This systematic review and meta-analysis examined predictive risk factors for tICH to characterize their roles in guidelines for the acute stage management of mTBI. A total of 17 studies spanning global research sites in North America, Europe and Asia were assessed for available data. No studies representing Africa, South America or Australasia were included in the final list of studies for review. Out of the eleven independent risk factors available for data synthesis, seven were shown in the meta-analysis to have predictive value for tICH and a few novel variables were shown to be statistically significant in individual studies. We believe that these findings may prove useful in the validation of existing guideline elements as well as provide justification for data elements in future studies. The risk factors we were able to provide a meta-analysis for are discussed in more detail below.

Signs of a skull base fracture

The strongest risk factor for tICH in our meta-analysis was signs of a skull base fracture, a known significant clinical sign present in the Canadian CT Head Rule [8], New Orleans CT Head Trauma Rule [14], CT in Head Injury Patients Rule [7], and the Scandinavian Neurotrauma Committee Guidelines [15]. The relative strength of this risk factor in our data must be interpreted with caution, as its low incidence leads to a wide confidence interval. Scandinavian guidelines does take the relative specificity of this predictor into account however, by including it among higher risk predictors where a 24 h minimum admission for in-hospital observation is recommended regardless of CT findings [15].

Decreased GCS

A post-traumatic decrease in GCS was a significant risk factor for tICH among the variables in our meta-analysis. Most existing TBI guidelines account for GCS reduction in some manner [8, 14, 15, 48], with some variations in recommendations ranging from Haydel et al. in the New Orleans CT Head Trauma Rule recommending all patients with GCS < 15 undergoing a CT [14] to Undén et al. in the Scandinavian Neurotrauma Committee’s guidelines suggesting a minimum of S100B sampling or 12 h of observation as possible alternatives to a head CT in low risk GCS 14 patients [15].

Loss of consciousness, post-traumatic amnesia and vomiting

LOC and PTA are both criteria in some management algorithms and a prerequisite for some definitions of mTBI [3, 8, 14, 49], and both were statistically significant predictors in our meta-analysis. There is some variation in the view on these predictors across different regions, such as in the Scandinavian Neurotrauma Committee’s guidelines for management of TBI which includes LOC and recurrent vomiting as indications for CT or S100B sampling, but not PTA [15]. There are some pragmatical difficulties in the accurate assessment of amnesia as a clinal variable as it may include any combination of transient, anterograde, retrograde amnesia, as well as coinciding with loss of consciousness. Foks et al. have shown in a head injury population with only GCS 15 patients with and without having undergone a head CT that PTA is associated with traumatic findings on CT with an OR of 3.8 (95% CI 2.9–4.9) and when combined with loss of consciousness the OR increases further to 4.1 (95% CI 3.1–5.3) [50]. Smits et al. have in a cohort of patients with GCS 15 and one additional risk factor shown that anterograde amnesia is not associated with hemorrhage, but persistent retrograde amnesia does show an association at OR 1.7 per 60 min of time with amnesia [51]. As expected, vomiting was a significant predictor of tICH in our meta-analysis. This is consistent with it being a known risk factor for tICH and is also included in existing management algorithms for tICH [8, 14, 15, 48].

Antiplatelet treatment and anticoagulation

Interestingly, though antiplatelet treatment was a significant risk factor in our meta-analysis, VKA and DOAC treatment were not. Existing guidelines caution careful management of patients on all categories of medication that impact hemostasis [8, 14, 15], and a number of studies including two independent meta-analyses verify the significance of antiplatelet treatment in the context of hemorrhage risk [52,53,54]. Additionally, the finding that antiplatelet treatment seem to outweigh anticoagulation in terms of tICH risk has also been observed in multiple recently published studies [40, 48].

One aspect we hoped to be able to investigate in our meta-analysis was the impact of the shift in anticoagulation prescription from vitamin K antagonists (VKA) to DOACs. The results from our meta-analysis suggest that VKA and DOAC treatment were similar in their risk profile for tICH and were both non-significant (OR 1.27, 95% CI 0.71–2.28 and OR 1.32, 95% CI 0.74–2.35 respectively). Several studies contained data on DOAC treated mTBI patients during our screening process, but did not fit our selection criteria due to patient selection (anticoagulanted patient subcohorts [55, 56]). However, these studies do suggest a lower tICH risk in patients on DOAC in comparison to patients on VKAs. Also, in a study updating the CT in Head Injury Patients Rule based on multicenter patient data by van den Brand et al. [48], the patient data suggested anticoagulation after the introduction of DOAC was no longer a predictor of tICH. Similarly, a systematic review synthesizing data from anticoagulated mTBI subpopulations by Karamian et al. showed an overall incidence of tICH in mTBI patients on DOACs of 6.4%, lower than in mTBI patients on VKAs at 10.5% [57].

Considering the otherwise predominant consensus in several existing guidelines that anticoagulation contributes to increase risk of traumatic hemorrhage, one possible explanation for these findings is that there is a tendency among clinicians to order CT scans for all patients on anticoagulation, regardless of severity of injury and a lack of symptoms or other factors to justify a radiological examination. These findings warrant consideration in future guideline updates as well as studies.

Headache and intoxication

The risk factors headache and intoxication are both included in the New Orleans Charity Head Trauma Rule [14], however both were not shown to be predictive of tICH in our meta-analysis. The inclusion of these variables is likely to be a factor in the consistent finding from previous validation studies that this guideline is low in sensitivity. This has been shown in pooled data in a systematic review and meta-analysis by Alzuhairy et al. to be 12.3% (95% CI 7.4–19.8%) [58]. Our results suggest that future guidelines that aim to reduce CT overuse in mTBI patients at the ED should prioritize other predictors than headache and intoxication, and these variables are not likely to be key candidates for standard data collection and confounding adjustment in future mTBI studies.

Male sex

Male sex is the only risk factor detected in the meta-analysis to not be currently in use in one or more existing guidelines for management of mTBI. However, this finding is consistent with data previously published in Dunning et al. [59]. The percentage of male patients across studies included in this review ranged from 47 to 72%, suggesting differences in cohort compositions between studies and geographical regions. It is possible that these findings are secondary to the male population being subject to confounding from for example different trauma mechanism, however this is generally not explored in our included studies. Even though it is unlikely that these findings will directly impact clinical guidelines considering the impracticality of unselectively providing CT head scans to all male patients, male sex could be a variable up for consideration in a system akin to the application of female sex in the CHA2DS2-VASc score for atrial fibrillation [60].

Age

Another risk factor we aimed to analyze in our meta-analysis was the aging population, and this has previously been partially accounted for in existing guidelines recommending thresholds of 60 [14] and 65 [8, 15] respectively. Two studies in our review reported data at the 65 year threshold (OR 1.14, 95% CI 0.66–1.96 [34] and OR 1.24, 95% CI 0.89–1.74 [37]), one study reported data at the 75 year threshold (OR 2.57, 95% CI 1.83–3.63 [43]), while an additional study reported data in terms of per year increase (OR 1.04, 95% CI 1.00–1.09 [47]). Surprisingly, Niklasson et al. found in their study population that an age threshold as low as 45 years of age proves to be a statistically significant predictor of tICH (adjusted OR 3.54, 95% CI 2.33–5.38) [40]. Though the variations in presentation made data synthesis unfeasible, the cumulative results suggest that though an older age appears to be a risk factor for tICH, there is insufficient data to support an optimal threshold for clinical application.

Biomarkers

Though some biomarkers were assessed in the studies included in our meta-analysis, there was not sufficient homogeneity of method or data to perform data synthesis on individual biomarkers. Acute and non-acute stage biomarker use in mTBI has been characterized in other systematic reviews [61, 62], demonstrating significant potential. However Visser et al. suggest that there is even more variety in methodology in this subgroup of studies in terms of time points (relative to trauma and sampling), controls, cut-offs, and management of samples, which further generates difficulty in drawing robust conclusions [62].

Scalp lesion

Galliazzo et al. reported the clinical finding of a scalp lesion as a statistically significant predictor of tICH [34]. In previous studies, the New Orleans CT Rule proposes that any sign of injury above the clavicle level as an indication for CT head scan in the GCS 15 mTBI population [14]. The association of scalp lesion with tICH suggests there may be additional approaches to risk assessment based on the external signs of injury using the localization of the injury that could be studied further.

Limitations

This systematic review with a meta-analysis was conducted while balancing two major factors: heterogeneity and available data quantity. Heterogeneity poses a concern to the generalizability of our findings, and the following are several examples of this in our study material.

Firstly, the definition of mTBI is also known to be highly variable in literature [3], with both GCS 13–15 and 14–15 being frequently used in combination with other symptoms. Secondly, individual studies varied in their application of inclusion and exclusion criteria, ranging from specific comorbidities such as mental disability that hinder assessment [32] to missing or incomplete medical records [40, 42]. Thirdly, there is also significant variation in the incidence of hemorrhage (3.7–33.9%), as well as in the mean and median ages (37 to 80.2 and 33 to 77, respectively) in the study cohorts. We have presented these factors for each included study in Table 1 to clarify the differences between studies.

These causes of heterogeneity are known issues that have been reported in previous meta-analyses on mTBI [53, 57, 59, 61, 63, 64]. However, in our study we have applied a more stringent approach to study selection in comparison to these studies. Though it is impossible to eliminate all heterogeneity, we sought to minimize the impact of these issues on our findings by excluding studies that contain patient selection criteria that affect risk factors of interest. An example of this is the exclusion of articles that only include subcohorts of mTBI patients on anticoagulation [55, 56]. This is performed at the cost of reducing the number of studies that can be included in the meta-analysis, and careful consideration was made on an article-by-article basis to determine the benefit of inclusion versus exclusion.

For future studies on mTBI and tICH, we suggest application of standardized methods of collection of predictive variables and more homogenized structures of confounding adjustment based on existing guidelines and data on risk factors. This form of standardization would help to eliminate persisting issues with both comparability and generalizability across mTBI studies as well as in future literature reviews.