Inclusion criteria were patients aged ≥ 18 years recruited consecutively at cardiology clinics, at very high CV risk based on the European Society of Cardiology11 defined as any one of the following: a) documented CV disease and at least 6 months since the last CV event; b) type 2 diabetes mellitus or type 1 diabetes mellitus with target organ damage; c) moderate to severe chronic kidney disease, or d) risk score > 10%. Another criterion was blood test results in the 3 months prior to recruitment. Patients were excluded if they had cancer or other disease that could confound study results.

We calculated the sample size from a previous study,8 in which 31.3% of high-risk patients achieved therapeutic targets. Using a binomial distribution, we estimated a sample size of 1360 patients, with ± 2.5% accuracy, to identify the percentage of patients at very high CV risk who achieve the therapeutic targets for LDL-C, with a 95% confidence interval (95%CI). Assuming that 2.5% of patients would not be evaluable, we therefore needed to recruit 1395 patients.

A total of 140 cardiologists were selected randomly from all the Spanish autonomous communities (Appendix). Each cardiologist was responsible for recruiting 10 consecutive patients. The recruitment period was from November 6, 2013 to July 31, 2014. The investigators collected study data in an electronic case report form. The data from the case report forms were entered in a database with internal consistency rules and ranges to ensure data quality, and data cleaning was performed.

We grouped patients by intensity of lipid-lowering therapy2: a) low-intensity (no treatment or daily dose of simvastatin 10 mg, pravastatin 10 to 20 mg, lovastatin 20 mg, fluvastatin 20 to 40 mg, pitavastatin 1 mg or ezetimibe 10 mg alone); b) moderate-intensity (atorvastatin 10-20 mg or rosuvastatin 5-10 mg, simvastatin 20-40 mg, pravastatin 40-80 mg, lovastatin 40 mg, fluvastatin XL 80 mg or pitavastatin 2-4 mg, or a low-intensity statin plus ezetimibe), and c) high-intensity (atorvastatin 40-80 mg, rosuvastatin 10-40 mg or any moderate-intensity statin plus ezetimibe).

The study primary endpoint was “adequate lipid control”. Following the European guideline criteria for CV prevention,1 we defined adequate control as LDL-C < 70mg/dL and divided the study population into 2 groups: LDL-C < 70mg/dL and LDL-C ≥ 70mg/dL.

Under the healthy habits section, alcohol consumption was defined in units (1 unit=1 beer, 1 glass of wine, 1/2 shot of liqueur or 1/2 whisky). More than 2 units/d was considered as excessive. Physical exercise was defined as walking for at least 30minutes per day or more than 2 sports sessions per week.

A short dietary intake questionnaire was used to analyze adherence to a Mediterranean diet.12

“Treatment after first visit” was reported as a variable to identify whether cholesterol-lowering therapy was up titrated.

The study was designed with 2 cross-sectional time points: at inclusion and at 1 year of follow up. This article refers to the first point only, because the second is still under way.

The sample was described using absolute and relative frequencies for dichotomous variables, and mean±standard deviation or median [interquartile range] for continuous variables, depending on whether or not they followed a normal distribution. For the between-group comparison (LDL-C < 70mg/dL and LDL-C ≥ 70mg/dL, subdivided into up titration and no titration), we used the chi-square test for categorical variables and the Student t test for continuous variables.

We analyzed independent predictors of therapeutic target achievement (primary endpoint) using a binary logistic regression model and LDL-C < 70mg/dL as the dependent variable. Our initial model included all variables showing a statistically significant association with the dependent variable in the bivariate analysis. Statistical significance was defined as P < .1. We used backward stepwise elimination and P entrance/exit tolerances of < .05 and > .1, respectively (with automated variable selection), to progressively eliminate variables until the model included only potential predictors of LDL-C < 70mg/dL with a statistical significance of P<.05. Potential effect modifiers were assessed using first-order interactions.

We then used the same procedure to analyze independent predictors of up titration after the first visit (increased intensity of the cholesterol-lowering therapy) in patients with LDL-C ≥ 70 mg/dL.

Finally, we examined variability by Spanish autonomous community in the prescription of high-intensity cholesterol-lowering therapy after the first visit, and whether such variability could be explained by the characteristics of the patients seen in each autonomous community or by the autonomous community size, expressed as total population. This analysis was performed only in autonomous communities that had enrolled at least 20 patients. We constructed a 3-step multilevel regression model13: in the first step we included a random constant only, to measure inter-community variability in patient rates with high-intensity cholesterol-lowering therapy. Second, we added various individual patient characteristics, to investigate whether the prescription differences by community could be explained by the characteristics of the patients seen in each community. We tested all the baseline characteristics that showed between-group differences with a statistical significance of P<.1. If autonomous community variability in the rate of patients with high-intensity cholesterol-lowering therapy fell to zero after we had adjusted by patient characteristics, variability would be due solely to the differences in the patients seen. Finally, in the third step we included the total population for each autonomous community. We estimated odds ratio (OR) as a measure of association. The multilevel regression models were constructed assuming independent covariance calculated with the R statistical package, version 0.98.953.

To measure the change in variability by autonomous community, at each step we calculated the percentage change in variance among autonomous communities in the most complex model vs the simplest. To measure the magnitude of the variance among autonomous communities, we estimated the intraclass correlation coefficient (ICC) and median OR (MOR). The ICC can be interpreted as the proportion of total variance in a selected variable that can be attributed to differences among autonomous communities. The MOR is defined as the median value of the estimated OR in the autonomous communities at “highest risk” and “lowest risk” after randomly selecting 2 autonomous communities repeated times. In this study, the MOR shows the extent to which the individual probability of receiving high-intensity cholesterol-lowering therapy is determined by the autonomous community where the patient is seen. A MOR equal to 1 would mean that there are no differences between autonomous communities in prescription rates. A MOR significantly greater than 1 would mean that there is some characteristic of the autonomous community that is relevant for explaining variations in the individual probability of receiving high-intensity cholesterol-lowering therapy, ie, some variability among autonomous communities remains unexplained. We used a 95%CI for ICC estimations and the Bayesian approach for MORs.

Analyses were calculated with the statistical packages SPSS 13.0 (Chicago, Illinois, United States) and R, version 0.98.953.

The most relevant serum lipid values for the entire study population (Table 1) were total cholesterol, 175±46 mg/dL; high-density lipoprotein cholesterol, 46±13 mg/dL; LDL-C, 94±44 mg/dL; triglycerides, 138±72 mg/dL; glycated hemoglobin, 6.4%±1.1%; ultrasensitive C-reactive protein, 2.5±4.2 mg/dL, and creatinine clearance (Cockroft), 87.2±58.5mL/min/1.73 m2.

Figure 1 shows the use of cholesterol-lowering agents by active substance. The most commonly-used drug was atorvastatin (47%), followed by rosuvastatin (20%) and simvastatin (19%). Among nonstatin cholesterol-lowering agents, ezetimibe was prescribed to 14% of the patients. A small number of patients, 4.7%, did not receive any cholesterol-lowering agent.

Figure 1.

Distribution of the prescription of cholesterol-lowering agents in the study patient population.

(0.14MB).

By therapy intensity, 45%, 45%, and 10% of patients received high-, moderate-, and low-intensity cholesterol-lowering therapy, respectively.

In the baseline blood tests for the entire study set, 279 patients (26%) had adequate LDL-C control, while the remaining 776 (74%) had LDL-C ≥ 70mg/dL. The clinical variables associated with LDL-C < 70 mg/dL are shown in Table 2. The following variables were of particular note: diabetes mellitus, coronary revascularization, better blood pressure control, lower heart rate, and, above all, high-intensity cholesterol-lowering therapy. In contrast, active smoking, a family history of dyslipidemia and concomitant carotid artery disease were associated with poor lipid control (LDL-C ≥ 70mg/dL). Table 2 also lists diet and exercise factors, and shows an association between good lipid control and gentle daily exercise, fruit, and wholegrain bread intake.

The independent predictors of LDL-C < 70 mg/dL in the multivariable analysis are shown in Table 3. The highest OR was associated with high-intensity cholesterol-lowering therapy. Lipid control was also better among diabetic patients, patients with diets containing less white bread and/or rice, and, finally, patients with a shorter history of dyslipidemia.

We analyzed prescription changes that physicians made for the 776 patients who had LDL-C ≥ 70mg/dL. No change was made in 70% of patients. Therapy intensity was increased in 26% and was reduced in 3%. The main variables associated with up titration of therapy were low-strength therapy at baseline (OR=5.05; 95%CI, 3.3-9.2) and the LDL-C value itself (OR=1.2; 95%CI, 1.01-1.02). Specifically, of the patients with baseline LDL-C ≥ 70 mg/dL, an LDL-C value of 140±43 mg/dL was found among patients whose therapy was up titrated vs a value of 104±36 mg/dL among those whose therapy remained unchanged (P<.001). We also analyzed physician characteristics and found a more proactive attitude to up titration among older physicians (P=.02).

Figure 2 shows a major difference in baseline LDL-C values by autonomous community. We also found differences when we investigated variability among autonomous communities in high-intensity cholesterol-lowering therapy prescription rates (Figure 3). In the unadjusted multilevel model, this variability was 75%, with ICC=0.14 (95%CI, 0.05-0.32) and MOR=4.02 (95%CI, 1.22-6.65) (Table 4). Various baseline patient characteristics were associated with a higher prescription rate of high-intensity cholesterol-lowering therapy: age, hypertension, smoking, history of myocardial infarction, and baseline LDL-C. Variability among autonomous communities remained unchanged after adjustment for these variables (ICC=0.14; 95%CI, 0.05-0.31; MOR=2.45; 95%CI, 1.21-6.4), showing there must be another factor unrelated to patient profiles that would explain this prescription rate. Finally, we adjusted by the “global population” variable by autonomous communities, and found that variability persisted (ICC=0.16; 95%CI, 0.05-0.35; MOR=3.28; 95%CI, 1.25-9.41), suggesting variability in the care process unexplained by individual patient characteristics or autonomous community population size.

Figure 2.

Low-density lipoprotein cholesterol values (mean±standard deviation) of study patients with coronary heart disease based on laboratory test results before the baseline visit by autonomous community of residence. LDL-C, low-density lipoprotein cholesterol.

(0.05MB).

Figure 3.

Variability in the prescription of cholesterol-lowering agents by Spanish autonomous community. Data collected at the baseline visit on therapy received until then.

(0.16MB).

Although we instructed investigators to recruit patients consecutively to the study, as specified in the protocol, we did not check whether this requirement was fulfilled, and therefore we cannot rule out selection bias. Also, investigators were not selected or stratified at random by care level. We tried to ensure that all care levels were represented, but we cannot guarantee they were represented equally.