Survival in rheumatoid arthritis: A population-based analysis of trends over 40 years

Abstract

Objective

To evaluate trends in and risk factors for mortality among patients with rheumatoid arthritis (RA) over a 40-year period.

Methods

A population-based inception cohort was assembled from among all Rochester, Minnesota residents ages ≥18 years who were first diagnosed with RA (fulfilling the 1987 American College of Rheumatology criteria for RA) between January 1, 1955 and December 31, 1994. Patients were followed up longitudinally through their entire medical records (including all inpatient and outpatient care by any provider) until death or migration from the county. Survival was described using the Kaplan-Meier method. Observed and expected survival were compared using the log-rank test, and standardized mortality ratios (SMRs) with expected survival were based on the sex and age of the study population and death rates from the Minnesota life tables. Cox proportional hazards models were used to estimate the influence of extraarticular manifestations and comorbidities, controlling for age, sex, body mass index (BMI), smoking, and rheumatoid factor positivity.

Results

Survival in this RA cohort was significantly lower than that expected in the population (P < 0.001) over the entire time period. Patients with RA were at significantly higher risk of death, with an SMR of 1.27 (95% confidence interval 1.13–1.41). Excess mortality among women was more pronounced than among men, with SMRs of 1.41 and 1.08, respectively. Presence of ≥1 extraarticular manifestation was the strongest predictor of mortality after adjusting for age, sex, BMI, smoking, and rheumatoid factor positivity.

Conclusion

Survival in RA patients is significantly lower than expected. The strongest predictors of survival appear to be those related to RA disease complications, specifically, extraarticular manifestations of the disease and comorbidities. More attention should be paid to mortality as an outcome measure in RA.

The first study of mortality in rheumatoid arthritis (RA), which appeared in 1953, reported an excess mortality of 29% among RA cases compared with controls (1). Since that time, there have been numerous other reports of excess mortality in RA (2–23). Overall, there seems to be little doubt that people with RA have a lower life expectancy compared with members of the general population of the same age and sex. However, less is known about the determinants of excess mortality in RA and whether this excess mortality has improved over time (22, 24–26). This is particularly relevant because management of RA has changed rather dramatically over the past 50 years. We sought to evaluate predictors of and trends in mortality in a large inception cohort of patients with RA first diagnosed between January 1, 1955 and December 31, 1994 with followup to January 1, 2000.

PATIENTS AND METHODS

The population of Rochester, Minnesota is well suited for an investigation of RA mortality because comprehensive medical records for all residents who have sought medical care for more than half a century are available. A records linkage system allows ready access to the medical records from all health care providers for the local population, including the Mayo Clinic and its affiliated hospitals, the Olmsted Medical Group and the Olmsted Community Hospital, local nursing homes, and the few private practitioners. The potential of this data system for use in population-based studies has been described previously (27, 28). This system ensures virtually complete information on vital status for all clinically recognized cases of RA among Rochester, Minnesota residents.

Using this data resource, an inception cohort of all cases of RA first diagnosed between January 1, 1955 and December 31, 1994 among Rochester, Minnesota residents ≥18 years of age was assembled as previously described (22, 29, 30). All cases fulfilled the 1987 American College of Rheumatology (ACR; formerly, the American Rheumatism Association) criteria for RA (31). Incidence date was defined as the first date of fulfillment of at least 4 of the 7 diagnostic criteria. All cases were followed up longitudinally through their entire medical records until death or migration from Olmsted County. All causes of death (including contributing causes of death) as reported in the medical record and/or the death certificate were collected for all cases.

Survival curves originating from the RA incidence date were estimated using the Kaplan-Meier method. Observed and expected survival were compared using the log-rank test. Standardized mortality ratios (SMRs) with expected survival were based on the sex and age of the study population and death rates from the Minnesota (white population) life tables.

Cox proportional hazards models were used to estimate the influence of a number of variables on survival from incidence date. These variables included the confounders assessed at baseline, which were age, sex, history of smoking, history of alcohol use, body mass index (BMI), and functional status based on the Steinbrocker criteria (32) (stages I or II versus stages III or IV), as well as factors assessed throughout the followup, which were rheumatoid factor positivity (≥40 IU/ml), leukopenia (<3,500/μl), erythrocyte sedimentation rate (>22 mm/hour for men and >29 mm/hour for women), rheumatoid nodules, rheumatoid lung disease, extraarticular manifestations (i.e., pericarditis, pleuritis, Felty's syndrome, major cutaneous vasculitis, neuropathies, scleritis, episcleritis, retinal vasculitis, glomerulonephritis, and other vasculitis [assessed using predefined criteria as previously described (33, 34)]), cardiovascular disease (i.e., previous or acute myocardial infarction, angina pectoris, coronary artery disease, arrhythmias, dysrhythmias, hypertension, congestive heart failure, pulmonary edema, rheumatic heart disease, valvular stenosis or insufficiency, and peripheral vascular disease), chronic pulmonary disease (i.e., asthma, chronic obstructive pulmonary disease, emphysema, chronic bronchitis, pneumoconioses), peptic ulcer disease (i.e., peptic, gastric, duodenal or gastrojejunal ulcers, esophagitis, esophageal ulceration, symptomatic hiatus hernia), all grades and complications of diabetes mellitus, cancer (i.e., any malignancy excluding basal cell nonmelanomatous skin cancer), renal disease (i.e., nephrotic syndrome, chronic renal failure, and/or uremia), liver disease (i.e., alcoholic cirrhosis, nonalcoholic cirrhosis, biliary cirrhosis, chronic hepatitis, hepatic coma, portal hypertension, other sequelae of chronic liver disease, esophageal varices), dementia (including Alzheimer's disease, other dementias, senility, and/or cerebral atrophy), cancer chemotherapy (i.e., the initiation of any therapeutic agent for cancer treatment), and antirheumatic medications (including intramuscular or oral gold, sulfasalazine, hydroxychloroquine, azathioprine, D-penicillamine, methotrexate, leflunomide, immunosuppressants, or corticosteroids).

Factors assessed throughout the followup were modeled as time-dependent covariates, in that an individual without one of the characteristics at diagnosis of RA could develop it during followup. All 2-way interactions among significant main effects were examined. Bootstrap sampling was used to validate variable selection.

RESULTS

The incidence cohort comprised 609 cases of RA, as defined by the 1987 ACR criteria for RA (31), that were first diagnosed between January 1, 1955 and December 31, 1994. These patients were Rochester, Minnesota residents who were ≥18 years of age at incidence date, of whom 26.9% (164) were male and 73.1% (445) were female. The mean age at incidence of RA in this population was 58 years and the mean followup time was 14.2 years. The overall age- and sex-adjusted annual incidence of RA among Rochester residents ≥18 years of age (1955–1994) was 44.6 per 100,000 population (95% confidence interval [95% CI] 41.0–48.2) (Table 1). A detailed description of the incidence patterns in this cohort are contained in our previous report (30).

Table 1. Characteristics of 609 Rochester, Minnesota residents (≥18 years of age) who first fulfilled the 1987 American College of Rheumatology criteria for rheumatoid arthritis between January 1, 1955 and December 31, 1994

	Value
Number of patients	609
No. (%) female	445 (73.1)
No. (%) male	164 (26.9)
Length of followup, mean ± SD years	14.2 ± 9.4
Age at incidence, years
Mean, median	58.0, 58.2
Minimum, maximum	18.5, 92.8
Incidence rate per 100,000 population  (95% confidence interval)	44.6 (41.0–48.2)

Survival in this RA cohort was significantly lower than that expected in the general population (P < 0.001) over the entire time period (Figure 1). When compared with individuals from the same population of the same age and sex who did not have RA, patients with RA were at significantly higher risk of death, with an SMR of 1.27 (95% CI 1.13–1.41). The SMR among women was higher than that among men (1.41 and 1.08, respectively) (Table 2). For example, these data indicate that the life expectancy for a 50-year-old white woman in Minnesota is 34 years, while the life expectancy for a 50-year-old white woman with RA is only 30 years. Likewise, the life expectancy of a 50-year-old white man in Minnesota is 27 years, while the life expectancy of a 50-year-old white man with RA is only 26 years. Thus, these population-based data indicate that the excess mortality in RA appears to be more pronounced in women compared with that in men.

Figure 1. Survival among Rochester, Minnesota residents first diagnosed with rheumatoid arthritis (RA) between January 1, 1955 and December 31, 1994 (n = 609), compared with expected survival.

Table 2. Standardized mortality ratios among 609 rheumatoid arthritis incidence cases compared with expected rates in the general population

	Standardized mortality ratio	95% confidence interval
Women	1.41	1.22–1.61
Men	1.08	0.86–1.32
Overall	1.27	1.13–1.41

We then categorized our incidence cohort according to decade of RA incidence and compared survival using the Kaplan-Meier method. As shown in Figure 2, the Kaplan-Meier survival curves of all 4 incidence cohorts of Rochester residents with RA appeared to be nearly identical. A Cox regression analysis examining the effect of year of RA incidence on survival failed to identify a statistically significant difference in survival in the more recent years (P = 0.38). Causes of death in our cohort included cardiovascular disease, infections, malignancies, pulmonary diseases, and cerebrovascular diseases.

Figure 2. Comparison of survival among 4 incidence cohorts of Rochester, Minnesota residents with rheumatoid arthritis (RA), grouped by decade of RA incidence. NS = not significant.

Finally, we examined predictors of mortality in RA using Cox proportional hazards models. After adjusting for age, sex, BMI, history of smoking, and rheumatoid factor positivity, the presence of ≥1 extraarticular manifestation was the strongest predictor of mortality in RA (Table 3). A number of comorbid diseases were also statistically significant predictors of mortality, as well as the use of cancer chemotherapy, a history of alcohol use, and corticosteroid use. There were no important 2-way interactions among the variables considered.

Table 3. Predictors of mortality in 609 incidence cases of rheumatoid arthritis

	Hazards ratio*	95% confidence interval
* From the Cox proportional hazards model adjusted for age, sex, body mass index, history of smoking, and rheumatoid factor positivity.
Extraarticular manifestations	4.4	3.2–6.3
Comorbidity
Cardiovascular disease	1.6	1.2–2.1
Renal disease	2.2	1.4–3.4
Liver disease	3.3	1.7–6.7
Dementia	3.7	2.7–5.0
Cancer, no chemotherapy	1.9	1.4–2.6
Cancer, with chemotherapy	6.9	3.6–12.9
History of alcohol use	2.7	1.8–4.1
Steroid use	1.5	1.2–2.0

DISCUSSION

These population-based incidence data indicate that survival in RA patients is significantly lower than that expected in the general population, and that there has been no evidence of improvement in survival over the past 4–5 decades. The strongest predictors of survival appear to be those related to RA disease complications, specifically, extraarticular manifestations of disease and comorbidities. The leading causes of death were cardiovascular diseases and infections.

Coste and Jougla examined annual age-specific proportional mortality ratios for RA in men and women according to age over a 20-year time period (35). Their findings were consistent with our own and, again, indicated that survival in RA has not improved over the past several decades. More recently, Lindqvist and Eberhardt examined mortality in RA patients with disease onset in the 1980s (25). The study population in that report included patients participating in a prospective study of outcome in RA at Lund University in Sweden. Interestingly, there were 18 deaths among the RA patients, compared with an expected 20 deaths, suggesting a trend toward improved survival in their RA cohort. Although this finding is intriguing, it almost certainly indicates referral bias, in that patients referred to this prospective study were likely to differ from those with RA in the general population at large. Moreover, that analysis was limited to 12 years and, as indicated in our own analyses (Figure 1), the excess mortality in RA does not become apparent until ∼8–10 years after disease onset.

Two recent studies of clinic-based cohorts (20, 21) concluded that patients diagnosed and treated early have better outcomes. Extensive treatment with disease-modifying antirheumatic drugs was also associated with decreased mortality risk (36). The influence of methotrexate response on mortality was examined by Krause and colleagues (37). These investigators found that patients with severe RA who do not respond to methotrexate treatment have a poor prognosis, with a >4-fold increase in mortality compared with the general population, whereas those RA patients who do respond to methotrexate have only a moderately increased mortality rate. These findings are consistent with our own, which indicated that patients with severe disease (i.e., those who are more likely to have extraarticular manifestations) have a >4-fold increased risk of death (Table 3). Our findings are also consistent with those of Turesson and colleagues (33), who reported a mortality rate ratio of 2.49 for patients with extraarticular manifestations.

A recent review (26) of the major studies conducted to date concluded that the apparent improvement in survival may not be due to more aggressive antirheumatic treatment, but rather, to referral selection bias, i.e., the more frequent use of inception cohorts rather than referral cohorts in recent years. Although the strengths of our study include a population-based design, the use of a standardized, systematic approach for case ascertainment, and the completeness of ascertainment for vital status, the results of this study must be interpreted in light of its limitations. Selection bias is a possible limitation; however, it is expected to be minimal in this study because all incident cases of RA diagnosed in the study population during the period of interest have been ascertained through a medical record linkage system. Although use of such a system for case ascertainment has the limitation that cases of RA that do not come to medical attention may have been missed, this is unlikely in RA because we believe that all cases of RA will eventually come to medical attention. Because some racial and ethnic groups are underrepresented in Rochester, Minnesota, where the population in 2000 was 90.3% white according to the US census data, the results of our population-based study are only generalizable to the US white population. Our results may also be criticized for having inadequate followup to reveal improvements in survival. However, improvements in survival have been documented in other chronic diseases over a similar time period (38). In addition, our own data show statistically significant improvements in survival over the past 25 years in systemic lupus erythematosus (39).

With inclusion of 18–34-year-old RA patients and longer followup of the same population-based inception cohort, our current SMRs are slightly lower than, but still within the same range as the values reported previously (22). The reason for the lower values is use of a slightly different comparison group for calculation of SMRs. Our previous SMRs were calculated using a population-based control group, whereas the current estimates are based on expected survival for the Minnesota population. We plan to continue to compare survival in our RA cohorts with expected survival for the Minnesota population at regular intervals in the future, in the hope that a followup study will reveal a more favorable survival experience for patients with RA. Our data do not include longitudinal information about functional status using common tools (such as the Health Assessment Questionnaire [40], for example), and it is quite possible that patients with RA, although not living longer, are indeed functioning at a higher level. Nonetheless, we believe that our findings of significantly poorer survival in RA cases compared with population controls, with no evidence of improvement over the past 4–5 decades, point to the need for more research on the determinants of mortality in RA and to the importance of assessing the long-term outcome, particularly with respect to mortality, related to new RA interventions.