A LASSO-derived model for the prediction of nonattainment of target LDL-C reduction with PCSK9 inhibitors in patients with atherosclerotic cardiovascular disease

Abstract

Background

Proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors have demonstrated significant efficacy in lowering low-density lipoprotein cholesterol (LDL-C) levels in patients with atherosclerotic cardiovascular disease (ASCVD), but some fail to achieve the target levels. This study aimed to explore the potential risk factors associated with this nonattainment of target LDL-C reduction (NTR-LDLC) and develop a prediction model.

Methods

The population was randomly divided into derivation and verification subsets in a 7:3 ratio. Utilizing the Least Absolute Shrinkage and Selection Operator (LASSO) regression, we filtered the variables within the derivation set. Subsequently, we assessed the model's predictive accuracy for the NTR-LDLC in both subsets through the application of decision curve analysis (DCA) and the plotting of receiver operating characteristic (ROC) curves.

Results

The study enrolled 748 patients, with 115 individuals experiencing NTR-LDLC. Using LASSO regression, five significant predictive factors associated with NTR-LDLC were identified: statin therapy, diastolic blood pressure (DBP), alanine aminotransferase (ALT), total cholesterol (TC), and LDL-C. Based on these results, a nomogram prediction model was constructed and validated, showing predictive accuracy with the area under the ROC curve (AUC) of 0.718 (95% confidence interval [CI]: 0.657 − 0.779) and 0.703 (95% CI: 0.605 − 0.801) for the derivation and validation sets, respectively.

Conclusions

This study presents a LASSO-derived predictive model that can be used to predict the risk of NTR-LDLC with PCSK9 inhibitors in patients with ASCVD.

Introduction

The prevalence of cardiovascular disease (CVD) continues to increase, with mortality rates remaining at the forefront, posing a significant threat to human health and quality of life, thus representing a pressing public health issue [1, 2]. Atherosclerosis serves as the core pathological mechanism of CVD, wherein elevated levels of low-density lipoprotein cholesterol (LDL-C) rank as the second leading risk factor, following hypertension [3]. Stringent blood lipid management significantly reduces the risk of cardiovascular events [4]. However, it is noteworthy that the achievement rate for the LDL-C target among individuals diagnosed with atherosclerotic cardiovascular disease (ASCVD) remains relatively low, at merely 26.6% [5]. Further research indicates that only 21%-34% of ASCVD patients undergoing lipid-lowering therapy are able to achieve the desired LDL-C target levels [6]. Therefore, achieving the LDL-C target lipid-lowering level remains crucial in the treatment of ASCVD patients.

Proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors, a novel class of monoclonal antibodies, have demonstrated remarkable efficacy in lowering LDL-C levels through their unique mechanisms of action [7]. These agents can further reduce LDL-C by 40% to 60% in additon to existing lipid-lowering therapies, providing a new treatment option for patients who fail to achieve LDL-C targets, and potentially enabling them to reach lipid management goals [8]. However, we observed that some patients could not adequately achieve the target LDL-C levels after treatment with PCSK9 inhibitors.

Previous studies [6, 9, 10] have explored the primary reasons for nonattainment of target LDL-C reduction (NTR-LDLC), which include the failure to initiate and intensify treatment according to guidelines, medication non-adherence, and the limitation of additional treatments for high-risk patients due to insurance denial. However, there is currently no systematic research on the factors influencing NTR-LDLC after the treatment of PCSK9 inhibitors. Therefore, the purpose of this study was to explore the potential risk factors associated with non-target LDL-C reduction in ASCVD patients treated with PCSK9 inhibitors, establish a prediction model based on the basic information and clinical characteristics of patients, effectively identify high-risk patients who are prone to fail to achieve the LDL-C targets, and provide recommendations on the use of PCSK9 inhibitor in ASCVD patients.

Methods

Study population

Patients were retrospectively enrolled from May 2019 to May 2023, according to the following inclusion criteria: 1) age > 18 years, 2) diagnosed with ASCVD, 3) continuous treatment with PCSK9 inhibitors for a minimum duration of 3 months, 4) at least one follow-up blood lipid test during the stable phase (≥ 3 months). Patients were excluded if they had the following conditions: 1) lack of baseline level prior to PCSK9 inhibitor use, 2) missing medical record information. This study was approved by the Ethics Committee of Beijing Anzhen Hospital and was conducted in accordance with the Declaration of Helsinki. According to the lipid reduction targets outlined in the Guidelines [11], NTR-LDLC was defined as LDL-C ≥ 1.8 mmol/L in patients with extremely high risk ASCVD and ≥ 1.4 mmol/L in thoses with ultra-high risk ASCVD, following at least 3 months of PCSK9 inhibitors therapy.

Statistical analysis

Statistical analyses were carried out utilizing R software 4.3.2. The population was randomly divided into derivation and verification sets in a 7:3 ratio. The Kolmogorov–Smirnov test was applied to assess the normality of continuous data distributions. For non-normally distributed continuous data, the median and interquartile range (IQR) were reported, and the Mann–Whitney U-test was employed for analysis. Categorical data were expressed as counts and percentages, analyzed using either the Chi-squared or Fisher’s exact test. A significance level of P < 0.05 was established. Data were collected from 41 variables, including demographics, comorbidities, laboratory examination, and concomitant medications. Variables with missing values exceeding 20% were excluded, while the remaining missing values were imputed using average values or modes. In the multivariable analysis of the derivation set, all potential predictors of NTR-LDLC were included in a logistic regression model. The LASSO regression method, implemented through the "glmnet" package, was utilized for variable selection. NTR-LDLC was the dependent variable (coded as 1 for patients with NTR-LDLC and 0 for patients with TR-LDLC). Ten-fold cross-validation was used to select the penalty term lambda (λ). The prediction performance of the fitting model was assessed using binomial deviation, with Lambda + min being chosen to avoid overfitting of the model.To evaluate the model's performance, we employed DCA curves, confusion matrices, and receiver operating characteristic (ROC) curves. The predictive accuracy of the logistic regression model for NTR-LDLC was analyzed for both the derivation and verification sets. both the derivation and verification sets.

Results

Patient characteristics

A total of 748 patients were ultimately enrolled in this study, with 115 patients (15.4%) failing to achieve the LDL-C reduction target despite the use of PCSK9 inhibitors. Figure 1 shows the cohort selection process for this study. The median age of the entire cohort was 57 years (ranging from 23 to 87), and the majority (78.5%) were male. Hyperlipidemia (91.6%) was the most common comorbidity, followed by hypertension (60.3%) and type 2 diabetes (33.2%). The overall population was randomly assigned to a training set (523 cases) and a validation set (225 cases) in a 7:3 ratio. As shown in Table 1, no significant differences were observed in the baseline characteristics between the two groups (P < 0.05).

Fig. 1

Flowchart of cohort selection

Table 1 Baseline demographic and clinical characteristics of patients in the derivation set and the validation set

LASSO regression analysis

Since there were many variables and relatively few cases, LASSO regression was used to extract features. We utilized ten-fold cross-validation to select the penalty term. Figure 2 presents the variables obtained from the LASSO regression. Given that lambda + 1se keeps only one variable, we chose lambda to fit the data more accurately. Considering the stability of the model, the minimum value (lambda + min) has the best model performance and retention of the five variables.The results indicated that the risk factors predictive of non-achievement of LDL-C reduction targets included statins therapy, diastolic blood pressure (DBP), alanine aminotransferase (ALT), total cholesterol (TC), and LDL-C. Details are shown in Table 2. Based on these five predictive factors, a nomogram prediction model for the risk of NTR-LDLC after using PCSK9 inhibitors was constructed, as shown in Fig. 3.

Fig. 2

The Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to extract features. A The LASSO regression filtering variable process. B The dashed line on the left revealed that 5 variables were retained at the point with the smallest error, and the dashed line on the right revealed that only one variable was retained within one standard error (1se) of the minimum error

Table 2 LASSO analysis of risk factors associated with NTR-LDLC after CABG

Fig. 3

Nomogram prediction model for the risk of not achieving LDL-C reduction targets after using PCSK9 inhibitors. DBP, diastolic blood pressure; ALT, alanine aminotransferase; TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; NTR-LDLC, nonattainment of target LDL-C reduction

Model validation and calibration

Model validation and calibration were performed to assess the predictive ability and accuracy of the NTR-LDLC risk model. ROC curves were generated to visualize the model's performance, with AUC values of 0.718 (95% CI: 0.657–0.779) for the derivation set (Fig. 4A) and 0.703 (95% CI: 0.605–0.801) for the validation cohort (Fig. 4B). Receiver operating characteristic (ROC) curves were plotted (Fig. 4). Table 3 presents the confusion matrices for both sets, providing insights into the model's classification accuracy. Further, DCA analysis was performed to assess the clinical utility of the model, as depicted in Fig. 5. The model demonstrated a sensitivity of 72.8% in the derivation set and 67.6% in the validation set, while specificity was 60.0% and 68.1%, respectively, as shown in Table 4. Similarly, the specificity in the derivation set and the validation set was 60.0% and 68.1%, respectively. Calibration curves were also plotted to evaluate the agreement between predicted and observed NTR-LDLC outcomes, with good calibration observed in both the derivation (Fig. 6A) and validation sets (Fig. 6B). Collectively, these results indicate that the model possesses satisfactory predictive capabilities.

Fig. 4

Receiver operating characteristic (ROC) curves for the derivation set (A) and the validation set (B)

Table 3 Confusion matrix diagram in the derivation set and the validation set

Fig. 5

Decision curve analysis of the model

Table 4 Validation value of the derivation set and the validation set

Fig. 6

The calibration curves of the derivation set (A) and the verification set (B)

Discussion

Our study has uncovered a significant finding: despite treatment with PCSK9 inhibitors, a proportion of 15.4% of ASCVD patients remained unable to attain the desired LDL-C level. By a LASSO regression analysis, we identified several risk factors that contribute to this failure. These include elevated levels of ALT, TC, and LDL-C, the absence of statin therapy, and a relatively low DBP. These insights provide crucial information for optimizing treatment strategies and improving patient outcomes in ASCVD management. And the model demonstrates good predictive performance in both the derivation and validation sets.

Elevated LDL-C levels are significantly associated with the risk of ASCVD, and the risk of major adverse cardiovascular events can be reduced by 22% for every 1 mmol/L reduction of LDL-C level [12]. Previous studies have found that even after receiving lipid-lowering treatment, the compliance rate of LDL-C level in patients is still less than 30%, and lipid management needs to be strengthened. Furthermore, the TERESA Real World study [13] revealed that over 70% of Spanish patients with high or very high cardiovascular risk, despite treatment with high-intensity statins with or without ezetimibe, failed to achieve the recommended LDL-C targets. Seijas-Amigo J, et al. While PCSK9 inhibitors, as a novel LDL-C lowering monoclonal antibody, may offer promising benefits for high-risk or very high-risk cardiovascular patients, there remains a risk of suboptimal lipid-lowering therapy.

A study [14] based on Australian patients with ACS revealed that several factors contribute to NTR-LDLC: under 65 years old, women, high cholesterol levels at admission and being discharged with less than 4 medications. These findings align with the outcomes of a European study [15]. Post-percutaneous coronary intervention (PCI) lipid monitoring showed that only 47.8% of patients attained the target LDL-C level within a year, and 23.4% had LDL-C below 1.4 mmol/L, with women being more prone to substandard results. Furthermore, patients who did not meet the standards had a lower proportion of high-intensity statin treatment. The DYSIS II Europe study [10] found that among patients with stable coronary heart disease receiving lipid-lowering treatment, factors such as female gender, stable angina, a history of heart failure, and smoking were negatively associated with achieving LDL-C targets. Conversely, chronic kidney disease, type 2 diabetes, and intensive statin therapy emerged as favourable factors that positively correlated with LDL-C attainment.

In a multivariate logistic regression analysis examining the factors influencing the failure to achieve lipid targets in acute coronary syndrome [16], it was revealed that the lack of intensified lipid-lowering therapy prescription and TC levels exceeding 4 mmol/L were significant risk factors for not achieving lipid targets. Consistent with our findings, our predictive model identified the non-combined with statins and high TC levels as risk factors for NTR-LDLC among ASCVD patients treated with PCSK9 inhibitors.

The previous studies [14, 16, 17] observed that women were negatively associated with lower rates of LDLc treatment target achievement. However, our study did not find a significant impact of gender on the outcome. We attribute this discrepancy to the different inclusion and exclusion criteria used in the studies, as the previous studies focused on patients who were NTR-LDLC after lipid-lowering therapy. Our study explored the factors associated with NTR-LDLC despite PCSK9 inhibitor treatment. PCSK9 inhibitors are novel lipid-lowering drugs that can further reduce LDL-C levels by approximately 50% on top of statin therapy, resulting in a more potent lipid-lowering effect. This may be explained by the phenomenon that PCSK9 inhibitors exhibit no significant difference in effectiveness between different genders. Additionally, we employed LASSO regression analysis to construct our predictive model. LASSO regression selects variables that have the greatest predictive power for the target variable while adjusting for complexity, and female was excluded as a variable with a relatively small weight in the LASSO regression.

High alanine aminotransferase (ALT) levels are a risk factor for NTR-LDLC, primarily because doctors are concerned about the adverse reactions of statins such as hepatotoxicity and myotoxicity. As a result, they prescribe low-dose statins or no statins to patients with high ALT levels, leading to a decreased LDL-C attainment rate in this patient population. This suggests that high ALT levels influencing treatment decisions highlights the delicate balance between statin use and hepatotoxicity risks in clinical practice. In addition, our study found that DBP levels influenced the LDL-C levels in patients. This could be attributed to the fact that elevated blood pressure, surpassing the normal range, prompts greater attention and prompt initiation of blood pressure-lowering treatment upon the emergence of clinical symptoms. As a result, patients adhere to a regular regimen of antihypertensive medications, which, when combined with lipid-lowering therapy, can lead to a more pronounced reduction in lipids [18]. This model allows clinicians to predict the LDL-C reduction outcomes for patients following treatment with PCSK9 inhibitors, taking into account their individual conditions and risk characteristics. As a result, it aids clinicians in devising more personalized treatment plans for patients, thereby enhancing the effectiveness of lipid-lowering therapy.

In addition to the risk factors we have identified in our study, recent studies have explored image and serum biomarkers in CVD and ASCVD risk prediction. Coronary artery calcium (CAC), as a key imaging biomarker, with a zero score indicating a relatively low atherosclerotic burden and potentially a lower risk period. Early detection using CAC can help identify individuals at risk before the manifestation of clinical symptoms, which is crucial for timely intervention [19, 20]. Serum biomarkers also play a role in ASCVD risk prediction, reflecting pathophysiological processes. In subsequent research, integrating these biomarkers into the risk model may improve the early detection and prognostic accuracy of ASCVD [21].

There are several limitations in this study. Firstly, its retrospective and single-center nature restricts the data analyzed to a single institution, potentially impacting the model's performance and risk factor assessment in diverse populations. Consequently, external validation is essential to evaluate the model's generalizability across different settings. Secondly, the study focused on available risk factorsand excluded variables where data were missing or overlooked, which may have compromised the model's predictive accuracy. Lastly, future prospective studies are crucial to determine the model's effectiveness in guiding clinical decisions regarding lipid-lowering therapy.

Conclusions

In this study, we utilized the lasso regression method and integrating patients' fundamental data, including statins use, DBP, ALT, TC, and LDL-C levels, to develop an intuitive and concise nomogram prediction model. This model aims to predict whether patients will encounter NTR-LDLC after receiving PCSK9 inhibitors. Through validation with the validation set, the model exhibited good predictive accuracy and effectiveness. Consequently, it can effectively assist physicians in developing lipid therapies for patients.