Parth J. Upadhyay (1), Aartje Maria van der Kuijl (1), Nienke J. Vet (2), Sebastiaan C. Goulooze (1), J.G. Coen van Hasselt (1) , Saskia N. de Wildt (2,3), Catherijne A.J. Knibbe (1,4), Elke H.J. Krekels (1)
(1) Division of Systems Biomedicine and Pharmacology, Leiden Academic Centre for Drug Research, Leiden University, Leiden, The Netherlands, (2) Department of Pediatrics, St. Antonius Hospital, Nieuwegein, The Netherlands, (3) Department of Pharmacology and Toxicology, Radboud University, Nijmegen, The Netherlands, (4) Department of Clinical Pharmacy, St. Antonius Hospital, Nieuwegein, The Netherlands
Introduction: Protocolised sedation in critically-ill children guides the selection of sedatives and their dose during mechanical ventilation. Treatment failure occurs when patients remain undersedated despite dose increases, which is treated with the addition of another sedative to the regimen. Treatment failure and subsequent addition of additional sedatives can happen repeatedly during long-term treatment, increasing the risk of self-extubation and patient discomfort. Early prediction of treatment failure is therefore important. We investigated the hazard of treatment failure in protocolized sedation in mechanically ventilated critically-ill children using a repeated time to event analysis.
Methods: Data from a randomized controlled trial[1] in mechanically ventilated children (n = 113) observed from admission until extubation, were included for analysis. An event was defined as the addition of a new sedative at least three hours after the administration of the previous additional sedative or after the highest dose escalation of the previous sedative. The event hazard for treatment failure was estimated using various hazard distributions such as constant, Gompertz and Weibull in NONMEM v7.4.3. Given the sparse number of events, log-normal inter-individual variability in baseline hazard was tested using Laplacian estimation, as well as other estimation methods better suited to sparse data, such as SAEM and importance sampling[2]. A simultaneous dropout hazard model was implemented to test whether patient’s extubation, either due to improvement in health or death, confounded the estimation of the event hazard. Hazard distributions similar to the event hazard were tested for the baseline dropout hazard structure. A covariate preselection was performed based on Schoenfeld-like residuals[3] followed by stepwise covariate modelling where significant covariates were incorporated in a forward inclusion step (p < 0.05) and retained if significant in a backward elimination step (p < 0.01) for the event as well as the combined event and dropout hazard models. Continuous and categorical covariates such as bodyweight, surgery and number of sedatives at trial commencement were tested. Time varying covariates such as disease severity were incorporated for within- and between- individual effect based on the method by Wählby et al. [4].The model was evaluated using visual predictive checks (VPC) and kernel-based visual hazard comparison (kbVHC) [5]. Bootstrap resampling of the final model was performed to assess model robustness.
Results: 67 patients had no events of treatment failure, 34 patients had at least one event, and 12 patients had two or more events. A Gompertz function best described the decreasing hazard (h) of treatment failure over time (h(t) = 0.00673*exp(-0.00358t)). Dropout hazard (dh) was described using a constant function (dh(t) = 0.00659). No statistically significant inter-individual variability in baseline hazard could be identified, therefore Laplacian estimation was used in the covariate analysis. Stepwise covariate analysis identified that increases in daily obtained disease severity (PELOD) score, indicating worsening health within the individual, was associated with an increased hazard of treatment failure (0.0522/unit increase in PELOD). Hazard of extubation (i.e. dropout)was also affected by disease severity, where decreases in PELOD score between and within individuals, indicating better health compared to the population or improvement within the individual, was associated with a higher hazard of dropout (-0.0478 and -0.0432 /unit increase of PELOD, respectively). Due to a small number of cases (n = 8), a separate dropout hazard for dying patients who dropped out without improvement in PELOD score, could not be estimated. Bootstrap resampling (n = 1000) of the final model identified robust parameter estimates with little bias (< 1%) except of the Gompertz base hazard model (38% and 113% for the two parameters). Both kbVHC and VPC demonstrated good agreement between observed and model predicted events.
Conclusion: Our RTTE analysis estimated a decreasing hazard of treatment failure and an increase in the hazard of treatment failure associated with worsening disease in critically-ill children receiving protocolized sedation during mechanical ventilation, which can be used to identify patients most at risk for requiring additional sedatives.
References:
[1] N. J. Vet et al., Intensive Care Med, 42:233–244, 2016
[2] K. E. Karlsson et al., AAPS J., 2011
[3] S. C. Goulooze et al. AAPS J., 2019
[4] U. Wählby et al. Br. J. Clin. Pharmacol., 2004
[5] S. C. Goulooze et al. AAPS J., 2018
Reference: PAGE 29 (2021) Abstr 9838 [www.page-meeting.org/?abstract=9838]
Poster: Drug/Disease Modelling - Paediatrics