Prediction of intrapartum fever using continuously monitored vital signs and heart rate variability

Abstract

Objectives: Neonatal early onset sepsis (EOS), bacterial infection during the first seven days of life, is difficult to diagnose because presenting signs are non-specific, but early diagnosis before birth can direct life-saving treatment for mother and baby. Specifically, maternal fever during labor from placental infection is the strongest predictor of EOS. Alterations in maternal heart rate variability (HRV) may precede development of intrapartum fever, enabling incipient EOS detection. The objective of this work was to build a predictive model for intrapartum fever.

Methods: Continuously measured temperature, heart rate, and beat-to-beat RR intervals were obtained from wireless sensors on women (n = 141) in labor; traditional manual vital signs were taken every 3-6 hours. Validated measures of HRV were calculated in moving 5-minute windows of RR intervals: standard deviation of normal-to-normal intervals (SDNN) and root mean square of successive differences (RMSSD) between normal heartbeats.

Results: Fever (>38.0 °C) was detected by manual or continuous measurements in 48 women. Compared to afebrile mothers, average SDNN and RMSSD in febrile mothers decreased significantly (p < 0.001) at 2 and 3 hours before fever onset, respectively. This observed HRV divergence and raw recorded vitals were applied to a logistic regression model at various time horizons, up to 4-5 hours before fever onset. Model performance increased with decreasing time horizons, and a model built using continuous vital signs as input variables consistently outperformed a model built from episodic vital signs.

Conclusions: HRV-based predictive models could identify mothers at risk for fever and infants at risk for EOS, guiding maternal antibiotic prophylaxis and neonatal monitoring.

Keywords: Neonatal early onset sepsis; continuous vital signs monitoring; heart rate variability; intrapartum fever; logistic regression; predictive modeling.

Figure 1.

Data and modeling summary schematics. (a) Patient recruitment and data refinement flowchart. (b) Logistic regression (LR) model development. Two LR models were developed to classify intrapartum fever and non-fever cases and to compare the value of continuous versus discrete vital sign data collection. Variables for the continuous LR model were temperature (T), heart rate (HR), standard deviation of normal-to-normal intervals (SDNN), and root mean square of successive differences (RMSSD) data averaged over the last 30 minutes before a specific timepoint, while the discrete LR model applied the most recent manually measured T and HR only, for the same timepoint. Four prediction time horizons were tested: 4–5, 3–4, 2–3, and 1–2 hours before fever onset. The LR models were validated by a four-fold leave-one-out cross-validation. All patients were shuffled, and each fold had approximately the same of number febrile and afebrile cases. Model performance was evaluated by calculated area under the curve (AUC) for receiver operating characteristic (ROC) and precision-recall (PR) curves.

Figure 2.

Example of recorded and calculated data. Shown is an example of all data corresponding to one patient. From top, temperature, heart rate, standard deviation of normal-to-normal intervals (SDNN), and root mean square of successive differences (RMSSD) were tracked during maternal labor, with t = 0 (vertical black solid line). Fever onset occurs approximately 128 minutes before delivery (vertical black dotted line). The blue traces for temperature and heart rate represent continuous data recorded by the wireless vital sign monitoring device, while the red trace shows manually taken temperatures. The yellow and purple traces for SDNN and RMSSD, respectively, are calculated from RR interval data recorded by the wireless vital sign monitoring device.

Figure 3.

Average heart rate variability (HRV) during labor. HRVs were averaged over all patients in afebrile (blue) and febrile (red) groups. For the febrile group, t = 0 was defined as onset of fever. For afebrile group, t = 0 was defined 2.4 hours prior to delivery, corresponding to the mean time of fever onset in the febrile group. Average maternal standard deviation of normal-to-normal intervals (SDNN) and root mean square of successive differences (RMSSD) showed significant separation between febrile and afebrile cohorts at 2 and 3 hours prior to fever onset, respectively. Average maternal LF/HF ratio did not show a significant difference between afebrile and febrile groups.

Figure 4.

Model performance illustrated by receiver operating characteristic (ROC) curves. ROC curves for predicting intrapartum fever at time horizon of (a) 4–5 hours and (b) 2–3 hours are shown. Dotted lines show performance evaluated for individual folds, and the solid line shows the average performance of the model with continuous (red) and discrete (blue) input variables. (c) The area under the curve (AUC) increases with decreasing time horizon values. Empty markers show AUC of individual folds, and filled markers show the average performance of the model with continuous (red) and discrete (blue) input variables.

Figure 5.

Model performance illustrated by precision-recall (PR) curves. PR curves for predicting intrapartum fever at time horizon of (a) 4–5 hours and (b) 2–3 hours are shown. Dotted lines show performance evaluated for individual folds, and the solid line shows the average performance of the model with continuous (red) and discrete (blue) input variables. (c) The area under the curve (AUC) increases with decreasing time horizon values. Empty markers show AUC of individual folds, and filled markers show the average performance of the model with continuous (red) and discrete (blue) input variables.