Phase-rectified signal averaging for intrapartum electronic fetal heart rate monitoring is related to acidaemia at birth

Authors

  • A Georgieva,

    Corresponding author
    1. Nuffield Department of Obstetrics & Gynaecology, University of Oxford, Women's Centre, John Radcliffe Hospital, Oxford, UK
    2. Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford, UK
    • Correspondence: Dr A Georgieva, Nuffield Department of Obstetrics & Gynaecology, University of Oxford, Level 3, Women's Centre, John Radcliffe Hospital, Oxford, OX3 9DU, UK. Email antoniya.georgieva@obs-gyn.ox.ac.uk

    Search for more papers by this author
  • AT Papageorghiou,

    1. Nuffield Department of Obstetrics & Gynaecology, University of Oxford, Women's Centre, John Radcliffe Hospital, Oxford, UK
    Search for more papers by this author
  • SJ Payne,

    1. Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford, UK
    Search for more papers by this author
  • M Moulden,

    1. Nuffield Department of Obstetrics & Gynaecology, University of Oxford, Women's Centre, John Radcliffe Hospital, Oxford, UK
    Search for more papers by this author
  • CWG Redman

    1. Nuffield Department of Obstetrics & Gynaecology, University of Oxford, Women's Centre, John Radcliffe Hospital, Oxford, UK
    2. Oxford Biomedical Research Centre, The Churchill Hospital, Oxford, UK
    Search for more papers by this author

Abstract

Objective

Recent studies suggest that phase-rectified signal averaging (PRSA), measured in antepartum fetal heart rate (FHR) traces, may sensitively indicate fetal status; however, its value has not been assessed during labour. We determined whether PRSA relates to acidaemia in labour, and compare its performance to short-term variation (STV), a related computerised FHR feature.

Design

Historical cohort.

Setting

Large UK teaching hospital.

Population

All 7568 Oxford deliveries that met the study criteria from April 1993 to February 2008.

Methods

We analysed the last 30 minutes of the FHR and associated outcomes of infants. We used computerised analysis to calculate PRSA decelerative capacity (DCPRSA), and its ability to predict umbilical arterial blood pH ≤ 7.05 using receiver operator characteristic (ROC) curves and event rate estimates (EveREst). We compared DCPRSA with STV calculated on the same traces.

Main outcome measure

Umbilical arterial blood pH ≤ 7.05.

Results

We found that PRSA could be measured in all cases. DCPRSA predicted acidaemia significantly better than STV: the area under the ROC curve was 0.665 (95% CI 0.632–0.699) for DCPRSA, and 0.606 (0.573–0.639) for STV (= 0.007). EveREst plots showed that in the worst fifth centile of cases, the incidence of low pH was 17.75% for DCPRSA but 11.00% for STV (< 0.001). DCPRSA was not highly correlated with STV.

Conclusions

DCPRSA of the FHR can be measured in labour, and appears to predict acidaemia more accurately than STV. Further prospective evaluation is warranted to assess whether this could be clinically useful. The weak correlation between DCPRSA and STV suggests that they could be combined in multivariate FHR analyses.

Introduction

Asphyxia at birth is confirmed by umbilical arterial blood gases: very low pH values (fetal acidaemia) result from prolonged oxygen deprivation.[1] Intrapartum fetal heart rate (FHR) monitoring aims to predict birth asphyxia and intervention to expedite delivery.

Electronic fetal monitoring (EFM) is currently assessed visually (subjectively) to determine the baseline rate, baseline variability, and presence of accelerations and decelerations;[2] however, subjective assessment has repeatedly been shown to be unreliable and error prone.[3] To address this issue, we and others are developing computerised (objective) analyses of FHR patterns in labour.[4, 5] Baseline variability has both short- and long-term components (STV and LTV, respectively), which are visually assessed as a unit in clinical practice.[2] Computerised analysis allows for much better resolution and the calculation of FHR features that cannot be assessed visually,[6-8] including STV. Throughout this article, STV refers to the computerised measurement and not its clinical equivalent (judged subjectively). STV is one of the strongest indicators of fetal acid–base status before labour.[6] It measures the average change in the FHR from one measurement to the next (either rising or falling).

A relatively new computerised measure of variability is phase-rectified signal averaging (PRSA). It was developed by Bauer et al. for the analysis of adult heart rate variability in electrocardiogram (ECG) time series,[9, 10] to estimate the mortality risk after myocardial infarction.[10] Unlike STV, PRSA measures signal variation in two directions: upwards, accelerative capacity (ACPRSA); and downwards, decelarative capacity (DCPRSA). ACPRSA is the average change in FHR when the signal rises from one point to the next. DCPRSA is the equivalent average when the signal falls. In general, ACPRSA and DCPRSA could be regarded as complementary halves of STV. A useful diagram explaining the precise calculation of ACPRSA and DCPRSA can be found in Huhn et al.[11] Huhn et al. and Lobmaier et al. adapted PRSA to analyse pre-labour FHR.[11, 12] Huhn et al. retrospectively studied 74 growth-retarded fetuses and 161 controls between 28 and 36 weeks of gestation.[11] ACPRSA discriminated growth-retarded fetuses from controls better than STV. Lobmaier et al. came to a similar conclusion in a prospective study of 39 cases versus 43 controls.[12]

PRSA has not been used for the analysis of intrapartum FHR, however. The aims of our study were to: (1) assess whether PRSA in intrapartum FHR traces relates to fetal acidaemia at birth; (2) quantify this relation, based on a large database; (3) compare it with computerised STV, a classical feature to which it is most closely related.

Methods

Case selection

We studied 7568 pregnancies, delivered in Oxford between April 1993 and February 2008 (the total number of deliveries in this period was 107 614, including antepartum stillbirths and births with before 24 weeks of gestation); these are a subset of a large Oxford archive of labour FHR traces, as described elsewhere.[5] The cases were selected to have at least 30 minutes of good-quality FHR trace (<50% signal loss) during the active second stage. Cases were also selected to have complete and reliable outcome information.[5] Further details on the case selection have been previously published.[5] We analysed the last 30 minutes of their FHR traces and the associated outcomes of the infants. All cases meeting the selection criteria were studied.

An umbilical arterial blood pH threshold of 7.05 was used to discriminate acidaemic (low arterial pH) from non-acidaemic (normal arterial pH) babies, as used in our previous studies.[5, 13] Blood was taken from a double-clamped cord into a heparinised syringe and measured within 15 minutes. An Instrumentation Laboratory blood gas analyser was used until 2002, before being replaced with a Radiometer ABL800 Flex. Only cases with paired blood samples were included (i.e. both arterial and venous). Cases in which the difference between venous and arterial values of pH was <0.01 (fifth centile), suggesting that the same cord vessel had been erroneously sampled twice, were excluded. Further details about data acquisition are given in Georgieva et al.[13]

The original 4–Hz sampling of the FHR signal was used. We implemented the algorithm as described by Bauer et al. and Huhn et al. to calculate the ACPRSA and DCPRSA elements for each case, and evaluate their relation to acidaemia at birth.[9, 11]

Assigning parameters to PRSA

ACPRSA and DCPRSA depend on two parameters: T and L. The parameter T is a measure of the smoothness (resolution) of the FHR signal that is most suitable for PRSA calculations. It can be expected that some degree of smoothing would be appropriate because the original signal is recorded by standard monitors at 4 Hz (i.e. four values per second), but the fetal heart beats more slowly than this. Also, micro-fluctuations might not yield useful information. These are smoothed when a larger value of T is used. The parameter L tells us how long before or after the increase (for ACPRSA) or fall (for DCPRSA) we need to look at the signal to extract important information. We tested various combinations of the parameters T and L, varying them between 5 and 110 in steps of 5. For each parameter combination we measured performance using receiver operator characteristic (ROC) and area under the curve (AUC) analysis, as well as by event rate estimate (EveREst) analysis, described below.

Statistical methods

The ROC curves were constructed in order to evaluate the diagnostic measures. ROC curves show the relationship between the true-positive rate and the false-positive rates. The diagnostic variable that gives the largest AUC is considered to have the best performance. We used stata 12/IC to calculate the AUC and confidence intervals, and the ‘roccomp’ function to calculate levels of significance between the AUCs of DCPRSA and STV.

The ROC curves do not show the probability that a diagnostic test will be correct; for this purpose, predictive values are needed. These depend on the prevalence of the condition being detected; this is relevant to this study, where acidaemia at birth has a very low incidence. Therefore, in addition to ROC curves, we used the EveREst plot.[13] It evaluates the performance of a diagnostic feature on a large scale over the total population of patients. The plot displays feature values sorted in ascending or descending order, which are in 20 exclusive groups, each containing 5% of all cases. The x-axis shows the median variable value for the patients in each of the 20 groups. The y-axis shows the percentage rate of events (low pH) in each of the 20 groups. Thus, an optimal diagnostic feature would have a 0% event rate on the left-hand side and a 100% event rate on the right. On EveREst plots we examined differences in predictive ability using a chi-square test for the comparison of proportions.

Correlations between DCPRSA, STV, and pH were calculated using Spearman's rho linear correlation coefficient. Two sided P levels were used throughout.

Results

Clinical and demographic characteristics of the study population have been reported in a previous publication.[5]

The ACPRSA and DCPRSA elements of PRSA were calculated for all cases. Both features performed similarly, and were highly correlated (R = 0.87, when comparing ACPRSA and DCPRSA for the respective optimal values for T and L). For simplicity, we present only the results for DCPRSA.

Varying the value of parameter T did not markedly change the predictive ability of DCPRSA (data not shown). For values of L above 30, the performance was roughly similar. No specific combination of values yielded a markedly better performance, which would indicate the algorithm was over-fitting to this specific data set. Thus the algorithm performed robustly with respect to parameters T and L, and similar results can be expected when DCPRSA is applied to new data. The best parameter combinations were found at T values of 5–10 and L values of 35–55. Thus, the combination of parameters T = 5 and L = 45 was selected for this study.

Increasing values of DCPRSA were related to low pH at birth. Figure 1 illustrates the difference between cases with very low or very high DCPRSA values. ROC curve analysis demonstrated that the AUC for DCPRSA (0.665, 95% confidence interval, 95% CI 0.632–0.699) was significantly greater than that for STV (0.606, 95% CI 0.573–0.639), with a difference between the two AUCs of 0.059 (95% CI 0.012–0.106; = 0.007; Figure 2A).

Figure 1.

Two 30-minute fetal heart rate (FHR) segments with DCPRSA values at the two ends of the spectrum: (A) very low DCPRSA value; (B) very high DCPRSA value.

Figure 2.

Assessment of prediction of arterial cord blood pH ≤ 7.05, comparing short-term variability (STV, red) and phase-rectified signal averaging (PRSA, black), using: (A) receiver operator characteristic (ROC) curve analysis; and (B) event rate estimate (EveREst) plots.

EveREst plots (Figure 2B) showed that the rate of acidaemia consistently increased with increasing values of both DCPRSA and STV. The cases in the highest 5% of DCPRSA had a 17.75% risk of low pH, which was significantly higher than for those in the highest 5% of STV, for which the risk was 11% (< 0.01; Figure 2B).

Although the correlation between DCPRSA and STV was statistically significant, the correlation was weak (r = 0.29), indicating that DCPRSA might be bringing additional useful information. The linear correlations between DCPRSA and STV with pH were −0.26 and −0.20, respectively.

Discussion

Main findings

In this study DCPRSA has been applied for the first time to analyse FHR in labour. Using data from 7568 deliveries, the most suitable PRSA parameters were determined. Increased DCPRSA indicated an increased risk for acidaemia at birth, with slightly higher predictive ability than STV. In addition, EveREst plots showed that DCPRSA offered a better predictive performance than STV. Another important finding is that DCPRSA was not highly correlated with STV, an important established FHR feature. This suggests that PRSA may yield additional useful information about fetal health, and that it may be possible to combine it with STV or other indices. Although the ROC curve is a useful and well-accepted method to assess diagnostic variables, the EveREst plot is more useful to relate a diagnostic variable with clinical practice, especially where the incidence of the event of interest is very low. For example, it can be seen from Figure 2B that a pre-emptive delivery in 5% of all cases (the ones with highest DCPRSA values) would be ‘correct’ in 17.75% occurrences.

Strength and limitations

One strength of our study is that we had a large data set that allowed for the robust examination of PRSA with respect to the parameters T and L; thus, we can expect similar results when PRSA is applied to new data. Our optimal parameter values were different from those reported by Huhn et al.[11] This may be because of the intrinsic differences between antepartum and intrapartum FHR traces, or because of the small data set used by Huhn et al.[11] Nevertheless, we found that the parameter combination reported by Huhn et al. would also perform relatively well for labour traces (data not shown).[11] Huhn et al. reported that ACPRSA yields better results than DCPRSA for diagnosing growth-retarded fetuses before labour;[11] however, our work shows ACPRSA and DCPRSA to be similar and highly correlated parameters for labour traces. A limitation of this study is that we considered only the outcome of low umbilical arterial pH. Other clinical outcomes, such as persistently low Apgar score, perinatal death, or brain damage, remain yet to be examined.

As mentioned in the Introduction, we are aware that it is now recommended that STV and LTV are considered together as a single index of baseline variability.[2] Whereas this is appropriate for visual assessment, computerised analysis allows the frequency components of baseline variation to be resolved and precisely measured, allowing the concept of STV to continue to be a valid feature in this context of measurement.

It can be viewed as a limitation of this study that we do not compare PRSA with all of the computerised features of the FHR. PRSA is by definition a variability measure, and it was only relevant to compare it with STV. The intention of this study was to establish whether PRSA can be useful for computerised FHR interpretation in labour. The bigger goal is ultimately to combine PRSA and STV with other computerised features. We do not propose or claim that traces should be judged by using only PRSA. Extensive further work will be carried out to establish how best PRSA can be combined with existing FHR measures.

The retrospective nature of our investigation is recognised along with potential sources for selection bias, as reported previously.[5] Our purpose was to examine whether PRSA relates to low pH at birth. We argue that any selection bias would not generally change this relationship, although it may have small effects on its magnitude.

Interpretation

We analysed end-of-labour traces, which are fundamentally different from pre-labour (antepartum) traces such as the ones used by Huhn et al.[11] Antepartum traces are more likely to be stable (constant mean and variance), and usually do not have decelerations, whereas end-of-labour traces are very likely to have FHR decelerations. Huhn et al. showed that ACPRSA in antepartum FHR traces is reduced in growth-retarded fetuses. It is known that reduced STV in antepartum traces is an ominous sign.[4, 6, 14] Our work shows that increased DCPRSA, and STV values in end-of-labour traces are related to acidaemia at birth. This inverse relationship has been previously shown for STV: the acute response to acute experimental hypoxaemia or repeated asphyxia in the term fetus is an increase in FHR variability, rather than a decrease.[15]

Future developments could be to investigate whether PRSA is associated with other computerised FHR assessments, such as: approximate entropy (ApEn),[7] which measures the regularity of the signals (i.e. a repetitive signal versus a chaotic one); the signal stability index, which measures the ability of the fetus to maintain its baseline heart rate;[8] or bivariate PRSA for labour monitoring, allowing the incorporation of uterine contraction signals, which stimulate changes in the fetal heart rate.[16] If DCPRSA is not correlated with these findings, the creation of a multivariate computerised FHR analysis tool would be a possibility.

Conclusion

Our results establish that PRSA could be valuable in labour FHR monitoring using computerised analysis: this must be validated using prospective data. We do not claim that PRSA should be used in isolation. Further research is necessary to combine this feature and multivariate analyses with other FHR features that reliably predict acidaemia.

Disclosure of interests

None to disclose.

Contribution to authorship

A.G. contributed to the study design, the maintenance of the FHR database, the FHR analysis and statistical analysis, and to writing the article. A.T.P. contributed to the study design, the statistical analysis, and to writing the article. S.J.P. contributed to the study design, the statistical analysis, and to writing the article. M.M. contributed to the study design, maintaining the database, and writing the article. C.W.G.R. initiated and maintained the digital archive, contributed to the study design, to the statistical analysis, and to the writing of the article.

Details of ethics approval

This study was approved by the national ethics committee: Proportionate Review Sub-Committee of the Newcastle & North Tyneside 1 Research Ethics Committee, REC reference 11/NE/0044.

Funding

A.G. is funded by the Henry Smith Charity and Action Medical Research. A.T.P. is supported by the Oxford Partnership Comprehensive Biomedical Research Centre, with funding from the Department of Health National Institute for Health Research (NIHR) Biomedical Research Centres funding scheme.

Acknowledgements

We are grateful to Dr Eduard Mulder (Utrecht University Hospital), and to Dr Gari Clifford and Mr James Williams (both from the Oxford Institute of Biomedical Engineering), for useful discussions on the topic.

Ancillary