Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Jan 1.
Published in final edited form as: Nurs Res. 2021 Mar-Apr;70(2):142–149. doi: 10.1097/NNR.0000000000000483

Predicting Infection in Very Preterm Infants: A Study Protocol

Robin B Dail 1, Kayla C Everhart 1, James W Hardin 2, Weili Chang 3, Devon Kuehn 3, Victor Iskersky 4, Kimberley Fisher 5, Heidi J Murphy 6
PMCID: PMC8044260  NIHMSID: NIHMS1636929  PMID: 33630537

Abstract

Background:

Neonatal sepsis causes morbidity and mortality in preterm infants. Clinicians need a predictive tool for the onset of neonatal infection to expedite treatment and prevent morbidity. Abnormal thermal gradients, a central-peripheral temperature difference (CPtd) of > 2°C or < 0°C, and elevated heart rate characteristic (HRC) scores are associated with infection.

Objective:

This article presents the protocol for the Predictive Analysis using Temperature and Heart Rate (PATH) study.

Methods:

This observational trial will enroll 440 very preterm infants to measure abdominal temperature (AT) and foot temperature (FT) every minute and HRC scores hourly for 28 days to compare to infection data. Time with abnormal thermal gradients (Model 1) and elevated HRC scores (Model 2) will be compared to the onset of infections. For data analysis, CPtd (AT-FT) will be investigated as two derived variables high CPtd (number/percentage of minutes with CPtd > 2°C) and low CPtd (number/percentage of minutes with CPtd < 0°C). In the infant-level model, the outcome yi will be an indicator of whether the infant was diagnosed with an infection in the first 28 days of life and the high CPtd and low CPtd variables will be the average over the entire observation period; logit(yi) = β0 + xiβ1 + ziγ. For the day-level model, the outcome yit will be an indicator of whether the ith infant was diagnosed with an infection on the tth day from t = 4 through t = 28 or the day that infection is diagnosed (25 possible repeated measures) logit(yit) = β0 + xitβ1 + zitγ. It will be determined whether a model with only high CPtd or only low CPtd is superior in predicting infection. Also, the correlation of abnormal HRC scores with high CPtd and low CPtd values will be assessed.

Discussion:

Study results will inform the design of an interventional study using temperatures and/or heart rate as a predictive tool to alert clinicians of cardiac and autonomic instability present with infection.

Keywords: body temperature, heart rate, infection, neonatal

Neonatal sepsis is an important cause of morbidity and mortality in the very preterm infant or infants less than 32 weeks gestational age (GA) and has been reported to occur in up to 41% of infants hospitalized in neonatal intensive care units ([NICU]; Greenberg et al., 2017). Blood stream infections in the preterm infant are categorized as early onset sepsis (EOS) or late onset sepsis (LOS). EOS occurs in the first 72 hr postbirth and is typically reflective of maternal infections encountered by the neonate prior to birth (Boghossian et al., 2013). LOS is acquired in the nosocomial hospital environment and can occur anytime during the hospitalization with the highest incidence between 10 and 22 days of life. Infants often exhibit subtle signs and symptoms of infection (Dong & Speer, 2015). While blood culture growth is the gold standard diagnostic tool to confirm neonatal sepsis, blood cultures often take up to 48–72 hr to grow pathogens and frequently do not grow anything even if sepsis is present (Agarwal et al., 2015; Silva et al., 2015; Stoll et al., 2015). With these realities, it is difficult for clinicians to know if and when to begin treatment with broad spectrum antibiotics. EOS and LOS can result in devastating neurodevelopmental sequela and death (Dong & Speer, 2015; Silva et al., 2015; Stoll et al., 2015); therefore, it essential that clinicians have a quick, efficient, and cost-effective biomarker to predict the onset of infection to expedite treatment.

Necrotizing enterocolitis (NEC) is a type of infection in the gastrointestinal (GI) tract, which occurs in 5%–15% of preterm infants (Stoll et al., 2010; Yee et al., 2012). NEC is considered a major morbid outcome in preterm infants and has a mortality rate as high as 50% (Blakely et al., 2005). The incidence of NEC increases with decreasing GA. The etiology of NEC is not clear; however, it may involve a combination of immaturity of the intestinal epithelial barrier and mucosal immune system—predisposing infants to bacterial invasion—particularly in the context of poor intestinal perfusion (Nankervis, et al., 2008; Nowicki, 2005). Abnormal postnatal circulation can lead to problems with feeding intolerance and intestinal motility, both of which are associated with NEC (Robel-Tillig et al., 2004). NEC is a significant problem for preterm infants, occurring in 1% to 28% of infants born less than 1,500 g in other countries, (Caplan & Jilling, 2001) and from 1% and 7% in infants weighing less than 1,000 g in the United States (Lin & Stoll, 2006). If NEC is identified early, then treatment of bowel rest, antibiotics, and radiological surveillance can be implemented.

A thermal gradient is the difference between body temperatures in different areas of the body. The central body to peripheral body temperature gradient can be measured using the central-peripheral temperature difference (CPtd), or abdominal temperature (AT) – foot temperature (FT). Large and very small thermal gradients, or a CPtd > 2˚C or < 0°C, have been found to be associated with infection in early preterm infants (Knobel-Dail, 2017; Leante-Castellanos et al., 2017; Ussat et al., 2015). A CPtd < 0°C is seen when the peripheral or foot temperature is warmer than the abdominal or central temperature and has mainly been seen in extremely premature infants (Knobel-Dail et al., 2017; Knobel et al., 2009; Lyon et al., 1997). Because surveillance of continuous body temperature, including the CPtd, promises to be a low-cost, quickly accessible biomarker of infection, our team is conducting an observational trial in 440 preterm infants less than 32 weeks GA to confirm the relationship between abnormal patterns of CPtd (> 2°C and/or < 0°C) and the onset of infection.

Reduced heart rate variability and transient decelerations, which are abnormal heart rate characteristics (HRC), also occur early in the course of neonatal sepsis (Fairchild, 2013; Griffin & Moorman, 2001). The HeRO monitor (Medical Predictive Science Corporation [MPSC], Charlottesville, VA) utilizes an algorithm that yields an “HRC score” using heart rate variability, decelerations, and standard deviations using the R-to-R interval times series from the electrocardiogram of neonatal monitors (Fairchild & Aschner, 2012). The HeRO monitor display has been utilized to predict infection in early preterm infants (Fairchild, 2013; Fairchild et al., 2017; Sullivan et al., 2014). A HeRO score of 1 indicates a low risk for infection and a 2 indicates a twofold increase of developing an infection within 24 hr (Hicks & Fairchild, 2013). This monitor is now in use in many NICUs, as an adjunct to the clinical signs that predict the onset of sepsis (personal communication; MPSC, Charlottesville, VA). The HeRO monitor is valuable for predictive monitoring; however, it must be purchased separately and used in conjunction with normal cardiac monitoring to calculate the HRC score using a sophisticated algorithm derived from multivariable logistic regression. The HeRO approach will be utilized as a valid, rigorous comparison model for our study of infants and body temperature to predict infection.

Results of this study will provide the evidence needed to move to an interventional study using continuous surveillance of central and peripheral temperature as a noninvasive assessment tool to alert clinicians of cardiac and autonomic instability present with infection. Clinicians will be able to use predictive monitoring of body temperature to decrease morbidity and mortality from infection and NEC by initiating earlier treatment for very preterm infants.

Objective

In this trial (2018–2023), we will examine averaged daily CPtd values with indicators of infection to determine if abnormal thermal gradients, or CPtd values of < 0˚C and/or > 2˚C predict the onset of infection (Model 1). In comparison, we will also examine averaged daily HRC scores generated by HeRO monitors as a predictor of infection (Model 2). Our trial will determine which model is superior and if a combined model, using both CPtd and HRC scores, or separate models improve the sensitivity and specificity of identifying infection through body temperature and/or heart rate changes. Aims for this study are:

Aim: 1.

In preterm infants, < 32 weeks GA, examine the temporal relationship between CPtd and diagnosis of infection. Hypothesis 1.1: During the first 28 days of life of very low birth weight (VLBW) preterm infants, an increase in CPtd > 2°C and/or CPtd < 0°C will be associated with an increased likelihood of diagnosed infection within 72 hr. Hypothesis 1.2: During their first 28 days of life, VLBW preterm infants who have a higher percentage of time with CPtd > 2°C and/or < 0°C are more likely to be diagnosed with NEC prior to discharge from the hospital.

Aim 2:

In VLBW preterm infants, < 32 weeks GA, examine the temporal relationship between HRC scores and diagnosis of infection. Hypothesis 2.1: Abnormal HRC scores (> 2) will be correlated with abnormal CPtd values (> 2°C or < 0°C). Hypothesis 2.2: VLBW infants diagnosed with infection within the first 28 days of life will have abnormal HRC and CPtd values within the preceding 24–72 hr of diagnosis.

Aim 3:

Develop a predictive model to determine what amount and duration of CPtd values best predict odds of infection diagnosis within 72 hr. We will then compare the CPtd and HRC predictive models using cross-validation with the data we collect. We will also look at a combined prediction model using HRC scores and CPtd values to see whether separate models are superior.

Exploratory Aim: 1.

Examine maternal factors (smoking, drug use, pre-eclampsia), VLBW preterm infant demographic factors (sex, race, GA, birth weight), infant clinical factors (patent ductus arteriosus [PDA], catecholamine infusions, caffeine administration) and research site as moderators to the relationship of body temperature and infection.

Methods

Study Design

This is an observational clinical trial over 5 years.

Setting and Sample Size

Setting

The study will take place at five tertiary NICUs in North and South Carolina: Duke University Hospital (DUH), East Carolina University (ECU) Vidant Medical Center, Medical University of South Carolina (MUSC), PRISMA Health, and University of North Carolina-Chapel Hill (UNC-CH). Each site has a site principal investigator (PI) and 50% effort research nurse coordinator (RNC). The University of South Carolina (UofSC), College of Nursing is the study coordinating and data analytic site and. home of the overall study PI, Dr. Robin Dail. MUSC is the central institutional review board (IRB) regulatory site.

Participants

Over the 4-year enrollment period we will enroll 440 preterm infants, taking into consideration that we may have up to 20% attrition, for a target sample of 350 with complete data. Infants may be included in the study if they (a) weigh 500 g to 1,500 g at birth, (b) are determined to be 24 to 32 weeks GA by obstetrical dating, (c) are born at the enrollment site, and (d) are available for enrollment within 6 hr of life. In our previous studies, we found that many times infants less than 24 weeks GA and less than 500 g were too unstable to enroll and often had skin that was too fragile to monitor with two additional temperature probes. Previously, we examined infants less than 29 weeks GA; however, infants do not have mature thermal stability until 32 weeks GA. Therefore, we are enrolling infants between 500 g and 1,500 g and between 24–32 weeks GA to include infants who vary in maturity and in clinical course. Infants must be born at the enrollment site to negate any environmental influences from a transport in an ambulance or helicopter influencing their body temperature and perfusion status. Exclusion criteria will include any major anatomical or cardiac defect known at birth, as these may increase the infant’s instability. Specifically, congenital heart disease has a known impact on circulation and increases risk of NEC (Nankervis et al., 2008).

Sample Size Considerations

The design for this study includes a time-to-event Cox proportional hazards regression model with a time varying covariate and a logistic regression analysis. We conservatively use 20% for infection group based on the national statistics that 20%–25% of the population who meet our inclusion criteria have infection (Agarwal et al., 2015; Stoll et al., 2015). For an infant-level logistic regression model (one observation per infant), we will have 80% power to detect an odds ratio of 1.40 associated with each of our main predictors of interest (CPtd is defined as the average/day over the entire observation period) at a level of significance of 5%—even if we have as much as 20% attrition (Aim 1). Knowing that an infant has 40% higher odds of an infection is clinically relevant. For a day-level logistic regression model with CPtd measures > 2°C or < 0°C defined over the preceding 3 days for Day 4 through the minimum of the day number on which an infection was diagnosed and Day 28, we will have 80% power to detect an odds ratio of 1.16 associated with each of our main predictors of interest at a level of significance of .05. This will so even if we have as much as 20% attrition (Aim 1). These effects correspond to a standardized effect of 0.16 and 0.06, respectively, which are smaller than Cohen’s small standardized effect size (Chen et al., 2010; Cohen et al., 2003). Notwithstanding protection of analyses for missing data, we will use multiple-imputation techniques in our analyses using Rubin’s method for combining analytic results across multiple imputed data sets. Similarly, we will have 80% power in Cox proportional hazards models to detect a hazard ratio of between 1.2 and 1.4 at level of significance 5%. This hazard ratio corresponds to being able to detect a 30% faster rate of infection due to increases in our main predictors of interest (CPtd > 2°C or < 0°C). For Hypothesis 2.1 of Aim 2, we will have 80% power to detect a correlation of 0.33 or higher at a 5% level of significance assuming 20% attrition and assuming a 20% infection rate; under the same assumptions, we will have 80% power to detect a standardized effect of 0.34 for Hypothesis 2.2 of Aim 2. The power achieved for this sample size is needed for primary Aim 3, which will develop the best function of CPtd > 2°C and CPtd < 0°C to create a prediction model of infection that might compete with the prediction model based on HRC.

Institution Review Board Procedures

This study was funded after the January 2018 mandate by the National Institutes of Health for all studies with multiple sites to use a single IRB of record (sIRB). We used SMART IRB (https://smartirb.org/) to track the reliance agreement for the sites and MUSC. The study protocol, consents, and instrumentation documents were submitted to the MUSC IRB by the UofSC study team.

Funding for this grant was received in September of 2018 and the overall study IRB approval was granted on November 4, 2018, which included site enrollment at MUSC. One site was approved in January 2019, one in February, and the two final sites were approved in March 2019. Because of the procedures in using sIRB and having five enrollment sites, each with their own IRB procedures, it took 6 months from funding receipt to approval of the last site. Enrollment of the first infant occurred in June 2019.

Procedures

Consent Procedures

All mothers who are hospitalized in the antepartum unit who are expected to deliver a premature infant, not in active labor, and meet enrollment criteria will be approached for consent. Surveillance for potential mothers to approach is accomplished according to site specific mechanisms, with a HIPAA waiver, ranging from using the electronic medical record to consulting with labor and delivery staff. Typically, a member from the site specific study team will be introduced to a mother from a member of the labor and delivery or antepartum staff, the study will be explained and questions answered, and if requested, the consent form will be left with the mother for consideration. From previous studies, we know that prenatal consent from three times as many mothers as infants is needed for enrollment, due to nonenrollment from loss to unexpected delivery during nights and holidays with no call to the study team and from mothers not delivering prior to 32 weeks GA. If an infant is born without an antepartum visit with the parents, a member of the research team will approach the parents after delivery within 6 hr as we are able to enroll infants within 6 hr of birth if one parent is available to consent for their infant.

Enrollment

The first infant enrolled at each site is deemed a pilot infant and data will not be analyzed. Each site will be supplied with sets of instruments to enroll between two and four infants at any given time. The RNC will periodically check with the bedside nurse during each week to ensure there are no questions. The RNC will periodically search the electronic health record for any infant in the study. Study data were collected and managed using REDCap electronic data capture tools hosted at UofSC (Harris et al., 2009). REDCap (Research Electronic Data Capture) is a secure, web-based application designed to support data capture for research studies, providing: 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources. Once an infant has completed the 28 days of physiologic data collection, the data logger and HeRO monitors will be removed from the infant’s bedside and there will be no more physiologic data collection. The RNCs will continue to search each study infant’s record for indications of infection, inclusive of NEC, to enter this information into the REDCap database.

Maternal History

Maternal history will be recorded from the obstetrical record to be used in analyses for interaction effects. Records will be searched for history of tobacco, alcohol or drug use during pregnancy, history of pre-eclampsia, chorioamnionitis, premature rupture of membranes, and infection or fever.

Infant Body Temperature

Infant body temperature will be measured by a thermistor attached to a data logger for continuous recording and data storage (Knobel, 2013). Thermistors are the preferred method for measuring skin surface temperature (Thomas, 2003). The RNC at each NICU site will enroll an infant by attaching a thermistor (disposable skin sensor, 499B, Cincinnati Sub Zero, Cincinnati, OH) to the infant’s abdominal/flank area (i.e., AT) and one to the foot (i.e., FT), then will connect each to one data logger (Squirrel SQ2010, Grant Instruments, Cambridge, England). Thermistors are almost identical to the standard of care thermistors used to monitor infant temperature in clinical care. By covering the thermistor on the abdomen with reflective tape, abdominal skin surface temperature is measured with a near zero heat flux so that the measure approaches core temperature (Simbrunner, 1995). The thermistor on the foot will be attached with the standard “tape” specified by the skin care protocol at each site. Both probe sites will be checked every 4–8 hr for skin breakdown by the infant’s bedside nurse during the study infants’ routine nursing care. Standard of care dictates that nurses rotate thermistor probes positions so that attention to study probes does not pose any additional care. The probe will be moved to an adjacent site if there is any sign of breakdown and the infants’ providers will be notified. The thermistors are attached to the data logger with a telephone jack connection which will be checked daily by the research coordinator at each site. Thermistors are labeled with the specified position so that there is integrity and reliability of data.

Data loggers attached to each thermistor will record temperature to the nearest 0.1°C, sampling every minute for 28 days. The data logger is small and able to be positioned outside the incubator. Clinicians in each unit will be blinded to temperatures from the research data loggers.

HeRO HRC Score

HRC scores (Fairchild, 2013; Fairchild & Aschner, 2012; Fairchild & O’Shea, 2010) will be recorded using a HeRO monitor which will be on a stand near the infant’s bedside. The HeRO monitors are being purchased for the study and each unit will receive between two and four monitors for the study. These instruments will be brought to the bedside, and a cable will be connected from the HeRO monitor to the standard cardiopulmonary monitor. The HeRO monitor is compatible with all monitors used in these five NICUs; however, cables are made specifically to connect to each brand of cardiopulmonary monitor. Units in this study are using Phillips, Space Labs, and Nihon Kohden cardiopulmonary monitors, each requiring a different connection cable. The HeRO monitor stores hourly HRC scores which are calculated from the previous 12 hr of R-to-R heart rate intervals. A Universal Serial Bus (USB) stick is inserted in the back of the HeRO monitor which will measure and store 28 days of data. These data will be entered into analysis in a predictive model to analyze the temporal relationship to diagnosis of infection as well as cross correlation with the predictive model between CPtd values and diagnosis of infection.

Infection

Infection will be defined as a diagnosis of EOS, LOS or NEC, and date of diagnosis of infection will be entered into the REDCap database for analysis with quantitative data. EOS will be defined as a condition meeting any of these indicators: mother has been diagnosed with positive chorioamnionitis, premature of rupture of membranes and probable infection suspected and infant is treated with antibiotics due to suspicion of infection beginning at NICU admission for at least 72 hr. For EOS, if the previously stated conditions are met, then the infant is declared infected for the purpose of analysis on date of birth. Blood culture results will be recorded but a positive blood culture is not absolutely needed for diagnosis of EOS; however, a positive blood, urine, or cerebral spinal fluid culture for bacteria or fungus at less than 7 days of age will be declared EOS. If the culture is positive after 72 hr, then the day it turns positive within the first 7 days will be the date infection is diagnosed.

LOS

LOS can include proven sepsis or clinical sepsis as defined by the Center for Disease Control ([CDC]; Garner et al, 1988). For proven LOS, an infant will have one of the following signs: fever, hypothermia, apnea or bradycardia, and a positive blood culture. The clinician must give antibiotics for a minimum of 5 days for proven LOS. We will include signs and symptoms of sepsis as well as have a positive culture for bacteria or fungus from blood, urine, or cerebral spinal fluid at 7 days of age or greater. Clinical sepsis does not require a positive culture. We will also document laboratory tests including complete blood count to look for leukopenia or leukocytosis and thrombocytopenia. We will also record episodes of hyperglycemia with a blood sugar value greater than 180 mg/dl to be used as additional evidence of sepsis. For analysis, the date of any positive culture is the date of diagnosis. If no cultures are found positive but antibiotics were administered for at least 5 days and symptoms are present, the date of suspected infection will be coded as such for analysis.

NEC

NEC will be defined based on modified Bell’s criteria, as is used by other large neonatal clinical studies such as those from the National Institute of Child Health and Human Development Neonatal Network (Bell et al., 1978). Infant participants will receive a diagnosis of NEC (Bell’s class II or IIIa) if there is an indication of abdominal distension, feeding intolerance, and/or blood in the stools as well as radiographic findings of suspected or definite pneumatosis intestinalis, portal venous gas, and/or pneumoperitoneum in the medical chart. The date of diagnosis for analysis will be the date of the positive radiograph with symptoms present.

The temporal relationship between the day of diagnosis of infection will be recorded for analysis with longitudinal measures of CPtd values and HRC scores over the infant’s first 28 days of life. Instances of NEC diagnoses will be recorded until discharge to use in analysis with abnormal CPtd and HRC scores to examine differences in infants who are diagnosed with NEC and those who are not.

Infant Demographics, Medical Course, and Clinical Factors

This information will be recorded through review of medical records by the RNC at each site. Race and biological sex will be determined from medical records. Race will be assigned by the mother’s declared race. Sex will be assigned by the admitting medical provider. General demographic information will also be recorded in a REDCap database for each study participant. The medical record will be used to provide the clinical context for each infant. Clinical context variables will be used as possible confounders or moderators and include presence of PDA, catecholamine infusions, caffeine administration, antibiotic administration, bilirubin levels and use of phototherapy. Records will be searched for EOS, LOS, and NEC indicators.

Potential Risks

This study is considered minimal risk because it uses noninvasive physiological monitoring; however, the participants of this study are a critical and unstable population because they are preterm infants. Infection and or death could occur during the hours of this study because these are normal risks to this population. It is extremely unlikely that this observational study could cause additional risks to the infant above the normal risks for a preterm infant admitted to a NICU. Thermistor probes are almost identical to the standard of care thermistor probes and will be positioned using tape adhesive approved by the sites for this population. Skin irritation is a potential risk from thermistors. We had no skin irritation from research probes that caused an infant to be withdrawn from our previous study. Laboratory results for routine NICU tests will be recorded from the medical chart. Either parent may remove the infant from the study at any time. A study infant’s attending physician may also remove an infant from the study at any time.

Data Analysis

Data Management

The RNC at each site will enter maternal and infant demographic and clinical data into the REDCap database developed for this study. The REDCap database will be accessed virtually, hosted on a secure server at UofSC, and password protected. The RNCs are trained to download temperature data from data loggers to the dedicated study laptop computer into a Microsoft™ Excel® file for each study infant; data will be downloaded daily to every few days and will be inspected to ensure data integrity, for a duration of 28 days. As such, a cable must be connected from the data logger to a computer to enable the download. The USB cable that will be connected to the back of the HeRO monitor will collect 28 days of data; data will be transferred by the RNC from the USB to the laptop computer when temperature data are downloaded. Physiologic data will be de-identified and emailed after every download to the RNC at UofSC. Clinical data will be entered into REDCap at each site and will be accessed by the study personnel at UofSC for data management. The data manager at UofSC will clean quantitative data and transfer into SAS (9.4) format data sets for this study. Each site has been assigned a study site number and each infant within that site will be assigned a unique study number including the site identifier. Study personnel at each site can only view REDCap data for their site; UofSC study personnel will have access to all sites’ data. Each day’s data for one infant has 1,440 measures for each temperature measure and 24 measures for HRC scores. Each 28-day data set for each infant will include a maximum of 40,320 measures for each temperature variable and 672 measures for HRC values. All infants will have missing data due to instruments being disconnected, infants being moved to a different bedside, or any other reason instrumentation is disconnected.

Data Analysis and Model Construction

Our interest for this study is the CPtd (AT-FT = CPtd). We will investigate that measure as two derived variables high CPtd (number/percentage of minutes with CPtd > 2°C) and low CPtd (number/percentage of minutes with CPtd < 0°C) which are continuous variables. Descriptive statistics will be computed on each of the variables. We will summarize continuous variables using measures of central tendency (mean and median) and measures of spread (standard deviation and range). Infection (EOS, LOS, or NEC) versus noninfection are the groups over which our main variables and covariates will be described using frequencies and cross tabulations. Because infection will be recorded as the day of diagnosis, we will bin our predictors of interest (high CPtd and low CPtd) over each day and treat that as a time-varying covariate (high CPtd and low CPtd for each defined epoch will reflect a value derived from the preceding 72 hr) in a proportional hazards model of hi(t) = h0(t)exp(xitβ1 + ziγ). In this case, the hazard hi(t) is a function of the baseline hazard h0(t) and is a nonnegative function of the time-varying covariate of interest xit (Aims 1 & 2). It is also a vector of moderators zi (mother smoked tobacco and/or used drugs during pregnancy, diagnosis of pre-eclampsia, sex and race of the child, birth weight in grams of the child, GA, PDA, catecholamine infusions, caffeine, and NICU site; Exploratory Aim A) with an associated parameter vector γ. We will estimate a separate Cox proportional hazards regression model for each time to infection event. Further, we will estimate separate models for the time varying covariates of interest xit (high CPtd and low CPtd); thus, there are two models in total (Aim 3). A significant positive estimated β^1 would indicate that infants with higher values of the predictors of interest (high CPtd or low CPtd) have a higher rate of infection.

We are also interested in modeling the likelihood of infection as a function of our covariates of interest (Aim 1 & Aim 2). In the infant-level model, the outcome yi will be an indicator of whether the infant was diagnosed with an infection in the first 28 days of life and the high CPtd and low CPtd variables will be the average over the entire observation period; logit(yi) = β0 + xiβ1 + ziγ (Aim 1). For the day-level model, the outcome yit will be an indicator of whether the ith infant was diagnosed with an infection on the tth day from t = 4 through t = 28 or the day that infection was diagnosed (25 possible repeated measures) logit(yit) = β0 + xitβ1 + zitγ. In either model in our multilevel analysis, our primary interest is in the Wald test of the null hypothesis H0:β1 = 0. As before, there are two models of interest (Aim 1 & Aim 2) as we will investigate whether a model with only high CPtd or only low CPtd is superior. The first hypothesis is an evaluation of the main effect of the predictor of interest (Aim 1). A significant positive estimated β^1 would indicate that infants with higher values of the predictor of interest (high CPtd or low CPtd) have higher odds of infection. Exponentiated estimated coefficients will be interpreted as odds ratios. We will also consider estimating repeated measures log-binomial generalized linear mixed models for which exponentiated estimated coefficients are interpreted as risk ratios. To investigate the hypotheses associated with Primary Aim 2, we will investigate models using the HRC scores. Also, we will formally assess the correlation of abnormal HRC scores with high CPtd and low CPtd values (Aim 3).

In each approach, we will estimate the fully adjusted models as described above, and we will estimate unadjusted models wherein we will not include the zitγ term of the model (Aim 3). We will investigate complete-case models as well as models based on missing data imputed using multiple imputation methods (Cohen et al., 2003). We will use sensitivity analyses for dropouts to determine whether the dropouts are systematic, completely at random, or missing at random. Several techniques will be used to examine the pattern of dropouts (Cohen et al., 2003). The level of statistical significance will be set at 0.05. Values that will be reported include beta coefficients, standard error of beta, and P-value.

Predictive models will be built utilizing high CPtd and low CPtd information defined over various lengths of preceding time (24 hr to 72 hr) to ultimately yield the optimal predictive model based on CPtd (Aim 3). Our predictive model will be compared to the HRC model utilizing a Monte Carlo cross-validation approach in which we will use 70% of the data for training and 30% for validation. If our CPtd predictive model improves the HRC model, it will serve as the foundation of a subsequent proposal to validate the CPtd predictive with new data. Finally, we will combine the HRC predictive covariates to the CPtd covariates to investigate whether the combined model offers further improvements to prediction. Again, we will investigate the optimal amount of lead time over which time varying predictive covariates should be measured.

Results

Participant recruitment and enrollment began in June 2019 and as of May 2020, 81 infants have been successfully enrolled. Analysis of study results is expected in fall of 2023 following study completion. With the onset of Covid-19 in March 2020, recruitment was suspended at three of five sites due to local regulations. As of June 2020, two of those sites have resumed enrollment with four of five sites in active study recruitment.

Discussion

Limitations

In our previous studies, we found that labeling thermistors as to their intended anatomical location prevented the nurses from attaching the probe in the wrong location, and we benefited that the telephone jack connection on our thermistors prevented them from being dislodged from the data logger. Even with labeled thermistors, we have had nurses attach thermistors to the wrong body part in this study. Therefore, our RNCs at each site are checking infants daily to ensure data integrity. If a thermistor is found on the unintended anatomical site, the data are coded as missing for that duration. We have standard operating procedures in place at each site to help prevent potential instrument problems. Each site has a team of RNCs who check equipment on infants daily. RNCs will work closely with site PIs to ensure rigor of study methods and data collection.

The ability to develop a predictive model using CPtd or HRC depend on associations between CPtd and HRC to be measured in Aims 1 & 2. Modeling will also be dependent on complete data trajectories and missing data could lessen our ability to make correlations. The study will enroll 440 infants to attempt to have 350 complete data sets for analysis.

Conclusion

Study results will provide the evidence needed to move to an interventional study using continuous surveillance of central and peripheral temperature as a noninvasive assessment tool to alert clinicians of cardiac and autonomic instability present with infection and predict infants at risk for NEC. Results will add to the science of using physiologic vital signs to predict the onset of infection, allowing algorithms to be translated into practice. Clinicians will benefit from this efficient and noninvasive tool to guide care of vulnerable premature infants and in doing so, morbidity and mortality can be reduced in this fragile population.

Acknowledgement:

Research reported in this publication was supported by the National Institute of Nursing Research of the National Institutes of Health, National Institute of Nursing Research, under Award Number 1R01NR017872-01A1. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

This protocol was supported by NIH/NCATS Colorado CTSA Grant Number UL1 TR002535. Its contents are the authors’ sole responsibility and do not necessarily represent official NIH views.

Footnotes

Ethical Conduct of Research: This study was approved under protocol # 82482 by the Institutional Review Board of Medical University of South Carolina, Charleston, SC, with initial approval date of November 6, 2018.

Clinical Trial Registration: This research study is not a clinical trial and has not been registered as such.

The authors have no conflicts of interest to report.

References

  1. Agarwal P, Sriram B, & Rajadurai VS (2015). Neonatal outcome of extremely preterm Asian infants < 28 weeks over a decade in the new millennium. Journal of Perinatology, 35, 297–303. 10.1038/jp.2014.205 [DOI] [PubMed] [Google Scholar]
  2. Bell MJ, Ternberg JL, Feigin RD, Keating JP, Marshall R, Barton L, & Brotherton T (1978). Neonatal necrotizing enterocolitis. Therapeutic decisions based upon clinical staging. Annals of Surgery, 187, 1–7. 10.1097/00000658-197801000-00001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Blakely M, Lally K, McDonald S, Brown RL, Barnhart D, Ricketts R, Thompson WR, Scherer LR, Klein MD, Letton RW. Chwals WJ, Touloukian RJ, Kurkchubasche AG, Skinner MA, Moss RL, & Hilfiker ML (2005). Postoperative outcomes of extremely low birth-weight infants with necrotizing enterocolitis or isolated intestinal perforation: A prospective cohort study by the NICHD Neonatal Research. Annals of Surgery, 241, 984–994. 10.1097/01.sla.0000164181.67862.7f [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Boghossian NS, Page GP, Bell EF, Stoll BJ, Murray JC, Cotten CM, Shankaran S, Walsh MC, Laptook AR, Newman NS, Hale EC, McDonald SA, Das A, & Higgins RD (2013). Late-onset sepsis in very low birth weight infants from singleton and multiple-gestation births. Journals of Pediatrics, 162, 1120–1124. 10.1016/j.jpeds.2012.11.089 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Caplan MS, & Jilling T (2001). New concepts in necrotizing enterocolitis. Current Opinion in Pediatrics, 13, 111–115. 10.1097/00008480-200104000-00004 [DOI] [PubMed] [Google Scholar]
  6. Chen H, Cohen P, & Chen S (2010). How big is a big odds ratio? Interpreting the magnitudes of odds ratios in epidemiological studies. Communication in Statistics—Simulation and Computation, 39, 860–864. 10.1080/03610911003650383 [DOI] [Google Scholar]
  7. Cohen J, Cohen P, West SG, & Aiken LS (2003). Applied multiple regression/correlation analysis for thebehavioral sciences (3rd ed.). Lawrence Erlbaum Associates. [Google Scholar]
  8. Dong Y, & Speer CP (2015). Late-onset neonatal sepsis: recent developments. Archives of Disease in Childhood—Fetal and Neonatal Edition, 100, F257–F263. 10.1136/archdischild-2014-306213 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Fairchild KD (2013). Predictive monitoring for early detection of sepsis in neonatal ICU patients. Current Opinion in Pediatrics, 25, 172–179. 10.1097/MOP.0b013e32835e8fe6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Fairchild KD, & Aschner JL (2012). HeRO monitoring to reduce mortality in NICU patients. Research and Reports in Neonatology, 2, 65–76. 10.2147/RRN.S32570 [DOI] [Google Scholar]
  11. Fairchild KD, Lake DE, Kattwinkel J, Moorman JR, Bateman DA, Grieve PG, Isler JR, & Sahni R (2017). Vital signs and their cross-correlation in sepsis and NEC: A study of 1,065 very-low-birth-weight infants in two NICUs. Pediatric Research, 81, 315–321. 10.1038/pr.2016.215 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Fairchild KD, & O’Shea TM (2010). Heart rate characteristics: Physiomarkers for detection of late-onset neonatal sepsis. Clinics in Perinatology, 37, 581–598. 10.1016/j.clp.2010.06.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Garner JS, Jarvis WR, Emori TG, Horan TC, & Hughes JM (1988). CDC definitions for nosocomial infections, 1988. American Journal of Infection Control, 16, 128–140. 10.1016/0196-6553(88)90053-3 [DOI] [PubMed] [Google Scholar]
  14. Greenberg RG, Kandefer S, Do BT, Smith PB, Stoll BJ, Bell EF, Carlo WA, Laptook AR, Sánchez PJ, Shankaran S, Van Meurs KP, Ball MB, Hale EC, Newman NS, Das A, Higgins RD, & Cotten CM (2017). Late-onset sepsis in extremely premature infants: 2000–2011. Pediatric Infectious Disease Journal, 36, 774–779. 10.1097/INF.0000000000001570 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Griffin MP, & Moorman JR (2001). Toward the early diagnosis of neonatal sepsis and sepsis-like illness using novel heart rate analysis. Pediatrics, 107, 97–104. 10.1542/peds.107.1.97 [DOI] [PubMed] [Google Scholar]
  16. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, & Conde JG (2009). Research electronic data capture (REDCap)—A metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics, 42, 377–381. 10.1016/j.jbi.2008.08.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hicks JH, & Fairchild KD (2013). Heart rate characteristics in the NICU: What nurses need to know. Advances in Neonatal Care, 13, 396–401. 10.1097/anc.0000000000000031 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Knobel-Dail RB, Sloane R, Holditch-Davis D, & Tanaka DT (2017). Negative temperature differential in preterm infants less than 29 weeks gestational age: Associations with infection and maternal smoking. Nursing Research, 66, 442–453. 10.1097/nnr.0000000000000250 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Knobel RB, Holditch-Davis D, Schwartz TA, & Wimmer JE Jr. (2009). Extremely low birth weight preterm infants lack vasomotor response in relationship to cold body temperatures at birth. Journal of Perinatology, 29, 814–821. 10.1038/jp.2009.99 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Knobel RB, Levy J, Katz L, Guenther B, & Holditch-Davis D (2013). A pilot study to examine maturation of body temperature control in preterm infants. Journal of Obstetric, Gynecologic & Neonatal Nursing, 42, 562–574. 10.1111/1552-6909.12240 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Leante-Castellanos JL, Martínez-Gimeno A, Cidrás-Pidré M, Martínez-Munar G, García-González A, & Fuentes-Gutiérrez C (2017). Central-peripheral temperature monitoring as a marker for diagnosing late-onset neonatal sepsis. Pediatric Infectious Disease Journal, 36, e293–e297. 10.1097/inf.0000000000001688 [DOI] [PubMed] [Google Scholar]
  22. Lin PW, & Stoll BJ (2006). Necrotising enterocolitis. Lancet, 368, 1271–1283. 10.1016/S0140-6736(06)69525-1 [DOI] [PubMed] [Google Scholar]
  23. Lyon AJ, Pikaar ME, Badger P, & McIntosh N (1997). Temperature control in very low birthweight infants during first five days of life. Archives of Disease in Childhood—Fetal and Neonatal Edition, 76, F47–F50. 10.1136/fn.76.1.f47 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Nankervis CA, Giannone PJ, & Reber KM (2008). The neonatal intestinal vasculature: Contributing factors to necrotizing entercolitis. Seminars in Perinatology, 32, 83–91. 10.1053/j.semperi.2008.01.003 [DOI] [PubMed] [Google Scholar]
  25. Nowicki PT (2005). Ischemia and necrotizing enterocolitis: Where, when, and how. Seminars in Pediatric Surgery, 14, 152–158. 10.1053/j.sempedsurg.2005.05.003 [DOI] [PubMed] [Google Scholar]
  26. Robel-Tillig E, Knüpfer M, Pulzer F, & Vogtmann C (2004). Blood flow parameters of the superior mesenteric artery as an early predictor of intestinal dysmotility in preterm infants. Pediatric Radiology, 34, 958–962. 10.1007/s00247-004-1285-6 [DOI] [PubMed] [Google Scholar]
  27. Silva SMR, Motta GDCD, Nunes CR, Schardosim JM, & Cunha MLCD (2015). Late-onset neonatal sepsis in preterm infants with birth weight under 1.500 g. Revista Gaúcha de Enfermagem, 36, 84–89. 10.1590/1983-1447.2015.04.50892 [DOI] [PubMed] [Google Scholar]
  28. Simbrunner G (1995). Temperature measurements and distribution of temperatures throughout the body in neonates. In Okken A & Koch J (Eds.), Thermoregulation of sick and low birth weight neonates (pp. 53–62). Springer Berlin Heidelberg. 10.1007/978-3-642-79934-1_5 [DOI] [Google Scholar]
  29. Stoll BJ, Hansen NI, Bell EF, Shankaran S, Laptook AR, Walsh MC, Hale EC, Newman NS, Schibler K, Carlo WA, Kennedy KA, Poindexter BB, Finer NN, Ehrenkranz RA, Duara S, Sánchez PJ, O’Shea TM, Goldberg RN, Van Meurs KP, … Higgins RD (2010). Neonatal outcomes of extremely preterm infants from the NICHD Neonatal Research Network. Pediatrics, 126, 443–456. 10.1542/peds.2009-2959 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Stoll BJ, Hansen NI, Bell EF, Walsh MC, Carlo WA, Shankaran S, Laptook AR, Sánchez PJ, Van Meurs KP, Wyckoff M, Das A, Hale EC, Ball MB, Newman NS, Schibler K, Poindexter BB, Kennedy KA, Cotten CM, Watterberg KL, D’Angio CT, … Higgins RD (2015). Trends in care practices, morbidity, and mortality of extremely preterm neonates, 1993–2012. JAMA, 314, 1039–1051. 10.1001/jama.2015.10244 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Sullivan BA, Grice SM, Lake DE, Moorman JR, & Fairchild KD (2014). Infection and other clinical correlates of abnormal heart rate characteristics in preterm infants. Journal of Pediatrics, 164, 775–780. 10.1016/j.jpeds.2013.11.038 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Thomas KA (2003). Preterm infant thermal responses to caregiving differ by incubator control mode. Journal of Perinatology, 23, 640–645. 10.1038/sj.jp.7211002 [DOI] [PubMed] [Google Scholar]
  33. Ussat M, Vogtmann C, Gebauer C, Pulzer F, Thome U, & Knüpfer M (2015). The role of elevated central-peripheral temperature difference in early detection of late-onset sepsis in preterm infants. Early Human Development, 91, 677–681. 10.1016/j.earlhumdev.2015.09.007 [DOI] [PubMed] [Google Scholar]
  34. Yee WH, Soraisham AS, Shah VS, Aziz K, Yoon W, Lee SK, & Canadian Neonatal Network. (2012). Incidence and timing of presentation of necrotizing enterocolitis in preterm infants. Pediatrics, 129, e298–e304. 10.1542/peds.2011-2022 [DOI] [PubMed] [Google Scholar]

RESOURCES