Performance of LDL-C only compared to the Dutch Lipid Clinic Network Score for screening of familial hypercholesterolaemia: the Austrian experience and literature review

Abstract

Aims

Familial hypercholesterolaemia (FH) is a severely underdiagnosed, inherited disease, causing dyslipidaemia and premature atherosclerotic cardiovascular disease. In order to facilitate screening in a broad clinical spectrum, we aimed to analyse the current yield of routine genetic diagnostics for FH and to evaluate the performance of the Dutch Lipid Clinic Network Score (DLCNS) compared to a single value, the off-treatment LDL-cholesterol exceeding 190 mg/dL.

Methods and results

We investigated all patients that underwent molecular genotyping routinely performed for FH over a 4-year period in two Austrian specialist lipid clinics. Variants reported in FH-causing genes including LDLR, APOB, PCSK9, LDLRAP, and APOE were collected and classified. For clinical classification, the DLCNS was calculated retrospectively and compared to the original scores documented in patient charts. Additionally, a literature review on comparisons of DLCNS to LDL-C was performed. Of 469 patients tested, 21.3% had a disease-causing variant. A median of 3 out of 8 (excluding genotyping results and LDL-C) DLCNS criteria were unavailable. DLCNS was documented in 48% of cases, with significant discrepancies compared to retrospective scoring (P < 0.001). DLCNS did not outperform off-treatment LDL-C alone (Δ = 0.006; P = 0.660), analogously to several reports identified in the literature. A single cut-off of 190 mg/dL LDL-C compared to DLCNS ≥ 6 showed excellent sensitivity (84.9% vs. 53.8%) and acceptable specificity (39.0% vs. 84.1%).

Conclusion

Missing criteria and severe discrepancies observed between retrospective and on-site scoring by treating physicians were highly prevalent, confirming limited utility of DLCNS in clinical routine and warranting a single off-treatment LDL-C cut-off of 190 mg/dL for enhanced index-case identification.

Graphical Abstract

Lay Summary

Familial hypercholesterolaemia as a genetic disorder driving the development of premature atherosclerotic cardiovascular disease needs intensified and reliable screening as it is severely underdiagnosed and undertreated. Definite diagnosis of Familial Hypercholesterolaemia is facilitated via genetic confirmation of an underlying variant. Genetic diagnostics however is costly and therefore dependent on the correct identification of patients at risk, using clinical tools such as the commonly used Dutch Lipid Clinic Network Score.

• In our study, however, its utility has been demonstrated to be severely limited by the highly prevalent lack of availability of critical information for the score in clinical routine.

• Furthermore, we found significant inconsistencies in scoring results between the treating physician and the retrospective evaluation by the study team, indicative of unreliable results produced by the score in clinical routine.

Therefore, we propose to promote the utilization of a more time-effective approach, which is the use of a single LDL-C value, allowing for the urgently needed surge in referrals to tertiary care centres for index-case identification through a low-threshold access.

Familial Hypercholesterolaemia, Austria, Spectrum, DLCNS, Yield, Performance

Introduction

Familial hypercholesterolaemia (FH) is an autosomal-dominant genetic disorder causing severe elevations of LDL-C and dramatically increased risk of premature atherosclerotic cardiovascular disease (ASCVD).¹ With an estimated prevalence of about 1:300,^2,3 it is the most common monogenic disorder worldwide. FH is underdiagnosed and undertreated worldwide,⁴ with undiagnosed patients being denied the opportunities for safe and effective therapy and its lifelong benefits.⁵ Diagnosis rates vary widely across Europe,^6,7 with most countries estimating rates between 5% and 10%.

The majority of genetically confirmed FH is caused by monogenic defects in the LDLR, APOB, and PCSK9 genes in the autosomal dominant form, or LDLRAP1 in the autosomal recessive form, with APOE being only a minor monogenic contributor. Large rearrangements in the LDLR gene have been found to account for 5–15% of the cases. In 20–30% of cases, the cause remained unexplained and a polygenic cause was found due to variants in a series of low-effect genes raising LDL-C levels to overlapping, yet in the majority of cases not as severe levels compared to monogenic FH.⁸

Despite the rising number of modifications or national adaptations for specific populations and alternative scores,^9–13 clinical diagnosis of FH is commonly facilitated by the Dutch Lipid Clinic Network Score (DLCNS),⁴ which is widely used in Austria¹⁴ and substantially influenced by the level of off-treatment LDL-C, the presence of physical signs of severe hypercholesterolaemia including tendon xanthomas and arcus lipoides juvenilis, as well as family history and patient history of premature ASCVD. A score of ≥6 results in a clinical diagnosis of FH and warrants diagnostic confirmation via genetic testing.⁴ As practical as the DLCNS appears in theory, it is limited in practice by its time consumption, especially when time is the main limitation in daily clinical care in middle Europe, as well as missing information of patients or off-treatment LDL-C.¹⁵ Furthermore, the DLCNS has never been validated in or calibrated for Austrian cohorts before, and the benefit of its use in contemporary cohorts has not been widely examined. Therefore, its use might potentially impede urgently needed referrals to specialist lipid clinics, crucial for overcoming the bottleneck of index-case identification and exponentiated detection via cascade screening.

Therefore, the aim of this study was to analyse the current yield of routine molecular genetic diagnostics for FH in our two lipid clinics at a university hospital and to evaluate the performance of the DLCNS in our cohort in order to identify strategies for the improvement of detection rates in the population.

Material and methods

Study population

All patients who routinely underwent molecular genetic testing for FH due to indication provided by the treating specialist during routine lipidologic workup in our two lipid clinics (Division of Endocrinology and Metabolism, Division of Cardiology) of the Vienna General Hospital, Austria over the period of 2018–2021 were included.

Laboratory biochemistry

Routine lipid lab tests, including LDL-C using Friedewald’s equation, were collected. Off-treatment LDL-C values were comprehensively investigated in electronic patient charts, with the highest one available being retrieved. LDL-C values in patients under stable lipid-lowering medication (LLM) were additionally imputed according to previous literature (see Supplemental Methods section of Supplementary material online).

Clinical classification

For clinical classification, the DLCNS was used.⁴ Scoring was performed by the treating physicians during clinical routine and retrospectively by the authors, by reviewing the electronic patient charts, extracting (1) clinical history, (2) physical examination, and (3) family history, up to the date of genetic testing, hence disregarding genetic test results. Cases lacking reports of LDL-C values of any kind or clinical assessment or documentation by physicians were excluded from scoring and further analyses. Cases with missing, incomplete, or equivocal reports on relevant criteria were considered unavailable and awarded 0 points in DLCN scoring in the respective category. Cases lacking documented ASCVD were considered disease-free in this regard. One point was awarded for any reported history of hypercholesterolaemia in a first-degree family member, instead of the original criteria (LDL-C > 95th percentile). For DLCN scoring, unimputed off-treatment LDL-C values were preferred over imputed values whenever available.

Molecular genetic analyses

For molecular genetic analyses, Amplicon Sequencing (SEQPRO Lipo IS; Progenika) on MiSeq (lllumina) was used. The genes analysed were LDLR, APOB, PCSK9, LDLRAP1, and APOE, with all exons +/−30 nucleotides intronic included in the analysis, except APOB (only c.10438–10757 of exon 26 and c.13009–13301 of exon 29) and APOE (only c.225–521). Multiplex ligation-dependent probe amplification (MLPA) was used for LDLR copy number analysis to detect large rearrangements (‘SALSA MLPA Probemix P062-D2 LDLR’). The Human Genome Variation Society nomenclature was used for reporting variants.

Variant assessment

Variants were interpreted and classified according to the current guidelines of the American College of Medical Genetics and Genomics (ACMG) using Alamut Visual Plus version v1.3, VarSome and ClinVar and were confirmed by experienced geneticists. The novelty of the variants was determined by their absence in HGMD, LOVD, and ClinVar and confirmed by a literature search conducted via Mastermind (see Supplemental Methods section of Supplementary material online). The reference sequences used for LDLR, APOB, PCSK9, LDLRAP1, and APOE were NM_000527.4 and NM_000384.2, NM_174936.3, NM_015627.2, NM_000041.2, respectively.

Data analysis

Statistical analyses were performed using IBM Statistical Package for the Social Sciences, version 29.0.0 (IBM SPSS Statistics for Macintosh, Armonk, NY: IBM Corp. USA). Distribution of data was assessed by Kolmogorov–Smirnov test. Continuous data were presented as means and standard deviations (SD) or medians and interquartile ranges (IQR), as appropriate. Student’s t-test, chi-squared test, Mann–Whitney-U test were used for comparing metric, dichotomous, and ordinal or non-normally distributed metric data, respectively. Wilcoxon signed-rank test was used for comparison of scorings, and paired t-test for comparisons of means. For investigating correlations, Spearman’s rho was used. Statistical significance was defined as P < 0.05, two-sided. Sensitivities and specificities were calculated via http://vassarstats.net/. Graphics were created using SPSS version 29.0.

Literature search

A literature research was conducted (see the Supplemental Methods section of the Supplementary material online and Supplementary material online, Figure S1).

Results

Patient characteristics

A total of 469 patients underwent genetic testing for FH. At the time of genetic testing, 43.3% were treatment-naive, 43.9%, 32.0%, and 4.5% were under statin-, ezetimibe- or PCSK9i therapy respectively, 1.3% underwent lipid apheresis and 5.1% used fibrates. For baseline characteristics of all patients, see Table 1. At the time of genetic testing, the median, current LDL-C was 150 (88–198) mg/dL. Unimputed, off-treatment LDL-C values were available in 62.5% of the patients, with a median level of 208 (185–243) mg/dL. When adding LDL-C values imputed via off-treatment values based on current medication, off-treatment LDL-C was available in 91.5% of the cohort.

Table 1

Patient characteristics of all patients at time of testing, of total cohort and per variant carrier status

	Total (n = 469)	Variant + (n = 100)	Variant − (n = 369)	P value
Sex, % male (n)	47.1 (221)	47.0 (47)	47.2 (174)	0.978
Age (years)	51.5 ± 13.6	47.6 ± 14.9	52.6 ± 13.1	0.001
BMI (kg/m2)	26.9 ± 4.8	26.4 ± 4.5	27.1 ± 4.9	0.257
Total cholesterol (mg/dL)	236 (164–297)	238 (179–333)	234 (160–289)	0.034
LDL-C (mg/dL)	150 (88–198)	163 (103–256)	147 (81–192)	<0.001
HDL-C (mg/dL)	52 (43–66)	51 (43–63)	53 (43–66)	0.469
Apo-A1 (mg/dL)	146 (129–171)	143 (127–162)	147 (130–172)	0.135
Apo-B (mg/dL)	133 (87–167)	134 (102–198)	133 (84–160)	0.004
Non-HDL-C (mg/dL)	181 (106–233)	189 (121–280)	180 (103–223)	0.018
Lipoprotein(a) (nmol/L)	43 (10–201)	43 (17–147)	43 (9–219)	0.651
Triglycerides (mg/dL)	129 (87–196)	102 (67–157)	135 (93–205)	<0.001
Creatine kinase (mg/dL)	105 (70–160)	100 (69–158)	106 (70–161)	0.532
Gamma-Glutamyl Transferase (U/L)	27 (16–43)	22 (15–33)	27 (17–46)	0.029
HbA1c (%)	5.5 (5.2–5.7)	5.4 (5.2–5.7)	5.5 (5.2–5.8)	0.166
Creatinine (mg/dL)	0.83 (0.72–0.98)	0.78 (0.71–0.87)	0.84 (0.72–0.99)	0.010
CHD, % (n)	30.5 (143)	37.8 (34)	32.5 (109)	0.387
CHD events, % (n)	23.7 (111)	26.0 (26)	23.0 (85)	0.530
Xanthelasma, % (n)	1.9 (9)	3.0	1.9	0.690
Treatment-naïve, % (n)	43.5 (204)	48.0 (48)	41.4 (153)	0.373
Nutraceuticals, % (n)	1.5 (7)
Ezetimibe only, % (n)	6.4 (30)
Bempedoic acid, % (n)	0.4 (2)
Statins only, % (n)	17.7 (83)
Statins + Ezetimibe, % (n)	23.2 (109)
PCSK9i only, % (n)	1.5 (7)
PCSK9i + Statins, % (n)	0.6 (3)
PCSK9i + Statins + Ezetimibe, % (n)	2.3 (11)
Fibrates, % (n)	5.1 (24)
Lipoprotein apheresis, % (n)	1.3 (6)

Values stated as means (±SD), medians (IQR), or per cent (count). P values for differences between variant-positive and variant-negative patients. P values <0.05 are highlighted in bold.

Excluding LDL-C and disregarding molecular genetic testing in our cohorts, a median of 3 (3–5) out of 8 criteria was unavailable for the DLCN scoring (see Table 2 and Figure 1). In comparison to non-carriers, carriers of disease-causing variants were younger (47.6 vs. 52.6 years, P = 0.001), showing higher median values of LDL-C (163 vs. 147 mg/dL; P < 0.001), ApoB (134 vs. 133 mg/dL, P < 0.001), non-HDL-C (189 vs. 180 mg/dL, P = 0.023), as well as lower median values of TG (102 vs. 135 mg/dL; P < 0.001) at the time of testing. Mean off-treatment LDL-C significantly differed between variant carriers and non-carriers [(262 vs. 201; P < 0.001), see Supplementary material online, Figure S2]. Reports of hypercholesterolaemia in first-degree relatives of full age, as well as <18 years of age were more prevalent in variant-positive patients (56% vs. 26% or 4% vs. 0.3%, respectively; P < 0.001). All other relevant patient and family histories (premature coronary heart disease (CHD); cerebral artery disease (CAD); peripheral artery disease (PAD), and pathognomonic signs) were not significantly more prevalent in the genetically FH group. Noteworthily, eight out of nine tendon xanthomas were reported in variant-negatives (see Table 2).

Figure 1

Number of criteria (beyond LDL-C) unavailable for scoring of the DLCNS in our cohort. Criteria included related to family history of first-degree relatives (4), clinical history (2), and physical examination (2).

Table 2

Proportion of unavailable criteria relevant to DLCNS and their prevalence (n) in total cohort and per variant carrier status

	Unavailable	Total (n = 469)	Variant+ (n = 100)	Variant− (n = 369)	P value
First-degree relative premature CHD, % (n)	28.8%	20.9 (98)	24.0 (24)	20.1 (74)	0.886
First-degree relative HCL, % (n)	61.4%	32.6 (153)	56.0 (56)	26.3 (97)	<0.001
First-degree relative TX/Arc lip, % (n)	98.7%	0.9 (4)	2.0 (2)	0.5 (2)	0.225
First-degree relative HCL <18y, % (n)	62.3%	1.1 (5)	4.0 (4)	0.3 (1)	<0.001
Premature CHD, % (n)	10.7%	19.2 (90)	26.0 (26)	17.3 (64)	0.144
Premature CAD/PAD, % (n)	16.4%	7.2 (34)	6.0 (6)	7.6 (28)	0.629
Tendon xanthoma, % (n)	43.9%	1.9 (9)	1.0 (1)	2.2 (8)	0.263
Arcus lipoides (<45y), % (n)	44.3%	0.2 (1)	1.0 (1)	0.0 (0)	0.083
Off-treatment LDL-C (mg/dL)a	37.5%	208 (185–243)	262 (218–315)	201 (180–228)	<0.001

Values stated as means (±SD), or per cent (counts), P values for differences between variant-positive and variant-negative patients. First-degree relative: family history of first-degree relative. P values <0.05 are highlighted in bold.

CHD, coronary heart disease; CAD, cerebral artery disease; PAD, peripheral artery disease; CVD, cardiovascular disease, premature defined as in men < 55 years or women < 60 years; HCL, hypercholesterolaemia; TX, tendon xanthoma; Arc lip, arcus lipoides.

^aUnimputed LDL-C values (as per patient history or current in untreated patients, excluding imputations).

Performance of the DLCNS

DLCNS based on actual off-treatment LDL-C values was retrospectively calculated in 61.8%, expanding to 90.8% when adding those with only estimated off-treatment values available based on current medication. Median DLCNS was 3, and the proportion of unlikely (DLCNS, <3), possible (DLCNS, 3–5), probable (DLCNS, 6–8), and definite (DLCNS, >8) FH before testing was 27.0%, 48.8%, 14.6% and 9.6%, respectively (see Supplementary material online, Figure S3). The percentage of patients with a clinical diagnosis of FH (DLCNS, ≥ 6) was 24.2%. Detection rate for causative variants was highest in the definite FH group with 56.1% carriers, 43.5% in the probable, 15.4% in the possible, and 9.6% in the unlikely group (see Supplementary material online, Figure S4), conferring to a variant detection rate of 48.5% in those scoring ≥6 in the DLCNS. The proportion of a clinical diagnosis of FH (DLCNS ≥ 6) in variant carriers was 53.8%, conferring to 46.2% not adequately identified by the DLCNS. The detection rate among off-treatment LDL-C > 190 mg/dL was 28.0%.

Supplementary material online, Table S1 shows sensitivities, specificities, and AUC for different comparators and their cut-offs. The best off-treatment LDL-C cut-off according to Youden’s index is >242 mg/dL, the closest top left is >231 mg/dL, and the one with the greatest sensitivity with a specificity >50% is >200 mg/dL (see Supplementary material online, Figure S5).

Receiver operator curve (ROC) analysis for off-treatment LDL-C values including imputed values compared to DLCNS for discriminating between variant carriers and non-carriers are shown in Figure 2. Both methods perform well with an AUC of 0.766 (CI: 0.708–0.824) for off-treatment LDL-C and 0.760 (CI: 0.703–0.816) for DLCNS, with no significant differences in discriminatory power between the 2 methods (Δ AUC: 0.006; CI: −0.021–0.034; P = 0.660). AUCs were higher when excluding those with imputed off-treatment LDL-C values available only, 0.801 (CI: 0.736–0.866) for LDL-C and 0.798 (CI: 0.738–0.859) for DLCNS, with no significant differences in discriminatory power between the two methods [(Δ AUC: 0.003; CI: −0.037–0.036; P = 0.894) see Supplementary material online, Figure S6]. For comparisons with unimputed and corresponding imputed LDL-C and DLCNS values, see Supplementary material online, Figure S7.

Figure 2

ROC curve of off-treatment LDL-C and DLCNS of total cohort including those with imputed values, for discriminating the presence of a causative variant in our cohort.

The number of missing criteria was significantly, though moderately, negatively correlated with the DLCNS (P < 0.001; s_ρ = −0.246).

The dichotomous cut-off of >190 mg/dL of off-treatment LDL-C compared with the dichotomous cut-off of DLCNS ≥3 points (possible FH) demonstrates that 29 patients (6.8% of the total cohort) classified for a score of ≥3 due to spare DLCNS criteria despite an off-treatment LDL-C below 190 mg/dL, 3 of which (10.3%) being variant carriers (3% of all variant carriers) as illustrated in Supplementary material online, Figure S8.

Comparing scoring by treating physicians with retrospective scoring by authors

Original DLCNS performed by the treating physician was available in 30% (Department of Endocrinology) and 62% (Department of Cardiology) of cases, in total 48%. Compared to retrospective scoring performed by the authors, original physician scoring was lower in 16% of cases (33); in 46% (97 cases), it was higher; and in 38% (79 cases), scoring was identical (P < 0.001). In line, clinical categories of DLCNS (unlikely, possible, probable, and definite) resulting from original physician scoring were lower compared to retrospective scoring in 10% (21 cases), higher in 23% (49 cases), and equal in 67% (139 cases), (P = 0.021).

Discrepancies between retrospective scoring by the authors and treating physicians were highly prevalent and highly discrepant, with −8 as the highest negative and +9 points as the highest positive discrepancy, resulting in a range of 17 and a mean of −0.5 points lower scoring by the authors. As illustrated in Figure 3, inter-observer discrepancies were present across the whole spectrum of missing criteria, without significant correlations (s_ρ = 0.031; P = 0.659) with respect to the number of missing criteria.

Figure 3

Discrepancies (Δ) of DLCN scoring between retrospective scoring by the authors minus scoring by the treating physician as function of number of missing criteria for DLCNS.

AUC of original DLCNS by treating physician was insignificantly lower than by retrospective scoring in the total cohort [0.692 (CI: 0.607–0.787) vs. 0.727 (CI: 0.646–0.807); ΔAUC: 0.035; CI: −0.034–0.104; P = 0.326; see Supplementary material online, Figure S9]. For comparisons with the total cohort, see Supplementary material online, Figure S10.

Genetic test results

Out of 469 patients that underwent genetic testing, 55 different variants which were either previously described as disease-causing or were newly identified as likely pathogenic or pathogenic could be found in 100 (21.3%) patients (see Supplementary material online, Table S2 and Supplementary material online, Table S3). Of the 55 different underlying variants, 51 were located in the LDLR gene (89 patients), one in the APOB gene (seven patients; E26: c.10580G > A, p.Arg3527Gln), one in the APOE gene (one patient, E4: c.500_502delTCC, p.Leu167del) and two in PCSK9 gene (three patients). No pathogenic variants were found in the LDLRAP1 gene. 60% (33) of the disease-causing variants were missense, 7.3% (4) nonsense variants, 10.9% (6) frameshift mutations, 7.3% (4) in-frame deletions, 10.9% (6) intronic/untranslated, and 3.6% (2) large rearrangements detected by MLPA. Two true homozygotes were identified (E5: c.798T > A, p.Asp266Glu (LDLR), FH Cincinnati-1; E2: c.81C > G, p.Cys27Trp (LDLR), FH San Francisco). In total, 44 variants of unknown significance (VUS) in LDLR, APOB, PCSK9, and LDLRAP were detected. All variants considered (likely) benign or of uncertain significance are listed in Supplementary material online, Table S4. One patient received an alternate diagnosis due to a pathogenic variant previously described in dysbetalipoproteinaemia (E4: c.460C > T, p.Arg154Cys, APOE).

Novel variants

In total, 12 variants not previously reported elsewhere have been identified (see Supplementary material online, Table S5), 2 of which are (likely) pathogenic. Five in the LDLR gene, 1 in APOB, 4 in PCSK9, and 2 in LDLRAP1. C.780_793del (LDLR), a small deletion resulting in a frameshift with a premature stop codon and c.1892_1906del (LDLR), an in-frame deletion, are considered to be (likely) pathogenic. The other novel variants are classified as variants of unknown significance (VUS) or (likely) benign and include three missense and one synonymous variants, three variants in untranslated regions, and five intronic variants. In silico analyses for the missense variants predominantly predict benign effects on the amino acid sequence. Splice site prediction tools predicted an influence on splicing for c.332A > C (LDLR), c.2139A > T (LDLR), and c.313 + 4A > C (LDLR).

Discussion

The current study confirmed our hypothesis that off-treatment LDL-C alone performs well compared to the DLCNS in clinical routine as a decision tool for suspecting and testing for familial hypercholesterinaemia. In order to prevent premature ASCVD, the identification of FH patients is necessary and should be easily performed at every point of care in the healthcare system. With a rate of detection in our clinic of 21.3%, we aimed to identify the reasons for the allocation of genetic testing and compare the score to off-treatment LDL-C cut-offs.

As observed in other studies, the lack of relevant data poses the risk of underestimation of the score.¹⁵ The number of missing criteria, as expected, led to lower DLCN scores, however did not seem to cause the discrepancies between retrospective scoring and scoring by treating physician, suggesting no significant bias due to loss of information owing to documentation. Partly, substantial discrepancies across the whole spectrum of total missing variables were observed.

Original DLCN scoring by the treating physician was available in only 48%, indicating limited utilization or feasibility in clinical routine. Despite the risk of omission of relevant criteria in documentation, which is inherent to retrospective studies relying on documentation, a substantial proportion of divergences derive from lower original than retrospective scoring. The proportion of lower-than-retrospective scorings and the lack of rising discrepancies with an increasing number of missing variables suggest that omission in electronic charts/neglected history taking can be ruled out as a significant confounder. Naturally, scores tend to be lower with a rising number of missing criteria, which can therefore be confounded by the point of time when the test is taken, as the availability of criteria and LDL-C values differ in time.¹⁶ Therefore, only findings up to the date of the genetic workup were included. Lower than retrospective ‘on-the-spot’-scoring suggests a failure to include relevant aspects by the treating physician, e.g. missed history taking or examination, or inaccurately identified original off-treatment LDL-C values, which were thoroughly investigated in this study, compared to the allegedly more prevalent use of evidentially less accurate, but time-saving imputed LDL-C in clinical routine, as suggested in Supplementary material online, Figure S7. Imputations reportedly pose the risk of overestimation of off-treatment LDL-C levels,¹⁵ were demonstrated to be inferior compared to imputed LDL-C in our report (see Supplementary material online, Figure S5) and were necessary in a substantial proportion of our cohort (29%). The need for imputation of pre-treatment values is prevalent, not only in clinical routine (necessary in 25%¹⁷) but also in systematic study settings (necessary in 15.4%;¹⁵ > 25%;¹⁸ 26%;¹⁹ and 46%²⁰) and owed to the oftentimes already established LLM at the time of referral with difficult to retrieve pre-treatment LDL-C values predating the current assessment by many years.²¹ A high proportion of cases were scored higher on-site than retrospectively, suggesting information relevant for scoring or higher off-treatment LDL-C on hand remained undocumented. Previously, a study demonstrated that DLCN scoring of identical clinical data tends to be prevalently discrepant between observers, suggesting high susceptibility to error.²²

Comparable to prospective studies, data on premature CHD was not unequivocally available in ∼10% of cases. This is owed to cases with missing age of onset (a proportion of those with long-standing CVD could not be verified to have had a certainly premature onset as per the strict definitions of DLCNS or not and could therefore not be classified with confidence); diagnostics ongoing and not completed at the time of testing, with ex post facto available results of those diagnostics being disregarded and the value categorized as missing, as only data available at the time of scoring/ordering the genetic test was included, to ensure comparability between retrospective and ‘on-the-spot’-clinician DLCN scoring, depicting the clinical reality which is oftentimes uncertainty about diagnoses. Third, past medical history is not comprehensively documented, therefore no certain rule-out or -in possible. Overall, history was ‘in dubio’ classified as missing in order not to compromise data quality. Crucial data on patient and family history relevant for scoring was unavailable in a substantial proportion of patients in our routine setting, with a family history of premature CVD or hypercholesterolaemia or pathognomonic signs missing in 28.8%, 61.4%, and 98.7%, respectively. Exact LDL-C values of relatives were unavailable in all except for sporadic cases. Therefore, any history of hypercholesterolemia reported in a first-degree family member was considered positive, instead of the stringent original criteria requiring exact values. Data on physical examination were unavailable in ∼44% of cases. Figures from prospective studies in similar populations²³ give reason to expect higher numbers of physical stigmata, suggesting underutilization of physical examination in clinical routine. Noteworthily, the vast majority of suspected tendon xanthomas in our cohort were described in patients without genetic confirmation of FH, suggesting difficulties in assessment across examiners, with increasing rareness and lack of objectifiable criteria for diagnosing xanthoma having been criticized before.^24,25 FH was recently demonstrated to be the strongest predictor for the presence of a positive family history of heart disease, yet only explaining 0.15% of positive family history of heart disease. Furthermore, the utility of self-reported family history is limited by inaccuracies due to inconsistent definitions for disease as well as the patients’ recall and awareness of their respective family history.²⁶

Overall, this suggests that relevant criteria for the DLCNS might not be feasibly collectible in clinical routine, as specific and complex information on family history is rarely known to the subjects and is time-consuming to investigate. However, even in the systematic collection via prospective study settings or cascade screening projects, as opposed to our clinical routine setting, missing information also amounts to substantial proportions, despite the targeted design.¹⁵

DLCNS did not show any significant advantage in predicting a positive genetic test result compared to off-treatment LDL-C alone in ROC analysis. Of the studies identified in our literature search^{17–20,27–33} (see Table 3), the majority similarly reported higher AUC values for LDL-C than for DLCNS, however mostly without statistically significant differences (if reported),^{17–20,27,30} except for one study by Noto et al²⁸ of well an investigated cohort subanalysed in those without any missing criteria, reporting significantly better performance of LDL-C alone. The largest study²⁰ reported marginal, non-significant differences in performance. Conversely, superior performances of DLCNS reported were either statistically insignificant,^31–33 or showed merely marginal differences in effect size (ΔAUC = 0.014) as in one study by Silva et al,²⁹ reaching statistical significance in a large cohort of a cascade screening project, as compared to the larger opposite effect size (ΔAUC = −0.075) in the much smaller subpopulation by Noto et al.²⁸ Possible reasons for the conflicting results include the extreme phenotype in the study by Silva or ethnicity, as it is generally unclear to what extent the DLCNS are applicable in particular populations outside the Western derivation populations.² Studies not reporting cut-offs of LDL-C only as compared to the whole spectrum of values were not included in the review for lack of comparability (see Supplementary material online, Figure S1).

Table 3

Literature reporting performance of DLCNS and LDL-C for the detection genetically verified FH

Author	Country	n	Designf	Cohort	Criteria	AUC DLCNS	AUC LDL-C	ΔAUCa	Significance
Li BV et al.27	New Zealand	213	Retrospective	Genotyping in lipid clinic	/	0.665	0.724	−0.059	n.r.
Molnar S et al.20	Germany	3267	registry	LURIC Study	Coronary angiography patients	0.675	0.678	−0.003	P = 0.952
Yip M-K et al.19	Hong Kong	31	Cross-sectional	Genotyping in lipid clinic	/	0.723	0.754	−0.031	n.r.
Fourgeaud M et al.32	France	147	Retrospective	Genotyping in lipid clinic	/	0.766	0.707	+0.059	P = 0.320
Noto D et al.28	Italy	836b	Retrospective	Genotyping in lipid clinic	>4.88 mmol/L LDL-C	0.662	0.737	−0.075	P = 0.007
Trinder M et al.18	Canada	626	Registry	FH registry	DLCNS ≥ 3	0.753	0.818	−0.065	n.r.c
Benedek P et al.33	Sweden	116	Registry	SWEDEHEART registry	ACS and TC > 7 mmol/L (or >4.9 mmol/L under LLM)	0.856	0.732	+0.124	P = 0.167
Chan DC et al.17	Australia	885	Cross-sectional	Genotyping in lipid clinic	DLCNS ≥ 3	0.816	0.835	−0.019	P = 0.164
Silva PRS et al.29	Brazil	753	Registry	Hipercol Brasil (Cascade Screening Programme)	LDL-C ≥ 210 mg/dL or suspected FH	0.744	0.730	+0.014	P = 0.014
Grenkowitz et al.30	Germany	206	Retrospective	Genotyping in lipid clinic	Clinical diagnosis of FH	0.789	0.799	−0.010	n.r.c
Mickiewicz et al.31	Poland	193	cross-sectional	genotyping in lipid clinic	LDL-C ≥ 4.9 mmol/L (190 mg/dL)	0.79d/0.86e	0.79d/0.84e	0.0d/+0.020e	n.r.c

P values <0.05 are highlighted in bold.

LLM, lipid-lowering medication; TC, total cholesterol; ACS, acute coronary syndrome

^aAUC DLCNS—AUC LDL-C.

^bSubanalysis performed in n = 203.

^cConfidence Intervals reported only.

^d< 40 years.

^e≥ 40 years.

^fas reported by the authors.

n.r. not reported.

With off-treatment LDL-C alone performing at worst only 0.021 points inferior to the DLCNS in ROC in our cohort, and off-treatment LDL-C alone being reported to be a good predictor in enriched cohorts,^20,28 it is tempting to speculate, whether recommendations regarding initiation of specialist work-up based on off-treatment LDL-C alone would be helpful, as limited utilization and feasibility of DLCNS might contribute to low number of referrals to specialist clinics and poor diagnosis rates.^34,35 Approaches using the >190 mg/dL cut-off, together with a simplified assessment of family history^21,36 or off-treatment LDL-C only have been proposed,²⁰ and have partly been demonstrated to show excellent agreement with DLCNS using LDL-C cut-points adapted to the population.²¹ Simplified and widespread pre-clinical stratification via off-treatment LDL-C alone might allow for a much-needed surge in referrals for consequent confirmation completed with genotyping in tertiary centres with expandable capacities for genetic testing.²⁰ The established and recommended off-treatment LDL-C cut-off of >190 mg/dL³⁷ performs well in our cohort as compared with DLCNS ≥ 6 regarding sensitivity and specificity, as it captures 84.9% of the affected patients. Additionally to the traditional methods, the cut-off with the greatest sensitivity with a specificity >50% (>200 mg/dL) was reported, as Youden’s index and closest-top-left weigh sensitivity and specificity equally. The cut-off of >200 mg/dL shows comparable sensitivity to the rather insensitive 190 mg/dL cut-off, however, with increased specificity. Further studies evaluating recommendations towards a simplified approach using off-treatment LDL-C for identification of potential index cases complemented by genotypic confirmation to aid in increasing diagnosis rates would be of great interest.

Limitations include lack of generalizability to the general population, as we investigated a pre-selected population which routinely underwent genetic testing due to the clinical judgement of the respective treating physicians and their reasonable suspicion of FH. Therefore, in our collective, there was a 27.8% positivity rate in those with an off-treatment LDL-C of >190 mg/dL compared to the reported 1.7–2.5% yield in unselected cohorts of >190 mg/dL.^1,38 Yields of molecular genetic testing are difficult to compare and remarkably inconsistent and range from 28% to 80% in enriched populations, depending on diagnostic criteria, study settings, pre-selection by clinical diagnosis and country.³⁹ The proportion of uninvestigable, essential criteria in clinical routine is substantial and prevents the DLCNS from a fair comparison with an off-treatment LDL-C of 190 mg/dL cut-off alone.

In conclusion, the utility of the currently established DLCNS is limited by difficulties in obtaining detailed information in clinical routine. The established cut-off of ≥6 failed to identify 46.2% of subjects with a genetic diagnosis of FH. We report significant discrepancies between retrospective scoring and original on-site scoring, regarding not only underestimation but also overestimation, suggesting high susceptibility to errors and varied subjective interpretation of criteria. Since obtaining the DLCNS was difficult even in a specialized clinic, scoring in a broader patient population in primary and secondary care might be not viable. Performance of off-treatment LDL-C only was not inferior to using the DLCNS. However, as with the prevalent scarcity of relevant information and consequential trending of the DLCNS towards approximating a score being driven solely by LDL-C, a sole LDL-C cut-off should be discussed by expert panels and communicated to physicians in a broader range.

Supplementary material

Supplementary material is available in European Journal of Preventive Cardiology.

Author contributions

F.M. wrote the original draft and assisted in conceptualization, data curation, and methodology; G.L., K.K.A., and F.P. assisted in data curation and validation; Y.W. conceived original idea and was responsible for conceptualization, supervision, validation, reviewing, and editing; K.W.A. and S.W.S. helped in supervising the project, validation, and reviewing; B.P.S. and S.P.R. validated genetic methodology and formal analyses; all authors critically revised for important intellectual content and approved integrity and accuracy the final manuscript.

Funding

None declared.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Author notes

Conflict of interest: The authors have no conflicts of interest to report.