Introduction

Mechanical ventilation (MV) is a lifesaving intervention for many preterm infants in the Neonatal Intensive Care Unit (NICU). Despite the lifesaving nature, neonatal MV is not without risk. In preterm infants in the NICU, prolonged duration of MV is associated with impaired brainstem development, abnormal white matter maturation, and lower scores on neurodevelopmental outcomes (NDO) testing at preschool age [1, 2]. For each additional day of MV in this population, the odds of neurodevelopmental impairment at 24 months of age increases by 8%, with the motor domain most significantly impacted [2]. While the strongest evidence exists for prolonged MV and impaired motor outcomes, other subdomains of NDO including cognitive and language outcomes have also been shown to be negatively affected by MV [3,4,5].

Unplanned extubations (UEs) are the most common adverse event occurring during MV in the NICU, noted in 14–41% of very low birth weight (VLBW) infants during their hospitalization [6, 7]. While common, UEs often result in significant short-term harm including oxygen desaturation, hemodynamic instability, and reintubation [8,9,10,11,12]. Infants who have UEs have also been observed to have poorer in-hospital outcomes including prolonged MV duration and increased length of stay compared to infants without UE in the NICU [13].

Although the short-term risks of UE are well known and their association with longer MV duration has been established, the association of UE with NDO in preterm infants has not yet been reported. Since UEs are associated with an increase in MV duration and each additional day of MV is associated with poorer NDO, we hypothesized that UEs may be associated with poorer NDO (potentially mediated by the MV duration). The objective of this study was to determine the association between UEs and NDO at 20–30 months corrected age (CA) in preterm infants.

Materials/subjects and methods

Study design/setting

We performed a retrospective, observational, matched cohort study [14] in the 98-bed, level IV regional academic NICU of the Vanderbilt University Medical Center (VUMC). All VLBW infants (<1500 g at birth) admitted to the NICU within 72 h of birth between February 1, 2014 and December 31, 2018, without chromosomal or congenital anomalies, who received MV through an oral endotracheal tube (ETT) for ≥1 h, and had NDO testing performed between 20 and 30 months CA at VUMC were eligible for inclusion in the cohort. These study years were chosen (and the subsequent convenience sample of infants) as UEs and NDOs were reliably collected during this time and these years allowed time to assess NDO testing prior to study analyses beginning. During the study period, the NICU had no ventilator weaning protocol or sedation guidelines. Consistent with evidence-based recommendations, ventilated infants did not routinely receive sedatives or opioid medications while intubated [15, 16]. The VUMC Institutional Review Board approved the study with a waiver of consent.

Exposure variable

The primary exposure was the occurrence of one or more UEs during the NICU hospitalization. The operational definition of a UE in the VUMC NICU during the study period was any removal of an ETT that the medical team did not plan including self-extubation by the patient, inadvertent dislodgement during routine nursing or medical care, or intentional removal in the setting of an acute resuscitation event or period of patient decompensation during which ETT placement and patency were in question. Elective reintubation to increase ETT size was not considered a UE. During the study period, the VUMC NICU used a standardized protocol to assess whether an ETT was in the trachea and patent consisting of auscultation for breath sounds, assessment for expiratory waveforms on the ventilator, and use of an end tidal CO2 detector. Direct laryngoscopy to determine if the ETT was in the trachea was not routinely used in the NICU during the study period [8, 17]. We used four data sources to prospectively collect UEs: (1) root cause analysis forms completed by the bedside team immediately after an UE, (2) daily medical record review of all intubated infants in our NICU, (3) review of all reports of UEs in our hospital’s voluntary incident reporting system, and (4) an ongoing record kept by the respiratory therapists of all UEs in the NICU. We previously reported the sensitivity of each of these sources to identify UE in our NICU, with daily medical records review being the most reliable method (detected 96% of UEs) [8]. A study team member (LDH) abstracted potential UEs from the four data sources, and these were reviewed monthly by a multidisciplinary group to determine if each event met objective criteria for UE. Infants in our NICU who had a UE were generally given a trial of non-invasive support unless they were physiologically unstable.

Outcome variable

Neurodevelopmental outcomes were evaluated using the Bayley Scales of Infant and Toddler Development-Third Edition (Bayley-III) at 2 years (20–30 months) CA [18]. The Bayley-III is a standardized instrument used to assess development across a number of domains including cognitive, language, and motor scored with a mean of 100 and SD of 15 and five norm referenced subscale scores: Cognitive, Expressive Communication, Receptive Communication, Fine Motor, and Gross Motor Subscales with a mean of 10 and SD of 3 [5, 18]. It is standard of care for all VLBW infants discharged from the VUMC NICU to have neurodevelopmental follow-up at the VUMC Neonatal Developmental Clinic following discharge from the NICU. Infants undergoing NDO testing were evaluated by a small group of certified examiners, including developmental and behavioral pediatricians as well as clinical psychologists. These assessors were blinded to the exposure variable (UEs), as these are not routinely documented in the NICU discharge summary. For each infant, we ascertained the Bayley-III motor composite score (including both gross and fine motor subdomain scaled scores), language composite score (including both receptive and expressive communication subdomain scaled scores),and cognitive outcome scores.

Matching procedure

The decision to perform matching (as opposed to more common techniques such as regression modeling) was to account for the exposure of MV duration as MV duration is associated with both the exposure (UE) and the outcome (NDO) and is potentially on the causal pathway. Given that infants who received MV for longer periods of time are at higher risk for both UE and poorer NDO, we chose to match cases and controls on multiple clinical characteristics including duration of MV at time of UE. Infants exposed to a UE were compared with similar infants who were “at risk” for UE by matching cases (+UE) 1:1 with controls (−UE) using a greedy nearest-neighbor matching algorithm without replacement [19] based upon gestational age (GA), Clinical Risk Index for Babies (CRIB) score [20] at birth, and cumulative duration of MV at the time of UE (for the cases). In cases where an infant was exposed to >1 UE (26% of the cohort), matching was performed based on the MV day of the first UE. For example, if an infant had a UE on day 5 of MV, all infants without a UE during their hospitalization with at least 5 MV days were identified and the infant with the most similar GA and CRIB score on MV day 5 was selected to form the matched pair. Matching was performed in Stata MP v17.1 (StataCorp, College Station, TX).

Statistical analysis

Descriptive statistics were generated for demographics, clinical characteristics at matching, and outcomes. The median and interquartile range (IQR) was computed for all continuous variables as were differences between cases and their matched controls. Nominal variables were represented using raw counts and percentages. Covariate imbalance was assessed before and after matching for all demographic and clinical characteristics using standardized differences[21]. Final regression models included all covariates with an absolute standardized difference ≥0.1 [22] after matching as sample size allowed. These included sex, birth weight, inborn status, race, antenatal steroid receipt, intraventricular hemorrhage (IVH), periventricular leukomalacia (PVL), necrotizing enterocolitis, and receipt of high frequency ventilation at time of matching. We calculated adjusted odds ratios (aORs) for all outcomes using multivariable ordinal cumulative probability models [23, 24] to account for nonindependence of the matched pairs. The aORs generated from these models describe the odds that cases had higher outcomes (e.g., Bayley-III composite motor scores, Bayley-III cognitive score, etc.) than their matched controls. All statistical procedures were performed in Stata MP v17.1 (StataCorp, College Station, TX) and a p value < 0.05 was considered significant.

Results

Baseline characteristics

During the study period, n = 736 VLBW infants were ventilated in the NICU with 533/736 (72%) infants meeting study inclusion criteria (Fig. 1). Of these infants, 92/533 (17%) had one or more UEs while in the NICU. A total of 42/92 (46%) infants with a UE had NDO data available for review. Thirty-nine infants did not have complete follow-up data documented in the Neonatal Development Clinic and 11 infants died before NICU discharge. The remaining 42 patients with a UE (cases) were matched to suitable controls. Though cases and controls showed improved covariate balance after matching, differences remained in several demographic and clinical characteristics. Study infants had a median gestational age of 26 weeks (IQR 25–27). Consistent with prior studies, infants with a UE had a longer post-matching duration of MV (median 11 vs. 4 days) compared to controls. Complete patient demographics and clinical characteristics are shown in Table 1. Timing of UEs, reasons for UE, and immediate consequences of UE are provided in Supplemental Table.

Fig. 1: Flow diagram of study subjects.
Fig. 1: Flow diagram of study subjects.
Full size image

179 infants met study inclusion criteria and had NDO testing available. Of these 179 infants, 42 (23%) had one or more UEs and a total of 42 matched pairs were formed.

Table 1 Demographics at birth, comorbidities, and variables at time of matching.
Full size table

Neurodevelopmental outcomes

Neurodevelopmental Outcomes are shown in Table 2. Infants with one or more UE had Bayley-III motor composite scores that had a lower median difference IQR of −6 [−18,10] compared to their matched cases. After adjusting for confounding, this difference was not statistically significant (aOR 0.58, 95% CI 0.25–1.38). Similarly, infants exposed to UE had lower Bayley-III gross motor scaled scores and lower Bayley-III fine motor scaled scores, however neither of these were significant in adjusted models (Table 2).

Table 2 Neurodevelopmental outcomes for matched pairs (n = 42 pairs).
Full size table

Cases (UE + ) had lower Bayley-III language composite scores with median difference IQR within matched pairs of −3 [−15,10] and Bayley-III cognitive scaled scores with median difference IQR within matched pairs of −1 [−3,2]. In adjusted models, cases (UE + ) had statistically lower odds of having the higher overall Bayley-III cognitive composite scores (aOR 0.34, 95% CI 0.14–0.8) and Bayley-III cognitive scaled score (aOR 0.32, 95% CI 0.14–0.74).

Discussion

We observed an association between UEs in the NICU in VLBW infants and lower Bayley-III cognitive scores at 2 years CA with cases (+UE) having lower cognitive scores compared to matched controls after adjusting for known confounders. Furthermore, consistent with prior studies, we found that infants who had one or more UEs in the NICU had a one-week longer MV duration than those who did not. We did not find a significant association between UEs and motor or language outcomes at 2 years CA after adjusting for known confounders.

Our study found a significant association between UE and lower cognitive scores at two years of age. Congruent with our study, multiple prior works have shown an association between MV duration and poorer cognitive outcomes [3,4,5]. A prospective study of 260 VLBW infants from birth to adulthood predicted a decrease of 0.43 IQ points for each day of MV despite adjusting for other confounders [25]. Additionally, a prospective longitudinal study found that duration of MV predicted mathematical abilities at eight years of age further suggesting that prolonged periods of mechanical ventilation after preterm birth may cause longer-term cognitive deficiencies [26].

While we found no association between UE and NDO in the motor domain, the strongest evidence in the literature supports an association between prolonged MV and impaired motor outcomes [1, 2]. Ventilated preterm infants are at risk for known complications that impact motor development such as intraventricular hemorrhage (IVH) and periventricular leukomalacia (PVL). Even in the absence of these known complications, infants on MV are at risk for neurodevelopmental impairment, specifically within the motor domain [5, 27]. A prospective study by Brown et al. found that a longer duration of ventilation in infants less than 32 weeks’ or less than 1500 g at birth was associated with poorer tone, reflexes, and spontaneous movement [27, 28] Our sample size may not have allowed the power needed to prove subtle differences in motor outcomes between cases and controls. We attempted to address this by including a cohort of nearly five years of ventilated infants. Even with this large period, only 42 matched pairs could be formed. Larger, multicenter cohort studies are needed to further define if an association between UE and motor outcomes truly exists.

Our study has important additional limitations which should be noted. First, our study may have been limited by residual confounding that we were unable to capture in our models. We utilized a matching algorithm to generate a more balanced cohort and to minimize confounders between infants who had a UE and those who did not. Though covariate balance improved following matching, cases and controls remained dissimilar in several measured areas. We attempted to address this limitation further by adjusting our analyses for covariates with known differences, but this may have been incomplete. Second, three of the four data streams we used to identify UEs were dependent on clinicians’ self-reports introducing the possibility of misclassification of UE. To avoid this potential misclassification, we reviewed medical records of all intubated infants to ensure complete capture of UEs. Third, due to the retrospective and pragmatic data sets used in our study, our conclusions may have been affected by our low follow-up rates at 24 months CA and the resulting small sample size. In post hoc analyses, the infants with NDO data available at 24 months CA appeared more ill at birth (poorer Apgar scores, lower GA) and likely had a higher baseline risk for poorer NDO (data not shown). The use of more controls for each case (e.g., 2:1 instead of 1:1) may have increased the ability to detect differences in NDO, however the size of our available cohort did not allow this technique. Additionally, though we replicated a previously reported association between UE and increased duration of MV, we cannot prove that a causal relationship between these two variables exists. Finally, although the Bayley-III is the most widely used scoring system for NDO, it is important to note that defining NDO in toddlers is imprecise and single timepoint outcomes at 2 years of age may not be predictive of longer-term neurodevelopmental outcomes [29,30,31,32].

Conclusion

In summary, our study found that unplanned extubations were associated with lower cognitive outcomes at two years CA. Additionally, consistent with prior studies, we found that infants who had one or more UEs in the NICU had longer duration of MV than those who did not. We did not observe a significant association between UEs and Bayley-III motor or language outcomes at two years CA. Based on our study’s limitations, it is possible that our findings may be due to chance. Larger, multicenter observational studies with longer-term follow-up should be designed to answer this important clinical question and build upon the findings from our study.

Our study suggests that, in addition to the known immediate and short-term adverse outcomes, UEs may have longer-term effects on NDO in VLBW infants. These results can be used to more optimally guide follow-up and early intervention programs, plan prospective interventions aimed to improve NDO, and reduce morbidity associated with these common adverse events in critically ill preterm infants.