Prognostic accuracy of heart rate variability analysis in neonatal encephalopathy – a systematic review

Vânia Oliveiraa, Rui Martinsb, Natasha Liowc, Justinas Teiserskasc, Wilhelm von Rosenbergd, Trisha Adjeid, Vijayakumar Shivamurthappac, Peter J Lallya, Danilo Mandicd, Sudhin Thayyila

a Centre for Perinatal Neuroscience, Imperial College London, London, UKb Royal London Hospital, London, UKc Neonatal Unit, Imperial Healthcare NHS Trust, London, UKd Electrical and Electronic Engineering, Imperial College London, London, UK

Correspondence Vânia Oliveira MSc, BScRoom 514, Hammersmith HouseCentre for Perinatal NeuroscienceImperial College LondonDu Cane RoadLondon W10HSE-Mail: v.oliveira@imperial.ac.ukTelephone: 020 3313 2473

1

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647

Key words: heart rate variability, neonatal encephalopathy, hypoxic ischemic encephalopathy

ABSTRACT

Background: Heart rate variability analysis offers real-time quantification of autonomic disturbance after perinatal asphyxia and hence may aid in disease stratification and prognostication after neonatal encephalopathy (NE).

Objective: To systematically review the existing literature on the accuracy of early heart rate variability (HRV) to predict brain injury and adverse neurodevelopmental outcomes after NE.

Design/Methods: We systematically searched the literature published between May 1947 and May 2018. We included all prospective and retrospective studies reporting HRV indices within the first 7 days of life in babies with NE and its association with adverse outcomes (defined as evidence of brain injury on magnetic resonance imaging and/or abnormal neurodevelopment at one year of age or more). We extracted raw data wherever possible to calculate the prognostic indices with confidence intervals.

Results: We retrieved 379 citations, of which five met the criteria. One further study was excluded due to analyzing an already included cohort. These four studies provided fata for 205 babies, of whom 80 (39%) had an adverse outcome. Prognostic accuracy was reported for 12 different HRV metrics and area under the curve (AUC) varied between 0.79 and 0.94. The best performing metric reported in included studies was relative power of high frequency band, with an AUC of 0.94.

ConclusionsHRV metrics are a promising bedside tool for early prediction of brain injury and neurodevelopmental outcome in babies with NE. Due to small number of available studies, their heterogeneity and methodological limitations, further research is needed to refine this tool so that it can be used in clinical practice.

2

484950515253545556575859606162636465666768697071727374757677787980818283

INTRODUCTION

Neonatal encephalopathy (NE) results primarily from a presumed lack of oxygen and blood flow to the fetal brain around the time of birth and accounts for one million deaths worldwide, every year [1] . Although rescue hypothermic neuroprotection improves survival and neurodevelopment outcome after NE, early identification of babies at risk of brain injury and adverse outcomes is challenging due to the evolving clinical picture. An abnormal voltage or pattern of amplitude integrated electroencephalography (aEEG) has been used as a bedside tool for quantifying brain injury and as an inclusion criterion for some of the cooling trials [2,3]. More recent data suggest that aEEG does not offer an added advantage over clinical examination [4] and its prognostic accuracy in cooled infants is poor [5].

As autonomic disturbance is the hallmark of perinatal hypoxic injury, heart rate variability (HRV) analysis may offer a promising solution. HRV analysis quantifies variations in heartbeat intervals, offering a non-invasive measure of autonomic function, as it reflects the actions of the sympathetic and parasympathetic nervous systems. The simplest measure of HRV is determined by the standard deviation of the difference between consecutive RR intervals (SDNN) but many other measures of HRV continue to be developed [6,7]. Currently, HRV analyses can include not only linear measures in time and frequency domains but also several non-linear metrics derived from complexity analysis [8].

Loss of HRV occurs in traumatic brain injury and during seizures [9,10], and is associated with brain injury and adverse neurodevelopment in very low birth weight babies [11,12]. In babies with birth asphyxia, the early identification of brain injury risk via HRV analysis could benefit patients by facilitating early disease stratification and implementation of the most appropriate treatment. Nonetheless, it is unclear what the prognostic value of these electrocardiographic (ECG) changes are in identifying babies at risk of brain injury, and whether these are useful in accurately predicting neurodevelopmental outcomes.

Our aim was to systematically review the available evidence about the prognostic accuracy of HRV analysis in predicting brain injury and adverse neurodevelopmental outcomes after NE.

METHODS

Search strategyTwo reviewers (VO/RM) independently searched the literature according to the strategy outlined in a predefined protocol, using Ovid SP (selecting Embase Classic+Embase, Global Health, HMIC Health Management Information Consortium, Ovid MEDLINE(R) ALL, Maternity & Infant Care Database), Web of Science, Scopus and the Cochrane Register of Diagnostic Test Accuracy Studies, between May 1947 and May 2018. The databases were searched using the following combination of keywords: [heart rate variability] AND [birth asphyxia OR neonatal brain injury OR

3

84858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130

hypoxic ischemic encephalopathy OR hypoxic ischaemic encephalopathy OR neonatal encephalopathy]. No language restrictions were applied. An example of a full search strategy is provided in Appendix 1. We also searched the reference lists of the retrieved articles. Our report follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [13].

Study selectionWe included all prospective and retrospective cohort or case-control studies comparing HRV of babies with NE, with either brain injury on neonatal magnetic resonance imaging (MRI) or an adverse neurodevelopmental outcome. All studies on term and late preterm babies (≥ 35 weeks) with NE, who were less than a week old at the start of ECG measurements, were eligible.

The index test was HRV analysis and included any metrics (time and frequency domain or complexity analysis), regardless of the method of ECG data collection (single or multiple leads; real time or archived data) or post-processing software.

The reference test was defined as either a structured neurodevelopmental examination at 12 months of age or later, or, as a surrogate biomarker, abnormalities on brain MRI performed within six weeks of birth, that are known to be associated with adverse neurological outcomes [14].

Data extraction and analysisTwo reviewers (VO/RM) independently screened the titles and abstracts of the retrieved articles, obtained full texts and online supplements of all relevant studies, extracted the data from each of the included studies, according to a pre-defined outcomes list. Any disagreement was addressed by consensus with a third author.

Wherever possible, we extracted the raw data to create 2x2 tables for calculation of diagnostic indices and summary receiver operating characteristic (ROC) curves or accuracy metrics that had not been reported. We contacted the corresponding authors and requested raw data, if these were not provided in the published paper. If these data were then still not available, we used the prognostic indices and thresholds which had been reported.

Assessment of risk of bias and statistical analysisWe described the characteristics of included studies and confounding factors. Two authors (NL/JT) independently appraised the evidence according to the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy [15] and summarized the risk of bias. Consensus was sought from a third author if disagreement occurred.

RESULTS

Included studiesThe database search resulted in 379 citations, 315 after removing duplicates. Eighty-two were screened and 12 full-text articles were assessed for eligibility (Figure 1).

4

131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177

Two studies did not have MRI or neurodevelopmental outcome measurements and hence were excluded. Five studies were excluded due to not examining the prognostic value of HRV and one due to including data overlapping with a cohort already included in this review. Therefore, we could extract either raw data or some prognostic indices from only four studies.

In total, these four studies included 228 patients, of which 205 had NE and ECG recordings/HRV calculated: 46(22.4%) with mild NE on admission, 88(42.9%) with moderate NE, 71(34.6%) with severe NE. In total 39% (80/205) of the babies had an adverse outcome, including a mortality of 12% (25/205).

One study reported data from babies admitted to the NICU before cooling therapy was standard care and three included babies who received cooling treatment. All four studies were observational and retrospective.

Characteristics of index and reference testsOf the four included studies, two used a structured neurodevelopmental examination (one with Bayley Scales of Infant Development III (BSID-III), one with Griffiths) as a gold standard indicator of adverse neurodevelopmental outcome; two used neonatal MRI within two weeks after birth (Table 1).

The ECG trace was obtained from bedside multiparametric monitors (n=100), combined electroencephalography monitor (NicoletOneTM, Vyasis Healthcare, San Diego CA) (n=61) and from HeRO monitor (n=67). Of the four studies included, ECG recordings were performed from ‘as soon as possible after birth’ up to day seven, although the earliest ECG data published refer to 24h of postnatal life. Three studies used in-house post-processing analysis pipelines, while one study used HeRO monitor scores (MPSC™, Charlottesville, USA).

All studies reported HRV data at 24h from birth (or day one). Three studies also reported data at different time-points: the first study at 48h, the second at 96h and a third at 7 days postnatal age. Consequently, pooling of data could only be attempted for the HRV metrics provided at 24h of postnatal age (although different metrics were reported at this time-point).

The relative power of low frequencies of the ECG spectrogram (rLF) was reported in two studies (both using 0.05-0.25Hz as a threshold for rLF). Otherwise, all other HRV metrics were reported only once across all studies. The selection method of reported HRV metrics was not pre-specified.

Prognostic accuracyTwelve different HRV metrics were reported by the included studies. The AUC for prognosticating an adverse outcome from these metrics ranged from 0.79 to 0.94. As presented in Table 2, the best performing feature reported was the relative power of the high frequency band (rHF) at 24-27h with an AUC of 0.94, sensitivity of 0.84 and specificity of 0.87[16]. This was based on a sample size of only 20 babies. High

5

178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223

frequency was defined in this study as ranging between 0.30Hz and 0.80Hz, although this threshold was different from what was reported in other studies.

In the work of Goulding and colleagues [17], the power of the low frequency band (LF) predicted death or adverse neurodevelopment with an AUC of 0.79 [95% confidence interval (CI) 0.57-0.94], although the LF differences between severity groups was no longer statistically significant when adjusted for phenobarbitone use.

The study by Vergales and colleagues [18] reported a AUC of 0.83 [CI not reported] for HRV score (proprietary), on day one, to predict death or abnormal MRI, although this was based on a device originally designed to predict sepsis onset in preterm infants [19]. The authors also highlight the difference of HRV between babies with moderate or severe encephalopathy and those with mild or no encephalopathy at 24h (15.8 vs 25.4 respectively, p=0.01 for heart rate characteristics (HRC) index and 2.48 vs 1.42, p<0.05 for HRV score (both proprietary metrics)). Equally, survivors had significantly higher variability than babies who died (HRV score 24.3 vs 10, p<0.05 and HRC index 2.85 vs 3.64, p<0.001).

In another study from Metzler and colleagues[20], the root mean square at short time-scale (RMSS) predicted basal ganglia injury on MRI with an AUC of 0.79 [CI not reported]. In this study, dichotomisation of the outcome excluded watershed/mild basal ganglia injury. RMSS and root mean square at long time-scale (RMSL) were significantly associated with MRI scores of brain injury, after adjusting for confounders.

Three studies also reported the prognostic accuracy of HRV metrics beyond 24h: AUC of 0.80 [0.62-1.00] for the power of very low frequency (VLF) at 48h [17]; 0.82 for LF at 93-96h; 0.81 for HF at 102-105h [16] and 0.75 at day 4-7 of life [18] for HRC index [CI not reported). Where more than one time-point was reported, 24h was the time-point of higher accuracy but no studies have reported data for less than 24h of postnatal age.

As several prognostic indices and confidence intervals were not reported in many studies, we extracted the raw data, where possible, to calculate those measures (Table 3 and Figure 2). Although meta-analysis was not appropriate due to heterogeneity of HRV metrics, to explore what could become the potential performance of HRV to predict adverse outcomes, we calculated the cumulative true/false positives/negatives and the associated prognostic values. There was a low risk of bias across the studies (Figure 3). Whereas all studies reported blinding to clinical information, only one study reported that index test assessors were blinded to the results of the comparator test and vice-versa. Missing data, artefacts in the ECG and suboptimal data collection were reported by all authors, since these were often obtained from archived data. Within the four selected studies, two [16,20] were from the same research group. It is unclear if there was any data overlap or if any participant data may have been included in both reports.

6

224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269

Additionally, given the good prognostic performance in all studies, publication bias cannot be excluded. The studies assessed the predictive value of different features of HRV, using different time points and gold standards for the assessment of outcomes. There was significant under-reporting, and pre-selection of specific HRV parameters was unclear across all studies.

Excluded studiesEight studies were excluded, of which two [21,22] measured the severity of brain injury with methods that did not meet the criteria for this review (Table 4). These studies reported significant differences in mean HRV between severity groups, as well as between survivors and non-survivors. In another study[23], several parameters were higher in the unfavourable than in the favourable outcome group at 24h, but not at 48h and 96h. The 24h result is contradictory to the findings of all other studies, but no estimates or hypothesis testing were reported and the authors’ primary aim was not to describe prognostic accuracy. The authors have informed us that they are currently examining a larger dataset with that aim. One excluded study [24], whose cohort overlapped with another study included in this review, reported an AUC of 0.66 which referred to the average of 60 ECG features examined. As individual accuracies were not reported ait is possible that this value is not representative of the best accuracy of individual parameters.

DISCUSSION

We have systematically reviewed the literature on the accuracy of HRV analysis to predict brain injury on MRI or adverse neurodevelopmental outcome in babies with NE. Although we found a very small number of studies and limiting designs, results are consistent throughout the studies to support that HRV metrics, acquired within days after birth, hold a good predictive value of later neurodevelopmental outcome. At its best performance, calculated HRV metrics had a sensitivity of 0.80, a specificity 0.90 and an AUC of 0.85 at 24h of age [16], which represents a sensitivity comparable to aEEG [5], but with a higher specificity.

Unfortunately, none of the studies analyzed HRV within six hours of birth, which is the time-point by which critical treatment decisions need to be made with regards to therapeutic hypothermia. This may be due to the logistic difficulties in obtaining consent for research soon after birth. Nevertheless, future prospective studies should attempt to recruit babies immediately after birth, so that the value of HRV to support early clinical decision can be examined.

It is important to mention that these studies used a dichotomized approach as has been traditional in NE outcomes research. This means babies with mild NE are grouped with those with normal neurodevelopmental outcome to define ‘favorable outcome’ and babies with moderate and severe NE are grouped to define ‘adverse outcome’. Whereas this approach has been common practice in many studies to date (because babies with mild NE were thought to be ‘intact survivors’), emerging evidence suggests that babies who suffer mild NE may not have an intact

7

270271272273274

275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315

neurodevelopment but instead present with more subtle delays later into childhood [25,26]. This systematic review, along with our recent survey highlighting clinician uncertainty about treatment choices in babies with mild encephalopathy [27], raises the importance of developing an accurate early bedside tool to identify babies at risk of adverse neurodevelopment. In fact, it underlines the point that severity stratification tools in NE must shift from the traditional dichotomous approach, to one that more accurately represents the full spectrum of NE.

As neurodevelopmental outcomes and severity of brain injury were dichotomized in all studies, current data does not inform us whether HRV analysis is sensitive to predict milder forms of adverse outcome. Equally, the fact that 35.7% of included infants had severe HIE may have increased the sensitivity of HRV to predict adverse outcome. It is therefore important to ascertain if the prognostic value of HRV is not driven by the inclusion of severe NE but also able to distinguish between mild, moderate and severe NE.

Once refined, HRV may become not only an accurate disease stratification tool but may also allow monitoring of the therapeutic effect of established and/or experimental therapies. Despite the number of ongoing trials exploring neuroprotective therapies, understanding which babies may benefit from each of those therapies is a question we will still struggle to answer if we lack accurate bedside tools for early disease stratification.

LimitationsThe studies included in our review were affected by heterogeneity of both the index and the reference test and selective reporting of outcome measures, which impeded a quantitative meta-analysis. Nonetheless, we tried to collect further information from the authors and extracted the data that allowed generating further prognostic metrics. The retrospective nature of most studies, and the fact that some were nested cohorts with small sample sizes, implied that adjusting for important confounders was not possible. Other reports have specifically addressed the effect of temperature on HRV[23,28], demonstrating contradictory results. In view of the evidence that HRV may be predictive of neurodevelopmental outcome despite confounding effects, future studies of HRV in babies with NE should examine important factors such as temperature, seizures, sedation and analgesia in order to optimize the accuracy and clinical translation of this tool.

CONCLUSION

Despite the small number of available studies and their heterogeneity, the studies included in our review suggest that HRV monitoring offers good prognostic accuracy to predict brain injury or adverse neurodevelopment in NE. Given its association with neurodevelopmental outcome, HRV analysis presents as a promising bedside tool for early severity stratification in these babies. However, further refinement of the technology and standardization of risk thresholds are required before clinical use. Future studies which examine this tool must adopt more robust designs that address

8

316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362

multiple confounders and more comprehensively describe prognostic accuracy over time, from birth and throughout the progression of encephalopathy.

AcknowledgementsWe would like to thank Dr Robert Goulding for providing raw data of their published study.

Funding VO is funded by the National Institute for Health Research (Doctoral Fellowship), PL is funded by the National Institute for Health Research (Clinical Trials Fellowship).

9

363364365366367368369370371372

REFERENCES

1. Lawn, J, Cousens, S, Zupan J. 4 million neonatal deaths: When? Where? Why? Lancet 2005; 365(891–900).

2. Azzopardi D V, Strohm B, Edwards a D, Dyet L, Halliday HL, Juszczak E, et al. Moderate hypothermia to treat perinatal asphyxial encephalopathy. N Engl J Med 2009; 361(14):1349–58.

3. Gluckman PD, Wyatt JS, Azzopardi D, Ballard R, Edwards AD, Ferriero DM. Selective head cooling with mild systemic hypothermia after neonatal encephalopathy: multicentre randomised trial. Lancet 2005; 365.

4. Natarajan G, Laptook A, Shankaran S. Therapeutic Hypothermia: How Can We Optimize This Therapy to Further Improve Outcomes? Clin Perinatol 2018 Jun 1; 45(2):241–55.

5. Chandrasekaran M, Chaban B, Montaldo P, Thayyil S. Predictive value of amplitude-integrated EEG (aEEG) after rescue hypothermic neuroprotection for hypoxic ischemic encephalopathy: a meta-analysis. J Perinatol 2017; 37(6):684–9.

6. Schwartz PJ, Garson A, Paul T, Stramba-Badiale M, Vetter VL, Villain E, et al. Guidelines for the interpretation of the neonatal electrocardiogram: A Task Force of the European Society of Cardiology. Eur Heart J 2002; 23(17):1329–44.

7. von Rosenberg W, Chanwimalueang T, Adjei T, Jaffer U, Goverdovsky V, Mandic DP. Resolving Ambiguities in the LF/HF Ratio: LF-HF Scatter Plots for the Categorization of Mental and Physical Stress from HRV. Front Physiol 2017; 8:360.

8. Lin P-F, Lo M-T, Tsao J, Chang Y-C, Lin C, Ho Y-L. Correlations between the Signal Complexity of Cerebral and Cardiac Electrical Activity: A Multiscale Entropy Analysis. Montoya ARH, editor. PLoS One 2014; 9(2):e87798.

9. Lowensohn, RI, Weiss, M, Hon E. Heart-rate variability in brain- damaged adults. Lancet. 1977; 1:626–8.

10. Doyle OM, Temko a, Marnane W, Lightbody G, Boylan GB. Heart rate based automatic seizure detection in the newborn. Med Eng Phys 2010; 32(8):829–39.

11. Hanna BD, Nelson MN, White-Traut RC, Silvestri JM, Vasan U, Rey PM, et al. Heart rate variability in preterm brain-injured and very-low-birth-weight infants. Biol Neonate 2000; 77(3):147–55.

12. Addison K, Griffin M, Moorman J, Lake D, O’Shea T. Heart rate characteristics and neurodevelopmental outcome in very low birth weight infants. J Perinatol 2009; 29:750–6.

13. Moher D, Liberati A, Tetzlaff J, Altman DG, Altman D, Antes G, et al. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. PLoS Medicine 2009; 6(7): e1000097.

14. Thayyil S, Chandrasekaran M, Taylor A, Bainbridge A, Cady E, Chong K. Cerebral magnetic resonance biomarkers in neonatal encephalopathy: a meta-analysis. Pediatrics 2010; 125.

15. Cochrane Screening and Diagnostic Tests. Handbook for DTA Reviews . [Internet] [cited 2018 May 1]. Available from:

10

373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419

http://methods.cochrane.org/sdt/handbook-dta-reviews16. Massaro, A N; Govindan, RB; Al-Shargabi, T; Andescavage, NN; Metzler, M;

Chang, T; Glass, P; Plesis A. Heart rate variability in encephalopathic newborns during and after therapeutic hypothermia. J Perinatol 2014; 34(11):836–41.

17. Goulding RM, Stevenson NJ, Murray DM, Livingstone V, Filan PM, Boylan GB. Heart rate variability in hypoxic ischemic encephalopathy: correlation with EEG grade and 2-y neurodevelopmental outcome. Pediatirc Res 2015; 77(5):681–7.

18. Vergales, Brooke D; Zanelli, Santina; Matsumoto, Julie; Goodkin, Howard; Lake, Douglas; Moorman, J.Randall; Fairchild K. Depressed Heart Rate Variability is Associated with Abnormal EEG, MRI, and Death in Neonates with Hypoxic Ischemic Encephalopathy. Am J Perinatol 2014; 31:855–62.

19. Fairchild K, Aschner. HeRO monitoring to reduce mortality in NICU patients. Res Reports Neonatol 2012; 65.

20. Metzler M, Govindan R, Al-Shargabi T, Vezina G, Andescavage N, Wang Y, et al. Pattern of brain injury and depressed heart rate variability in newborns with hypoxic ischemic encephalopathy. Pediatr Res 2017; 82(3):438–43.

21. Goulding RM, Stevenson NJ, Murray DM, Livingstone V, Filan PM, Boylan GB. Heart rate variability in hypoxic ischemic encephalopathy during therapeutic hypothermia. Pediatr Res 2017;81(4):609–15.

22. Aliefendioglu, Didem; Dogru, Tolga; Albayrak, Meryem; DibekMisirhoglu, Emine; Sanh C. Heart rate variability in neonates with hypoxic ischemic encephalopathy. Indian J Pediatr 2012; 79(11):1468–72.

23. Vesoulis Z, Rao R, Trivedi S, Mathur A. The effect of therapeutic hypothermia on heart rate variability. J Perinatol 2017;37:679–83.

24. Temko A, Doyle O, Murray D, Lightbody G, Boylan G, Marnane W. Multimodal predictor of neurodevelopmental outcome in newborns with hypoxic-ischaemic encephalopathy. Comput Biol Med 2015;63:169–77.

25. Murray DM, O ’connor CM, Ryan CA, Korotchikova I, Boylan GB. Early EEG Grade and Outcome at 5 Years After Mild Neonatal Hypoxic Ischemic Encephalopathy. Pediatrics 2016; 138(4).

26. Jacobs SE, Morley CJ, Inder TE, Stewart MJ, Smith KR, McNamara PJ, et al. Whole-Body Hypothermia for Term and Near-Term Newborns With Hypoxic-Ischemic Encephalopathy. Arch Pediatr Adolesc Med 2011 Aug 1; 165(8):692.

27. Oliveira V, Singhvi DP, Montaldo P, Lally PJ, Mendoza J, Manerkar S, et al. Therapeutic hypothermia in mild neonatal encephalopathy: a national survey of practice in the UK. Arch Dis Child - Fetal Neonatal Ed 2017; 103(4):F388-F390.

28. Massaro AN, Campbell HE, Metzler M, Al-Shargabi T, Wang Y, du Plessis A, et al. Effect of Temperature on Heart Rate Variability in Neonatal ICU Patients With Hypoxic-Ischemic Encephalopathy. Pediatr Crit Care Med 2017; 18(4):349–54.

29. Massaro AN, Campbell HE, Metzler M, Al-Shargabi T, Wang Y, du Plesis A, et al. Effect of Temperature on Heart Rate Variability in Neonatal ICU Patients with

Hypoxic-Ischemic Encephalopathy. Pediatr Crit Care Med 2017; 18(4):349–54.30. Lasky RE, Parikh NA, Williams AL, Padhye NS, Shankaran S. Changes in the

PQRST intervals and heart rate variability associated with rewarming in two

11

420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466

newborns undergoing hypothermia therapy. Neonatology 2009; 96(2):93–5. 31. Javorka K, Lehotska Z, Kozar M, Uhrikova Z, Kolarovszki B, Javorka M, et al.

Heart Rate Variability in Newborns. Physiol Res 2017; 66:203–14. 32. Schneebaum Sender N, Govindan RB, Sulemanji M, Al-Shargabi T, Lenin RB,

Eksioglu YZ, et al. Effects of regional brain injury on the newborn autonomic nervous system. Early Hum Dev 2014; 90(12):893–6.

33. Matic V, Cherian PJ, Widjaja D, Jansen K, Naulaers G, Huffel S Van, et al. Heart Rate Variability in Newborns with Hypoxic Brain Injury , [In] Oxygen Trasnport to Tissue XXXV. advances in Experiemental Medicine and Biology 2013; 789:43–8.Springer, New York.

12

467468469470471472473474475476477478

TABLES, FIGURES AND APPENDICES

Appendix 1: Example of a full search strategy

Table 1: Characteristics of the included studies Table 2: Prognostic accuracy of the best performing HRV metrics originally reported by the included studies.Table 3: Summary of the calculated prognostic metrics. Table 4: Excluded studies

Figure 1: Flow chart of the literature reviewFigure 2: Schematic summary of the accuracy of the calculated prognostic metricsFigure 3: Assessment of the risk of bias in included studies

13

479480481482483484485486487488489490491492

APPENDIX 1: Example of full search strategy

Ovid SP

Resources selected:http://ovidsp.tx.ovid.com/sp-3.31.1b/ovidweb.cgi?&S=LOEPFPJDBEDDFJFNNCEKJBIBEJMNAA00&Database+Field+Guide=4

Embase Classic+Embase 1947 to 2018 May 01, Global Health 1973 to 2018 Week 18, HMIC Health Management Information Consortium 1979 to May 2018, Ovid MEDLINE(R) ALL 1946 to May 01, 2018, Maternity & Infant Care Database (MIDIRS) 1971 to May 2018

Searches:1# heart rate variability.mp. [mp=ti, ab, hw, tn, ot, dm, mf, dv, kw, fx, dq, bt, id, cc, nm, kf, px, rx, an, ui, sy]

2# hypoxic ischaemic encephalopathy.mp. [mp=ti, ab, hw, tn, ot, dm, mf, dv, kw, fx, dq, bt, id, cc, nm, kf, px, rx, an, ui, sy]

3# hypoxic ischemic encephalopathy.mp. [mp=ti, ab, hw, tn, ot, dm, mf, dv, kw, fx, dq, bt, id, cc, nm, kf, px, rx, an, ui, sy]

4# birth asphyxia.mp. [mp=ti, ab, hw, tn, ot, dm, mf, dv, kw, fx, dq, bt, id, cc, nm, kf, px, rx, an, ui, sy]

5# neonatal encephalopathy.mp. [mp=ti, ab, hw, tn, ot, dm, mf, dv, kw, fx, dq, bt, id, cc, nm, kf, px, rx, an, ui, sy]

6# 2 or 3 or 4 or 5

7# 1and 6

14

493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527

Table 1: Characteristics of the included studies.

Study Design Population NE stagen (%) TH n/N Index test Age at

index test Gold standardAge at gold

standard

Prevalence of adverse outcome

Massaro[16]

retrospective cohort

NE, 36+, NICHD criteria

for cooling

Moderate 11(55%)Severe 9(45%) yes 20/20 rHF

rLF 5-41h Death or BSID-II <2SD 15 months 10/20 (50%)

Metzler[20]

retrospective cohort

NE, 35+, NICHD criteria

for cooling

Moderate 55 (74%)Severe 19 (26%) yes 74/80

RMSSRMSL

rLF24-27h

Death of predominant BG

or global injury on MRI

10 days 24/74 (32%)

Goulding[17]

retrospective cohort

NE, 37+, 2 of the following: pH<7.1, Apgar<=6 at 5min, lac >7mmol or abn. neurology/seizures

Mild 22 (50%)Moderate 9 (20%)Severe 13 (30%)

no 44/61

SDNNTINNVLF

LF, HF

12-48hDeath, CP or Griffiths <87 24 months 20/44 (45%)

Vergales[18]

retrospective cohort

NE, 34+, NICHD criteria

for cooling

Normal 5 (8%)Mild 19 (28%)

Moderate 13(19%)Severe 30 (45%)

yes 67/67 HRC indexHRV 1-7 days

Death or "moderate" or

"severe" injury on MRI

7 days 26/67 (29%)

Table 1: Abn.: abnormal; TH: therapeutic hypothermia; n/N (babies included/total reported in the study); Index test: HRV metrics for which prognostic accuracy metrics were reported.rHF: Relative power of high frequency band; rLF: relative power of low frequency band; RMSS: root mean square at short time-scale (15-50beats); RMSL: root mean square at long time-scale (100-150beats); SDNN: standard deviation of the RR intervals; TINN: triangular interpolation of the NN intervals histogram; VLF: power of very low frequency band; LF/HF: low/high frequency power ratio; HRC index: heart rate characteristics index (proprietary); HRV: Heart Rate Variability (proprietary); BSID: Bayley Scales of Infant Development

15

528529

530531532533

534535536537538539

Table 2: Prognostic accuracy of the best performing HRV metrics originally reported by the included studies.

Massaro 2014[16]

Metzler 2016 [20]

Goulding 2015[17]

Vergales 2014[18]

AUC [95% confidence interval] 0.94 0.79 0.79 [0.57-0.94] 0.83Sensitivity 0.84 0.72 NR NRSpecificity 0.87 0.86 NR NRPPV 0.87 0.74 0.57 NRNPV 0.9 0.79 0.91 NROR [95% confidence interval] NR NR NR 3.19 [1.3-7.8]

AUC: area under the receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value: OR: odds ratio. NR: data not reported.

16

540541

542

543

Table 3: Summary of the calculated diagnostic metrics.

1st Author Metric TP FN FP TN Sensitivity Specificity PPV NPV AUC OR

Metzler 2016 [20] RMSS 15 9 6 38 0.62 [0.41-0.81] 0.86 [0.73-0.95] 0.71 [0.48-0.89] 0.81 [0.67-0.91] 0.74 [0.63-0.86] 10.56 [3.27-34.07]Metzler 2016 [20] RMSL 16 8 11 33 0.67 [0.45-0.84] 0.75 [0.60-0.87] 0.59 [0.39-0.78] 0.81 [0.65-0.91] 0.71 [0.59-0.82] 6 [2.05-17.56]Metzler 2016 [20] rLF 17 7 14 30 0.71 [0.49-0.87] 0.68 [0.52-0.81] 0.55 [0.36-0.73] 0.81 [0.65-0.92] 0.70 [0.58-0.81] 5.2 [1.79-15.1]Massaro 2014 [16] rLF 9 1 3 7 0.90 [0.56-1.00] 0.70 [0.35-0.93] 0.75 [0.43-0.95] 0.88 [0.47-1.00] 0.80 [0.62-0.98] 21 [2.13-.]Massaro 2014 [16] rHF 8 2 1 9 0.80 [0.44-0.98] 0.90 [0.56-1.00] 0.89 [0.52-1.00] 0.82 [0.48-0.98] 0.85 [0.69-1.00] 36 [3.24-.]Goulding 2015 [17] SDNN 9 1 8 9 0.90 [0.56-1.00] 0.53 [0.28-0.77] 0.53 [0.28-0.77] 0.90 [0.56-1.00] 0.71 [0.56-0.87] 10.12 [1.28-.]Goulding 2015 [17] TINN 8 2 7 10 0.80 [0.44-0.98] 0.59 [0.33-0.82] 0.53 [0.27-0.79] 0.83 [0.52-0.98] 0.69 [0.52-0.87] 1.94 [1.02-3.71]Goulding 2015 [17] VLF 8 2 6 11 0.80 [0.44-0.98] 0.65 [0.38-0.86] 0.57 [0.29-0.82] 0.85 [0.55-0.98] 0.72 [0.55-0.90] 7.33 [1.26-40.51]

Goulding 2015 [17] LF 9 1 7 10 0.90 [0.56-1.00] 0.59 [0.33-0.82] 0.56 [0.30-0.80] 0.91 [0.59-1.00] 0.74 [0.59-0.90] 12.86 [1.61-.]

CUMULATIVE 100 32 63 157 0.74[0.67-0.80]

0.79 [0.73-0.84]

0.66 [0.59-0.73]

0.79 [0.73-0.84]

0.73 [0.69-0.77]

7.48[4.88-11.47]

Study without raw dataVergales 2014 [18] HRV NR NR NR NR NR NR NR NR 0.83 3.16 [1.30-7.80]

HRC index NR NR NR NR NR NR NR NR 0.81 6.11 [0.97-68.54]Top table presents data from the three studies for which prognostic metrics were calculated from available raw data or author information. Bottom table presents data for the remaining study. Numbers in between square brackets report confidence intervals. TP: true positives; FN: false negatives; FP: false positives; TN: true negatives; AUC: area under the receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value: OR: odds ratio; RMSS: root mean square at short time-scale (15-50beats); RMSL: root mean square at long time-scale (100-150beats); rLF: relative power of low frequency band; rHF: relative power of high frequency band; SDNN: standard deviation of the RR intervals; TINN: triangular interpolation of the NN interval histogram; VLF: power of very low frequency band; LF: power of high frequency band; HRV : Heart Rate Variability score (proprietary); HRC index: heart rate characteristics index (proprietary).

Table 4: Excluded studies.

17

544545

546547548

549550551552553

554555556557558

1st Author Country n Reason for exclusion Summary of findings

Massaro[29] USA 51 Correlation between HRV and temperature not between

HRV and severity of NE. Not relevant

Goulding [21] Ireland 118 Severity of NE classified per EEG, not MRI or NDO.

HRV higher in TH group than preTH group. Significant difference between mild and moderate NE in preTH group. Significant difference between moderate and severe NE in TH group

Aliefendioglu[22] Turkey 22 Group differences based on Sarnat, not MRI or NDO. Significant difference between control group and moderate/severe NE group

Significant difference between cases who died and cases who survived

Lasky [30] USA 2 Describes changes in Q-T segment with cooling. Not relevant

Kozar [31] Slovakia 3 Correlation between HRV and temperature not between

HRV and severity of NE. Not relevant

Schneebaum[32] USA 40 Examines lateralization of sympathetic and

parasympathetic function. No prognostic metrics. Not relevant

Matic[33] Netherlands 19 Book chapter. Classification with machine learning. No

prognostic metrics. Linear discriminant analysis (LDA) 80%.

Vesoulis[23] USA 16 Describes effect of time and temperature, on HRV no

prognostic metrics

SDNN, TINN, LF, HF, HF/LF higher in unfavorable outcome group than in favorable outcome group, at 24h, but not at 48 and 96h (no estimates or hypothesis testing)

Temko[24] Ireland 38 Same cohort as Goulding 2015 AUC of average HRV metrics = 0.66

HRV: heart rate variability score; TH: therapeutic hypothermia; NE: neonatal encephalopathy; SDNN: standard deviation of the RR intervals; TINN: triangular interpolation of the NN interval histogram; LF: power of the low frequency band; HF: power of the high frequency band; HF/LF high frequency-low frequency ratio.

18

559

560561

Figure 1: Flow chart of the literature review.

Figure 1 legend: PRISMA diagram of literature search performed by two independent reviewers.

Records screened(n = 82)

Records identified through database searching

(n=379)

Full-text articles excluded (n = 8)2 did not use MRI/NDO5 not relevant/no prognostic metrics1 repeated cohort

Studies included in qualitative synthesis

(n = 4)

Full-text articles assessed for eligibility

(n = 12)

Records after duplicates removed(n=315)

Records excluded(abstract not relevant)(n = 70)

Records excluded(title not relevant)(n = 233)

19

562563564565566567568569570571572573574575576577578579580581582583584585586587588589590

Figure 2: Schematic summary of the accuracy of the calculated metrics.

Figure2: PPV: Positive Predictive Value; NPV: Negative Predictive Value; AUC: Area Under (ROC) Curve; DOR: diagnostic Odds Ratio; RMSS: root mean square at short time scale; RMSL: root mean square at long time scale; rLF: relative power of the low frequency band; rHF: relative power of the high frequency band; SDNN: standard deviation of the RR intervals; TINN: triangular interpolation of the RR intervals histogram; VLF: power of the very low frequency band; HRC index: heart rate characteristics index (proprietary); HRV: heart rate variability score (proprietary).

20

591592

593594595596597

598599600

Figure 3: Assessment of the risk of bias in included studies.

Figure 3: Risk of bias of included studies. Pre-specification of HRV thresholds was not reported in any of the studies as well as blinding to the results of reference/index tests.

21

601

602603604