Overall, our analyses of several alternative models provide support for the unidimensional structure of the IKDC with some qualifications, in a population of young active patients with ACL tears.
Based on exploratory analyses of the IKDC, it appears as though two distinct factors may be present when used with ACL patients, in contrast to the intended unidimensional structure. Higgins and colleagues concluded that a two-factor model of the IKDC was most appropriate after using EFA, CFA, and item response theory to investigate data from 1517 patients with various knee problems(4). The proposed two-factor structure included a symptom and knee articulation factor (items 2, 3, 4, 6, 9a-i, 10b), and an activity level factor (items 1, 5, 7 and 8). This model was replicated in a recent study of 319 ACL patients using Bayesian structural equation modeling, which ultimately concluded poor structural validity of the IKDC in ACL patients(5). While the same structure emerged in our own two-factor EFA of the IKDC, we felt that the grouping of items 1, 5, 7, 8 and 10 may be strongly related to similar linguistics and response options amongst questions, rather than distinct questionnaire constructs.
However, when confirmatory analyses were conducted, both the single factor and alternative two-factor structure of the IKDC required too many modifications to obtain adequate fit for us to have confidence in the utility of these models for our patient population. Accepting the one- or two-factor structures of the IKDC with the required modifications would complicate scoring and convolute interpretability of IKDC in young active patients with ACL tears. The pattern of item correlations suggested by modification indices during the CFA process made it clear that items worded similarly or sharing the same number of response options tended to group together. This is intuitive and may indicate clustering of items based on linguistic factors rather than constructs of the questionnaire. This analysis outlines potential issues with multiple rating scales in the same unidimensional outcome as consideration when developing future measures. When creating new orthopaedic PROMs, it may be helpful to standardize the response options and linguistic style of questions to avoid such item clusters.
Finally, a bifactor model was run to test the hypothesis that there is indeed one overall general factor of “symptoms, function and sports activity” as intended, with residual associations between subsets of items similarities in with linguistics, response options, or content. There were four content factors created: symptoms (items 2, 3, 4, 6), activity level (items 1, 5, 7, 8, 10b), activities of daily living (ADLs) (items 9a-f), and sport (items 9g-i). The symptoms factor covers patients’ pain, stiffness, and locking/catching of the knee. The activity level factor consists of questions about the activity level a patient can participate in without various symptoms. The ADL and sport factors are distinct in the types of activities being inquired about. We believe that this distinction is important for our population of interest and a consideration that was missing from the previously proposed two-factor structure.
The bifactor model showed acceptable fit indices without modifications, unlike the previously tested structures. Each IKDC item showed adequately strong (≥ 0.35) associations with the general factor, and generally weaker and inconsistent associations with the secondary, more specific content factors. Importantly, measures of explained systematic, reliable, and common variance (ωH, ωR, and ECV, respectively) were adequately high for the general factor but were low for all four secondary factors. This suggests that while similarly worded items or those with similar content tended to group together, they did not necessarily represent distinct constructs that should form their own subscales. Additionally, only the general factor reached the threshold of factor determinacy (0.9) that would recommend its score be used. The score for the general factor is the total IKDC score currently in use. The secondary sport and activity level factors showed some evidence of reliability and adequacy as independent factors, however, the estimates are not substantial or consistent enough that we would advocate the use of these subscale scores. Taken together, the reliability indices suggest that the general factor stands largely on its own but that there may be value in looking at the sport and activity level factors, as some people may show differentiation on these three factors (e.g., a high score on two and moderate score on the third). Overall, these analyses suggest that the IKDC can be deemed “essentially unidimensional”.
Therefore, in the context of young active ACL patients, we propose a bifactor model structure for the IKDC whereby one general factor best represents the items as intended, but item clusters caused by differences in content and item construction (wording and response options) are accounted for to obtain adequate fit. Clinically, this means that clinicians and researchers can continue to use the IKDC as it was designed, as a unidimensional instrument for young active patients with ACL tears. What the bifactor model adds is acknowledgement of item clusters depicting symptoms, activity level, ADLs, and sport. Clinicians and researchers can look at these clusters of items more closely to determine what secondary factor may be driving a high or low score, and patient progress or lack thereof. Importantly, these factors should not be treated as subscales, and scoring of the IKDC has been validated in this population to remain as is. Forming subscales in addition to the total score creates a multi-factor correlated model, with resultant scores which would be highly correlated and redundant, because they would include a mix of both the general and secondary factors. The bifactor model therefore does not change use or scoring of the IKDC in young active patients with ACL tears but increases confidence in the utility of one total score and adds further nuance to the interpretability of the scale for these patients.
Limitations
Only baseline, and not post-operative data was used for this analysis. Assessing additional time points and the invariance of the proposed structure across time is beyond the scope of the initial CFA for this outcome measure covered in this paper. Therefore, we cannot be sure that the bifactor model and essential unidimensionality shown here would have adequate fit in other datasets at later timepoints following ACL injury.