Joint awareness, function, and pain measures characterise 70% of the variance in 29 knee arthroplasty outcome measures: A principal component analysis

Objectives:
To identify and categorise a reduced set of meaningful outcome measures, by examining and attributing the total variance within the multiplicity of questions in patient reported outcome measures (PROMs) through principal component analysis.

Methods:
A dataset comprising 841 unilateral TKAs performed between 2021 and 2022 at a single centre with complete six-week and one-year post-operative patient-reported outcome measures (PROMs) was collected retrospectively. Principal component analysis was used to identify the components necessary to explain 60-70% of variance within the 29 variables of the Forgotten Joint Score (FJS), Oxford Knee Score (OKS) and EuroQoL five-dimension health questionnaire (EQ5D-5L). A variable was considered to meaningfully contribute to a component if the magnitude of its load was at least 0.5.

Results:
In the cohort, 431 (51. 2%) of the patients were female and 391 (46.5%) of patients were less than 70 years old. The cohort mean (SD) age was 69.8 (8.7) and mean (SD) BMI was 31.4 (4.8). The BMI distribution was: normal weight - 76 (9%), over-weight-270 (32.1%), obese class I -292 (34.7%), obese class II -168 (20%) and obese class III- 35(4.2%).
Three principal components explained 61.6% of the variance in the six-weeks’ data while two components explained 67.5% of variance in the one-year data. However, for further comparative analysis, three components accounting for 70.5% of variance were extracted for the one-year data.
The first principal component was categorised as a construct for “joint awareness” having all the subscales of the FJS loading saliently on it. Consequently, it can be considered equivalent to the FJS and it explained 48% and 58% of variance in six-weeks and one-year data respectively.
The second and third principal components can be categorised as “function” and “pain” components accounting for 9% and 4% of variance for six-weeks data and 9% and 3% for one-year data, respectively. Cross-loadings were observed on the function and pain components particularly from OKS subscales indicating some overlap between pain and function. OKS-2 (washing) and OKS-3 (transport) loaded saliently on the function component in both 6-weeks and one-year data and only OKS-8 (night pain) loaded saliently on the pain component. The remaining OKS subscales cross-loaded on function and pain components. Moreover, OKS-7 (kneeling) showed inconsistency in its cross-loading at the different time-points (on pain and function at 6 weeks, on function and joint awareness at one-year). Also, the communality values of OKS-7 (kneeling) were particularly low in both six-weeks (0.269) and one-year (0.397) data.
All EQ5D-5L subscales loaded saliently on the function component (2nd principal component) in the six-weeks data (except EQ-5D pain/discomfort which loaded on the pain component) and one-year data.

Conclusion
Three principal components can explain up to 70% of the variance in 29 PROM scores and were categorised as “joint awareness”, “function” and “pain” components. The FJS has clearly provided additional value in the assessment of patient outcomes, its contribution has been validated by this analysis and supports ongoing collection.