Oral glucocorticoid therapy and all-cause and cause-specific mortality in patients with rheumatoid arthritis: a retrospective cohort study

Rheumatoid arthritis (RA) is a chronic inflammatory disease which affects between 0.5 and 1 % of the adult population worldwide [1‐3]. Oral glucocorticoid (GC) therapy was introduced as a treatment for patients with RA nearly 60 years ago [4] and is still used widely. Around one third of patients with RA are current users, and two thirds of patients have ever used GCs [5]. GCs improve symptoms of active RA through reducing joint pain, swelling and stiffness [6]. However, there are some concerns about their potential side effects including cardiovascular (CV) events, diabetes, infection, fracture, and cataracts [7‐11], many of which are associated with an increased risk of mortality.

Previous studies have investigated the association between GC therapy and mortality, mostly focusing on all-cause mortality, though some have investigated CV mortality [5, 12‐16]. Findings from these studies are not consistent. GCs have been associated with an increased risk of all-cause mortality in some [12‐15, 17, 18] but not all studies [16, 19, 20], with similar inconsistency for CV mortality [12, 14, 16]. Very few studies have examined other cause-specific mortality. In studies that consider dose, some have suggested no association with doses <5 mg prednisolone equivalent [12, 13], reflecting either a lack of significant side effects at this dose or perhaps a favourable balance between side effects and positive anti-inflammatory properties.

There are important methodological issues when considering GC exposure and mortality, including confounding by indication—whereby GC therapy is given to patients with high disease severity and high disease severity is itself associated with increased mortality. However, studies have rarely considered a form of protopathic bias we will call ‘perimortal bias’, where illness in the latter stages of life influences GC exposure, and which consequently might affect the observed relationship between GC use and death. For example, if a patient were to develop cancer, GC therapy may be prescribed to treat the malignancy and a resultant association would be observed between GCs and (cancer-specific) mortality. The aim of this study was to investigate all-cause and cause-specific mortality in association with various models of oral GC exposure in patients with RA, and to explore and control for the possible existence of perimortal bias.

There were 37,983 patients identified with a diagnosis of RA, of whom 21,355 were eligible for ONS linkage. After applying the exclusion criteria, the cohort reduced to 16,762 patients (Fig. 2). Table 1 summarizes the patient characteristics of the whole cohort, ever GC users and non-users. 70 % of patients were female, with similar proportions in the GC and non-GC groups. Mean age at baseline was 60.2 years [standard deviation (SD): 14.6].

There were 8367 (50 %) patients who received at least one prescription for oral GCs. These patients were on average 4 years older, more likely to have received GCs in the 1 year prior to RA (44 vs. 6 %, respectively) and had higher DMARD use during follow-up compared to non-users. The mean baseline Charlson comorbidity index was slightly higher in ever users compared with never users (1.39 vs. 1.25) (Table 1).

During active GC prescriptions, the mean current daily dose (PED) was 7.5 mg (SD 6.9 mg). The mean cumulative dose (PED) among the 8367 patients who received GC therapy was 5.3 g (SD 6.0 g).

During a total of 111,099 person years, 2996 patients died (median follow-up of 6.1 years per person), giving an all-cause mortality rate of 27.0 deaths per 1000 person-years (pyrs) (95 % CI 26.0–28.0) (Table 2). In those never exposed to GCs the mortality rate was 15.5 deaths per 1000 pyrs, compared to 44.0 deaths per 1000 pyrs in those ever exposed to GCs.

Overall the most common cause of death was cardiovascular disease, followed by neoplasms and respiratory diseases. The underlying causes of death in the “other causes” category were mostly musculoskeletal (28.0 %). Figure 3 shows the cumulative incidence curves from Fine-Gray models [27] for the four categories of cause-specific mortality. Ever users had higher mortality rates in each cause-specific category compared to never users. For each category the mortality rate for ever users was higher than never users from the start of follow-up, and the mortality rate was consistent through follow-up for both exposed and unexposed groups.

Cardiovascular mortality rates were 15.8 deaths per 1000 pyrs in ever users compared to 6.4 deaths per 1000 pyrs in never users. Within this chapter, ischemic heart disease had the highest mortality rate for both ever and never users. Neoplasms had the second highest mortality rate for ever GC users. Conversely, the second highest mortality rate for never users was other causes of death. Respiratory diseases had the lowest mortality rate in both ever and never GC users (Table 2).

Table 3 shows the associations between oral GC use and risk of all-cause and cause-specific mortality, estimated through six alternative Cox models, adjusted for age, gender, smoking status, SES, prior cumulative dose of GC, baseline Charlson comorbidity index, time-varying NSAID use and time-varying DMARD use.

BMI was the only potential confounder with higher than five percent of missing data (Table 1). When it was included in a complete case analysis of model 1 it did not alter the hazard ratio for GC use and was not significantly associated with mortality and so was not included in the fully adjusted models. Smoking and SES had <5 % missing data and were included in the fully adjusted models. All models consistently showed that risk of death was associated with GC use and increased with higher dosages of GCs. There was a nearly twofold greater risk of all-cause mortality in ever users, compared to never users (HR 1.97, 95 % CI 1.81–2.15). For cause-specific mortality, ever users had over a three times higher risk of death from neoplasms compared to never users (HR 3.20, 95 % CI 2.66–3.86). For both all-cause and cause-specific mortality, a similar pattern was seen for current use, though the point estimates were lower. For each 5 mg increase in GC dose there was a 33 % increased risk of all-cause mortality compared to non-users (HR 1.33, 95 % CI 1.30–1.35). Similar increased risks were seen for each of the cause specific mortality categories, with the highest risk seen for neoplasms (HR 1.46, 95 % CI 1.42–1.49).

The categorisation of current daily dose showed that for all-cause mortality, CV mortality and mortality due to respiratory diseases, a dose below 5 mg per day was not associated with an increased risk of death. Furthermore, for neoplasms and ‘other causes’, a dose of below 7.5 mg per day was not associated with an increased risk of death. However, as current daily dose increased above these doses, so too did the risk of death. Comparing between the hazard ratios for cause-specific mortality, the risk of cardiovascular mortality was notably lower than for the other three categories of death, for current GC dose above 7.5 mg.

There was a 6 % increased risk of all-cause mortality for each 1000 mg/day increase in cumulative dose since cohort entry (HR 1.06, 95 % CI 1.05–1.07). Similar increases in risk were seen for each cause-specific mortality category. Categorisation of cumulative dose showed a dose response increased risk of all-cause mortality in each category of cumulative dose, with risk of death increasing with increased categories of cumulative dose. The exception to this was for other causes of death, where there was not an increased risk of death from other causes with cumulative doses up to 3054.9 mg (Table 3). Additional adjustment for mean number of rheumatology outpatient visits per year and mean number of GP visits per year in general increased the risk of all-cause mortality and cause-specific mortality, but did not alter the significance, except for the lowest current dose category (0-4.9 mg) where a significantly reduced risk of mortality due to neoplasms was seen (Online resource Table A1).

This study examined the association between oral GC therapy and mortality rates in a cohort of patients with RA in the UK. Ever GC use and current GC use was associated with an increased risk of all-cause mortality and cause-specific mortality, with a largely consistent dose–response effect. An increase in current dose of 5 mg per day was associated with an increased risk of death, however categorisation showed that taking <5 mg per day at the time of death did not increase the risk of all-cause mortality or cause-specific mortality, and taking <7.5 mg per day at the time of death did not increase this risk of death from neoplasms or other non-CV and non-respiratory causes. In addition, moderate to high doses of GC therapy were associated with a lesser risk of CV deaths compared to neoplasm, respiratory and other causes of death, which might suggest GC therapy has a less harmful effect on CV mortality.

The study showed that perimortal bias partially explained some of the results, especially at higher doses. Perimortal bias is important to consider for a number of reasons. GCs can be used to treat diseases that might develop through the course of follow-up, and where that disease is the leading cause of death. For example, if a patient were to develop a malignancy, they might start GC therapy as part of their cancer treatment which would lead to a positive association between GCs and (cancer-related) mortality. Similarly, end of life care might lead to a switch from disease-modifying anti-rheumatic drug (DMARD) therapy (that requires regular blood monitoring) to GC therapy, again generating an association between GC use and death.

When GC use in the 6 months prior to death was removed, the association between ever GC use and all-cause mortality remained significant, but the risks were reduced. This was mainly influenced by the large reduction in risk of death from neoplasms, where there is a clear possibility of perimortal bias: GCs are prescribed as a treatment for cancer [31]. Initial signals of possible perimortal bias were evident in the magnitude of the association between high-dose GCs and risk of death due to neoplasm (HR 31.3, 95 %CI 23.5–41.9).

The all-cause mortality rate for this study was 27 deaths per 1000 person-years, and the cardiovascular mortality rate was 10 deaths per 1000 person-years. This was higher than a recent cohort study in the UK (Norfolk Arthritis Registry (NOAR)) [32] where rates were 20–21 per 1000 person-years and 7–8 per 1000 person years for all-cause mortality and cardiovascular mortality respectively. This would be expected as NOAR includes patients with early inflammatory arthritis, whereas this study included patients with a higher baseline age and with RA only, and therefore more severe disease.

Our findings are in agreement with some previous studies [5, 12‐15, 17] which have investigated all-cause mortality or CV mortality in association with GC use. Caplan et al. [5] found an increased risk of death with current GC use, with an adjusted odds ratio of 2.2 (95 % CI 1.9–2.7) and an increased risk of death with increasing duration of GC treatment. del Rincon et al. [12] found a GC dose-dependent increase in death from all causes (HR 1.07 per 1 mg/day (95 % CI 1.05–1.08) and CV cause with a similar point estimate. They also showed that there was a dose response association for cumulative dose for all-cause and CV mortality with a threshold of 40 g. Listing et al. [13] showed that GC doses higher than 5.0 mg per day were significantly associated with increased all-cause mortality, independent of disease activity. Treatment with prednisolone higher than 15 mg per day was associated with 3.4-fold (95 % CI 2.01–5.86) increased risk of all-cause mortality compared with non-use. Our findings of probable perimortal bias, however, might suggest that the hazard ratios reported in these previous studies are over-estimates of the true effect.

An important finding of this study was the absence of an association between both all-cause and cause-specific mortality and GC doses lower than 5 mg per day. This may reflect either a low risk of adverse events at this dose, or at least a favourable balance between the harms and the biologically plausible benefits through their anti-inflammatory effects [33]. This finding replicates similar findings from Listing et al. [13] and del Rincon et al. [12], which showed that doses lower than 5 and 8 mg PED, respectively, had no association with mortality risk.

The strengths of this study are firstly its size, with nearly 3000 deaths in 16,762 patients. This meant the study had greater power to detect differences in mortality rates and allowed us to explore cause-specific mortality. We were thus able to see an increased rate of respiratory deaths, accepting the possibility of perimortal bias but also likely driven by a causal increased risk of respiratory infection [9]. Second, the study used linkage to the national mortality register, providing robust and complete information on cause of death for all patients in the study. Third, time-varying covariates for DMARDs, NSAIDs and GCs, were used to allow more accurate estimation compared to time-independent variables for these drugs. Fourth, a range of patterns of GC use were explored including GC use, GC daily dose, and cumulative dose since cohort entry and their categorical variables compared with non GC use in association with risk of death. This approach allowed some consideration of the impact of dose, duration and timing of treatment on mortality risk. For example, the finding that the highest current dose category was associated with very high HRs for neoplasm, respiratory and other causes of death, whilst the highest quartile had notably lower HRs, suggests that high doses may be used at the end of life when cumulative exposure is less of an issue. We also explored possible perimortal bias which has not been considered in previous studies. Moreover, we examined oral GC therapy in association with cause-specific mortality beyond CV mortality which has not been investigated in previous studies.

There were some limitations with the study. The prescription data from the CPRD dataset are reliable in terms of drugs prescribed, but does not cover drugs prescribed in secondary care only, such as biologic DMARDs, or over the counter use of NSAIDs; although it has been shown that biologic DMARDs are not associated with an increased mortality compared to standard DMARDs [34]. In addition there may have been some exposure misclassification because of assumptions in data preparation, missing data, patient adherence, injectable steroids and hospital administered GC, although the latter is likely to be minimal as UK rheumatologists typically make recommendations for oral GC treatment to GPs. Like all observational studies, the impact of confounding and bias needed consideration. A range of possible confounders were adjusted for, including time-varying exposure to DMARDs and NSAIDS, and healthcare utilisation variables as surrogate measures of RA disease severity. Although we didn’t have direct measures of disease severity, previous studies that did adjust for clinician-reported disease severity found a persistent association between GC use and mortality [13]. It is thus likely that there was some residual confounding by disease severity. In terms of possible residual or unmeasured confounders affecting the results, of which disease severity is one, sensitivity analyses showed that these would need to be very large to fully explain the results. So even though, for example, high cumulative disease severity has been shown to be associated with lymphoma [35] this would not fully explain the results seen. Adjusting for the Charlson comorbidity index at baseline was expected to control for key comorbidities that contribute to an increased risk of mortality. The main difference at baseline between GC users and non-GC users was prior GC use, which was much higher in GC users (44 vs. 6 %). It may have been that this group was particularly susceptible to death, and any association between GC use and death may have been exaggerated. However, prior GC use was adjusted for so the results should not be biased.

It is very challenging to understand the true causal relationship between oral GC use and mortality from an observational study due to the complex relationships between the indication for treatment (that changes through time) and the outcome, as well as the granularity of the data from a population necessarily large to support the analysis. Nonetheless, despite this blurring of causality by bias and confounding, some important messages emerge. Doses <5 mg were not associated with an increased risk of death. This absent risk is not explained by confounding by disease severity (where you would expect mortality to be higher in the treated compared to non-treated), or by perimortal bias where you would again expect an increased risk compared to non-use. The lower dose-specific hazard ratios for cardiovascular deaths compared to the hazard ratios seen for the other causes of death raises the interesting hypothesis that GC therapy might have a beneficial effect on cardiovascular mortality; yet a protective effect is impossible to conclude with certainty as there is a statistically significant increased risk for CV mortality with all doses above 5 mg. Disentangling these complex factors is impossible, but the large population observational research raises questions that can feed back into more targeted studies, both basic science and epidemiological.

This study has found that GC use is associated with an increased risk of death in RA, both all-cause and cause-specific mortality, which is partially explained by perimortal bias. Importantly, doses of below 5 mg PED were not associated with an increased risk of death. There is a suggestion that GCs may have a less harmful effect on CV mortality compared to their association with other cause-specific mortality, but targeted research is required to examine this signal further.