Journal: International Journal of Clinical and Health Psychology : IJCHP

Article Title: Optimizing pediatric “Mild” traumatic brain injury assessments: A multi-domain random forest analysis of diagnosis and outcomes

doi: 10.1016/j.ijchp.2025.100600

Figure Lengend Snippet: Results from the Random Forest analyses for the final diagnostic models at each Visit (V) for classifying pediatric “mild” traumatic brain injury patients versus healthy controls. The top row shows receiver operating characteristics (ROC) results including area under the curve (AUC), balanced accuracy (BA), sensitivity, and specificity for each visit. The bottom row displays the variable importance (VIMP) score for each variable in the final models at each visit. Feature selection for the final model occurred based on whether the lower bound of the 90 % VIMP confidence interval was greater than 0 % (selected = blue; not selected = red). For Visit 1, current/retrospective clinical-ratings predominated for feature selection along with memory and executive function on paper-and-pencil cognitive tests. In contrast, symptom provocation ratings from multiple neurosensory tests were eliminated in the final model. Somatic complaints (sleep and headache), overall post-concussive symptom burden, and performance on memory tests exhibited the best diagnostic accuracy at visits 2 and 3. The following abbreviations are included in the figure: Post-Concussive Symptoms (PCS), Quality of Life (QoL), Retrospective (R), Immediate (IR) and Delayed Recall (DR) from the Hopkins Verbal Learning Test Revised (HVLT-R), Stroop Inhibition (Stroop I), Symptom Provocation (SP), King-Devick (KD), Reaction Time (RT), Visual Motion Sensitivity (VMS), Near Point Convergence (NPC), Vertical Saccades (V Sac), Monocular Accommodative amplitude (MA), Vertical Vestibular Ocular Reflex (V VOR), Tandem Gait (TG), Horizontal Vestibular Ocular Reflex (H VOR), Double Dorsal Foot Stretch (DDFS).

Article Snippet: Current self-reported clinical-ratings (except depression) and all self-reported retrospective-ratings, and performance on tests of immediate and delayed recall (HVLT-R), inhibition (Stroop), and rapid number naming (King-Devick), were retained for the final model. Somatic symptoms (pain, headache, PCS, sleep) were among the best predictors.

Techniques: Diagnostic Assay, Selection, Inhibition

Journal: International Journal of Clinical and Health Psychology : IJCHP

Article Title: Optimizing pediatric “Mild” traumatic brain injury assessments: A multi-domain random forest analysis of diagnosis and outcomes

doi: 10.1016/j.ijchp.2025.100600

Figure Lengend Snippet: Results from the Random Forest analyses for the final outcome models at each Visit (V) for classifying pediatric “mild” traumatic brain injury patients with poor versus favorable outcomes. The top row shows receiver operating characteristics (ROC) results including area under the curve (AUC), balanced accuracy (BA), sensitivity, and specificity for each visit. The bottom row displays the variable importance (VIMP) score for each variable in the final models. Feature selection for the final model occurred based on whether the lower bound of the 90 % VIMP confidence interval was greater than 0 % (selected = blue; not selected = red). Across all three visits, current clinical-ratings, in particular somatic complaints (headache and sleep) and emotional distress, together with neurosensory symptom provocation measures, predominated for feature selection. In contrast, injury severity characteristics and performance-based cognitive measures tended to be eliminated in the final models. The following abbreviations are included in the figure: Post-Concussive Symptoms (PCS), Quality of Life (QoL), Retrospective (R), Symptom Provocation (SP), Tandem Gait (TG), Visual Motion Sensitivity (VMS), Monocular Accommodative amplitude (MA), Horizontal and Vertical Vestibular Ocular Reflex (H VOR; V VOR), King-Devick (KD), Error (Er), Horizontal and Vertical Saccades (H Sac; V Sac), Double Dorsal Foot Stretch (DDFS), Smooth Pursuit (Sm Pur), Near Point Convergence (NPC), Number of Previous Injuries (NumPrevInj), Stroop Inhibition (Stroop I), Identification (IDN), One-card Learning (OCL), Detection (DET), Reaction Time (RT), Accuracy (ac), Loss of Consciousness/Posttraumatic Amnesia (LOC/PTA).

Article Snippet: Current self-reported clinical-ratings (except depression) and all self-reported retrospective-ratings, and performance on tests of immediate and delayed recall (HVLT-R), inhibition (Stroop), and rapid number naming (King-Devick), were retained for the final model. Somatic symptoms (pain, headache, PCS, sleep) were among the best predictors.

Techniques: Selection, Inhibition