Glucocorticoid Dose Thresholds Associated With All-Cause and Cardiovascular Mortality in Rheumatoid Arthritis

Authors


Abstract

Objective

To delineate daily and cumulative glucocorticoid dose thresholds associated with increased mortality rates in rheumatoid arthritis (RA).

Methods

We studied RA patients recruited from rheumatology clinics. Annually, we assessed the glucocorticoid dose, demographic, socioeconomic, clinical, and laboratory features of RA, cardiovascular (CV) risk factors, and vital status. We used Cox proportional hazards regression to assess associations between the daily or cumulative glucocorticoid dose and death, adjusting for potential confounders and for the propensity to receive glucocorticoids. We tested strata of the glucocorticoid dose to delineate the threshold associated with death.

Results

We studied 779 RA patients with a total of 7,203 person-years of observation, during which 237 of them died, yielding a mortality rate of 3.2 per 100 person-years (95% confidence interval [95% CI] 2.8–3.7). One hundred twenty of the deaths were due to CV causes, yielding a CV mortality rate of 1.8 (95% CI 1.5–2.1). Exposure to glucocorticoids was associated with a dose-dependent increase in death from all causes, with a ratio (HR) of 1.07 per mg of prednisone per day (95% CI 1.05–1.08). Compared to patients who were not receiving corticosteroids, the minimum daily prednisone dose threshold associated with an increase in all-cause mortality was 8–15 mg, with an adjusted HR of 1.78 (95% CI 1.22–2.60). For the cumulative dose of glucocorticoids, the minimum dosage associated with all-cause mortality was 40 gm (HR 1.74 [95% CI 1.25–2.44]).

Conclusion

Glucocorticoid use in RA is associated with a dose-dependent increase in mortality rates, with a daily threshold dose of 8 mg, at which the number of deaths increased in a dose-dependent manner. These findings may assist clinicians in selecting the appropriate glucocorticoid dosage for RA patients who require these agents.

Glucocorticoids can benefit patients with rheumatoid arthritis (RA) ([1-3]). However, the use of these agents is a double-edged sword, with significant toxicity counterbalancing their beneficial effects ([4, 5]). A troubling but little discussed consequence of glucocorticoid use is increased mortality rates ([6-8]), which could be related to their cardiovascular toxicity ([9, 10]). The dose-dependent nature of most glucocorticoid side effects ([4]) suggests that toxicity may be avoided if the dosage is maintained under certain thresholds. We are not aware of any previous efforts to delineate the dose-response association of glucocorticoids with death in patients with RA.

Our objective was to define thresholds in the daily and cumulative glucocorticoid doses associated with death in patients with RA. We studied a cohort of RA patients we have followed closely for more than a decade, with annual physical and laboratory measurements of joint inflammation and damage, glucocorticoid use, cardiovascular risk factors, and mortality rates. This allowed us to estimate the association between glucocorticoids and death, with adjustment for RA severity and for the patient's propensity to receive glucocorticoids.

PATIENTS AND METHODS

Patients

Between January 1996 and April 2001, we recruited from 6 local rheumatology clinics consecutive patients who met the American College of Rheumatology 1987 criteria for the classification of RA ([11]). During a comprehensive baseline evaluation and subsequent annual followup assessments, we collected the variables described below.

Comorbidity assessment

To assess the extent of comorbidity, we used the Duke Severity of Illness (DUSOI) checklist ([12]), as previously described ([13]). For these assessments, a physician compiled a list of the patient's active health problems and, using a scale of 0–4, assigned a severity value to each problem for each of the following 4 dimensions: symptoms, complications, prognosis, and expected response to treatment. The 4 values were then summed, divided by 16, and multiplied by 100, yielding a range of 0–100 for the severity of each problem, with higher scores indicating greater severity. The highest-ranking problem was given the most weight, with progressively diminishing contributions from each succeeding health problem, according to the following formula:

display math

where Dxmax is the highest-ranking health problem, and Dxn is each subsequent problem according to its severity ranking (n). Physicians who scored the DUSOI were trained according to the standardized scoring protocol.

To quantify the burden of illness outside of RA (the comorbidities), we eliminated the values for RA; this scale we called the COMDUSOI. We also used the scores due to RA alone to measure RA severity; this scale we called the RADUSOI. We have previously shown that these variables are reproducible (interrater reliability 0.87), are associated with concurrent measures of RA severity, and are able to predict deaths in an RA cohort ([13]).

Assessment of RA manifestations

A physician assessed patients for tenderness or pain on motion, swelling, or deformity in 48 joints, as well as for the presence of extraarticular subcutaneous nodules ([14]). Reliability coefficients were 0.94 for tenderness/pain on motion, 0.90 for swelling, and 0.98 for deformity. We measured the erythrocyte sedimentation rate (ESR) using the Westergren method and serum rheumatoid factor (RF) using the latex technique. To quantify RA severity, we used the RA ratings of the DUSOI as described above ([13]). The RADUSOI is reproducible (interrater intraclass correlation coefficient 0.87), associated with concurrent measures of RA disease severity, and predictive of death ([13]).

Socioeconomic status

We used self-reported years of education, monthly income, and current or past occupation to compute a socioeconomic status score according to the method described by Nam and Powers ([15]). We divided the score into 3 ascending socioeconomic status categories: lower (scores of 1–24), middle (scores of 25–46), and upper (scores of 47–91).

Anthropometrics

We used the body mass index (BMI; calculated as the weight [kg] ÷ height [m2]) to classify patients into 1 of 4 categories: underweight (BMI <20), normal weight (BMI ≥20 to <25), overweight (BMI ≥25 to <30), and obese (BMI ≥30).

Assessment of cardiovascular risk factors

We defined hypertension according to the presence of 1 of the following 4 criteria: a physician's diagnosis recorded in the medical records, the use of antihypertensive medications, a diastolic blood pressure ≥90 mm Hg, or a systolic blood pressure ≥140 mm Hg. Diabetes mellitus was defined as the presence of 1 of the following 3 criteria: a physician's diagnosis recorded in the medical records, use of antidiabetic medications, or a fasting blood sugar level of ≥126 mg/dl. We defined hypercholesterolemia according to 1 of the following 3 criteria: a physician's diagnosis recorded in the medical records, use of lipid-lowering medications, or a fasting plasma cholesterol level of ≥200 mg/dl. Patients were classified as current smokers if they smoked at the time of the initial study visit and as former smokers if they had quit.

Ascertainment of deaths

We learned of deaths through family, friends, obituaries, neighbors, physicians, and the public health department. Online mortality databases were searched monthly using ad hoc computer programs. In April and May of 2010, we made further efforts to ascertain vital status by contacting patients directly. For those we could not contact, we reviewed their medical records to determine the date of their last contact with the health system. All deaths were confirmed by death certificate. Deaths were considered cardiovascular-related if the International Classification of Diseases, Ninth Revision (ICD-9) code for the immediate or underlying cause of death shown on the death certificate was between 390 and 459.

Ascertainment of glucocorticoid use

We assessed glucocorticoid exposure as described previously ([16]). Briefly, from a list of current medications elicited from patients at the time of the study visit and supplemented by a review of pharmacy and medical records, we determined whether patients were receiving glucocorticoids and, if so, what their daily dose was. We estimated the cumulative oral glucocorticoid dose by multiplying the current daily dose by the number of days since glucocorticoid treatment was first initiated. In most cases, glucocorticoids were initiated prior to the initial study visit. For alternate-day or other nondaily schedules, we averaged the dose to obtain the daily amount. We considered patients who received glucocorticoids less often than monthly as not receiving glucocorticoids. At each followup visit, we calculated the cumulative dose since the previous visit. The total cumulative glucocorticoid dose is provided by the summed dose over each interval since glucocorticoid initiation (expressed in prednisone equivalents). As a measure of the intensity of exposure, we divided the cumulative glucocorticoid dose at each observation by the length of time since glucocorticoids were initiated (expressed as grams per year).

Statistical analysis

We used a person-years approach to calculating the incidence rates and 95% confidence intervals (95% CIs) ([17]). We graphed survival using the Kaplan-Meier method, testing differences using the log-rank test ([18]). Cox proportional hazards regression with time-varying covariates was used to assess the association between daily or cumulative glucocorticoid dose and all-cause mortality or CV mortality, adjusting for potential confounders ([19]). All variables that could change over time, with the exception of smoking, were allowed to vary over time in the models. Variables were added in groups in 4 consecutive models. Model 1 included the daily or cumulative glucocorticoid dose. Model 2 included the model 1 variable plus age at onset of RA, RA duration, sex, race/ethnicity (non-Hispanic white versus other), and the Nam and Powers socioeconomic status categories. Model 3 included model 2 variables plus tender, swollen, and deformed joint counts, subcutaneous nodules, the ESR, and the RADUSOI disease severity scale. Model 4 included model 3 variables plus diabetes mellitus, hypercholesterolemia, hypertension, cigarette smoking, and BMI.

In an alternative approach to account for potential confounding by indication, we estimated the patient's propensity to receive glucocorticoids and used the propensity score as a covariate in Cox models of the glucocorticoid–mortality association. The propensity score was the predicted probability that a patient would be taking glucocorticoids at a given visit (assessed for each visit), which was calculated from a logistic regression model using the independent variables listed in model 4 above ([20]). We examined the performance of the propensity score in equalizing groups of patients defined by glucocorticoid exposure, by comparing the 2 groups of patients with adjustment and without adjustment for the score.

To identify thresholds above which mortality risk increased, we categorized the daily and cumulative glucocorticoid doses into various quantiles of their respective distributions and tested the quantiles as categorical variables for association with death. We aimed to find the smallest dose category associated with a rise in mortality rates. We tested separate models for death from all causes and for death from CV causes. We tested the assumption of proportional hazards for all variables in the Cox models by testing the hypothesis that the slope of their Schoenfeld residuals over time was zero ([21]), and by visually inspecting plots of the survival probability against time. Variables that did not meet the proportional hazards assumption were retested after grouping into quantiles or, alternatively, using stratified Cox regression models with the unmodified variable ([22]). All analyses were conducted using the Stata software package, version 9.0 for Windows (StataCorp).

RESULTS

We recruited 779 patients with RA; these patients have been described previously ([13, 16]). Table 1 shows their characteristics at the first visit and at the pooled followup visits, according to whether or not they were receiving glucocorticoids at the time of recruitment. A total of 393 patients (50%) were receiving glucocorticoids at baseline. On average, patients receiving glucocorticoids had a significantly longer duration of RA, more deformities, and more severe disease, as measured by the RADUSOI, but were not significantly different in other respects from those not receiving glucocorticoids at baseline. The mean ± SD daily dose of prednisone (or prednisone equivalents) at baseline was 6.9 ± 5.0 mg (median 5 mg [range 1–60 mg]). The mean ± SD cumulative dose of glucocorticoids among the 522 patients who received glucocorticoids was 4.2 ± 47.3 gm (median 1.4 gm [range 5 mg to 87 gm]).

Table 1. Characteristics of the 779 RA study patients at the baseline and followup visits, by glucocorticoid exposure*
 Patients receiving glucocorticoids at baseline visitPooled followup visits at which patients were receiving glucocorticoids
NoYesPadj.NoneSomeAllPadj.
  1. P values were adjusted (Padj) for the propensity to receive glucocorticoids; see Patients and Methods for a description of the propensity estimates. Padj values for the pooled followup visits are for the comparison of the some visits and all visits groups, respectively, versus the no visits group. RA = rheumatoid arthritis; ESR = erythrocyte sedimentation rate; RF = rheumatoid factor; RADUSOI = Duke Severity of Illness due to RA alone; CV = cardiovascular; BMI = body mass index.
  2. aUnadjusted P ≤ 0.01 versus the group not receiving glucocorticoids at baseline or versus the group with no pooled followup visits at which patients were receiving glucocorticoids.
  3. bUnadjusted P ≤ 0.05 versus the group with no pooled followup visits at which patients were receiving glucocorticoids.
  4. cUnadjusted P ≤ 0.001 versus the group not receiving glucocorticoids at baseline or versus the group with no pooled followup visits at which patients were receiving glucocorticoids.
No. of patients or no. of patients/no. of visits386393257/1,451303/1,837219/1,259
Age at RA onset, mean ± SD years44 ± 1444 ± 13
RA duration, mean ± SD years10 ± 1012 ± 10a0.313.6 ± 10.115.5 ± 11.815.7 ± 130.6b0.8
No. (%) men102 (26)126 (32)385 (26)598 (32)404 (32)
No. (%) white134 (35)138 (35)492 (33)669 (36)387 (30)
RA manifestations       
Tender joint count, mean ± SD15 ± 1314 ± 1212.1 ± 12.115.2 ± 13.3c15..3 ± 12.8a0.4, 0.1
Swollen joint count, mean ± SD8 ± 78 ± 73.6 ± 4.84.9 ± 5.8c4.8 ± 5.0c0.04, 0.5
Deformed joint count, mean ± SD9 ± 1011 ± 12a0.712.7 ± 11.714.8 ± 12.1b16.9 ± 12.8c0.8, 0.5
No. (%) with nodules112 (29)122 (31)372 (25)534 (29)379 (30)
ESR, mean ± SD mm/hour40 ± 2341 ± 2238.8 ± 25.241.0 ± 25.843.1 ± 26.0b0.1
No. (%) RF positive280 (81)306 (81)866 (68)1208 (74)919 (82)c0.2
RADUSOI score, mean ± SD47 ± 1251 ± 13c0.447.9 ± 12.051.1 ± 11.8c53.5 ± 12.5c0.5, 0.9
CV risk factors       
Diabetes mellitus, no. (%)54 (14)47 (12)322 (22)362 (19)218 (17)
Hypercholesterolemia, no. (%)30 (8)28 (7)437 (30)604 (32)394 (31)
Hypertension, no. (%)191 (49)196 (50)915 (63)1130 (61)752 (59)
Past smoking, no. (%)155 (40)161 (41)496 (39)701 (43)477 (42)
Current smoking, no. (%)74 (19)79 (20)276 (19)358 (19)252 (20)
BMI, mean ± SD kg/m228 ± 628 ± 629.6 ± 6.829.0 ± 5.828.8 ± 6.8
Glucocorticoids       
Daily dose, mean ± SD mg0.06.9 ± 5.00.03.3 ± 4.96.5 ± 3.9
Cumulative dose, mean ± SD gm12.5 ± 16.40.02.3 ± 5.46.9 ± 8.8
Intensity of exposure, mean ± SD gm/year0.01.8 ± 1.70.02.7 ± 3.36.1 ± 4.2

By the censoring date of May 31, 2010, the 779 patients had accrued 7,203 person-years of observation, an average of 9.2 years per person (median 10.3 years [range 1 day to 14.2 years]). During this time, 237 patients died, which yielded a mortality rate of 3.2 deaths per 100 person-years (95% CI 2.8–3.7). The patients who received glucocorticoids during the observation period experienced 142 deaths during 3,291 person-years, yielding a mortality rate of 4.3 per 100 person-years (95% CI 3.6–5.0). The 257 patients who did not receive glucocorticoids during the pooled followup visits experienced 95 deaths during 3,912 person years, yielding a mortality rate of 2.4 per 100 person-years (95% CI 1.9–2.9). The incidence rate ratio between the two groups was 1.77 (95% CI 1.36–2.32; P ≤ 0.001).

There were 120 deaths from CV causes. Combined with the 542 patients who were alive at the last followup prior to the censoring date (i.e., excluding the 117 patients who died from non-CV causes, so that only living patients were included in the referent category), the total observation time for death from CV causes was 6,525 patient-years, a CV mortality rate of 1.8 per 100 person-years (95% CI 1.5–2.1). Of the 120 deaths from CV causes, 72 during 2,879 person-years occurred among glucocorticoid-exposed patients, a rate of 2.5 per 100 person-years (95% CI 1.9–3.1). Among patients not exposed to glucocorticoids, there were 48 CV deaths in 3,646 person-years, or 1.3 per 100 person-years (95% CI 0.9–1.7). The incidence rate ratio for CV-related deaths associated with glucocorticoid exposure was 1.89 (95% CI 1.30–2.79).

Deaths from all causes and from CV causes were also quantitatively associated with glucocorticoid exposure. In the case of the daily glucocorticoids, the unadjusted hazard ratio (HR) for all-cause mortality was 1.07 per milligram of daily prednisone (95% CI 1.05–1.08). For deaths from CV causes, the HR was 1.08 per milligram of daily prednisone (95% CI 1.05–1.12). Death was also associated with the cumulative glucocorticoid dose. For all-cause mortality, the HR was 1.04 per each 10 gm of cumulative exposure to prednisone or equivalent (95% CI 1.03–1.05). The HR for CV mortality per 10 gm of prednisone was nearly identical, at 1.04 (95% CI 1.03–1.06). The intensity of glucocorticoid exposure was also significantly associated with mortality. For all-cause mortality, the HR was 1.12 per gm/year of glucocorticoid exposure (95% CI 1.10–1.14), while for CV mortality the HR was 1.16 (1.12–1.19).

We tested the above associations after adjusting for a series of covariates that could potentially confound the association between glucocorticoid exposure and death. The resulting HRs are shown in Figure 1. The association between deaths from all causes and deaths from CV causes and exposure to glucocorticoids remained significant after adjustment for all of the above covariates, suggesting that the risk estimates are not confounded by the covariates. This was the case for the daily prednisone dose, the cumulative exposure, and the exposure expressed in grams per year.

Figure 1.

Hazard ratios (HRs) and 95% confidence intervals (95% CIs) for deaths from all causes (left) and deaths from cardiovascular (CV) causes (right) associated with the daily dose, the cumulative dose, and the grams per year dose of glucocorticoids in patients with rheumatoid arthritis (RA). HRs from 4 different Cox proportional hazards models are shown for each dose category. Model 1 included only the glucocorticoid dose. Model 2 included the glucocorticoid dose plus demographic and socioeconomic variables. Model 3 included all variables in model 2 plus the clinical and laboratory manifestations of RA. Model 4 included all variables in model 3 plus the CV risk factors. See Patients and Methods for a detailed description of the variables. For the daily dose, the HRs are per milligram. For the cumulative dose, the HRs are per decile of the cumulative exposure distribution. The reference group consisted of patients who did not receive glucocorticoids. Horizontal solid line and broken lines represent gridlines.

We also tested models that included both the daily and the cumulative glucocorticoid doses during the same model. These revealed that both measures of glucocorticoid exposure were associated with death from all causes independently of each other, remaining significant through all 4 adjustment models. In the case of death from CV causes, both glucocorticoid exposure measures were likewise significantly associated independently of each other throughout the 4 adjustment models. The only exception was for the cumulative glucocorticoid exposure, which lost significance in model 3, only to regain a significant association in model 4.

To delineate a threshold exposure level associated with an increased risk of death, we categorized the daily and cumulative glucocorticoid doses in various ways. For the daily glucocorticoid dose, empirical categories were consistent with the commonly prescribed tablet formulations of prednisone.

The categorization that provided the lowest daily dose threshold included the following 5 dose strata: no daily glucocorticoids (referent), <5 mg, 5–7.5 mg, 8–15 mg, and >15 mg of glucocorticoids per day. The unadjusted HRs for these categories are shown in Figure 2. Each higher stratum displayed a progressively stronger association with mortality rates, suggesting a biologic gradient effect and reaching significance at 8–15 mg/day. Deaths from all causes and deaths from CV causes displayed similar associations with the daily prednisone strata.

Figure 2.

Estimates of the threshold of the daily dose of glucocorticoids associated with death from all causes (left) and deaths from CV causes (right) in RA. HRs from 4 different Cox proportional hazards models are shown for each dose category. See Figure 1 and Patients and Methods for a detailed description of the variables. The reference group consisted of patients who did not receive glucocorticoids. Horizontal solid line and broken lines represent gridlines. See Figure 1 for definitions.

For the cumulative glucocorticoid exposure, we tested progressively smaller categories from quartiles to deciles, entering each as a categorical variable in the Cox models, aiming to identify the lowest cumulative exposure associated with death. The lowest stratum of exposure associated with increased mortality rates was the fourth exposure quartile, corresponding to ≥40 gm. Exposures above this threshold were associated with an increase in all-cause and CV mortality rates. The association of these cumulative glucocorticoid exposure strata with death from all causes and death from CV causes is shown in Figure 3. For the number of gram-years, the threshold appeared to be at ≥7 gm/year (Figure 4).

Figure 3.

Estimates of the threshold of the cumulative dose of glucocorticoids associated with death from all causes (left) and deaths from CV causes (right) in RA. HRs from 4 different Cox proportional hazards models are shown for each dose category. See Figure 1 and Patients and Methods for a detailed description of the variables. The reference group consisted of patients who did not receive glucocorticoids. Horizontal solid line and broken lines represent gridlines. See Figure 1 for definitions.

Figure 4.

Estimates of the threshold of the grams per year dose of glucocorticoids associated with death from all causes (left) and deaths from CV causes (right) in RA. HRs from 4 different Cox proportional hazards models are shown for each dose category. See Figure 1 and Patients and Methods for a detailed description of the variables. The reference group consisted of patients who did not receive glucocorticoids. Horizontal solid line and broken lines represent gridlines. See Figure 1 for definitions.

In addition to the above multivariable models that considered confounders individually, we calculated a glucocorticoid propensity score in order to control for potential confounding by indication. The purpose of the propensity score was to quantify the characteristics that may have led physicians to prescribe glucocorticoids to the patients in this noninterventional study, characteristics that could influence a patient's risk of death and, thus, confound the association between death and glucocorticoid exposure that we had aimed to estimate. When used as a covariate, the propensity score adjusts for differences in the measured characteristics between patients who are and those who are not exposed to glucocorticoids.

We generated the propensity score from a logistic regression of the use/no use of glucocorticoids at the time of each study visit, with the model 4 variables as predictors. The model had a C statistic of 0.65, and at a cutoff probability of 0.52, the propensity score correctly classified 61% of the visits with respect to glucocorticoid use and was normally distributed. We tested the performance of the propensity score in reducing differences in clinical manifestations between patients who were and those who were not taking glucocorticoids. When included as a covariate in comparisons of the clinical manifestations between patients who were and those who were not taking glucocorticoids, the propensity score resulted in a loss of significance of observed differences, suggesting that the score adequately represented differences between the two groups.

Table 2 shows propensity-adjusted P values for comparisons of patients with various glucocorticoid exposure levels. When we added the propensity score to the Cox mortality models that tested the association between glucocorticoid exposure and death, the associations remained significant, suggesting that they were not due to confounding by the indication to receive glucocorticoids.

Table 2. Daily and cumulative glucocorticoid dose thresholds for death from all causes and death from CV causes, adjusted for the propensity to receive glucocorticoids*
Glucocorticoid exposureAll-cause mortalityCV mortality
No. of deathsNo. of person-yearsHR (95% CI), adjusted for glucocorticoid propensityNo. of deathsNo. of person-yearsHR (95% CI), adjusted for glucocorticoid propensity
UnadjustedAdjustedUnadjustedAdjusted
  1. Hazard ratios (HRs) and 95% confidence intervals (95% CIs) were determined using Cox proportional hazards models. The glucocorticoid propensity score (probability of receiving glucocorticoids) was predicted from a logistic regression model using receipt or nonreceipt of glucocorticoids at the study visit as the dependent variable and age at onset of rheumatoid arthritis (RA), RA duration, sex, race, socioeconomic status, counts of tender, swollen, and deformed joints, presence of subcutaneous nodules, presence of rheumatoid factor, the erythrocyte sedimentation rate, RA severity, presence of diabetes mellitus, hypercholesterolemia, and hypertension, cigarette smoking, and body mass index as independent variables.
  2. aThis variable (daily dose <5 mg) diverged from the proportional hazards assumption in the model for cardiovascular (CV) mortality (21). Pooled with the 5–7-mg stratum, it satisfied the assumption, with an adjusted HR of 1.39 (95% CI 0.80–2.42) for the pooled variable. All other variables in the models shown met the proportional hazards assumption.
Daily dose        
None953,9121.0 (referent)1.0 (referent)483,6461.0 (referent)1.0 (referent)
<5 mg226471.38 (0.87–2.20)1.19 (0.74–1.90)165852.07 (1.17–3.65)a1.82 (1.03–3.22)
5–7 mg691,8761.51 (1.11–2.06)1.21 (0.88–1.66)281,6471.28 (0.80–2.05)1.07 (0.67–1.72)
8–15 mg426762.58 (1.79–3.71)1.78 (1.22–2.60)245753.17 (1.94–5.17)2.27 (1.36–3.79)
≥15 mg9913.98 (2.01–7.90)2.83 (1.41–5.66)4724.08 (1.47–11.32)3.21 (1.14–8.97)
Cumulative dose        
None612,5351.0 (referent)1.0 (referent)312,3721.0 (referent)1.0 (referent)
<9 gm231,4980.63 (0.39–1.03)0.59 (0.36–0.95)141,4300.95 (0.47–1.95)0.70 (0.37–1.31)
9–39.9 gm501,4951.38 (0.95–2.01)1.12 (0.76–1.64)231,3230.93 (0.45–1.87)1.08 (0.63–1.88)
≥40 gm1031,6752.46 (1.78–3.39)1.74 (1.25–2.44)521,3982.32 (1.36–3.96)2.05 (1.29–3.27)
Dose/time        
None612,5351.0 (referent)1.0 (referent)312,3721.0 (referent)1.0 (referent)
<1.98 gm/year201,5920.51 (0.30–0.85)0.48 (0.29–0.80)131,5290.62 (0.32–1.20)0.60 (0.31–1.16)
1.98–5.08 gm/year481,6151.23 (0.84–1.80)0.99 (0.67–1.45)231,4051.26 (0.73–2.16)1.04 (0.60–1.81)
≥5.08 gm/year1081,4602.93 (2.14–4.03)2.11 (1.51–2.94)531,2173.12 (1.99–4.88)2.35 (1.48–3.75)

DISCUSSION

We examined the association between exposure to glucocorticoids and mortality rates in a well-characterized RA patient cohort that we have monitored for nearly a decade. Both all-cause mortality and CV mortality were associated with the use of glucocorticoids. This was the case whether glucocorticoids were considered qualitatively as a yes/no exposure or quantitatively, seeking a dose-response pattern. The daily dose of glucocorticoids, the cumulative dose, and the intensity of exposure, as estimated by the dose divided over time, were all associated with death, and the role of each of these exposure measures was independent of the other. Moreover, glucocorticoids were associated with death independently of a comprehensive battery of potential confounders, including RA disease activity and severity as well as CV risk factors, all of which were ascertained by physicians who directly evaluated the patients.

Our findings extend those of previous studies on several fronts ([6, 7, 23-26]). Although earlier studies reported a dose-dependent association between the use of glucocorticoids and the presence of CV disease ([9, 10]), we are not aware of previous estimates of the daily and cumulative glucocorticoid dose thresholds associated with death. Sihvonen and colleagues ([8]), without providing exact dosing data, found that the use of low-dose glucocorticoids for more than 10 years was associated with higher mortality rates than were shorter exposures. Caplan and colleagues ([27]) also noted an association between increasing duration of glucocorticoid treatment and death. Other investigators have used prescription records to examine the question of dose. Wei and colleagues ([9]) reported that prednisone doses in excess of 7.5 mg raised the risk of CV disease and deaths from all causes. Davis et al ([10]) reported that a daily glucocorticoid dose of 7.5 mg or higher was associated with increased CV risk in RF-positive, but not RF-negative, RA patients ([10]). Those authors did not report the CV risk associated with other dose strata.

We used narrower dose strata than were used in previous studies, and we tested each of them as categorical variables as compared to the patients who had not taken glucocorticoids. This allowed us to pinpoint mortality thresholds more precisely and to explore a broader range of dose-response categories. We found that in the lowest daily exposure levels, death rates were not significantly different from those among patients who had not been exposed to glucocorticoids, but that in progressively higher dose strata, the risk of death increased. This gradient, or dose-response, pattern was readily apparent for both the daily and cumulative exposures. Despite differences in our target population and study design, our findings are consistent with those of the previous reports ([8-10]).

Our results suggest that daily prednisone doses of 7.5 mg or less in RA patients may be “safe” from the standpoint of mortality risk, but that higher doses are associated with a progressively higher risk of dying. Similarly, a cumulative prednisone dose <40 gm was associated with a mortality risk that was no different from that in the unexposed patients. Above this threshold, however, mortality rates increased. It is instructive to consider how long it would take to reach the 40-gm cumulative dose threshold with various daily prednisone dosing regimens. With 5 mg/day, a commonly used dosage, it would take 21.9 years. With 10 mg/day, the time is halved to 10.9 years, and with 3 mg, the interval is 36.5 years. In RA, which is a chronic disease, such durations of exposure are commonly reached.

Our findings further support the notion that clinicians should prescribe the lowest dose of glucocorticoids for the shortest time possible ([28]). The finding that low cumulative doses (<9 gm) were associated with a significant reduction in death from all causes may seem unexpected; however, the potential benefits of a small amount of glucocorticoids in RA have been discussed ([29]), and it is possible that a low dose of steroids may help some patients.

One important aspect by which our study advances earlier research is that we systematically examined all patients physically, performing careful, standardized, yearly assessments of the clinical and laboratory manifestations of RA and CV risk factors. The examining physicians graded RA severity according to a validated quantitative scale ([13]). We used these RA features to control for the possibility that the glucocorticoid associations we observed would be confounded by the RA disease manifestations. As was the case with previous studies, the possibility that we had influenced the patients' glucocorticoid treatment assignments did not occur, since these agents were prescribed solely at the discretion of each patient's personal physician. This makes our estimates, as well as earlier ones, prone to confounding by indication: Physicians most likely prescribed glucocorticoids to the patients with the most severe manifestations ([30]).

Since the disease manifestations themselves may be associated with atherosclerosis, CV disease, and increased mortality rates in RA, it was important to account for them in any evaluation of the toxic effects of glucocorticoids. We used these individually ascertained clinical variables as time-varying covariates in Cox regression models, to minimize the potential that our estimates were confounded by disease severity. We also used these data to score the patients' propensity to be receiving glucocorticoids at each visit. These propensity scores, which were used as covariates in multivariable models, compensate for the effect of nonrandom treatment assignment ([31]). Upon adjustment for the measured covariates and the propensity score, there was minimal change in the risk of death associated with glucocorticoids. The mortality risk associated with glucocorticoids remained significantly increased, suggesting that any confounding of our estimates by disease severity is not major.

One potential bias with respect to the cumulative glucocorticoid dose is that early death limits the total exposure. Although we are unable to estimate the influence of this bias, its likely effect would be to attenuate the association between high cumulative exposures and death. The observed dose-dependent association between the cumulative dose of glucocorticoids and death suggests that the effect of this potential bias is not a major one. An additional limitation of this analysis is that we did not account for the use of low-dose aspirin.

Conclusions from our study must nevertheless be tempered by the limitations inherent in a noninterventional study design that is not able to account for confounders that we did not measure. Despite the above statistical efforts to adjust for nonrandom treatment assignment, readers should keep in mind that our findings do not take the place of a well-designed randomized trial. However, because such a trial of the effect of glucocorticoids on mortality rates is unlikely to be performed in the near future, the present data may provide the best estimate for some time to come. A larger sample size would likely have produced narrower risk strata and therefore different thresholds. Thus, care should be taken in extrapolating our data to an individual patient's risk, and it is likely always best to aim for the lowest possible glucocorticoid dosage.

We conclude that exposure to glucocorticoids is associated with an increased risk of death in RA. However, the increased mortality rates occur only among patients taking daily prednisone doses that exceed 8 mg or cumulative doses of ≥40 gm. Clinicians may wish to avoid exceeding these thresholds in order to avoid exposing their patients to an increased risk of death. Further research is needed estimate the extent to which glucocorticoid doses below these thresholds are beneficial to patients' survival.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Del Rincón had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Del Rincón, Escalante.

Acquisition of data. Del Rincón, Restrepo, Erikson.

Analysis and interpretation of data. Del Rincón, Battafarano, Escalante.

Ancillary