1 of 32

Extubation Readiness in the �NICU

2 of 32

Outline

A review of neonatal ventilation
The negative effects of prolonged mechanical ventilation
Factors associated with successful extubation
Extubation readiness tools and algorithms
Summary

3 of 32

Content prepared by

�Dr. Lindsey Knake MD

@LindseyKnake

4 of 32

A review of neonatal ventilation

5 of 32

And when Elisha came into the house, behold the child was dead… He went up and lay upon the child and put his mouth upon his mouth… and the flesh of the child waxed warm… and the child opened his eyes

- Old Testament 2 Kings 4:32-35

“

6 of 32

Assisted Ventilation of the Neonate 6^th ed

7 of 32

Robertson, A. F. Reflections on Errors in Neonatology: II. The “Heroic” Years, 1950 to 1970. J Perinatol 23, 154–161 (2003).

�

8 of 32

Robertson, A. F. Reflections on Errors in Neonatology: II. The “Heroic” Years, 1950 to 1970. J Perinatol 23, 154–161 (2003).

�

9 of 32

Dreyfuss D et al. Am Rev Respir Dis. 1988;137:1159-1164

10 of 32

Shalish, W., Keszler, M., Davis, P. G. & Sant’Anna, G. M. Decision to extubate extremely preterm infants: art, science or gamble? Archives Dis Child - Fetal Neonatal Ed 107, 105–112 (2022).

�

Decision to extubate extremely preterm infants: art, science or gamble? Wissam Shalish ,1 Martin Keszler ,2 Peter G Davis ,3 Guilherme M Sant’Anna

Great review of all the studies showing the harmful effects of extubation failure
Jensen article has been quoted frequently when looking at the harmful effects of prolonged mechanical ventilation since this was a large study that adjusted for the cumulative duration of mechanical ventilation in <1000g neonates

Original hypothesis of this article was that the number of ventilation course and reintubations would increase risk of BPD after adjusting for the duration of mechanical ventilation it was found that duration of mechanical ventilation was actually the stronger predictor of both outcomes. The risk of BPD remained increased only among infants exposed to 4 or more ventilation courses.

However the other three studies studies do show increased death with early reintubations, with some of those studies reporting deaths from IVH, massive pulmonary hemorrhage or NEC.
This is why many centers may be more cautious on extubating the "nano-premies" 22-24 weeks until they are older, more stable, and you know the PDA is not causing a huge problem.

11 of 32

The right time to extubate

12 of 32

Shalish, W., Keszler, M., Davis, P. G. & Sant’Anna, G. M. Decision to extubate extremely preterm infants: art, science or gamble? Archives Dis Child - Fetal Neonatal Ed 107, 105–112 (2022).

�

13 of 32

Shalish, W., Keszler, M., Davis, P. G. & Sant’Anna, G. M. Decision to extubate extremely preterm infants: art, science or gamble? Archives Dis Child - Fetal Neonatal Ed 107, 105–112 (2022).

�

14 of 32

Shalish, W. et al. Age at First Extubation Attempt and Death or Respiratory Morbidities in Extremely Preterm Infants. J Pediatrics (2022)

15 of 32

Shalish, W. et al. Age at First Extubation Attempt and Death or Respiratory Morbidities in Extremely Preterm Infants. J Pediatrics (2022)

16 of 32

Kanbar, L. J. et al. Automated prediction of extubation success in extremely preterm infants: the APEX multicenter study. Pediatr Res 1–9 (2022)

17 of 32

Shalish, W., Latremouille, S., Papenburg, J. & Sant’Anna, G. M. Predictors of extubation readiness in preterm infants: A systematic review and meta-Analysis. Archives Dis Child - Fetal Neonatal Ed 104, F89--F97 (2019).

�

18 of 32

Shalish, W., Latremouille, S., Papenburg, J. & Sant’Anna, G. M. Predictors of extubation readiness in preterm infants: A systematic review and meta-Analysis. Archives Dis Child - Fetal Neonatal Ed 104, F89--F97 (2019).

�

19 of 32

Kanbar, L. J. et al. Automated prediction of extubation success in extremely preterm infants: the APEX multicenter study. Pediatr Res 1–9 (2022)

scatter plot from the Figure from Apex study supplemental material

Remember, this was a study that enrolled infants only if the clinician deemed the infant was ready for extubation.
Not surprising the >28 week and >1100g infants successfully extubated (however, hard to say that would be universal across all neonates).

However, what's more intriguing is the variability in the smaller babies. We see that there are some 24 week sub-600 gram infants that were successfully extubated and some 27 week 1000g babies who failed.
This study had 23% of infants <1250 g who failed extubation when clinicians thought they were clinically ready.
Thus, clinical expertise alone still resulted in 23% failure rate which is why the quest for a clinical decision support tool to add in this decision to extubate exists.

20 of 32

Extubation readiness tools & algorithms

21 of 32

Shalish, W., Latremouille, S., Papenburg, J. & Sant’Anna, G. M. Predictors of extubation readiness in preterm infants: A systematic review and meta-Analysis. Archives Dis Child - Fetal Neonatal Ed 104, F89--F97 (2019).

�

2017 paper: Predictors of extubation readiness in preterm infants: a systematic review and meta-analysis: Wissam Shalish, Samantha Latremouille, Jesse Papenburg, Guilherme Mendes Sant’Anna

The conclusion of this was review is that there is a lack of strong evidence to support using extubation readiness tests in preterm infants.
Current predictors have low overall accuracy and add little benefit in the identification of extubation failures.
Although SBTs are attractive assessment tools, higher quality studies are needed to determine the best duration, level of PEEP and definition of test pass/failure to guide their use in the most vulnerable infants (become unstable during the tests likely because SBTs may cause atelectasis ).

Review shows the sensitivity/specificity of different tests (overall higher sensitivity (high true positive) meaning we can predict who will succeed extubation but lower specificity (not great at predicting the true negatives and who will actually fail). Also not good to have too many false negatives because if we say this baby is going to fail extubation but actually doesn't then the clinician will keep the baby intubated unnecessarily. )\

studies have looked at tidal volume, spontaneous minute ventilation, variability index of different breathing patterns, SBT, SBT + variability index of breathing patterns, diaphragmatic tension and tension time index of respiratory muscles.

22 of 32

Shalish, W. et al. Assessment of Extubation Readiness Using Spontaneous Breathing Trials in Extremely Preterm Neonates. Jama Pediatr 174, 178--185 (2020).

�

23 of 32

MUELLER, M. et al. Predicting Extubation Outcome in Preterm Newborns&colon; A Comparison of Neural Networks with Clinical Expertise and Statistical Modeling. Pediatr Res 56, 11–18 (2004).

�

The first model is actually from 2004. Predicting extubation outcomes in Preterm Newborns: A comparison of Neural Networks with Clinical Expertise and Statistical Modeling https://www.nature.com/articles/pr2004166

Artificial neural network used to predict extubation success 183 infants BW 900 -1500g
Extubation failure if reintubated within 48 hours (controversial definition of extubation failure as we've discussed before) with 19% extubation failure rate
51 clinical variables chosen by physicians as important variables (easily available at the bedside and routine evaluations such as RT vent setting recordings)
When using all 51 variables the model was overfitting for the training data (meaning the AUC of the ROC curve reached almost 1 or perfect for the training data but fell to 0.5 or "chance" for the test data).
Best balance was to use 13 variables for the model (4 input variables with most sensitive so consider most important were pH, SaO2, GA, PEE) AUC for the ROC for both training and test was around 0.8.

24 of 32

Mueller, M., Wagner, C. C., Stanislaus, R. & Almeida, J. S. Machine learning to predict extubation outcome in premature infants. 2013 Int Jt Conf Neural Networks Ijcnn 2013, 1–6 (2013).

�

Almost the same group of authors 9 years later, now with more patients and including all preterm babies. 486 patients with 12% extubation failure
Tested different type of machine learning models this time, including support vector machine (SVM), naive Bayesian classifier (NBC), boosted decision trees (BDT), and again artificial neural network and multivariable logistic regression algorithms.
Tried both full data sets of 54 variables or only statically significant input variables but validation data AUC was never over 0.8
Due to the uniformly low performance across all algorithms with AUC < 0.8 no decision support tool was developed
Limitations may have been due to incorrect data being included in the algorithm, changing practices and weaning protocols in the NICU during this time period (different patient population including all premies instead of the 900 - 1500g patients in the previous paper), only 12% failure rate for the ML algorithm to learn on, and finally the authors also mentioned that maybe reading of pages from handwritten medical records (data quality issues of missing forms, data points, illegible handwriting, and data entry errors that required extensive data cleaning and imputation of missing valves).

25 of 32

Gupta, D. et al. A predictive model for extubation readiness in extremely preterm infants. J Perinatol 39, 1663--1669 (2019).

Not a machine learning model but a model that did create a prediction algorithm available
This was a retrospective study that included infants with birth weights ≤1250 g at Wayne State University, who were admitted to a tertiary NICU over a 7-year period, received mechanical ventilation and had an elective extubation attempt within 60 days of age. Perinatal and periextubation characteristics were compared in the successful and failed extubation groups.
312 were included and considered extubation success if >72 hours extubated. Extubation succeeded in 73% and failed in 27%. Adjusted factors associated with successful extubation included greater gestational age, chronologic age, pre-extubation pH and lower preextubation FiO2, along with lower “peak” respiratory severity score in the first 6 h of age.
Found that the model had sensitivity/specificity of 87%/53%, 76%/66%, and 54%/81% for cutoffs of 60%, 70%, and 80% probability of extubation success, respectively.

26 of 32

Gupta, D. et al. A predictive model for extubation readiness in extremely preterm infants. J Perinatol 39, 1663--1669 (2019).

http://www.extubation.net

27 of 32

Chakraborty, M., Watkins, W. J., Tansey, K., King, W. E. & Banerjee, S. Predicting extubation outcomes using the Heart Rate Characteristics index in preterm infants: a cohort study. Eur Respir J 56, 1901755 (2020).

28 of 32

Chakraborty, M., Watkins, W. J., Tansey, K., King, W. E. & Banerjee, S. Predicting extubation outcomes using the Heart Rate Characteristics index in preterm infants: a cohort study. Eur Respir J 56, 1901755 (2020).

29 of 32

https://www.heroscore.com/extubation/

30 of 32

Kanbar, L. J. et al. Automated prediction of extubation success in extremely preterm infants: the APEX multicenter study. Pediatr Res 1–9 (2022)�

Used data from the APEX study which looked at SBTs in <28 GA infants prior to planned physician extubation to look at 5-min ET cpap trials (recording cardiopulmonary signals)
259 infants from 5 NICUs (2013-2018) with BW <1250g
they included cardio respiratory data for 1 hour before extubation and SBT

They did used machine learning to help choose principle component features meaning the features that seemed most correlated to predicting extubation success were included int he model then there was a two part classifier where the ML model first the high risk population was selected useing a GA and BW cut off by the machine.

Best outcomes:

The classifier had a sensitivity 70% , specificity 75% with AUC of 0.75
As an additional clinical-decision tool, the classifier would have led to an increase in extubation success from 82% to 93% but misclassified 60 infants as failures who would have been successfully extubated
Did perform better for infants extubated in the first 7 days of life maybe less clinically ill or complicated neonates

Summary: overall high sensitivities but low specificities (meaning good at predicting which babies will succeed but not as good at identifying the extubation failures)

31 of 32

Summary

Neonatal mechanical ventilation has greatly improved in the past 70 years

Current methods include lung protective strategies to avoid volutrama

Prolonged mechanical ventilation and failed extubations can both increase risk for BPD and prolonged hospitalizations
There are no consensus guidelines for predictors for extubation success

Spontaneous breathing trials caused clinical instability in preterm neonates

Current prediction models have high sensitivity but low specificity

Multi-center research with larger patient populations are needed to improve extubation models.

32 of 32

Thank You

www.the-incubator.org