2.1 Participants

The ProMote study is an ongoing longitudinal study of mothers and their premature neonates conducted at the Department of Neonatology/Neonatal Intensive Care Unit of the University General Hospital of Heraklion. Collection of the data presented here lasted from 10^th November 2023–31st March 2025. The study protocol [20] has been approved by the Research Ethics Committee of the University of Crete (103/22.09.2023, 158/15.12.2023 and 38/15.02.2024) and by the Scientific Council, according to the positive recommendation of the Ethics Committee, and the Board of Directors of the General University Hospital of Heraklion (26636/2.10.2023 and 35546/23.10.2023). This research has been performed in accordance with the Declaration of Helsinki. Mothers were informed by the researcher and then through an explanatory letter, allowing time for reflection. Written informed consent was signed by all participating mothers, for themselves and their newborns.

A hundred and thirty-two (132) preterm neonates hospitalized in the NICU were included in the study. Exclusion criteria for the neonates were as follows: the presence of perinatal asphyxia; neurological pathologies; malformation syndromes and major congenital malformations; sensory deficits; metabolic genetic disease; central nervous system infection. Prematurity-associated morbidities correlate with autonomic development in premature infants and may have a greater impact on the extrauterine maturation of this system than birth gestational age [1,3].

Of the 132 recruited neonates, 44 (33.3%) were further excluded due to low signal quality that precluded reliable HRV analysis. Comparison between included and excluded neonates according to birth characteristics (type of delivery, gestational age, weight for gestational age, birth weight, birth height, head circumference, sex, prematurity category and twinship) shows that there are the following variations: a) excluded neonates group included less appropriate for gestational age neonates compared to included neonates (41 vs 72); and b) excluded neonates group included more very preterm neonates (10 vs 5), less moderate (8 vs 29) and less late preterm neonates (24 vs 53) compared to included neonates group (Table 1). Thus, although exclusion was based solely on objective signal quality criteria, the excluded neonate group may represent infants with more fragile or unstable clinical conditions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Comparison of birth characteristics between included and excluded infants from HRV analysis due to poor signal quality.

https://doi.org/10.1371/journal.pone.0339681.t001

Initially, neonates were categorized in four groups according to World Health Organization guidelines (https://www.who.int/news-room/fact-sheets/detail/preterm-birth) as extremely preterm (GA less than 28 weeks), very preterm (GA 28 to less than 32 weeks), moderate preterm (GA 32 to less than 34 weeks) and late preterm (GA 34 to less than 37 weeks). In the current analysis, extremely, very and moderate preterm neonates were all included in Group A (N = 31), while late preterm neonates were included in Group B (N = 57).

It has to be noted that the following restrictions imposed this grouping:

1. a). the number of extremely (<28 weeks) and very preterm neonates (28–31 weeks) was very low (N = 6) and this did not permit a grouping of these two prematurity categories in one group (e.g., Group A).
2. b). According to evidence from the Hellenic Statistical Authority, preterm births are decreasing longitudinally in the region of Heraklion, Crete (https://www.statistics.gr/el/statistics/-/publication/SPO03/-) in which the reference hospital is located,
3. c). In the course of subject recruitment, a number of mothers who gave birth to preterm neonates in the reference hospital were either excluded from the study due to the fact that they did not intent to breastfeed, or denied to participate (see [20] for the published protocol of the study with the inclusion and exclusion criteria).
4. d). There is evidence that, in Greece, from 1991 to 2022 the average annual percent range (AAPC) for late preterm births (34–36 weeks) has increased and it is higher in comparison to moderate preterm births (5.8 vs 4.9) while the rates of extremely (<28 weeks) and very preterm births (28–31 weeks) saw slower growth with AAPCs of 2.2 and 0.7, respectively ( [21]). These trends may justify the high number of late preterm births in the NICU of the University General Hospital of Heraklion, Crete, Greece, from which our sample has been recruited.

2.2 Heart rate variability

2.2.3 Data preprocessing and feature extraction.

All ECG recordings went through structured preprocessing and feature-extraction pipeline. First, we applied band-pass filtering to reduce low-frequency drift and high-frequency noise in each raw signal. Additionally, we corrected baseline wander by fitting and subtracting a high-order polynomial trend from the filtered waveform [26]. This two-step method was especially beneficial for neonatal ECGs, which often have significant motion artifacts and noise from involuntary movements. Subsequent to preprocessing, we determined the R–R intervals on each ECG channel by applying Pan–Tompkins–based peak detection algorithm [27] implemented in NeuroKit2 [28]. The detected R-wave positions were then converted into RR intervals by computing the temporal differences between consecutive peaks. We identified and corrected ectopic beats using a median-based sliding window approach. Specifically, we compared each RR interval to the median of its five adjacent intervals. Deviations exceeding 30% of the local median were flagged as ectopic and replaced by the average of the two neighboring intervals, effectively smoothing abrupt anomalies while preserving normal physiological variability [29].

Calculated HRV features were based on time-domain indices, frequency-domain values and non-linear measurements (Table 2) [2,11,14,16,21,25,30–36].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. HRV metrics measured in preterm infants.

https://doi.org/10.1371/journal.pone.0339681.t002

2.3 Statistical analysis

Statistical analyses were performed using standardized procedures to ensure valid group comparisons and correlation assessments. First, continuous variables were screened for distributional shape using the Shapiro–Wilk test and skewness as shown in S1 Table in S1 File. Variables with Shapiro–Wilk p > 0.05 and |skewness| < 1 were classified as approximately normal to guide the choice of parametric versus non-parametric methods for subsequent analyses. To facilitate comparability across variables expressed in different units, HRV features and GA were standardized as SD scores (mean = 0, SD = 1) prior to analysis. Categorical variables are summarized using their frequency and % percentage.

To examine between-group differences (Group A vs Group B), we initially employed univariate regression (feature ~ Preterm) where coefficients represent standardized mean differences [37]. To evaluate whether these models could be validly interpreted as parametric tests, large sample, linearity, normality and homoscedasticity are required [38]. As all variables, except DFA, are conventional linear-domain indices [39], assumptions of linearity and adequate sample size were met. Then we examined residuals for normality and homoskedasticity with Shapiro–Wilk, Breusch–Pagan, and skewness/kurtosis. In case residuals met all assumptions, regression results could have been retained; otherwise, the Wilcoxon rank-sum (Mann–Whitney U test) was the most proper formal test, following Rosner’s approach [40]. Furthermore, associations of HRV features with GA and PMA were quantified using Spearman’s correlation coefficient (r, two-sided p), stratified by group to avoid masking stage-dependent patterns.

All analyses were two-tailed, with p-values < 0.05 considered statistically significant. Where multiple HRV indices were compared, findings were interpreted with caution to account for potential inflation of type I error due to multiple testing. Statistical analyses were conducted using R version 4.3.2.

4.1 Limitations

Data analysis was restricted to recordings with acceptable signal quality, which effectively reduced the usable dataset and may have introduced a slight bias [19]. Compared to the included neonates, the excluded neonates group included fewer appropriate for gestational age neonates, more very preterm neonates, fewer moderate, and fewer late preterm neonates. Although exclusion was based solely on objective signal quality criteria, the excluded neonate group may represent infants with more fragile or unstable clinical conditions. As a result, our final analytic sample may disproportionately represent clinically more stable preterm neonates. This means that the present findings should be interpreted as primarily applicable to this subgroup, and caution is warranted when generalizing to the broader preterm population, especially to extremely and very preterm neonates.

We did not control and record in detail neonates’ state of alertness [23]. This may affected observed variations in HRV since HRV values according to neonate behavioral states [quiet sleep (S1), active sleep (S2), quiet awake (S3), active awake (S4)] showed that SDNN tended to increase over S1 to S4, with S1 values significantly lower than those of S2, S3 and S4. On the contrary, RMSSD tended to decrease over states S1 to S4, with S1 values significantly higher than those of S3 and S4 [22]. Future studies should address these limitations by incorporating objective state monitoring methods, such as video-based behavioral scoring, actigraphy, or polysomnography, to control for state-dependent variability in HRV. In addition, multi-site protocols would strengthen the generalizability of findings by capturing infants across diverse clinical environments. Together, these methodological refinements will improve the robustness of HRV as a biomarker of autonomic maturation in preterm populations. The reliance on brief recordings means that longer-term fluctuations were not captured, and short recording epochs might not fully represent the infant’s overall autonomic regulation [14].

Grouping of the sample at 32 weeks GA, instead of 34 weeks, would have been robustly justified due to the fact that parasympathetic nervous system maturation undergoes accelerated development at 25–32 weeks [7,9,10,16]. With this grouping, we would expect to find that younger preterms of Group A (<32 weeks) would have less mature ANS at PMA 35–36 weeks compared to preterms of a higher GA (>32 weeks). Meanwhile, it has to be noted that in our sample the number of extremely (<28 weeks) and very preterm neonates (28–31 weeks) was extremely low (N = 6) and this did not permit a grouping of these two categories (extremely and very preterm neonates) on their own as a total in one group (e.g., Group A).

Furthermore, this study was conducted in a single NICU setting, which inherently constrains the external validity of the results, as different NICUs can have varied caregiving protocols and ambient conditions that influence neonate physiology. Multi-site protocols would strengthen the generalizability of findings by capturing infants across diverse clinical environments. Together, these methodological refinements will improve the robustness of HRV as a biomarker of autonomic maturation in preterm populations.

4.2 Comparison of key findings to results of previous relevant research

Our study demonstrated that preterm neonates with higher gestational age at birth (Group B) exhibited elevated time-domain measures (HTI, SDNN, RMSSD, pNN50) and higher frequency-domain indices (LF, LF/HF, TP) compared with those born earlier (Group A). These findings support previous evidence showing that advancing gestational age is associated with greater parasympathetic modulation and more balanced sympathovagal control [1,7,9,16,19,41]. Importantly, the consistency of our results with prior studies reinforces the view that HRV is a reliable marker of autonomic nervous system maturation in this population.

Our results on RMSSD, but not on LF and LF/HF, are consistent with some earlier studies [17,42]. This suggests that while parasympathetic indices such as RMSSD may develop predictably with advancing GA, markers of sympathovagal balance may be more sensitive to methodological context or sample characteristics. The fact that our findings for LF and LF/HF do not align with studies such as those by Fyfe et al. [13] or Patural et al. [18] underlines that clinically reliance on single frequency-domain indices may not be sufficient for evaluating autonomic maturation, and that composite profiles integrating both time- and frequency-domain features are more informative.

We showed that preterm neonates of Group A exhibited higher DFA_alpha and DFA_alpha2 values compared to preterms of Group B. Previous research on non-linear HRV measures has yielded mixed findings [16,17,21,42]. Our results align with some of these studies [16,21,42]. Given that DFA_alpha2 describes long-term fluctuations and long-range dependence [14], higher DFA_alpha2 values of preterms of Group A imply stronger long-range correlations, suggesting less mature physiological system. This highlights the clinical importance of including non-linear metrics in HRV analysis, as they may capture subtle aspects of dysmaturation that linear indices alone cannot detect. Given that little is known about non-linear HRV measures of preterm infants close to their theoretical full term [17], the developmental trajectory of non-linear metrics remains insufficiently understood, and our results underscore the need for further longitudinal work.

We found minimal group differences in HF values. While this aligns with certain studies [1,18], it contradicts several others that reported stronger group differences [7,9,13,16–20,42]. This inconsistency is noteworthy: it underscores that parasympathetic maturation may be particularly sensitive to contextual factors such as sample morbidity, timing of data collection relative to accelerated vagal developmental periods, or NICU environmental stressors. In our study, the relatively low morbidity of the cohort and the timing of HRV recordings at 35–36 weeks PMA may explain the absence of marked HF differences. Clinically, this implies that HF alone may not always serve as a reliable marker of maturation, especially before the second vagal acceleration phase around 37–38 weeks [7–11]. Instead, a broader set of HRV indices may be required to accurately monitor autonomic development in this transitional period.

In summary, by situating our findings in the context of previous evidence, we highlight both convergences and divergences that have important implications. The agreements strengthen confidence in HRV as a biomarker of autonomic maturation, while the discrepancies point to unresolved questions about how prematurity, environmental exposures, and methodological approaches interact to shape ANS development. These insights underscore the necessity for gestational age–specific analyses and justify HRV’s role as a tool for guiding neurodevelopmental monitoring and intervention strategies in preterm infants.

4.3 Interpretation of key findings

Regarding time-domain measures, SDNN provides an estimate of global variability as it is influenced by both SNS and PNS. RMSSD reflects the parasympathetic modulation being associated with sinus respiratory arrhythmia. pNN50 is closely correlated with parasympathetic nervous system activity [2,14,21,42]. For all these time-domain HRV metrics, higher values reflect higher variability, which is more prevalent in healthy states [2]. HR increase indicates a rise in SNS activity and it corresponds to cardiac-linked sympathetic predominance associated with decreased vagal activity [21,43]. The HTI indicates a measure of overall variability during the recording period [2,41]. In terms of frequency-domain findings, LF power in our cohort can be interpreted as reflecting baroreflex-related modulation influenced by both sympathetic and parasympathetic activity [14,16,30]. The LF/HF ratio, although debated, provided additional insight into shifts in sympathovagal balance across gestational age groups. In our study, its elevation in late preterm infants suggests a developmental trajectory toward more balanced autonomic regulation [14] TP is highly correlated to both SNS and PNS activity [31]. Regarding non-linear metrics, detrended fluctuation analysis (DFA_alpha) estimates the long-range dependence and correlation properties of the signal using a self-similarity – autocorrelation parameter (a). DFA_alpha2 describes long-term fluctuations in heart rate and negatively correlates with increasing HRV [14,16].

On this basis, the above-mentioned group variations, which favor Group B compared to Group A in time-domain (RMSSD, SDNN, pNN50, HTI and TP) and in certain frequency-domain (LF, LF/HF) metrics imply that Group B compared to Group A, exhibited significantly higher HRV magnitudes and greater parasympathetic tone. Further, we indicated that preterm neonates born earlier than 34 weeks gestational age showed higher mean heart rates and stronger long-range correlations suggesting less randomness and reduced adaptability, thus a less mature physiological system. Taken together, Group A infants (born earlier) may display persistent immaturity and higher baseline heart rates, whereas Group B infants (born later) show evidence of vagal catch-up and more balanced sympathovagal control. This reversal suggests that HRV development is non-linear, with different maturational “phases” across gestational groups rather than a uniform trajectory. This pattern aligns with developmental physiology, where parasympathetic tone accelerates after ~25–32 weeks GA and again around 37–38 weeks GA. Thus, persistent immaturity of Group A infants (particularly those born before 32 weeks GA) may be due to the fact that not only they missed both the first and the second period of in-uterus PNA accelerated development but also in the first days of their life they are exposed to extra-uterus NICU multiple stressors that can interfere with autonomic maturation. This implies that modeling ANS development requires approaches that capture non-linear, stage-dependent changes rather than assuming linear progression across gestational ages.

Variations in HRV metrics between Groups A and B, which favored Group B in a transitional period of ANS development, may be due to a combination of their different developmental period in uterus during gestation, specific in-utero stressors linked with the preterm birth, nutritional, environmental or iatrogenic stress in the ex utero environment [1,7–11,16,19].

4.4 Developmental implications

a) Understanding ANS Development and Maturation in a Transitional Window: The absence of HRV indices’ increase of Group A preterms by 35 weeks PMA (when still in NICU) suggests a developmental lag that likely continues until they reach a more advanced PMA though relevant results are mixed [13,44]. Clinically, this emphasizes the need for careful monitoring of preterm infants’ autonomic status; their fragile ANS may benefit from supportive interventions during the NICU stay to foster maturation; b) Connections to Neurodevelopmental Trajectories: Evidence shows that decreased neonatal HRV has been associated with adverse neurodevelopmental outcomes in preterm populations [3]. Our finding of a strong positive correlation between GA and vagal indices therefore not only marks a physiological maturation, but also could signal which infants might have smoother versus more vulnerable developmental trajectories. By understanding that HRV features such as pNN50, HTI, or DFA_alpha, are early indicators of ANS maturity, we gain insight into one piece of the complex puzzle of how prenatal and immediate postnatal physiology can influence long-term neurodevelopment. This bridges biological findings with developmental outcomes, reinforcing the idea that early physiology can shape – and be shaped by – the caregiving environment, with lasting consequences for neurodevelopment; c) Implications for NICU Care and Early Interventions: Preterm infants in NICU are subject to numerous stressors in the NICU at a time when their ANS is highly vulnerable [15]. Our results, which demonstrate the gap in HRV development between more and less mature preterms, contribute to the dialogue and calls for early interventions specifically targeting autonomic function, for instance, interventions that promote vagal tone – potentially through feeding practices [45], kangaroo care [46], Music Therapy [47], maternal affective touch [48], spontaneous interactions with parents [49] and parent-infant cardiac synchronization [50] – to reinforce family-centered care in the NICU [51].