Genetic risk factors for drug-induced long QT syndrome: findings from a large real-world case–control study

Abstract

Aim:

Drug-induced long QT syndrome (diLQTS), an adverse effect of many drugs, can lead to sudden cardiac death. Candidate genetic variants in cardiac ion channels have been associated with diLQTS, but several limitations of previous studies hamper clinical utility.

Materials & methods:

Thus, the purpose of this study was to assess the associations of KCNE1-D85N, KCNE2-I57T and SCN5A-G615E with diLQTS in a large observational case–control study (6,083 self-reported white patients treated with 27 different high-risk QT-prolonging medications; 12.0% with diLQTS).

Results:

KCNE1-D85N significantly associated with diLQTS (adjusted odds ratio: 2.24 [95% CI: 1.35–3.58]; p = 0.001). Given low minor allele frequencies, the study had insufficient power to analyze KCNE2-I57T and SCN5A-G615E.

Conclusion:

KCNE1-D85N is a risk factor for diLQTS that should be considered in future clinical practice guidelines.

Plain language summary

Some medications can lead to a condition called drug-induced long QT syndrome (diLQTS), which can be a serious abnormal heart rhythm in some patients. In our research, we explored three specific changes in DNA related to the electrical function of the heart (KCNE1-D85N, KCNE2-I57T, SCN5A-G615E) and their link to diLQTS. Our study revealed a connection between KCNE1-D85N and diLQTS. This study emphasized the importance of including KCNE1-D85N in the medical guidelines to help identify patients at risk of diLQTS. We were unable to identify the connection of KCNE2-I57T and SCN5A-G615E with diLQTS, due to a low number of carriers in the study.

The QT interval on an electrocardiogram (ECG) measures the time period it takes for the heart ventricles to complete their electrical excitation: spanning between the depolarization and repolarization phases of its cardiomyocytes action potentials [1,2]. QT interval depends on heart rate and is often corrected (QTc) with various possible methods [3], with the Bazett method being the most commonly used in clinical practice [4]. Normal QTc interval ranges are approximately 350 to 450 ms for adult males and 360 to 460 ms for adult females [5]. QTc intervals ≥500 ms increase the risk of experiencing cardiac events, such as syncope, aborted cardiac arrest, torsades de pointes (TdP) or sudden cardiac death, by tenfold [6]. QT prolongation can be either congenital (cLQTS) or acquired. cLQTS occurs in approximately 1 in 2000 individuals, and it is caused by specific genetic mutations in ion channels that are involved in the depolarization and repolarization phases of the ventricular action potential [7]. Acquired long QT syndrome is primarly caused by several commonly used US FDA-approved drugs, including antiarrhythmics, antibiotics, antipsychotics and antidepressants [8]. Drug-induced long QT syndrome (diLQTS), characterized by delayed cardiac repolarization due to drugs, primarily results from the inhibition of the rapid potassium inward current (I_Kr) from the Kv11.1 potassium channel [9]. This susceptibility is attributed to the larger inner cavity [10], aromatic residues [11] and hydrophobic pockets [12] in the Kv11.1 channel. diLQTS is a potentially life-threatening adverse drug reaction that can ultimately lead to TdP. diLQTS has also been associated with significantly longer hospital stay (11.5 vs 5.5 days), and tripled risk of all-cause in-hospital mortality among critically ill patients [13]. Although clinical risk factors for diLQTS have been identified, such as female sex, hypokalemia, hypomagnesemia and congestive heart failure [1], they do not explain all risks for diLQTS [14]. Therefore, there is a critical need to identify additional risk factors of diLQTS, so that serious arrhythmias and sudden death can be prevented in patients.

Current evidence strongly suggests that genetics plays a role in the risk of diLQTS [15–17]. A family study revealed significantly higher risk of diLQTS in first-degree relatives of patients with cLQTS compared with the general population [18]. Despite the shared clinical manifestations of cLQTS and diLQTS, recent research indicates that cLQTS mutations are identified in only a minority of diLQTS cases (~10–20%) [19]. This suggests that other genetic variants, characterized by smaller effect sizes and not associated with monogenic cLQTS, likely contribute to the occurrence of diLQTS. According to a recent position statement by the European Society of Cardiology, the heritability of QT interval duration in the general population is about 35% [20], but the heritability estimate of diLQTS remains unknown. Candidate gene studies have primarily focused on variants affecting cardiac ion channels (e.g., KCNE1-D85N, KCNE2-T8A, SCN5A-L1825P, SCN5A-G615E, KCNE2-I57T, KCNE1-D76N), and several uncommon variants (minor allele frequency [MAF] <5%) have been associated with diLQTS [16,17]. However, these previous studies had several limitations that hampered the clinical utility of the pharmacogenetics of diLQTS, such as small sample size, lack of replication and/or assessment in real-world clinical settings. Therefore, this study aims to address the aforementioned limitations by assessing the independent association of candidate genetic variants in cardiac ion channels with the risk of diLQTS in a large observational case–control study in real-world clinical settings.

Materials & methods

Study design

This is a single-center, retrospective case–control study using clinical and genomic data from the Michigan Genomics Initiative (MGI) [21]. MGI is the genomic biobank that integrates whole genome array and imputed genotype data with electronic health records (EHR) at the University of Michigan Health System (also called ‘Michigan Medicine’). MGI participants are primarily recruited while awaiting a diagnostic or interventional procedure either at a preoperative appointment or on the day of their operative procedure at Michigan Medicine. Eligible patients aged 18 years or older who were prescribed at least one dose of a high-risk QT-prolonging drug (see ‘Drug Exposure’ section below for more details) between 1 March 2001 and 30 September 2022 were selected from MGI for this study. QTc data was available starting in 2012. Patients were excluded from the study if they: were diagnosed with congenital long QT syndrome (cLQTS); had any QTc >500 ms before any treatment with a high-risk QT-prolonging drug; were diagnosed with left bundle branch block; and/or used a pacemaker. The study was carried out in accordance with the Declaration of Helsinki and was approved by the local Institutional Review Board with a waiver of informed consent.

Subject selection

Patients included in the study were selected from Michigan Medicine's EHR with MGI genetic data (Figure 1). The initial eligibility criteria involved individuals who had been prescribed at least one dose of a high-risk QT-prolonging drug (Table 1) during the study period. Subsequently, we refined the patient sample by excluding those without any ECG measurements. A single ECG during the prescription for a high-risk drug was sufficient for inclusion in the study, since QTc >500 ms is an established risk factor for severe arrythmias. Moreover, most patients did not have an ECG measured prior to a prescription for a high-risk drug. Pediatric patients were excluded from the study. Given our focus on diLQTS, we also excluded patients with a QTc >500 ms before any treatment with a high-risk QT-prolonging drug, with a diagnosis of congenital long QT syndrome, or patients who were not prescribed a QTc-prolonging drug at the time of their maximum QTc values (sometimes the maximum QTc was measured during gaps in time between prescriptions for QT-prolonging drugs). Additionally, considering the impact of pacemakers and left bundle branch block on QT interval measurement accuracy, we also excluded patients with these conditions from this study. In order to ensure independence among the participants, pairwise genetic relatedness was also considered during patient selection. Patients with close relatives within the sample (kinship coefficients greater than 0.125, in other words, first or second-degree relatives) were randomly selected from each related pair and excluded. Additionally, 59 patients were also excluded because, even though they were genotyped by MGI, their genotype data did not meet quality control standards. The genotype data for the three candidate variants analyzed: KCNE1-D85N, KCNE2-I57T and SCN5A-G615E (see the ‘Candidate variant selection’ section for detail) were checked across all racial groups. However, these variants were not found in any non-white race groups. Consequently, these three variants were exclusively analyzed within the white patient subgroup in our sample, comprising a total of 6083 patients.

Figure 1. . Flow chart with patient selection.

cLQTS: Congenital long QT syndrome; ECG: Electrocardiogram; EHR: Electronic health record; LBBB: Left bundle branch block; MGI: Michigan genomics initiative.

Table 1. . List of high-risk QT-prolonging drugs with a known risk of TdP on the US market from CredibleMeds.

Antiarrhythmics Quinidine Sotalol Dofetilide Disopyramide Flecainide Amiodarone Ibutilide Dronedarone Procainamide	Antibiotics Clarithromycin Azithromycin Levofloxacin Erythromycin Moxifloxacin Ciprofloxacin	Antipsychotics Haloperidol Droperidol Chlorpromazine Pimozine Thioridazine Antidepressants Citalopram Escitalopram	Antifungals Fluconazole Pentamidine Cholinesterase inhibitor Donepezil Opioid agonist Methadone
Phosphodiesterase inhibitors Cilostazol Anagrelide	Antimalarials Chloroquine Hydroxychloroquine Anesthetics Propofol Sevoflurane Cocaine	Anti-cancer Oxaliplatin Vandetanib Arsenic trioxide Cesium Chloride Mobocertinib	Vasodilator Papaverine Antiemetics Ondansetron

Data taken from [8].

Data collection

Patients were initially identified using the DataDirect system [22], a self-serve tool that enables access to discrete clinical data, such as demographics, anthropometry, vital signs, diagnoses, procedures, exams, medications (ordered and administered) and labs (ordered and results). To ensure compliance with exclusion criteria, the Electronic Medical Record Search Engine (EMERSE) was utilized to search through clinical notes (dictated or typed) for terms [23], from cardiology to pathology. Data collection focused around two points in time: the start date for the first prescription for a QT-prolonging drug (Table 1), and the index date was defined as the date of the highest QTc value (Bazett) measured during any high-risk QT-prolonging drug prescription. Baseline refers to the time period prior to the start of any prescription for a QT-prolonging drug. Other medications were collected within a 1-year window around the index date. Age was determined at the index date. The identification of comorbidities relied on ICD-9 and -10 diagnosis codes and lab values, subsequently calculated using Elixhauser and Charlson scoring systems, as previously described [24].

Drug exposure

The list of high-risk QT-prolonging drugs was sourced from the evidence-based list compilation in CredibleMeds [25], a federally-funded, expert-curated, nonprofit website dedicated to foster safe medication practices. CredibleMeds classifies medications into conditional, possible, or known risk for TdP. To prioritize drugs with the highest level of evidence of association with TdP risk, we selected only those categorized as known risk for TdP. The high-risk QT-prolonging drugs list, detailed in Table 1, guided patient selection and subsequent analysis. Medication records were collected regardless of route of administation, formulation, dose or frequency. Therefore, exposure to high-risk QT-prolonging drugs was defined as the presence or absence of any listed medication in Table 1 within the patient's medication list. Additionally, drug–drug interactions with these QT-prolonging drugs were defined as concurrent and systemic use of either cytochrome P450/p-glycoprotein (CYP/p-gp) inhibitors or inducers with known major interaction potential, as outlined by Micromedex^® [26].

Outcomes

The primary outcome was the presence of QTc prolongation, defined as a change of >60 ms from the baseline QTc and/or an absolute QTc value ≥500 ms while patients were prescribed any high-risk QT-prolonging drugs specified in Table 1. Controls had maximum QTc values <500 ms during their prescription(s) for QT-prolonging drug(s). The QTc interval measurements were automatically recorded using a computer-based, FDA-approved electrocardiogram (ECG) system (General Electric [GE] MUSE™ Cardiology Information System) [27], which is the standard system utilized in routine clinical practice at Michigan Medicine. The QT interval measurements were obtained from an average across the 12 leads of the ECG [27], with the primary outcome corrected for heart rate using the Bazett formula (QTc = QT/√ RR). However, recognizing that Bazett's formula tends to overestimate QT duration at extreme heart rates [28], sensitivity analyses were also conducted using two additional QT correction methods: Fridericia (QTc = QT/RR^1/3) [29] and Framingham (QTc = QT + 0.154 × (1 – RR)) [30]. Additionally, if the QRS complex exceeded 120 ms, further QTc correction was performed using QTc = QTc-(QRS-100), in line with established literature [31].

Candidate variant selection

A semi-quantitative scoring system to evaluate the strength of evidence for pharmacodynamic genetic variants as risk factors for diLQTS (i.e., characterized as limited, moderate, strong or definitive evidence) was established [17]. Among the evaluated variants, one displayed definitive evidence (KCNE1-D85N), one exhibited strong evidence (KCNE2-T8A), while four showcased moderate evidence (SCN5A-L1825P, SCN5A-G615E, KCNE2-I57T, KCNE1-D76N). The rest of the 107 genetic variants had limited evidence. For this study, we specifically selected the genetic variants with at least moderate strength of evidence for association with diLQTS, covered by MGI genotyping or imputation, totaling three: KCNE1-D85N, KCNE2-I57T and SCN5A-G615E.

Genomic data

Genomic data were made available by the MGI, as previously published [21]. Briefly, all genotyping was performed at the UM Advanced Genomics Core lab with standard quality checks [32], and using Illumina Infinium CoreExome v12.1 bead arrays^® (Illumina, CA, USA). Quality checks were routinely conducted in batches of genotyped samples, leading to samples excluded from the study based on the following criteria: participant withdrawal, genotype-inferred sex mismatch or missing self-reported sex, atypical sex chromosomal aberration, high kinship coefficient (≥0.45) with a different sample, sample-level call-rate below 99%, technical duplicates or twin sample with higher call rate, contamination level exceeding 2.5%, call rate on any individual chromosome ≤95%, or sample processed in a flagged DNA extraction batch due to technical issue. Imputation was performed using the world-famous Michigan Imputation Server [33] with the Haplotype Reference Consortium r1.1 (HRC) [34] as the reference panel. Standard post-imputation filters were applied to remove poorly imputed variants (r² <0.3 and MAF <0.01%) to ensure a high-quality dataset. The genotypes for all three candidate genetic variants analyzed were imputed. Pairwise genetic relatedness among all patients in the sample was analyzed using Kinship-based INference for GWAS (KING) v2.1.3 [35]. Patients with close relatives within the sample, with kinship coefficients greater than 0.125 (i.e., first or second degree relatives), were randomly selected from each related pair and excluded from the sample.

Statistical analysis

Patients were stratified into two groups based on the primary outcome: without prolonged QTc and with prolonged QTc. Categorical variables were represented as counts and percentages and compared between subgroups using the χ² test (or Fisher's exact test, when necessary). The distribution of the continuous variables was assessed using the Kolmogorov-Smirnov test and visual inspection of distribution plots. Normally distributed continuous variables were represented as mean ± standard deviation (SD) and compared between groups using the Student's t-test. Non-normally distributed continuous variables were represented as median ± interquartile range (IQR) and compared between groups using the Mann-Whitney U Test.

Univariable logistic regression models were used to assess the independent association of clinical variables with prolonged QTc presence. Variables showing significance (p < 0.05) between the groups were considered as clinical predictors, and were included in logistic regression models as covariates for the genetic variants. Given that several clinical variables significantly differed between groups, a propensity score was calculated for each patient using all variables with p < 0.05 [36]. Matching the two groups (with and without prolonged QTc) 1:1 by propensity score formed a new propensity-matched sample. The MatchIt package for RStudio was used to calculate propensity scores and match patients. Subsequently, multivariable logistic regression models were used to assess the association of each of the three genetic variants with diLQTS risk. Odds ratios (OR) and respective 95% confidence intervals (95% CI) were calculated for the unmatched sample in unadjusted and propensity score adjusted models (Models 1 and 3), and for the propensity-matched sample in unadjusted models (Model 2). Model 3 was considered the primary results due to its superior statistical power compared with Model 2. Unlike Model 1, Model 3 incorporated adjustments with clinical covariates. In consideration of the low MAF for each variant, the three candidate genetic variants was determined a priori to be tested using the dominant genetic model (i.e., major allele homozygotes vs heterozygotes + minor allele homozygotes). To minimize potential population stratification, self-identified race groups were analyzed separately. A Bonferroni-corrected p-value of 0.0167 (0.05÷3) for each candidate genetic variant was a priori set as the threshold for statistical significance. Given this Bonferroni-corrected alpha, total sample size available for analysis (n = 6,083), event rate defined by the Bazett QT correction method, and MAF observed for each variant (Table 2), this analysis had 80% power to detect odds ratios of 2.21, 9.03 and 44.85 for KCNE1-D85N, KCNE2-I57T and SCN5A-G615E, respectively. Furthermore, an exploratory analysis was conducted based on specific drug class and individual drugs, if at least 500 patients were treated with any of the high-risk QT-prolonging drugs. All statistical analyses were performed using R version 4.2.2.

Table 2. . Allele frequency, genotype frequency and Hardy-Weinberg equilibrium assessment of the candidate genetic variants for diLQTS among self-reported white patients.

SNPs	Allele frequency in non-Finnish European population in gnomAD	Allele frequency observed in sample	Genotype frequencies, n (%)	HWE (p-value)
SCN5A-G615E rs12720452	0.0005178	0.000329	CC: 6079 (99.9) CT: 4 (0.1) TT: 0 (0.00)	0.980
KCNE2-I57T rs74315448	0.001045	0.000986	TT: 6071 (99.8) TC: 12 (0.2) CC: 0 (0.00)	0.938
KCNE1-D85N rs1805128	0.01223	0.009206	CC: 5971 (98.1) CT: 112 (1.9) TT: 0 (0.00)	0.469

HWE: Hardy-Weinberg equilibrium.

Results

Clinical characteristics

A total of 6989 eligible patients met the study inclusion/exclusion criteria (Figure 1): 6083 (87.0%) self-reported to be White, 565 (8.1%) African–American, 111 (1.6%) Asian, 46 (0.7%) American Indian or Alaska Native, 2 (0.03%) Native Hawaiian or Other Pacific Islander, and 182 (2.6%) unknown race. Genotype data for all three candidate variants, KCNE1-D85N, KCNE2-I57T and SCN5A-G615E, were checked across all racial groups, but none of these variants were found in any of the non-white race groups. This finding is consistent with the reported MAF in gnomAD, where these three variants are most frequent among European populations and rare in non-European populations. Consequently, these three variants were exclusively analyzed within the white patient subgroup in our study. The genotype and allele frequencies for these three genetic variants among white patients are shown in Table 2. All genotype frequencies were in Hardy-Weinberg equilibrium with p-values >0.05, and all of the allele frequencies were similar to those previously reported for Europeans in gnomAD.

The primary outcome of QTc prolongation occurred in 12.0% of all patients, with a prevalence of 11.4% in women and 12.8% in men (p = 0.103 for sex-based difference). The clinical characteristics of both groups are shown in Table 3. Overall, before propensity score matching, patients with prolonged QTc exhibited a significantly higher prevalence of electrolyte disturbances (hypokalemia, defined as potassium <3.5 mEq/l and hypocalcemia, defined as calcium <8.5 mg/dl), renal (chronic kidney disease), liver and cardiovascular conditions (congestive heart failure, coronary artery syndrome, hypertension, peripheral vascular disease, arrhythmias and history of myocardial infarction and stroke), chronic obstructive pulmonary disease, and diabetes mellitus compared with patients without prolonged QTc. Moreover, patients with prolonged QTc had a significantly higher prevalence of history of alcohol consumption, use of loop diuretics, digoxin and beta-blockers, than those without prolonged QTc. Elixhauser and Charlson comorbidities scores were also significantly higher in patients with prolonged QTc than those without prolonged QTc. In contrast, patients without prolonged QTc had a higher prevalence of concomitant usage of over two QTc-prolonging drugs, compared with those with prolonged QTc. There were no significant differences observed in any of the other variables assessed such as age, sex, body mass index, hypothyroidism or cancer. For the subset of patients that had baseline ECG data available (N = 433, 7% of total sample size), the median number of days [interquarile range] between the baseline QTc and the index date was 610 [1011] and 844 [1458] for those without QTc prolongation and with prolonged QTc, respectively (p = 0.519). The time between the start date of the QT-prolonging drug and the index date was also similar between the two groups (Median [IQR]: 1622 [1395] for those without QTc prolongation and 1607 [1457] for those with QTc prolongation; p = 0.964). For the purpose of matching and adjustment, age was dichotomized at 68 years old because age at 68 years old has been previously associated as an independent risk factor for diLQTS in hospitalized patients [37]. After applying a 1:1 propensity score matching, 733 patients were included in each group, effectively eliminating all of the significant differences in the clinical characteristics. In addition, no significant differences were found in clinical characteristics between carriers and non-carriers of the three candidate genetic variants (except for age and use of digoxin, which were significantly higher in the four carriers of SCN5A-G165E; Supplementary Table 1).

Table 3. . Clinical characteristics compared between patients without and with prolonged QTc in the unmatched and propensity-matched samples.

Characteristics	Unmatched sample		p-value	Propensity-matched sample		p-value
	Without prolonged QTc	With prolonged QTc		Without prolonged QTc	With prolonged QTc
Patients, n (%)	5,350 (88.0)	733 (12.0)	__	733 (50.0)	733 (50.0)	__
Aged 68 years or older, n (%)	2,967 (55.5)	403 (55.0)	0.838	297 (40.5)	263 (35.9)	0.076
Female, n (%)	2,985 (55.8)	385 (52.5)	0.103	373 (50.9)	385 (52.5)	0.565
Hypokalemia, n (%)	325 (6.1)	103 (14.1)	<0.001	86 (11.7)	103 (14.1)	0.212
Hypocalcemia, n (%)	1,005 (18.8)	298 (40.7)	<0.001	317 (43.2)	298 (40.7)	0.341
Hypomagnesemia, n (%)	1,059 (19.8)	128 (17.5)	0.149	155 (21.1)	128 (17.5)	0.085
BMI (kg/m2), mean (SD)	30.3 (7.5)	29.9 (7.9)	0.343	30.7 (7.5)	29.9 (7.9)	0.070
CHF, n (%)	369 (6.9)	148 (20.2)	<0.001	137 (18.7)	148 (20.2)	0.509
HTN, n (%)	3,010 (56.3)	495 (67.5)	<0.001	509 (69.4)	495 (67.5)	0.465
CAD, n (%)	2,279 (42.6)	387 (52.8)	<0.001	393 (53.6)	387 (52.8)	0.794
Hypothyroidism, n (%)	754 (14.1)	105 (14.3)	0.911	119 (16.2)	105 (14.3)	0.345
History of MI, n (%)	374 (7.0)	96 (13.1)	<0.001	91 (12.4)	96 (13.1)	0.754
DM, n (%)	1,097 (20.5)	222 (30.3)	<0.001	234 (31.9)	222 (30.3)	0.535
DM complicated, n (%)	353 (6.6)	87 (11.9)	<0.001	91 (12.4)	87 (11.9)	0.810
History of Stroke, n (%)	529 (9.9)	121 (16.5)	<0.001	121 (16.5)	121 (16.5)	1.000
CKD, n (%)	707 (13.2)	184 (25.1)	<0.001	202 (27.6)	184 (25.1)	0.313
Severe liver disease, n (%)	82 (1.5)	30 (4.1)	<0.001	35 (4.8)	30 (4.1)	0.612
Liver disease, n (%)	598 (11.2)	115 (15.7)	0.001	134 (18.3)	115 (15.7)	0.211
COPD, n (%)	429 (8.0)	96 (13.1)	<0.001	97 (13.2)	96 (13.1)	1.000
Cancer, n (%)	385 (7.2)	51 (7.0)	0.887	53 (7.3)	51 (7.0)	0.925
PVD, n (%)	458 (8.6)	113 (15.4)	<0.001	118 (16.1)	113 (15.4)	0.774
Arrhythmia, n (%)	974 (18.2)	223 (30.4)	<0.001	209 (28.5)	223 (30.4)	0.456
History of alcohol consumption, n (%)	189 (3.5)	47 (6.4)	<0.001	44 (6.0)	47 (6.4)	0.829
Loop diuretic, n (%)	659 (12.3)	261 (35.6)	<0.001	254 (34.7)	261 (35.6)	0.743
Digoxin, n (%)	21 (0.4)	16 (2.2)	<0.001	15 (2.0)	16 (2.2)	1.000
Beta-blockers, n (%)	1,453 (27.2)	341 (46.5)	<0.001	350 (47.7)	341 (46.5)	0.676
>2 QT-prolonging drugs, n (%)	2,771 (51.8)	272 (37.1)	<0.001	274 (37.4)	272 (37.1)	0.957
Elixhauser score (IQR)	7.0 (13.0)	11.0 (14.0)	<0.001	12 (14.0)	11 (14.0)	0.547
Charlson score (IQR)	3 (5.0)	4 (5.0)	<0.001	5 (5)	4 (5)	0.727
Days between start date of QT-prolonging drug and maximum QTc value (IQR)	1622 (1395)	1607 (1457)	0.964	1739 (1368)	1607 (1457)	0.108

Bolded p-values <0.05.

BMI: Body mass index; CAD: Coronary artery disease; CHF: Congestive heart failure; CKD: Chronic kidney disease; COPD: Chronic obstructive pulmonary disease; CrCl: Creatinine clearance; DM: Diabetes mellitus; HTN: Hypertension; IQR: Interquartile range; MI: Myocardial infarction; PVD: Peripheral vascular disease.

Drug exposure

Out of the 40 drugs marketed in the US that are classified as known risk for TdP by CredibleMeds (Table 1), 27 (67.5%) were prescribed in our sample (Table 4). Dofetilide, dronedarone, sotalol and papaverine were prescribed significantly more frequently for those who had prolonged QTc than those without prolonged QTc. In contrast, azithromycin, ciprofloxacin, propofol and ondansetron were prescribed significantly more frequently for those without prolonged QTc than those with prolonged QTc.

Table 4. . Comparison of QTc-prolonging drugs prescribed to patients without and with prolonged QTc.

Drug class	Drugs, n (%)	Without prolonged QTc (n = 5,350) (88.0%)	With prolonged QTc (n = 733) (12.0%)	p-value
Antiarrhythmics	Disopyramide	7 (0.1)	0 (0.0)	0.604
	Dofetilide	14 (0.2)	24 (2.6)	<0.001
	Dronedarone	2 (0.0)	3 (0.3)	0.021
	Flecainide	40 (0.7)	10 (1.1)	0.212
	Quinidine	0 (0.0)	1 (0.1)	0.137
	Sotalol	39 (0.7)	21(2.3)	<0.001
	Procainamide	2 (0.0)	1 (0.1)	0.357
Antibiotics	Azithromycin	556 (9.5)	26 (2.8)	<0.001
	Ciprofloxacin	607 (10.4)	54 (5.8)	<0.001
	Clarithromycin	4 (0.1)	1 (0.1)	0.521
	Levofloxacin	290 (5.0)	34 (3.7)	0.102
	Moxifloxacin	11 (0.2)	1 (0.1)	1.000
	Erythromycin	11 (0.2)	3 (0.3)	0.426
Antidepressants	Citalopram	414 (7.1)	43 (4.6)	0.007
	Escitalopram	382 (6.5)	49 (5.3)	0.167
Cholinesterase inhibitor	Donepezil	46 (0.8)	10 (1.1)	0.332
Opioid agonist	Methadone	54 (0.9)	10 (1.1)	0.587
Antipsychotics	Haloperidol	66 (1.1)	12 (1.3)	0.787
	Chlorpromazine	9 (0.2)	2 (0.2)	0.655
Antifungals	Fluconazole	388 (6.7)	58(6.3)	0.712
	Pentamidine	2 (0.0)	2 (0.2)	0.093
Antimalarial	Hydroxychloroquine	173 (3)	33 (3.6)	0.378
Anesthetic	Propofol	2,263 (38.8)	275 (29.7)	<0.001
Antiemetic	Ondansetron	4315 (74.0)	537 (58.0)	<0.001
Oncology	Oxaliplatin	16 (0.3)	2 (0.2)	1.000
Vasodilator	Papaverine	38 (0.7)	21 (2.3)	<0.001
Phosphodiesterase Inhibitor	Cilostazol	32 (0.5)	2 (0.2)	0.312

Bolded p-values indicate p less than 0.05.

Associations of candidate genetic variants with the risk of drug-induced long QT syndrome 

The association of KCNE1-D85N (rs1805128) and KCNE2-I57T (rs7415448) with the risk of diLQTS, as defined by Bazett's QT correction method is displayed in Figure 2. KCNE1-D85N met the Bonferroni-corrected level of statistical significance (p < 0.0167) in all models: Model 1 (p = 0.006), Model 2 (p = 0.009) and Model 3 (p = 0.001). The odds ratios for KCNE1-D85N were 1.9, 2.9 and 2.2 in Models 1, 2, and 3 respectively. KCNE2-I57T did not meet the Bonferroni-corrected level of statistical significance in any of the models using the Bazett QTc correction method (p > 0.1 in all 3 models). The odds ratios for KCNE2-I57T were 2.4, 1.5 and 1.5 in Models 1, 2 and 3, respectively. Only four total patients carried the SCN5A-G615E (rs12720452) variant, and none of the carriers had prolonged QTc. Thus, SCN5A-G615E was not included in Figure 2. However, complete regression results for all three variants are displayed in Supplementary Table 2.

Figure 2. . Forest plot of the logistic regression models assessing the association of candidate genetic variants with diLQTS in white patients (dominant genetic model).

Model 1: Unmatched sample with unadjusted model (total n = 6,083). Model 2: Propensity-matched sample with unadjusted model (total n = 1466). Model 3: Unmatched sample with model adjusted for propensity score (total n = 6083). *p < 0.0167 (Bonferroni correction for multiple comparisons) The squares represent the odds ratio of the individual candidate gene variant in the three models and the horizontal lines indicate the 95% confidence interval.

OR: Odds ratio.

Since Bazett's formula for correcting QT interval tends to overestimate the QT duration at extreme heart rates, sensitivity analyses were also performed using the Fridericia and Framingham QT correction methods. The findings are shown in Supplementary Table 3. Regardless of the QT correction method, none of the four SCN5A-G615E carriers were in the prolonged QTc group. Therefore, the results for SCN5A-G615E were similar across the three QTc correction methods. However, the results for KCNE1-D85N and KCNE2-I57T slightly differed among the three QTc correction methods. KCNE1-D85N did not meet the Bonferroni-corrected level of statistical significance in any of the three models with Fridericia QT correction method or Model 1 or Model 2 with the Framingham correction method (p ≥ 0.04). However, KCNE1-D85N was still statistically significant in Model 3 with the Framingham QT correction method (OR: 2.25; 95% CI: 1.11–4.16; p = 0.015). In contrast, even though KCNE2-I57T was not statistically significant in any of the models using the Bazett correction method, KCNE2-I57T was statistically significant in Model 1 with the Fridericia QT correction method (OR: 6.53; 95% CI: 1.74–20.80; p = 0.002) and Models 1 and 3 with the Framingham QT correction method (KCNE2-I57T in Model 3 with Framingham: OR: 6.70; 95% CI: 1.51–24.85; p = 0.006).

Given that KCNE1-D85N was the most frequent of the three variants, we also conducted exploratory analyses for KCNE1-D85N by specific drug class and individual drugs if at least 500 patients were treated with any of the individual drugs. The findings are shown detailed in Supplementary Figure 1. Overall, all p-values exceeded 0.3 across various individuals drugs (ondasentron, azithromycin, propofol, ciprofloxacin) and drug classes (antiarrhythmics, antibiotics, antidepressants/antipsychotics, antifungals). Although these subgroup analyses are underpowered, these results do not suggest that KCNE1-D85N affects risk differently among different drugs/drug classes.

Discussion

To the best of our knowledge, this is the largest study that investigated the association of candidate genetic variants with the risk of diLQTS. The largest previous candidate gene studies with a similar endpoint all had <100 cases [38–40]. We analyzed three candidate genetic variants with moderate-to-high level of prior evidence for their association with diLQTS in a previous literature review and had genotype data available in our study: KCNE1-D85N, KCNE2-I57T and SCN5A-G615E [17]. Our results further support KCNE1-D85N as significant risk variant for diLQTS. The study had insufficient power to provide precise estimates for KCNE2-I57T and SCN5A-G615E.

KCNE1-D85N & drug-induced long QT syndrome 

KCNE1 encodes the beta subunit of voltage-gated potassium channels, a pivotal player in modulating the alpha subunit of the cardiac channel encoded by KCNQ1. Together, KCNQ1 and KCNE1 form a complex that generates the slowly activating potassium current, also known as I_Ks. This current, in conjunction with the rapidly activating potassium current (I_Kr), results in cardiac repolarization. D85N leads to a substitution of asparagine with aspartic acid in the C-terminal of the beta subunit. When examined in heterologous expression models, KCNE1-D85N abolished up to 50% of KCNQ1-encoded currents in the absence of drugs [41]. Additionally, in vitro studies have found that the cell membrane expression of KCNE1-D85N is 20% lower than that of the wild-type [42]. While there is no direct effect between D85N and drugs, the susceptibility to diLQTS in D85N carriers may be attributed to a diminished ‘repolarization reserve’. Repolarization reserve refers to redundancy in mechanisms compensating for the inhibition of Kv11.1 [43]. These in vitro findings have been translated in both clinical candidate gene association studies [44–46] and a whole-exome sequencing study [47], thereby confirming the association between KCNE1-D85N and diLQTS. On the other hand, several other candidate gene studies did not identify the association between the D85N variant and diLQTS [40,48–56]. This lack of association may be attributed to the small sample size of cases in the negative studies (largest sample size consisting of only 77 cases) and the low MAF of this variant (only ~1% in Europeans). Consequently, it is plausible that these studies might be underpowered and have introduced type 2 errors. By analyzing this variant in a much larger sample size herein, we heighten the confidence that the association between KCNE1-D85N and diLQTS is not either spurious or a false-positive, especially after meeting the Bonferroni corrected level of significance. Moreover, an added strength of our investigation lies in utilizing authentic patient data from real-world clinical settings, diverging from the highly controlled clinical trial in a previous study [54]. This improvement markedly reinforces the relevance and applicability of our findings in real-world clinical settings.

KCNE2-I57T & drug-induced long QT syndrome

The KCNE2 gene, very similar to KCNE1, encodes the beta subunit of voltage-gated potassium channels including the hERG channel, which generate I_Kr [57]. The I57T variant occurs in the transmembrane segment of the accessory β-subunit, and can decrease I_Kr by 34% [38]. Previous candidate gene [38,58] and in vitro studies supporting functional effects [38,59–61] have found KCNE2-I57T to be significantly associated with diLQTS. However, there are other studies that have yielded contradictory findings regarding this variant [40,44,45,47,52,53,62]. In our study, KCNE2-I57T did not meet the Bonferroni-corrected level of statistical significance in any of the models using the Bazett QTc correction method. However, some of the sensitivity analyses using the Fridericia and Framingham QTc correction methods were statistically significant. These discrepancies arising from our findings for the I57T variant could be explained by the decrease in statistical power due to its lower allele frequency (approximately 0.1% MAF in Europeans). In contrast to the D85N variant, the occurrence of the I57T variant was notably rarer, approximately 9.5-times less frequent within our study. Therefore, whether or not the I57T variant is a definitive risk factor for diLQTS remains unclear. To determine the ultimate implication of this variant in diLQTS, it is imperative to perform further investigations encompassing sample sizes that exceed that of our current study.

SCN5A-G615E & drug-induced long QT syndrome

SCN5A encodes the α-subunit of the cardiac sodium channel, crucial for the rapid depolarization upstroke of the cardiac action potential [63]. The G615E occurs in the intracellular domains of the sodium channel α-subunits. However, there have been conflicting findings regarding its functional impact in experimental studies [39,64]. While one study in Xenopus oocytes indicated that G615E reduced recovery times from fast inactivation, affecting cardiac sodium channel activity [64], a contradictory study by Yang et al. revealed that this variant did not alter the amplitude or voltage dependence of sodium channel activation/inactivation in mammalian expression vectors, suggesting it played no role in the diLQTS phenotype [39]. Several earlier candidate gene association studies reported an association of G615E with diLQTS [39,62,65]. However, in various other candidate gene association studies, G615E was not found in any diLQTS cases [40,44,45,49,53,55,62,66–68], possibly due to its rarity (approximately 0.05% MAF in Europeans). Previous pharmacogenomic analyses indicated that rare genetic variants might have much larger effect sizes than common ones [69]. Therefore, while a larger expected effect size for G615E might compensate for lower statistical power due to its very low MAF, our study's power estimate demonstrated that even a large sample size like ours was severely underpowered to detect a significant association for G615E, even with a potentially substantial effect size. Thus, more studies are needed to confirm the association of SCN5A-G615E with diLQTS risk. Despite the very small number of carriers, it is worth noting that none of the four G615E carriers in our study exhibited prolonged QT.

Clinical implication

The ultimate goal concerning KCNE1-D85N involves the clinical implementation of pharmacogenetic testing to prevent diLQTS. Currently, there is a lack of clinical practice guidelines or recommendations from authorities like the US FDA and Clinical Pharmacogenetics Implementation Consortium regarding the management of carriers of this variant. A recent position statement by the European Society of Cardiology emphasized the relevance of personalized treatment and risk stratification for diLQTS [20], but it also stressed the need for comprehensive validation studies to effectively translate KCNE1-D85N into clinical practice. Conducting randomized clinical trials to showcase the clinical utility of pharmacogenetic testing for these variants is probably not feasible due to their low MAF [70]. Despite this, the wealth of evidence from comprehensive studies aligned with our large-scale investigation suggest that KCNE1-D85N warrants formal evaluation for publication in a clinical practice guideline, and possibly for clinical implementation as well. The recent, landmark clinical trial published by the Ubiquitous Pharmacogenomics Consortium demonstrated the feasibility and efficacy of a pharmacogenetic testing panel in reducing the risk of severe adverse drug reactions [71], suggesting that a pharmacogenomic panel incorporating D85N would be feasible with minimal upfront costs and lifelong applicability for the patient. This approach is suggested to minimize prescription complications and to improve healthcare with substantial cost-benefit perspective.

Limitations

This study has several limitations that require careful consideration. Primarily, it was a retrospective and observational study conducted within a single health system, possibly restricting the generalizability of our findings. A baseline ECG, in other words, prior to the prescription of any high-risk drug, was not available for most patients. Thus, even though we excluded patients with a known diagnosis for long QT syndrome, we assumed that most patients did not have prolonged QTc prior to the prescription. Our reliance on comorbidity data extracted from the EHR using ICD-9 and ICD-10 codes is known to be not entirely accurate. Patients were categorized based on self-identified race groups instead of ancestry, although there is a very high agreement between self-identified white patients in MGI and European ancestry [21]. MGI participants are predominantly recruited while awaiting a diagnostic or interventional procedure; as a result, our sample is not a random representation of patients at Michigan Medicine, potentially introducing selection biases. Additionally, our study focused solely on drugs posing a high risk of TdP, whereas there are over 100 drugs with a conditional risk for TdP [8], which were not included in our analysis. Furthermore, the data collected regarding patients' medication lists encompassed only prescribed medications, and assessing adherence to these prescriptions was not feasible. Our analysis only considered pharmacodynamic variants with at least moderate prior evidence of association with diLQTS; thus excluding the potential contribution of other pharmacodynamic variants with lower levels of prior evidence as well as pharmacokinetic variants, which might also influence diLQTS risk [17].

While Bazett's formula is the most widely adopted in clinical practice and is preferred for long QT syndrome types 1 and 2 [72], our study recognizes its tendency to overestimate the number of patients with potentially dangerous QTc prolongation. Notably, the Fridericia and Framingham correction formulae have demonstrated superior rate correction and a better estimate of 30-day and 1-year mortality [73]. The robustness of our findings is evident in the consistency of associations for KCNE1-D85N in Model 3 across the three QT correction methods, as shown in Supplementary Table 3. The results in Model 3 of the Framingham study with Bazett further reinforces the reliability of our conclusions. However, it is worth noting that although the p-value for D85N with Fridericia in Model 3 did not achieve statistical significance, the odds ratio was nearly 2. While the result did not reach conventional significance levels, the proximity to significance suggests that KCNE1-D85N, when using Fridericia, could potentially be a risk factor for diLQTS, and our study was underpowered to detect it. Finally, QT prolongation, the focus of our study, serves as a surrogate measure for more severe clinical outcomes, such as TdP and sudden cardiac death.

Conclusion

This is the largest study of candidate genetic variants in cardiac ion channels associated with the risk of diLQTS to date. KCNE1-D85N significantly associated with diLQTS risk, while the study was underpowered to determine associations for KCNE2-I57T and SCN5A-G615E with diLQTS. Incorporating these findings into a clinical pharmacogenomic program may translate to preventing adverse events.

Summary points.

• KCNE1-D85N, KCNE2-I57T and SCN5A-G615E have been previously associated with drug-induced long QT syndrome (diLQTS), but several limitations of previous studies hamper clinical utility.

• The purpose of this study was to assess the associations of KCNE1-D85N, KCNE2-I57T, and SCN5A-G615E with diLQTS in a large observational case–control study (6,083 self-reported white patients treated with 27 different high-risk QT-prolonging medications).

• A significant association between KCNE1-D85N and diLQTS was identified (adjusted odds ratio: 2.24 [95% CI: 1.35–3.58]; p = 0.001).

• This study was underpowered to determine the associations of KCNE2-I57T and SCN5A-G615E with diLQTS.

• Incorporating KCNE1-D85N into a clinical pharmacogenomic program may prevent diLQTS.