Page 88 - The value of total hip and knee arthroplasties for patients

Chapter 5
Recently, two of such studies have been published. De Rooij et al13 investigated the conceptual overlap between cognitive concepts in patients with chronic widespread pain and found that 16 different cognitive subscales could be reduced to three factors namely 1. negative emotional cognitions, 2.active cognitive coping and 3.control belief and expectations of chronicity. Campbell et al6 studied the conceptual overlap of psychological constructs in low back pain patients and found that 20 subscales of psychological questionnaires could be reduced into four factors namely 1.pain-related distress, 2.cognitive coping, 3.causal beliefs and 4.perceptions of the future. Our study also addresses this issue; however we had a slightly different approach. De Rooij et al and Campbell et al performed factor analyses on a subscale level thereby aiming to identify the most complete though comprehensive set of cognitive (de Rooij) or psychological (Campbell) constructs.We however assessed whether individual item of questionnaires measuring the constructs of interest indeed load on the factors as intended by the developers of the subscale. Our approach therefore, may be seen as the first in a two-step approach in examining overlap between constructs. Once distinctive measurement has been established on an item level, a next step could then be assessing overlap between subscales as de Rooij and Campbell did.
Strengths and Limitations
A strength of this study is that the most widely accepted measurement instruments that aim to measure the included constructs were used, ensuring comparability to future CFA’s in other patient groups. A limitation of this study is the limited sample size for CFA’s, which made us decide to test our primary hypotheses on the complete sample of THA and THA patients. A multigroup analysis was only done to test for factorial invariance between the THA and TKA group. Although the FI models converged well and results suggest that strict invariance holds for our data, we do recognize that these analyses may be slightly underpowered and therefore these results should be interpreted with caution. Factorial invariance testing showed similar factorial structures in both groups implying that the constructs measured in this study have the same meaning in both patient groups, thereby suggesting generalizability of our results. Another limitation is the high percentage of missing responses on one of the items of the HS, which was caused by a printing error in the questionnaire.The WLMSV estimator in Mplus statistical software incorporates missing data by pairwise presence, though this is under the assumption that missing data are missing completely at random (MCAR). Because of the reason of the
86