Temporal Trends in Heart Rate Characteristic Index Preceding Necrotizing Enterocolitis in Preterm Infants

Abstract

Objective

To evaluate temporal trends in heart rate variability as defined by the Heart Rate Characteristic index (HRCi) preceding necrotizing enterocolitis (NEC) in preterm infants.

Study Design

Retrospective cohort study of neonates with NEC (Bell Stage II/III) at a tertiary NICU from 2020–2024. Continuous HRCi data were extracted for 15 days before and 4 days after NEC diagnosis. HRCi trajectories were analyzed using generalized additive models, Wilcoxon signed-rank tests, and linear regression. Logistic regression models were fit to daily HRCi with NEC diagnosis day (Day 0) as the binary outcome, adjusting for gestational age, small for gestational age (SGA) status, and birth weight Z-scores. Model discrimination was assessed using receiver operating characteristic curves and area under the curve (AUC).

Results

Among 51 infants with NEC, median birth weight was 900 g and gestational age 27.7 weeks. HRCi rose progressively two weeks before onset of NEC, with a significant increase three days before diagnosis (median difference 0.31; p = 0.031). Logistic regression showed HRCi remained independently associated with NEC diagnosis day (OR = 1.28; 95% CI: 1.12–1.45; p < 0.001), while gestational age and birth weight Z-score were not. Within-patient AUCs were modest (0.65–0.66); exploratory comparison to the broader NICU population yielded higher discrimination (AUC = 0.84).

Conclusion

Rising HRCi preceded NEC diagnosis, with significant elevation three days before onset. HRCi demonstrated independent temporal association with NEC, though modest discrimination reflects the exploratory design. Prospective validation with matched controls and confounder adjustment is required to establish clinical utility.

Introduction

Necrotizing enterocolitis (NEC) is the most serious gastrointestinal disease affecting preterm and low birth weight infants. Despite modern diagnostics and therapy, NEC has mortality rates ranging from 20% to 50% [1–3]. Beyond the high mortality, NEC is associated with significant morbidity among survivors, including gastrointestinal complications and long-term neurodevelopmental impairment [4]. Recognized NEC risk factors include prematurity, intestinal dysbiosis, limited human milk feeding, and neonatal sepsis [5].

The Heart Rate Characteristic index (HRCi, aka HeRO Score; Medical Predictive Science Corporation, Charlottesville, VA) is a bedside monitoring system that analyzes heart rate beat-to-beat variability integrated with heart rate decelerations, displaying a continuous risk score. Originally developed for early detection of late onset sepsis in preterm infants, HRCi monitoring has been associated with reduced sepsis-related mortality in neonates, with one study noting a decrease from 19% to 11% among infants with culture-confirmed sepsis [6,7]. Emerging literature suggests that systemic inflammation and autonomic dysregulation play a role in NEC pathogenesis, with coincident suppression of heart rate variability [8]. This supports a plausible physiologic link between early subclinical NEC and changes in HRCi.

No specific physiologic marker currently exists for early NEC detection before clinical signs emerge, and evaluation is typically prompted by symptoms such as abdominal distension, feeding intolerance, or apnea. Although NEC and neonatal sepsis share overlapping pathophysiology and frequently co-occur [9,10], there is a paucity of research evaluating heart rate variability as an early indicator of NEC before clinical signs present, with prior studies exploring NEC as part of a composite outcome [11–15]. Ultrasonography, Near-Infrared Spectroscopy (NIRS) and physiologic biomarkers have been investigated as possible predictors of NEC onset, but no approach has achieved widespread clinical acceptance and implementation [16]. We hypothesized that HRCi would rise preceding clinical NEC diagnosis, reflecting early autonomic dysregulation associated with evolving intestinal inflammation. Specific aims included characterizing temporal HRCi trends in the period before NEC diagnosis in preterm neonates and evaluating the association between HRCi elevation and NEC onset.

Methodology

This retrospective cohort study included neonates diagnosed with necrotizing enterocolitis at Atrium Health Wake Forest Baptist Brenner Children’s Hospital between January 1, 2020, and December 31, 2024. Patients were identified from the Children’s Hospitals Neonatal Consortium (CHNC, Dover, DE) database, and included only infants with confirmed NEC Bell Stage II or III [17]. For the purposes of temporal analysis, “Day 0” was defined as the date of NEC diagnosis, marked by radiographic confirmation and initiation of medical or surgical treatment for NEC. All HRCi values were extracted and analyzed relative to Day 0, with days preceding diagnosis labeled as negative (Day −15 to Day −1) and days following diagnosis labeled as positive (Day +1 to Day +4).

Of the 62 infants diagnosed with NEC during the study period, 51 had continuous HRCi monitoring data available (Figure 1). The HRCi value is derived from heart rate variability analysis, incorporating measures such as reduced variability and transient decelerations. The monitoring system generates scores hourly from 0 to a maximum of 7, with <1 considered baseline, 1–2 indicating increased risk, and >2 regarded as a significant predictor of sepsis in prior studies [6,18]. HCRi is routinely generated and presented at the bedside to clinicians for all patients in the Baptist NICU, and data were mapped to individual bed assignments to ensure accurate temporal alignment with clinical events. Eleven patients were excluded as they were transferred to the Wake Forest Baptist NICU from an outside facility without prior HRCi monitoring, were not matched to the correct bed assignment, or presented from home without corresponding pre-diagnostic physiologic trend data necessary for this analysis.

Figure 1. Study flow diagram.

CONSORT-style flow diagram showing identification, exclusion criteria, and final cohort inclusion. Of 62 neonates diagnosed with Bell Stage II/III NEC during the study period, 51 had continuous HRCi monitoring data available for analysis. Eleven patients were excluded due to outside facilities without prior HRCi monitoring (n=8), inability to match to correct bed assignment (n=2), or presentation from home without pre-diagnostic physiologic data (n=1).

Abbreviations: HRCI, Heart Rate Characteristic index; NEC, necrotizing enterocolitis

For the 51 included patients, HRCi data were extracted for the 15-day period immediately preceding and 4 days after Day 0 (NEC diagnosis day). This window was selected to avoid the immediate postnatal period when heart rate variability is affected by transitional physiology and autonomic immaturity, which can result in unreliable HRCi values unrelated to evolving disease [6,10]. The 15-day pre-diagnosis period allowed focus on HRCi deviations more likely to represent true physiologic change associated with NEC pathogenesis, though for infants developing NEC in the first two weeks of life, this window partially overlapped with the early postnatal transition. Additional clinical and demographic variables obtained from the CHNC Neonatal Database included birth weight (BW), gestational age (GA), sex, race, and in-hospital mortality. NEC diagnosis, staging, and outcomes were verified by author HM through manual review of the CHNC dataset in conjunction with institutional records.

Descriptive statistics were used to summarize demographic and clinical characteristics of the cohort. For each infant, the maximum observed HRCi value per calendar day was extracted, capturing the highest physiologic risk signal. A generalized additive model (GAM) was used for descriptive visualization of smoothed trends in daily maximum HRCi, with 95% confidence bands plotted to illustrate variability from Day −15 to Day +4. Daily HRCi values were aggregated into 2-day bins for visualization to reduce high-frequency variability in the data, as 1-day bins produced markedly noisier trends. This approach preserved the underlying trajectory while improving interpretability of the temporal pattern. GAMs allow for flexible modeling of non-linear relationships between predictors and the outcome using smooth functions of covariates while preserving model additivity [19]. To contextualize these trends, we also extracted maximum daily HRCi values from all infants in the NICU who had active HRCi monitoring during the same study period, regardless of diagnosis. These values were used to derive the 25th, 50th, and 75th percentile reference distributions representing the range of HRCi typically observed in the unit. Because the reference cohort included infants across a range of postnatal ages and clinical conditions, the percentiles were derived from the full distribution without age-adjustment; the goal was to provide a real-world baseline for physiologic variability rather than a matched control group. For the exploratory ROC comparison, HRCi values from NEC patients on Days −1 and 0 were compared with HRCi values from all other infants with active HRCi monitoring during the same period who did not have a diagnosis of NEC. This analysis was descriptive and intended to contextualize NEC-associated HRCi elevation rather than to provide a validated predictive model. HRCi from the three days immediately preceding diagnosis were compared with those from the prior week using the Wilcoxon signed-rank test. Finally, a simple linear regression was performed to assess the association between days from NEC diagnosis and HRCi, modeling HRCi as a function of time relative to NEC diagnosis over the 15-day pre-diagnosis period.

Logistic regression models were constructed with Day 0 (NEC diagnosis day) as the binary outcome and daily maximum HRCi as the primary predictor. These models were designed to characterize the temporal association between HRCi and NEC diagnosis using within-patient comparisons, rather than to develop a clinical prediction tool. Each infant’s daily HRCi values from Days −15 to 0 were included, comparing values on Day 0 to their own earlier measurements. This approach allowed us to evaluate whether HRCi elevation is temporally linked to NEC onset, rather than to assess predictive accuracy for clinical decision-making at the bedside. Birth weight Z-scores were calculated using Fenton growth curve parameters (see Supplementary Methods) [20]. Infants with birth weight below the 10^th percentile (Z < −1.28) were classified as small for gestational age (SGA). To control for potential confounding by baseline physiologic characteristics, models were sequentially adjusted for GA, SGA status, and birth weight Z-score. The within-patient design inherently minimized confounding by fixed infant characteristics, as each infant served as their own control. Model performance was evaluated using ROC curves and AUC. This analytic approach used repeated daily HRCi measurements from the same infants. Logistic regression models were fit without adjustment for within-patient clustering, which may bias standard errors; however, given the exploratory aim of characterizing temporal association rather than developing predictive models, we prioritized interpretability. Time-varying confounders such as feeding advancement, concurrent infections, transfusion events, and medication exposures could not be included due to inconsistent temporal resolution in retrospective data collection and represent important limitations to address in prospective studies.

Data analysis included date formatting, deduplication, and alignment of the date of first charted clinical concern for NEC (“Day 0”). Missing data were addressed using complete case analysis. No imputation methods were employed, and sensitivity analyses were not performed to assess impact of missing data, which represents a limitation of this exploratory analysis. Statistical analyses were conducted using R software. This study was approved by the Wake Forest University Institutional Review Board with a waiver of informed consent as a retrospective study.

Results

Cohort Characteristics

A total of 51 neonates with Bell Stage II or III NEC were included in the final analyses. The median birth weight (BW) was 900 g (interquartile range [IQR]: 685, 1275), and the median gestational age (GA) at birth was 27 weeks 5 days (IQR: 24 weeks 6 days, 31 weeks 3 days). 31 infants were male (61%), 22 infants were Black (43%), 16 were White (31%), and 1 was Asian (2%). 12 infants (24%) had no race category selected or had been categorized as “Other” (Table 1).

Table 1.

Clinical and demographic characteristics of neonates with Bell Stage II or III NEC (n = 51).

Characteristic	Value
Birth weight, g	900 (685–1275)
Gestational age at birth, weeks	27.7 (24.8–31.4)
Male sex, n (%)	31 (61%)
Female sex, n (%)	20 (39%)
Race/Ethnicity, n (%)
– Black	22 (43%)
– White	16 (31%)
– Asian	1 (2%)
– Other/Unknown	12 (24%)
NEC onset postnatal age, days	21 (11–35)
Surgical NEC, n (%)	17 (33%)
In-hospital mortality, n (%)	8 (16%)

Values are presented as median (interquartile range) or number (percentage), as appropriate.

Abbreviations: NEC, necrotizing enterocolitis.

In-hospital mortality of the cohort was 16% (8 of 51 infants). 17 infants (33%) underwent surgical intervention for NEC. The median age at NEC onset was 21 days (IQR: 11, 35) and was significantly earlier in male infants compared to female infants (12.5 days vs 33.5 days, respectively; p = 0.0007, Wilcoxon rank-sum test). Wilcoxon rank-sum testing revealed no statistically significant differences in age at NEC onset among infants who died, were discharged, or were transferred (all adjusted p > 0.29). Although the median age at NEC onset was earlier among infants who died (postnatal day 13) compared to those who survived to discharge (postnatal day 25), this difference was not statistically significant.

HRCi Trends

HRCi values across the cohort ranged from 0.16 to 7.00 (the system maximum), with a median HRCi of 2.18 (IQR: 1.06, 4.34). Smoothed trends using generalized additive modeling (GAM) revealed a consistent increase in maximum daily HRCi approaching NEC diagnosis. To contextualize HRCi elevation in NEC patients, the 25th, 50th, and 75th percentiles of maximum daily HRCi from all infants in the NICU monitored with HRCi during the same study period were plotted, representing the baseline distribution of physiologic variability in the unit (Figure 2). This comparison demonstrated that NEC patients had elevated HRCi beginning up to 15 days prior to clinical diagnosis, with values consistently exceeding NICU-wide median thresholds. Comparing the three days immediately preceding NEC diagnosis (Days −3 to 0) with the Days −10 to −4, median HRCi increased from 2.33 to 2.64 (median paired difference of 0.31, p = 0.031 by Wilcoxon signed-rank test). Linear regression analysis confirmed a statistically significant positive trend in HRCi as NEC diagnosis approached (p < 0.001).

Figure 2. Maximum daily Heart Rate Characteristic index (HRCi) in necrotizing enterocolitis patients increased leading up to first clinical concern (Day 0).

Boxplots display the distribution of maximum daily HRCi values aggregated into 2-day bins (e.g., Days −15 to −14, Days −13 to −12), with the horizontal black line within each box representing the median value for that time period. Individual daily HRCi measurements are overlaid as points with grayscale shading indicating risk level: light gray represents low infectious risk (HRCi < 2), medium gray represents intermediate risk (HRCi 2–6), and dark gray to black represents high infectious risk (HRCi ≥ 6). The thick black curved line shows a smoothed trend using a generalized additive model, with the surrounding gray shaded area representing the 95% confidence interval. The vertical solid black line at x = 0 marks Day 0, defined as the day of NEC diagnosis when clinical evaluation was initiated and radiographic confirmation obtained. Three horizontal dark gray reference lines indicate the 25th percentile (lower dashed line), 50th percentile/median (middle solid line), and 75th percentile (upper dashed line) of HRCi values from all infants with active HRCi monitoring in the NICU during the study period, providing baseline context for typical physiologic variability. The progressive upward trend demonstrates that NEC patients exhibited elevated HRCi values beginning as early as Day −15, with continued increase approaching clinical diagnosis.

Abbreviations: HRCi, Heart Rate Characteristic index; NEC, necrotizing enterocolitis; GAM, generalized additive model; CI, confidence interval; NICU, neonatal intensive care unit.

In an exploratory analysis, the HRCi trajectories of infants with surgical NEC (n = 17) were compared to those with medical NEC (n = 34). Across the entire 15-day pre-diagnosis window, median HRCi values were similar between groups (2.14 for surgical vs. 2.19 for medical NEC). In the three days immediately preceding diagnosis (Days −3 to 0), surgical cases demonstrated slightly higher median HRCi (2.57, IQR: 1.27, 5.43) compared to medical cases (2.38, IQR: 1.09, 5.01), though distributions overlapped considerably. Given the modest sample size and overlapping distributions, we did not perform statistical testing. These preliminary observations suggest that HRCi elevation occurs across the spectrum of NEC severity, with no clear threshold distinguishing medical from surgical disease in the pre-diagnostic period.

Regression Modeling

Logistic regression models were constructed to characterize the association between HRCi and NEC diagnosis day (Table 2). In Model 1, which included HRCi alone, increasing HRCi was significantly associated with NEC diagnosis (odds ratio [OR] = 1.23 per unit increase; 95% confidence interval [CI]: 1.09–1.39; p = 0.001). In Model 2, which additionally adjusted for gestational age and SGA status, HRCi remained statistically significant (OR = 1.28; 95% CI: 1.12–1.46; p < 0.001). Gestational age showed a borderline association (OR = 1.07 per week; 95% CI: 1.00–1.16; p = 0.059), while SGA was not predictive (OR = 1.21; 95% CI: 0.47–3.13; p = 0.69). In Model 3, which included HRCi, gestational age, and birth weight Z-score, HRCi again remained independently associated with NEC (OR = 1.28; 95% CI: 1.12–1.45; p < 0.001). This indicates that each 1-unit increase in HRCi was associated with a 28% increase in the odds of that day being the NEC diagnosis day. For example, an HRCi increase from 2.0 to 3.0 would correspond to 28% higher odds of Day 0 versus earlier days. Gestational age demonstrated a non-significant trend (OR = 1.07; 95% CI: 0.99–1.16; p = 0.078), and birth weight Z-score was not independently associated with NEC onset (OR = 0.96; 95% CI: 0.64–1.42; p = 0.83).

Table 2.

Logistic regression models assessing association between HRCi and NEC diagnosis day (Day 0).

	Predictor	Estimate	SE	OR	95% CI	p-Value
Model 1	HRCi	0.206	0.063	1.23	1.09–1.39	0.00106
Model 2	HRCi	0.246	0.067	1.28	1.12–1.46	<0.001
	Gestational Age (weeks)	0.072	0.038	1.08	1.00–1.16	0.0595
	SGA (<10th percentile)	0.192	0.48	1.21	0.47–3.11	0.6891
Model 3	HRCi	0.244	0.067	1.28	1.12–1.45	<0.001
	Gestational Age (weeks)	0.072	0.041	1.07	0.99–1.16	0.0781
	Birthweight Z-score	−0.044	0.202	0.96	0.64–1.42	0.8291

Abbreviations: HRCi, Heart Rate Characteristic index; NEC, necrotizing enterocolitis; SE, standard error; OR, odds ratio; CI, confidence interval; GA, gestational age; SGA, small for gestational age.

The three logistic regression models yielded modest AUC values, with AUCs of 0.65 for the HRCi-only model, 0.66 for the model including gestational age and SGA, and 0.66 for the model including gestational age and BW Z-score (Figure 3). These AUC values indicate modest discrimination and reflect limited standalone predictive utility for clinical decision-making. The modest performance is expected given the within-patient design, which compared each infant’s diagnosis day to their own earlier values rather than distinguishing NEC cases from non-NEC controls. HRCi remained the primary variable associated with NEC onset, with limited additional value from growth-related covariates. In an exploratory comparison, HRCi values on Days −1 to 0 in infants with NEC were compared with HRCi values from all other infants in the NICU who had active HRCi monitoring during the same period and did not have NEC (n=4407 infants across varying postnatal ages, gestational ages, and clinical conditions). This descriptive analysis yielded an AUC of 0.838, indicating that HRCi levels in the immediate pre-diagnostic period were generally higher among infants who developed NEC than among the broader monitored NICU population (Figure 4). This analysis was intended to contextualize HRCi elevation rather than serve as a formal internal validation model.

Figure 3. Receiver operating characteristic curves for logistic regression models predicting necrotizing enterocolitis diagnosis day (Day 0).

Three ROC curves are shown representing different logistic regression models. Model 1 (solid black line): HRCi only, AUC = 0.65. Model 2 (dashed black line): HRCi + gestational age + small for gestational age, AUC = 0.66. Model 3 (dotted black line): HRCi + gestational age + birth weight Z-score, AUC = 0.66. Diagonal reference line (dashed gray line) represents no discrimination (AUC = 0.5). Models were designed to characterize temporal association within NEC patients (comparing Day 0 to earlier days) rather than predict NEC occurrence.

Abbreviations: ROC, receiver operating characteristic; NEC, necrotizing enterocolitis; HRCi, Heart Rate Characteristic index; AUC, area under the curve; SGA, small for gestational age.

Figure 4. Exploratory ROC analysis: HRCi values in NEC patients vs. broader NICU population.

ROC curve comparing maximum daily HRCi values from Days −1 to 0 in infants with NEC (n = 51) to HRCi values from all other infants with active HRCi monitoring during the same period who did not have NEC (n = 4407). AUC = 0.838. This was a descriptive, uncontrolled comparison without matching for gestational age, postnatal age, or clinical acuity. While this finding suggests potential discriminatory ability, it should not be interpreted as validated predictive performance. Formal case-control studies with appropriate matching are required before clinical application.

Abbreviations: ROC, receiver operating characteristic; NEC, necrotizing enterocolitis; HRCi, Heart Rate Characteristic index; AUC, area under the curve.

Discussion

In this retrospective cohort study, we identified a distinct pattern of rising HRCi among preterm neonates in the days leading up to clinical signs of NEC and confirmatory diagnosis. Notably, our findings demonstrate that HRCi begins to elevate as early as two weeks before clinical NEC onset, with a statistically significant increase in the three days immediately preceding diagnosis. This trend supports our hypothesis that heart rate variability, captured by the HRCi, may reflect early subclinical changes in autonomic function preceding the overt clinical signs of NEC. These findings are consistent with prior literature associating systemic inflammation and inflammatory mediators with autonomic dysregulation prior to the development of both NEC and sepsis [1,4,21]. Although finer temporal resolution (e.g., 1-day bins) might capture short-term variability better, the limited sample size produced unstable day-to-day estimates. Aggregating over 2-day intervals allowed clearer visualization of consistent directional trends while minimizing noise. Importantly, this study was not designed to evaluate HRCi as a diagnostic test on the day of NEC onset or to compare it directly with bedside clinical assessment. Day 0 served as the clinical anchor to examine HRCi changes in the days leading up to NEC. The clinical relevance of HRCi may lie in the trend of progressive elevation, particularly within the three-day window before NEC presents, rather than the single value at diagnosis. Our findings show a gradual rise in HRCi in the week preceding NEC evaluation, suggesting autonomic dysregulation may begin before overt clinical signs. Future studies designed to provide real-time predictive validation will require inclusion of infants evaluated for NEC who do not develop the disease, which was beyond the scope of this initial trend characterization analysis.

NEC patients demonstrated persistently elevated HRCi starting as early as Day −15 when cohort HRCi were compared to percentile benchmarks derived from the general NICU population at our institution. This suggests a potentially heightened baseline inflammatory state or increased physiologic stress in these infants well before the onset of NEC. Such early elevations could be attributable to comorbid conditions (e.g., respiratory distress syndrome, indwelling catheters, or subclinical infection), or underlying predisposition to gut dysbiosis and disease, or impaired autonomic regulation [1,4]. Further investigation is needed to test these hypotheses systematically and determine whether high early HRCi may serve as a marker of early NEC development.

Another novel observation in our analysis was the downtrend in HRCi beginning on Day 0, the day NEC was first clinically suspected and treatment was initiated. This pattern could reflect a physiologic response to early intervention, including supportive care or initiation of antibiotics similar to the trend noted after sepsis diagnosis [7]. The precise clinical significance of this day 0 inflection point remains unclear and merits further investigation, particularly its utility in monitoring treatment response.

Several time-varying clinical factors may influence HRCi trajectory in the pre-NEC period and warrant consideration in interpreting our findings. Feeding advancement, a central component of NEC pathogenesis, may coincide with early autonomic changes reflected in HRCi elevation. Similarly, blood transfusions—a recognized NEC risk factor—could contribute to inflammatory signaling and subsequent HRCi rise. Concurrent infectious processes, whether clinically recognized or subclinical, may independently elevate HRCi and confound its specificity for NEC. Additionally, the clinical practice of withholding feeds or initiating empiric antibiotics in response to concerning signs may itself alter the HRCi trajectory, creating a complex bidirectional relationship between clinical intervention and physiologic monitoring. While our study design focused on characterizing the temporal pattern of HRCi change relative to NEC diagnosis, future prospective work incorporating detailed daily clinical variables—including feeding volumes, transfusion timing, antibiotic exposure, and laboratory markers—will be essential to discriminate between these relationships and assess the independent contribution of HRCi to NEC prediction.

HRCi trends demonstrated a significant independent association with NEC diagnosis day in our logistic regression models; however, the modest area under the curve (AUC = 0.66) reflects the exploratory, within-patient design of this analysis and should not be interpreted as predictive performance for clinical decision-making. These models were designed to assess temporal association—whether HRCi systematically increases as NEC approaches—rather than to provide a validated prediction tool for identifying which infants will develop NEC. The modest AUC is expected given that we compared each infant’s diagnosis day to their own earlier days, rather than comparing NEC patients to non-NEC controls. However, exploratory ROC analysis comparing HRCi among NEC patients to all other monitored NICU infants without NEC yielded a higher AUC of 0.838. This suggests potential discriminatory ability when comparing NEC to non-NEC infants, however, this was a descriptive, uncontrolled comparison without matching for gestational age, postnatal age, or clinical acuity, and should not be interpreted as validated predictive performance. Formal case-control or prospective cohort studies with appropriate matching and validation are required before clinical application. Our findings align with recent multicenter work by Kausch et al. demonstrating that cardiorespiratory monitoring (heart rate and oxygen saturation via the Pulse Oximetry Warning System) predicted NEC with AUC = 0.758, similar to sepsis prediction (AUC = 0.804) [22]. Like our results, they observed dynamic rises in physiologic scores preceding clinical diagnosis, reinforcing that NEC and sepsis share autonomic signatures detectable through continuous monitoring. Despite these limitations, HRCi trends may have clinical value when integrated with other risk assessment tools, consistent with prior studies showing that the utility of HRCi is enhanced when interpreted alongside clinical judgment and other risk factors [13,23,24]. Low HRCi may also have clinical utility in reassuring clinicians that sepsis risk is low, highlighting the negative predictive value for infection of HRCi. Our analysis identified that male infants developed NEC significantly earlier than female infants (median 12.5 vs 33.5 days, p = 0.0007). This is consistent with prior literature demonstrating that male preterm infants have higher rates of respiratory distress syndrome, sepsis, and overall mortality compared to females, potentially reflecting differences in immune system maturation, inflammatory responses, or gut microbiome development [25]. Whether sex-specific differences in autonomic maturation or HRCi trajectories contribute to this earlier onset in males warrants further investigation.

These findings highlight the potential of HRCi monitoring as a non-invasive, real-time adjunct to existing NEC surveillance strategies. When combined with clinical risk factors, biomarkers, and imaging modalities such as abdominal ultrasound or near-infrared spectroscopy (NIRS), heart rate variability metrics have the potential to strengthen a multimodal approach for earlier NEC detection [26,27].

Our exploratory comparison of surgical versus medical NEC cases revealed similar HRCi elevation patterns in both groups, suggesting that autonomic dysregulation reflected by HRCi may be a common feature of NEC pathophysiology regardless of ultimate disease severity. While surgical cases showed slightly higher median HRCi in the immediate pre-diagnosis period (2.57 vs. 2.38), the overlapping distributions suggest HRCi alone may not reliably distinguish which infants will require surgical intervention. Future larger studies should investigate whether combining HRCi with other clinical parameters, laboratory markers, or imaging findings could improve prediction of disease severity and guide treatment intensity.

Limitations:

This study has several limitations that should be considered when interpreting our findings.

Study Design:

As a single-center retrospective analysis, the generalizability of our findings is limited. However, our patient population is a mix of local urban/suburban cities and rural referral regions, representing a cross-section of neonates across a large, heterogeneous state, in contrast to cohorts from large urban centers that may reflect more specialized or atypical populations. The absence of a matched non-NEC control group for formal predictive modeling constrains interpretation of specificity and sensitivity. Nonetheless, we captured and analyzed all NEC cases in our unit, and HRCi was interpreted relative to baseline values established in non-ill infants. A logical next step would be to examine infants with rising HRCi who underwent a NEC evaluation but were ultimately found to be negative, as in a case-matched or cohort structured study this would allow assessment of false-positive signal.

Data Availability:

Our retrospective dataset lacked detailed data on concurrent sepsis evaluations, culture results, antibiotic exposures, pain medication administration, and hemodynamic instability or shock – all of which can affect autonomic tone and heart rate variability. NEC diagnosis was established using Bell staging criteria based on radiographic and clinical findings documented in the CHNC database; however, given the known overlap between NEC and sepsis, we cannot definitively attribute HRCi elevation to NEC-specific pathology versus concurrent infection. We also did not incorporate time-varying clinical variables such as feeding status and transfusion events, which may interact with autonomic function and contribute to HRCi changes. Furthermore, our selection of a 15-day pre-diagnosis window assumed sufficient postnatal maturation for reliable HRCi interpretation. However, 25% of our cohort developed NEC before 15 days of life, meaning the earliest monitoring days for these infants overlapped with the early postnatal transition period when HRCi may be less reliable. This may have affected our ability to detect pre-NEC changes in the youngest infants, and may have introduced heterogeneity in baseline HRCi interpretability.

Modeling Approach:

Our logistic regression models were fit without adjustment for within-patient clustering, which may bias standard errors; however, given the exploratory aim to characterize temporal association rather than develop predictive models, we prioritized interpretability. Future work should employ mixed-effects models or generalized estimating equations (GEE) to appropriately account for correlation structure. Additionally, our modeling approach did not include internal validation (e.g., cross-validation or bootstrapping) or formal calibration analysis. Critically, our logistic regression models were designed to characterize temporal association within NEC patients rather than to develop or validate a clinical prediction model; the modest AUC values reflect this exploratory design and should not be interpreted as predictive performance metrics for prospective NEC surveillance.

Physiologic Specificity:

The HRCi is inherently non-specific and reflects autonomic dysregulation from multiple inflammatory and infectious processes, not exclusively NEC. Changes in HRCi can be influenced by concurrent conditions such as sepsis, respiratory distress, hemodynamic instability, or medication effects, which may confound associations specifically focused on NEC pathophysiology.

Future Directions:

As an exploratory, single-cohort study, we focused on temporal association rather than model development or generalizability. Future prospective studies should systematically capture detailed infectious disease variables, medication exposures, hemodynamic parameters, and time-varying clinical data to isolate NEC-specific HRCi changes from other factors affecting autonomic function. Such studies should adhere to TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) and PROGRESS (Prognosis Research Strategy) guidelines, incorporating appropriate sample size calculations, internal and external validation, calibration assessment, and prospective testing in independent cohorts to establish clinical utility before implementation [28–30].

Conclusion:

Rising HRCi preceded clinical NEC diagnosis in preterm neonates, with significant elevation in the three days before clinical recognition. These findings support HRCi as an early physiologic marker of autonomic dysregulation associated with NEC pathogenesis, establishing temporal plausibility for its use in early detection. However, modest predictive performance (AUC=0.66) in our within-patient analysis, lack of non-NEC control group, and inability to account for concurrent sepsis and time-varying confounders limit immediate clinical applicability.

Prospective validation studies are needed and should: (1) include matched NEC-negative controls to assess specificity and false-positive rates, (2) systematically capture concurrent sepsis evaluations, feeding practices, medication exposures, and hemodynamic data, (3) employ appropriate statistical methods (mixed effects models or GEE) to account for clustered data, (4) evaluate HRCi as part of a multimodal risk assessment combining physiologic monitoring with biomarkers and imaging, and (5) adhere to TRIPOD/PROGRESS guidelines with internal and external validation before clinical implementation. Until such validation is completed, HRCi should be considered as part of a comprehensive clinical assessment rather than a standalone predictive tool for NEC.