Construct validity of the Physiotherapy Evidence Database (PEDro) quality scale for randomized trials: Item response theory and factor analyses

Abstract

Background: There is an agreement that the methodological quality of randomized trials should be assessed in systematic reviews, but there is a debate on how this should be done. We conducted a construct validation study of the Physiotherapy Evidence Database (PEDro) scale, which is widely used to assess the quality of trials in physical therapy and rehabilitation.

Methods: We analyzed 345 trials that were included in Cochrane reviews and for which a PEDro summary score was available. We used one- and two-parameter logistic item response theory (IRT) models to study the psychometric properties of the PEDro scale and assessed the items' difficulty and discrimination parameters. We ran goodness of fit post estimations and examined the IRT unidimensionality assumption with a multidimensional IRT (MIRT) model.

Results: Out of a maximum of 10, the mean PEDro summary score was 5.46 (SD = 1.51). The allocation concealment and intention-to-treat scale items contributed most of the information on the underlying construct (with discriminations of 1.79 and 2.05, respectively) at similar difficulties (0.63 and 0.65, respectively). The other items provided little additional information and did not distinguish trials of different quality. There was substantial evidence of departure from the unidimensionality assumption, suggesting that the PEDro items relate to more than one latent trait.

Conclusions: Our findings question the construct validity of the PEDro scale to assess the methodological quality of clinical trials. PEDro summary scores should not be used; rather, the physiotherapy community should consider working with the individual items of the scale.

Keywords: item response theory; physiotherapy; randomized clinical trials; risk of bias; study quality scale; validation.

Figure 1

Items of the Physiotherapy Evidence Database (PEDro) scale and their distribution in the validation dataset of 345 trials. Item 1 (eligibility criteria) does not contribute to the total score

Figure 2

Item characteristic curves from the two‐parameter logistic model [Colour figure can be viewed at http://wileyonlinelibrary.com]

Figure 3

Item information functions (IIF) from the two‐parameter logistic model for all PEDro items. Items 3 and 9 contributed the most information but at the same trial quality. The coverage for qualities larger than 2 was poor [Colour figure can be viewed at http://wileyonlinelibrary.com]