Body temperature instability and respiratory morbidity in the very low birth weight infant: a multiple case, intensive longitudinal study

This was a secondary analysis of data from 12 very premature infants enrolled in the parent study, “Body Temperature and Vasomotor Tone in Preterm Infants,”(NIH/NINR: 1R15NR012157-0; RWJF: 68041), an intensive exploratory study conducted at a North Carolina university hospital from 2010–2013 that examined vasomotor tone maturation and associated morbidity and mortality. Eligibility criteria for the parent study were infants < 29 weeks gestational age (GA) at birth and a birthweight (BW) of < 1200 grams, and it included 30 infants: 22 infants completing all data collection [28]. This secondary analysis included 12 infants from the parent study. Infants transferred before 34 weeks postmenstrual age (PMA) (two weeks prior to BPD diagnoses) (N = 5) or discharged before 36 weeks PMA (N = 1) were excluded from this secondary analysis. An additional infant (N = 1) was excluded because some data were no longer accessible for analysis. Because birthweight is a significant risk factor for BPD and thermal instability [5, 16], a sample that equally represented three birthweight categories (< 800 grams, 800–900 grams, > 900 grams) was then chosen from the remaining infants (N = 15). This sample (N = 12) included four infants per birthweight category. The parent study and this secondary analysis was approved by Duke Health Institutional Review Board.

Infants were stabilized under radiant heat, then transferred to a pre-warmed Draeger Caleo incubator within 4–6 hours of life (Knobel-Dail et al., 2017). Per study site standard of care, infants were placed on servo-control mode with a setting of 36.5 °C to 37.2 °C, with a relative humidity setting based on birthweight (≤ 750 grams = 75–80%; >750 grams = 0-40%). Using a skin temperature probe, servo-control mode continuously monitors skin temperature, adjusting heat output to achieve or maintain the servo temperature setting. The servo temperature setting (servo set point) was retrieved from the electronic health record (EHR). Minute to minute abdominal temperatures were measured by covered Y series Steri-Probe® skin temperature probes, which were covered with a reflective cover (Model 499B, Cincinnati Sub-Zero, Cincinnati, OH) [28]. Nurses repositioned infants and examined skin at probe sites throughout each day according to standard care [28]. Temperature probe positions were changed a minimum of every 6–8 hours, with the infant’s skin condition recorded on the infant’s case study log [28]. With secure and appropriated probe placement, the skin temperature of VLBW infants is an adequate proxy for their body temperature [29, 30]. The thermistor was attached to a 4-channel data logger, model SP-1400-44Y (Veriteq Instruments; Vaisala; Richmond, British Columbia, CA), which sampled and stored temperatures every minute for 2 weeks. Body temperatures below 33.0^o C and above 39.0^o C were removed during parent study data cleaning, as they were thought to represent errors in data collection (i.e., insecure probe placement or probe removal for repositioning) [28]. These temperatures were marked as missing and were excluded from analysis. Heart rate (HR) was monitored continuously with the infant’s General Electric Healthcare cardiopulmonary monitor. Oxygen saturations (SpO₂) were collected using Masimo Radical-7 pulse oximeters (Masimo Corporation, Irvine, CA) with 10-second averaging. Missing SpO₂ measurements or measurements equal to zero were excluded from analysis. All minute-to-minute measurements (~ 20,000 measures per data type) were downloaded into a laptop computer. Additional respiratory data, including apnea episodes, FiO₂ requirement, RS settings (i.e., PIP, PEEP, respiratory rate (RR)), surfactant and diuretic dosing, and FiO₂ requirement at 28 days and 36 weeks PMA were extracted from EHR review. Infant demographics and additional risk factors for acute respiratory morbidity and CLD were also extracted from the EHR, including: GA, BW, sex, maternal diagnosis of chorioamnionitis, Apgar scores at 1 and 5 minutes, infant infection (i.e., sepsis or presumed sepsis), small for gestational age (SGA), intrauterine growth restriction (IUGR), and diagnosis of patent ductus arteriosus (PDA). See Table 1 for analysis variables and their definitions.

Abdominal temperature, HR, SpO₂, and EHR data for each infant were concatenated and archived in a SAS case dataset (one dataset per infant) using ^®SAS statistical software version 9.4 (Cary, NC). The 12 infant datasets were then merged into a final analysis dataset. Analysis variables in this final dataset included measures of body temperature, biomarkers for acute respiratory morbidity, chronic respiratory morbidity risk, and covariates. The covariates SGA and IUGR were combined into a single covariate during analysis as infants with IUGR in this study also met criteria for SGA (i.e., BW < 10% on growth curve).

Descriptive statistics were used to summarize infant characteristics and analysis variables during the first 14 days of life (DOL). Non-directional statistical tests were performed with the level of significance set at 0.05 for each test. The significance level was not adjusted for the multiple outcomes and tests for this exploratory analysis of 12 VLBW infants. Effect sizes and their 95% confidence intervals (CIs) were calculated to address clinical relevance.

Hierarchical multi-level, mixed-effects models for intensive longitudinal data were used to examine the association between the infant body temperatures during the first 14 DOL and both CRMR and biomarkers for acute respiratory morbidity. Random coefficients regression models (RRMs), a type of multi-level, mixed-effects model for intensive longitudinal data, were used for continuous outcomes, while Generalized Linear Mixed Models (GLIMMIX) were applied for binary outcomes. A multi-level approach was applied because level-3 of such models allowed us to adjust for nesting within each infant. The multi-level models also included six infant covariates: (1) infant GA, (2) BW, (3) Apgar score at 5 minutes, (4) infection, (5) sex, and (6) surfactant dosing. SGA/IUGR was omitted as a covariate due to its collinearity with GA (Spearman r_s = 0.66). Similarly, Apgar score at one minute (Spearman r_s = -0.68), chorioamnionitis (Spearman r_s = 0.55), and PDA (Spearman r_s = -0.52) were excluded due to collinearity with surfactant. The fixed effects in the multi-level modes were the measured outcomes (i.e., CRMR, desaturations, B/D, apnea, increase in FiO2 requirement, or increase in RS) and the six infant covariates, while the random effect was the infant.

RRMs were used to determine and describe the trajectory of infant body temperatures across the 14 days. The fixed effects were DOL and the six covariates, while random effects were infant and the infant-by-DOL interaction trajectories. Outcome was body temperature across the 14 days, which did not significantly change across days (p = 0.526), and none of the covariates were significantly associated infant body temperature (all p > 0.05). Since there was not a significant change in body temperature across days, DOL was not included as a predictor in the multi-level model analysis models.

Descriptive statistics were used to examine the association between episodes of hypothermia and hyperthermia with the outcome of CRMR. Minute-by-minute data with missing temperatures were excluded from the analysis. Hypothermic and hyperthermic episodes were analyzed by infant case, then merged for group analysis (N = 12). Following removal of missing abdominal temperatures, an average of 22.3 hours (20.3–23.1 hours) per infant per day were assessed over the 14 days. As a group (N = 12), infants spent almost one-third of the assessed time hypothermic (Median = 30.5%, IQR = 26.7% − 49.1%), and the median duration of each hypothermic episode was 26.8 minutes (IQR = 20.2–38.7). Infants without CRMR (N = 04) spent more time hypothermic (Median = 51.8%, IQR = 35.7–63.6%) than infants with CRMR (N = 08) (Median = 27.6%, IQR = 26.7–34.2%), and the duration of hypothermic events were longer in infants without CRMR (Median = 42.4 min, IQR = 26.1–62.7 min) than infants with CRMR (Median = 22.4 min, IQR = 20.2–34.4 min). The occurrence of hyperthermia was similar among infants with and without CRMR (CRMR: Median = 6.3%, IQR = 4.4% − 10.5%; without CRMR: Median = 6.5%, IQR 3.4–9.5%), yet the duration of hyperthermic episodes was shorter among infants with than without CRMR (CRMR: Median = 36.3 minutes, IQR = 30.6–55.7; without CRMR: Median = 47.2 min, IQR = 39.1–50.9).

Multi-level models were then used to examine the association between infant body temperatures and CRMR. Infant body temperatures were not significantly associated with the outcome of CRMR (p = 0.2765). However, the adjusted mean body temperature of infants with CRMR was euthermic, while the adjusted mean body temperature of the infants without CRMR were hypothermic. Very small effect sizes for these statistically significant relationships were observed (both Cohen’s d < ± 0.20). All covariates were non-significant (all p > 0.05). Table 3 presents the adjusted means and details of RRM.

Multi-level models were also used to examine the association between infant body temperatures and desaturations and B/D. Desaturations and B/Ds were significantly associated with lower body temperatures (both p < 0.0001). Very small effect sizes for these statistically significant relationships were observed (both d < ± 0.20). All covariates were non-significant (all p > 0.05). Table 3 presents the adjusted means and details of RRM.

Bivariate regression and GLIMMIX were used to examine the association between infant body temperatures and documented episodes of apnea, increases in FiO₂ requirement, and increases in RS. Each episode included the recorded respiratory event (e.g. apnea), as well as the 15 minutes preceding and the five minutes following each event. These episodes were created to assess the relationship between the respiratory event and it’s preceding temperatures.

Table 4 shows the results of regression analysis of apnea, increase in FiO2 requirement, increase in RS, and associated covariates in relation to infant body temperature. Apnea was significantly associated with warmer body temperatures (p = 0.0318), and all covariates were significant. Increase in FiO₂ requirement (p = 0.0298) and increase in RS (p < 0.0001) were significantly associated with decreased body temperatures. All covariates were significantly associated with all infant body temperature (all p < 0.0001), with larger BW, younger GA, higher Apgar scores, male sex, and surfactant dosing associated with higher body temperatures, while presence of infection was associated with decreased body temperatures.

Desaturations, B/D, increases in FiO₂ requirement, and increases in RS were significantly associated with hypothermia and hyperthermia (all p < 0.0001). The adjusted odds of hypothermia was 1.3 times higher during desaturations, 2.2 times higher during B/D, and 1.2 times higher during increases in FiO2 requirement and increases in RS. The adjusted odds of hyperthermia was 1.2 times lower during desaturations, 1.4 times lower during B/D, and 2.2 times lower during increases in RS. See Fig. 1 for a summary of the aORs. The episodes of increased FiO₂ requirement during hyperthermia were too few to analyze. Covariates were associated with hypothermia, including increased GA during B/D (p = 0.0057), lower Apgar scores during increases in FiO2 requirement (p = 0.0220), and smaller BW during increases in RS (p = 0.0390). Covariates associated with hyperthermia included lower Apgar score during desaturation (p = 0.0170), increased Apgar score during B/D (p = 0.0170), and smaller BW during increases in RS (p = 0.0051). The remaining covariates were not significantly related to hypothermia or hyperthermia (all p > 0.05).