Impact of a posttraumatic cerebral infarction on outcome in patients with TBI: the Italian multicenter cohort INCEPT study

A traumatic brain injury (TBI) is a leading cause of mortality and morbidity mostly among young people; although, its incidence is increasing in older people, particularly in high-income countries [1]. The outcome of a TBI depends on several factors, including the patients’ characteristics, disease severity at admission, and complications arising during its clinical course. Multivariable prognostic models such as the International Mission for Prognosis and Clinical Trial design in TBI (IMPACT) have shown that most prognostic information is contained in a core set of three predictors: age, the Glasgow Coma Scale (GCS) motor score (GCSm), and pupillary reactivity [2]. IMPACT also provided an extended prognostic model, which adds the brain computed tomography (CT) classification and secondary cerebral insults, such as hypoxia and hypotension, to the core variables. Both core and extended IMPACT models focus on identifying prognostic factors at baseline and do not include predictors from the intensive care unit (ICU) stay [3]. Posttraumatic cerebral infarction (PTCI) is a common complication of a TBI in the acute stage of the disease. A PTCI is frequent in patients who die after a moderate or severe TBI, with a reported incidence in post-mortem studies up to 90%. The ante-mortem occurrence rate of a PTCI has been assessed in six single-center studies, of which only one was prospective, and the rate varied between 1.9% and 20.3% [4‐9]. None of these studies investigated whether PTCI adds incremental value to current prognostic models. Therefore, we planned a multicenter, prospective observational cohort study in patients with a moderate or severe TBI to investigate: (1) the impact of PTCI on the 6-month outcome evaluated by the Glasgow Outcome Scale (GOS), (2) if the PTCI adds incremental value beyond that provided by the IMPACT prediction models on the GOS at 6 months, and (3) the occurrence of PTCI in the study population.

In the present prospective observational cohort study (ClinicalTrials.gov Identifier: NCT02430324), we included all consecutive adult patients aged ≥ 16 years with a moderate (post-resuscitation GCS 12 to 9) or severe (post-resuscitation GCS 8 to 3) TBI who were admitted to the ICUs of nine Italian trauma centers from December 2009 to December 2012. Exclusion criteria were a history of cerebral ischemia, CT evidence of brain ischemia at admission, absence of invasive intracranial pressure (ICP) monitoring, and patients with a GCS score of 3 and unreactive pupils.

During the study period, 487 patients with a TBI were admitted to the 9 participating ICUs with a final 143 patients (29.3%) enrolled in the study (Fig. 1). Of these, 47 (32.9%) patients developed a PTCI. There were no differences in patients with and without PTCI in terms of age, severity of the TBI, presence of SAH or EPH on the admission brain CT, hospital LOS, and days of mechanical ventilation (Table 1, Table 2). The incidences of intra-hospital hypotension and hypoxia, pupillary light reflex abnormalities, and evacuated mass lesions (defined according to Marshall brain CT classification) were higher in patients developing a PTCI (Table 1). A total of 94 cerebral infarctions developed in 47 patients, 81 were territorial (86.2%) and 8 were watershed (8.5%). Five infarctions could not be attributed either to territorial or watershed types. Territorial infarctions were in the area of the MCA (n = 17; 18.1%), ACA, (n = 18; 19.1%), PCA (n = 21; 22.3%), LSA (n = 8; 8.5%), TPA (n = 7; 7.4%), BA (n = 3; 3.2%), SCA (n = 3; 32%), PICA (n = 1; 1.1%), and anterior communicating artery (AcoA) (n = 3; 3.2%). Watershed infarctions were in the boundary zones (n = 3; 3.2%) and terminal zones (n = 5; 5.3%). The mean (SD) onset time of a PTCI was 6.2 (11.4) days with an early peak within 24 h (21 cases, 39.6%), a late peak between 3 and 7 days (14 cases, 26.4%), and 4 cases (7.6%) between 24 and 48 h, Fig. 2.

After the central revision of the entire brain CT dataset, we correctly identified 10 patients with PTCI, missed at the initial evaluation; conversely, 7 patients initially classified as PTCI were subsequently classified as non-PTCI.

In the simple ordered logistic regression, the GOS was significantly worse in patients with a PTCI than patients without a PTCI, with a higher proportion of patients with a severe disability and death and a lower proportion of patients with good recovery and moderate disability (Table 2). In the multiple ordered logistic regression, a PTCI was retained in both the core and the extended models (Table 3a, b) as an independent predictor of the GOS. The predictive performances of the obtained models (Fig. 3) were good and increased significantly when PTCI was added to the IMPACT core model (AUC = 0.73, 95% C.I. 0.66–0.82 increased to AUC = 0.79, 95% CI 0.71–0.83; 0.0007) and the extended model (AUC = 0.74, 95% C.I. 0.65–0.81 increased to AUC = 0.80, 95% C.I. 0.69–0.85; p = 0.00008). The results were replicated after correcting the AUC for optimism (Fig. 4 and Table 4). Patients with a PTCI showed higher ICU mortality (10 patients [21.3%] vs one patient [1.0%], p < 0.0001) as well as higher 6-month mortality (13 patients [27.7%] vs 7 patients [7.3%], p < 0.0001); whereas hospital mortality did not differ between the two groups (Table 2).

In this multicenter prospective cohort study, we found that a PTCI is an independent predictor of an unfavorable 6-month outcome and its addition to the IMPACT core and extended models increased their performance in predicting the GOS. Moreover, we confirmed that PTCI is a frequent complication occurring in more than one-third of patients suffering severe or moderate TBI. Most of the PTCIs were territorial, affecting one or more cerebral artery territories, and developed early during the ICU stay.

This is the first prospective study showing a PTCI has an independent effect on a patient’s long-term outcome. Previous studies were either post-mortem neuropathological investigations or ante-mortem retrospective clinical investigations. Among the latter, the GOS was assessed in four single-center studies at 3 months [5] or 6 months [4, 9]; while in one study, the timing of the GOS was not reported [8]. These studies showed increased morbidity [4], increased mortality [9], increased morbidity and mortality [8], or no difference [5] in patients with a PTCI compared with patients without. In three of these studies [4, 5, 9], the impact of a PTCI was assessed while considering the role of other predicting variables, such as age and GCS, using multiple regression analysis. However, none of these studies demonstrated that a PTCI adds value to risk prediction models that include validated factors, as we showed here. We added PTCI to the core and extended IMPACT models, which have been extensively validated with various datasets. The IMPACT models focus on baseline prognostic factors and do not include variables that develop throughout the disease process [3]. Therefore, our research expands the results of IMPACT, showing that patients developing a PTCI during the acute stage of disease have a five-fold increased risk for a poor outcome, independently from important factors such as age, motor score, pupillary reactivity, hypotension and hypoxia, brain CT, and the presence of posttraumatic SAH or EDH.

The AUC increased significantly from 0.73 to 0.79 and from 0.74 to 0.80 when the PTCI was added to the core and extended IMPACT models, respectively; although, these are already strong predictive models. Since the increase in the AUC greatly depends on the strength of the baseline model, the stronger the baseline model, the lesser the expected increase in the AUC [26]. This result further confirms that a PTCI is an important predictor of outcome in patients with a moderate or severe TBI. Our finding that a PTCI is a key independent predictor of long-term morbidity in TBI survivors is highly clinically plausible. Residual morbidity in patients suffering from cerebral infarction is high, with 13% of survivors discharged to institutional care [27]. Almost half of all elderly people suffering from an ischemic stroke have hemiparesis and cognitive impairment causing moderate to severe disability [28]. With severe stroke requiring ICU admission and mechanical ventilation, as many as two-thirds of surviving patients are left with a severe persisting disability [29]. The ICU mortality was significantly higher in patients with a PTCI, suggesting that a PTCI might be a proxy of the severity of the TBI; however, a multivariable analysis could not be performed due to the small number of deaths.

We found that PTCI was more frequent than in our previous retrospective study (32.9% vs 19.1%) [4]. In prospective cohort studies, the selection of patients through application of the inclusion and exclusion criteria is more accurate compared to retrospective studies, as it is the measurement of exposures before the outcome occurs, thereby establishing temporality and outcome. The availability of newer brain CT scanners, with higher sensitivity for detecting brain ischemia, may have also played a role. Not least, the centralized revision of all brain CTs increased the detection of PTCI, as 10 more PTCI cases were identified compared to diagnoses made locally by participating centers. This happened without sacrificing specificity because 7 patients initially classified as PTCI were subsequently classified as non-PTCI after centralized revision.

Our study has strengths and limitations. Strengths include the prospective multicenter nature of the study with central readings of all brain CT scans, as well as rigorous statistical methodology.

The identification of PTCI in ambiguous cases was greatly aided by the centralized evaluation of the neuroimaging data, allowing the assessment of the entire neuroradiological history of each patient, with immediate comparison of any questionable finding in a given CT scan to several previous and following exams, thus making the final diagnosis of cerebral infarction much more reliable.

The main limitation is the absence of detailed neurological and neuropsychological assessments of patients at long-term follow-up, limiting our understanding of the relative contribution of the primary traumatic and secondary ischemic brain damage to the persisting disability. Another limitation is that we did not assess the causes of mortality, and hence, we cannot exclude that life-sustaining therapies were withdrawn in patients with PTCI, leading to a self-fulfilling prophecy bias of the outcome prediction model. Moreover, results became available to the participating centers only after the entire brain CT data set of each patient was transferred to the coordinating center where the diagnosis of PTCI was definitely determined.

Our study is the first multicenter study demonstrating that a cerebral infarction, indicating posttraumatic brain damage, is an independent predictor of long-term disability when added to validated prediction models. This is in line with the recommendation of the Lancet Neurology Commission that prognostic models for TBI patients should include dynamic predictors that develop during the disease course [1]. Future research should externally validate this finding in larger studies with adequate power and accurate neurological and neuropsychological assessments of patients at longer-term follow-up. This would assess its generalizability and recommend the inclusion of PTCI in the list of measurable, clinically relevant variables that improve prognostication and contribute to a comprehensive definition of the diversity of post-TBI disability and the needs for personalized rehabilitation. The global burden of TBI has continuously increased in the last 25 years [30], and the prevention of residual disability is a major concern. Future studies should prioritize at risk-patients identification, along with effective prevention strategies to be deployed before the cerebral infarction is fully established.

The findings provide evidence that PTCI is a common complication in patients suffering from a moderate or severe TBI and is an independent risk factor for long-term disability. The addition of PTCI to the IMPACT core and extended predictive models significantly increased their performance in predicting the GOS.