Analysis of Internal Construct Validty of the SRS24 Questionnaire: Paper #18

Rothenfluh, Dominique A. MD, PhD; Neubauer, Georg MD; Min, Kan MD

Spine: Affiliated Society Meeting Abstracts:
Podium Presentation Abstracts
Author Information


Summary: The SRS‐24 questionnaire does not meet properties of internal construct validity. Adjustments are recommended for a valid linear and unidimensional change score

Introduction: The SRS‐24 questionnaire is widely used for outcome assessment in adolescent idiopathic scoliosis patients. While the questionnaire was validated using methods of classical test theory, internal construct validity has never been shown. Internal construct validity, i.e. unidimensionality and linearity, is a fundamental arithmetic requirement and needs to be shown for a scale for summating any set of Likert‐type items

Methods: 250 SRS‐24 questionnaires distributed to 116 patients pre‐operatively and at follow‐up were analyzed. All patients were diagnosed with adolescent idiopathic scoliosis. The questionnaires were subjected to Rasch analysis using the RUMM2020 software package. Rasch analysis involves an iterative process on which items are analyzed for psychometric properties such as threshold ordering, linearity, unidimensionality and reliability indices, contrasting traditional methods

Results: The SRS‐24 questionnaire as a summated score is multidimensional and shows misfit to the Rasch model and therefore lacks linearity. Disordered thresholds between response options were found in 10/24 items. The response options had to be collapsed in order to achieve threshold ordering. While the 7 different domains still showed misfit to the Rasch model, 4/7 were unidimensional as would be expected. Unidimensionality and linearity, i.e. fit to the Rasch model, could only be achieved for an aggregate score by separating pre‐ and postoperative items and omitting items which caused model misfit. Reducing the questionnaire to 7 pre‐operative items (p=0.062; 1.28% t‐tests) and 5 postoperative items (p=0.267; 3.70% t‐tests) yields model fit and unidimensionality for both summated scores. The person‐separation indices (PSI) were 0.67 and 0.69 respectively, corresponding to the Cronbach alpha of the original questionnaire

Conclusion: The SRS‐24 score is a non‐linear and multidimensional construct. In principle, adding the items into a single value is therefore not supported and invalid. Making profound changes to the questionnaire, yields a score which fulfills the properties of internal construct validity and supports its use a change score for outcome measurement

Significance: Internal construct validity is fundamental requirement for calculation of change scores

© 2010 Lippincott Williams & Wilkins, Inc.