https://orcid.org/0000-0002-8274-6468
Abstract
To perform a network meta-analysis (NMA) on the efficacy of antiosteoporotic interventions in the prevention of vertebral and non-vertebral fractures in adult patients taking glucocorticoids (GCs).
We performed NMAs based on a prospectively developed protocol. A librarian-assisted database search of MEDLINE, EMBASE, Web of Science, Cumulative Index of Nursing and Allied Health Literature (CINAHL), the Cochrane Central Register of Controlled Trials (CENTRAL) and Chinese databases was conducted for randomized controlled trials (RCTs) comparing antiosteoporotic interventions in adult patients taking GCs. Outcomes were vertebral and non-vertebral fracture incidences.
We included 56 RCTs containing 6479 eligible patients in our analysis. We found that alendronate and teriparatide were associated with decreased odds of both vertebral and non-vertebral fractures. Denosumab and risedronate were associated with decreased odds of vertebral fractures, while etidronate, ibandronate and alfacalcidol were associated with decreased odds of non-vertebral fractures. We observed low network heterogeneity as indicated by the I2 statistic, and we did not detect evidence of publication bias. All outcomes were based on a moderate quality of evidence according to GRADE.
Bisphosphonates, teriparatide and denosumab are associated with decreased odds of fracture in patients undergoing GC therapy. Vitamin D metabolites and analogues (e.g. alfacalcidol) may have greater anti-fracture efficacy compared with plain vitamin D.
The International Prospective Register of Systematic Reviews (PROSPERO)—CRD42019127073
Bisphosphonates, teriparatide and denosumab are associated with decreased odds of vertebral and/or non-vertebral fractures in patients undergoing GC therapy.
Vitamin D metabolites and analogues (e.g. alfacalcidol) may have greater anti-fracture efficacy compared with plain vitamin D in patients undergoing GC therapy.
Introduction
Glucocorticoids are a class of steroid medications with excellent immunosuppressive properties and can inhibit a wide variety of inflammatory mediators [1]. They are commonly prescribed to patients with chronic inflammatory or autoimmune diseases, such as SLE and asthma [2, 3]. Because of these applications, the chronic use of glucocorticoids is prevalent in clinical settings. In a US estimate of the prevalence of oral glucocorticoid usage, it was reported that over 2 500 000 patients had taken oral glucocorticoids from 1999 to 2008, with 28% of patients having used oral glucocorticoids for five or more years [4].
However, glucocorticoids can cause a range of serious side effects, including cardiovascular and immune disorders [5]. Glucocorticoids can also facilitate osteoclast differentiation and inhibit osteoblast proliferation, causing increased bone resorption, leading to glucocorticoid-induced osteoporosis (GIOP) [6]. Bone weaknesses can occur with daily glucocorticoid doses as low as 2.5 mg of prednisone and the risk of bone fractures increases significantly during the first 3–6 months of glucocorticoid use [7, 8]. As a result, 30–50% of patients experience osteoporotic fractures after being placed on glucocorticoids [9–11].
Previous systematic reviews and evidence-based guidelines have provided physicians and patients with recommendations on the use of antiosteoporotic treatment options to manage GIOP and GIOP-related fractures. The latest guidelines, published by the ACR in 2017, had recommended vitamin D, calcium supplementation and oral bisphosphonates as first-line treatments for GIOP [12]. However, the ACR did not specify the type of bisphosphonate that physicians should use, nor did it evaluate all available antiosteoporotic treatment options. A previous network meta-analysis (NMA) on preventing GIOP in oral glucocorticoid users identified risedronate, etidronate and teriparatide as having an association with decreased fracture risk and increased BMD [13]. But the study was not able to establish the efficacy of several treatment arms, including denosumab, calcitonin and ibandronate. This was due to the study’s narrow inclusion criteria (such as including double-blinded randomized controlled trials only), which resulted in low sample sizes and statistical power [14].
NMA allows for a unified analysis of randomized controlled trial (RCT) data and offers rankings of multiple treatment regimens based on direct and indirect evidence. However, large sample sizes must be incorporated into the analysis to yield precise results and narrow confidence intervals [15]. With this principle in mind, we conducted a comprehensive systematic review and NMA of RCTs in adult patients undergoing glucocorticoid therapy evaluating the effects of antiosteoporotic interventions on fracture incidences.
Methods
We conducted a systematic review and NMA in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) incorporating NMA of health care interventions [16]. This study is prospectively registered on PROSPERO—CRD42019127073. We followed a prospectively developed protocol [17] to complete this review.
Study identification
We conducted a librarian-assisted search of the following databases on 24 May 2019 with an updated search on 23 October 2019: (i) MEDLINE, (ii) EMBASE, (iii) Web of Science, (iv) Cumulative Index of Nursing and Allied Health Literature (CINAHL), and (v) the Cochrane Central Register of Controlled Trials (CENTRAL).
We systematically searched the following Chinese databases on 24 May 2019 with an updated search on 23 October 2019 using a Chinese search strategy: (i) Wanfang Data, (ii) Wanfang Med Online, (iii) China National Knowledge Infrastructure (CNKI), and (iv) Chongqing VIP Information (CQVIP).
The search strategy used for the database searches can be found in Supplementary Material S13–S14, available at Rheumatology online.
Eligibility criteria
We included all available antiosteoporotic pharmacological interventions, including bisphosphonates, denosumab, calcitonin, raloxifene, teriparatide, calcium and vitamin D or vitamin D analogues. Complementary and alternative medicine, as well as traditional Chinese medicine according to the National Institutes of Health (NIH) criteria (https://nccih.nih.gov/health/integrative-health) were not evaluated with the exception of vitamins and calcium.
We included parallel RCTs that satisfied the following criteria in our analysis: they (i) included adult patients (age ≥18) who were undergoing glucocorticoid therapy at baseline, who planned to start glucocorticoid treatments during the study period, or who had used glucocorticoids for at least 3 months in the year prior to baseline; (ii) compared two antiosteoporotic pharmacological interventions, or one intervention with no treatment/placebo; and (iii) reported vertebral or non-vertebral fracture incidences. We did not place restrictions on the blinding status of the RCT, the route of glucocorticoids or antiosteoporotic interventions, nor the language of the publication.
Study selection
Six authors (J.W.D., E.H., Z.S., E.Z., K.K., A.W.) performed title and abstract screening independently and in duplicate using Rayyan (https://rayyan.qcri.org) based on the eligibility criteria described above. Abstracts deemed relevant by both reviewers were then entered into an in-duplicate full-text screening process. We resolved disagreements by recruiting a third author to reach consensus. Figure 1 shows the PRISMA flowchart [18] of our study selection process.
PRISMA flowchart for the identification and selection of randomised controlled trials
CENTRAL: the Cochrane Central Register of Controlled Trials; CINAHL: Cumulative Index of Nursing and Allied Health Literature; CNKI: China National Knowledge Infrastructure; CQVIP: Chongqing VIP Information; PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses.
Outcome measures
Our outcomes of interest are incidences of vertebral and non-vertebral fractures, defined as per individual study criteria. We performed sensitivity analyses limiting the inclusion criteria to different definitions of vertebral fracture to examine the impact of different fracture definitions on the reported vertebral fracture incidence.
Data extraction
Two reviewers carried out data extraction independently and in duplicate using standardized extraction sheets developed a priori, and conflicts were resolved by recruiting a third author to review the data. We extracted information on the publications’ meta-data, methodology, baseline characteristics and outcome measures at the end of the studies’ follow-up period according to the intention-to-treat principle. We contacted corresponding authors to obtain missing or unpublished data whenever necessary.
Risk of bias
We examined risk of biases in included studies using the Cochrane Collaboration’s tool for assessing risk of bias in randomized trials [19]. We assessed the risk of bias from the following seven domains: (i) random sequence generation, (ii) allocation concealment, (iii) blinding of participants and personnel, (iv) blinding of outcome assessment, (v) incomplete outcome data, (vi) selective reporting, and (vii) other biases. For other biases, we examined study funding sources, authors’ conflicts of interest, study sample sizes and follow-up durations.
We rated the risk of bias in each domain as high, unclear or low. We then designated the average risk of bias rating across the seven domains as the overall risk of bias. We resolved disagreements on the risk of bias assessment by recruiting a third author to reach consensus.
Data synthesis
We conducted all statistical analysis using R 3.5.1 (https://www.r-project.org), and we performed Bayesian NMA using the R package gemtc 0.8–3 [20]. We expressed treatment effects of vertebral and non-vertebral fracture incidences as odds ratios (ORs), and we used the surface under the cumulative ranking curve (SUCRA) score to establish ranking for treatment arms [21]. We used gemtc to estimate pooled ORs comparing active interventions vs no treatment and their corresponding 95% credible intervals (CrIs). CrIs can be considered as analogous to CI and can be interpreted similarly [22]. We assessed the presence of heterogeneity between trials using Cochran’s Q test with a significance level of P < 0.10 as recommended by the Cochrane Handbook for Systematic Reviews of Interventions [23]. Heterogeneity was quantified using the I2 statistic [23, 24]. We interpreted 30% < I2 < 75% as moderate heterogeneity and I2 ≥ 75% as serious heterogeneity, as recommended by the Cochrane Handbook [23].
We drew comparison-adjusted funnel plots to detect small-study effects as a signal of publication bias within our networks [25]. Treatment arms were sorted according to their efficacy by SUCRA values. We quantitatively assessed the symmetry of the funnel plots using Egger’s regression test [26]. We further assessed the quality of our networks using Confidence in Network Meta-Analysis (CINeMA) [27], in accordance with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework [28].
To facilitate network connections, we disregarded adjuvant therapies that are common in all treatment arms within individual studies, and we combined all treatment arms involving vitamin D and vitamin D analogues (i.e. alfacalcidol and calcitriol) into a single treatment arm. We also combined monofluorophosphate with sodium fluoride to form a single fluoride treatment arm. To examine the impact of these simplifications, we performed a sensitivity analysis without merging the vitamin D and fluoride treatment arms. The results of these analyses are reported in Supplementary Material S11, available at Rheumatology online.
We performed meta-regression analyses on each study-level covariate among (i) percentage of females, (ii) percentage of post-menopausal females, (iii) mean age, (iv) cumulative GC dosage, and (v) fracture prevalence at baseline as specified in our protocol [17]. We reported the results of our meta-regressions as beta coefficients in Supplementary Material S10, available at Rheumatology online, and we consider a covariate to be important if the 95% CrI of its coefficient crosses a null value of 0.
Results
Characteristics of included trials
We included 56 RCTs published between 1983 and 2019 and including 6479 patients in our NMA. See Table 1 and Supplementary Material S1–5, available at Rheumatology online for detailed patient and trial characteristics. Forty RCTs compared an antiosteoporotic with a placebo or untreated group, and 16 RCTs were head-to-head comparisons. A majority of the included trials (n = 39, 69.6%) had a low overall risk of bias (as shown in Fig. 2).
Risk of bias graph for included studies
Characteristics of included trials and patients
| Characteristic | Total (n = 56) |
|---|---|
| Number of eligible patients, median (range) | 73 (12–771) |
| Year of publication, median (range) | 2004 (1983–2019) |
| Mean follow-up period, median (range), days | 365 (56–2555) |
| Intervention categories evaluated, no. of trials/no. of analysed patients | |
| Bisphosphonates | 46/3508 |
| Vitamins/calcium | 15/585 |
| Teriparatide | 5/437 |
| Conjugated oestrogen | 1/13 |
| Raloxifene | 1/50 |
| Denosumab | 2/352 |
| Fluorides | 2/35 |
| Calcitonin | 3/62 |
| Thiazide | 1/11 |
| No treatment | 40/1426 |
| Outcomes analysed, no. of trials/no. of analysed patients | |
| Vertebral fractures | 53/6333 |
| Non-Vertebral fractures | 39/4880 |
| Patient characteristics | |
| Female, median (range), % | 68 (0–100) |
| Postmenopausal females, median (range), % | 46 (0–100) |
| GC users at baseline, median (range), % | 100 (0–100) |
| Mean age, median (range), years | 57 (31–72) |
| Mean cumulative GC dosage at baselinea, median (range), mg prednisone | 705 (0–55 100) |
| Characteristic | Total (n = 56) |
|---|---|
| Number of eligible patients, median (range) | 73 (12–771) |
| Year of publication, median (range) | 2004 (1983–2019) |
| Mean follow-up period, median (range), days | 365 (56–2555) |
| Intervention categories evaluated, no. of trials/no. of analysed patients | |
| Bisphosphonates | 46/3508 |
| Vitamins/calcium | 15/585 |
| Teriparatide | 5/437 |
| Conjugated oestrogen | 1/13 |
| Raloxifene | 1/50 |
| Denosumab | 2/352 |
| Fluorides | 2/35 |
| Calcitonin | 3/62 |
| Thiazide | 1/11 |
| No treatment | 40/1426 |
| Outcomes analysed, no. of trials/no. of analysed patients | |
| Vertebral fractures | 53/6333 |
| Non-Vertebral fractures | 39/4880 |
| Patient characteristics | |
| Female, median (range), % | 68 (0–100) |
| Postmenopausal females, median (range), % | 46 (0–100) |
| GC users at baseline, median (range), % | 100 (0–100) |
| Mean age, median (range), years | 57 (31–72) |
| Mean cumulative GC dosage at baselinea, median (range), mg prednisone | 705 (0–55 100) |
Cumulative GC dosages were converted to prednisone equivalent doses using the conversion table in the Oxford Handbook of Critical Care, 3rd edition [29].
Characteristics of included trials and patients
| Characteristic | Total (n = 56) |
|---|---|
| Number of eligible patients, median (range) | 73 (12–771) |
| Year of publication, median (range) | 2004 (1983–2019) |
| Mean follow-up period, median (range), days | 365 (56–2555) |
| Intervention categories evaluated, no. of trials/no. of analysed patients | |
| Bisphosphonates | 46/3508 |
| Vitamins/calcium | 15/585 |
| Teriparatide | 5/437 |
| Conjugated oestrogen | 1/13 |
| Raloxifene | 1/50 |
| Denosumab | 2/352 |
| Fluorides | 2/35 |
| Calcitonin | 3/62 |
| Thiazide | 1/11 |
| No treatment | 40/1426 |
| Outcomes analysed, no. of trials/no. of analysed patients | |
| Vertebral fractures | 53/6333 |
| Non-Vertebral fractures | 39/4880 |
| Patient characteristics | |
| Female, median (range), % | 68 (0–100) |
| Postmenopausal females, median (range), % | 46 (0–100) |
| GC users at baseline, median (range), % | 100 (0–100) |
| Mean age, median (range), years | 57 (31–72) |
| Mean cumulative GC dosage at baselinea, median (range), mg prednisone | 705 (0–55 100) |
| Characteristic | Total (n = 56) |
|---|---|
| Number of eligible patients, median (range) | 73 (12–771) |
| Year of publication, median (range) | 2004 (1983–2019) |
| Mean follow-up period, median (range), days | 365 (56–2555) |
| Intervention categories evaluated, no. of trials/no. of analysed patients | |
| Bisphosphonates | 46/3508 |
| Vitamins/calcium | 15/585 |
| Teriparatide | 5/437 |
| Conjugated oestrogen | 1/13 |
| Raloxifene | 1/50 |
| Denosumab | 2/352 |
| Fluorides | 2/35 |
| Calcitonin | 3/62 |
| Thiazide | 1/11 |
| No treatment | 40/1426 |
| Outcomes analysed, no. of trials/no. of analysed patients | |
| Vertebral fractures | 53/6333 |
| Non-Vertebral fractures | 39/4880 |
| Patient characteristics | |
| Female, median (range), % | 68 (0–100) |
| Postmenopausal females, median (range), % | 46 (0–100) |
| GC users at baseline, median (range), % | 100 (0–100) |
| Mean age, median (range), years | 57 (31–72) |
| Mean cumulative GC dosage at baselinea, median (range), mg prednisone | 705 (0–55 100) |
Cumulative GC dosages were converted to prednisone equivalent doses using the conversion table in the Oxford Handbook of Critical Care, 3rd edition [29].
Vertebral fracture incidences
Figure 3A shows the network diagram for the outcome of vertebral fracture incidence. Figure 4A shows the forest plot for the outcome of vertebral fracture incidence in 52 trials including 6129 eligible patients. One trial, comparing alfacalcidol and vitamin D, was only included in the sensitivity analysis [30]. Eight (15.4%) studies defined vertebral fracture according to the semiquantitative grading scheme developed by Genant et al. [31], two (3.8%) studies used the spinal deformity index (SDI) developed by Minne et al. [32], and three (5.8%) studies only recorded symptomatic vertebral fractures validated by radiographs. A majority of the studies (n = 21, 40.4%) adopted variations of definitions proposed by Riggs et al. [33], which defined vertebral fractures as a 15–25% (or a 4–4.6 mm) decrease in the anterior, central or posterior vertebral height. Nineteen (36.5%) studies did not report their definition for vertebral fractures (see Supplementary Material S5, available at Rheumatology online).
Network diagrams showing comparisons between treatment arms in fracture incidence outcomes
(A) Network diagram for vertebral fracture incidence. (B) Network diagram for non-vertebral fracture incidence. The size of the treatment nodes represents the relative sample size of the treatment arm, and the thickness of the edge connections represents the relative number of direct connections between the connected nodes.
Forest diagrams for estimates of odds ratios in fracture outcomes
Antiosteoporotic interventions were compared with no treatment with heterogeneity measures Cochran’s Q (PQ) and I2 statistics. (A) Forest diagram for vertebral fracture incidence. (B) Forest diagram for non-vertebral fracture incidence. ORs < 1 indicate beneficial treatment effects compared with no treatment. OR: odds ratio.
Alendronate (OR 0.48, 95% CrI: 0.27, 0.95, SUCRA 0.564), denosumab (OR 0.32, 95% CrI: 0.12, 0.86, SUCRA 0.712), risedronate (OR 0.50, 95% CrI: 0.31, 0.84, SUCRA 0.554) and teriparatide (OR 0.14, 95% CrI: 0.058, 0.37, SUCRA 0.884) were all associated with reductions in the odds of vertebral fractures compared with the untreated group which ruled out the possibility of no difference. According to SUCRA ranking, teriparatide ranked first in terms of reductions in the odds of vertebral fractures, followed by raloxifene (OR 0.15, 95% CrI: 0.0074, 2.8, SUCRA 0.754) and thiazide (OR 0.22, 95% CrI: 0.0067, 5.8, SUCRA 0.750). Table 2 gives the full list of SUCRA scores and treatment rankings. We did not detect the presence of heterogeneity in our network (I2 = 0%, PQ = 0.96), and we did not observe asymmetry in the comparison-adjusted funnel plots as validated using Egger’s test (P = 0.73, see Supplementary Material S8.1, available at Rheumatology online). Overall, the outcome of vertebral fracture incidence is based on a moderate quality of evidence due to the presence of indirectness and imprecision (see Supplementary Material S7, available at Rheumatology online). Meta-regression analyses revealed no significant correlations between study-level covariates and the estimated treatment effects, as the CrIs of regression coefficients all crossed the line of no effect (see Supplementary Material S10, available at Rheumatology online).
Treatment ranking based on SUCRA scores
| Intervention arm | Vertebral fracture incidence | Non-vertebral fracture incidence | ||
|---|---|---|---|---|
| SUCRA | Ranking | SUCRA | Ranking | |
| Alendronate | 0.564 | 7 | 0.711 | 2 |
| Calcitonin | 0.429 | 11 | 0.387 | 9 |
| Calcium | — | — | 0.368 | 11 |
| Clodronate | 0.614 | 6 | — | — |
| Conjugated oestrogen | 0.302 | 14 | 0.521 | 7 |
| Denosumab | 0.712 | 4 | 0.249 | 12 |
| Etidronate | 0.452 | 10 | 0.619 | 4 |
| Fluoride | 0.112 | 17 | 0.248 | 13 |
| Ibandronate | 0.352 | 13 | 0.715 | 1 |
| Pamidronate | 0.536 | 9 | 0.615 | 5 |
| Raloxifene | 0.754 | 2 | — | — |
| Risedronate | 0.554 | 8 | 0.378 | 10 |
| Teriparatide | 0.884 | 1 | 0.647 | 3 |
| Thiazide | 0.750 | 3 | — | — |
| No treatment | 0.273 | 15 | 0.487 | 8 |
| Vitamin D | 0.180 | 16 | 0.557 | 6 |
| Vitamin K | 0.625 | 5 | — | — |
| Zoledronic acid | 0.407 | 12 | — | — |
| Intervention arm | Vertebral fracture incidence | Non-vertebral fracture incidence | ||
|---|---|---|---|---|
| SUCRA | Ranking | SUCRA | Ranking | |
| Alendronate | 0.564 | 7 | 0.711 | 2 |
| Calcitonin | 0.429 | 11 | 0.387 | 9 |
| Calcium | — | — | 0.368 | 11 |
| Clodronate | 0.614 | 6 | — | — |
| Conjugated oestrogen | 0.302 | 14 | 0.521 | 7 |
| Denosumab | 0.712 | 4 | 0.249 | 12 |
| Etidronate | 0.452 | 10 | 0.619 | 4 |
| Fluoride | 0.112 | 17 | 0.248 | 13 |
| Ibandronate | 0.352 | 13 | 0.715 | 1 |
| Pamidronate | 0.536 | 9 | 0.615 | 5 |
| Raloxifene | 0.754 | 2 | — | — |
| Risedronate | 0.554 | 8 | 0.378 | 10 |
| Teriparatide | 0.884 | 1 | 0.647 | 3 |
| Thiazide | 0.750 | 3 | — | — |
| No treatment | 0.273 | 15 | 0.487 | 8 |
| Vitamin D | 0.180 | 16 | 0.557 | 6 |
| Vitamin K | 0.625 | 5 | — | — |
| Zoledronic acid | 0.407 | 12 | — | — |
SUCRA: surface under the cumulative ranking curve.
Treatment ranking based on SUCRA scores
| Intervention arm | Vertebral fracture incidence | Non-vertebral fracture incidence | ||
|---|---|---|---|---|
| SUCRA | Ranking | SUCRA | Ranking | |
| Alendronate | 0.564 | 7 | 0.711 | 2 |
| Calcitonin | 0.429 | 11 | 0.387 | 9 |
| Calcium | — | — | 0.368 | 11 |
| Clodronate | 0.614 | 6 | — | — |
| Conjugated oestrogen | 0.302 | 14 | 0.521 | 7 |
| Denosumab | 0.712 | 4 | 0.249 | 12 |
| Etidronate | 0.452 | 10 | 0.619 | 4 |
| Fluoride | 0.112 | 17 | 0.248 | 13 |
| Ibandronate | 0.352 | 13 | 0.715 | 1 |
| Pamidronate | 0.536 | 9 | 0.615 | 5 |
| Raloxifene | 0.754 | 2 | — | — |
| Risedronate | 0.554 | 8 | 0.378 | 10 |
| Teriparatide | 0.884 | 1 | 0.647 | 3 |
| Thiazide | 0.750 | 3 | — | — |
| No treatment | 0.273 | 15 | 0.487 | 8 |
| Vitamin D | 0.180 | 16 | 0.557 | 6 |
| Vitamin K | 0.625 | 5 | — | — |
| Zoledronic acid | 0.407 | 12 | — | — |
| Intervention arm | Vertebral fracture incidence | Non-vertebral fracture incidence | ||
|---|---|---|---|---|
| SUCRA | Ranking | SUCRA | Ranking | |
| Alendronate | 0.564 | 7 | 0.711 | 2 |
| Calcitonin | 0.429 | 11 | 0.387 | 9 |
| Calcium | — | — | 0.368 | 11 |
| Clodronate | 0.614 | 6 | — | — |
| Conjugated oestrogen | 0.302 | 14 | 0.521 | 7 |
| Denosumab | 0.712 | 4 | 0.249 | 12 |
| Etidronate | 0.452 | 10 | 0.619 | 4 |
| Fluoride | 0.112 | 17 | 0.248 | 13 |
| Ibandronate | 0.352 | 13 | 0.715 | 1 |
| Pamidronate | 0.536 | 9 | 0.615 | 5 |
| Raloxifene | 0.754 | 2 | — | — |
| Risedronate | 0.554 | 8 | 0.378 | 10 |
| Teriparatide | 0.884 | 1 | 0.647 | 3 |
| Thiazide | 0.750 | 3 | — | — |
| No treatment | 0.273 | 15 | 0.487 | 8 |
| Vitamin D | 0.180 | 16 | 0.557 | 6 |
| Vitamin K | 0.625 | 5 | — | — |
| Zoledronic acid | 0.407 | 12 | — | — |
SUCRA: surface under the cumulative ranking curve.
The sensitivity analysis without treatment arm merging, which included 53 trials with 6333 patients, did not change the treatment ranking significantly (see Supplementary Material S11.1, available at Rheumatology online). Vitamin D analogues (i.e. alfacalcidol and calcitriol) ranked higher when compared with vitamin D, with alfacalcidol ranking the highest out of the three, but all three arms still ranked lower than the untreated arm. Similarly, sodium fluoride ranked higher than monofluorophosphate, but both arms ranked lower than the untreated arm. There were no differences in the level of heterogeneity in the network (I2 = 0%).
We could not complete the sensitivity analysis with limitations on vertebral fracture definitions due to insufficient network connections for studies using Minne’s SDI, using Genant’s semiquantitative grading scheme, or that only recorded symptomatic vertebral fractures. We were not able to establish ranking for denosumab, vitamin K, pamidronate, zoledronic acid, ibandronate and conjugated oestrogen when we limited the fracture definition to the criteria developed by Riggs et al. We did observe changes in treatment rankings when we limited the definition to that of Riggs et al.; most notably, alendronate ranked first in the sensitivity analysis, whereas teriparatide ranked fourth (see Supplementary Material S12, available at Rheumatology online). These changes indicate that different fracture definitions may have an effect on the reported incidences of vertebral fractures in our included studies. We did not observe differences in the level of heterogeneity in the network (I2 = 0%).
Non-vertebral fracture incidences
Figure 3B shows the network diagram for the outcome of non-vertebral fracture incidence. Figure 4B shows the forest plot for the outcome of non-vertebral fracture incidence in 38 trials including 4676 patients. One trial, comparing alfacalcidol and vitamin D, was only included in the sensitivity analysis [30]. No treatment arms were associated with a reduction in the odds of non-vertebral fractures which ruled out the possibility of no difference. Ibandronate (OR 0.51, 95% CrI: 0.087, 3.2, SUCRA 0.715) ranked first, followed by alendronate (OR 0.63, 95% CrI: 0.29, 1.4, SUCRA 0.711) and teriparatide (OR 0.73, 95% CrI: 0.31, 1.7, SUCRA 0.647) according to SUCRA. Table 2 gives the full list of SUCRA scores and treatment rankings. We did not detect the presence of heterogeneity in our network (I2 = 0%, PQ = 0.90), and we did not observe asymmetry in the comparison-adjusted funnel plots as validated using Egger’s test (P = 0.42, see Supplementary Material S8.2, available at Rheumatology online). Overall, the outcome of vertebral fracture incidence is based on a moderate quality of evidence due to the presence of indirectness and imprecision (see Supplementary Material S7, available at Rheumatology online). Meta-regression analyses revealed no significant correlations between study-level covariates and the estimated treatment effects, as the CrIs of regression coefficients all crossed the line of no effect (see Supplementary Material S10, available at Rheumatology online).
The sensitivity analysis without treatment merging included 38 trials and 4852 patients. One trial, comparing calcitriol and conjugated oestrogen, was excluded due to insufficient network connections [34]. We found that the treatment effect of vitamin D decreased significantly, while alfacalcidol appears to be superior in terms of non-vertebral fracture reduction compared with vitamin D and the untreated arms (see Supplementary Material S11.2, available at Rheumatology online). Sodium fluoride was the only fluoride arm in the network. There were no changes in network heterogeneity (I2 = 0%).
Discussion
The results of our systematic review and NMA show that the use of antiosteoporotic interventions is likely to achieve decreases in the incidence of vertebral and non-vertebral fractures among patients undergoing glucocorticoid therapy. We found that bisphosphonates, including alendronate and risedronate, are effective at decreasing the odds of vertebral fracture incidences, but their efficacy may be lower compared with teriparatide and denosumab. We found that teriparatide may be able to reduce the odds of vertebral fractures by 84%, while denosumab can decrease the odds by 68%. In comparison, alendronate and risedronate can reduce the odds of vertebral fractures by 52% and 50%, respectively. However, these results should be interpreted with caution due to differences in vertebral fracture definitions.
Raloxifene, thiazide, vitamin K, clodronate and pamidronate may also be effective at reducing the odds of vertebral fracture, but the sample sizes included in these treatment arms are very small, and therefore the estimated treatment effects for these arms may not be reliable. In the case of thiazide, for example, the effective sample size was only 11 patients from a study published in 1989 [35]. In a large observational study of diuretic use and risk of osteoporotic fractures involving 55 780 women, it had been shown that thiazide may actually increase the risk of osteoporotic vertebral fracture rather than decrease fracture risk [36]. Therefore, more large scale RCTs are needed to establish the true efficacy of these small sample size treatment arms in patients taking glucocorticoids.
We also found that ibandronate, alendronate, etidronate and teriparatide may be effective at reducing non-vertebral fractures compared with no treatment. Ibandronate, alendronate and etidronate may reduce fracture odds by 49%, 37% and 21%, respectively, while teriparatide can reduce fracture odds by 27%. Pamidronate and conjugated oestrogen did show a decrease in fracture odds as well, but findings in these arms should be interpreted with caution due to low sample sizes. We found that vitamin D may decrease the odds of non-vertebral fractures, but our sensitivity analysis showed that the treatment effect of vitamin D may have been exaggerated by alfacalcidol. In our sensitivity analysis, we found that alfacalcidol is able to reduce the odds of non-vertebral fractures by 21%.
Our findings regarding the efficacy of teriparatide and bisphosphonates at reducing fractures in GC patients supports the conclusion from previous NMAs regarding the treatment of GIOP [13, 37]. In addition, we also identified denosumab as effective at vertebral fracture reduction, and alfacalcidol as effective at non-vertebral fracture reduction. Currently, the latest ACR guidelines on the management of GIOP have recommended the use of vitamin D, calcium and bisphosphonates in adult GC patients for fracture prevention [12]; however, results from our study and previous reviews show that teriparatide and denosumab may serve as potential replacements for bisphosphonates, and thus these treatments warrant further investigations using large scale RCTs [13, 37, 38]. Furthermore, results from our sensitivity analyses suggest that vitamin D analogues (e.g. calcitriol) and active metabolites (e.g. alfacalcidol) may be more efficacious at fracture prevention compared with plain vitamin D (i.e. cholecalciferol). Future guidelines and trials should consider the use of calcitriol or alfacalcidol in the management of GIOP rather than plain vitamin D.
To our knowledge, this is the most comprehensive NMA for examining pharmacological therapies in treating GIOP-related fractures. More RCTs with large sample sizes are needed to determine the efficacy of interventions such as raloxifene. Future RCTs should adopt similar definitions of vertebral fractures to reduce the inconsistencies and heterogeneity in future systematic reviews and meta-analyses.
Data availability statement
All results from our analyses are published in the Supplementary Material, available at Rheumatology online. Aggregated patient data supplied by the included RCTs and extracted by our investigators are available upon reasonable request. We are happy to provide these data to help validate and replicate the findings of our study. Please contact the corresponding or first author for data requests.
Acknowledgement
We would like to acknowledge Mr Wenteng Hou of McMaster University and Mrs Jannusha Panicker of University of Ottawa, for their assistance during the data extraction portion of this study. Their willingness to donate their time has been very much appreciated. J.W.D. made significant contributions to the conception and design of the work, performed data analyses, drafted the work, substantially reviewed the work, and participated in article screening, data extraction and risk of bias analyses. E.H. and Z.S. made significant contributions to the conception of the work, drafted the work, substantially reviewed the work, and participated in article screening, data extraction and risk of bias analyses. E.Z., K.K. and A.W. substantially reviewed the work and participated in article screening, data extraction and risk of bias analyses. S.S. made significant contributions to the methodology of the work, substantially reviewed the work, and performed all database searches. W.C., J.D. and G.G. made significant contributions to the conception of the work, drafted the work, substantially reviewed the work and made revisions to the final work. All authors read and approved the final manuscript.
Funding: No specific funding was received from any bodies in the public, commercial or not-for-profit sectors to carry out the work described in this article.
Disclosure statement: The authors declare no conflicts of interest.
Supplementary data
Supplementary data are available at Rheumatology online
References
Deng J, Huang E, Silver Z et al. Optimal pharmacological management and prevention of glucocorticoid-induced osteoporosis (GIOP): protocol for a systematic review and network meta-analysis. medRxiv
Comments