Stochastic Resonance Effects on Apnea, Bradycardia, and Oxygenation: A Randomized Controlled Trial

Clinical Setting

Study infants were cared for in the NICU (level IIIB designation) at Beth Israel Deaconess Medical Center (BIDMC), a Harvard Medical School teaching hospital.

Study Design

This was a randomized crossover study in which the patient served as his or her own control to compare the effects of SR stimulation for each infant.

Study Population

Enrollment occurred from April 2012 to July 2014. Eligible infants were born at a gestational age of <36 weeks, with a postmenstrual age (PMA) of <45 weeks at the time of study (calculated by adding gestational age at birth in weeks to the number of weeks since birth). Infants had at least 1 clinically documented apnea, bradycardia, and/or oxygen desaturation event before enrollment. We excluded infants who were receiving mechanical ventilation or continuous positive airway pressure at the time of the study. There were no birth weight restrictions. All decisions pertaining to diagnosis and treatment of the enrolled infant, including treatment with caffeine and/or supplemental oxygen, were at the discretion of the local clinician providing care in the NICU. The BIDMC Institutional Review Board approved the study, and Harvard Medical School Institutional Review Board ceded review.

Study Protocol

All families provided written informed consent to participate in the study. No remuneration was offered or provided to any family for participating. The protocol consisted of a baseline period of at least 1 hour with no SR stimulation followed by up to two 3- or 4-hour intervention periods (Fig 1). During an intervention period, each infant received 30-minute alternating intervals of SR stimulation and no SR stimulation. The first intervention period was randomly assigned to begin with stimulation either on or off (Fig 1), whereas the second intervention period automatically began with the opposite on/off state as the previous intervention. The session ended with a baseline period of at least 1 hour with no stimulation. The infant’s vital signs were simultaneously and continuously recorded during the entire session. All other aspects of the infant’s care, including feeding, clinical procedures, and assessments, were unaffected by participation in this study.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Schematic of a common protocol sequence.

SR Stimulation

We replaced the standard mattress in the infant’s crib or isolette with a custom-built SR mattress. SR stimulation was delivered in the form of gentle, low-amplitude displacements at the surface of the mattress with a frequency bandwidth of 30 to 60 Hz and displacements of 10 to 20 microns. In 13 infants we used a signal generator (Balance Engineering, Lexington, MA) and mattress (TheraSound), which provided vibration of the entire bedding surface, described previously by Bloch-Salisbury et al.⁸ In the remaining 23 infants, we used the Wyss Institute–developed signal generator and mattress, which contains an isolation unit that significantly dampens vibration to the rostral third of the mattress.⁹ Otherwise, the 2 mattresses provided identical stimulation. The equipment underwent rigorous validation procedures at the Wyss Institute, including frequent maintenance checks throughout the study period.

Data Acquisition

Study data were recorded by using the VueLogger Patient Monitoring System, a novel data-acquisition system, developed by the Wyss Institute. The VueLogger is a portable, computer-based system that connects to the existing Philips “Intellivue” patient bedside monitor used in the NICU, allowing it to retrieve and record the same physiologic signals that are gathered for clinical purposes without placing additional sensors on the patient. The VueLogger records the signals captured by the patient monitor without alteration or filtering.

The following clinical signals were recorded: heart rate (HR), respiratory rate (RR), electrocardiogram, peripheral capillary oxygen saturation, and pulse plethysmography (Nellcor oximeters). The VueLogger also recorded the mattress vibration signal, as well as the light and sound levels in the crib/isolette. A sound meter (Extech 407764) to record sound intensity and a light meter (Hioki 3423 Lux HiTester) to measure brightness and changes in light levels were both placed next to the infant. Sound and light were continuous measures. Data were recorded at sampling rates that varied from 1 Hz (oxygen saturation) up to 500 Hz (electrocardiogram) and streamed to the VueLogger’s hard-disk at the bedside, and then exported for analysis. Clinical monitoring information (eg, HR, RR) was collected directly from the patient bedside monitor without alteration and was used to retrospectively determine the incidence of relevant events. Our review and analysis indicated no systematic issues with the collection, management, or quality of the data acquired with the use of the VueLogger system.

Variables

We had 3 outcome variables, oxygen desaturation, bradycardia, and apnea events, for which we studied the number and duration of events. For oxygen desaturation and bradycardia events, we also studied the intensity of the events. We defined our outcomes on the basis of BIDMC NICU clinical definitions. An oxygen desaturation event was defined as oxygen saturation <87% for infants <35 weeks’ PMA and 90% for infants with a PMA of ≥35 weeks. A bradycardia event was defined as an HR of <100 beats per minute for infants <34 weeks’ PMA and <80 beats per minute for infants with a PMA of ≥34 weeks. For oxygen desaturation and bradycardia, there was no specified length of the event (ie, it had to last for at least 1 second to meet the definition). If there were successive values that fell below the threshold, they were counted as part of 1 event (ie, if the oxygen level dropped below the threshold for 10 seconds in a row without rising above the threshold, that would be counted as 1 not 10 events). Apnea was defined as a pause in respiration (ie, RR of zero) of ≥10 seconds.

The duration of an event was defined as how many seconds the event lasted before the parameter returned to normal. For both oxygen desaturation and bradycardia events, we also studied intensity as a composite outcome that took into account how far below the threshold the parameter dropped and how long the event lasted. Intensity was calculated as the area under the curve for the event. All outcome parameters were ascertained automatically (ie, no manual coding of outcomes) by computer code during analysis.

Other neonatal variables were obtained from the medical record, including information about the infant at birth (ie, weight and gestational age) and at the time of study (ie, current weight, PMA, and whether the infant was being treated with supplemental oxygen and/or caffeine). Race and ethnicity were per maternal self-report on the birth certificate.

Statistical Analysis

We compared each infant’s SR stimulation “on” periods with the SR stimulation “off” periods, such that each infant served as his or her own control. We used Poisson regression with random effects for both bivariate and multivariable analyses. For each individual infant we analyzed the number, duration, and intensity of oxygen and bradycardia desaturation events. To demonstrate an effect of SR stimulation, analysis of the measured data would have to show that the number, duration (in seconds), and intensity (in cumulative percentage below the locally defined limit) of the clinically defined outcome were lesser, shorter, and weaker, respectively, for those events when SR stimulation was on. Independent clinical variables (eg, current weight, PMA, ± oxygen, ± caffeine) that were statistically significant in bivariate analyses were individually added to the significant SR stimulation Poisson models to assess the effect of SR stimulation, controlling for the selected clinical variable.

Mixed Poisson modeling is an appropriate statistical modeling because the “mixed” aspect allows fitting random-effects terms for each infant within each session to absorb individual differences in risk of events. Counts are all positive integers, and for rare events, the Poisson distribution (rather than the normal) is more appropriate because the Poisson mean is >0. The typical Poisson regression model expresses the log outcome rate as a linear function of a set of predictors. Poisson modeling is classically a method used to test for differences in outcome measures that are explicitly count variables (ie, whole numbers) that may be applied to dependent variables that are continuous as well.¹⁰ A Poisson regression is a test of whether the change in regression parameters will influence the probability of a unit increase. In this sense, Poisson regression is a close cousin of the logistic regression that is not limited to modeling the occurrence or nonoccurrence of a single outcome. We used Poisson regression to model the effect of SR stimulation on the occurrence of each outcome, confirming that residuals were uncorrelated across model-predicted values according to assumptions of Poisson modeling. Poisson regression was conducted with SAS version 9.4 (2000; SAS Institute, Cary, NC) by using the proc glimmix procedures.