Rheumatoid arthritis (RA) is associated with an increased risk of dying. This risk is elevated when RA is severe, as defined by the presence of advanced joint damage, functional limitations, disability, extraarticular disease, and rheumatoid factor positivity (1–8). Because RA is a joint-centered disease that, in most patients, does not affect the viscera, its increased mortality engenders questions about the cause of death. In fact, patients with RA die from a wide variety of illnesses (2, 6, 9). Their burden of coexistent diseases, or comorbidity, is also higher than expected as compared with people of the same age and sex without RA (10). Comorbidity has a negative influence on the physical and psychosocial condition of RA patients (11), and may be a major factor explaining the disease's high death rate (12). Although both disease severity and comorbidity have recognized associations with mortality in RA, their independent contribution to the death risk among RA patients has received less attention.
In the present analysis, we carefully measured disease severity and comorbidity in a sample of RA patients. We hypothesized that mortality would be associated with disease severity in RA, independent of comorbidity.
- Top of page
- PATIENTS AND METHODS
We enrolled 779 patients. Their baseline characteristics, according to vital status as of April 30, 2002, are shown in Table 1. The followup time period ranged from 0.1 year to 6.3 years (mean 2.52 years), for a total period of observation of 2,315 person-years. During this time, 75 patients died (9.6%), for a mortality rate of 3.2 per 100 patient-years (95% CI 2.6–4.1).
Table 1. Baseline characteristics of the rheumatoid arthritis (RA) study sample, according to vital status at followup
|Individual characteristic||Vital status at followup*||P|
|Living (n = 704)||Dead (n = 75)|
|Sociodemographic variables|| || || |
| Age, mean ± SD years||54 ± 13||67 ± 9||≤0.001|
| Men, n (%)||195 (28)||34 (45)||0.001|
| Years of education, mean ± SD||11 ± 4||9 ± 4||≤0.001|
|Race/ethnic background, n (%)|| || ||0.5|
| Hispanic||391 (56)||43 (57)|| |
| White||243 (35)||29 (39)|| |
| African American||50 (7)||3 (4)|| |
| Asian||14 (2)||0 (0)|| |
| Other ethnic group||6 (1)||0 (0)|| |
|Clinical features|| || || |
| Years with RA, mean ± SD||11 ± 10||16 ± 13||≤0.001|
| Tender joint count, mean ± SD||15 ± 13||17 ± 14||0.1|
| Swollen joint count, mean ± SD||8 ± 7||7 ± 7||0.3|
| Deformed joint count, mean ± SD||9 ± 10||18 ± 14||≤0.001|
| Subcutaneous nodules, n (%)||206 (29)||27 (36)||0.2|
| Rheumatoid factor positive, n (%)||620 (88)||64 (85)||0.7|
|Functional status and disability level|| || || |
| Steinbrocker functional class, n (%)|| || ||≤0.001|
| Class I||160 (23)||3 (4)|| |
| Class II||359 (51)||27 (36)|| |
| Class III||161 (23)||29 (39)|| |
| Class IV||24 (3)||16 (21)|| |
| Modified Health Assessment Questionnaire, mean ± SD score||1.8 ± 0.7||2.3 ± 0.9||≤0.001|
|Comorbidity instruments|| || || |
| COMDUSOI, mean ± SD†||46 ± 22||66 ± 19||≤0.001|
| Charlson Comorbidity Index, n (%)|| || ||≤0.001|
| 1||419 (60)||22 (29)|| |
| 2||169 (24)||29 (39)|| |
| ≥3||116 (17)||24 (32)|| |
|RA disease severity instruments|| || || |
| RADUSOI, mean ± SD‡||48 ± 13||57 ± 16||≤0.001|
| Global disease severity, mean ± SD score||2.8 ± 2.1||4.8 ± 2.9||≤0.001|
The inter- and intrarater reliability of the patient-averaged DUSOI was 0.87 and 0.90, respectively. For the total Charlson Comorbidity Index score, the interrater reliability was 0.94. These reliability estimates may reflect a lower bound because, in our study, all comorbidity scores underwent a second level of review by one or more physicians who discussed the case prior to the final score. Graphs depicting the frequency distribution of the RA severity and comorbidity scores among the 779 patients are shown in Figure 1. The Spearman's correlation between the 2 RA severity scales, the RADUSOI and the RA global severity rating, was 0.55 (P ≤ 0.0001), while the correlation between the 2 comorbidity scales, the COMDUSOI and the Charlson Index, was 0.45 (P ≤ 0.0001). The correlations between the comorbidity scales and RA disease severity scales were modest, with r values ≤0.23 in the 4 pairwise comparisons between both comorbidity scales and both RA severity scales.
Figure 1. Frequency distribution of the rheumatoid arthritis (RA) Duke Severity of Illness Checklist (RADUSOI), the RA physician-rated global severity measure, the non-RA Comorbidity DUSOI (COMDUSOI), and the Charlson Comorbidity Index scores among patients with RA (see Patients and Methods for detailed descriptions of each measure).
Download figure to PowerPoint
The 2 RA disease severity scales correlated, to a similar extent, with the other concurrently ascertained variables that reflect the intensity of disease activity or the extent of joint damage (Table 2). The correlation was strongest with the measurements of joint damage and physical disability. Nevertheless, both RA severity scales also correlated modestly, but significantly, with the measures of disease activity. The presence of subcutaneous nodules, a marker of severe disease, was also associated with significantly higher scores on both RA severity scales (mean ± SD RADUSOI score 54 ± 13 among patients with nodules versus 47 ± 13 among those without nodules [P ≤ 0.0001]; mean ± SD global RA severity score 3.9 ± 2.6 among patients with nodules versus 2.6 ± 2.0 among those without nodules [P ≤ 0.0001]).
Table 2. Correlation between RA disease severity measures and criterion standards*
|Criterion||Disease severity measure|
|RADUSOI||Global RA severity|
|Tender joint count||0.23||0.19|
|Swollen joint count||0.19||0.18|
|Deformed joint count||0.55||0.67|
|Erythrocyte sedimentation rate†||0.17||0.23|
|Steinbrocker functional classification||0.44||0.68|
|Modified Health Assessment Questionnaire score||0.34||0.38|
Both comorbidity scales and RA severity scales displayed strong bivariate associations with the probability of survival, and these associations were statistically significant. Figure 2 plots the survival function according to comorbidity categories, and similarly, Figure 3 plots the survival function according to disease severity.
Figure 2. Estimated survival curves for each comorbidity scale, using the Kaplan-Meier technique. Top, Survival in relation to the COMDUSOI categories. The difference in survival probability between the 3 categories was significant (log-rank χ2 [with 2 degrees of freedom] = 29.24, P ≤ 0.0001). Bottom, Survival in relation to the Charlson Comorbidity Index categories. Patients with a Charlson score of 1 had RA only, without other comorbidities scored on this scale (log-rank χ2 [with 2 degrees of freedom] = 33.18, P ≤ 0.0001). Values next to the categories are the number of patients in each category in each year of followup. See Figure 1 for definitions.
Download figure to PowerPoint
Figure 3. Estimated survival curves for each RA severity scale, using the Kaplan-Meier technique. Top, Survival according to the RADUSOI categories. The difference in survival probability between the 3 categories was significant (log-rank χ2 [with 2 degrees of freedom] = 15.03, P ≤ 0.0001). Bottom, Survival according to the RA global severity scale (log-rank χ2 [with 2 degrees of freedom] = 33.5, P ≤ 0.0001). Values next to the categories are the number of patients in each category in each year of followup. See Figure 1 for definitions.
Download figure to PowerPoint
We used multivariate Cox regression to explore the independent influences of RA severity and comorbidity on survival. We tested 4 models, each of which included one comorbidity scale and one disease severity scale. Our objective with this approach was to show that the effect of severity and comorbidity on mortality did not depend on the type of measurement scale. We included age, sex, and disease duration as covariates in these models to control for the possibility of confounding by these potentially important mortality predictors. We found significant independent associations with mortality for both of the disease severity scales and both of the comorbidity scales. The age-, sex-, and disease duration–adjusted hazard ratios for each comorbidity and severity measure are provided in Table 3. These hazard ratios should be interpreted as the proportional change in the risk of death associated with a 1-unit change in the independent variable. Thus, a hazard ratio of 1.03 associated with the COMDUSOI represents a 3% increase in the risk of death per unit increase in that scale, which ranged from 0 to 98 in our study. In the case of the Charlson Index, which in the present study ranged from 1 to 9, each unit increase was associated with an even larger increase in the hazard ratio for death, because Charlson units represent a greater proportion of the scale's range.
Table 3. Independent association of RA disease severity and comorbidity with mortality*
|Variable||Risk of death|
|Model 1||Model 2||Model 3||Model 4|
|Severity scale|| || || || |
| RADUSOI (range 6–93)||1.02 (1.00–1.04)†||1.03 (1.02–1.05)‡||–||–|
| Global severity (range 0–10)||–||–||1.20 (1.10–1.31)‡||1.29 (1.18–1.41)‡|
|Comorbidity scale|| || || || |
| COMDUSOI (range 0–98)||1.04 (1.02–1.05)‡||–||1.03 (1.02–1.05)‡||–|
| Charlson Index (range 1–9)||–||1.34 (1.15–1.56)‡||–||1.32 (1.13–1.54)|
|Likelihood ratio chi-square§||126.7||109.5||137.3||124.4|
We also estimated the diagnostic accuracy of these baseline measurements as predictors of death during the observation period. We modeled death as a logistic function of age, sex, disease duration, and the comorbidity and severity scores. We then estimated the predicted probability of death for each person, based on each of the models tested, and calculated the area under the ROC curve for the prediction. Models including both a severity and a comorbidity variable had significantly greater area under the ROC curve (P ≤ 0.005 for the increase in ROC area) than did a model that included only age, sex, and disease duration. This is shown graphically in Figure 4.
Figure 4. Receiver operating characteristics (ROC) curves of logistic models, using age + sex + disease duration alone as mortality predictors (broken jagged line in Top and Bottom) (area under the ROC curve 0.79, 95% confidence interval [95% CI] 0.75–0.84) versus age + sex + disease duration + disease severity + comorbidity. Top, Solid jagged line represents age + sex + disease duration + COMDUSOI + RADUSOI (area under the ROC curve 0.84, 95% CI 0.80–0.89). Bottom, Solid jagged line represents age + sex + disease duration + RA global severity + Charlson Comorbidity Index (area under the ROC curve 0.83, 95% CI 0.79–0.87). See Figure 1 for other definitions.
Download figure to PowerPoint
Because of the possibility that our inclusion of deaths that occurred within 1 year of recruitment in these analyses could overestimate the influence of comorbidity and disease severity, we repeated all of the analyses after excluding the 19 patients who died within 1 year. All of our findings remained essentially unchanged following this exclusion, without loss of significance of comorbidity or disease severity as death predictors in RA.
- Top of page
- PATIENTS AND METHODS
Mortality in this RA sample was associated with disease severity in a manner that was independent of the burden of coexistent disease. This finding is important because it further situates RA as a systemic illness with pathophysiologic consequences that transcend the joint. Markers of disease severity were among the first factors that were recognized to increase the risk of death among RA patients (1, 3, 4, 11). However, early studies of survival in RA did not include formal measures of coexistent disease, leaving open the possibility of confounding by this variable. This omission may, in part, explain why the high mortality risk in this disease is under-recognized outside of rheumatology, since it is not intuitively apparent why a disease that targets joints should also increase the risk of death.
More recently, a study in which comorbidity was measured carefully demonstrated that RA is indeed accompanied by a greater presence of coexistent illness, as compared with the observations in age- and sex-matched subjects without RA (10, 12). This confirmed that, as in other diseases, comorbidity in RA has a strong negative influence on survival (10, 12). The contribution of our study has been to provide an estimate of the independent effect of both disease severity and comorbidity on RA survival.
We used 2 validated scales to measure comorbidity. The Charlson Comorbidity Index is a checklist of 18 conditions, each weighted by the mortality risk conferred to hospitalized people (16). This widely used index is simple and does not require the investigator to judge the severity of each condition, which makes it convenient and facilitates its implementation for research. However, the Charlson Index is not sensitive to variation in severity within a given condition, and it limits investigators to 18 conditions. The DUSOI complements the Charlson Index by providing an open-ended way to classify a condition's severity. This allows the clinician-investigator to record and severity-score any health problem. In our study, both the DUSOI and the Charlson Index had excellent reliability.
We extracted the component attributable to RA alone from the DUSOI, which we called the RADUSOI. We also recalculated the score excluding the RA values, which we called the COMDUSOI. Our purpose in doing these operations was 2-fold. First, we wanted to remove the substantial contribution of RA to the DUSOI score because it interfered with our goal of separating RA severity from comorbidity. In removing the RA contribution from the DUSOI, our second goal became a byproduct in that we were able to calculate the sum of the scores for symptoms, complications, prognosis, and treatability of RA to produce the RADUSOI, which thus provided a reasonable RA severity index. This was borne out by the good correlation of the RADUSOI with the deformed joint count, Steinbrocker functional class, and M-HAQ score (Table 2). The RA global severity assessment, scored on a 0–10 scale by the examining clinician, also correlated well with concurrent variables that reflected RA severity. All 4 of the scales used in our study were strongly linked to survival in bivariate analyses, even when only deaths that occurred ≥1 year after the beginning of the observation period were considered. Adjustment for the potential confounding effects of age, sex, and disease duration did not remove this association, which is an important observation, particularly given the likelihood of an association between age and comorbidity.
The predictive accuracy of a multivariate model such as that tested here is provided by the area under its ROC curve. Without any information about a patient, random choice would give an accuracy of 50% in predicting death, equal to a coin flip. We have shown that information on a patient's age, sex, and disease duration increases the predictive accuracy to 79%, and that adding comorbidity and disease severity further raises the accuracy to 84%. This compares favorably to the accuracy of frequently used diagnostic systems (27).
Our aim with these analyses was to test the independent effect of disease severity and comorbidity as determinants of mortality in RA. We thus employed what could be described as a purely biomedical model, in that it included only disease-related variables in the analysis. Although we documented a significant association between these variables and mortality, there are likely to be additional factors associated with mortality in RA that we did not consider here. Examples of factors that would fit within a bio-psychosocial health paradigm, and which may also affect mortality in RA, include educational level, marital status, psychological limitations, and other psychosocial indicators (24). Exploring how these factors influence mortality after accounting for disease severity and comorbidity represents an interesting avenue for future research.
Some aspects of our study merit a cautionary note in interpretation. We recruited the patients in our sample from clinics where they had a scheduled appointment with a rheumatologist. Thus, our findings are most applicable to the type of patient with established RA who is usually seen in a rheumatologist's office for continuing care. The clinical facilities from which we recruited patients provided us with the full range of RA disease severity and comorbidity. This enabled us to test valid hypotheses about the influence of disease severity and comorbidity on mortality (28). However, it should be noted that the prevalence and distribution of comorbidity in our sample may differ from that of the total population of RA patients.
We recruited a significant proportion of our patients after their RA had been present for a number of years, which allowed us to study the full range of disease-induced damage. However, this may also have introduced bias by excluding patients who died or whose disease remitted before we began recruitment. Replicating our analysis in an inception cohort, however, would likely require a substantially larger sample and a longer followup period in order for comorbidity, disease-induced damage, and mortality to accrue to an extent sufficient to test our hypotheses.
Our findings have potential clinical implications. The observation that comorbidity is a major contributor to mortality in RA should serve to focus attention on other diseases that may affect RA patients. Although it is not a widespread practice, a significant proportion of rheumatologists fulfill a primary care role for RA patients (29). Many RA patients also look to the rheumatologist as their principal doctor, even when a primary care physician has been assigned to them. In light of our findings, rheumatologists should continue to keep a watchful eye on their patients' nonrheumatologic coexistent illnesses.
It is not clear how the severity of RA could lead to an increased mortality risk in a manner independent of comorbidity. Plausibly, severe RA can mask the recognition of comorbid conditions associated with mortality, and these may not be picked up by comorbidity instruments. Alternatively, severe RA can be associated with occurrence of conditions that cause sudden, unexpected death, which would also not appear in comorbidity scales. It is also possible that severe RA leads to a greater exposure to medications and their side effects. Neither of the comorbidity measurement scales we used provide for direct scoring of drug toxicity.
The high death risk associated with RA may also be a consequence of an excess allostatic load, or global adverse physiology caused by the presence of a chronic illness (30). Stress induced by RA or other persistent unfavorable conditions may lead to a maladaptive response by the neuroendocrine and other systems. This process, over time, is believed to increase the cumulative biologic risk to the cardiovascular and other systems (31). In-depth physiologic measurements are needed to capture this variable (32), and further study is needed to understand its possible role in explaining mortality and other outcomes in RA.
Finally, it is also possible that we overestimated the effect of comorbidity on mortality. Some of the comorbidities we encountered (for example, osteoporosis, diabetes mellitus) may be linked to RA and its treatment. We did not attempt to tie individual comorbidities to RA. It is likely that this would have resulted in a stronger association between RA severity and mortality than that observed in this study.
We conclude that mortality in RA is independently associated with the severity of the disease and with the extent of the comorbid disease burden. Further studies are needed to understand the mechanisms whereby disease severity can lead to increased mortality in RA.