CORR Insights®: Can the Knee Outcome and Osteoarthritis Score (KOOS) Function Subscale Be Linked to the PROMIS Physical Function to Crosswalk Equivalent Scores?

Where Are We Now?

Patient-reported outcome measures (PROMs) now are the prime method to evaluate clinical outcomes in orthopaedics. There are various PROMs available, each with its own particular emphasis. But not all regions, institutions, and authors use the same ones. Still, as long as the PROM chosen is appropriate, validated, and robust for the research question, then the results of an individual study are likely to be meaningful. Problems arise, however, if we pool the results of multiple studies to perform meta-analyses, which have greater power to answer clinical questions, if the source studies in those meta-analyses used dissimilar PROMs.

One solution is to insist that everyone uses the same core outcomes dataset, but this would require us to agree on a select choice of tools, and it’s impossible to mandate agreement on points like this. Should a study use the Knee Outcome and Osteoarthritis Score (KOOS) or the Forgotten Joint Score-12? PROMIS or WOMAC? It depends on the goals of each author group, and achieving consistent, broad, and stable consensus seems unlikely.

To address this problem, researchers started mapping the results of different PROMs to each other. These “crosswalk” tables are valuable tools because they allow conversions of scores on one PROM to scores on another; this is helpful for clinicians who are interpreting studies that used tools other than the ones they use every day, and they’re critically important to meta-analysts who seek to pool data across studies that used a variety of PROM instruments.

In the current study, Heng et al. [3] report such a crosswalk, linking the KOOS functional subscale to the PROMIS physical function score. This work, in principle, allows us to interchangeably calculate PROMIS function scores from the KOOS Activities of Daily Living score and vice-versa, which will be helpful in facilitating data pooling of function scores, as well as in clinical practice.

Where Do We Need To Go?

While this study [3] is an excellent mathematical model, that is literally what it is. The authors used an equipercentile approach, linking both unsmoothed and log linear smoothed score distributions to contrast the two PROM scores, which were obtained from a single dataset. Their sensitivity analyses highlight the robustness of their modeling, but this does not ensure wider generalizability. Of note, the dataset they used was a > 90% white nonethnic patient sample. We do not know if this is representative of the wider arthroplasty population within the United States or internationally. Recent evidence suggests substantial cross-cultural variation in PROM interpretation and responses [2], so while this is an excellent research tool, future studies should seek to generalize its findings in other populations.

In addition, we need to develop crosswalks among other outcomes scores if we want to make comparisons across the various different scores that are in common use. Importantly, crosswalks only work between scores that assess the same general domain of interest. We cannot, for example, make a valid link between the KOOS pain subscore and the PROMIS function score. Still, there are a large number of commonly used outcomes tools with no available crosswalks among them. Researchers have begun the process of deriving crosswalk tables [1, 5, 6], but there is more work to be done.

There will inevitably be more new scores in the future. When these are developed, researchers will need to add more crosswalks to our existing tools. But as of this writing, no single PROM has proven superior to others in measuring outcome constructs (individual aspects like function or pain or joint awareness), so researchers should continue using current outcome score preferences until a new tool eventually usurps older ones.

Each crosswalk will also need external validation, and ideally, this will be done using diverse clinical populations to ensure both validity and wider generalizability. We don’t know the extent to which demographic factors such as age, gender, and ethnicity influence crosswalks, but we do know that different groups respond slightly differently to PROM questions [4]. This likely will influence the mapping process unless researchers can obtain a large and widely representative sample of crosswalks. For example, if there are five different studies on KOOS to PROMIS crosswalks using different population groups that report different models, then we will have a choice of five different crosswalk models to use in clinical practice. How do we decide which is correct? Perhaps we need to pool all the data in a meta-analysis to create a single agreed-upon crosswalk for that score-score interaction.

Ultimately, we will need a library of interconnected scores. This will be a perpetual task because the library will need to be updated every time a new PROM is introduced and enters popular use. It is also important to note this would apply across the full spectrum of procedures where we collect outcomes with PROMs, including knee interventions, hips, shoulders, and elbows.

How Do We Get There?

Creating a library of crosswalks and validating them across diverse clinical populations is a big test for the research community considering the number of PROMs in use in orthopaedics. This project requires multiple research groups with sufficient expertise to complete the task.

Linking all score data in the literature is a daunting prospect as well. To do this, researchers should start by linking the major PROMs currently in use and then move to comparisons with more obscure and legacy instruments. This is logistically practical in areas where we have large datasets and where we have groups of patients who have completed multiple PROMs that assess the same construct at the same time. It may be harder though to access reliable data in sufficient volume to link scores for less frequently performed interventions. Frustratingly, these are the very procedures in most need of crosswalks because there are less outcome data available to evaluate the intervention outcomes. We should also consider demographic effects. We can do this by validating models published with homogeneous populations in a more diverse and reflective sample to ensure wider generalizability. This will need to be done before we can consider the use of such algorithms to guide clinical care of management decisions.