|Perel, P., Edwards, P., Wentz, R., & Roberts, I. (2006). Systematic review of prognostic models in traumatic brain injury. BMC Medical Informatics and Decision Making, 6, 38.
|Glasgow Outcome Scale, models, statistical, brain injury, diagnosis, brain complications, mortality, prognosis, treatment outcome
Background: Traumatic brain injury (TBI) is a leading cause of death and disability world-wide. The ability to accurately predict patient outcome after TBI has an important role in clinical practice and research. Prognostic models are statistical models that combine two or more items of patient data to predict clinical outcome. They may improve predictions in TBI patients. Multiple prognostic models for TBI have accumulated for decades but none of them is widely used in clinical practice.
Objectives: To critically assess existing prognostic models for TBI.
Search strategy: Studies that combine at least two variables to predict any outcome in patients with TBI were searched in PUBMED and EMBASE. Studies conducted prior to 1990 were excluded because patient management and diagnostic techniques may have changed since this time. Studies that investigate more than one variable but do not combine them for obtaining a prediction were excluded. The reference lists of included studies were inspected for further possible studies meeting the inclusion criteria.
Selection criteria: Two reviewers independently examined titles, abstracts and keywords of records from electronic databases, for eligibility. The full text of all potentially relevant records was obtained and two reviewers independently assessed whether each met the pre-defined inclusion criteria. A third reviewer resolved disagreement.
Data collection and analysis: The authors analyzed the quality of the prognostic models (i.e. the internal and external validity) included in this systematic review using an 18-item. One reviewer extracted the information from each study for assessing the quality of reporting in each of the questions.
Main results: A total of 53 reports including 102 models were identified. Almost half (47%) were derived from adult patients. Three quarters of the models included less than 500 patients. Most of the models (93%) were from high-income countries. Logistic regression was the most common analytical strategy to derived models (47%). In relation to the quality of the derivation models (n: 66), only 15% reported less than 10% pf loss to follow-up, 68% did not justify the rationale to include the predictors, 11% conducted an external validation and only 19% of the logistic models presented the results in a clinically user-friendly way.
Conclusions: Prognostic models are frequently published but they are developed from small samples of patients, their methodological quality is poor and they are rarely validated on external populations. Furthermore, they are not clinically practical, as they are not presented to physicians in a user-friendly way. Finally because only a few are developed using populations from low and middle-income countries, where most of trauma occurs, the generalizability to these setting is limited