Canagliflozin Amputation Signal Persisted in FAERS After FDA Removed the Boxed Warning

The FDA removed canagliflozin's amputation boxed warning in August 2020. Caliper's analysis of FAERS post-removal data 2020-09 through 2026-05 shows the disproportionality signal persists at PRR = 4.25 vs other SGLT2 inhibitors — a 4× higher amputation reporting rate after the warning was eliminated.

Pre-registration sha1: 8280ccc10160be0b76561b4b7030d0afa299439e · Finding card sha1: 20cad3aa91ea6c788aa21c51b023c614b0622036

What the data shows

Canagliflozin (Invokana, Janssen/J&J) received an FDA boxed warning for amputation risk in 2017 based on the CANVAS trial. In August 2020, after CANVAS-R and CREDENCE showed no excess amputation risk in selected patient populations, the FDA removed the boxed warning. Caliper analyzed FAERS reports from 2020-09 through 2026-05 — 5+ years of post-removal real-world data — and found the disproportionality signal versus other SGLT2 inhibitors (empagliflozin, dapagliflozin, ertugliflozin) persisted at PRR = 4.25 (95% CI 3.69–5.25). The signal did not disappear after the warning was removed.

Specific signals that persisted

Adverse event	n_canagliflozin	n_controls	PRR	χ²
Diabetic foot infection	33	11	40.5	307
Leg amputation	47	48	13.2	262
Foot amputation	25	36	9.4	105
Toe amputation	63	97	8.8	258
Diabetic foot (general)	30	47	8.6	118
Amputation (unspecified)	10	19	7.1	30
Gangrene	38	78	6.6	117

The diabetic foot infection signal (PRR 40.5) is the most striking

The diabetic foot infection rate for canagliflozin is 40× higher than other SGLT2 inhibitors in post-removal FAERS data. Diabetic foot infections are the immediate precursor to amputation. This suggests the underlying pathology that the original 2017 warning targeted is still being captured — even after the warning text was removed.

What this means

For patients

Patients on canagliflozin should be aware that real-world post-2020 reporting data continues to show amputation events at higher rates than other SGLT2 inhibitors
Patients with peripheral artery disease, prior amputation, neuropathy, or diabetic foot ulcer history may want to discuss canagliflozin vs alternatives with their physician

For physicians

The 2020 boxed warning removal should not be interpreted as proof the signal resolved
FAERS data suggests vigilance for amputation risk in canagliflozin patients remains warranted
Diabetic foot care protocols for canagliflozin patients deserve particular attention

For Janssen / Johnson & Johnson

Active pharmacovigilance for the amputation/diabetic foot signal would strengthen post-2020 safety messaging
A retrospective comparative analysis of canagliflozin vs empagliflozin/dapagliflozin amputation rates in claims data would provide higher-quality evidence than FAERS alone

For FDA

The 2020 boxed-warning-removal decision is questioned by post-market FAERS data showing signal persistence at PRR 4.25 vs other SGLT2s
A 2026 reassessment of the canagliflozin amputation safety signal, incorporating 5+ years of post-removal real-world data, would be timely

Methodology (transparent)

Data: openFDA FAERS API, accessed 2026-05-17. Period-stratified queries.
Case drug: canagliflozin (Invokana, Janssen/J&J)
Within-class controls: empagliflozin (Jardiance), dapagliflozin (Farxiga), ertugliflozin (Steglatro)
Periods: PRE = 2017-08 to 2020-08 (boxed warning era); POST = 2020-09 to 2026-05 (post-removal era)
Method: per-AE Proportional Reporting Ratio with 95% CI via Reporting Odds Ratio log-normal, Yates-corrected chi-square. Evans criteria (PRR ≥ 2, χ² ≥ 4) for signal detection.
Pre-registration locked before any data extraction

Caveats and limitations

FAERS is voluntary reporting; absolute rates are not population risks
Notoriety bias from the former boxed warning may artificially elevate canagliflozin reporting throughout both periods, though the within-class comparison partially controls for this
Channeling bias post-2020: prescriber selection may have shifted canagliflozin to specific patient subgroups
Cannot establish causation; this is a signal-detection analysis, not a causal claim
Pre-period rate (36.6%) was inflated by notoriety bias; focus is on post-period rate (4.33%) as the meaningful population-level signal