Risk of amputation with canagliflozin across categories of age and cardiovascular risk in three US nationwide databases: cohort study.

OBJECTIVE: To estimate the rate of lower limb amputation among adults newly prescribed canagliflozin according to age and cardiovascular disease.
DESIGN: Population based, new user, cohort study. DATA SOURCES: Two commercial and Medicare claims databases, 2013-17. PARTICIPANTS: Patients newly prescribed canagliflozin were propensity score matched 1:1 with patients newly prescribed a glucagon-like peptide-1 (GLP-1) receptor agonist. Hazard ratios and rate differences per 1000 person years were computed for the rate of lower limb amputation in the following four groups: group 1, patients aged less than 65 years without baseline cardiovascular disease; group 2, patients aged less than 65 with baseline cardiovascular disease; group 3, patients aged 65 or older without baseline cardiovascular disease; group 4, patients aged 65 or older with baseline cardiovascular disease. Within each group, pooled hazard ratio and rate difference per 1000 person years were calculated by meta-analysis. INTERVENTION: Canagliflozin versus a GLP-1 agonist. MAIN OUTCOME MEASURES: Lower limb amputation requiring surgery.
RESULTS: Across the three databases, 310 840 propensity score matched adults who started canagliflozin or a GLP-1 agonist were identified. The hazard ratio and rate difference per 1000 person years for amputation in adults receiving canagliflozin compared with a GLP-1 agonist for each group was: group 1, hazard ratio 1.09 (95% confidence interval 0.83 to 1.43), rate difference 0.12 (-0.31 to 0.55); group 2, hazard ratio 1.18 (0.86 to 1.62), rate difference 1.06 (-1.77 to 3.89); group 3, hazard ratio 1.30 (0.52 to 3.26), rate difference 0.47 (-0.73 to 1.67); and group 4, hazard ratio 1.73 (1.30 to 2.29), rate difference 3.66 (1.74 to 5.59).
CONCLUSIONS: The increase in rate of amputation with canagliflozin was small and most apparent on an absolute scale for adults aged 65 or older with baseline cardiovascular disease, resulting in a number needed to treat for an additional harmful outcome of 556 patients at six months (that is, 18 more amputations per 10 000 people who received canagliflozin). These results help to contextualize the risk of amputation with canagliflozin in routine care. © Author(s) (or their employer(s)) 2019. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Introduction

Sodium glucose cotransporter 2 (SGLT2) inhibitors decrease plasma glucose by inhibiting its reabsorption at the proximal tubule and by lowering the renal threshold for glucose excretion.1 Canagliflozin reduces the risk of myocardial infarction, stroke, and cardiovascular mortality.2 Additional benefits of the drug include a reduction in the rate of hospital admission for heart failure and renal failure, and reductions in systolic blood pressure, weight, and proteinuria.2 3 SGLT2 inhibitors, however, are also associated with an increased risk of important adverse events (eg, diabetic ketoacidosis, urogenital infections).4 5 The CANVAS (CANagliflozin cardioVascular Assessment Study) Program also detected an increased risk of amputation with canagliflozin in comparison with placebo.2

CANVAS consisted of two randomized cardiovascular outcome trials that included adults with established cardiovascular disease or multiple risk factors for cardiovascular disease.2 Canagliflozin was associated with a nearly twofold increased risk of lower limb amputation compared with placebo (6.3 v 3.4 per 1000 person years). The increased risk of amputation was surprising, because it had not been reported in previous clinical trials of canagliflozin or other SGLT2 inhibitors.6 As the mean age of patients included in the CANVAS Program was 63 years, about 10 years older than patients in previous trials, the risk of amputation could possibly be confined to older adults.6 The increased risk of amputation in CANVAS might also be related to the higher baseline cardiovascular risk of the patients included compared with previous trials of canagliflozin.6

Since the results of CANVAS, many observational studies have quantified the rate of amputation with canagliflozin. Two studies, started by the manufacturers of canagliflozin, did not find a higher rate of amputation in analyses that focused on canagliflozin.7 8 Two other studies, using data from the insurance claims database MarketScan up until 2015, found an increased rate of amputation with SGLT2 inhibitors (most of which were canagliflozin), but because of the paucity of events the confidence intervals were wide9 and crossed 1.0 for one of the studies, allowing for a possible null effect.10 In all four studies, the average age of patients was about 55 and most patients had no baseline history of cardiovascular disease.

As age and cardiovascular disease are two of the strongest risk factors for amputation among adults with diabetes, it is important to assess the potentially increased risk of amputation with canagliflozin among older adults and patients at higher risk of cardiovascular disease.11 12 13 We considered including other SGLT2 inhibitors but were underpowered to do so. Furthermore, canagliflozin was the most commonly dispensed SGLT2 inhibitor in the United States during the period of our study. In addition, after publication of the CANVAS Program, patients at higher risk of amputation could have been directed towards receiving empagliflozin or dapagliflozin because neither drug has been shown to be associated with an increased risk of amputation.14 15 Thus we conducted an observational cohort study using three nationwide databases in the US with data up until 2017 to estimate the rate of lower limb amputation among patients newly prescribed canagliflozin, according to age and baseline cardiovascular risk.

Methods

Study population

We conducted a population based, new user, cohort study using three US healthcare claims databases: Optum Clinformatics Data Mart Database, IBM MarketScan, and Medicare fee-for-service (parts A/B/D). The databases contain deidentified longitudinal, individual level data on healthcare use, inpatient and outpatient diagnoses, diagnostic tests and procedures, outpatient laboratory results (available for about 40% of patients in Optum and 6% of patients in MarketScan), and pharmacy dispensing of drugs. MarketScan and Optum primarily include adults with employer based health plans. The average age of adults is generally higher in Optum than in MarketScan, because about 10% of adults in Optum have Medicare Advantage (a supplemental insurance plan for patients in Medicare).16 Medicare predominantly includes adults aged over 65 years. Collectively, these three databases include over 100 million people in the US.

We compared data for adults with diabetes mellitus type 2 newly prescribed canagliflozin or a glucagon-like peptide-1 (GLP-1) agonist between 29 March 2013 (date of approval of canagliflozin) and the last available data (Optum: 30 September 2017, MarketScan: 31 December 2017, Medicare: 31 December 2016). The CANVAS trial was published online in June 2017, and thus the data for the vast majority of patients included in our study predated that date.2 Patients with diabetes mellitus type 2 were identified using ICD-9 (international classification of diseases, ninth revision) and ICD-10 codes (appendix table 1). The cohort entry date was the date of the first prescription for canagliflozin or a GLP-1 agonist. New users of these drugs were defined as those without a previous prescription for an SGLT2 inhibitor or GLP-1 agonist in the preceding 180 days. GLP-1 agonists were chosen as the comparator because they are also a second line treatment for diabetes, have similar cardiovascular benefits, and have not been associated with an increased risk of amputation.17 18

Patients prescribed both canagliflozin and a GLP-1 agonist on the cohort entry date were excluded. Patients with any of the following characteristics in the 180 days before cohort entry were also excluded: insufficient baseline healthcare data (that is, <180 days), human immunodeficiency virus, end stage renal disease, cancer, or type 1 diabetes. These characteristics were selected because they could be associated with directing drugs to a specific patient group. For example, some patients with type 1 diabetes in the US are prescribed SGLT2 inhibitors off label.19 Similar to the CANVAS Program, patients with a prior amputation were not excluded.

Cohort follow-up

Follow-up began on the day after cohort entry (date of first prescription) and continued until the end of the study, end of enrollment in a continuous health plan, occurrence of a study outcome, discontinuation of the initial drug (or switching to, or adding, the comparator drug), or death. A drug was considered to be discontinued if 60 days had elapsed after the expiration of the last prescription’s supply without the prescription being refilled.

Study outcome

The primary outcome was lower limb amputation requiring surgery based on an ICD-9/10 procedural code or a current procedural terminology code (appendix 2).7 20 These codes are commonly used in observational studies,7 8 9 and a validation study showed sensitivity above 94% and positive predictive value above 80%.21

Baseline covariates

All covariates were assessed during the 180 days before cohort entry, including the index date. Data were collected for each patient (table 1 and appendix 3) on the basis of diagnoses and procedures recorded during health encounters, including chronic medical conditions, diabetes severity, amputation risk, healthcare use, recent clinic appointments, drugs for treatment of diabetes and non-diabetes related drugs. These variables were selected a priori based on available literature, clinical experience, and expert opinion.22 23 24

Table 1

Selected baseline patient characteristics after propensity score matching across three US databases. Data are number (%) unless otherwise specified

	Age <65						Age ≥65
	No baseline cardiovascular disease			Baseline cardiovascular disease			No baseline cardiovascular disease			Baseline cardiovascular disease
	GLP-1 (n=80 640)	Canagliflozin (n=80 640)		GLP-1 (n=10 763)	Canagliflozin (n=10 763)		GLP-1 (n=44 522)	Canagliflozin (n=44 522)		GLP-1 (n=19 495)	Canagliflozin (n=19 495)
Female sex	39 896 (49.5)	39 841 (49.4)		3764 (35.0)	3788 (35.2)		24 625 (55.3)	24 603 (55.3)		8325 (42.7)	8291 (42.5)
Age (mean; SD)	51.93 (8.41)	51.95 (8.35)		55.84 (6.63)	55.86 (6.46)		71.11 (5.36)	71.11 (5.34)		72.60 (6.08)	72.62 (6.17)
Hypertension	53 480 (66.3)	53 591 (66.5)		9178 (85.3)	9213 (85.6)		30 954 (69.5)	31 014 (69.7)		17 020 (87.3)	17 021 (87.3)
Hyperlipidemia	54 381 (67.4)	54 528 (67.6)		8755 (81.3)	8781 (81.6)		27 208 (61.1)	27 213 (61.1)		14 708 (75.4)	14 753 (75.7)
Coronary artery disease	0	0		7695 (71.5)	7707 (71.6)		0	0		14 523 (74.5)	14 531 (74.5)
Cerebrovascular disease	0	0		1017 (9.4)	1024 (9.5)		0	0		1626 (8.3)	1640 (8.4)
Peripheral vascular disease	0	0		2077 (19.3)	2111 (19.6)		0	0		4119 (21.1)	4144 (21.3)
Prior amputation	0	0		207 (1.9)	215 (2.0)		0	0		348 (1.8)	366 (1.9)
Diabetic nephropathy	3809 (4.7)	3861 (4.8)		802 (7.5)	815 (7.6)		2648 (5.9)	2677 (6.0)		1515 (7.8)	1576 (8.1)
Diabetic retinopathy	2988 (3.7)	2967 (3.7)		702 (6.5)	698 (6.5)		1423 (3.2)	1422 (3.2)		920 (4.7)	892 (4.6)
Diabetic neuropathy	7999 (9.9)	8009 (9.9)		2248 (20.9)	2240 (20.8)		4875 (10.9)	4870 (10.9)		3542 (18.2)	3552 (18.2)
Insulin use	19 300 (23.9)	19 330 (24.0)		3641 (33.8)	3700 (34.4)		12 797 (28.7)	12 847 (28.9)		7240 (37.1)	7237 (37.1)
Metformin use	50 133 (62.2)	50 157 (62.2)		6466 (60.1)	6455 (60.0)		26 418 (59.3)	26 451 (59.4)		10 598 (54.4)	10 563 (54.2)
DPP4 inhibitor use	13 764 (17.1)	14 666 (18.2)		1963 (18.2)	2047 (19.0)		11 769 (26.4)	11 696 (26.3)		5322 (27.3)	5315 (27.3)
Diuretic use	11 199 (13.9)	11 373 (14.1)		3091 (28.7)	3159 (29.4)		12 082 (27.1)	11 949 (26.8)		8645 (44.3)	8662 (44.4)
Endocrinology clinic visit	14 758 (18.3)	15 064 (18.7)		2347 (21.8)	2422 (22.5)		8264 (18.6)	8283 (18.6)		3896 (20.0)	3931 (20.2)
Vascular or orthopedic surgery clinic visit	7354 (9.1)	7332 (9.1)		1516 (14.1)	1530 (14.2)		6483 (14.6)	6461 (14.5)		3818 (19.6)	3879 (19.9)
Inpatient hospital admission	1401 (1.7)	1387 (1.7)		1064 (9.9)	1092 (10.1)		2624 (5.9)	2587 (5.8)		4894 (25.1)	4832 (24.8)
Outpatient visits, mean (SD)	4.24 (3.06)	4.26 (3.19)		6.51 (4.49)	6.55 (4.52)		5.57 (4.18)	5.56 (4.33)		7.80 (5.26)	7.81 (5.38)
Hemoglobin A1c (%), mean (SD)*	8.77 (2.01)	8.88 (1.86)		8.76 (1.93)	8.83 (1.81)		8.32 (1.67)	8.55 (1.65)		8.40 (1.65)	8.49 (1.63)
Creatinine (mg/dL), mean (SD)*	0.87 (0.25)	0.85 (0.22)		0.97 (0.34)	0.93 (0.26)		1.01 (0.34)	0.94 (0.27)		1.11 (0.39)	1.03 (0.31)

DPP4=dipeptidyl peptidase 4; GLP-1=glucagon-like peptide-1 analogue; SD=standard deviation.

Baseline characteristics before matching are provided in the appendix.

Creatinine and hemoglobin A1c values were available only for a subset of patients (appendix). For patients younger than 65, laboratory results were available for about 19% of patients. For patients aged 65 and older, laboratory results were available for about 9% of patients.

Statistical analysis

We used propensity score matching to adjust for confounding. The probability of a prescription fill for canagliflozin versus a GLP-1 agonist was calculated through a multivariable logistic regression model, which contained all potential confounders at baseline except laboratory values—for example, hemoglobin A1c and creatinine, because these results were not available for all patients. A caliper of 0.05 on the propensity scale was used for matching with a 1:1 ratio. Covariate balance between the matched exposure groups was assessed using standardized differences. A standardized difference of 0.1 or less indicates negligible differences between groups.25

Within each database, separate propensity scores were estimated and used for 1:1 matching initiators of canagliflozin or a GLP-1 agonist in each of four mutually exclusive groups: group 1, patients aged less than 65 without baseline cardiovascular disease; group 2, patients aged less than 65 with baseline cardiovascular disease; group 3, patients aged 65 and older without baseline cardiovascular disease; group 4, patients aged 65 and older with baseline cardiovascular disease. Cardiovascular disease was defined as a diagnosis during the baseline period of one or more of coronary artery disease, stroke or transient ischemic attack, heart failure, peripheral vascular disease, or amputation.

After propensity score matching, we calculated the rate per 1000 person years of the primary outcome among exposure groups in each of the four groups of age and cardiovascular disease within each database. We also calculated rate differences per 1000 person years and 95% confidence intervals, and estimated hazard ratios and 95% confidence intervals, using proportional hazards models without further adjustments. Within each of the four groups of age and cardiovascular disease, hazard ratios and rate differences per 1000 person years from each database were then combined using a random effects meta-analysis that weighted studies by the inverse of their variance using the DerSimonian and Laird approach.26 Estimated variances for rates assumed a Poisson distribution. We evaluated the presence of effect measure modification of canagliflozin across 10 categories of age, baseline cardiovascular disease, and database on a relative and absolute scale using the Wald test for homogeneity (χ2 with nine degrees of freedom).27 This analysis allowed us to estimate the relative effect of cardiovascular disease and age on the rate of amputation. We provided both relative risk estimates and risk difference estimates to help contextualize the potential risk of amputation for both a population and an individual patient. Measurement of the risk in absolute terms is critical for a well informed benefit-risk assessment.28

To assess the ability of correctly reproducing a known association in our data, we also assessed the risk of hospital admission for heart failure as a positive control, because previous studies have shown a reduction in heart failure with SGLT2 inhibitors, but not with GLP-1 agonists.2 20 29 30 All analyses were conducted using the validated Aetion platform, version 3.1231 and R version 3.1.2.

Patient and public involvement

No patients were involved in planning the research question, outcome measures, study design, study implementation, or interpretation or writing up of the results.

Results

Study cohort and patient characteristics

From MarketScan, Optum, and Medicare we identified 496 642 people who satisfied study inclusion and exclusion criteria (appendix fig 1) before propensity score matching. Patient characteristics were well balanced after propensity score matching (table 1). Appendix table 3 shows complete patient characteristics within each database before and after propensity score matching. In general, patients who were prescribed a GLP-1 agonist were more likely to be female, have chronic kidney disease, or to have filled a prescription for insulin or a diuretic agent during the baseline period. Patients who were newly prescribed an SGLT2 inhibitor were less likely to have a diagnosis related to obesity or to have seen an endocrinologist, and more likely to have received a filled prescription for a dipeptidyl peptidase 4 inhibitor at baseline. The mean follow-up was approximately 275 days (appendix table 4). After propensity score matching, we identified 155 420 patients who were prescribed canagliflozin and 155 420 who were prescribed a GLP-1 agonist.

Patients aged less than 65

After propensity score matching, of patients under the age of 65 years without baseline cardiovascular disease, 98 adults who were prescribed a GLP-1 agonist had a lower limb amputation (1.55 per 1000 person years) compared with 117 adults prescribed canagliflozin (1.67 per 1000 person years; table 2). This result corresponded to a hazard ratio of 1.09 (95% confidence interval 0.83 to 1.43, I2=0%; fig 1) and to an absolute rate difference of 0.12 per 1000 person years (95% confidence interval −0.31 to 0.55; fig 2). Appendix table 5 shows the rates of amputation for patients from each database. For patients younger than 65 years with baseline cardiovascular disease, 69 adults prescribed a GLP-1 agonist had a lower limb amputation (8.57 per 1000 person years) compared with 87 adults prescribed canagliflozin (9.74 per 1000 years). This result corresponded to a hazard ratio of 1.18 (0.86 to 1.62, I2=0%; fig 1) and to an absolute rate difference of 1.06 per 1000 person years (−1.77 to 3.89; fig 2).

Table 2

Propensity score matched rate of amputation, hazard ratios, and rate differences stratified by age and baseline cardiovascular disease

	MarketScan			Optum			Medicare			Overall
	GLP-1	Canagliflozin		GLP-1	Canagliflozin		GLP-1	Canagliflozin		GLP-1	Canagliflozin
Patients without baseline cardiovascular disease, aged <65:
No of patients	56 292	56 292		24 348	24 348		NA	NA		80 640	80 640
No of person years	46 364	50 450		16 875	19 822		NA	NA		63 239	70 272
No of amputations	74	82		24	35		NA	NA		98	117
Rate per 1000 person years	1.60	1.63		1.42	1.77		NA	NA		1.55	1.67
Patients with baseline cardiovascular disease, aged <65:
No of patients	6925	6925		3838	3838		NA	NA		10 763	10 763
No of person years	5473	6038		2579	2896		NA	NA		8052	8934
No of amputations	41	50		28	37		NA	NA		69	87
Rate per 1000 person years	7.49	8.28		10.86	12.77		NA	NA		8.57	9.74
Patients without baseline cardiovascular disease, aged ≥65:
No of patients	6715	6715		8026	8026		29 781	29 781		44 522	44 522
No of person years	5957	5993		5318	5838		21 724	22 818		32 999	34 649
No of amputations	12	6		2	8		39	67		53	81
Rate per 1000 person years	2.01	1.00		0.38	1.37		1.8	2.94		1.61	2.34
Patients with baseline cardiovascular disease, aged ≥65:
No of patients	2954	2954		3605	3605		12 936	12 936		19 495	19 495
No of person years	2513	2504		2273	2526		8775	9136		13 561	14 166
No of amputations	7	15		8	16		59	101		74	132
Rate per 1000 person years	2.79	5.99		3.52	6.33		6.72	11.06		5.46	9.32

GLP-1=glucagon-like peptide-1 agonist; NA=not applicable.

Fig 1

Forest plot for propensity score matched rate of amputation for adults younger than 65 without cardiovascular disease, younger than 65 with cardiovascular disease, aged 65 and older without cardiovascular disease, and aged 65 and older with cardiovascular disease. GLP-1=glucagon like peptide-1

Fig 2

Forest plot for propensity score matched rate difference of amputation for adults younger than 65 without cardiovascular disease, younger than 65 with cardiovascular disease, aged 65 and older without cardiovascular disease, and aged 65 and older with cardiovascular disease. GLP-1=glucagon like peptide-1

Patients aged 65 and older

Among patients aged 65 and older without baseline cardiovascular disease, 53 adults who were prescribed a GLP-1 agonist had a lower limb amputation (1.61 per 1000 person years) compared with 81 adults prescribed canagliflozin (2.34 per 1000 person years). This result corresponded to a hazard ratio of 1.30 (95% confidence interval 0.52 to 3.26, I2=68%; fig 1) and to an absolute rate difference of 0.47 (−0.73 to 1.67; fig 2) per 1000 person years. For patients aged 65 and older with baseline cardiovascular disease, 74 adults prescribed a GLP-1 agonist had a lower limb amputation (5.46 per 1000 person years) compared with 132 adults prescribed canagliflozin (9.32 per 1000 person years). This result corresponded to a hazard ratio of 1.73 (1.30 to 2.29, I2=0%; fig 1) and to an absolute rate difference of 3.66 per 1000 person years (95% CI 1.74 to 5.59; fig 2). Figure 3 shows Kaplan-Meier plots for amputation for adults younger and older than 65 and with and without cardiovascular disease. The evidence for heterogeneity across the 10 groups formed by age, baseline cardiovascular disease, and database, was modest on a relative scale, with overall test of heterogeneity not reaching significance (P=0.14); evidence for significant heterogeneity was shown on an absolute scale (P=0.01). These data suggest that the combination of both factors (that is, age and cardiovascular disease) meaningfully increased the absolute rate of amputation.

Fig 3

Kaplan-Meier plot for amputation after propensity score matching for adults (A) younger than 65 years without cardiovascular disease, (B) younger than 65 years with cardiovascular disease, (C) aged 65 and older without cardiovascular disease, and (D) aged 65 and older with cardiovascular disease. GLP-1=glucagon like peptide-1

Validation against a known causal association

The propensity score matched rate of hospital admission due to heart failure was lower for patients prescribed canagliflozin than for those receiving GLP-1 agonists across the four groups (appendix table 6). For example, among patients under 65 without baseline cardiovascular disease, the hazard ratio was 0.50 (95% confidence interval 0.29 to 0.88) for canagliflozin compared with a GLP-1 agonist.

Discussion

In this longitudinal study of over 300 000 propensity score matched adults with type 2 diabetes mellitus given routine care, an increased rate of amputation associated with canagliflozin, compared with a GLP-1 agonist, was small. The rate was most apparent on an absolute scale among adults aged over 65 with baseline cardiovascular disease, resulting in nearly four more amputations in 1000 person years. This observation, in addition to the rate of amputation calculated in the other three groups included in our study, which resulted in 0.1 to 1.1 additional amputations in 1000 person years, helps to put into context the rate of amputation associated with the initiation of canagliflozin in routine care and can inform decision making between physicians and patients before this drug is started.

Assessing the incremental risk of amputation associated with the use of new drugs for diabetes is important, because adults with diabetes have a markedly higher rate of amputation than adults without diabetes.32 33 Lower limb amputation is also associated with considerable disability, morbidity, and mortality.34 35 For example, in adults over the age of 65, amputation is associated with a one year mortality of about 50%.34 35

We observed a similar rate of amputation for adults aged less than 65 prescribed canagliflozin or a GLP-1 agonist, regardless of baseline cardiovascular disease, which is in agreement with previous real world studies of canagliflozin.7 8 10 It is unlikely that our findings are explained by changes in prescribing of canagliflozin on the basis of the potential risk of amputation, because almost all data from our study preceded publication of the CANVAS Program. Unlike CANVAS, we were unable to include a placebo group because we used data from routine care. We could have compared adults with diabetes who were prescribed canagliflozin with adults with diabetes who did not receive the drug. Previous research, however, has shown that a non-user comparator group can lead to several forms of bias, including confounding and immortal time bias.36 Our findings are also consistent with the initial clinical trials of canagliflozin, which typically included patients aged less than 65 and in whom baseline cardiovascular disease was uncommon.6

The higher rate of amputation with canagliflozin in our study was most apparent among adults aged 65 or older who had baseline cardiovascular disease. This rate is in line with a recent observational study of 12 629 veterans from the Veterans Affairs database (mean age 66) prescribed an SGLT2 (58% were prescribed canagliflozin) compared with adults not prescribed an SGLT2 (odds ratio 1.99, 95% confidence interval 1.12 to 3.51).20 The rate of amputation we observed among adults aged 65 and older with baseline cardiovascular disease corresponded to a number needed to treat for an additional harmful outcome of 556 (95% confidence interval 301 to 3596) at six months, calculated based on the Altman and Andersen approach.37 Thus it is expected that one additional person will incur an amputation for every 556 adults receiving canagliflozin rather than a GLP-1 agonist over six months (that is, 18 more amputations per 10 000 people who received canagliflozin).

Our results are similar to those of the CANVAS Program, but they differ from the recently published CREDENCE trial (canagliflozin and renal outcomes in type 2 diabetes and nephropathy).3 The CREDENCE trial randomized 4401 adults with diabetes who had albuminuric chronic renal failure to either canagliflozin or placebo and observed a similar rate of amputation with canagliflozin and placebo (hazard ratio 1.11, 95% confidence interval 0.70 to 1.37).3 In May 2016, however, two years after the start of the CREDENCE trial, an amendment was issued that “asked investigators to examine patients’ feet at each trial visit and temporarily interrupt the assigned treatment in patients with any active condition that might lead to amputation.”3 Although this was necessary to prevent potential amputation with canagliflozin, it could also have driven the null finding. Another explanation for the lack of an increased risk of amputation with canagliflozin could be related to baseline cardiovascular risk. For example, only half of the included patients in the CREDENCE trial had cardiovascular disease at baseline compared with two thirds in the CANVAS Program.2 3

Established cardiovascular disease is one of the strongest risk factors for amputation among adults with diabetes.11 33 38 In our study, the overall rate of lower limb amputation was highest among patients prescribed canagliflozin with baseline cardiovascular disease (age <65 years: 9.74 per 1000 person years, age ≥65 years: 9.32 per 1000 person years). By contrast, the overall rate of amputation for adults without established cardiovascular disease prescribed canagliflozin was lower (age <65 years: 1.67 per 1000 person years, age ≥65 years: 2.34 per 1000 person years).7 8 10 Taken together, these facts suggest that established cardiovascular disease might be a stronger predictor for amputation than older age.

The cardiovascular outcome trials for empagliflozin and dapagliflozin found no increased risk of amputation compared with placebo.14 29 Some have speculated that unique pharmacologic properties of canagliflozin could explain the increased risk of amputation that has not been seen for empagliflozin or dapagliflozin.29 39 One hypothesis is that canagliflozin has a more potent diuretic effect than empagliflozin or dapagliflozin, and the resultant hemoconcentration and hyperviscosity could increase the risk of peripheral tissue ischemia.39 This mechanism has also been reported to explain the increased rate of amputations seen with some diuretic agents.40 In vitro studies have shown that canagliflozin, but not dapagliflozin or empagliflozin, inhibits both glutamate dehydrogenase and complex I of the mitochondrial electron transport chain.41 This dual inhibition has been shown to cause intracellular accumulation of various amino acids which can induce cytotoxicity.41 It is unknown if either of these purported mechanisms explains the observed increased risk or if there is another mechanism yet to be identified.

Limitations

A limitation of our study is that it was not randomized, and thus is prone to residual confounding. Furthermore, our study lacked an adequate sample size to individually analyze other SGLT2 inhibitors (that is, empagliflozin, dapagliflozin). A recent cohort study identified a higher rate of amputation among adults prescribed empagliflozin or dapagliflozin (mean age 61 years) than those prescribed GLP-1 agonists, but few people receiving an SGLT2 inhibitor had an amputation (n=27), as expressed in the study’s wide confidence intervals (hazard ratio 2.32, 95% confidence interval 1.37 to 3.91).42 Another limitation is that we selected GLP-1 agonists rather than another class of oral drugs for diabetes (eg, dipeptidyl peptidase 4 (DPP4) inhibitors). Previous work shows that patients prescribed SGLT2 inhibitors and GLP-1 agonists are of similar age, whereas adults prescribed a DPP4 inhibitor tend to be older and are more likely to have reduced renal function.43 44 Furthermore, both SGLT2 inhibitors and GLP-1 agonists have been shown to reduce the risk of cardiovascular events in randomized controlled trials, and thus in comparison with DPP4 inhibitors, it is expected that these two classes of drugs are more likely to be prescribed to patients with similar baseline cardiovascular risk. This choice of prescription is supported by the strong similarity between these two groups of patients (that is, those receiving GLP-1 agonists compared with SGLT2 inhibitors) before propensity score matching.

An additional limitation of our study is that we lacked complete data on the severity of the diabetes for all patients (eg, duration of diabetes, hemoglobin A1c) and other risk factors for amputation (eg, body mass index, duration of smoking). A recent study showed that propensity score matching of other diabetes related severity and comorbidity variables can achieve balance for such unmeasured characteristics, including duration of diabetes, body mass index, and smoking.45 We do not expect that these differences would be large enough to affect the risk of amputation because the comparator drug treatments (that is, GLP-1 agonists) are prescribed to patients with similar baseline characteristics. Furthermore, we stratified by variables strongly associated with amputation (eg, older age) and by cardiovascular disease, which is colinear with diabetes severity. Another important limitation of our study is that tests for interaction are low powered and we cannot perfectly separate cardiovascular risk from older age since the two are highly positively correlated.46

Conclusion and future implications

Our results suggest that the increased risk of amputation with canagliflozin is small and most apparent on an absolute scale among adults aged 65 or older with baseline cardiovascular disease, resulting in nearly four more amputations in 1000 person years with canagliflozin than with a GLP-1 agonist, corresponding to a number needed to treat for an additional harmful outcome of 556 patients at six months. These findings help to contextualize the risk of amputation in routine care and can help patient-physician decision making before prescribing canagliflozin.

The cardiovascular outcome trial for canagliflozin identified a nearly twofold increased risk of lower limb amputation with canagliflozin than with placebo

The mean age of patients included in the CANVAS Program was 63 years, about 10 years older than patients in previous trials, suggesting that the risk of amputation could be confined to older adults

The increased risk of amputation in CANVAS could also be related to the higher baseline cardiovascular risk of patients included in that trial in comparison with previous trials of canagliflozin

This study included 310 840 propensity score matched adults who were newly prescribed canagliflozin or a glucagon-like peptide-1 (GLP-1) agonist

The increased risk of amputation with canagliflozin was most apparent among older adults who had cardiovascular disease at baseline

In this population the hazard ratio was 1.73 (95% confidence interval 1.30 to 2.29), corresponding to a rate difference of 3.66 (95% confidence interval 1.74 to 5.59) per 1000 person years compared with GLP-1 agonists