Composite Scores: What’s Behind the Number?
From Dafne Capelusnik
As rheumatology fellows, we quickly become familiar with an alphabet soup of composite scores: DAS28, SDAI, DAPSA, PASDAS, SLEDAI… We learn their cut-offs, what constitutes response or remission, and when they should guide treatment decisions. But we may spend much less time asking a simple question: what does that final number actually represent?
In “Use of multidimensional composite scores in rheumatology: parsimony versus subtlety”, Robert Landewé and Désirée van der Heijde challenge the intuitive idea that combining more information into a single score necessarily provides a better picture of our patient.
Composite scores have undeniable advantages. By combining several measures, they can reduce variability, improve the signal-to-noise ratio, and increase statistical power (exactly what we want in a clinical trial). But this parsimony comes at a price, especially in clinical practice: subtlety. By reducing complexity to a single number, we may lose information about what is actually happening in an individual patient.
Rheumatic diseases are inherently multidimensional, although to very different degrees. SLE is perhaps the clearest example: myositis, nephritis, arthritis, skin or haematological involvement may appear independently and do not necessarily change together. PsA provides another intuitive example: arthritis, enthesitis, dactylitis, skin and nail disease may coexist, but improvement in one domain does not guarantee improvement in another. A multidimensional composite nevertheless attempts to summarise these potentially independent dimensions into a single number.
But multidimensional composites are only part of the story. Even a composite designed to measure a single dimension can become more difficult to interpret when the context in which it is used changes. And the challenge becomes particularly relevant when that number is not only used to describe disease, but also to guide treatment decisions, as happens in treat-to-target strategies.
Consider the DAS. Originally developed to assess disease activity in patients with active Rheumatoid Arthritis, its components reflected different perspectives of predominantly the same dimension (arthritis). But the context in which we use these scores has changed. In a patient with little objective inflammation, persistent pain or tender joints may reflect other mechanisms, such as central sensitisation, rather than ongoing inflammation. A high score may therefore persist despite little residual inflammation. Landewé and van der Heijde describe this as “dimension creep”: a measure that was originally largely unidimensional may acquire additional dimensions when used in a different clinical context.
Another important lesson is to distinguish the disease process from its consequences. Inflammation happens now, whereas damage and functional impairment may partly reflect what happened in the past. Yet some composites mix them together. The authors use the HAQ (Health Assessment Questionnaire) within the ACR20 response criteria and MDA (Minimal Disease Activity) as an example. Functional impairment may closely reflect inflammation in early active disease, but the same HAQ score can have a very different meaning in a patient with longstanding, currently quiescent disease, in whom part of that impairment may be irreversible.
This leads to perhaps the most clinically relevant lesson of the paper: a useful clinical trial endpoint is not necessarily a good treatment target. Two patients may have the same “high” composite score (or fail to reach the same target) for entirely different reasons. If residual pain, irreversible functional impairment, or another non-inflammatory component is driving the score, simply escalating anti-inflammatory treatment may not help and could even lead to overtreatment.
The score tells us how high. Its components may tell us why.
Perhaps the most useful habit this paper taught me is therefore not to stop at the final number. When interpreting a composite score, ask: What are its components? Are they measuring the same construct? Are we mixing the disease process with its consequences? Was this score developed for the context in which I am using it? And, most importantly, what is actually driving the score in this patient?
Composite scores help us simplify complex diseases. They are invaluable tools, particularly in clinical research, but simplification should not replace clinical interpretation. As Landewé and van der Heijde remind us, parsimony in outcome assessment can come at the cost of clinical subtlety.
References
Landewé RBM, van der Heijde D. Use of multidimensional composite scores in rheumatology: parsimony versus subtlety. Ann Rheum Dis. 2021 Mar;80(3):280-285. doi: 10.1136/annrheumdis-2020-216999. Epub 2020 Nov 3. PMID: 33144304.