Which factors predict incident HF in patients with T2D?

A newly developed model that included NT-proBNP, TnT, and BMI accurately predicted new-onset HF in patients with T2D, including those treated with SGLT2 inhibitors.

This summary is based on the publication of Said F, Arnott C, Voors AA, et al. - Prediction of new-onset heart failure in patients with type 2 diabetes derived from ALTITUDE and CANVAS. Diabetes Obes Metab. 2024 Jul;26(7):2741-2751. doi: 10.1111/dom.15592

Introduction and methods

Background

To predict the risk of developing HF in patients with T2D, a few risk scores have been developed [1-5]. However, these risk scores incorporated different predictors, some of which are not readily available in daily clinical practice, such as additional ECG parameters.

Aim of the study

The study aim was to identify predictive markers of new-onset HF in T2D patients at high risk of CV events, including those treated with an SGLT2 inhibitor, create an accurate and clinically applicable new-onset HF prediction model using routinely measured variables, and provide a web-based risk prediction engine to facilitate clinical use of the model.

Methods

To create a derivation model for new-onset HF, the authors used data from the ALTITUDE (Aliskiren Trial in Type 2 Diabetes Using Cardiorenal Endpoints) trial. This multicenter, double-blind, placebo-controlled, parallel-group, phase 3 RCT comprised 8561 T2D patients with evidence of albuminuria or CVD [6]. In the current analysis, 5081 patients with baseline NT-proBNP measurements and no history of HF were included.

The new-onset HF prediction model was externally validated in the placebo arm of the CANVAS (Canagliflozin Cardiovascular Assessment Study) trial (n=4330), a double-blind, placebo-controlled, phase 3 RCT that enrolled patients with inadequately controlled T2D (HbA1c: 53–91 mmol/mol) and symptomatic ASCVD or high CVD risk [7]. In this analysis, 994 placebo-treated patients with baseline NT-proBNP measurements and no HF history were included.

To test the hypothesis that the prediction model retains its predictive ability in high-risk patients after treatment with canagliflozin, the performance of the prediction model was also evaluated in 1668 patients randomized to the treatment arm of the CANVAS trial, at baseline and 1 year after canagliflozin treatment. In addition to baseline NT-proBNP measurements and no HF history, these patients had no HF hospitalization at 1-year follow-up.

Outcome

The endpoint was new-onset HF, defined as adjudicated, unplanned, first-time hospitalization for HF.

Main results

Internal and external validation of prediction model

• In the ALTITUDE cohort, the following variables were identified as independent predictors of new-onset HF: higher log NT-proBNP level, use of calcium channel blockers, higher log urine albumin-to-creatinine ratio, higher BMI, higher log HbA1c level, higher log TnT level, older age and lower haematocrit level.
• In a multivariable model, the strongest predictor of hospitalization for new-onset HF was NT-proBNP level (HR per SD increase of log NT-proBNP: 2.58; 95%CI: 2.19–3.04; P<0.001).
• This model had good discriminative ability (C-statistic: 0.828; 95%CI: 0.801–0.855). Internal validation of the model showed a C-statistic of 0.822, with adequate calibration over tertiles of predicted risk (χ²: 4.953; P=0.084).
• External validation in the placebo arm of the CANVAS trial showed a C-statistic of 0.800, with adequate calibration over predicted risk tertiles (χ²: 1.446; P=0.485).

Risk score formula

• The authors developed a risk score formula for calculating a weighted score for the prediction of new-onset HF. This formula included the following variables: age, BMI, calcium channel blocker use, urine albumin-to-creatinine ratio, HbA1c, hematocrit, hs-TnT, and NT-proBNP. The Risk Score Calculator is available at https://groningencardiology.shinyapps.io/shiny_app.
• In the ALTITUDE cohort, the optimal cut-off point for identifying patients at highest risk of new-onset HF was 9.29 (sensitivity: 0.78; specificity: 0.72). When this cut-off point was applied to the CANVAS placebo population, the sensitivity was 0.96 and the specificity 0.30.

Performance of prediction model after canagliflozin treatment

• NT-proBNP level, TnT level, and BMI were significant and independent predictors of new-onset HF in the ALTITUDE cohort and the CANVAS treatment arm (all P<0.05).
• At baseline, the C-statistic in this population was 0.838 (95%CI: 0.793–0.883), which increased to 0.847 (95%CI: 0.792–0.902) after 1 year of canagliflozin treatment (P for difference<0.001).

Conclusion

Higher NT-proBNP levels, higher TnT levels, and higher BMI were independent and externally validated predictors of new-onset HF in T2D patients, including those treated with SGLT2 inhibitors. Other significant predictors in the derivation cohort were age, calcium channel blocker use, urine albumin-to-creatinine ratio, HbA1c, and hematocrit.

Find this article online at Diabetes Obes Metab.

References

1. Berg DD, Wiviott SD, Scirica BM, et al. A biomarker-based score for risk of hospitalization for heart failure in patients with diabetes. Diabetes Care. 2021;44(11):2573-2581. Available from. https://diabetesjournals.org/care/article/44/11/2573/138510/ABiomarker-Based-Score-for-Risk-of
2. Pandey A, Vaduganathan M, Patel KV, et al. Biomarker-based risk prediction of incident heart failure in pre-diabetes and diabetes. JACC Hear Fail. 2021;9(3):215-223. Available from. https://linkinghub.elsevier.com/retrieve/pii/S2213177920305916
3. Segar MW, Patel KV, Hellkamp AS, et al. Validation of the WATCH-DM and TRS-HF DM risk scores to predict the risk of incident hospitalization for heart failure among adults with type 2 diabetes: a multicohort analysis. J Am Heart Assoc. 2022;11(11):e024094. Available from. https://www.ahajournals.org/doi/10.1161/JAHA.121.024094
4. Lin Y, Shao H, Shi L, Anderson AH, Fonseca V. Predicting incident heart failure among patients with type 2 diabetes mellitus: the DM-CURE risk score. Diabetes Obes Metab. 2022;24(11):2203-2211. Available from. https://dom-pubs.pericles-prod.literatumonline.com/doi/10.1111/dom.14806
5. Segar MW, Vaduganathan M, Patel KV, et al. Machine learning to predict the risk of incident heart failure hospitalization among patients with diabetes: the WATCH-DM risk score. Diabetes Care. 2019;42(12):2298-2306. Available from. https://diabetesjournals.org/care/article/42/12/2298/36259/Machine-Learning-to-Predict-the-Risk-of-Incident
6. Parving H-H, Brenner BM, McMurray JJV, et al. Aliskiren trial in type 2 diabetes using cardio-renal endpoints (ALTITUDE): rationale and study design. Nephrol Dial Transplant. 2009;24(5):1663-1671. Available from. https://academic.oup.com/ndt/article-lookup/doi/10.1093/ndt/gfn721
7. Neal B, Perkovic V, Mahaffey KW, et al. Canagliflozin and cardiovascular and renal events in type 2 diabetes. N Engl J Med. 2017;377(7):644-657. Available from. http://www.ncbi.nlm.nih.gov/pubmed/28605608