Derivation And Validation Of A Clinical Risk Model For Subclinical Heart Failure In Type 2 Diabetes Mellitus

Presenting Author(s)

• 

  Alice C. Cowley, MB

  Clinical Research Fellow
  University of Leicester
  Leicester, England, United Kingdom

Primary Author(s)

• 

  Alice C. Cowley, MB

  Clinical Research Fellow
  University of Leicester
  Leicester, England, United Kingdom

Co-Author(s)

• GG

  Gaurav Gulsin, PhD

  Clinical Lecturer
  University of Leicester
  Leicester, United Kingdom
• AD

  Abhishek Dattani, MBBS

  Clinical Research Fellow
  University of Leicester
  Leicester, United Kingdom
• JY

  Jian L. Yeo, MD

  Clinical Research Fellow
  University of Leicester, United Kingdom
• JB

  Joanna Bilak, PhD

  Cardiology registrar
  University of Leicester
  Leicester, United Kingdom
• TY

  Thomas Yates

  University of Leicester, United Kingdom
• MD

  Melanie Davies, MD

  Professor of Diabetes Medicine
  University of Leicester
  Groby Road, United Kingdom
• GM

  Gerry P. McCann, MD

  Professor
  University of Leicester
  Leicester, England, United Kingdom
• EB

  Emer M. Brady, PhD, BSc

  Post doctoral research fellow
  University of Leicester
  Leicester, England, United Kingdom

Background: Screening for asymptomatic left ventricular (LV) dysfunction is strongly advocated in type 2 diabetes (T2D), due to robust associations with incident heart failure. We hypothesized that a clinical risk prediction model would have good discrimination for identifying reduced left ventricular (LV) global longitudinal strain (GLS) in adults with T2D and no prevalent cardiovascular disease.

Methods: Adults with T2D and no prior history, clinical symptoms or signs of CV disease were prospectively recruited to a single-centre (NCT03132129). Subjects underwent comprehensive phenotyping including cardiac MRI with feature tracking strain for GLS. A priori, a total of 13 candidate predictors were pre-selected based on expert clinical consensus and prevailing evidence: age, sex, ethnicity, body mass index, history of hypertension, blood pressure, heart rate, glycosylated haemoglobin, diabetes medications, diabetes duration, urinary albumin: creatinine ratio and estimated glomerular filtration rate. Automated backwards selection was performed on the original and each imputed dataset (n=20) to create a risk prediction model. Assuming a prevalence of reduced GLS of 50%, the minimum sample size required for new model development with up to 13 candidate predictor parameters for potential inclusion, an expected R of 30%, and 90% power would be 385 at alpha=0.05 to detect an area under the curve (AUC) >0.7. The derived model was internally validated using bootstrapping and performance statistics adjusted for optimism. The final model was externally validated, and performance statistic reported (c-index).

Results: Model derivation cohort included 530 T2D (61% male; age 63 ± 7 years) and external validation cohort included 154 T2D (58% male; age 57 ± 10 years). Reduced GLS (< 16.6% based on UK Biobank) was present in 54% (n=285, table 1) of the derivation and in 36% of the validation cohort(n=56). Eight predictors remained in the final model: age, sex, ethnicity, systolic and diastolic blood pressure, heart rate and diabetes medications with a c-index of 0.760 (0.730, 0.791); adjusted to 0.750 (0.751, 0.779). Good discriminatory power was retained on external validation (c-index of 0.737 [0.658, 0.815]) (Figure 1).

Conclusion: A high proportion of asymptomatic adults with T2D have reduced GLS, and can be identified using a simple clinical risk prediction model with good accuracy.

Characteristics of the groups defined by GLS threshold

Characteristic	GLS < 16.6 %n=285 (54%)	GLS ≥ 16.6%n=245 (46%)	p-value
Demographics
Age, years	63 ± 7	64 ± 7	0.020
Sex (male)	205 (72%)	116 (47%)	< 0.001
Ethnicity
White-European	203 (71%)	155 (63%)	0.143
South Asian	67 (24%)	75 (31%)
Other	15 (5%)	15 (6%)
Diabetes Duration, years	10 (5,15)	9 (5,15)	0.605
HbA1c, %	7.4 ± 1.3	7.2 ± 1.2	0.040
Cardiac Investigation
LV mass/height, g/m	70.2 ± 14.7	63.9 ± 12.1	< 0.001
LV mass:volume, g/mL	0.93 ± 0.16	0.89 ± 0.12	0.004
LV ejection fraction, %	64.1 ± 7.2	70.0 ± 5.9	< 0.001
GLS, %	14.6 ± 1.6	18.4 ± 1.3	< 0.001
LA volume/height, mL/m	34.2 ± 11.7	35.1 ± 10.6	0.322
ECV, %	26.8 ± 2.8	27.3 ± 2.9	0.033
LGE (yes)	67 (24%)	41 (17%)	0.046
MPR*	2.87 ± 0.84	2.67 ± 0.75	0.005
CT Calcium score	63.5 (0, 324)	33.5 (0, 271)	0.332
*Available n=482CT, computed tomography; ECV, extracellular volume; GLS, global longitudinal strain; HbA1c, glycosylated haemoglobin; LA, left atrial; LV left ventricular; LGE, late gadolinium enhancement; MPR, myocardial perfusion reserve

Performance statistics of the model predicting subclinically reduced GLS in T2D on internal and external validation