A systematic review identified several prognostic models that may help clinicians predict which pediatric patients are at risk for persistent symptoms in the weeks and months following a concussion. While more validation is needed, the Predicting and Preventing Postconcussive Problems in Pediatrics (5P) clinical risk score had the strongest evidence among models examined.
The review was conducted following the recommendations of the Cochrane Prognosis Methods Group and adhered to the TRIPOD-SR guidelines. Initially identifying 17,433 references, only 6 studies ultimately met the inclusion criteria after screening. The review specifically focused on models predicting delayed recovery 28 days to 1-year post-injury, with prediction made within 28 days of injury.
The six included studies were published between 2014-2022, with participant enrollment occurring from 2010-2019. Sample sizes ranged from 125 to 2,006, and the prevalence of persistent symptoms ranged from 17.6% to 57.8% across studies. Four studies used the Concussion in Sport Group consensus definition of concussion, and all incorporated some form of the International Classification of Diseases, 10th Revision (ICD-10) criteria for defining persistent symptoms, though with variable operationalization.
Logistic regression was the most common modeling approach (five studies). The number of candidate predictors examined ranged from 4 to 46, with events per variable (EPV) ranging from 2.4 to 18.5 in model development studies. Two of the 3 external validation studies had over 100 outcome events.
The 5P score, which incorporates factors like age, sex, prior concussions, migraine history, balance, and acute symptoms, was the only model externally validated in both an emergency department and concussion clinic setting, up to 10 days post-injury. A meta-analysis of three studies found it had an aggregate area under the curve (AUC) of 0.69, considered a small-to-medium effect size for predicting risk.
The review also examined the Buffalo Concussion Physical Exam delayed recovery (BCPE RDR) risk score, which incorporates factors like time since injury, prior concussions, and vestibular and ocular assessments. It was internally validated with an AUC of 0.86. Ten unnamed models by Bressan et al used components of child SCAT tests as predictors and were internally validated with AUCs ranging from 0.51 to 0.68. An unnamed model by Grubenhoff et al using only acute symptoms as predictors had no validation performed, with an AUC of 0.68 in the derivation cohort.
Five models underwent some form of validation, with the 5P score being the most extensively validated through one temporal external validation by the original authors and two independent external validations. Bootstrapping was the most common internal validation method.
Model discrimination was reported via the AUC in all studies, but calibration statistics were only reported in two studies. None of the studies evaluated the potential clinical utility of the models.
The most common predictors across models were headache, noise sensitivity, and fatigue. Only the 5P and BCPE RDR scores incorporated non-symptom predictors like demographics, prior medical history, and physical exam findings.
All studies were rated as high risk of bias. Common issues included determining outcomes with knowledge of predictor information, small sample sizes, and suboptimal handling of continuous variables and missing data. The authors used a modification of the GRADE approach to assess the quality of evidence for prognostic factor studies.
Limitations of the review include the heterogeneity in outcome definitions and lack of data on model performance in specific subpopulations. The authors noted that a formal GRADE process for rating the quality of prognostic model studies does not yet exist. They recommended achieving consensus on operationalized criteria for persistent post-concussive symptoms to facilitate further progress in prediction model development. The review also highlighted the need for more rigorous external validations, particularly in non-emergency department settings.
Conflict of interest disclosures can be found in the study in Pediatrics.
