A modified algorithm with lipoprotein(a) added for diagnosis of familial hypercholesterolemia

Abstract Background Previous studies have observed that high level of lipoprotein (a) [Lp(a)] was common in the phenotypic familial hypercholesterolemia (FH) and may explain part of the clinical diagnosis of FH. Hypothesis We aim to develop a modified model including Lp(a) and compare its diagnostic performance with Dutch Lipid Clinic Network (DLCN) criteria. Methods Data of 10 449 individuals were utilized for the model establishment (7806 for derivation and 2643 for validation) from January 2011 to March 2018. The novel score model was modified on the basis of DLCN. Furthermore, 718 patients were screened for LDLR, APOB, and PCSK9 gene mutations. Results The novel modified model consisted of untreated low‐density lipoprotein cholesterol (LDL‐C) level, Lp(a), personal premature coronary heart disease (CHD), tendon xanthomas and family history of CHD and/or hypercholesterolemia. It has shown high discrimination (area under curve [AUC] 0.991, 95% confidence interval [CI[ 0.988‐0.994, P < .001) for distinguishing clinical FH from non‐FH diagnosed using DLCN. Furthermore, a concordance analysis was performed to compare the modified model with DLCN and it showed a good agreement with DLCN (κ = 0.765). External validation of the novel model also showed good accordance (κ = 0.700). Further genetic analysis showed that the agreements between the new model and mutation improved a little compared to that between DLCN and mutation. Conclusions The novel modified model, including Lp(a), could provide new insights into FH diagnosis in Chinese population with more concerns on the patients with high level of Lp(a).

lipoprotein cholesterol (LDL-C) and thus premature coronary heart disease (CHD). 1 Patients with homozygous phenotype even could suffer cardiovascular morbidity and mortality in their childhood. 2,3 Despite of increasing awareness of FH, it has still been underdiagnosed and undertreated worldwide, partly attributing to the complexity and disunity of the current diagnostic criteria and underutilized genetic testing. 4 Until now, Dutch Lipid Clinic Network (DLCN) criteria, Simon Broome Register (SBR), Make Early Diagnosis-Prevent Early Death (MEDPED), criteria given by the 2015 American Heart Association scientific statement and other national diagnostic algorithms have been adopted worldwide. [5][6][7][8] However, there are several limitations in regard of their utilization in the clinical practice. First, the cut-off value of LDL-C may be only applicable to specific populations. Second, a comprehensive family history of dyslipidemia and/or CHD is usually unavailable or inaccurate. 9 Third, the discordance between different clinical diagnostic criteria and heterogeneity of phenotype and genotype often confused the physicians, 10 especially for those without their own diagnostic guidelines including China. Last but not least, the calculation is complex and could not promote diagnosis conveniently in the primary care.
With the deeper understanding of FH, more factors have been described responsible for the manifestations, among which lipoprotein (a) [Lp(a)] has received a lot of attention. 11,12 Previous studies have demonstrated significantly higher level of Lp(a) in patients with FH compared to the non-FH and its independent role in the risk stratifications. 13,14 The biochemical measurement of Lp(a) may explain 5% to 20% prevalence of the suspected FH, especially for those with negative FH-causing mutations. 15 Thus, we try to propose a potential approach to integrate Lp(a) into the diagnosis and investigate its performance.
Several novel model tools have been developed to simplify and improve the diagnosis of FH. 16 From March 2011 to March 2018, we consecutively recruited the subjects who were referred for coronary angiography (CAG) in this study as previously described. 19 Patients were excluded if they: (a) with secondary cause of dyslipidemia including severe thyroid, liver, and renal dysfunction; (b) without LDL-C measurement; (c) without Lp(a) measurement. As a result, a total of 10 449 participants at the time of the analysis were enrolled.
The adult patients were diagnosed as definite or probable FH according to DLCN criteria with a score ≥ 6. Clinical data of each participant were collected by physicians and experienced nurses, including the prior lipid levels and use of lipid-lowering medications, family and personal history of dyslipidemia and CHD as well as presence of tendon xanthoma and corneal arcus. For patients on lipid-lowering medications and without available untreated lipid profiles, their untreated LDL-C levels were adjusted using a correction factor depending on the type and potency of the lipid-lowering drugs. 20 Of the 10 449 patients, we further enrolled 718 patients with LDL-C levels above 4.5 mmol/L for a genetic testing. Patients were mutation positive if they carried pathological mutations in LDLR, APOB and PCSK9 genes as described in our previous studies. 19

| Biochemical examination
After an overnight fast, blood samples were collected from cubital veins for biochemical measurements as described in our previous publications. 21

| Model establishment
We developed the novel diagnostic criteria including Lp(a) on the basis of DLCN. First, we redefined the cut-off value of LDL-C according to Chinese data considering of the ethnic differences. Data from a cohort of patients without lipid-lowering therapy (LLT) were used to determine the 95th percentile for LDL-C. Furthermore, we calculated the best cut-off value of LDL-C for predicting FH and the value was 4.68 mmol/L. considering the relatively lower level of lipid in Chinese population, we adjusted the score classification and gave no score to the patients with LDL-C < 4.7 mmol/L. Second, Lp(a) level ≥ 22 mg/dL was given one point. Third, the premature CHD was still given two points, which was defined as male patients with on-set CHD younger than 55 years old and female patients younger than 60 years old. Of note, the xanthomas were specially referred to tendon xanthomas. Other manifestations including xanthelasma do not have high specificity although with a higher prevalence. Last but not least, as an inherited disorder, the family history is important for assisting diagnosis. In the new modified score model, we still kept one point for the family history of CHD or hypercholesterolemia. Finally, when the total score was above 6, the patient was defined as FH as same as DLCN criteria.

| Statistics
The statistical analysis was performed using SPSS version 21 software (SPSS Inc., Chicago, Illinois). Continuous variables were presented as mean ± standard deviation (SD) or median (Q1-Q3 quartiles) according to their distributions. Otherwise, categorical variables were shown as number (percentage). The differences of clinical and biochemical parameters between two groups were assessed using Student's t test, Mann-Whitney U test or χ² test appropriately. In the current analysis, the prior 75% samples of the study population according to the recruitment time were used as the derivation cohort, while the remaining 25% samples were used to validate the model as the validation cohort. Patients with a score < 6 using DLCN criteria were designated as negative cases for calculating sensitivity, specificity, positive predictive value, and negative predictive value. The Cohen kappa (κ) coefficient was applied to evaluate the agreement between the new FH definition, DLCN criteria and gene mutation. In detail, κ > 0.8 indicated excellent agreement, 0.6 to 0.8 indicated good agreement, 0.4 to 0.6 indicated moderate agreement, and < 0.4 indicated poor agreement. 22 The discriminative power was further evaluated using the receiver operating characteristics (ROC) curve and the new modified score model was compared with different models by the c statistics with 95% confidential interval (CI).

| Baseline characteristics
The clinical characteristics of the derivation population and validation population were shown in Table 1. For the derived data set with a total of 7806 patients, there were 253 patients diagnosed with FH using DLCN. Compared with non-FH subjects, patients with FH were significantly younger and suffered more premature CHD (66% vs 33.8%, P < .001), while the prevalence of CHD was not significantly different (85.4% vs 85%, P = 0.874). In addition, only 12 (4.7%) patients present tendon xanthoma, which was highly specific in FH. Three patterns of statin use (high-, moderate-, and low-intensity patterns) were defined on the basis of the statin type and potency as described previously. 19 Obviously, there were much more patients with FH under high-intensity statin treatment. In spite of moderate and high-intensity statin treatment, patients with FH had significantly higher level of LDL-C, TC, apoB, and Lp(a) levels than those with non- FH (all P < .001). Similar results were presented in the validation population.

| Screening criteria for FH
The novel modified model was shown in Table 2. The 95th percentile for LDL-C in the untreated cohort was 4.83 mmol/L and the cut-off value of LDL-C for predicting FH was 4.68 mmol/L. The distributions of LDL-C according to lipid-lowering treatment were shown in Figure S1.
Along with the DLCN criteria, examination of existing databases confirmed that Lp(a) levels ≥22 mg/dL was the best cut-off value for a clinical diagnosis of FH ( Figure 1). Thus, patients with LDL-C < 4.7 mmol/L were given no point but patients with Lp(a) level ≥ 22 mg/dL were given one point. The DLCN criteria were also provided in Table S1.

| Predictive performance for FH
The predictive capacity of the new modified model was evaluated using c statistic (Figure 2). the LDL-C cut-points of DLCN criteria has been changed because the date deriving from western populations may not be adopted to Chinese population due to the relatively lower level of LDL-C in China. 23 The FH diagnostic guideline issued by Japan applies LDL-C ≥ 180 mg/dL (4.6 mmol/L) as the cut-point. 24 Korean scientists found that best LDL-C threshold for predicting putative mutation was 225 mg/dL (5.8 mmol/L) while the conventional diagnostic criteria had low specificities. 25 Thus, we downgraded the LDL-C cut-points based on the Chinese data. Second, the physical signs including xanthomas and corneal arcus are uncommon despite of high specificity. 26   We acknowledge the limitations in the current study. First, the new score model was modified mainly according to our experiences in the clinical practice and supported by our own data. Further analysis to establish a more scientific model is essential. Second, the DLCN criteria are not applicable for children. So is the new modified model.

| CONCLUSION
We