Romosozumab Cardiovascular Safety Magnitude

What the evidence says

Randomized trials give a split picture that depends on the comparator. Against placebo, the 36-month follow-up of the FRAME trial found balanced adjudicated serious cardiovascular adverse events: 3.6% in the romosozumab-to-denosumab arm and 3.5% in the placebo-to-denosumab arm. Against alendronate, the ARCH trial found positively adjudicated serious cardiovascular adverse events during the first year were numerically more frequent in the romosozumab group. The sources here give no event counts or rates for ARCH, so the size of that imbalance is not established on this page. One review describes the trial pattern as a numerical increase "with no discernable etiology".

Pooled analyses mostly find no measured effect. A 2023 network meta-analysis found active treatments did not increase serious cardiovascular adverse events versus placebo or other comparators. A 2024 network meta-analysis found no differences in major cardiovascular events against placebo and similar profiles for romosozumab, denosumab, teriparatide and alendronate. A 2025 network meta-analysis found no difference in cardiovascular mortality or major cardiovascular events overall, and no excess signal against comparators other than alendronate. A placebo-controlled meta-analysis found no increase in cardiovascular adverse events within 12 months. Two other meta-analyses point the other way: one found a "probable higher risk" from up to three RCTs, and one found more adjudicated cardiovascular serious adverse events and cardiovascular deaths on romosozumab 210 mg than placebo, without a statistically distinguishable difference. The certainty ratings in the network meta-analyses are moderate to low.

Real-world cohorts mostly show no increase. Against denosumab, the adjusted hazard ratio for major adverse cardiovascular events was 1.42 (95% CI 0.64–3.19) at 1 year and 1.51 (0.79–2.88) at 3 years, with incidence of 9.20 versus 6.43 per 1,000 person-years at 1 year. The confidence intervals are wide and the event counts are small. Against teriparatide, one propensity-matched cohort found a lower hazard for major adverse cardiovascular events (HR 0.82, 95% CI 0.70–0.97) and myocardial infarction (HR 0.77, 95% CI 0.63–0.95). A population-based cohort found no difference over 365 days. These designs are observational, so residual confounding is possible.

Spontaneous-reporting data show elevated reporting. FAERS analyses report reporting odds ratios (RORs) of 2.38 to 4.90 across myocardial infarction, stroke, cardiovascular death and cardiac failure. These measure reporting disproportionality, not incidence, and they cannot establish causality. Genetic evidence supplies a mechanism. Variants that mimic sclerostin inhibition, the target of romosozumab, lower fracture risk and raise the risk of myocardial infarction, coronary revascularization and coronary artery disease. This is indirect evidence. It is a reason the signal might be real, not a measure of the drug's effect.

Where the evidence splits

Trials versus alendronate and trials versus placebo disagree in direction. ARCH shows numerically more serious cardiovascular events on romosozumab in the first year, while FAFRAME shows balance against placebo. The sources suggest a comparator difference, but they do not state a design reason for it. One network meta-analysis notes the inconsistency with the ARCH finding and says real-world confirmation is needed. Another suggests no excess signal against comparators other than alendronate. Whether the ARCH imbalance reflects romosozumab, alendronate, or chance is unexplained by the sources on this page.

Pooled analyses split on direction. Three network meta-analyses and one placebo-controlled meta-analysis find no increase. Two meta-analyses find a probable or numerically higher risk. The sources do not state which trials, follow-up windows or outcome definitions drive the difference. The split is unexplained by the sources on this page.

Cohorts versus teriparatide split. One matched cohort finds fewer events on romosozumab (HR 0.82), while a population-based cohort finds no difference over 365 days. Both are observational. The sources do not say which design feature explains the difference.

Pharmacovigilance versus adjudicated trials split. FAERS shows elevated reporting, while the FRAME trial shows balance against placebo. The sources do not reconcile these. One FAERS analysis notes the overall elevation was primarily driven by Japanese reports, and 164 of 206 MACE reports (79.6%) came from Japan. Spontaneous reporting also lacks exposure denominators.

Thresholds and dose-response

No threshold established in the current sources.

Open empirical questions

• What are the absolute event counts and rates behind the ARCH first-year imbalance versus alendronate, and does it persist beyond year one?
• Do patients with pre-existing cardiovascular disease or diabetes face a different romosozumab-attributable risk from the one seen in the general trial population? The sources show higher absolute event rates in those groups but no comparison to controls.
• What explains the opposite results of pooled analyses, and which trials, windows or outcome definitions drive each?
• Do real-world cohorts with larger event counts and longer follow-up narrow the wide confidence intervals against denosumab and teriparatide?