Samples from 1,287 participants in the treatment arm and 1,256 in the placebo arm at baseline and at year 1 were analyzed, totaling 60% of the original cohort (Reduced CREDENCE cohort). Overall, there were 850 males and 437 females in the canagliflozin arm, and 836 males and 420 females in the placebo group (P = 0.783). Among the measured plasma substrates, the strongest associations were between FFAs and glycerol (r = 0.33 at baseline and r = 0.66 at year 1, P < 0.0001 for both), and between β-OH and AcAc (r = 0.44 at baseline and r = 0.46 at year 1, P < 0.0001 for both). Intraindividual correlations for each substrate over time (baseline and 1 year) were remarkably robust (with r values ranging from 0.40 to 0.60) regardless of treatment assignment.

Renal endpoint versus hHF. In the Reduced CREDENCE cohort, the effect of canagliflozin treatment on the primary composite endpoint (n = 330 first events) closely reproduced the result of the Full CREDENCE cohort (n = 585 first events) (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.180637DS1). In the full dataset, time to first composite renal endpoint was significantly (P < 0.0001) longer (707 [IQR 420] days) than time to first hHF (556 [IQR 438] days). Risk of first composite renal endpoint was significantly greater in participants who had had a first episode of hHF (n = 87) as compared with those who had not (n = 143, HR 2.92 [95% CI: 2.16–3.94], P < 0.0001).

Baseline substrates versus outcomes. In univariate Cox analysis, participants in the bottom tertile of baseline fasting plasma lactate concentrations were at significantly higher risk for both the primary composite and the renal composite endpoint (Figure 1). In multivariate Cox models using all 6 measured substrates (Model I), lactate was inversely associated with both the primary endpoint and the renal endpoint; additionally, lactate was negatively associated with all-cause death (Table 1). Notably, lactate was inversely associated with each of the individual components of the composite renal endpoint (Table 2).

Baseline substrates versus outcomes. Kaplan-Meier plot of time to first primary composite endpoint (top) and time to composite renal endpoint (bottom) by tertile of baseline fasting plasma lactate concentrations.

Multivariate associations of baseline substrates with outcomes

Association of baseline substrates with individual renal outcomes

When Model I was adjusted for the full set of covariates and potential confounders (Model II), lactate still emerged as a negative predictor of the primary composite endpoint, hHF/CVD, and all-cause death (Table 1 and Figure 2).

Forest plot showing the multivariate Cox proportional hazards model of the association between several covariates, including baseline fasting plasma lactate and free fatty acids, with time to first primary composite endpoint. Only covariates reaching statistical significance are plotted; additional covariates with no statistically significant differences are not plotted. SD, standard deviation; Ln, natural logarithm; BMI, body mass index; HbA1c, glycated hemoglobin A1c; eGFR, estimated glomerular filtration rate; UACR, urinary albumin-to-creatinine ratio; LDL-C, low-density lipoprotein cholesterol; FFA, free fatty acids.

In the present study cohort, patients with first on-trial hHF had a 3-fold higher history of HF (33% vs. 12%, P < 0.0001) and lower baseline FFA (398 [IQR 334] vs. 462 [IQR 340] μmol/L, P = 0.0031) than patients without this outcome. Most notably, baseline fasting plasma FFAs showed essentially the same pattern of associations as plasma lactate levels, namely, a graded increase in risk of the composite renal endpoint and hHF/CVD across decreasing concentration tertiles in univariate Cox analysis (Figure 3), and an independent negative association with the primary and renal endpoints and hHF/CVD in the fully adjusted Cox model (Table 1 and Figure 4). In the latter analysis, however, FFAs were not significant predictors of all-cause death. Along with lactate, plasma FFAs too were significant independent predictors of the components of the composite renal endpoint (Table 2).

Kaplan-Meier plots of time to first composite renal endpoint and time to first hospitalized congestive heart failure by tertile of baseline fasting plasma FFA concentrations.

Forest plots showing the multivariate Cox proportional hazards model of the association between several covariates, including baseline fasting plasma lactate and free fatty acids, with time to first of the composite of hospitalized congestive heart failure (upper panel) or cardiovascular death and time to all-cause death (lower panel). Only covariates reaching statistical significance are plotted; additional covariates with no statistically significant differences are not plotted. SD, standard deviation; Ln, natural logarithm; HbA1c, glycated hemoglobin A1c; UACR, urinary albumin-to-creatinine ratio; LDL-C, low-density lipoprotein cholesterol; CVD, cardiovascular disease; HF, heart failure; FFA, free fatty acids.

To identify the phenotype of participants with different baseline lactate or FFA levels, we compared clinical and metabolic features across tertiles of the 2 substrates (Table 3). For the lowest versus the highest FFAs, participants were more often males, slightly younger, and with a longer duration of diabetes. The urinary albumin-to-creatinine ratio (UACR) was higher and other substrates (glucose, glycerol, β-OH, and AcAc) were lower, as was a prior history of HF. Of note, background insulin use was much higher, while metformin or sulfonylurea use was lower. For lactate, the pattern was remarkably overlapped with that for FFAs, except for a lesser use of statins and anti-thrombotics.

Clinical phenotype of participants by tertile of baseline FFA and by tertile of baseline lactate

Treatment. In univariate analysis, treatment led to significant protection not only against the primary composite endpoint (Supplemental Figure 1), but also against the composite renal endpoint (HR 0.68 [95% CI: 0.51–0.89]) and its components (HR 0.69 [95% CI: 0.50–0.95]) for CV and renal death, HR 0.59 [95% CI: 0.43–0.80] for serum creatinine doubling, HR 0.69 [95% CI: 0.50–0.95] for end-stage kidney disease, and HR 0.56 [95% CI: 0.38–0.83] for end-stage kidney disease with an estimated glomerular filtration rate (eGFR) below 15.

As compared with placebo, canagliflozin treatment led to a decrease in plasma glucose, and a rise in the plasma concentrations of FFA, glycerol, β-OH, and AcAc (Table 4). Using year 1 substrate concentrations, multivariate Cox models (Model I) indicated that higher levels of FFA and lactate were independently associated with significantly reduced risk of both the primary and the renal endpoint (Supplemental Table 1). With full adjustment for all covariates (Model II), lactate — but not FFA — was still an inverse predictor of both the primary and the renal endpoint.

Plasma substrate measurements

In the 96 patients experiencing a first nonfatal myocardial infarction, neither FFA nor lactate (baseline or year 1 samples) was a significant predictor, nor did canagliflozin treatment protect against this outcome.

To characterize the specific impact of canagliflozin treatment on outcomes, we calculated the changes in substrate levels between baseline and year 1, and related the top quartile of such changes to main endpoints in Cox models adjusting for baseline substrate values and treatment itself. The results (Table 5) show that each substrate, except glycerol, predicted an approximately 30% reduction in relative risk of both the primary composite endpoint and hHF/CVD independently of treatment.

Relationship of canagliflozin-induced changes in plasma substrate concentrations to outcomes

Non-responder analysis. Canagliflozin significantly reduced eGFR slope (–2.0 [IQR –4.7] vs. –3.5 [IQR –6.2] mL/min/1.73 m2/year of placebo, P < 0.0001). By defining non-responders as those participants in the drug treatment arm falling in the top quartile of the slope distribution (i.e., an eGFR decline greater than –4.77 mL/min/1.73 m2/year), non-responders had a much higher risk of composite renal, but also a higher risk of hHF (Figure 5). Including both non-responder status and drug treatment in a multi-adjusted Cox model for the composite renal as well as the full list of risk factors, the multi-adjusted HR was 9.79 [95% CI: 7.40–12.95] for the former and 0.94 [95% CI: 0.76–1.16] for the latter. For the hHF endpoint, non-responder status had an HR of 2.52 [95% CI: 1.89–3.35], while drug treatment had an HR of 0.68 [95% CI: 0.52–0.90]; in neither model was there a significant interaction between responder status and drug treatment. There were 29% non-responders in the bottom baseline lactate third versus 23% in the top third (P = 0.0110); likewise, there were 30% non-responders in the bottom baseline FFA third versus 21% in the top third (P = 0.0004).

Kaplan-Meier plot of time to first composite renal endpoint and time to first hospitalized congestive heart failure by treatment responder status in the treatment arm (see text for definition of responder).