Lack of Prognostic Model Validation in Low Back Pain Prediction Studies: A Systematic Review. Issue 8 (August 2018)

• Record Type: Journal Article
• Authors: McIntosh, Greg; Steenstra, Ivan; Hogg-Johnson, Sheilah; Carter, Tom; Hall, Hamilton
• Is Part Of: Clinical journal of pain. Volume 34:Issue 8(2018)
• Publication Date: 2018-08
• Subjects: back pain -- prediction -- prognosis -- systematic review -- validation; Pain -- Periodicals; Pain -- Treatment -- Periodicals; Analgesia -- Periodicals; 616.047205
• DOI: 10.1097/AJP.0000000000000591
• Languages: English
• ISSNs: 0749-8047

Abstract

Abstract : Objective: The objective of this study was to investigate the frequency with which prediction studies for low back pain outcomes utilize prospective methods of prognostic model validation. Method: Searches of Medline and Embase for terms "predict/predictor, " "prognosis, " or "prognostic factor." The search was limited to studies conducted in humans and reported in the English language. Included articles were all those published in 2 Spine specialty journals (Spine and The Spine Journal) over a 13-month period, January 2013 to January 2014. Conference papers, reviews, and letters were excluded. The initial screen identified 55 potential studies (44 in Spine, 11 in The Spine Journal); 34 were excluded because they were not primary data collection prediction studies; 23 were not prediction studies and 11 were review articles. This left 21 prognosis papers for review, 19 in Spine, 2 in The Spine Journal. Results: None of the 21 studies provided validation for the predictors that they documented (neither internal or external validation). On the basis of the study designs and lack of validation, only 2 studies used the correct terminology for describing associations/relationships between independent and dependent variables. Discussion: Unless researchers and clinicians consider sophisticated and rigorous methods of statistical/external validity for prediction/prognostic findings they will make incorrect assumptions and draw invalid conclusions regarding treatment effects and outcomes. Without proper validation methods, studies that claim to present prediction models actually describe only traits or characteristics of the studied sample.