Measuring Disease Activity and Damage with Validated Metrics: A Systematic Review on Mortality and Damage in Systemic Lupus Erythematosus

Objective. To identify the effect of disease activity and damage, measured by validated indices, on mortality and damage accrual, in order to inform upcoming Canadian systemic lupus erythematosus (SLE) recommendations. Methods. Following GRADE (Grading of Recommendations Assessment, Development and Evaluation) methodology to fill in evidence-to-decision tables to create recommendations for “minimal investigations needed to monitor SLE patients at baseline and subsequent visits,” a systematic literature review was performed. The effect of disease activity and damage, measured by validated metrics, on mortality and damage was systematically reviewed, with metaanalyses performed when available. Results. A title/abstract screen of 5599 articles identified 816 articles for full paper review, with 102 meeting inclusion criteria and 53 with extractable data. Thirty-three articles describing outcomes related to disease activity and 20 articles related to damage were identified. Mortality was associated with higher SLE Disease Activity Index-2000 scores in 6 studies (HR 1.14, 95% CI 1.06–1.22) and higher Systemic Lupus International Collaborating Clinics/ACR Damage Index scores in 6 studies (HR 1.53, 95% CI 1.28–1.83). Higher SLE Activity Measure scores were associated with increased risk of damage in 3 studies (OR 1.06, 95% CI 1.04–1.08). British Isles Lupus Assessment Group was associated with mortality in 1 study with HR of 1.15. Conclusion. Active SLE disease and damage are associated with and predict greater mortality and damage. The use of validated disease activity and damage metrics is important in the assessment of disease activity and damage and will inform upcoming Canadian recommendations for the assessment of SLE.

Systemic lupus erythematosus (SLE) is a complex autoimmune disease that includes targeted autoantibody production, resulting in heterogeneous disease phenotypes 1 . Manifestations vary considerably, requiring careful in-depth clinical assessments to clearly define the degree of disease activity. Disease damage is another important domain in the assessment of patients with SLE, driven both by the burden of disease activity, glucocorticoids, and immunosuppressant exposure 2 . As a result, the contribution of both disease activity and damage leading toward further damage and mortality in SLE is increasingly recognized by physicians treating SLE.
Despite the important relative contributions of disease activity and damage toward SLE outcomes, clinical assessment of patients varies among rheumatologists. Formalized assessments with the use of validated outcome measures of disease activity and damage are largely performed in tertiary care centers that are systematically collecting SLE outcomes for research. A practice pattern survey of Canadian rheumatologists revealed that most were not formally evaluating disease activity or damage using standardized metrics [e.g., validated disease activity instruments such as the SLE Disease Activity Index (SLEDAI)] 3 . This is in contrast to other rheumatic diseases such as rheumatoid arthritis (RA) or spondyloarthritis (SpA), in which validated composite measures of disease activity have been integrated more commonly into clinical practice to facilitate treat-to-target care and medication reimbursement.
Recommending regular performance of validated measures of SLE disease activity and damage in clinical practice is complicated because of many factors including appropriateness in the clinic setting and rheumatologists' familiarity with available instruments. Moreover, the association between the measured constructs in these validated measures and the important physician-driven outcomes of mortality and damage in the short term and long term require analysis. Therefore, the goal of this systematic literature review was to collect existing evidence for the effect of disease activity and damage when measured with validated metrics on mortality and damage in patients with SLE. This summary of evidence will inform upcoming Canadian recommendations for the assessment of SLE using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) methodology.

MATERIALS AND METHODS
A survey of 175 Canadian rheumatologists to evaluate practice patterns for the diagnosis, monitoring, and treatment of patients with SLE was completed in 2012 and served as the basis for future Canadian SLE recommendations 3 . The question of disease activity/damage evaluation as part of patient assessment/monitoring over time arose from this survey, serving as the basis for this systematic literature review. Patient-reported outcomes (PRO) were not included in this analysis owing to the heterogeneity of outcome measures, and difficulties applying this in a clinical context, knowing that PRO and disease activity/damage are independent domains in the assessment of SLE and in general not associated with disease activity/damage. Moreover, disease flares were discussed but not included in the measured outcomes because of the difficulties in finding homogeneous definitions among studies.
Search strategy. The effects of disease activity and damage, using validated metrics, on the outcomes of damage accrual [measured by the Systemic Lupus International Collaborating Clinics/American College of Rheumatology Damage Index (SDI)] and mortality were assessed. A list of candidate disease activity and damage measures was identified by the group through the American College of Physicians 4 Journal Club, UptoDate, and existing systematic reviews on the subject. A final list of candidate measures of disease activity/damage was circulated and discussed by teleconference by the group and submitted to the librarian (TC) as part of the search criteria. Relevant articles were identified by searching OVID Medline (1946 to July 2016), OVID Embase (1974 to July 2016), and the Cochrane Library (inception to July 2016). Conference abstracts retrieved from Embase were also reviewed. The search strategy was broad to address 2 PICO (population, intervention, comparison, outcome), the first relating to validation studies for disease activity and damage measures (not the subject of this review). The second PICO and focus of this review evaluated the following: population = SLE; intervention = disease activity and damage scores; comparison = self; outcome(s) = worse disease activity, damage, and mortality.
Terms were searched as keywords and/or subject headings as appropriate (Supplementary Data 1, available with the online version of this article) and duplicate references were removed within RefWorks. Studies were limited to those involving humans, and exclusions included non-English abstracts, case reports, editorials, and review articles. Bibliographies of review and guideline articles were hand-searched for articles meeting the inclusion criteria. The dataset from this search was used to inform GRADE recommendations for SLE assessment and monitoring in Canada. Ethics approval was not required because this was a systematic literature review.
Data collection and analysis. Title/abstract screening was performed by 2 reviewers (SOK/AB) and full paper review divided among SOK/JM/ZT/NT/ JP/ZA/AB. Disagreements at the title screening and full paper level were discussed and consensus reached. A prespecified Excel spreadsheet for inclusion/exclusion criteria and data extraction was created to reduce data extraction error. Data extraction included publication information (year/author), study site(s), study design, patient population, sample size, outcome measure(s) used (e.g., specific disease activity or damage measure), comparator measures, and outcomes assessed [mortality, damage (free text for specific damage data)].
Data analyses included descriptive statistics. Mortality and damage outcomes were pooled using a DerSimonian Laird random effects model when enough data were available for studies using validated disease activity and damage measures (BV). The statistical heterogeneity was assessed with the I 2 test statistic with the following interpretations: 25-49% low heterogeneity; 50-74% moderate, and ≥ 75% high heterogeneity 5 . Pooled relative risks (RR) and OR with 95% CI comparing dead versus surviving patients with higher disease activity and damage were also calculated. Data from cohorts with multiple studies were included only once. Study quality was assessed using the Newcastle-Ottawa scale for observational studies 6 .

RESULTS
We identified 816 papers for full review. Fifty-three met the specific inclusion criteria for mortality and damage outcomes from high disease activity or damage scores after removing 33 for evaluating PRO (e.g., the Medical Outcomes Study Short Form-36 questionnaire, fatigue). These were the most common reasons for data exclusion: (1) not meeting the inclusion criteria (664 articles); (2) not being in English (18 articles); (3) being reviews (14 articles); and (4) involving rheumatic diseases other than SLE (14 articles (23) of studies were prospective, 9 were retrospective, 1 study was from a randomized control trial, 1 from an SLE clinical trials registry, and 2 were observational cohorts not otherwise specified. Higher mortality was noted in 6 studies 14,22,23,35,37,38 evaluating the effect of higher SLEDAI scores at baseline or over time (weight-adjusted HR 1.14, 95% CI 1.06-1.22; Figure 2). Sensitivity analyses evaluating the effect of higher baseline SLEDAI and higher time-adjusted SLEDAI [e.g., adjusted mean SLEDAI (AMS)] on mortality demonstrated a greater but not statistically significant mortality risk, with HR of 1.13 (95% CI 0.99-1.28) for baseline SLEDAI 35,37,38 and HR 1.19 (95% CI 0.99-1.43) for time-adjusted SLEDAI (data not shown) 14,20,21,22,23 .
A greater risk of mortality was found in patients with SLE with over 10 years of followup 14,22,23,38 (HR 1.13, 95% CI 1.06-1.21) compared to those with < 10 years of followup 35,37 in a sensitivity analysis (HR 1.98, 95% CI 0.53-7.49). Although increasing age is a significant risk factor for mortality, the majority of these patients were young adults, in whom the relative contribution of age to death is less significant. Three studies ( Figure 2) demonstrated higher SLEDAI scores in dead versus living patients with SLE 29,32,39 , and the odds of mortality were greater (OR 1.13, 95% CI 1.06-1.22) in SLE patients with higher SLEDAI disease activity 25 .
Higher disease activity scores were also associated with greater damage in patients with SLE ( Figure 3). Three studies 18,24,25 demonstrated greater damage in patients with higher SLEDAI scores at baseline with OR 1.08 (95% CI 1.03-1.12; Figure 3). Four studies 14,15,21,30 demonstrated  Many other studies could not be combined to produce meaningful metaanalyses; however, those studies that involve pediatric and adult SLE international cohorts in both outpatient and hospital settings do support the association between higher disease activity (using SLEDAI, Mexican SLEDAI, BILAG, SLAM, or ECLAM) and the outcomes of mortality and damage accrual, respectively (Table 1). For example, in the Hopkins Lupus Cohort, AMS was a significant independent predictor of mortality (HR 1.23, 95% CI 1.14-1.33) 19 . Hospitalized Chinese patients with SLE had a higher mortality (HR 1.64) when discharged with a SLEDAI > 8 17 , while a higher SLEDAI at diagnosis for a retrospective Chinese cohort of late-onset SLE (50 yrs or older) was an independent predictor of mortality (OR 1.091, p = 0.003) 25 . In confirming that the SDI score increased by 0.13 per year, Petri, et al also demonstrated that damage risk was greater for patients with higher disease activity (SELENA-SLEDAI); however, this association was not significant after adjusting for corticosteroid use 31 . SLE damage measures. The majority of studies evaluating the effect of damage on mortality and further damage involved the SDI (22 studies), with 14 prospective and 6 retrospective observational cohorts, 1 clinical trial registry, and 2 observational studies not otherwise specified. Specifically, 7 studies 10,15,35,48,49,55,59 found a significant effect on mortality with worse damage as measured by the SDI either at baseline or over time (HR 1.44, 95% CI 1.29-1.61; Figure 4). Sensitivity analyses were performed evaluating the effect on mortality of damage (measured by the SDI) at baseline and over time and confirmed HR of 1.35 (95% CI 1.25-1.46) and 1.57 (95% CI 1.29-2.11), respectively (data not shown). Four studies 23,32,39,51 demonstrated worse SDI in dead versus living patients; however, the data were statistically insignificant ( Figure 4). Although this was not statistically significant, the sample size and the heterogeneity of the included studies (as demonstrated by I 2 = 44%) could have affected these results. In addition, 2 studies 18,46 independently demonstrated greater odds of early damage in patients with baseline damage (worse SDI scores); however, the statistical effect was lost in the metaanalysis (OR 1.20, 95% CI 0.87-1.66) owing to high heterogeneity (I 2 = 66%; Figure 5  last measured SDI scores were higher in dead versus living pediatric patients with SLE in a Norwegian cohort 24 . A Swedish prospective inception cohort found that the SDI at 5 years postdiagnosis was predictive of survival and mortality up to a median of 7 years of followup 57 , while the Erlangen cohort from Nuremberg 53 confirmed that an increase of 2 or more points of SDI from the first to the third year of disease was prognostic for mortality (RR 7.7, p < 0.0001). Renal damage at 1 year was predictive of death within 10 years of diagnosis in a Pakistani SLE cohort 58 .

DISCUSSION
This systematic review suggests that disease activity and damage, as measured through various validated outcome measures, is overall associated with increased mortality and damage accrual. In particular, the effect of several different disease activity scores on the risk of mortality and damage accrual was consistently significant across the metaanalyses and largely confirmed by the independent studies. Overall, the risks of mortality and damage accrual were significantly higher with higher damage scores at baseline and over time, a result also supported by several individual studies. These results likely confirm what many SLE physicians have seen or demonstrated in their clinical practice and/or research settings. The multiple observational cohort studies suggest that disease activity and damage beget more damage and death. Therefore, physicians involved in SLE care should consider proper evaluation of disease activity and damage through a formalized method such as any of the disease activity and damage indices evaluated in our study (Supplementary Table 1, available with the online version of this article).
However, several issues arise from this systematic literature review that complicate this assumption. Despite identifying over 50 studies, the ability to evaluate the various outcome measures and pool studies for metaanalyses was challenging and limited largely to descriptive reporting. Reasons include the significant heterogeneity between cohorts and the objectives of each study, and differences in how and when specific outcome measures might have been collected in a particular cohort. Cohort descriptions varied in their detail, often with a wide range of disease activity or damage at baseline and followup. This may partially explain the variability in significant results when evaluating the effect of disease damage on mortality and damage accrual in the metaanalyses. Moreover, studies conflicted regarding the effect of damage on further damage accrual and even when comparing dead versus living patients based on differences in past damage. The effect of glucocorticoids and how they were analyzed in the studies may have contributed to varying effects of disease activity and damage on mortality and damage accrual. Great difficulty arose in separating out adult versus pediatric SLE, although attempts were made to include both in this review.
Regarding the completed metaanalyses, statistical heterogeneity (computed with I 2 ) and heterogeneity in the characteristics of the studied cohorts were consistent challenges, reflecting among other things the unequal weight among studies within the analysis. For example, when evaluating the effect of higher disease activity scores on mortality, the combined studies included multiple international cohorts from China, Brazil, Europe, and Canada, and included hospitalized and outpatient groups and combined pediatric and adult patients, which may imply differences in SLE phenotypes and therefore disease severity.
This review was not meant to evaluate the validation and comparability of disease activity and damage metrics. However, it demonstrated that consistency and comparability exist across different metrics in evaluating for important outcomes in SLE. The choice of which measure to use will be influenced by many factors. Some important examples include feasibility in the clinical setting (e.g., time, paper vs electronic records, language used) to familiarity of clinic staff with a particular measure. Many components of the SLE disease activity and damage metrics are recorded in a good medical history and physical examination, which should be standard for SLE patients with the potential for multisystem disease. Moreover, adoption of standardized disease activity measures has already become integrated into rheumatology clinics beyond academic settings (e.g., the 28-joint count Disease Activity Score in RA, the Bath Ankylosing Spondylitis Disease Activity Score in SpA). The studies in our review were never designed to answer our specific PICO search strategy, and this likely contributed to the great heterogeneity seen in the results. Nevertheless, qualitatively the results of different studies based on different metrics of disease activity and damage, along with the results of the metaanalysis and the sensitivity analyses, were concordant. Clearly the results highlighted the effect of disease activity (at baseline and/or over time) on damage and mortality, and the effect of damage on mortality and further accrual of damage over time. A multicenter study to answer these questions with greater consistency in the metrics used to measure disease activity and damage would assist in addressing this more directly. Moreover, such a study might address how specific disease activity scores might be used in clinical practice, and what targets are appropriate in identifying low disease activity or remitted states, as is well-recognized in the defined disease activity score targets for RA.
This systematic literature review was conducted to inform a broader set of recommendations for the assessment of patients with SLE in Canada, focusing specifically on what to perform in the monitoring of our patients. This review confirmed that increased disease activity and damage, measured by validated metrics, were associated with further damage and increased mortality. This body of evidence was generated from studies using validated measures of disease activity and damage at baseline and on followup. Thus, it would be helpful to use validated metrics in the assessment and monitoring of SLE. The use of validated metrics enables scientists, trialists, industry people, and policy makers to qualify and evaluate the implication of a specific construct on other outcomes. When deciding on the use of a specific metric, one should consider its administrative (time, scoring, complexity) and cost burden, and the preparedness and skill of the assessor on the selected metric. Measurement of a health state (disease activity, damage) is essential in daily practice and research 63 , and Lord Kelvin stated, "when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot, your knowledge is of a meager and unsatisfactory kind 64 ." The results will be included as part of the evidence-to-decision table in upcoming recommendations, which will incorporate other important domains that must be considered when developing recommendations according to GRADE.

ONLINE SUPPLEMENT
Supplementary material accompanies the online version of this article.