Impact of implementing Dutch versus European guideline risk factor targets in older patients with ischaemic heart disease

We used data from the PHARMO Database Network, a population-based network of electronic healthcare databases that combines ongoing anonymous data collection from different primary and secondary healthcare settings in the Netherlands. The longitudinal nature of the PHARMO Database Network enables follow-up of 7 million patient journeys through different healthcare settings, via linkage of national databases and registries. Study populations are created from the Database Network using the data sources needed to address specific study objectives. In general, the population included in the PHARMO Database Network is representative of the Dutch population with respect to age and gender, but the database contains a relatively large proportion of individuals > 80 years, possibly due to an increased use of healthcare in older age [9].

For the current analysis, healthcare settings included in the data extraction file from the PHARMO Database Network were general practices, clinical laboratories and hospitals. A retrospective cohort was extracted that included patients aged 71–80 years who were hospitalised for IHD in 2017, 2018 or 2019. For the primary analysis, we excluded patients with missing LDL‑c and/or SBP measurements. The use of the study-specific dataset from the PHARMO Database Network was controlled by an independent privacy and governance board, the Compliance Committee [9].

We analysed data as of 1 January 2017, with individual index dates defined as the first hospitalisation with a primary diagnosis of IHD. Follow-up was until death or end of study (31 December 2020). We extracted individual-level data from the general practitioner’s notes on the patient’s age, sex, International Classification of Primary Care (ICPC) codes for medical history, smoking status, body mass index (BMI) and blood pressure measurements. Furthermore, we collected information on recurrent hospitalisations and outpatient clinic visits with corresponding primary diagnosis and laboratory results. Medical history was defined using ICPC codes and/or International Classification of Diseases (ICD)–10 codes for primary diagnosis of outpatient clinic visits and hospitalisations.

IHD was defined as ICD-10 codes I20–I25 and ICD‑9 codes 410–414, with the exception of the following ICD-10 and ICD‑9 codes for non-atherosclerotic disease: I201 (coronary spasm), I240 (thrombosis not resulting in myocardial infarction), I241 (Dressler’s syndrome), I254 and 414.10/414.11 (coronary aneurysm or dissection) and I255 (ischaemic cardiomyopathy).

The European LDL‑c target was defined as < 1.4 mmol/l according to the 2019 ESC/EAS Guidelines for the management of dyslipidaemias, [4] and the European SBP target was defined as < 130 mm Hg according to the 2021 ESC Guidelines on CVD prevention in clinical practice [5]. The Dutch LDL‑c and SBP targets were defined as < 2.6 mmol/l and < 140 mm Hg, respectively, according to the 2018 Dutch guidelines on cardiovascular risk management [6].

The primary outcome was the potential benefit (expressed as gain in cardiovascular event-free years) of reaching LDL‑c and SBP targets according to Dutch versus European prevention guidelines. Secondary outcomes included the potential decrease of 10-year and lifetime risks of major recurrent CV events and the percentage of patients with LDL‑c and SBP measurements who reached LDL‑c and SBP targets according to Dutch versus European guidelines.

Patient characteristics of the study cohort and of patients excluded due to missing values were compared to investigate potential healthy participant selection bias using t-tests for continuous variables, Wilcoxon-rank sum tests for non-continuous variables and Fisher exact tests for categorical variables, with a two-sided p-value < 0.005 considered to be statistically significantly different.

To determine the plausibility of genuine missing values for LDL‑c and SBP, instead of being a result of limited data availability, we examined the presence of other laboratory results pertaining to these patients. First, the percentage of patients reaching LDL‑c and SBP targets according to Dutch versus European guidelines was calculated. Second, 10-year and lifetime (i.e. until the age of 90 years) risks of recurrent myocardial infarction, stroke or vascular death were estimated using the Fine and Grey SMART-REACH model—a competing risks prediction model used for secondary prevention—based on the following CVD risk factors: age, sex, last smoking status, diabetes mellitus, number of CVD manifestations (coronary artery disease, cerebrovascular disease and/or peripheral artery disease), atrial fibrillation, heart failure (HF), geographical region (the Netherlands), and lowest systolic blood pressure, total cholesterol level and creatinine level available. Total cholesterol was calculated as follows: lowest available LDL-c + high-density lipoprotein cholesterol +0.2*triglycerides. Third, the effect of reaching the LDL‑c or SBP target was calculated by applying meta-analysis—derived hazard ratios to each individual’s estimated cardiovascular risk [10, 11]. For every 10-mm Hg reduction of the patient’s current SBP, the SMART-REACH model assumes a hazard ratio of 0.80. [10] For every mmol/l reduction of the patient’s current LDL‑c level, a hazard ratio of 0.78 was applied [11]. Estimates of risk and treatment effect were performed for the overall cohort and for subgroups that did not reach Dutch or European targets for both LDL‑c and SBP.

All analyses were stratified by sex. Missing risk factors needed for the SMART-REACH risk estimation were imputed using mean/median imputation, which has been shown to be a robust method for dealing with missing values in CVD prediction models [12]. Numbers of missing predictors, imputation values and a comparison of levels in complete cases versus the imputed cohort can be found in Table S1 in the Electronic Supplementary Material. A sensitivity analysis was performed for complete cases. All statistical analyses were performed using R statistical software (version 4.1.3).

Of the 3003 patients hospitalised for IHD from 2017 through 2019, 1186 (39%) had missing LDL‑c and/or SBP measurements, of whom 523 (17% of the overall cohort) missed both values. In total, 1817 patients (36% women) for whom both measurements were available were eligible for primary outcome analysis (Fig. 2). Median age during index hospitalisation was 74 years (interquartile range (IQR): 72–77), and median follow-up duration was 33 months (IQR: 24–41), with 52 patients dying before the end of study period.

Tab. 1 compares characteristics of patients without SBP and/or LDL‑c measurements (n = 1186) to those included in the primary outcome analysis. Patients with missing values were more likely to have a first diagnosis of acute coronary syndrome (50% vs 42%) or a history of HF (19% vs 13%), and they had a shorter median follow-up period (25 vs 33 months) due to a higher mortality rate (10% vs 3%). For 96% of the patients with missing LDL‑c and/or SBP values, other laboratory results were available during follow-up. In this group, 245 (21%) had an LDL‑c measurement (mean ± standard deviation (SD): 1.9 ± 0.8 mmol/l) (but no SBP measurement), of whom 80% met the Dutch target and 22% met the European target. For SBP, 418 (35%) had a measurement (mean ± SD: 129 ± 21 mm Hg) (but no LDL‑c measurement), of whom 69% met Dutch and 52% met European targets.

Distributions of LDL‑c and SBP levels are shown in Fig. 3. Mean ± SD SBP was 125 ± 16 mm Hg, and median LDL‑c level was 1.8 mmol/l (IQR: 1.4–2.2). Dutch targets for LDL‑c and SBP were both reached by 84%, whereas European targets were reached by 23% and 61%, respectively (Tab. 2). With these risk factor levels, median 10-year and lifetime risks for recurrent events were 29% (IQR: 24–36%) and 39% (IQR: 35–46%), respectively, for the overall population.

On top of current risk levels, reaching Dutch or European targets would result in a gain of median 0.0 (IQR: 0.0–0.1) and 0.5 (IQR: 0.2–1.2) event-free years, respectively. In 1281 patients reaching Dutch but not European targets, the additional effect of reaching European targets was 0.6 years (IQR: 0.3–1.0). The greatest effect could be reached in patients who currently did not meet Dutch targets for both LDL‑c and SBP (n = 501), i.e. a gain of median 0.6 (IQR: 0.2–1.2) and 1.7 (IQR: 1.2–2.5) event-free years by reaching Dutch versus European targets, respectively (Fig. 4).

Mean SBP levels and targets were comparable for both sexes. However, LDL‑c levels were higher in women compared with men (median (IQR): 1.9 (1.5–2.4) vs 1.7 (1.4–2.1); p < 0.01), and therefore fewer women reached Dutch (77% vs 88%; p < 0.01) and European (19% vs 25%; p = 0.02) LDL‑c targets (Tab. 2). Furthermore, women had a longer event-free life expectancy (median (IQR): 86 (84–87) vs 83 (82–84) years; p < 0.01). Consequently, the potential lifetime benefit of (strict) risk factor management was estimated to be greater in women than men (p < 0.01 for all comparisons) (Tab. 2).

There are several strengths to our study. First, we analysed a large, real-world database of healthcare settings, which included general practices, clinical laboratories and hospitals [9]. The data linkage in the PHARMO Database Network provides a unique opportunity to fill gaps in patient journeys and investigate real-world treatment levels, which cannot be addressed in most single (unlinked) databases. Second, as the PHARMO Database Network follows patients regardless of age and sex or other in- or exclusion criteria, our study had minimal selection bias. Finally, we calculated risk estimations using the innovative and validated SMART-REACH model, which supports treatment (goal) decisions with outcomes that are relevant to clinical practice.

Some aspects of our study deserve consideration. First, our primary outcome was not observed mortality and morbidity but contained an estimate, with inherent uncertainty thereof. Second, we were limited to including patients from the PHARMO Database Network for whom all required data linkages were accessible, thus restricting our sample size. Furthermore, we had to exclude patients without LDL‑c and SBP measurements from our primary outcome analysis, which may have introduced healthy participant selection bias. However, we have provided insight into this in Tab. 1. Finally, we did not include information on the type of medication prescribed and dispensed, which could have provided insight into treatment effects and unavoidable or justifiable causes of unmet treatment targets.