Op-brhe130493 963..964

In this issue of Rheumatology, Stolwijk et al. [1] evaluate the validity of the self-administered co-morbidity questionnaire (SCQ) developed with the aim of assessing co-morbidity in AS. The article could not have a better justification and timing. Rheumatologists have witnessed remarkable advances in the last decade regarding the management of rheumatic diseases, mainly in the field of measurement and in the development of safe and effective therapies. These advances are vastly supported by clinical trials methodology where patients are unique: both free of troublesome co-morbidity and active enough in their disease as to warrant starting a new treatment. But real life is slightly different. Not all patients are as active as those eligible for clinical trials and patients have their own and unique comorbidity combinations that affect their lives. It is well known that eligibility criteria in randomized controlled trials clearly influence the response and the safety of interventions [2]. But also, as the authors point out, ‘comorbidities can affect the detection, prognosis, therapy and outcome of a single condition at any moment’, therefore ‘co-morbidity should be considered as a confounder, effect modifier or predictor in studies on outcomes in AS’ [1]. Co-morbidity is therefore the great confounder in the observational world; it is a must variable in the multivariable analysis. But what is the best way to measure it? There are several approaches to measure co-morbidity. There is a clear overview by Sarfati [3] in which these approaches are explained in relation to the oncology literature, but these are applicable to rheumatology as well. We can measure co-morbidity as (i) counts of individual conditions, (ii) organor system-based approaches, (iii) weighted indices, (iv) case-mix approaches and (v) overall measures of physical health status. The procedure to select an approach for a particular study will rely upon the feasibility of the measure or on data availability. For instance, co-morbidity indices that require access to clinical notes and training for abstractors are more time consuming to use, and some indices are better than other for administrative databases [4]. The greatest difficulty of all is the absence of a gold standard to guarantee that the instrument is really measuring what it is intended to measure [3]. Therefore, despite the fact that most indices have good enough face and content validity, criterion validity is difficult to study. In addition, if a study focuses on a particular disease, an index that has been developed specifically for that disease may be the most appropriate. As no co-morbidity index was fully validated in AS, Stolwijk et al. [1] chose a generic index, which is a selfreport questionnaire with 13 common medical conditions that produces a score between 0 and 45 (0 39 in a shorter version), the SCQ. As neither its criterion nor its construct validity had been studied, the authors evaluated these [1]. The set of hypotheses the authors base their study on is sound and the gold standard they use can be considered fair, as the patients self-report their co-morbidities, and a blind review of their clinical charts by two independent assessors seems good enough to ascertain co-morbid conditions. Other analyses they perform, like testing the concordance of the SCQ-AS with other co-morbidity indices used in rheumatology, or testing the association of the score with expected correlates—namely age or physical function—are also acceptable. So now we can say we have a validated tool to measure co-morbidity in AS studies, but how does this translate into practical terms? Interestingly, the SCQ (Table 1) was developed as a generic index by rheumatologists [5]. They included items on rheumatic and musculoskeletal diseases, while they left out other co-morbidities, perhaps not as common. Nevertheless, patients—do not forget this is a self-administered tool—can add up to three non-specific medical problems to the conditions list. Also, if we look at the diseases included, one can say that there is too much generalisation (any heart disease, any lung disease, kidney stones count as kidney disease). Interestingly, the SCQ is weighted by the limitations on activities due to each co-morbidity; it is not just the co-morbidity, it is its impact. The development of the original SCQ, which is not fully accepted by primary care physicians [6], included a validation against the Charlson co-morbidity index, which is perhaps the most frequently used index [7], and the results were poor. The same applies for the present study in this issue [1]. It is true that the Charlson index is not as good at predicting poor function as it is at predicting death, and that the items are not equal between the Charlson index and the SCQ. What are the implications of this? If the objective of a particular observational study is to measure mortality, the SCQ may not be as good as the Charlson index to adjust for co-morbidity with a risk of mortality. If, on the other hand, the objective of the particular