Impact of glycemic control on the progression of aortic stenosis: a single-center cohort study using a common data model

This study utilized Observational Health Data Sciences and Informatics (OHDSI) open source software and the Observational Medical Outcomes Partnership (OMOP)-CDM version 5.3. The CDM comprised de-identified patient-level EHR data from outpatients and inpatients was obtained between April 2003 and July 2019, routinely collected during medical services at Seoul National University Bundang Hospital. The study protocol was approved by our institutional review board (B-1812–510-111) on December 26, 2018, which waived the requirement of informed consent. All aspects of the study were conducted in accordance with the Declaration of Helsinki.

From the CDM database, we extracted patients with mild (defined as a Vpeak/] of 2.0–3.0 m/sec) or moderate AS (defined as a Vpeak of 3.0–4.0 m/sec) at baseline, and for whom follow-up echocardiography was performed at an interval of ≥ 6 months (n = 5,764) (Fig. 1). Considering the intermittent changes in the antidiabetic medications and the fluctuations of HbA1c levels, we selected the first 2 echocardiograms (baseline showing mild to moderate AS, and follow-up at an interval of ≥ 6 months), even in patients who underwent three or more echocardiographic examinations.

The exclusion criteria were as follows: (1) age < 19 years; (2) open heart surgery or transcatheter valve implantation before or during follow-up; (3) longest interval between the paired echocardiogram evaluations was less than 6 months; and (4) unavailable glycated hemoglobin (HbA1c) data (Fig. 1). After applying these exclusion criteria, a total of 1,364 patients were included in the final analysis.

In the present study, the severity of AS was defined using the Vpeak, for which the Logical Observation Identifiers Names and Codes (LOINC) was used to represent the echocardiographic finding in the CDM (Supplementary Table S1). For the exclusion of patients who underwent aortic valve replacement, the Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT) were used to define the procedure. Other echocardiographic parameters, including left ventricular (LV) end-diastolic and end-systolic dimensions, LV end-diastolic and end-systolic volumes, LV mass index, the mean pressure gradient across the AV (meanPG), and the AV area (AVA) were also mapped to the LOINC concepts in the CDM (Supplementary Table S1).

The overall attrition diagram and a detailed list of concept sets and International Classification of Disease (ICD) codes used for constructing target definitions and exclusion criteria are provided in Supplementary Table S1. Patients with DM were designated by the SNOMED-CT, current use of antidiabetic medication, and a fasting glucose ≥ 126 mg/dL or HbA1c ≥ 6.5%. “Well-controlled DM” was defined as a mean HbA1c level < 7.0% during the study period, and “poorly-controlled DM” was defined as a mean HbA1c level ≥ 7.0% during the study period. Patients meeting the above eligibility criteria and definitions were identified using ATLAS version 2.8.0 and Structured Query Language (SQL) codes; outliers, as determined by R programming version 4.0.3 (The R Foundation for Statistical Computing, Vienna, Austria), were excluded. The presence of hypertension, dyslipidemia, heart failure, and atrial fibrillation was defined using the SMONED-CT for CDM condition data, and the use of medication was defined using RxNorm, RxNorm Extension, and Anatomical Therapeutic Chemical Classification System (ATC) for CDM drug exposure data.

All echocardiographic images were obtained using a standard ultrasound machine with a 2.5-MHz probe. Standard techniques were used to obtain M-mode, 2-dimensional, and Doppler measurements, in accordance with the American Society of Echocardiography’s guidelines [10, 11]. The Vpeak was recorded using the apical, right parasternal, or suprasternal window yielding the highest velocity signal. The AV meanPG was calculated using a modified Bernoulli equation, and the AVA was estimated from the continuity equation using the left ventricular outflow tract diameter and flow velocity.

The primary outcome was the AS progression rate, calculated as the annualized change in the Vpeak (△Vpeak/year). Annualized changes in the meanPG (△meanPG/year) and AVA (△AVA/year) were also considered; however, due to missing AVA values in the dataset, the change in AVA was not assessed as a study outcome.

In order to evaluate the impact of the presence of DM and the degree of glycemic control on the AS progression rate, the study population was divided into three groups: patients without DM (no DM; n = 1,027), patients with well-controlled DM (HbA1c < 7.0%; n = 193), and patients with poorly-controlled DM (HbA1c ≥ 7.0%: n = 144).

Continuous variables are presented as the median with interquartile range (IQR) after testing for normality with Shapiro–Wilk test, and categorical variables as frequencies. Differences between groups were evaluated using the Kruskal–Wallis test or one-way analysis of variance (ANOVA) with Bonferroni correction for multiple comparisons for continuous variables, and the χ² test for categorical variables. Bonferroni correction was performed to compare between the three groups as a post hoc test if the ANOVA test result was significant, in order to avoid bias of multiple testing.

Trends in the AS progression rate (△Vpeak/year) according to the presence of DM and degree of glycemic control, as well as the baseline Vpeak, were assessed using the penalized smoothing spline methods. The association between the degree of glycemic control and AS progression rate (△Vpeak/year) was assessed using linear regression modeling, using the △Vpeak/year as a dependent variable. Because the value of the dependent variable was numerically small, the dependent variable (△Vpeak/year) was analyzed in units of ‘cm/sec/year’. Demographic factors, laboratory findings including the mean HbA1c level during follow-up, and echocardiographic findings were used as independent variables. Because of the number of variables used in the univariable linear regression analysis was very large compared to the number of stud population, and there is possible multicollinearity between the variables, variables with a P-value of < 0.1 on univariable analysis were entered into the multivariable regression analysis, using the stepwise backward elimination with the Akaike Information Criterion (AIC) method [12, 13]. Additionally, as the highest △Vpeak/year tertile among patients with DM in the present study was 0.2 m/sec/year, patients with △Vpeak/year values > 0.2 m/sec/year were considered to indicate accelerated AS progression. The association between the degree of glycemic control and accelerated AS progression (the dependent variable; defined as △Vpeak/year values > 0.2 m/sec/year) was analyzed using a single-level logistic regression modeling, using demographic factors, laboratory findings and echocardiographic findings as independent variables. Variables with a P-value of < 0.1 on univariable analysis were entered into the multivariable regression analysis, using the stepwise backward elimination with the AIC method [12, 13].

Baseline characteristics, laboratory findings, and echocardiography data of the total study population are summarized in Table 1. The median age was 74 (IQR 65–80) years and 47% were male. Compared to those without DM (group 1), the patients with DM had higher frequencies of comorbidities, which were more frequent in those with poorly-controlled DM (group 3) than in those with well-controlled DM (group 2).

The median HbA1c levels were 5.8% (IQR 5.4–6.1) in patients without DM, 6.4% (IQR 5.9–6.8) in patients with well-controlled DM, and 7.6% (IQR 7.0–8.6) in patients with poorly-controlled DM. At baseline, the median Vpeak, meanPG, and AVA were 2.5 m/sec (IQR 2.2–2.9), 13.1 mmHg (IQR 10.2–19.4), and 1.3 cm² (IQR 1.1–1.6) in total study population, and the severity of AS was similar between the 3 groups.

The rate of AS progression was different between the groups: the △Vpeak/year was 0.06 m/sec/year (IQR -0.04–0.23) in patients without DM, 0.07 m/sec/year (IQR 0.00–0.21) in patients with well-controlled DM, and 0.09 m/sec/year (IQR 0.00–0.27) in those with poorly-controlled DM (overall p = 0.015) (Table 1). The rate of AS progression assessed by the changes in the meanPG and AVA showed similar trends of a lower AS progression rate in patients without DM, and a higher AS progression rate in patients with DM, especially in those with poorly-controlled DM.

In order to evaluate the overall impact of glycemic control on the rate of AS progression, the △Vpeak/year and △meanPG/year were plotted according to the mean HbA1c levels in Fig. 2. The AS progression rate was proportional to the degree of glycemic control, showing a more rapid progression in those with a higher HbA1c level. The association between the degree of glycemic control and the rate of AS progression (△Vpeak/year; cm/sec/year) was further assessed in linear regression analyses (Table 2 and Supplementary Table S2). In the multivariable regression model, the higher mean HbA1c level during follow-up (β = 2.620; 95% CI 0.732–4.507; p = 0.007), higher Vpeak at baseline (β = 5.574; 95% CI 1.329–9.818; p = 0.010), higher total cholesterol level at baseline (β = 0.063; 95% CI 0.005–0.120; p = 0.034), and the presence of coronary artery disease (β = 6.716; 95% CI 1.344–12.088; p = 0.014) were significantly associated with the higher rate of AS progression; whereas the presence of heart failure was negatively associated with the rate of AS progression (Table 2).

Considering that the baseline Vpeak was significantly associated with the rate of AS progression, we compared the rate of AS progression between the patients without DM, those with well-controlled DM, and those with poorly-controlled DM, across the range of baseline Vpeak (Fig. 3). As the baseline Vpeak increases, the rate of AS progression rate showed a tendency to gradually increase (Fig. 3A). Of note, compared to those without DM, the increase in the AS progression rate according to the baseline Vpeak value showed a steeper pattern in patients with DM, and the slope was the steepest in those with poorly-controlled DM. The same trend was observed, when the AS progression rate was assessed using the changes in the meanPG (Fig. 3B).

Using the △Vpeak/year value of > 0.2 m/sec/year as the cutoff for accelerated AS progression, we performed logistic regression analyses to investigate the association between the glycemic control and the accelerated AS progression (Table 3 and Supplementary Table S3). In the multivariable logistic regression analysis, a + 1%-unit increase in the mean HbA1c level was associated with a 27% higher risk of accelerated AS progression (adjusted OR = 1.267 per + 1%-unit increase in HbA1c; 95% CI 1.106–1.453; p < 0.001). This association maintained its statistical significance when the glycemic control level was used as a categorical variable (mean HBA1c ≥ 7.0%) (adjusted OR = 1.524; 95% CI 1.010–2.285; p = 0.043).