Sex and age do not modify the association between glucocorticoids and bone mineral density in patients with rheumatoid arthritis: a cross-sectional study

Background It is unclear whether sex or age modify the association of glucocorticoid (GC) use with reduced bone mineral density (BMD) in patients with rheumatoid arthritis (RA). Methods We studied cross-sectional data of RA patients with current or previous GC treatment in a single center cohort study (Rh-GIOP cohort). Our primary outcome was the minimum T-score (measured by DXA) of either lumbar spine, total femur, or femoral neck. Current GC dose was the main exposure; cumulative GC dose and cumulative duration of GC use were also assessed. Following a predefined statistical analysis plan, linear regression analyses with adjustment for confounders assessed whether the association of GC use with BMD was modified by sex (men versus women) or age (≥ 65 versus < 65 years). Results Four hundred eighty-three patients with RA (mean age 64 ± 12 years, 80% women) were included. 33% were not currently taking GCs, 32% were treated with a dose of 5 mg/d prednisone equivalent and 11% with more than 7.5 mg/d. 23% of patients had osteoporosis by DXA (minimum T-score ≤ -2.5). The slope, i.e., the association between changes in minimum T-scores with 1 mg/d change in current GC dose, was similar in men and women (-0.07 and -0.04, respectively; difference -0.03 [-0.11 to 0.04]; p for interaction = 0.41). Slopes were also similar for elderly and non-elderly patients (-0.03 and -0.04, respectively; difference -0.01 [-0.06 to 0.05]; p for interaction = 0.77). Using cumulative dose and duration of use as exposures did not lead to substantial changes of these results. Conclusions In our sample, the association of GC use with reduced BMD in RA was not modified by sex or age. Supplementary Information The online version contains supplementary material available at 10.1186/s13075-023-03083-x.


Section 2: Introduction
Background Rheumatoid arthritis (RA) is a chronic inflammatory autoimmune disease with potentially negative effects on the quality of life and life expectancy of affected patients. [1] Various disease-related factors are responsible for this, such as pain and reduced physical performance but also associated comorbidities such as osteoporosis (OP). [2] Glucocorticoids (GCs) are widely used to treat RA and have potent anti-inflammatory effects. [3][4][5][6][7][8] According to current guidelines, GC therapy should primarily be used as bridging therapy. [9] This means that GCs should only be used for a short period of time to rapidly reduce disease activity at treatment initiation. Rapid tapering follows to avoid potential GC-associated side effects, including OP. However, some patients experience a flare-up of disease activity upon discontinuation of GCs. Others may achieve remission or low disease activity but remain chronically dependent on low-dose GCs. Although the role of low-dose GCs as long-term therapy for RA is considered controversial [10], real-life data shows that GCs are still often continued for years. [5,[11][12][13] The benefits and risks of long-term therapy with GCs must be assessed accurately: It is known that GC-associated side effects increase in frequency and severity with higher dose and longer duration of therapy. One of the most worrying side effects of GC therapy is OP, as reported by both patients and rheumatologists. [14] RA regularly leads to a localized and generalized loss of bone substance and a reduction in bone quality; thus, OP is considered a complex comorbidity of the disease. [15] [16] The generally increased systemic inflammatory activity in RA and the upregulation of proinflammatory cytokines are associated with reduced bone formation and increased bone resorption. For example, Schett et al. found an increased risk of nontraumatic fractures at even minimal elevations in serum C-reactive protein. [17] GCs are also often cited a cause for the high prevalence of OP in RA patients. Namely, OP associated with GCs is thought to be the most common form of secondary OP. [18] GCs have a direct inhibitory effect on osteoblasts and additionally lead to osteoclast activation (at least initially), which increases bone loss. [18,19] However, it should be taken into consideration that the anti-inflammatory effects of GCs can also have a bone-protective impact. [10] In addition, their analgesic and stiffness-reducing effects may lead to more exercise and thus have a positive impact on OP. [20]

Rationale for this study
To accurately weigh the risk-benefit ratio of GCs and specific anti-osteoporotic therapy in rheumatic diseases, many patient-specific factors need to be taken into account, such as nicotine abstinence and physical training (as protective factors) or inflammation and estrogen deficiency (as risk factors). [10] In the international widely used Fracture Risk Assessment Tool (FRAX) of the University of Sheffield, RA is listed as a risk factor, as is treatment with GCs. [21] However, a major weakness of this tool is that GC therapy is not differentiated -for example, multiple variables should be included such as current and cumulative dose -but GCs are only treated as a dichotomous variable ("yes/no"). There are some ideas on how to correct for GC dosages (also from the developers of FRAX), [22] but this was not directly imbedded in the FRAX online calculator.
The exact impact of GCs on the risk of OP in RA has not yet been adequately studied. GCs have long been viewed solely as risk factors. [23,24] However, the evidence has mostly arisen from observational studies which are prone to "confounding by indication" [25]. Patients with higher disease activity are generally more likely to take higher doses of GCs. This makes it difficult to separate the effects of high disease activity from the effects of GC therapy. While it is known that both factors influence BMD, the question of whether GCs in young and elderly patients, and in men and women, have a similar effect on osteoporosis, remains insufficiently answered. Performing an investigation to see if there are differences in those subpopulations could contribute to a more individualized approach in the therapy of RA in the future ("personalized medicine").

Objectives
In this study, we aim to assess the impact of age (A) and sex (S) on the effects GC (G) exert on the dependent variable bone mineral density (BMD, Y) in patients with RA, i.e., we want to look for an interaction between age and GCs (A×G), and sex and GCs (S×G) while taking the main effects into consideration (A, S, and G, respectively). This will be done by analyzing cross-sectional baseline data from the Rh-GIOP cohort (https://clinicaltrials.gov/ct2/show/NCT02719314

Trial design -Rh-GIOP Cohort
In 2015, the Rh-GIOP cohort study (Glucocorticoid-Induced Osteoporosis in Patients with chronic inflammatory Rheumatic Diseases or Psoriasis -Rh-GIOP) was initiated (registered at clinicaltrials.gov: NCT02719314; positive vote of the local ethics committee of Charité Universitaetsmedizin -Berlin: EA1/367/14]). RA patients at Charité -Universitätsmedizin Berlin who have a history of GC use and who have an indication for osteoporosis diagnostics (according to the German umbrella osteology association [DVO] guidelines) have been included in the study and were systematically investigated. We also included patients in control groups, e.g. patients without GCs and/or without an inflammatory rheumatic disease, and patients with psoriasis (without arthritis). In this study, we include a subpopulation of the Rh-GIOP cohort (see section 5: Trial population).
Various relevant variables are collected at each visit (Box 1). Thus, to our knowledge, Rh-GIOP is the largest prospective cohort study of osteoporosis in patients with rheumatic diseases worldwide in which such a large number of parameters relevant to bone health are specifically recorded over such a long time (and data collection is still ongoing). In a first large cross-sectional study of our extensive database, low GC doses of < 5mg/d prednisolone equivalent appear to be "relatively safe" for bone health (manuscript currently in peer-review).

Sample size
For the present cross-sectional study, we will include all patients with a diagnosis of RA which are included until the database lock (performed on March 16 th , 2022).

Confidence intervals and P values
All P values and confidence intervals will be two sided. We will not apply explicit adjustments for multiplicity, rather we will keep the number of tests at an absolute minimum (formal significance tests only for comparisons indicated in Table 2). Descriptive statistics will be used to summarize the collected relevant demographic variables and disease characteristics enabling an assessment of the balance across the two stratification variables (Sex and Age group). These data will be presented in the primary manuscript as outlined in Table 1. Categorical data will be described using numbers and percentages. Normally distributed continuous data will be described using means and SDs, whereas continuous data that are skewed will be described using medians and interquartile ranges. Statistical inference (i.e., from statistical tests) will only be reported for the analyses described in Table 2.

Analysis populations
The primary analyses will be based on the Intention to Monitor (ITM) population, i.e., based on the Full Analysis Set. Full Analysis Set will be defined as the eligible individuals agreeing to participate in

Section 5: Trial population
A flow diagram will be used to visualize and transparently report the progress through the phases of enrolment, data availability and subsequent data analysis; see Figure 1.

Screening data
In the Rh-GIOP cohort, we include all patients (provided they meet the eligibility criteria written down below) who have a physician referral for a bone density measurement. They are included consecutively if they give informed consent.

Eligibility
Eligible patients will have a physician diagnosis of RA. They need to be currently taking GCs or have a history of GC use. Patients need to have an indication for osteoporosis diagnostics (according to the German umbrella osteology association [DVO]). Breastfeeding and lactating women are excluded, as are patients unable to provide informed consent for any reason. The primary analyses will be based on all patients with an RA diagnosis. We will exclude patients with a potential other cause for a secondary OP, namely patients with a history of multiple myeloma, hyperthyroidism (including grave's disease), and hyperparathyroidism. Also, we will exclude patients with high GC doses that are not usually used for longer periods of time (i.e., >15mg/d prednisone equivalent).

Recruitment
Recruitment was taking place monocentric in the Department of Rheumatology and Clinical Immunology, Charité -Universitätsmedizin Berlin, a tertiary care university hospital. We include both in-and outpatients.

Baseline patient characteristics
We will report baseline patient characteristics in a Table 1, for the whole group of RA patients, and then stratified by sex (male vs. female) and age (elderly (≥65 years) vs. non-elderly (<65 years)).

Outcome definitions
The main outcome measure is the minimum T-score observed (selected from lumbar spine, total femur, or femoral neck, whichever is lowest). There will be one primary analysis investigating the interaction between current GC dose and sex, and between current GC dose and age. 1 GC: Depending on the analysis, either current or cumulative GC dose or cumulative duration of GC use

Missing data
Missing data is for various reasons difficult to avoid. We will apply the analysis framework suggested by White et al (2011) in which missing data related to the ITM approach depend on making plausible assumptions about the missingness of the data and including all participants in subsequent sensitivity analyses. [26] Missing data on the outcomes and covariates will be handled using multiple imputation (5 imputations), as well as the observed dataset. This approach should be valid assuming the data is Missing At Random (MAR).

De-confounding variables
Confounding variables are independent variables other than the exposure variable (GC) that are correlated to the outcome of the study. Unaccounted for confounding variables prevent us from measuring the true impact the GC exposure has on an outcome.
Thus, adjustment methods attempt to correct for the assignment mechanism (exposure to higher GC dose) by finding control units similar to treatment units. Prespecifying possible deconfounders, we will use the following pragmatic definition of what makes a confounding variable (C):

•
The Covariate (C) is an ancestor (cause) of the outcome (Y)

•
The Covariate (C) probably causes the exposure (X; GC group)

•
The Covariate (C) is not a descendant (effect) of the exposure (X) or outcome (Y) (De-) Confounding factors: The challenge with observational data is that treatments are not applied randomly, leading to selection bias and confounding variables. We will adjust for potential confounders listed in box 2 above.

Additional analyses
Sensitivity analyses: Instead of independent variable current GC dose: cumulative GC dose / duration of GC therapy; then interaction term for sex*cumulative dose, age*cumulative dose, sex*duration of GC therapy, and age*duration of GC therapy, respectively.
Instead of dependent variable minimum T-Score: lumbar spine T-Score, and total hip T-score (of both right and left femur).
Furthermore, the primary analyses will also be conducted based on available data ('as observed').

Statistical software
R software or similar will be used for all analyses.   Table 1: Patient Characteristics (for all patients, then separately for women and men, and separately for elderly (≥65 years) vs. nonelderly (<65 years).

Figure 2:
Scatterplot showing the association between current GC dose and minimum T-score for sex subgroups (panel a) and age subgroups (panel b), (y-axis: minimum T-score; x-axis: current GC dose; color coding of dots: sex or age subgroups; separate regression lines for sex or age subgroups with respective 95% confidence intervals) Table 2: Association between current GC dose (linear) and minimum T-score for sex and age subgroups (Results of multiple regression with minimum T-score as the dependent variable, current GC dose and with and without above-mentioned confounders as independent variables [i.e., adjusted and crude, respectively], and an interaction between sex and current GC dose and age and current GC dose; multiple imputation will be applied)