies (21%) enrolled patients who were free of medications. Four studies studied samples of temporal lobe epilepsy, one of frontal lobe epilepsy, one of juvenile myoclonic epilepsy, one of Dravet syndrome, one of West syndrome, one of patients that had SUDEP (retrospective study), and the remaining 10 studies enrolled patients with more than one epilepsy syndrome (Table 1).

Meta-analyses revealed that epilepsy patients presented lower HF [g −0.69; 95% confidence interval (CI) −1.05 to −0.33], suggesting lower vagal activity in these patients (Fig. 2A). Interestingly, we did not observe significant differences in LF, a measure of sympathetic activity, in epilepsy patients versus controls (g −0.18; 95% CI −0.71 to 0.35) (Fig. 2B). Analyses showed significant between study heterogeneity for HF (χ2 = 121, p < 0.01; I2 = 86%) and LF (χ2 = 257, p < 0.01; I2 = 93.4%), although their Egger’s test was not significant (p = 0.09, p = 0.67; respectively) and also the visual inspection of the Begg’s funnel plot was not suggestive of publication bias (Fig. S1). In addition, the sensitivity analysis for HF and LF did not show that the influence of one particular study could be driving the results (Fig. S2): for LF, score changes ranged from 0.5 in the upper bound of the 95% CI after the exclusion of the study of Persson et al. (2006) to a −0.85 in the lower bound of the 95% CI after excluding the study of Harnod et al. (2009); for HF, the 95% CI upper bound changed from −0.33 to −0.25 after the exclusion of the study of Persson et al. (2006) and the 95% CI lower bound changed from −1.05 to −1.13 after the exclusion of the study of Hattori et al. (2007).

image

Figure 2.   Forest plot graphs for the comparison between epilepsy patients and healthy controls using (A) high-frequency (HF) and (B) low-frequency (LF) values. We used a random-effects model and Hedges’g was the measure of effect size (black squares); error bars represent the 95% CI. A negative effect indicates lower values in epilepsy compared to controls. No significant differences were observed for LF, although epilepsy patients presented lower HF values.

Download figure to PowerPoint

Eleven studies compared patients receiving treatment versus in the absence of treatment, encompassing 143 patients receiving treatment and 207 patients without treatment. Five studies evaluated VNS effects and six evaluated AEDs effects. Mean age of participants was 33.8 years (SD 16.7 years; median 33 years, interquartile range 18.35–38.7 years) and 42% were female (SD 14.6%, median 41%, interquartile range 31–54%) (Table 1). For the HF index, meta-analysis showed no significant differences between groups (g = −0.05; 95% CI = −0.37 to 0.27), even when stratifying for AEDs and VNS interventions (Fig. 3A). In this analysis, between-study heterogeneity was low (I2 = 37.4%) and not significant (χ2 = 14.4, p = 0.11). Sensitivity analysis and Begg’s funnel plot also did not suggest effect or publication bias (Figs S1 and S2). For the LF index, again no significant differences were observed between groups (g = 0.1; 95% CI −0.13 to 0.33), although there was a trend for lower LF scores in drug versus drug-free group (g = 0.16; 95% CI −0.11 to 0.43), suggesting suppressed sympathetic activity during pharmacotherapy (Fig. 3B). No significant between-study heterogeneity was observed (χ2 = 6.7, p = 0.67, I2 = 0), or for other tests such as sensitivity analysis, and Begg’s funnel plot were suggestive of bias (Figs S1 and S2).

image

Figure 3.   Forest plot graphs for the comparison of patients receiving versus not receiving treatment, using the (A) high-frequency (HF) and (B) low-frequency (LF) values. Results were subanalyzed according to the type of treatment: vagus nerve stimulation (VNS) and antiepileptic drugs (AEDs). A negative effect size indicates lower values in the treated group. No differences were observed between groups, although a trend for higher values in the AED group was observed in the LF analysis.

Download figure to PowerPoint

Four studies compared well-controlled versus refractory patients (Table 1). There were 78 well-controlled and 90 refractory patients. Their mean age was 26.7 years (SD 5.12, median 24 years, interquartile range 13–35.5 years) and 48% (SD 18%, median 48%, interquartile range 43–80%) were female. No significant differences between groups were observed for the HF (g = −0.5; 95% CI −1.2 to 0.21) and the LF (g = −0.19; 95% CI −1.02 to 0.64) indexes (Table 2). Sensitivity analysis and funnel plot did not suggest bias effect (Table 2).

Table 2.   Summary of the main study results
ComparisonAnalysisStatisticsLF indexHF index
  1. The values of the standardized effect size (Hedges’g) with the 95% CI are shown for the main meta-analysis that evaluated changes in low-frequency (LF) and high-frequency (HF) values in three comparisons (each of which represented in one row). Other statistics are shown, such as p-values for the Egger’s test (assessment of publication bias); I2, χ2, and p-values for heterogeneity; and β and p-values for meta-regression. Negative effect-size values represent lower values in the first (compared to the second) group.

Patients versus controlsEffect sizeg (95% CI)−0.18 (−0.71 to 0.35)−0.69 (−1.05 to −0.33)
HeterogeneityI2, χ2, p93%; 257; <0.0186%; 121; <0.01
Egger’s testp0.670.09
Meta-regressionAge –β, p−0.03, 0.3<0.01; 0.98
 Gender –β, p−0.99; 0.7−0.22; 0.88
With versus without treatmentEffect sizeg (95% CI)0.1 (−0.13 to 0.33)−0.05 (−0.37 to 0.27)
HeterogeneityI2, χ2, p0; 6.7; 0.6737%; 14.4; 0.11
Egger’s testp0.710.48
Meta-regressionAge –β, p−0.01; 0.60.02; 0.12
 Gender –β, p0.42; 0.691.93; 0.14
Well-controlled versus refractoryEffect sizeg (95% CI)−0.19 (−1.02 to 0.64)−0.5 (−1.2 to 0.21)
HeterogeneityI2, χ2, p84%;19;<0.0178%;14; <0.01
Egger’s testp0.460.9
Meta-regressionAge –β, p−0.12; 0.27−0.03; 0.79
 Gender –β, p−1.4; 0.79−0.08; 0.95

Finally, meta-regression results showed no particular influence of age and gender for all analyses (Table 2).

Other analyses

Table 3 shows the analyses performed for the other time- and frequency-domain indexes, as well as the results for heterogeneity and Egger’s test. Of interest, patients with epilepsy presented lower SDNN (Hedges’g = −0.68; 95% CI −1.08 to −0.27) and lower RMSSD (g = −0.52; 95% CI −0.87 to −0.17), suggesting, respectively, lower HRV and lower vagal activity in patients compared to controls. In addition, refractory patients showed lower SDNN values (g = −0.44; 95% CI = −0.76 to −0.13). Other analyses were not significant. In addition, between-study heterogeneity was low and not significant for most analyses, and Egger’s test was not significant for all analysis. Finally, nonlinear HRV analysis was not performed, as only two studies were available at each comparison; RMSSD analysis and LF/HR ratio were also not performed when assessing refractoriness for the same reason.

Table 3.   Summary of the secondary study results
Meta-analysisPatients versus controlsWith versus without treatmentWell-controlled versus refractory
Effect sizeHeterogeneityEgger’s testEffect sizeHeterogeneityEgger’s testEffect sizeHeterogeneityEgger’s test
g (95% CI)I2, χ2, p-valuep-valueg (95% CI)I2, χ2, p-valuep-valueg (95% CI)I2, χ2, p-valuep-value
  1. The values of the standardized effect size (Hedges’s) with the 95% CI are shown for other secondary meta-analysis that evaluated changes in time-domain indexes (SDNN, standard deviation of normal-to-normal intervals; RMSSD, root mean square of successive differences) and the LF/HF ratio. Other statistics are shown, such as p-values for the Egger’s test (assessment of publication bias); I2, χ2, and p-values for heterogeneity; and β and p-values for meta-regression. Negative effect-size values represent lower values in the first (compared to the second) group. N/A refers to analysis not performed due to the low number of studies.

SDNN−0.67 (−1.07 to −0.27)87%,110, <0.010.17−0.19 (−0.61 to 0.23)56%;14; 0.030.72−0.44 (−0.76 to −0.13)0%;2.9; 0.410.75
RMSSD−0.52 (−0.87 to −0.17)64%;16; 0.010.77−0.14 (−0.53 to 0.24)5%;3; 0.360.45N/AN/AN/A
LF/HF ratio0.01 (−0.3 to 0.5)83%; 66; <0.010.41−0.21 (−0.76 to 0.34)77%;40; <0.010.52N/AN/AN/A

Discussion

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

Main findings

The main finding of our study is that patients with epilepsy present lower HF values, a marker of impaired vagal activity that is consistently associated with increased cardiovascular risk and arrhythmias (Thayer et al., 2010). The net effect size obtained was medium to large (−0.69), confirming that this difference is clinically relevant, and obtained after reviewing 19 case–control studies that enrolled patients with different epilepsy syndromes and from infancy to adulthood. Results were robust according to sensitivity and funnel plot analysis; and secondary time-domain analysis confirmed the hypothesis of lower vagal activity, as RMSSD and SDNN values were also found to be lower in epilepsy.

Interestingly, meta-regression showed no significant effect of age and gender, conversely to other studies that enrolled healthy samples (Umetani et al., 1998; Antelmi et al., 2004). One likely explanation is that our meta-analysis was composed mainly of case–control studies that matched subjects by gender and age—in fact, most studies presented a male-to-female ratio of around 1 (Table 1), thus controlling for a gender effect. In addition, gender effects on HRV are relatively subtle and usually seen when stratifying for age—this was not possible in our study, as it would require assess to individual patient data. For age, it should be noticed that the studies enrolled young samples (mean age of 26 years), whereas age effects in HRV are significant especially for subjects older than 30–40 years (Umetani et al., 1998; Antelmi et al., 2004).

Our study showed no significant differences between patients and controls for LF, an index of sympathetic activity; conversely, a trend for lower LF ratios was identified in treated patients, especially for those using pharmacotherapy (compared with VNS). Moreover, no changes in HF values were observed during treatment. That is, not only did we find that epilepsy treatment did not positively influence vagal dysfunction (by increasing HF values), but also possible negative influence on sympathetic activity (by increasing LF values). This finding highlights the need for screening for cardiac conditions (cardiovascular diseases, arrhythmias, and so forth) in epilepsy patients before treatment, and also for further studies to investigate which drugs are associated with detrimental changes in HRV.

Finally, no HRV differences were found between well-controlled and refractory patients. Here, one possible explanation is an underpowered analysis due to the low number of included studies (i.e., a false negative finding). Along these lines, a “ceiling effect” could have occurred as both groups were using pharmacotherapy, hindering additional effects of refractoriness in HRV. Another hypothesis is lack of assay sensitivity, which can take place when comparing two similar groups without a control arm. Hence, theoretically both groups could present pathologic HRV changes without significant overt differences between them.

Limitations

For some analyses, between-study heterogeneity was high, probably due to different sample characteristics (e.g., age, epilepsy syndrome, and therapeutic intervention). To deal with this, we used a random-effects model to calculate the pooled effect size, which is used when heterogeneity is significant. Moreover, sensitivity analysis did not show that our results were driven by a particular study, as the exclusion of any of them would not change the results. Further, Begg’s funnel plot did not detect a publication bias and showed a fairly symmetrical distribution. Finally, secondary time-domain analyses showed results that corroborated the main study findings.

Another issue is that several meta-analyses were performed, increasing the risk of false-positive results. However, we limited the number of analysis to a minimum, as we did not perform meta-analysis when three or less studies were available. Moreover, main findings were robust and congruent with secondary results.

Finally, we were not able explore whether time period of HRV assessment (i.e., day vs. night measurement) influenced the results, as studies have not systematically reported this variable. As there is evidence that HRV features depend on daytime (Ronkainen et al., 2005), future studies, especially those that employed 24 h-Holter, should report HRV in different time periods.

Autonomous nervous system dysfunction in epilepsy

Several factors are responsible for the sympathovagal imbalance observed in epilepsy during and between seizures (Sevcencu & Struijk, 2010). HRV changes during the interictal period (demonstrated by several studies and corroborated by the present work) might be related to chronic structural changes in autonomic centers, which are continuously stimulated or inhibited by repetitive seizures (Wasterlain et al., 1993). Although we could not perform meta-analysis for long-term HRV changes in epilepsy due to the low number of data sets, it is worth noting that Suorsa et al. (2011) observed detrimental HRV changes in patients with refractory epilepsy after a 6-year follow-up. Interestingly, well-controlled patients with less than one seizure per month did not show HRV changes, suggesting that repetitive seizures (or drug polytherapy that was more prevalent in refractory patients) might play a role in interictal decreased HRV.

In addition, the lateralization hypothesis supports that right-sided epilepsy focus results in increased sympathetic activity (tachycardia and high LF) and left-sided in increased vagal activity (bradycardia and high HF) (Yoon et al., 1997; Mayer et al., 2004; Panchani et al., 2011). Although we could not control HRV changes regarding laterality (as studies have not described this parameter), we found evidence for lower HF values and similar LF values compared to controls. Therefore, our findings do not support the lateralization hypothesis. Nonetheless, others (Jallon, 1997; Tinuper et al., 2001; Sevcencu & Struijk, 2010) have shown that not only the side of epileptic focus regulates sympathovagal balance but also the consecutive activation of other autonomic centers that are not necessarily close to the focus.

In addition, treatment effects on epilepsy’s autonomic dysfunction should be considered. Although some AEDs have been associated with near-fatal and fatal arrhythmias (Nilsson et al., 1999; Langan et al., 2005), recently this relationship has been placed under dispute (Sander, 2003; Colugnati et al., 2010). In our study, we could not conclude that AED effects on HRV were detrimental, as only a trend for higher LF values was observed. It should be underscored, though, that other variables such as type of AED, drug dosage, and period of use have not been controlled in our analysis. As previous observations indicated that drug therapy, especially polytherapy, is an independent risk factor for SUDEP (Surges et al., 2009b; Hesdorffer et al., 2011), further studies investigating HRV changes due to therapeutic interventions are necessary to investigate treatment effects of autonomic imbalance. Along these lines, it is interesting to note the mechanistic effects of VNS treatment. This form of noninvasive neuromodulation consists in electrical stimulation of the left vagus nerve, thus modulating primarily vagal activity, which might be related to the observed absence of sympathetic (LF) effects. Interestingly, a recent animal study observed that VNS improved cardiac autonomic control and attenuated heart failure development in a canine high-rate ventricular pacing model (Zhang et al., 2009). Cardiac benefits of VNS have also been observed in seizure-induced rat models (Sahin et al., 2009) and patients with major depression (Sperling et al., 2011).

Increased cardiovascular risk in epilepsy

Lower HRV is associated with several cardiovascular conditions such as congestive heart failure (Cygankiewicz et al., 2008) and myocardial infarction (Liew, 2010). A recent prospective study found that lower HRV was associated with mortality in diabetic subjects, and also that lower LF at baseline was an independent mortality predictor after a 5-year follow-up (May & Arildsen, 2011). Therefore, our results demonstrated that patients with epilepsy present increased cardiovascular risk, including mortality, as variables such as HF, SDNN, and RMSSD were found to be lower in these subjects.

Cohort studies have showed that lower HRV in patients after acute myocardial infarction or unstable angina predicted short-term mortality, in which ventricular arrhythmias is an important cause (Lanza et al., 1999, 2006). Therefore, our findings support the hypothesis that autonomic cardiac dysfunction might play a role in SUDEP etiology. In fact, studies have been investigating possible HRV markers for SUDEP, with mixed results: One recent study (DeGiorgio et al., 2010) showed that higher SUDEP scores were associated with lower RMSSD values (a marker of vagal activity), whereas another study (Surges et al., 2009a) did not observe an association between HRV parameters in SUDEP patients as compared to controls. A putative mechanism bridging lower HF and SUDEP could be a decreased “cardiac resilience” in the periictal period: as seizures usually cause sinus tachycardia and/or tachyarrhythmia, impaired vagal activity would fail in restoring normal heart rhythm and thus favor the appearance of ventricular arrhythmias (triggering factor). Still, during the interictal period, decreased vagal tonus (and possibly an increased sympathetic tonus) could favor a “proarrhythmic” state (predisposing factor) and facilitate the development of arrhythmias.

Conclusion

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

To conclude, the results of our meta-analysis showed that epilepsy is associated with reduced HRV values, that is, decreased vagal activity. In addition, not only do AEDs not restore normal HRV values, but also a trend for higher sympathetic activation was observed. Taken together, our findings have important clinical and research implications. First, we helped to clarify some mechanisms of HRV changes in epilepsy, which foments further research in SUDEP and epilepsy-related arrhythmias. Second, we showed that patients with epilepsy present reduced HRV values. Considering that HRV is a predictor of cardiovascular morbidity and mortality in patients with heart conditions, our findings provide additional evidence that patients with epilepsy might present an increased cardiovascular risk due to autonomic imbalance—although further studies investigating HRV parameters in chronic epilepsy patients with cardiac conditions are warranted. Third, we suggest that HRV should be used as a tool to assess cardiovascular risk in epilepsy, as HRV is easily measured with available ECG devices and, thus, could be useful to identify patients with increased cardiovascular and/or SUDEP risk. Finally, our findings indicated that different antiepileptic interventions have distinct outcomes in HRV and, therefore, future studies should identify which ones are particularly hazardous for cardiovascular risk.

Disclosures

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

References

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Disclosures
  8. References
  9. Supporting Information

Figure S1. Funnel plot (publication bias assessment) of the effect size for each comparison: low frequency (LF) for epilepsy versus healthy (A) and treatment versus no treatment (B); and high-frequency (HF) for epilepsy versus healthy (C) and treatment versus no treatment (D).

Figure S2. Sensitivity analyses for comparisons between epilepsy versus healthy controls (upper side) and patients with versus without treatment (lower side) for low- and high- frequency values (LF and HF), respectively on the left and right side.

FilenameFormatSizeDescription
EPI_3361_sm_FigS1.tif1520KSupporting info item
EPI_3361_sm_FigS2.tif1520KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.