Tomas, Carissa W.;
Fitzgerald, Jacklynn M.;
Bergner, Carisa;
Hillard, Cecilia J.;
Larson, Christine L.;
deRoon‐Cassini, Terri A.
Machine learning prediction of posttraumatic stress disorder trajectories following traumatic injury: Identification and validation in two independent samples
Sie können Bookmarks mittels Listen verwalten, loggen Sie sich dafür bitte in Ihr SLUB Benutzerkonto ein.
Medientyp:
E-Artikel
Titel:
Machine learning prediction of posttraumatic stress disorder trajectories following traumatic injury: Identification and validation in two independent samples
Beteiligte:
Tomas, Carissa W.;
Fitzgerald, Jacklynn M.;
Bergner, Carisa;
Hillard, Cecilia J.;
Larson, Christine L.;
deRoon‐Cassini, Terri A.
Erschienen:
Wiley, 2022
Erschienen in:Journal of Traumatic Stress
Sprache:
Englisch
DOI:
10.1002/jts.22868
ISSN:
0894-9867;
1573-6598
Entstehung:
Anmerkungen:
Beschreibung:
<jats:title>Abstract</jats:title><jats:p>Due to its heterogeneity, the prediction of posttraumatic stress disorder (PTSD) development after traumtic injury is difficult. Recent machine learning approaches have yielded insight into predicting PTSD symptom trajectories. Using data collected within 1 month of traumatic injury, we applied eXtreme Gradient Boosting (XGB) to classify admitted and discharged patients (hospitalized, <jats:italic>n</jats:italic> = 192; nonhospitalized, <jats:italic>n</jats:italic> = 214), recruited from a Level 1 trauma center, according to PTSD symptom trajectories. Trajectories were identified using latent class mixed models on PCL‐5 scores collected at baseline, 1–3 months posttrauma, and 6 months posttrauma. In both samples, nonremitting, remitting, and resilient PTSD symptom trajectories were identified. In the admitted patient sample, a unique delayed trajectory emerged. Machine learning classifiers (i.e., XGB) were developed and tested on the admitted patient sample and externally validated on the discharged sample with biological and clinical self‐report baseline variables as predictors. For external validation sets, prediction was fair for nonremitting versus other trajectories, areas under the curve (AUC = .70); good for nonremitting versus resilient trajectories, AUCs = .73–.76; and prediction failed for nonremitting versus remitting trajectories, AUCs = .46–.48. However, poor precision (< .57) across all models suggests limited generalizability of nonremitting symptom trajectory prediction from admitted to discharged patient samples. Consistency in symptom trajectory identification across samples supports prior studies on the stability of PTSD symptom trajectories following trauma exposure; however, continued work and replication with larger samples are warranted to understand overlapping and unique predictive features of PTSD in different traumatic injury populations.</jats:p>