Abstract
Background
Epidural labor analgesia is associated with intrapartum maternal fever, termed epidural‐related maternal fever (ERMF). Reliable risk stratification tools to support intrapartum monitoring and management are limited. We aimed to develop and internally validate a multivariable prediction model for ERMF among parturients receiving programmed intermittent epidural bolus (PIEB).
Methods
We conducted a single‐center retrospective cohort study of term (37–41 weeks), singleton, cephalic pregnancies with vaginal delivery who received PIEB between January and December 2024. ERMF was defined as axillary temperature ≥38.0°C confirmed by two measurements 30 min apart after PIEB initiation. Candidate predictors were prespecified based on clinical plausibility. Univariable and multivariable logistic regression were performed with multicollinearity assessed by variance inflation factor. Model discrimination was evaluated by area under the receiver operating characteristic curve (AUC) and precision–recall AUC (PR‐AUC). Internal validation used 1000 bootstrap resamples. Calibration was assessed by bootstrap‐corrected calibration curves and Brier score. Clinical utility was examined using decision curve analysis (DCA).
Results
Among 2800 women, 308 (11.0%) developed ERMF. Independent predictors included longer first‐stage labor duration, primiparity, artificial rupture of membranes, amniotic fluid contamination ≥grade 2, and higher cumulative oxytocin dose. The model achieved an AUC of 0.75, PR‐AUC of 0.254, and Brier score of 0.090; bootstrap validation showed stable performance. DCA suggested net benefit over treat‐all/treat‐none strategies across a wide range of threshold probabilities.
Conclusion
A multivariable logistic regression model using routinely available intrapartum variables demonstrated stable discrimination, calibration, and potential clinical utility for ERMF risk assessment and bedside implementation.
Keywords: epidural analgesia, epidural‐related maternal fever, labor analgesia, logistic regression, maternal fever, obstetric anesthesia, risk prediction model
1. BACKGROUND
Epidural labor analgesia is widely regarded as the most effective and safest method for obstetric analgesia and is considered the “gold standard” for pain relief during labor. 1 Accumulating evidence suggests a significant association between epidural labor analgesia and intrapartum maternal fever, a phenomenon known as epidural‐related maternal fever (ERMF). 2 Reported ERMF incidence varies substantially; approximately 10%–25% of women receiving epidural labor analgesia develop fever, 3 , 4 , 5 whereas the fever rate among women without epidural analgesia is only 0.8%–2.4%. 6 , 7 , 8 , 9 This variability might be partly attributable to differences in study design and fever definitions. Currently, no safe and effective strategy has been established to prevent ERMF. Because intrapartum fever, regardless of etiology, is associated with adverse maternal and neonatal outcomes, women who develop intrapartum fever should undergo standardized evaluation and management (including differentiation of isolated maternal fever from suspected/confirmed intra‐amniotic infection) whether or not epidural labor analgesia is used, to minimize unnecessary interventions while ensuring maternal–neonatal safety. 3
The pathophysiology of ERMF remains incompletely understood. Two main hypotheses have been proposed: (i) a sterile inflammatory response, in which local anesthetics (e.g., bupivacaine) might inhibit caspase‐1 activity and reduce the release of antipyretic cytokines (e.g., interleukin‐1 receptor antagonist) from peripheral leukocytes, thereby increasing body temperature 10 , 11 , 12 ; and (ii) impaired thermoregulation, whereby neuraxial blockade affects sweating and vasomotor responses, limiting heat dissipation and leading to temperature elevation. 13 , 14 , 15 These mechanisms might act synergistically to contribute to intrapartum fever.
Prior studies have suggested that prolonged labor, primiparity, oxytocin use, premature rupture of membranes, and greater gestational age might be associated with ERMF; however, their independent effects remain uncertain, and systematic risk prediction models are still lacking. 8 , 9 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 If high‐risk women could be identified early during labor, more intensive temperature surveillance and targeted preventive measures could be implemented.
Therefore, this study aimed to identify key risk factors for ERMF through a retrospective analysis of a large cohort and to develop and internally validate a clinically practical prediction model, providing a simple and operable risk assessment tool to optimize safety management for epidural labor analgesia.
2. MATERIALS AND METHODS
This was a single‐center retrospective cohort study conducted at Hubei Maternal and Child Health Hospital, a tertiary Grade A specialized women's and children's hospital in Wuhan, China. The study period spanned from January 1, 2024, to December 31, 2024. The analysis was restricted to women with term (37–41 weeks), singleton, cephalic pregnancies who underwent vaginal delivery and received programmed intermittent epidural bolus (PIEB) during labor. A total of 3038 women who underwent vaginal delivery during the study period were initially screened. Among them, 2869 received epidural labor analgesia. After applying predefined inclusion and exclusion criteria (detailed below), 2800 women were included in the final analytic cohort. The study protocol was approved by the institutional ethics committee (Approval No. 2025‐226‐01). Written informed consent was waived due to the retrospective nature of the study.
2.1. Study population
Inclusion criteria were women aged 18–50 years; singleton, cephalic, term pregnancy (gestational age ≥37 and ≤41 weeks); vaginal delivery; live fetus; voluntary choice and receipt of PIEB meeting clinical indications; and American Society of Anesthesiologists physical status I–II. Exclusion criteria were severe preeclampsia/HELLP (hemolysis, elevated liver enzymes and low platelets) syndrome/severe cardiopulmonary disease; known or clinically suspected infectious disease, including chorioamnionitis (because placental histopathology was not routinely performed for all cases in this retrospective cohort, any case documented in the medical record as “clinically suspected” or “diagnosed” chorioamnionitis was excluded to minimize confounding by infection) and other prenatally diagnosed infections (e.g., urinary or respiratory tract infection); use of antibiotics, systemic corticosteroids, or non‐steroidal anti‐inflammatory drugs of analgesia only after full cervical dilation (10 cm); baseline temperature ≥37.5°C before PIEB; and incomplete electronic medical records preventing retrieval of key labor‐related variables.
2.2. Epidural labor analgesia
All epidural catheterizations were performed by attending‐level or senior anesthesiologists. Epidural placement was conducted at the L2–L3 or L3–L4 interspace, with catheter insertion depth of 4–5 cm into the epidural space. The analgesic solution consisted of 0.08% ropivacaine combined with sufentanil 0.5 μg/mL and was administered using a PIEB regimen. Prior to initiation, a 3 mL test dose of 1% lidocaine was administered. Initial epidural dosing included a 10 mL loading dose. The default background infusion rate was set at 1 mL/h to maintain catheter patency. PIEB parameters were configured as follows: bolus dose 12 mL, bolus interval 60 min, patient‐controlled bolus 9.9 mL, lockout interval 15 min, and maximum hourly dose 35 mL. Each infusion reservoir contained 100 mL of analgesic solution and was replaced as needed. Dose adjustments were performed according to a standardized institutional protocol based on maternal pain score (visual analogue scale [VAS] ≥4), inadequate analgesia, or excessive motor block. Adjustment included modification of bolus volume (8–15 mL) and/or bolus interval (50–60 min) at the discretion of the attending anesthesiologist. During labor analgesia, anesthesiologists and midwives jointly monitored maternal blood pressure, heart rate, temperature, and fetal heart rate.
2.3. Data sources and variable definitions
2.3.1. Cohort flow
Of 2869 women who received PIEB, 69 were excluded due to incomplete electronic medical records or missing key labor‐related variables required for model construction (e.g., labor stage duration, oxytocin exposure, and membrane rupture status), yielding a final analytic cohort of 2800 women.
2.3.2. Data extraction
Clinical data were retrospectively extracted from the hospital electronic medical record system and verified through structured manual data quality checks. Extracted data included maternal demographics, peripartum variables, laboratory indices, and neonatal outcomes.
2.3.3. Outcome definition
Epidural‐related maternal fever was defined as axillary temperature ≥38.0°C occurring after initiation of PIEB and confirmed by two consecutive measurements 30 min apart. Only post‐PIEB temperature elevations were considered. Temperature was first measured upon admission to the labor ward and routinely recorded every 2 h; if ≥37.3°C, monitoring increased to every 30 min until delivery. Measurements were performed and documented by midwives in the electronic labor ward system as part of routine care. Labor room temperature was maintained at 24–26°C. Group assignment (ERMF vs. non‐ERMF) was determined retrospectively in 2025 based on 2024 temperature records. Treating clinicians were unaware of study grouping.
2.3.4. Predictor definitions
Baseline variables included maternal age, height, weight, gestational age, gravidity, and parity (primiparity defined as parity = 1). Pre‐analgesia heart rate, fetal heart rate, VAS score, and fibrinogen (within 24 h before delivery) were also recorded.
Intrapartum variables included labor stage durations, total labor duration, epidural analgesia duration, pre‐analgesia uterine contraction duration and interval, cervical dilation rate, oxytocin use and cumulative dose, membrane rupture status (premature rupture of membranes [PROM] or artificial rupture of membranes [ARM]), amniotic fluid contamination (≥grade 2 defined as moderate‐to‐heavy meconium‐stained fluid), and postpartum length of stay.
The first stage of labor was defined as the interval from regular contractions with progressive cervical dilation to complete dilation (10 cm). Labor durations were retrieved directly from the electronic medical record system, where they are automatically generated based on clinically documented time points entered by midwives. Cumulative oxytocin dose was calculated as the total administered international units during labor based on infusion pump records. Neonatal outcomes included birth weight and 1‐ and 5‐min Apgar scores (≤7 recorded). Variables were classified as baseline (pre‐analgesia) or dynamic (intrapartum).
2.3.5. Missing data
Missingness was minimal. A small proportion of baseline variables (height, weight, body mass index (BMI), Bishop score, and length of stay) had <3% missing values. All key predictors and the primary outcome were complete. Missing baseline values were imputed using K‐nearest neighbor imputation (k = 5) with z‐score standardization of continuous variables. Given the low level of missingness, multiple imputation was not performed.
2.4. Statistical analysis
Normality of continuous variables was assessed using the Shapiro–Wilk test. Normally distributed continuous variables are presented as mean ± standard deviation (SD) and compared using Student's t‐test; non‐normally distributed continuous variables are presented as median (interquartile range, IQR) and compared using the Mann–Whitney U‐test. Categorical variables are presented as number (percentage) and compared using the χ2‐test or Fisher's exact test, as appropriate.
Univariable logistic regression was first performed to evaluate associations between candidate predictors and ERMF. Variables with P < 0.10 in univariable analyses or deemed clinically important were entered into a multivariable logistic regression model. Before model construction, multicollinearity was assessed using the variance inflation factor (VIF); VIF <5 was considered indicative of no substantial collinearity, and no significant multicollinearity was identified. Multivariable logistic regression was then used to identify independent predictors of ERMF, reporting adjusted odds ratios (aORs) with 95% confidence intervals (CIs). Discrimination was evaluated using the area under the receiver operating characteristic curve (AUC). Internal validation was conducted using 1000 bootstrap resamples to estimate optimism and derive an optimism‐corrected AUC. Calibration was assessed using bootstrap‐corrected calibration curves and the Brier score.
Clinical utility was evaluated with decision curve analysis (DCA) by comparing net benefit across a range of threshold probabilities against “treat‐all” and “treat‐none” strategies. All tests were two‐sided, and P < 0.05 was considered statistically significant. All analyses were performed using Python 3.11 (pandas, statsmodels, scikit‐learn, matplotlib).
3. RESULTS
3.1. Baseline characteristics
A total of 2800 women receiving epidural labor analgesia were included (Table 1), among whom 308 (11.0%) developed ERMF. Baseline characteristics by ERMF status are shown in Table 1. Compared with women without ERMF, those with ERMF were slightly younger (29.39 ± 2.98 vs. 30.10 ± 3.34 years, P < 0.001) and had a slightly higher BMI (26.7 ± 2.9 vs. 26.2 ± 3.1 kg/m2, P = 0.003). Women with ERMF had lower gravidity and a markedly higher proportion of primiparity (92.5% vs. 69.8%, P < 0.001). Oxytocin use was more frequent in the ERMF group (88.6% vs. 70.7%, P < 0.001), and both cumulative oxytocin dose and exposure time were higher. ARM and amniotic fluid contamination ≥grade 2 were more common in the ERMF group (both P < 0.001), whereas PROM did not differ significantly between groups. The ERMF group had longer durations of the first and second stages of labor and longer total labor duration, as well as longer epidural analgesia duration (all P < 0.001). Cervical dilation rate was lower in the ERMF group (P < 0.001). Admission Bishop score, baseline pulse, fibrinogen level, and baseline VAS score did not differ significantly.
TABLE 1.
Baseline characteristics of parturients and neonates stratified by epidural‐related maternal fever.
| Characteristic | Total (n = 2800) | Fever (n = 308) | Non‐fever (n = 2492) | P‐value |
|---|---|---|---|---|
| Maternal age (years) | 30.0 ± 3.3 | 29.39 ± 2.98 | 30.10 ± 3.34 | <0.001 |
| Maternal height (cm) | 161.90 ± 5.0 | 161.36 ± 4.24 | 161.94 ± 4.8 | 0.045 |
| Maternal weight (kg) | 69.0 ± 8.7 | 69.68 ± 8.5 | 68.9 ± 8.8 | 0.104 |
| Body mass index (BMI, kg/m2) | 26.3 ± 3.1 | 26.7 ± 2.9 | 26.2 ± 3.1 | 0.003 |
| Gestational week (weeks) | 39 (38, 40) | 39.00 (39, 40) | 39.00 (38, 40) | 0.051 |
| Gravidity [median (IQR)] | 1 (1, 2) | 1 (1, 1) | 1 (1, 2) | < 0.001 |
| Primiparity (n %) | 2024 (72.3%) | 285 (92.5%) | 1739 (69.8%) | <0.001 |
| Use of oxytocin (n %) | 2035 (72.7%) | 273 (88.6%) | 1762 (70.7%) | <0.001 |
| Cumulative oxytocin dose (IU) | 3.04 (0.00–10.93) | 7.03 (1.43–15.06) | 2.65 (0.00–10.23) | < 0.001 |
| Cumulative oxytocin exposure (min) | 166.50 (0.00–410.00) | 283.50 (94.75–540.00) | 150.00 (0.00–390.00) | <0.001 |
| Artificial rupture of membranes (ARM) | 855 (30.5%) | 157 (51.0%) | 698 (28.0%) | <0.001 |
| Premature rupture of membranes (PROM, n %) | 737 (26.3%) | 71 (23.1%) | 666 (26.7%) | 0.189 |
| Amniotic fluid contamination ≥grade 2 | 538 (19.2%) | 93 (30.2%) | 445 (17.9%) | <0.001 |
| First stage of labor (min) | 477.13 ± 194.78 | 590.47 ± 175.03 | 463.12 ± 192.54 | <0.001 |
| Second stage of labor (min) | 28.00 (19.00–40.00) | 33.00 (24.00–51.25) | 27.00 (18.00–39.00) | <0.001 |
| Third stage of labor (min) | 6.00 (4.00–9.00) | 6.00 (5.00–10.00) | 6.00 (4.00–9.00) | 0.013 |
| Total labor duration (min) | 508.00 (364.75–650.00) | 625.00 (520.00–750.00) | 486.00 (350.00–635.00) | <0.001 |
| Duration of epidural analgesia (min) | 425.66 ± 168.07 | 548.31 ± 131.53 | 410.50 ± 165.88 | <0.001 |
| Cervical dilation rate (cm/h) | 2.00 (1.25–3.00) | 1.40 (1.04–2.00) | 2.00 (1.33–3.28) | <0.001 |
| Baseline bishop score | 6.12 ± 0.41 | 6.15 ± 0.42 | 6.12 ± 0.40 | 0.185 |
| Baseline maternal heart rate (beats/min) | 86.00 (80.00–88.00) | 86.00 (80.00–89.00) | 86.00 (80.00–88.00) | 0.342 |
| Fibrinogen (FIB, g/L) | 3.63 ± 0.86 | 3.60 ± 0.92 | 3.63 ± 0.85 | 0.535 |
| Baseline VAS score | 8.07 ± 1.16 | 8.04 ± 1.10 | 8.08 ± 1.17 | 0.554 |
| Apgar score at 1 min | 10.00 (9.00–10.00) | 9.00 (9.00–10.00) | 10.00 (9.00–10.00) | <0.001 |
| Apgar score at 5 min | 10.00 (10.00–10.00) | 10.00 (10.00–10.00) | 10.00 (10.00–10.00) | 0.003 |
| Apgar score ≤7 at 1 min (n %) | 15 (0.5%) | 3 (1.0%) | 12 (0.5%) | 0.208 |
| Apgar score ≤7 at 5 min (n %) | 1 (0.0%) | 0 (0.0%) | 1 (0.0%) | 1.000 |
| Fetal weight (g) | 3216.3 ± 380 | 3263.5 ± 389 | 3210.5 ± 379 | 0.025 |
| Postpartum hospital stay (days) | 2.77 (2.57–3.00) | 2.75 (2.58–3.14) | 2.78 (2.56–2.99) | 0.317 |
Note: Data are presented as mean ± standard deviation (SD) for normally distributed continuous variables, median (interquartile range, IQR) for non‐normally distributed continuous variables, and number (percentage) for categorical variables. Group comparisons were performed using Student's t‐test for normally distributed continuous variables, the Mann–Whitney U‐test for non‐normally distributed continuous variables, and the χ2‐test or Fisher's exact test for categorical variables, as appropriate. Primiparity was defined as parity equal to 1, whereas multiparity was defined as parity ≥2. Amniotic fluid contamination ≥grade 2 indicates moderate to severe amniotic fluid contamination.
Abbreviations: ARM, artificial rupture of membranes; BMI, body mass index; ERMF, epidural‐related maternal fever; FIB, fibrinogen; PROM, premature rupture of membranes; VAS, visual analogue scale.
Regarding neonatal outcomes, the ERMF group had slightly lower 1‐min Apgar scores (P < 0.001), whereas 5‐min Apgar scores were comparable. The proportions of low Apgar (≤7) and neonatal asphyxia did not differ significantly. Birth weight was slightly higher in the ERMF group (P = 0.025), and postpartum length of stay was similar between groups.
3.2. Risk factor analysis for epidural‐related maternal fever
Univariable and multivariable logistic regression analyses were used to identify risk factors associated with ERMF (Table 2). In univariable analyses, several peripartum factors were significantly associated with ERMF. Longer first‐stage labor duration (odds ratio [OR] = 1.003, 95% CI 1.003–1.004, P < 0.001), greater cumulative oxytocin dose (OR = 1.049, 95% CI 1.036–1.062, P < 0.001), ARM (OR = 2.672, 95% CI 2.102–3.397, P < 0.001), primiparity (OR = 5.366, 95% CI 3.479–8.276, P < 0.001), and amniotic fluid contamination ≥grade 2 (OR = 1.990, 95% CI 1.528–2.591, P < 0.001) were each associated with increased ERMF risk. Gravidity was inversely associated with ERMF (OR = 0.570, 95% CI 0.475–0.684, P < 0.001). Gestational age (P = 0.103) and PROM (P = 0.168) were not statistically significant in univariable analyses.
TABLE 2.
Risk factors associated with epidural‐related maternal fever (ERMF).
| Characteristic | N | Event N | Univariable | Multivariable | ||
|---|---|---|---|---|---|---|
| OR (95% CI) | p | OR (95% CI) | p | |||
| Gestational week (weeks) | 2800 | 308 | 1.089 (0.983–1.207) | 0.103 | 0.979 (0.879–1.091) | 0.704 |
| Cumulative oxytocin dose (IU) | 2800 | 308 | 1.049 (1.036–1.062) | <0.001 | 1.021 (1.007–1.036) | 0.004 |
| First stage of labor (min) | 2800 | 308 | 1.003 (1.003–1.004) | <0.001 | 1.002 (1.001–1.003) | <0.001 |
| Primiparity | ||||||
| No | 776 | 23 | — | — | ||
| Yes | 2024 | 285 | 5.366 (3.479–8.276) | <0.001 | 3.339 (1.943–5.738) | <0.001 |
| PROM | ||||||
| No | 2063 | 237 | — | — | ||
| Yes | 737 | 71 | 0.821 (0.621–1.086) | 0.168 | 1.205 (0.858–1.691) | 0.281 |
| Amniotic fluid contamination ≥grade 2 | ||||||
| No | 2262 | 215 | — | — | ||
| Yes | 538 | 93 | 1.99 (1.528–2.591) | <0.001 | 1.840 (1.388–2.440) | <0.001 |
| ARM | ||||||
| No | 1945 | 151 | — | — | ||
| Yes | 855 | 157 | 2.672 (2.102–3.397) | <0.001 | 2.345 (1.770–3.107) | <0.001 |
| Gravidity | 2800 | 308 | 0.57 (0.475–0.684) | <0.001 | 0.964 (0.777–1.197) | 0.742 |
| Maternal age (years) | 2800 | 308 | 0.934 (0.899–0.97) | <0.001 | — | |
| Maternal height (cm) | 2800 | 308 | 0.98 (0.958–1.001) | 0.064 | — | |
| Maternal weight (kg) | 2800 | 308 | 1.0 (0.998–1.001) | 0.806 | — | |
| Baseline maternal heart rate (beats/min) | 2800 | 308 | 1.001 (0.996–1.008) | 0.628 | — | |
| Baseline Bishop score | 2800 | 308 | 1.22 (0.919–1.621) | 0.169 | — | |
| Baseline VAS score | 2800 | 308 | 0.971 (0.877–1.075) | 0.571 | — | |
| Total duration of labor (min) | 2800 | 308 | 1.003 (1.002–1.003) | <0.001 | — | |
| Duration of epidural analgesia (min) | 2800 | 308 | 1.005 (1.004–1.005) | <0.001 | — | |
| Uterine contraction duration (s) | 2800 | 308 | 0.975 (0.928–1.024) | 0.314 | — | |
| Uterine contraction interval (min) | 2800 | 308 | 1.149 (1.013–1.302) | 0.031 | — | |
| Uterine contraction intensity (mmHg) | 2800 | 308 | 0.997 (0.987–1.007) | 0.561 | — | |
| Fetal weight (g) | 2800 | 308 | 1.0 (1.0–1.001) | 0.021 | — | |
| Fetal heart rate (beats/min) | 2800 | 308 | 1.007 (0.993–1.022) | 0.34 | — | |
| Use of oxytocin | ||||||
| No | 765 | 35 | — | — | ||
| Yes | 2035 | 273 | 3.232 (2.249–4.643) | <0.001 | — | |
| Cumulative oxytocin exposure (min) | 2800 | 308 | 1.002 (1.001–1.002) | <0.001 | — | |
| Fibrinogen (g/L) | 2800 | 308 | 0.954 (0.831–1.096) | 0.51 | — | |
Note: Univariable logistic regression was performed for all candidate variables. Variables with a p‐value <0.10 in univariable analysis and/or those considered clinically relevant were included in the multivariable logistic regression model. Adjusted odds ratios (aORs) were estimated from the multivariable model including primiparity, artificial rupture of membranes, cumulative oxytocin dose, amniotic fluid contamination ≥ grade 2, duration of the first stage of labor, gestational age, and premature rupture of membranes. Continuous variables were analyzed per unit increase as indicated. Odds ratios are presented with 95% confidence intervals. N represents the number of participants included in each analysis, and events represent the number of women who developed maternal fever.
Abbreviations: ARM, artificial rupture of membranes; PROM, premature rupture of membranes; VAS, visual analogue scale.
After excluding collinear variables, predictors with P < 0.10 in univariable analyses and clinically important variables were entered into the multivariable model. Longer first‐stage labor duration (aOR = 1.002, 95% CI 1.001–1.003, P < 0.001), greater cumulative oxytocin dose (aOR = 1.021, 95% CI 1.007–1.036, P = 0.004), primiparity (aOR = 3.339, 95% CI 1.943–5.738, P < 0.001), ARM (aOR = 2.345, 95% CI 1.770–3.107, P < 0.001), and amniotic fluid contamination ≥grade 2 (aOR = 1.840, 95% CI 1.388–2.440, P < 0.001) remained independent predictors. In contrast, gestational age (aOR = 0.979, 95% CI 0.879–1.091, P = 0.704), PROM (aOR = 1.205, 95% CI 0.858–1.691, P = 0.281), and gravidity (aOR = 0.964, 95% CI 0.777–1.197, P = 0.742) were not independently associated with ERMF in the multivariable model.
3.3. Discrimination and calibration of the epidural‐related maternal fever prediction model
Discrimination was evaluated using the receiver operating characteristic curve (Figure 1a). The model achieved an AUC of 0.75 in the overall cohort, indicating moderate‐to‐good discrimination between women with and without ERMF.
FIGURE 1.
Discrimination performance of the epidural‐related maternal fever (ERMF) prediction model. (a) Receiver operating characteristic (ROC) curve demonstrating the discriminative ability of the model, with an area under the curve (AUC) of 0.75, indicating moderate discrimination. (b) Precision–recall (PR) curve illustrating model performance under class imbalance, with an AUC of 0.25. The PR curve highlights the trade‐off between precision and recall across different decision thresholds and provides complementary information to the ROC curve in the presence of a relatively low event rate.
Given the relatively low event rate (approximately 11%), a precision–recall (PR) curve was additionally generated to assess performance under class imbalance (Figure 1b). The PR‐AUC was 0.254, exceeding the baseline level corresponding to the event rate (11%), suggesting added value in identifying high‐risk individuals.
Calibration was assessed using a calibration plot (Figure 2b). Overall, predicted probabilities agreed well with observed event probabilities: the logistic calibration line was close to the ideal reference line, and the nonparametric smooth curve showed no pronounced systematic deviation, suggesting no substantial overfitting or underfitting. Calibration statistics further showed a calibration intercept near 0 and slope near 1, with a Brier score of 0.090, indicating good overall calibration. Spiegelhalter's calibration test (S:P = 0.814) also supported good agreement between predicted and observed risks (Tables S1–S3).
FIGURE 2.
(a) Decision curve analysis (DCA) illustrating the clinical utility of the model. The y‐axis represents the number of women classified as high risk per 1000 individuals. The upper x‐axis indicates the threshold probability, and the lower x‐axis shows the corresponding cost–benefit ratio. The solid red line represents the number of women identified as high risk by the model, while the dashed blue line indicates the number of events among those classified as high risk. (b) Calibration plot comparing predicted and observed probabilities of ERMF. The gray diagonal line indicates perfect calibration. The solid black line represents logistic calibration, and the dashed black line represents nonparametric calibration. Triangles indicate grouped observations. Closer alignment of the calibration curves with the ideal line indicates better agreement between predicted and observed risks.
3.4. Clinical utility (decision curve analysis)
Decision curve analysis demonstrated that across a wide range of threshold probabilities (approximately 0.01–0.60), the model provided greater net benefit than “treat‐all” and “treat‐none” strategies (Figure 2a), suggesting potential clinical value for risk stratification and management optimization.
Based on the predictors identified from multivariable logistic regression (first‐stage labor duration, primiparity, ARM, amniotic fluid contamination ≥grade 2, cumulative oxytocin dose, gestational age, gravidity, and PROM), we constructed an ERMF risk prediction model and presented it as a nomogram for bedside application (Figure 3).
FIGURE 3.
Nomogram for predicting epidural‐related maternal fever (ERMF) among parturients receiving epidural analgesia. The nomogram was developed based on a multivariable logistic regression model incorporating duration of the first stage of labor, premature rupture of membranes (PROM), artificial rupture of membranes (ARM), primiparity, gestational age, amniotic fluid contamination ≥grade 2, gravidity, and cumulative oxytocin dose. For clinical use, the value of each predictor corresponds to a specific point score on its respective axis. The total score is obtained by summing the individual points, which can then be translated into an estimated probability of intrapartum maternal fever using the risk scale at the bottom of the nomogram.
4. DISCUSSION
In this single‐center retrospective cohort study, we developed and internally validated a dynamic multivariable prediction model for ERMF during epidural labor analgesia and implemented the model as a bedside nomogram. Eight candidate predictors were prespecified based on prior literature and clinical plausibility, including gestational age, cumulative oxytocin dose, first‐stage labor duration, primiparity, PROM, amniotic fluid contamination ≥grade 2, ARM, and gravidity, rather than being selected solely through data‐driven procedures. Because cervical dilation rate is strongly correlated with labor duration and might introduce collinearity, we prioritized labor duration, which is more clinically interpretable and more consistently documented. Multivariable analysis identified prolonged first‐stage labor, primiparity, ARM, amniotic fluid contamination ≥grade 2, and cumulative oxytocin dose as independent predictors, whereas PROM, gestational age, and gravidity were not independently significant after adjustment, suggesting that their effects might be mediated through labor prolongation and/or increased intervention intensity. The model showed stable discrimination (AUC ≈ 0.75; optimism‐corrected AUC ≈ 0.739), good calibration (Brier ≈0.09), and favorable clinical net benefit across a broad range of thresholds on DCA.
In recent years, several ERMF prediction models have been proposed. Niu et al. developed a nomogram in 1692 term vaginal deliveries with epidural analgesia, incorporating PROM, first‐stage duration, vaginal examination frequency, and inflammatory markers (IL‐6, SII, and NLR), with good discrimination. 26 However, reliance on inflammatory biomarkers that are not routinely measured in many labor units might limit real‐time bedside applicability. Liu et al. 27 proposed a two‐stage approach (“baseline” vs. “whole‐process”) emphasizing that risk should be dynamically updated as labor information accumulates. Guo et al. used multiple machine learning algorithms and reported discrimination, calibration, and DCA in a comprehensive framework. 28 While these studies are methodologically informative, some models are complex or depend heavily on laboratory indices, potentially limiting interpretability and dissemination. Our model offers several practical strengths: (i) all predictors are readily available at the bedside without requiring non‐routine biomarkers such as IL‐6/SII; (ii) oxytocin exposure is quantified using cumulative dose rather than a simple binary indicator, better reflecting clinical exposure intensity; and (iii) we report discrimination, calibration, and net benefit concurrently, avoiding over‐reliance on AUC alone. As with most published models, external validation remains necessary; nevertheless, bootstrap internal validation and systematic reporting of calibration and net benefit were used to mitigate overfitting concerns.
First‐stage labor duration was among the most robust predictors in our model. Prior studies consistently show that prolonged labor is associated with higher intrapartum fever risk, potentially due to increased maternal metabolic demand, heat generation from sustained uterine contractions, and altered thermoregulation associated with prolonged neuraxial analgesia. 18 , 22 , 29 Earlier work also suggests that labor duration, temperature measurement site, and temperature trajectories influence the onset and magnitude of intrapartum fever. 30
Artificial rupture of membranes and amniotic fluid contamination ≥grade 2 were also independently associated with ERMF. ARM might increase intrauterine exposure and alter the intrauterine environment, while meconium‐stained or contaminated amniotic fluid might indicate heightened inflammatory activity; together these factors might contribute to inflammatory temperature elevation even when no clear clinical infection is identified. 31 Cumulative oxytocin dose showed an independent positive association with ERMF, consistent with prior reports that oxytocin intensifies uterine activity, might prolong labor in dysfunctional patterns, and increases maternal metabolic load. 32 , 33 Using the cumulative dose as a continuous measure might better capture a dose–response relationship than a dichotomous “use vs. non‐use” variable.
We position this model as a dynamic intrapartum risk assessment tool. As labor progresses and interventions accumulate, predicted ERMF probability can be updated in real time to support risk stratification and to adjust the intensity of monitoring and evaluation rather than serving as a one‐time antepartum prediction. This concept aligns with previously proposed dynamic or staged modeling strategies. 27 For women in higher‐risk ranges, the model output might prompt more frequent temperature and vital sign surveillance, earlier assessment for infectious versus non‐infectious etiologies, and earlier multidisciplinary communication among obstetric, anesthetic, and neonatal teams.
Importantly, the model provides individualized ERMF probability but does not support using a fixed threshold to determine antibiotic administration. According to the American College of Obstetricians and Gynecologists, intrapartum fever should be classified as isolated maternal fever versus suspected or confirmed intra‐amniotic infection, and management should integrate clinical findings rather than rely solely on temperature criteria. 34 In routine practice, intrapartum maternal fever, regardless of etiology, prompts intensified monitoring and clinical evaluation, with antibiotics reserved for cases where infection is suspected. In the absence of infection and with reassuring fetal status, management is primarily supportive, including increased oral hydration, physical cooling measures, and antipyretic medication as needed. Although ERMF is often non‐infectious and self‐limited, previous studies have reported that approximately 90% of ERMF cases resolve within a few hours after delivery, 35 suggesting a predominantly transient course. Nevertheless, its occurrence frequently leads to increased maternal investigations and neonatal sepsis evaluation, potentially resulting in greater antibiotic exposure. 36 Current evidence suggests that prophylactic acetaminophen does not effectively prevent epidural‐related fever, 37 and no single preventive strategy has demonstrated consistent benefit. 38 Therefore, our model is intended to support risk stratification and process optimization, enhancing surveillance in high‐risk women while helping to avoid unnecessary interventions.
This study has limitations. As a retrospective single‐center analysis, residual confounding cannot be fully excluded, including subclinical infection and unmeasured factors, and external validation is lacking. In addition, we did not include a non‐epidural control group; therefore, we cannot disentangle the relative contributions of ERMF versus fever associated with dysfunctional labor. Future work should externally validate the model in multicenter cohorts and explore more granular time‐varying information or dynamic modeling approaches to improve generalizability and bedside utility. 39
5. CONCLUSION
We developed and internally validated an ERMF prediction model based on routinely available peripartum variables. The model demonstrated stable discrimination, good calibration, and potential clinical utility, and it can be implemented as a bedside nomogram to support intrapartum risk assessment and refined peripartum management for early identification of ERMF.
AUTHOR CONTRIBUTIONS
Conception of study design: Junyang Ma, Jing Cheng. Study design: Junyang Ma, Ping Lu. Data analysis: Junyang Ma, Meiqi Sun. Drafting the manuscript: Junyang Ma. Critical revision for important intellectual content: all authors.
CONFLICT OF INTEREST STATEMENT
The authors have no competing interests to declare.
ETHICS STATEMENT
This study was approved by the Ethics Committee of Hubei Maternal and Child Health Hospital (Approval No. 2025‐226‐01), with a waiver of informed consent.
CONSENT TO PARTICIPATE
The requirement for informed consent was waived by the Institutional Review Board due to the retrospective and anonymized nature of the study.
Supporting information
Data S1. STROBE Statement—Checklist of items that should be included in reports of cohort studies.
Table S1. Standardized coefficients and adjusted odds ratios (ORs) of predictors in the ERMF model.
Table S2. Overall performance metrics of the ERMF prediction model.
Table S3. Assessment of multicollinearity among predictors included in the multivariable ERMF prediction model.
DATA AVAILABILITY STATEMENT
Due to restrictions imposed by the hospital ethics committee and potential concerns regarding patient privacy, the datasets generated and/or analyzed during the current study are not publicly available. However, data might be made available from the corresponding author upon reasonable request and with appropriate ethical approval.
REFERENCES
- 1. Anim‐Somuah M, Smyth RM, Cyna AM, Cuthbert A. Epidural versus non‐epidural or no analgesia for pain management in labour. Cochrane Database Syst Rev. 2018;5(5):Cd000331. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Fusi L, Steer PJ, Maresh MJ, Beard RW. Maternal pyrexia associated with the use of epidural analgesia in labour. Lancet. 1989;1(8649):1250‐1252. [DOI] [PubMed] [Google Scholar]
- 3. Patel S, Ciechanowicz S, Blumenfeld YJ, Sultan P. Epidural‐related maternal fever: incidence, pathophysiology, outcomes, and management. Am J Obstet Gynecol. 2023;228(5s):S1283‐S1304.e1281. [DOI] [PubMed] [Google Scholar]
- 4. Seiler FA, Scavone BM, Shahul S, Arnolds DE. Maternal fever associated with continuous spinal versus epidural labor analgesia: a single‐center retrospective study. Anesth Analg. 2022;135(6):1153‐1158. [DOI] [PubMed] [Google Scholar]
- 5. Ward C, Caughey AB. Does the presence of epidural analgesia reduce the risk of neonatal sepsis in the setting of an intrapartum fever? s. 2022;35(11):2110‐2115. [DOI] [PubMed] [Google Scholar]
- 6. Agakidis C, Agakidou E, Philip Thomas S, Murthy P, John Lloyd D. Labor epidural analgesia is independent risk factor for neonatal pyrexia. J Matern Fetal Neonatal Med. 2011;24(9):1128‐1132. [DOI] [PubMed] [Google Scholar]
- 7. Greenwell EA, Wyshak G, Ringer SA, Johnson LC, Rivkin MJ, Lieberman E. Intrapartum temperature elevation, epidural use, and adverse outcome in term infants. Pediatrics. 2012;129(2):e447‐e454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Gross JB, Cohen AP, Lang JM, Frigoletto FD, Lieberman ES. Differences in systemic opioid use do not explain increased fever incidence in parturients receiving epidural analgesia. Anesthesiology. 2002;97(1):157‐161. [DOI] [PubMed] [Google Scholar]
- 9. Herbst A, Wølner‐Hanssen P, Ingemarsson I. Risk factors for fever in labor. Obstet Gynecol. 1995;86(5):790‐794. [DOI] [PubMed] [Google Scholar]
- 10. Del Arroyo AG, Sanchez J, Patel S, et al. Role of leucocyte caspase‐1 activity in epidural‐related maternal fever: a single‐centre, observational, mechanistic cohort study. Br J Anaesth. 2019;122(1):92‐102. [DOI] [PubMed] [Google Scholar]
- 11. Ackland GL, Van Duijvenboden S, Abbott TEF, Gutierrez Del Arroyo A, Wilson MJ, David AL. Interleukin‐1 receptor antagonist, mode of analgesia and risk of caesarean delivery after onset of labour: a mendelian randomisation analysis. Br J Anaesth. 2022;128(1):89‐97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Wohlrab P, Boehme S, Kaun C, et al. Ropivacaine activates multiple proapoptotic and inflammatory signaling pathways that might subsume to trigger epidural‐related maternal fever. Anesth Analg. 2020;130(2):321‐331. [DOI] [PubMed] [Google Scholar]
- 13. Horn EP, Schroeder F, Gottschalk A, et al. Active warming during cesarean delivery. Anesth Analg. 2002;94(2):409‐414. table of contents. [DOI] [PubMed] [Google Scholar]
- 14. Segal S. Labor epidural analgesia and maternal fever. Anesth Analg. 2010;111(6):1467‐1475. [DOI] [PubMed] [Google Scholar]
- 15. Kellogg DL Jr. In vivo mechanisms of cutaneous vasodilation and vasoconstriction in humans during thermoregulatory challenges. J Appl Physiol (1985). 2006;100(5):1709‐1718. [DOI] [PubMed] [Google Scholar]
- 16. Lieberman E, Lang JM, Frigoletto F Jr, Richardson DK, Ringer SA, Cohen A. Epidural analgesia, intrapartum fever, and neonatal sepsis evaluation. Pediatrics. 1997;99(3):415‐419. [DOI] [PubMed] [Google Scholar]
- 17. Mayer DC, Chescheir NC, Spielman FJ. Increased intrapartum antibiotic administration associated with epidural analgesia in labor. Am J Perinatol. 1997;14(2):83‐86. [DOI] [PubMed] [Google Scholar]
- 18. Reilly DR, Oppenheimer LW. Fever in term labour. J Obstet Gynaecol Can. 2005;27(3):218‐223. [DOI] [PubMed] [Google Scholar]
- 19. Törnell S, Ekéus C, Hultin M, Håkansson S, Thunberg J, Högberg U. Low Apgar score, neonatal encephalopathy and epidural analgesia during labour: a Swedish registry‐based study. Acta Anaesthesiol Scand. 2015;59(4):486‐495. [DOI] [PubMed] [Google Scholar]
- 20. Vinson DC, Thomas R, Kiser T. Association between epidural analgesia during labor and fever. J Fam Pract. 1993;36(6):617‐622. [PubMed] [Google Scholar]
- 21. Douma MR, Stienstra R, Middeldorp JM, Arbous MS, Dahan A. Differences in maternal temperature during labour with remifentanil patient‐controlled analgesia or epidural analgesia: a randomised controlled trial. Int J Obstet Anesth. 2015;24(4):313‐322. [DOI] [PubMed] [Google Scholar]
- 22. Philip J, Alexander JM, Sharma SK, Leveno KJ, McIntire DD, Wiley J. Epidural analgesia during labor and maternal fever. Anesthesiology. 1999;90(5):1271‐1275. [DOI] [PubMed] [Google Scholar]
- 23. Sharma SK, Alexander JM, Messick G, et al. Cesarean delivery: a randomized trial of epidural analgesia versus intravenous meperidine analgesia during labor in nulliparous women. Anesthesiology. 2002;96(3):546‐551. [DOI] [PubMed] [Google Scholar]
- 24. Goetzl L, Cohen A, Frigoletto F Jr, Lang JM, Lieberman E. Maternal epidural analgesia and rates of maternal antibiotic treatment in a low‐risk nulliparous population. J Perinatol. 2003;23(6):457‐461. [DOI] [PubMed] [Google Scholar]
- 25. Mantha VR, Vallejo MC, Ramesh V, Phelps AL, Ramanathan S. The incidence of maternal fever during labor is less with intermittent than with continuous epidural analgesia: a randomized controlled trial. Int J Obstet Anesth. 2008;17(2):123‐129. [DOI] [PubMed] [Google Scholar]
- 26. Niu DG, Wang L, Wei RY, et al. Development and validation of a nomogram for predicting intrapartum fever in parturients with epidural analgesia. Front Med (Lausanne). 2025;12:1668597. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Liu B, Ling L, Wei D, et al. Predictive models for intrapartum maternal fever: development and validation of pre‐analgesia and labor process indicators. Medicine (Baltimore). 2025;104(25):e42939. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Guo X, Zhang H, Mei H. Machine learning algorithms to predict epidural‐related maternal fever: a retrospective study. Front Pharmacol. 2025;16:1614770. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Goetzl L. Maternal fever in labor: etiologies, consequences, and clinical management. Am J Obstet Gynecol. 2023;228(5s):S1274‐s1282. [DOI] [PubMed] [Google Scholar]
- 30. Mercier FJ, Benhamou D. Hyperthermia related to epidural analgesia during labor. Int J Obstet Anesth. 1997;6(1):19‐24. [DOI] [PubMed] [Google Scholar]
- 31. ACOG clinical practice update: update on criteria for suspected diagnosis of intraamniotic infection. Obstet Gynecol. 2024;144(1):e17‐e19. [DOI] [PubMed] [Google Scholar]
- 32. Hermesch AC, Kernberg AS, Layoun VR, Caughey AB. Oxytocin: physiology, pharmacology, and clinical application for labor management. Am J Obstet Gynecol. 2024;230(3s):S729‐s739. [DOI] [PubMed] [Google Scholar]
- 33. Kim SH, MacIntyre DA, Firmino Da Silva M, et al. Oxytocin activates NF‐κB‐mediated inflammatory pathways in human gestational tissues. Mol Cell Endocrinol. 2015;403:64‐77. [DOI] [PubMed] [Google Scholar]
- 34. Committee opinion No. 712: intrapartum Management of Intraamniotic Infection. Obstet Gynecol. 2017;130(2):e95‐e101. [DOI] [PubMed] [Google Scholar]
- 35. Gonen R, Korobochka R, Degani S, Gaitini L. Association between epidural analgesia and intrapartum fever. Am J Perinatol. 2000;17(3):127‐130. [DOI] [PubMed] [Google Scholar]
- 36. Goetzl L, Cohen A, Frigoletto F Jr, Ringer SA, Lang JM, Lieberman E. Maternal epidural use and neonatal sepsis evaluation in afebrile mothers. Pediatrics. 2001;108(5):1099‐1102. [DOI] [PubMed] [Google Scholar]
- 37. Goetzl L, Rivers J, Evans T, et al. Prophylactic acetaminophen does not prevent epidural fever in nulliparous women: a double‐blind placebo‐controlled trial. J Perinatol. 2004;24(8):471‐475. [DOI] [PubMed] [Google Scholar]
- 38. Cartledge A, Hind D, Bradburn M, et al. Interventions for the prevention or treatment of epidural‐related maternal fever: a systematic review and meta‐analysis. Br J Anaesth. 2022;129(4):567‐580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. STROBE Statement—Checklist of items that should be included in reports of cohort studies.
Table S1. Standardized coefficients and adjusted odds ratios (ORs) of predictors in the ERMF model.
Table S2. Overall performance metrics of the ERMF prediction model.
Table S3. Assessment of multicollinearity among predictors included in the multivariable ERMF prediction model.
Data Availability Statement
Due to restrictions imposed by the hospital ethics committee and potential concerns regarding patient privacy, the datasets generated and/or analyzed during the current study are not publicly available. However, data might be made available from the corresponding author upon reasonable request and with appropriate ethical approval.