In this large registry-based multicentric cohort study of 459 patients with acute tSCI, we found that 72% of patients were liberated from MV and discharged alive from their primary ICU; 12% died, the majority of whom within 28 days from the initiation of MV; 16% were transferred, still ventilated, primarily to other acute-care or rehabilitation facilities.
According to our prediction model, key factors associated with lower probability of successful liberation from MV at discharge from primary ICU were the occurrence of a severe trauma as measured by a high ISS, a non-blunt injury, the presence of a cervical lesion, a clinically complete syndrome on admission, and increasing age. The prediction score built from the linear combination of these predictors was reasonably accurate in predicting weaning outcomes, higher values being associated with increased probability of weaning failure. With the addition of the number of comorbidities, variables relevant for weaning success were also able to predict weaning at any time point considering the competing risk of death.
The proportion of patients that could be weaned from MV in ICU after an acute tSCI in our study (72%) is slightly higher compared to the meta-analytic proportion we found in patients discharged from ICUs from our recently published systematic review (63.1%, 95%CI 45.2–77.9%) (17). This proportion is considerably higher when compared to the proportion of patients discharged from ICUs of Level I trauma centers only (27–29), where it ranged between 16% (27) and 40–45% (28, 29). However, the probability of weaning success appears similar when considering the studies conducted in cohorts comparable to our cohort with regards to patients and injury baseline characteristics, even if considerably smaller (30–32). These findings suggest that baseline characteristics are essential in influencing weaning outcomes. Heterogeneity of care among trauma centers, non-uniform definition of weaning success (and use of weaning protocols), and possible dissimilar discharge criteria could also participate in the dispersion of results. Moreover, our registries did not capture measures of pulmonary mechanical insufficiency, inadequate pulmonary toilet, or sedation which can influence extubation and weaning failures, as suggested in a previous cohort of tSCI patients (33).
The two registries used, St. Michael’s Trauma Registry and RHSCIR, slightly differed with regards to the mechanisms of injury, distribution of the level of injury, proportion and type of extra-spine lesions, and proportion of traumatic brain injury, as well as some baseline characteristics. Despite prevalent cervical lesions (expected to have a lower probability of weaning success), no patients with lumbar lesions, and a global higher number of patients with extra-spine lesions, RHSCIR cohort had a higher probability of weaning success at day 14 and 28, but not at ICU discharge. This may allow for further consideration on the effects of baseline characteristics on weaning outcomes, which may gradually soften over time. Previous studies have shown a relationship between patient’s and lesion baseline characteristics and weaning outcomes. Liebscher et al. (34) found a higher ISS to be associated with a reduced probability of weaning success, Yu et al. (35) the presence of a cervical lesion to predict weaning failure. Moreover, our findings of a complete lesion associated with prolonged requirement for MV has also been previously noted (28).
Our study is based on one of the largest available cohorts of ventilated patients with tSCI, from multiple, independent centers across Canada, allowing for the development of more robust and reliable prediction models for weaning success and time to liberation.
We used the entire cohort for model development and internal validation, following methodological recommendations (21). The robustness of our models is further supported by the consistency of our predictors in predicting weaning success regardless of the time point of assessment (at Day 28 and at ICU discharge) and across weaning-related outcomes (weaning success and time to weaning accounting for the competing risk of death). The few previous studies building regression models often happened to select, as predictors, factors differing across various analyses, and whose clinical correlation with weaning outcomes in some cases appears scarcely plausible (17). Small sample sizes (underpowered to build a prediction model), patient heterogeneity, and initial variable selection methods (e.g., based on series of univariate regression analyses) have led to the selection of inconsistent sets of candidate predictors (27, 30, 34–36).
From the coefficients of our logistic regression model, we derived the BICYCLE score, a prediction score for weaning success at ICU discharge. The ISS, the most frequently reported score in traumatic SCI literature, which performs well as a measure of mortality (37), showed poor performance in predicting weaning outcomes. On the contrary, the BICYCLE score represents an early and valuable prediction tool incorporating the main factors determining weaning success after tSCI, all easily determined after admission. Our score could provide clinicians with a reasonably reliable tool for informing patients and families about the possible prognosis and might help with decision-making about further management of patients with SCI. Use of the BICYCLE score may also assist in directing the level of care required by tSCI patients (ICU versus lower intensity units), and in stratifying tSCI patients when conducting research in this population. The different score quartiles appeared to reflect clearly different incidences of weaning and risks of death (Figs. 2 and 1b). The main advantages of the BICYCLE score consist in the ready availability of the variables used to compute it, the possibility to assess it at an early stage (potentially on ICU admission), and its ease of computation.
Among the variables selected in the prediction model for weaning success and entered in the score, the presence of a cervical lesion deserves particular attention. As displayed in Fig. 1a, the level of the injury alone already provides indications on patient’s prognosis. Patients with cervical SCI experienced the lowest probability of weaning at every time point and seemed to show the highest mortality, although between-group differences for mortality were non-significant (> 0.05 for all comparisons). Among patients with a cervical SCI, those with a C3-or-above injury level experienced weaning exclusively within the first 3 weeks of hospitalization, with no further improvement in the subsequent weeks. Conversely, up to 70% of patients with a C4-or-below injury could be weaned, the cumulative incidence of weaning continuing to increase up to 70 days from ICU admission.
Variables that resulted as relevant predictors of the outcome of weaning success appeared to also predict time to weaning accounting for the competing risk of death. Yet, besides the five variables composing the BICYCLE, the number of comorbidities also appeared to play a role. This suggests that, while not carrying a significant weight on the probability of weaning, the number of comorbidities becomes relevant in influencing the time required to be weaned and mortality.
This study has limitations. First, in the absence of an external validation, generalizability might potentially be limited by specific aspects related to the Canadian healthcare system. Calibration might be needed to ensure good predictive performance in other settings. Second, St. Michael’s Trauma Registry and RHSCIR, like all databases with routinely collected data, have missing values. Moreover, as the data were collected for purposes other than the assessment of weaning outcomes, the set of available variables was potentially not exhaustive for our purpose. Third, we acknowledge that we have no information in terms of weaning outcomes once discharged of the 16% of patients that were transferred still ventilated. However, it must be noted that only 23 of them (corresponding to 5% of the total number of patients enrolled) were discharged within 12 days of ICU admission (the median time to liberation from MV in our cohort). Hence, weaning success after transfer could be assumed as unlikely with consequently low probability of affecting the analysis. Finally, we have no data on readmissions or on the need to restore MV in patients who were discharged from ICU alive and no longer ventilated; analogously, we have only incomplete information about their survival once discharged. These types of information would have given more of a context to our score and would have reassured us on the accuracy (and therefore usefulness) of our model in predicting the ultimate weaning outcome. However, unlike other healthcare contexts, SCI patients in Canada are not likely to experience early discharge due to a relative paucity of long-term acute care facilities, and therefore, the outcome at ICU discharge is a reasonable proxy indicator of patients’ ultimate outcome.