Multiple Vertebral Fractures After Denosumab Discontinuation: FREEDOM and FREEDOM Extension Trials Additional Post Hoc Analyses

ABSTRACT

It is uncertain whether the risk of vertebral fracture (VF) and multiple vertebral fractures (MVFs; ≥2 VFs) after denosumab (DMAb) discontinuation is related to treatment duration. A prior analysis of Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months (FREEDOM) and FREEDOM Extension trials did not find a relationship with DMAb duration and may have underreported MVF incidence because it included women who did not have radiographs. In this post hoc exploratory analysis, the crude incidence and annualized rates of VF and MVF were determined in patients with ≥7 months' follow‐up and ≥1 spine radiograph after discontinuing placebo or DMAb. A multivariate analysis was performed to identify predictors of MVF. Clinical characteristics of patients with ≥4 VFs were explored. This analysis included women who discontinued after placebo (n = 327) or DMAb either from FREEDOM or FREEDOM Extension (n = 425). The DMAb discontinuation group was subsequently dichotomized by treatment duration: short‐term (≤3 years; n = 262) and long‐term (>3 years; n = 213) treatment. For any VF, exposure‐adjusted annualized rates per 100 patient‐years (95% confidence interval [CI]) were 9.4 (95% CI, 6.4–13.4) for placebo, 6.7 (95% CI, 4.2–10.1) for short‐term DMAb, and 10.7 (95% CI, 7.4–15) for long‐term DMAb. Annualized rates for MVF were 3.6 (95% CI, 1.9–6.3), 2.9 (95% CI, 1.4–5.4), and 7.5 (95% CI, 4.8–11.1), respectively. Annualized rates for ≥4 VFs were 0.59 (95% CI, 0.1–2.1), 0.57 (95% CI, 0.1–2.1), and 3.34 (95% CI, 1.7–6.0), respectively. In a multivariate regression model, DMAb duration was significantly associated with MVF risk (odds ratio 3.0; 95% CI, 1.4–6.5). Of 15 patients with ≥4 VFs, 13 had DMAb exposure (mean ± standard deviation [SD], 4.9 ± 2.2 years). The risk of MVF after DMAb discontinuation increases with increased duration of DMAb treatment. Patients transitioning off DMAb after 3 years may warrant more frequent administration of zoledronic acid or another bisphosphonate to maintain bone turnover and bone mineral density (BMD) and prevent MVF. © 2022 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).

Introduction

Discontinuation of denosumab (DMAb) results in a rapid increase in the biochemical indices of bone turnover, which exceed baseline values within 9 months and remain elevated for up to 30 months after the last dose of DMAb.^{( 1} , ² , ^{3 )} In association, bone mineral density (BMD) is also lost rapidly and returns to baseline or below baseline levels within 1–2 years.^{( 1} , ² , ³ , ^{4 )} The risks of a vertebral fracture (VF) and multiple vertebral fractures (MVFs; defined as ≥2 VFs) also increase rapidly over 6–18 months after the last dose of DMAb.^{( 3} , ⁵ , ⁶ , ⁷ , ^{8 )} A prior post hoc study of the Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months (FREEDOM) and FREEDOM Extension trials estimated the risks of new or worsening VF after DMAb discontinuation.^{( 5 )} In that study, the annualized rate of MVF was higher after discontinuing DMAb (4.2%) compared with discontinuing placebo (Pbo; 3.2%). Furthermore, of the women who had any VF, the proportion who developed MVF was higher in those who discontinued DMAb versus Pbo (61% versus 39%).

Several lines of evidence suggest that a longer DMAb duration is associated with a higher rebound bone turnover and higher bone loss after discontinuation.^{( 4} , ⁷ , ⁹ , ¹⁰ , ¹¹ , ^{12 )} Longer DMAb treatment duration is associated with greater bone loss even when bisphosphonate treatment is administered at the time of discontinuation.^{( 9} , ¹¹ , ^{13 )} Furthermore, a longer duration of DMAb treatment has been associated with a higher number^{( 14} , ^{15 )} and earlier development^{( 16 )} of MVFs. After a comprehensive review of these findings, a recent European position statement recommended a more intensive strategy after DMAb discontinuation in patients who have had DMAb treatment beyond 2.5 years.^{( 11 )}

The prior post hoc study of the FREEDOM and FREEDOM Extension trials did not find a relationship between DMAb treatment duration and risk of VF or MVF after DMAb discontinuation.^{( 5 )} A major limitation of this analysis was that follow‐up radiographs were not available for the majority of patients who discontinued DMAb in the extension trial. Most VFs do not produce clearly defined symptoms at the time of the event.^{( 17 )} In fact, the vast majority of the radiographs (>97%) taken after discontinuation of medication were obtained at routine scheduled appointments, and not for intercurrent back pain or any other clinical VF symptomatology. In the prior analysis, those women who did not undergo radiographic examination were assumed to have no VF. Therefore, the calculated fracture rates based on the results from this larger group of women without available radiographs are likely to underestimate the risk of VF and MVF.

The current post hoc analyses aimed to evaluate the rates of VF and MVF in women for whom radiographs were available and to evaluate more comprehensively the impact of DMAb treatment duration as a predictor of VF or MVF after treatment discontinuation. Another objective of these analyses was to determine if there were any clear clinical characteristics, most importantly treatment duration, which could predict the risk of ≥4 VFs. This latter category of fractures was also predefined in the prior analysis; however, treatment duration was not analyzed in that subgroup of women.

Patients and Methods

Study design and patients

The study designs of the FREEDOM trial (NCT00089791) and its open‐label extension (NCT00523341) have been published previously.^{( 18} , ^{19 )} The original post hoc analysis of risk of VF and MVF after DMAb discontinuation included women enrolled in FREEDOM and FREEDOM Extension trials who had received at least two doses of DMAb or Pbo and subsequently discontinued treatment but presented for at least one appointment 7 or more months after the last administered dose.^{( 5 )} The off‐treatment follow‐up period began 7 months after the last dose. In the current analyses, we assessed women who met the criteria for the original post hoc analysis and who also had at least one spine radiograph available after medication discontinuation. The protocol from the FREEDOM and FREEDOM Extension studies specified that radiographs be obtained at years 1, 2, 3, 5, 6, 8, and 10. All participants considered in the analyses here had at least one radiograph after treatment discontinuation and results were compared with the last on‐treatment radiograph; patients who discontinued DMAb were dichotomized into short‐term (≤3 years) or long‐term (>3 years) groups based on the duration of DMAb treatment in the FREEDOM and FREEDOM Extension trials. A longer mean interval between radiographs was observed for the >3 years DMAb group as no radiographs were obtained at years 4, 7, and 9. In the DMAb Discontinuation group, the median (Q1, Q3) intervals between the last on‐treatment radiograph and first off‐treatment radiograph were 12.1 (11.6, 13.4) months for the ≤3 years DMAb cohort and 24.0 (19.2, 24.9) months for the >3 years DMAb cohort; for the Pbo Discontinuation group, the median (Q1, Q3) interval between the last on‐treatment and first off‐treatment radiographs was 12.0 (11.7, 12.4) months.

We defined prevalent and incident fractures as per the original FREEDOM and FREEDOM Extension trials and the prior analysis by Cummings and colleagues^{( 5} , ^{18 )} and Bone and colleagues^{( 19 )} Prevalent fracture of a vertebral body was defined as at least a semiquantitative grade 1 fracture at baseline. Prevalent fractures were reported for the cohorts considered here at both the FREEDOM baseline and at the time of treatment discontinuation. An incident off‐treatment VF was defined as a semiquantitative grade 1 increase compared to that vertebra at the time of treatment discontinuation for any vertebra between T₄ through L₄ (inclusive). This included vertebrae originally classified as normal (grade 0; new fractures) and vertebrae with at least a grade higher fracture than baseline (worsening fractures). We defined MVFs after treatment discontinuation as at least two new and/or worsening VFs during the follow‐up period. We defined incident ≥4 VFs as at least four new and/or worsening VFs during the follow‐up period.

Statistical analyses

All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC, USA). Crude incidence and annualized exposure‐adjusted rates of any VF and MVF were determined in women who had available spine radiographs. A multivariate logistic regression analysis was performed to identify predictors of MVF. Potential predictors of off‐treatment VF and MVF were included in multiple logistic regression models with stepwise selection procedure as reported.^{( 5 )} The treatment group was included at all steps. Covariates considered included factors from FREEDOM baseline, factors associated with treatment during FREEDOM/FREEDOM Extension, and factors assessed after treatment discontinuation. The covariates from FREEDOM baseline included were prevalent VF; prior nonvertebral fracture; age; body mass index; and T‐scores at the lumbar spine (LS), total hip (TH), and femoral neck (FN). Covariates associated with treatment during FREEDOM/FREEDOM Extension included treatment duration; bisphosphonate use before or during FREEDOM; annualized on‐treatment TH and FN BMD percentage change; incident VF; and incident nonvertebral fracture. The covariates assessed after treatment discontinuation included age at treatment discontinuation, prevalent VF at treatment discontinuation (which included baseline VF and incident VF during FREEDOM and FREEDOM Extension), off‐treatment follow‐up duration, use of postdiscontinuation osteoporosis therapy if initiated prior to the off‐treatment VF, and annualized off‐treatment TH and FN BMD percentage change between last on‐treatment and last off‐treatment BMD measurement, if available.

Clinical characteristics of patients with ≥4 VFs were evaluated, including baseline age and prevalent VF at entry into FREEDOM, treatment duration, prevalent VF at treatment discontinuation, glucocorticoid use while on treatment, and glucocorticoid use after treatment discontinuation.

Results

Patient disposition and characteristics

Figure 1 shows the derivation of patients included in these analyses; the total number of patients who discontinued from Pbo was 470, and the total number who discontinued DMAb was 327 during the FREEDOM trial and 678 from the FREEDOM Extension. Of those patients, 143 from the Pbo group did not have an X‐ray after treatment discontinuation and were excluded from these analyses; 122 patients who discontinued DMAb during the FREEDOM trial and 408 who discontinued during the FREEDOM Extension study were excluded as they did not have an X‐ray. The remaining subgroups who did have an X‐ray included 327 patients who discontinued Pbo (hereafter referred to as the “Pbo Discontinuation” group), and a total of 475 patients who discontinued DMAb (referred to hereafter as the “DMAb Discontinuation” group). We subsequently dichotomized the 475 women from the DMAb Discontinuation group into those who discontinued after short‐term DMAb use (≤3 years treatment; n = 262) and those who discontinued after long‐term DMAb use (>3 years treatment; n = 213).

Fig. 1.

Patient disposition. Flow diagram shows disposition of patients from FREEDOM or its Extension through treatment discontinuation and follow‐up. The off‐treatment follow‐up period begins from the last dose plus 7 months through the end of study. DMAb = denosumab; EXT = extension; FU = follow‐up; Pbo = placebo.

In Table 1, the term “FREEDOM baseline characteristics” refers to characteristics of the participants at the time of enrollment into the original FREEDOM trial. In the lower half of Table 1, the characteristics of study participants at the time of DMAb or Pbo discontinuation are presented. Table 1 compares the characteristics of Pbo Discontinuation participants who had X‐rays available after treatment discontinuation with those who were excluded because X‐rays were not available, both regarding baseline characteristics at entry into FREEDOM and characteristics at the time of treatment discontinuation. Table 1 also compares the characteristics of DMAb Discontinuation participants who had radiographs available after treatment discontinuation with those who were excluded because X‐rays were not available, both regarding baseline characteristics at entry into FREEDOM and characteristics at the time of treatment discontinuation.

Table 1.

Characteristics of Patients at FREEDOM Baseline and at the Time of Discontinuation

	No off‐treatment X‐ray a		With off‐treatment X‐ray a
Characteristic	Pbo Discontinuation group (n = 143)	DMAb Discontinuation group (n = 530)	Pbo Discontinuation group (n = 327)	DMAb Discontinuation group (n = 475)
Characteristics of patients at FREEDOM baseline
Age at baseline (years)	73.7 ± 5.3	73.9 ± 5.2*	72.7 ± 5.1	71.9 ± 4.8
Baseline vertebral fracture (%) b	25.2	24.2	26.3	24.4
Baseline lumbar spine T‐score	−2.9 ± 0.8	−2.8 ± 0.7	−2.8 ± 0.7	−2.8 ± 0.7
Baseline total hip T‐score	−2.0 ± 0.8	−2.0 ± 0.8*	−2.1 ± 0.9	−1.9 ± 0.8
Characteristics of patients at discontinuation
Overall treatment duration (months), median (IQR)	19.8 (13.4, 25.4)	49.8 (26.3, 73.4)	19.8 (18.7, 30.7)	31.2 (19.4, 70.0)
Age at discontinuation (years)	75.5 ± 5.4	79.4 ± 5.6*	74.6 ± 5.1	76.4 ± 5.3
Prevalent vertebral fracture at discontinuation (%) c	26.6	28.9	35.8 †	27.8
Prevalence of ≥4 vertebral fractures at discontinuation (%)	0*	0.9*	0.6	0.1
Lumbar spine T‐score at discontinuation	−2.9 ± 0.8	−2.2 ± 0.9	−2.8 ± 0.8	−2.3 ± 0.9
Total hip T‐score at discontinuation	−2.1± 0.8	−1.7 ± 0.9	−2.2 ± 0.9	−1.6 ± 0.9
Median off‐treatment follow‐up time (months), median (IQR) d	1.7 (0.5, 3.6)*	1.8 (0.6, 4.6)*	11.5 (5.2, 17.6)	11.4 (4.5, 23.0)

Unless otherwise noted, data are presented as mean ± SD for continuous variables and as % for categorical variables. For T‐score at the time of discontinuation, DXA data were taken from the last available DXA before off‐treatment start date.

DMAb = denosumab; IQR = interquartile range; Pbo = placebo; SD = standard deviation.

^a Patients with ≥7 months of follow‐up after the last dose of Pbo or DMAb.

^b Vertebral fracture confirmed by radiograph at FREEDOM baseline.

^c Vertebral fracture confirmed by radiograph any time before treatment discontinuation.

^dBeginning at 7 months after the last DMAb/Pbo injection.

^* p value <0.05 (^† p = 0.0518) between corresponding group (Pbo Discontinued without X‐ray versus with X‐ray; DMAb Discontinued without X‐ray versus with X‐ray) based on two‐sample t tests (for continuous variables) or Pearson chi‐square tests (for categorical variables) comparing patients with off‐treatment X‐ray to patients without off‐treatment X‐ray within each treatment group. Fisher's exact test was used for estimating the prevalence of ≥4 vertebral fractures during the on‐treatment period due to the low number of events.

Comparison of Pbo discontinuation groups with and without off‐treatment X‐rays and DMAb discontinuation groups with and without off‐treatment X‐rays

When comparing baseline characteristics of participants who discontinued Pbo and did not have an X‐ray after treatment discontinuation with those who did have an X‐ray and were thus included in the analyses here, age, VF prevalence, and LS and TH T‐scores were not different between the respective Pbo populations. The DMAb Discontinuation participants without an X‐ray were on average older (73.9 versus 71.9 years old) and had a lower TH T‐score (−2.0 versus −1.9) than the DMAb Discontinuation participants who did have an X‐ray.

For comparisons of characteristics at the time of treatment discontinuation (lower half of Table 1), there were no significant differences between Pbo Discontinuation patients who had an X‐ray versus Pbo Discontinuation patients who were excluded because they did not have an X‐ray, although VF prevalence was higher (35.8% in the Pbo Discontinuation group with an X‐ray versus 26.6% in the Pbo Discontinuation group without X‐ray; p = 0.0518). Very few patients in any of the cohorts had ≥4 VFs at the time of treatment discontinuation; however, the prevalence was slightly higher in the Pbo Discontinuation cohort with an X‐ray (0.6%) versus those without (0%). The prevalence of ≥4 VFs was also slightly lower in the DMAb Discontinuation cohort with an X‐ray (0.1%) versus those without an X‐ray (0.9%). Furthermore, DMAb Discontinuation patients who did not have an X‐ray were older (79.4 versus 76.4 years old) and had a slightly lower TH T‐score (−1.7 versus −1.6) than the DMAb Discontinuation group with X‐rays included here.

Comparison of Pbo and DMAb discontinuation cohorts with off‐treatment X‐rays

When comparing the two Discontinuation cohorts with available X‐rays that were included in these analyses, the Pbo Discontinuation group was slightly older and had a slightly lower TH BMD than the DMAb Discontinuation Group at FREEDOM baseline (p < 0.05 for both).

When comparing the two discontinuation cohorts analyzed here at the time of treatment discontinuation, there were significant differences, as expected, based on treatment assignment: the DMAb Discontinuation group was slightly older, and had longer mean treatment duration, lower mean VF prevalence, and higher mean BMD levels at both the spine and hip compared with the Pbo Discontinuation group (p < 0.05 for all except prevalence of ≥4 VFs at discontinuation).

The median off‐treatment follow‐up period (beginning 7 months after the last dose) was similar between the DMAb Discontinuation and Pbo Discontinuation groups, with X‐rays 11.5 months after DMAb discontinuation and 11.4 months after Pbo discontinuation. These treatment periods were much longer than those seen in the discontinuation cohorts without X‐rays, where follow‐up periods were <2 months in both discontinuation cohorts.

Comparison of discontinuation cohorts included in the analysis versus the full FREEDOM population

The baseline FREEDOM characteristics of the two cohorts included in the analyses here were also compared with the baseline FREEDOM characteristics of the rest of the FREEDOM population (N = 7006; Table S1). In the remainder of the FREEDOM population, the mean age was 72 years, VF prevalence was 23.5%, and mean LS and TH T‐scores were −2.8 and −1.9, respectively. There were no significant differences between the remaining FREEDOM population and the Pbo Discontinuation and DMAb Discontinuation groups included in these analyses, except that the Pbo Discontinuation group had a slightly lower baseline TH T‐score (−2.1; p < 0.05).

Comparison of short‐term and long‐term DMAb discontinuation groups

Table 2 presents patient characteristics for the DMAb Discontinuation group after dichotomizing by treatment duration. At FREEDOM baseline, the short‐term DMAb group was 1 year older than the long‐term DMAb group (72.4 versus 71.3 years, respectively; p < 0.05) but there were no other significant differences. Consistent with the pre‐planned dichotomous grouping, the median DMAb treatment duration was 19.8 months for the ≤3‐year (short‐term) treatment cohort and 67.7 months for the >3‐year (long‐term) treatment cohort.

Table 2.

Characteristics of DMAb Discontinuation Participants with X‐rays at FREEDOM Baseline and at the Time of Treatment Discontinuation Dichotomized by DMAb Duration ≤3 years versus >3 years

	DMAb Discontinuation group a
Characteristic	DMAb duration ≤3 years (n = 262)	DMAb duration >3 years (n = 213)
Characteristics of patients at FREEDOM baseline
Age at baseline (years)	72.4 ± 5.0*	71.3 ± 4.4
Baseline vertebral fracture (%) b	24.0	24.9
Baseline lumbar spine T‐score	−2.8 ± 0.7	−2.8 ± 0.6
Baseline total hip T‐score	−1.9 ± 0.8	−1.8 ± 0.9
Characteristics of patients at discontinuation
Overall treatment duration (months), median (IQR)	19.8 (18.2, 25.6)*	67.7 (55.1, 85.4)
Age at discontinuation (years)	74.9 ± 5.0*	78.3 ± 4.9
Prevalent vertebral fracture at discontinuation (%) c	27.9	27.7
Prevalence of ≥4 vertebral fractures at discontinuation (%)	0	0.5
Lumbar spine T‐score at discontinuation	−2.6 ± 0.8*	−1.9 ± 0.9
Total hip T‐score at discontinuation	−1.7 ± 0.9*	−1.4 ± 0.9
Median off‐treatment follow‐up time (months), median (IQR) d	14.0 (5.0, 22.2)*	9.3 (2.2, 32.1)

Unless otherwise noted, data are presented as mean ± SD for continuous variables and as % for categorical variables.

DXA data were taken from the last available DXA before off‐treatment period start date.

DMAb = denosumab; IQR = interquartile range; SD = standard deviation.

^a Patients who discontinued DMAb in FREEDOM or Extension with ≥7 months of follow‐up after the last dose of DMAb and ≥1 spine X‐ray.

^b Vertebral fracture confirmed by radiograph at FREEDOM baseline.

^c Vertebral fracture confirmed by radiograph any time before treatment discontinuation.

^d Beginning at 7 months after the last DMAb injection.

^* p value < 0.05 based on two‐sample t tests (for continuous variables) or Pearson chi‐square tests (for categorical variables) comparing the ≤3‐year group to the >3‐year group. Fisher's exact test was used for estimating the prevalence of ≥4 vertebral fractures during the on‐treatment period due to the low number of events.

At the time of discontinuation, long‐term DMAb discontinuers were older and had higher LS and TH BMD levels (p < 0.05), consistent with the longer course of DMAb treatment prior to discontinuation. Also, at treatment discontinuation, VF prevalence was similar in the short‐term and long‐term DMAb Discontinuation groups and prevalence of ≥4 VFs was similarly very low. Median follow‐up periods after discontinuation were slightly longer in the short‐term DMAb Discontinuation group.

Crude incidence and exposure‐adjusted annualized rates of VF, MVF, and ≥4 VFs in the Pbo discontinuation and DMAb discontinuation groups

VF rates were higher than those reported in the prior analysis.^{( 5 )} For any VF, the crude incidence was 9.5% after Pbo and 11.8% after DMAb discontinuation (Table 3). For MVF, the crude incidence was 3.7% after Pbo and 7.2% after DMAb discontinuation. In both the original and the current analysis, the proportion of total VF that was MVF was higher after stopping DMAb (61%) compared with the proportion observed after stopping Pbo (about 39%).^{( 5 )} The incidence of women with ≥4 VFs was 0.6% after Pbo discontinuation and 2.7% after DMAb discontinuation.

Table 3.

Crude Incidence and Exposure‐Adjusted Annualized Rates of VF, MVF, and ≥4 VFs in the Pbo and DMAb Discontinuation Groups

	Pbo Discontinuation group (N = 327)		DMAb Discontinuation group (N = 475)
Parameter	Crude incidence n (%)	Exposure‐adjusted rate per 100 patient‐years (95% CI)	Crude incidence n (%)	Exposure‐adjusted rate per 100 patient‐years (95% CI)
Vertebral fracture	31 (9.5)	9.4 (6.4–13.4)	56 (11.8)	8.7 (6.5–11.2)
Multiple vertebral fracture	12 (3.7)	3.6 (1.9–6.3)	34 (7.2)	5.1 (3.6–7.2)
≥4 Vertebral fractures	2 (0.6)	0.6 (0.1–2.1)	13 (2.7)	1.9 (1.0–3.3)

CI = confidence interval; DMAb = denosumab; MVF = multiple vertebral fracture; n = number of patients with fracture; N = total number of patients; Pbo = placebo; VF = vertebral fracture.

Annualized rates for any VF (per 100 patient‐years; 95% confidence interval [CI]) were 9.4 (95% CI, 6.4–13.4) after Pbo and 8.7 (95% CI, 6.5–11.2) after DMAb discontinuation. Annualized rates for MVF were 3.6 (95% CI, 1.9–6.3) after Pbo and 5.1 (95% CI, 3.6–7.2) after DMAb discontinuation. Annualized rates for ≥4 VFs were 0.6 (95% CI, 0.1–2.1) after Pbo and 1.9 (95% CI, 1.0–3.3) after DMAb discontinuation.

Crude incidence and exposure‐adjusted annualized rates of VF, MVF, and ≥4 VFs with DMAb discontinuation group categorized by duration ≤3 years versus >3 years and by crossover versus continuous groups

In the DMAb Discontinuation group, the incidence of any VF in women who received short‐term DMAb was similar to those who received Pbo (8.4% and 9.5%, respectively); however, women who had received long‐term DMAb had a higher incidence of VF (16.0%). For MVF, in the DMAb Discontinuation group, the incidence again appeared similar between short‐term DMAb and Pbo Discontinuation groups (3.8% and 3.7%, respectively) but was greater with long‐term DMAb treatment (11.3%). This was also true for risk of ≥4 VFs; 11 of 15 of these occurred in women after long‐term DMAb exposure (incidence 5.2% after long‐term treatment).

As shown in Fig. 2A , estimated annualized rates (per 100 person‐years; 95% CI) for any VF were higher after more than 3 years of DMAb (10.7; 95% CI, 7.4–15.0) than for short‐term treatment (6.7; 95% CI, 4.2–10.1). Short‐term treatment did not increase MVF rates more than Pbo (3.0 [95% CI, 1.4–5.4] for short‐term treatment versus 3.6 [95% CI, 1.9–6.3] for Pbo), but treatment beyond 3 years was associated with an approximate doubling of the rate (7.5; 95% CI, 4.8–11.1) (Fig. 2B ). The rate of occurrence of ≥4 VFs was 0.59 (95% CI, 0.1–2.1) after Pbo, 0.57 (95% CI, 0.1–2.1) after short‐term DMAb, and 3.34 (95% CI, 1.7–6.0) after long‐term DMAb exposure (Fig. 2C ).

Fig. 2.

Exposure‐adjusted annualized rates of (A) any vertebral fracture, (B) multiple vertebral fractures, and (C) ≥4 vertebral fractures in Pbo and DMAb Discontinuation groups categorized by duration ≤3 years versus >3 years. Bar graphs show exposure‐adjusted annualized rates (per 100 patient‐years) of any vertebral fracture, multiple vertebral fractures, and ≥4 vertebral fractures in Pbo Discontinuation (n = 327), DMAb Discontinuation ≤3 years (n = 262), and DMAb Discontinuation >3 years (n = 213) groups. Number in parentheses represents the number of patients with fracture. DMAb = denosumab; Pbo = placebo.

Predictors of off‐treatment MVF based on multivariate logistic regression model

Variables that were previously found to be predictors of MVF remained predictive of MVF in this cohort of patients with available radiographs and the effect sizes were also similar (including prevalent VF at treatment discontinuation, hip bone loss rate, and off‐treatment duration) (Table 4). Although treatment duration was not a predictor in the original analysis, it was found to be a predictor of MVF in this sample of participants (odds ratio: 1.2/year of DMAb treatment; 95% CI, 1.0–1.4). Among the 475 patients who discontinued DMAb, DMAb treatment duration (>3 years versus ≤3 years) was found to be predictive of MVF after discontinuation (odds ratio 3.0; 95% CI, 1.4–6.5) based on logistic regression model without interaction. To further evaluate the robustness of this analysis, we performed a sensitivity analysis that included age at the beginning of the off‐treatment period as an additional covariate and found that none of the estimates changed more than 5% (Table S2).

Table 4.

Significant Predictors of Off‐Treatment Multiple Vertebral Fractures Based on a Multivariate Logistic Regression Model

Significant covariates	Patients with off‐treatment X‐ray (n = 736) a Odds ratio (95% CI)
Prior VF (Yes versus No) b	3.14 (1.54–6.39)
Off‐treatment duration (per year)	1.35 (1.08–1.68)
On‐treatment duration (per year)	1.20 (1.03–1.40)
Off‐treatment annualized total hip BMD loss (per 1%) c	1.19 (1.08–1.32)

BMD = bone mineral density; CI = confidence interval; VF = vertebral fracture.

^a 736 patients included 301 patients who discontinued placebo and 435 patients who discontinued denosumab and had available off‐treatment annualized total hip BMD change assessments.

^b “Prior VF” includes any prevalent VF confirmed by radiograph any time before treatment discontinuation.

^c “Off‐treatment annualized total hip BMD loss” was defined as annualized percent change in total hip BMD after treatment discontinuation, ie, between the last on‐treatment and off‐treatment BMD assessments.

Characteristics of patients with ≥4 VFs

Of the 15 women with ≥4 VFs, 2 (0.6%) had been on Pbo and 13 (2.7%) had been on DMAb (difference p = 0.005 by exact binomial test) (Table 3). Mean ages were very similar (DMAb Discontinuation Group = 70.6 years; Pbo Discontinuation Group = 70 years). The mean DMAb treatment duration in patients who had ≥4 VFs was 4.9 ± 2.2 years. Only two of the 13 DMAb patients with ≥4 VFs had been on short‐term DMAb; both received four doses in the 2 years of DMAb treatment. At FREEDOM baseline, one of the Pbo patients had a baseline prevalent VF and six of the 13 DMAb‐treated patients had a prevalent VF. At treatment discontinuation, prevalent VF was still seen in one of two Pbo‐treated patients and in seven of 13 DMAb‐treated patients. Both Pbo‐treated patients with ≥4 VFs were treated with glucocorticoids (during the trial and after treatment discontinuation). One of the 13 patients with ≥4 VFs from the DMAb group had glucocorticoid exposure during the trial, but no glucocorticoid treatment after DMAb discontinuation. Glucocorticoid treatment included intermittent oral doses (no continuous use for >6 months), one intraarticular administration, and one intravenous administration.

Discussion

The risk of VF and MVF after DMAb discontinuation increases with the duration of treatment, and patients taking DMAb for more than 3 years have a high risk of VF (10.7 per 100 patient‐years) and MVF (7.5 per 100 patient‐years) after discontinuing DMAb. In women who had long‐term DMAb exposure (>3 years), the annualized rate of MVF after discontinuation was more than twofold higher compared with those with shorter (≤3 years) DMAb exposure. Consistent with this finding, the presentation of ≥4 VFs was almost solely a function of long‐term treatment duration and one‐half of the women who sustained ≥4 VFs had no prior VF before or during DMAb treatment. The only Pbo Discontinuers who had ≥4 VFs were glucocorticoid‐treated. This analysis differs from the previously published analysis from the FREEDOM and FREEDOM Extension trials with respect to the influence of treatment duration. In the prior analysis, a large number of women who did not have X‐rays were considered to be free of fracture; this analysis included only those patients who had radiographs. Moreover, here, our focus was on treatment duration, and long‐term use was defined as more than 3 years of treatment.

In addition to the new finding that treatment duration is a predictor of MVF after DMAb discontinuation, other variables that were previously found to be predictors remained predictive in the cohort analyzed here (including prevalent VF at treatment discontinuation, hip bone loss rate, and off‐treatment duration) and effect sizes were similar. Of the other variables, such as younger age, lower body mass index (BMI), and renal insufficiency, found to contribute to MVF risk in some^{( 12 )} but not all studies,^{( 8 )} age and BMI were not identified as predictors in this analysis and renal insufficiency was not included in the model.

Our finding that discontinuation‐associated VF and MVF risk increase with DMAb treatment duration is consistent with prior observations. The number and/or incidence of MVF appears to be higher and earlier after a longer course of therapy.^{( 15} , ^{16 )} Bone turnover marker levels and bone loss rates may be greater upon discontinuation of longer duration of DMAb treatment.^{( 4} , ⁹ , ¹⁵ , ^{20 )} In patients treated with DMAb and aromatase inhibitor therapy, after discontinuation, risks of VF and MVF were higher in patients who stopped DMAb and VF occurred at an earlier time in patients who had longer treatment duration. A retrospective chart review of VF risk after treatment discontinuation by Burckhardt and colleagues^{( 8 )} did not show an association between DMAb treatment duration and VF risk; however, that study included only clinically diagnosed VF. Furthermore, almost 20% of the patients included in that study were treated with DMAb for the breast cancer indication, not osteoporosis. Also, the mean treatment duration in our study was longer than in this retrospective study (44.6 months versus 35 months). Thus, it is possible that Burckhardt and colleagues^{( 8 )} did not have a large enough number of patients with longer treatment duration to see the effect of treatment duration on VF risk after discontinuation. Most importantly, almost 70% of the women received bisphosphonates at DMAb discontinuation and bisphosphonates were very effective at minimizing VF risk after DMAb discontinuation in that population. In contrast, in our study, only 25.5% of patients who discontinued DMAb received a subsequent osteoporosis treatment, of whom 86.0% received bisphosphonates. This is consistent with other studies suggesting that the risks of VF and MVF are low in patients who receive zoledronic acid after DMAb discontinuation.^{( 11} , ¹² , ¹³ , ^{21 )} Still the effectiveness of zoledronic acid to maintain BMD after DMAb discontinuation appears related to DMAb treatment duration.^{( 9} , ¹¹ , ^{13 )} Our findings are consistent with the recent recommendations by the European Calcified Tissue Society (ECTS) that stress the importance of DMAb treatment duration as a predictor of MVF risk after discontinuation and recommend a more intensive strategy upon withdrawal after long‐term use.^{( 11 )}

A potential mechanism for the increased bone resorption after DMAb discontinuation could be through an increased number of osteoclast precursor cells as seen in the blood samples of a small number of women with postmenopausal osteoporosis obtained 6 months after discontinuing DMAb treatment compared with historical controls.^{( 22 )} Upon DMAb discontinuation, with loss of receptor activator of nuclear factor κB ligand (RANKL) suppression, these accumulated precursor cells can develop into osteoclasts, leading to a profound increase in bone turnover.^{( 7 )} In addition, recently, osteomorphs were described as an end‐stage cell derived from osteoclasts after bone resorption. The development of these distinct non‐resorbing cells is under RANKL control. With RANKL suppression, osteomorphs accumulate and with release of the suppression, the cells can be recycled back into active osteoclasts.^{( 23 )} As both the accumulation of osteoclast precursors and postresorption osteomorphs might be related to DMAb duration, they both could play a role in contributing to excessive bone remodeling, bone loss, and MVF after discontinuation of more prolonged DMAb administration.

This analysis has important limitations. The FREEDOM trial was not designed to answer discontinuation questions and did not provide a long‐term Pbo group to compare with the long‐term DMAb exposure group. The subset of patients included in this study is defined by treatment discontinuation, which is a postrandomization factor, and such analysis can introduce potential bias.^{( 24} , ^{25 )} Using propensity score based methods to balance the treatment duration groups was not feasible because patients who did not continue into the Extension study were all categorized into the ≤3 year duration group, which could distort the propensity model. However, we accounted for confounders by utilizing a step‐wise model selection approach consistent with the original analysis presented by Cummings and colleagues^{( 5 )} and performed a sensitivity analysis to confirm the robustness of this model.

Moreover, there was an additional decrease in the sample size compared with the original post hoc analysis and an even greater sample size drop in the analyses of the effect of treatment duration. We excluded patients who did not have any off‐treatment X‐rays from the analysis; thus, we cannot verify if those excluded patients had a similar probability of developing VF or MVF during the follow‐up period. The cohorts analyzed here were not identical to the patients who discontinued and were excluded because of lack of X‐rays. However, the differences seen with regard to both baseline FREEDOM characteristics and characteristics at the time of treatment discontinuation would, if anything, bias towards seeing a lower MVF risk in longer term DMAb Discontinuers versus Pbo Discontinuers. For example, the discontinuers in the Pbo group who had X‐rays were older and had lower BMD than the discontinuers from the Pbo group who did not have X‐rays. Similarly, the higher (marginally significant) prevalent VF rate at the time of treatment discontinuation in the Pbo Discontinuation group with X‐rays would predict a higher MVF risk in the Pbo group. The differences in characteristics between DMAb Discontinuers who had X‐rays versus those who were excluded (ie, younger age, higher BMD) would also predict, if anything, a lower MVF rate in those who were included. Differences in characteristics between short‐ and long‐term treated DMAb Discontinuers, such as higher spine and hip BMD in the long‐term DMAb Discontinuers would also bias towards seeing a lower MVF rate in those patients. In fact, we saw the opposite. The long‐term DMAb Discontinuers had a longer period of time between the last on‐treatment X‐ray and first off‐treatment X‐ray compared with the short‐term DMAb group. However, based on the average known rate of VF while on long‐term DMAb, the higher MVF rate in the long‐term DMAb Discontinuation group could not have been related to a longer period of time between X‐rays while DMAb treatment was ongoing.

Although the ECTS has recommended more intensive bisphosphonate treatment after discontinuation of patients who have had exposure beyond 2.5 years,^{( 11 )} this guidance was not available when we prepared the analysis plan for our study. We chose 3 years as the cutoff point because it was the original length of the FREEDOM trial and is the treatment duration for most of the osteoporosis treatment studies in the literature. However, with our 3‐year cutoff, the majority of women (55.4%) actually received five denosumab injections, not six, because the cutoff was based on the amount of time on DMAb and not on the number of DMAb doses. No patients classified as “≤ 3 years use” could have had more than six DMAb treatments. Therefore, the 2.5‐year versus 3‐year cutoffs are probably very similar. The analyses here did not have the design or power to conclude that less than 3 years of DMAb is safe regarding the risk of MVF after discontinuation.

The strengths of this study include a more accurate assessment of the true rate of VF and MVF based on patients who had the required diagnostic testing for this outcome. Furthermore, the patients who had radiographs and were evaluated here had longer follow‐up periods, again likely contributing to greater accuracy in assessing rates of the specified outcomes. We have evaluated the influence of treatment duration by dichotomizing duration at ≤3 years versus >3 years as well as by multivariate step‐wise regression analysis. We have performed sensitivity analyses to support the accuracy of our conclusions. We have identified treatment duration as an important factor associated with the presentation of ≥4 VFs.

Patients who are prescribed denosumab should be carefully followed up to be certain that they receive doses as scheduled or are contacted and evaluated within 1 month of the date that a dose was due. In patients who have achieved treatment goals or who have developed medical or personal reasons warranting DMAb discontinuation, treatment with an alternative osteoporosis therapy is required. Imaging to identify prevalent VF should be performed when discontinuation is being considered since this is a major risk predictor for MVF after discontinuation; however, prevalent VF at treatment discontinuation was only seen in about one‐half of the women who sustained ≥4 VFs after discontinuation. Our findings add to the growing body of literature confirming that more intensive assessment and treatment might be required in those who discontinue DMAb after more than 3 years of treatment. More frequent administration of bisphosphonates, for example, zoledronic acid every 6 months, might be needed to maintain bone turnover and BMD, and ultimately minimize the risk of MVF after discontinuation of long‐term DMAb.^{( 9} , ^{21 )} Zoledronic acid is the most well studied among the agents prescribed subsequent to DMAb discontinuation,^{( 9} , ¹⁰ , ¹³ , ^{21 )} but alendronate should be considered in clinical settings where treatment with zoledronic acid is not available or possible.^{( 26 )} When possible, referral to a specialist familiar with serial measurements of bone turnover to guide therapeutic decisions might help optimize management of these patients.

Author Contributions

Felicia Cosman: Conceptualization; methodology; writing – original draft; writing – review and editing. Shuang Huang: Conceptualization; formal analysis; methodology; writing – original draft; writing – review and editing. Michele McDermott: Conceptualization; methodology; writing – original draft; writing – review and editing. Steven R. Cummings: Conceptualization; methodology; writing – original draft; writing – review and editing.

Conflicts of Interest

FC has received grants/research support from Amgen and Radius Health; received consulting fees from Amgen, Radius Health, Enterabio, and Biocon; and has served on speakers' bureaus for Amgen and Radius Health. SH and MMcD are employees and stockholders of Amgen. SRC has received grants/research support and consulting fees from Amgen.

Supporting information

Table S1. FREEDOM baseline characteristics of Placebo and Denosumab Discontinuation patients with off‐treatment X‐ray who were included in this analysis, and of the remaining FREEDOM patient population.

Table S2. Sensitivity analysis with age at start of off‐treatment period as an additional covariate.

Data Availability Statement

Qualified researchers may request data from Amgen clinical studies. Complete details are available at the following: https://wwwext.amgen.com/science/clinical‐trials/clinical‐data‐transparency‐practices/clinical‐trial‐data‐sharing‐request/