Clinical and genetic determinants of cardiomyopathy risk among hematopoietic cell transplantation survivors

Abstract

Cardiomyopathy has been recognized as a complication after hematopoietic cell transplantation (HCT). Using a nested case-cohort design, we examined the relationships between demographic, therapeutic, and selected cardiovascular disease risk factors among 1-year HCT survivors who developed cardiomyopathy before (n=43) or after (n=89) one year from HCT as compared to a randomly selected subcohort of survivors without cardiomyopathy (n=444). Genomic data were available for 79 cases and 267 non-cases. Clinical and genetic covariates were examined for association with the risk of early or late cardiomyopathy. Clinical risk factors associated with both early and late-onset cardiomyopathy included anthracycline exposure ≥250 mg/m² and pre-existing hypertension. Among late-onset cardiomyopathy cases, the development of diabetes and ischemic heart disease further increased risk. We replicated several previously reported genetic associations among early-onset cardiomyopathy cases, including rs1786814 in CELF4, rs2232228 in HAS3, and rs17863783 in UGT1A6. None of these markers were associated with risk of late-onset cardiomyopathy. A combination of demographic, treatment, and clinical covariates predicted early-onset cardiomyopathy with reasonable accuracy (AUC 0.76, 95% CI 0.68–0.83), but prediction of late cardiomyopathy was poor (AUC 0.59, 95% CI 0.53–0.67). The addition of replicated genetic polymorphisms did not enhance prediction for either early or late-onset cardiomyopathy. Conventional cardiovascular risk factors influence the risk of both early and late-onset cardiomyopathy in HCT survivors. While certain genetic markers may influence the risk of early-onset disease, further work is required to validate previously reported findings and to determine how genetic information should be incorporated into clinically useful risk prediction models.

INTRODUCTION

Improvements in hematopoietic cell transplantation (HCT) outcomes have resulted in an expanding population of long-term survivors, many of whom experience significant late effects secondary to pre-transplant and transplant-related treatment exposures.¹ Among these, cardiovascular disease has been increasingly recognized as a significant cause of morbidity and mortality following HCT.² Previously our group described the increased incidence of late post-transplant cardiovascular mortality and morbidity related to ischemic heart disease, stroke, and cardiomyopathy or heart failure when compared to the normal population.³ Additionally, we observed a significantly higher incidence of related conditions that predispose toward more serious cardiovascular disease including hypertension, dyslipidemia, and diabetes. More recently, we and others described the impact of these and other conventional cardiac risk factors on the risk for significant cardiac events including ischemic heart disease, stroke, cardiomyopathy, and cardiac mortality.^4–7

While treatment exposures and the presence of conventional cardiovascular risk factors may help determine the degree of long-term cardiac risk, these factors do not entirely account for the inter-individual variability in cardiomyopathy risk. Numerous genetic polymorphisms have been reported to influence cardiomyopathy risk.^8–18 Many of these studies are based on relatively small numbers of cardiomyopathy cases, and the polymorphisms reported have been inconsistently replicated across studies. In this updated analysis, we expand upon the number of cardiomyopathy cases available from our prior analyses and assess the influence of treatment exposures, conventional cardiovascular risk factors (hypertension, dyslipidemia, and diabetes), and select genetic polymorphisms reported to be influential in the literature.

METHODS

Study population

A variety of data sources previously described in detail were used to ascertain potential cardiomyopathy cases among all 1-year survivors of autologous and allogeneic HCT treated at the Fred Hutchinson Cancer Research Center (FHCRC) from 1970 through 2010 (n=6,903). These included: 1) a National Death Index linkage of deaths from 1979 to 2006;¹⁹ 2) linkage to Washington State hospital discharge and death registry records from 1987 to 2008 (of HCT recipients who were state residents at time of transplant);³ and 3) responses to a patient survey on cardiovascular health sent to all living HCT survivors in 2010 and 2011.⁷ Potential cardiomyopathy cases from the National Death Index and Washington State registry data were identified based on selected International Classification of Diseases-9^th revision codes (Appendix Table 1) or equivalent 10^th revision codes. Cases defined by these administrative data or by patient self-report alone, were further reviewed using available medical records and included in the analysis only if clinical documentation supported the diagnosis. If medical records were unavailable, cases were accepted if the patient was taking medication(s) used to treat heart failure (as distinct from treatment for other conditions such as hypertension alone), or if administrative records corroborated patient self-report. We excluded cardiomyopathy attributable to a transient event (e.g. sepsis) with subsequent recovery of heart function, cardiomyopathy due to diastolic dysfunction alone, and cardiomyopathy with a history of amyloidosis. While imaging (e.g. echocardiogram) data were not required for our case definition, such information along with medication information and initial date of cardiomyopathy onset were abstracted when available.

Individuals who did not respond to the survey, or were not Washington state residents, or had not died would not have been ascertained by these data sources. Therefore, we also reviewed available medical records of 415 additional individuals who otherwise met our eligibility criteria and who had existing genome wide association study (GWAS) data available.²⁰ Overall, through these methods, we were able to determine the cardiomyopathy status of 4,026 of 6,903 ≥1-year survivors (58.3%; Figure 1), and verified 132 cardiomyopathy cases for analysis. Of these, 25 were diagnosed before HCT and 18 were diagnosed within the first year after HCT. Given limited resources that precluded an ability to perform detailed medical record review for exposure assessment on all 4,026 members of this cohort, we used an alternative nested case-cohort study design.²¹ The final analytic population thus included all validated cardiomyopathy cases (n=132), plus a random 10% sample selected from each of our four data sources to serve as the study subcohort (n=444, after excluding overlap with 17 cases; Figure 1).

FIGURE 1.

Study population and data sources for nested case-cohort study of cardiomyopathy among ≥1-year hematopoietic cell transplant survivors.

Exposure information

As previously described,⁶ standard transplant exposures (donor type, stem cell source, and conditioning regimen including total body or lymphoid irradiation) were supplemented with pre-transplant exposures ascertained from the medical records. These included anthracyclines, radiotherapy, and baseline medication use for hypertension, dyslipidemia, or diabetes. Recognizing that there is some uncertainty regarding the most appropriate equivalence formula, to be consistent with prior analyses, anthracycline doses were converted to the equivalent doxorubicin dose: daunorubicin*0.83, epirubicin*0.67, idarubicin*5, and mitoxantrone*4.^22;23 Radiotherapy records were abstracted and exposures classified by the body area exposed: brain, chest, and abdomen/pelvis. Fields that spanned multiple body areas, e.g. spine (chest, and/or abdomen/pelvis depending on extent), had all relevant areas coded as exposed. Post-transplant exposures included systemic immunosuppressive therapy for chronic GVHD, disease relapse, use of medications for hypertension, dyslipidemia, and diabetes post-transplant based on the medical records, and the development of ischemic heart disease (angina, coronary artery disease, myocardial infarct) per the medical records.

Identification of published single nucleotide polymorphisms (SNP) associated with cardiomyopathy and heart failure

To identify candidate SNPs for replication, we reviewed PUBMED for articles published on “cardiomyopathy”, “heart failure”, or “systolic dysfunction” among HCT survivors or cancer survivors as of June 2014. Results from one subsequent abstract presented at a national meeting was also included.¹⁸ Studies were then graded in relation to their similarity with our study population (Appendix Table 2). Category 1 studies were those that featured HCT survivors with similar case ascertainment.¹⁶ Category 2 studies were those that featured non-HCT cancer survivors with similar case ascertainment.^{8;9;11–14;17;18} Category 3 studies were those that featured non-HCT cancer survivors with case-status based primarily on echocardiographic changes.^10;15 Separately, we also identified a fourth category of studies consisting of GWAS focused on cardiomyopathy occurring in the general Caucasian population, restricted to phenotypes distinct from hypertrophic and ischemia-associated cardiomyopathies.^24–29

Sample preparation, genotyping, and imputation

Host DNA, in the form of blood mononuclear cells (for some, subsequently immortalized as EBV-transformed B-lymphocyte cell lines) had been prospectively banked at the time of transplant for 341 (59%) of our subcohort and cases. DNA was isolated using Puregene kits (Qiagen) as previously described, and had been genotyped as part of an existing GWAS dataset on three genotyping platforms: Affymetrix 5.0 Human GeneChip (1,048 HCT recipients; 98 used in this analysis), Illumina 1M Quad (1,904 HCT recipients; 128 in this analysis) and Illumina 2.5M (688 HCT recipients; 115 in this analysis) BeadArrays.²⁰ Amplification and hybridization for the Affymetrix 5.0 array was performed at the Affymetrix Service Laboratory (Santa Clara, CA), and amplification and hybridization for the Illumina BeadArrays was performed by the FHCRC Genomics Shared Resource lab. Data quality was assessed via three different methods: the Affymetrix Bayesian Robust Linear Model with Mahalanobis distance classifier (BRLMM) based “QC call rate”, the clustering call rate and a PCR-based ABO and XY genotyping based sample verification method.²⁰

The genotypes of the candidate SNPs were determined separately for each platform using the BRLMM algorithm for Affymetrix array³⁰ and the GeneCall algorithm for the Illumina arrays.³¹ Candidate SNPs not genotyped on the array were imputed using the 1000 Genomes Project Phase 1 SNPs as a reference panel and the software IMPUTE v2 (http://mathgen.stats.ox.ac.uk/impute/impute_v2.html).³² The posterior probability of the most probable genotype was calculated as the probability of observing an unobserved genotype at the imputed locus, given all of the observed genotypes in the flanking region. The imputed SNP genotype was retained only if the average posterior probability exceeded 0.8, and was treated as missing if the average posterior probability was <0.8. The call rate reflected the percentage of non-missing genotypes in each platform, and SNPs were treated as “failure” and excluded from analyses if the call rate was <0.85. Data quality for genotyping and imputation are summarized in Appendix Table 3.

Four SNPs had low imputed call rates in the Affymetrix array (rs1786814 [CELF4], rs1883112 [NCF4], rs13058338 [RAC2], and rs2235487 [HN1L]), and their values from this array were not used in analysis (Appendix Table 3). All SNPs imputed from the denser Illumina arrays had call rates ≥97%, including those with low call rates using Affymetrix, and were included in the analysis.

Statistical analyses

Because our overall cohort was conditioned on a minimum 1-year HCT survival, a priori, two models were created separately to examine patients who developed cardiomyopathy before (n=43; logistic regression) and after (n=89; Cox proportional hazards regression) one year from HCT in relation to clinical and genetic risk factors. This approach was used because time-to-event analyses of survivors diagnosed with cardiomyopathy before 1-year post-HCT could be confounded by survival bias, since we did not ascertain cardiomyopathy cases or non-cases (denominator subjects) who died during the first year after HCT. Logistic regression therefore was used to examine the prevalence of prior cardiomyopathy diagnosed before 1-year post-HCT. Both logistic and Cox models were adjusted for sex, age at and year of transplant, anthracycline dose (none, <250, ≥250 mg/m²; to match prior studies and new cardiomyopathy surveillance recommendations^4;33), and pre-transplant chest radiation exposure (yes/no). HCT-related exposures were also adjusted for in the Cox models, and included total body irradiation (TBI), donor type (autologous vs. allogeneic), history of chronic GVHD requiring systemic immunosuppressive therapy (yes/no), and any relapse of the pre-transplant underlying disease after HCT (modeled as a time-dependent covariate).

Secondary models examined any additional influence of hypertension, dyslipidemia, or diabetes as baseline covariates at time of transplant (yes/no, defined by being on medication for the respective condition). Among survivors who developed cardiomyopathy ≥1-year post-transplant, we also examined the influence of these same conditions, plus ischemic heart disease as time-dependent covariates. For our Cox models, robust variance estimators were used to account for the nested case-cohort design, and follow-up was censored on December 31, 2012.³⁴ Death from causes other than the specific outcome of interest was treated as a competing risk event.³⁵

In our analysis of candidate SNPs identified in HCT or cancer survivors (category 1–3 studies), we attempted to replicate the genetic and statistical models used in the original publications as closely as possible. When applicable, survivors without anthracycline exposure were excluded. In these analyses, only associations with a one-sided p<0.05 were considered replicated. For general population studies (category 4), we applied Fisher’s exact test (two-sided p-values) to the distribution of alleles among cases and non-cases. We examined survivors who developed cardiomyopathy before and after 1-year post-transplant separately for all analyses. Because our goal was the replication of previously published associations with individualized models designed to approximate those used in each prior publication, we did not adjust for multiple comparisons.

Finally, receiver operating characteristic curves and the corresponding areas under the curve (AUC) were examined to determine the ability of clinical and genetic covariates to predict the risk of early or late cardiomyopathy. Risk scores based on summing the coefficient values from the logistic and Cox models for which the odds ratios (OR) or hazard ratios (HR) values were ≥1.3, respectively, were tested for their predictive power. All analyses were performed using Stata, version 14 (StataCorp, College Station, TX). All study procedures were approved by the FHCRC institutional review board.

RESULTS

Among the study population, cardiomyopathy cases (n=132) were slightly older than non-cases (n=444) at the time of transplant (46.9 vs. 43.3 years) but had similar follow-up duration (10.6 vs. 10.1 years). The distribution of treatment era varied somewhat among non-cases and those who developed cardiomyopathy before 1-year post-HCT (n=43) vs. later (n=89), with early cardiomyopathy more common in the recent era (79.1% of cases transplanted from 2000–2010) while delayed cardiomyopathy occurred more frequently among patients treated earlier (77.5% transplanted from 1970–1999; Table 1). Demographic and clinical characteristics of cases who developed cardiomyopathy prior to transplant (n=25) vs. within 1-year post-transplant (n=18) were similar (data not shown). Among all cardiomyopathy cases, most could be classified as American College of Cardiology Foundation/American Heart Association (ACCF/AHA) stage C heart failure due to a history of prior or current symptoms (62.1%), and over 80% of all cases were receiving chronic medical therapy for cardiomyopathy (Table 2). Although individuals with genetic data (n=341) differed in some respects from those without genetic data (n=235; e.g., more recent transplant era, less likely autologous transplant recipients, less commonly exposed to anthracyclines), the distribution of cardiomyopathy characteristics between the two groups was similar (Table 2; Appendix Table 4).

TABLE 1.

Clinical characteristics of ≥1-year hematopoietic cell transplantation (HCT) survivors analyzed (n=576).

Characteristics	Non-cases N=444 (%)*	Cardiomyopathy cases
		Before 1-year post-HCT N=43 (%)*	After 1-yr post-HCT N=89 (%)*
Female	211 (47.5)	21 (48.8)	45 (50.6)
White race, non-Hispanic ethnicity	387 (87.2)	38 (88.4)	84 (94.4)
Age at transplant, years
<20	78 (17.6)	3 (7.0)	10 (11.2)
20–39	116 (26.1)	10 (23.3)	18 (20.2)
40–59	195 (43.9)	21 (48.8)	55 (61.8)
≥60	55 (12.4)	9 (20.9)	6 (6.7)
Year of transplant
1970–1989	53 (11.9)	3 (7.0)	18 (20.2)
1990–1999	156 (35.1)	6 (14.0)	51 (57.3)
2000–2010	235 (52.9)	34 (79.1)	20 (22.5)
Underlying diagnosis
Acute leukemia, myelodysplasia	177 (39.9)	23 (53.5)	25 (28.1)
Chronic leukemia	91 (20.5)	3 (7.0)	17 (19.1)
Lymphoma	77 (17.3)	8 (18.6)	33 (37.1)
Multiple myeloma	46 (10.4)	6 (14.0)	8 (9.0)
Solid tumor	24 (5.4)	0	4 (4.5)
Non-malignant disorder	29 (6.5)	3 (7.0)	2 (2.2)
Anthracycline, mg/m2†
None	189 (42.8)	9 (21.4)	25 (28.1)
<250	132 (29.9)	9 (21.4)	25 (28.1)
≥250	121 (27.4)	24 (57.1)	39 (43.8)
Chest radiotherapy
Pre-transplant	36 (8.1)	6 (14.0)	7 (8.0)
Total body irradiation (TBI)	243 (54.9)	25 (58.1)	58 (65.2)
Cyclophosphamide conditioning	301 (67.8)	11 (25.6)	64 (71.9)
Donor
Autologous	138 (31.1)	14 (32.6)	40 (44.9)
Related allogeneic	180 (40.5)	15 (34.9)	30 (33.7)
Unrelated allogeneic	126 (28.4)	14 (32.6)	19 (21.3)
History of chronic GVHD	207 (47.6)	20 (50.0)	32 (37.2)
History of post-transplant relapse	89 (20.0)	1 (2.3)	17 (19.1)
Pre-transplant cardiovascular conditions‡
Hypertension	45 (10.2)	14 (32.6)	11 (12.5)
Dyslipidemia	19 (4.3)	6 (14.0)	3 (3.4)
Diabetes	21 (4.7)	2 (4.7)	6 (6.7)
≥1-year post-transplant cardiovascular conditions‡
Hypertension	129 (29.5)	-	45 (50.6)
Dyslipidemia	101 (23.2)	-	44 (49.4)
Diabetes	69 (15.6)	-	35 (39.3)

^*Percentages exclude those with missing information.

^†Anthracycline doses were converted to the equivalent doxorubicin dose: daunorubicin*0.83, epirubicin*0.67, idarubicin*5, and mitoxantrone*4

^‡Defined by being on medication(s) for the specified condition. Estimates shown are adjusted for demographic and treatment exposures mentioned above for each model.

TABLE 2.

Characteristics of validated cardiomyopathy cases among ≥1-year hematopoietic cell transplantation (HCT) survivors (n=132).

Characteristics	All cases N=132 (%)	Cases before 1-year N=43 (%)	Cases after 1-year N=89 (%)
ACCF/AHA heart failure stage
B – asymptomatic left ventricular dysfunction	15 (11.4)	4 (9.3)	11 (12.4)
C – prior/current heart failure symptoms	82 (62.1)	33 (76.7)	49 (55.1)
B or C – exact stage not specified	18 (13.6)	2 (4.7)	16 (18.0)
D – refractory end stage heart failure	15 (11.4)	4 (9.3)	11 (12.4)
Unknown	2 (1.5)*	0	2 (2.2)
Receiving chronic medical therapy
Yes	108 (81.8)	35 (81.4)	73 (82.0)
No	17 (12.9)	5 (11.6)	12 (13.5)
Unknown	7 (5.3)†	3 (7.0)	4 (4.5)
Ejection fraction
≤45% documented	76 (57.6)	36 (83.7)	40 (44.9)
≤40% documented	70 (53.0)	33 (76.7)	37 (41.6)
Other/unknown	49 (37.1)‡	5 (11.6)	44 (49.4)
Cardiomyopathy onset
Before transplant	25 (18.9)	-	-
Between transplant and 1-year	18 (13.6)	-	-
After 1-year	89 (67.4)	-	-

ACCF/AHA, American College of Cardiology Foundation/American Heart Association

^*Medical records note cardiomyopathy but other details unavailable.

^†Includes 2 patients with unknown heart failure stage; others stage C (n=4) or D (n=1).

^‡40 of 49 receiving chronic medical therapy for cardiomyopathy.

Higher anthracycline exposure and pre-existing hypertension were the primary risk factors identified among survivors who developed cardiomyopathy before 1-year post-HCT (OR>4; p<0.01; Table 3). Associations for chest radiotherapy or pre-existing dyslipidemia and diabetes were not statistically significant. Estimates for anthracycline exposure remained robust even after adjustment for hypertension, dyslipidemia, and diabetes. We found no evidence for significant statistical interaction between anthracycline dose and hypertension (p>0.50 for all interaction terms tested).

TABLE 3.

Association between select exposures with early and later onset cardiomyopathy risk among ≥1-year HCT survivors.

Exposure	Before 1-year OR (95% CI)*	After 1-year HR (95% CI)†
Anthracycline, mg/m2
None	1.0 (ref)	1.0 (ref)
<250	1.4 (0.5–3.6)	2.4 (1.2–4.9)
≥250	4.0 (1.7–9.1)	2.6 (1.3–5.2)
Chest radiotherapy
Pre-transplant	1.3 (0.5–3.5)	0.9 (0.3–2.3)
Total body irradiation (TBI)	-	1.3 (0.7–2.6)
Autologous donor	-	2.5 (1.1–5.8)
History of chronic GVHD	-	1.3 (0.6–2.8)
History of post-transplant relapse	-	1.7 (0.8–3.6)
Pre-transplant cardiovascular conditions
Hypertension	4.2 (1.8–9.8)	1.8 (0.7–4.6)
Dyslipidemia	2.2 (0.7–7.1)	1.1 (0.3–4.0)
Diabetes	0.3 (0.05–1.3)	1.7 (0.5–5.7)
≥1-year post-transplant cardiovascular conditions
Hypertension	-	2.1 (1.1–4.1)
Dyslipidemia	-	1.6 (0.8–3.3)
Diabetes	-	2.9 (1.3–6.5)

HR, hazard ratio; OR, odds ratio

^*Logistic regression model adjusted for sex, age at and year of HCT, anthracycline dose, pre-transplant chest radiotherapy (not TBI).

^†Cox proportional hazards models adjusted for same covariates as logistic regression models, plus TBI, donor status, history of chronic GVHD, and history of post-transplant relapse.

Among survivors who developed cardiomyopathy beyond one year from HCT, anthracycline exposure and any history of autologous transplantation (vs. allogeneic) were statistically significant risk factors (Table 3). Pre-transplant cardiovascular conditions were no longer associated, but post-transplant development of hypertension and diabetes were strongly associated with subsequent cardiomyopathy (HR >2; p<0.05). As before, we found no evidence for significant statistical interaction between these traits and prior anthracycline exposure (hypertension, p=0.22; diabetes, p=0.68). Post-transplant ischemic heart disease (n=48) also was an important risk factor for subsequent cardiomyopathy (HR 5.4, 95% CI 2.2–13.3), an association that remained robust even after adjustment for post-transplant hypertension, dyslipidemia, and diabetes (HR 3.9, 95% CI 1.4–11.0). Finally, female gender and exposure to cyclophosphamide-containing conditioning regimens were not associated with early or late-onset cardiomyopathy in any analyses (data not shown). If early- and late-onset cardiomyopathy cases were combined in a single logistic regression model, results were generally similar (Appendix Table 5).

Among early-onset cardiomyopathy cases with genetic data, we replicated the association with rs17863783 in UGT1A6 (OR 19.5, 95% CI 3.5 to 110.5; 1-sided p<0.001; Table 4). Two previously published gene-anthracycline interactions also were replicated (p-value for interaction 0.02 for both): rs2232228 in HAS3 (anthracycline doses >250 mg/m2, AG vs. GG: OR 21.8, 1-sided p=0.02) and rs1786814 in CELF4 (anthracycline doses >300 mg/m2, OR 22.2, 1-sided p=0.01). For rs2232228, the AA vs. GG comparison restricted to anthracycline doses >250 mg/m2 also was associated with increased, albeit imprecise risk (OR 6.1, 95% CI 0.3 to 129.1; 1-sided p=0.12). However, none of these associations were replicated among late-onset cardiomyopathy cases using either our primary proportional hazards models or secondary logistic regression models adjusting for the same covariates as in the original reports. The only association with borderline replication was an increased risk with rs10836235 in CAT (CC genotype associated with HR 2.5 [95% CI 0.9 to 7.4]; 1-sided p-value=0.05). If early- and late-onset cardiomyopathy cases were combined in a single logistic regression model, results were generally similar with the main effect association for rs17863783 (1-sided p<0.001) and interactions with anthracycline dose for rs1786814 (1-sided p=0.01) and rs2232228 (AG genotype, 1-sided p=0.02) still meeting replication thresholds (Appendix Table 5).

TABLE 4.

Replication of single nucleotide polymorphisms (SNP) previously associated with cardiomyopathy in hematopoietic cell transplant (HCT) and/or non-HCT cancer survivors.

Gene (SNP)	Reference	Genotype [R=risk*]	Non-cases† N (%)	Cardiomyopathy before 1-year			Cardiomyopathy after 1-year
				Cases, N (%)†	OR (95% CI)‡	p-value§	Cases, N (%)†	HR (95% CI)**	p-value§
ABCC2 (rs8187710)	Armenian 2013	GG	124 (92.5)	21 (95.5)	1.0 (ref)	0.64	29 (87.9)	1.0 (ref)	0.11
		GA/AA [R]	10 (7.5)	1 (4.5)	0.7 (0.1–6.5)	(0.73)	4 (12.1)	2.7 (0.6–12.8)	(0.21)
CAT (rs10836235)	Rajic 2009	CT/TT	36 (26.9)	4 (18.2)	1.0 (ref)	0.19	5 (15.2)	1.0 (ref)	0.05
		CC [R]	98 (73.1)	18 (81.8)	1.7 (0.5–7.1)	(0.39)	28 (84.8)	2.5 (0.9–7.4)	(0.10)
CBR3 (rs1056892)	Blanco 2008, 2012	AA/GA	160 (61.1)	14 (51.9)	1.0 (ref)	0.19	28 (53.8)	1.0 (ref)	0.19
		GG [R]	102 (38.9)	13 (48.1)	1.5 (0.6–3.3)††	(0.38)	24 (46.2)	1.3 (0.7–2.6)	(0.37)
CELF4 (rs1786814)	Wang 2015	GA/AA	55 (35.6)	11 (47.8)	1.0 (ref)	0.09	11 (28.2)	1.0 (ref)	0.88
		GG [R]	99 (64.3)	12 (52.2)	1.9 (0.7–5.1)‡‡	(0.18)	28 (71.8)	0.6 (0.3–1.4)	(0.25)
HAS3 (rs2232228)	Wang 2014	GG	55 (21.5)	4 (14.8)	1.0 (ref)	0.07	11 (22.4)	1.0 (ref)	0.31
		GA[R]	104 (40.6)	16 (59.3)	2.7 (0.8–9.4)§§	(0.13)	23 (46.9)	1.3 (0.5–3.2)	(0.62)
		AA [R]	97 (37.9)	7 (25.9)	1.4 (0.4–5.8)	0.31 (0.61)	15 (30.6)	1.0 (0.4–2.8)	0.48 (0.96)
HFE (rs1799945)	Armenian, Lipshultz 2013	CC	95 (72.0)	15 (68.2)	1.0 (ref)	0.58	27 (81.2)	1.0 (ref)	0.95
		CG/GG [R]	37 (28.0)	7 (31.8)	0.9 (0.3–2.6)	(0.83)	6 (18.2)	0.4 (0.1–1.2)	(0.10)
NCF4 (rs1883112)	Wojnowski 2005	AA	20 (22.0)	2 (10.5)	1.0 (ref)	0.86	3 (12.0)	1.0 (ref)	0.86
		AG/GG [R]	71 (78.0)	17 (89.5)	0.4 (0.1–2.1)	(0.28)	22 (88.0)	0.5 (0.1–1.9)	(0.28)
RAC2 (rs13058338)	Armenian 2013	TT	55 (51.4)	12 (60.0)	1.0 (ref)	0.88	16 (61.5)	1.0 (ref)	0.93
		TA/AA [R]	52 (48.6)	8 (40.0)	0.5 (0.2–1.6)	(0.25)	10 (38.5)	0.2 (0.02–1.7)	(0.14)
SLC28A3 (rs7853758)	Visscher 2012, 2013	GG [R]	101 (75.4)	15 (68.2)	1.0 (ref)	0.81	24 (72.7)	1.0 (ref)	0.56
		GA/AA***	33 (24.6)	7 (31.8)	1.5 (0.6–3.8)	(0.38)	9 (27.3)	1.1 (0.5–2.2)	(0.89)
UGT1A6 (rs17863783)	Visscher 2012, 2013	GG	130 (97.0)	18 (81.8)	1.0 (ref)	<0.001	31 (93.9)	1.0 (ref)	0.42
		GT [R]	4 (3.0)	4 (18.2)	19.5 (3.5–110.5)	(0.001)	2 (6.1)	1.4 (0.2–10.9)	(0.84)
		TT	0	0	-		0	-

HR, hazard ratio; OR, odds ratio; n/e, not estimable

^*As reported in the referenced publication.

^†Denominator varies based on the population characteristics of the original study.

^‡Logistic regression models adjusting for covariates specified in the original publications.

^§One-sided p-value (2-sided value in parentheses).

^**Cox regression models adjusted for same baseline covariates as logistic models; additional adjustment for post-transplant relapse did not change the significance of any estimate.

^††If restricted to anthracycline doses 1–250 mg/m², OR 2.2 (95% CI 0.3, 14.0; 1-sided p=0.20).

^‡‡Interaction between SNP and anthracycline dose, p=0.02; if limited to anthracycline doses >300 mg/m², OR 22.2 (95% CI 1.5, 339.2; 1-sided p=0.01).

^§§Interaction between AG genotype and anthracycline dose, 1-sided p=0.01; if analysis restricted to anthracycline doses >250 mg/m², OR for AG genotype 21.8 (95% CI 1.2–386.4; 1-sided p=0.02), and OR for AA genotype 6.1 (95% CI 0.3–129.1; 1-sided p=0.12). OR for AG genotype if concurrent adjustment for pre-transplant hypertension and diabetes 3.3 (95% CI 0.8–12.6; 1-sided p=0.04).

^***Per additive model with “A” allele as protective.

Finally, no cardiomyopathy SNP identified in the general population (category 4 studies) was consistently associated in univariate analyses of our case-cohort population (Appendix Table 6). Rs9262636 in HCG22 was possibly differentially distributed among cardiomyopathy cases who presented after one year (p=0.02 overall; p=0.08 if GG vs. AA and AG), but not among early-onset cardiomyopathy cases (p=0.87).

Clinical and genetic covariates were then examined with regards to their ability to predict either early or late cardiomyopathy (Table 5). Among survivors analyzed for risk of early cardiomyopathy, the addition of pretransplant hypertension and dyslipidemia significantly improved prediction (AUC 0.76 vs. 0.68; difference in model fit, p=0.02). The addition of candidate SNPs with marginal associations (ORs ≥1.3) did not further improve prediction; restriction of SNPs to only those with a one-sided p-value <0.1 provided similar results (data not shown). The ability of risk scores derived from the early cardiomyopathy model to predict late cardiomyopathy was poor (AUC~0.6). However, risk scores weighted by coefficients derived from the proportional hazards model specifically examining late cardiomyopathy (limited to exposures present before 1-year post-HCT) also did not improve prediction (AUC 0.61, 95% CI 0.55 to 0.68).

TABLE 5.

Prediction of subsequent risk of cardiomyopathy, area under the curve (95% CI).

Model components	Cardiomyopathy before 1 year		Cardiomyopathy after 1 year
	All patients N=479	Genetic data only N=288	All patients N=525	Genetic data only N=313
Model 1: gender, anthracycline dose category, and pre-transplant chest radiation based on coefficients from early cardiomyopathy logistic regression model	0.68 (0.60–0.76)	0.71 (0.62–0.81)	0.60 (0.54–0.67)	0.57 (0.49–0.66)
Model 2: model 1 plus pretransplant hypertension and dyslipidemia	0.76 (0.68–0.83)*	0.78 (0.69–0.87)†	0.59 (0.53–0.67)	0.60 (0.51–0.68)
Model 3: model 2 plus all category 1–3 SNPs with OR’s ≥1.3	-	0.76 (0.65–0.86)	-	0.58 (0.49–0.67)
Model 4: gender, anthracycline dose category, total body irradiation, donor status, chronic GVHD, pre-transplant hypertension and diabetes, and post-transplant relapse based on coefficients from late cardiomyopathy proportional hazards model	-	-	0.61 (0.55–0.68)	0.62 (0.54–0.71)

^*Difference in model fit, compared with Model 1, p=0.02.

^†Difference in model fit, compared with Model 1, p=0.09.

DISCUSSION

While chemotherapy and radiotherapy-associated cardiomyopathy is a well characterized complication in the general cancer survivor population, data among HCT survivors are limited.² The 132 cardiomyopathy cases reported in this study, even when divided into early and late-onset cases, still represent one of the largest samples among HCT survivors, with most cases (72%) classified as ACCF/AHA heart failure stage C or D. Among pre-transplant therapeutic exposures, our results confirm that cumulative anthracycline dose predicts both early and late-onset cardiomyopathy among HCT survivors.^4;6 Our findings also verify the significance of conventional cardiovascular risk factors, especially hypertension, diabetes, and ischemic heart disease, which have been demonstrated to increase the risk of cardiomyopathy in HCT survivors and in the general cancer survivor population.^4–7;36

Importantly, we limited misclassification by using chart review to ensure that anti-hypertensive agents were not used solely as treatment for cardiomyopathy. However, some misclassification is likely still present, as medication status alone does not account for undiagnosed or undertreated hypertension. Nevertheless, such misclassification would be expected to be conservative, biasing results towards the null. Studies of HCT and other cancer survivors suggest that underdiagnosis and undertreatment of these cardiovascular risk factors occur frequently.^6;37;38 Thus, these cardiovascular risk factors remain important targets for further study and potential intervention among HCT survivors.

Given the limited power of our genetic analysis, we did not attempt to discover new genetic associations, but instead focused on replicating previously reported associations. With approximately 20 to 50 cases with available genetic data, we generally would have had sufficient power (1-sided p-value <0.05) to identify risk ratios >3 across allelic frequencies ranging from 0.1 to 0.3 for dominant and allelic genetic models, but not for recessive models.³⁹ Given the relatively small sample sizes in the literature, many of the risk ratios reported by others have generally been in this range or higher.^{8–10;13;16;17} However, lack of replication by our study for some previously reported associations in cancer survivors with risk ratios <3 (e.g., rs1799945 in HFE, rs1883112 in NCF4, rs13058338 in RAC2) may very well reflect insufficient power.^8;15;16 This same consideration could also explain why associations for non-hypertrophic and non-ischemic cardiomyopathy reported in the general population were not statistically significant in our study. Additionally, although we attempted to recreate the statistical models used by the original studies, inherent differences in study design and study population may also limit our ability to fully replicate prior findings.

Nevertheless, we found some evidence that certain SNPs may influence risk of early cardiomyopathy among HCT survivors. Notably, the SNPs that demonstrated some statistical evidence for association in our study were those previously identified in either large SNP array studies^13;14;17 or through GWAS,¹⁸ supporting the rigor of those study designs. Specifically, we replicated the significance of the rs17863783 genotype within UGT1A6, which has been associated with anthracycline-related cardiomyopathy across several Canadian and Dutch cohorts of childhood cancer survivors.^13;14 UGT1A6 is a UDP glucuronosyltransferase with a potential role in anthracycline metabolism.¹⁴ Altered glucuronidation related to the reduced enzyme activity associated with the rs17863783 genotype could lead to intracellular accumulation of toxic anthracycline metabolites.^40;41 We also replicated the increased cardiomyopathy risk associated with the GG genotype of CELF4 compared with AA or GA genotypes in patients receiving >300mg/m² of anthracyclines, as identified by a Children’s Oncology Group-sponsored GWAS.¹⁸ CELF4 is implicated in alternative splicing of cardiac troponin T, a component of the myocardial sarcomere critical to sarcomeric assembly and myocardial contractility.⁴² Finally, similar to findings from the same Children’s Oncology Group study and replicated within a mostly adult cohort of HCT survivors^16;17, we observed that among patients who received >250mg/m² of anthracyclines, the AG genotype of rs2232228 in HAS3 had a significantly increased cardiomyopathy risk compared to those with the GG genotype. The HAS3 gene encodes for a hyaluronan synthase, which produces hyaluronan, a component of extracellular matrix with a critical role in tissue organization following injury.¹⁷

Despite having a slightly larger sample size (although still with limited statistical power), we were largely unable to replicate these genetic associations for late cardiomyopathy. It is possible that genetic predisposition may have a more influential role among those with earlier-onset cardiomyopathy, a phenomena seen across many other disease states.⁴³ Among survivors with later-onset cardiomyopathy, other exposures such as the subsequent development of hypertension, diabetes, and ischemic heart disease are important, and are likely linked to different genetic factors not evaluated in this study. The importance of these post-transplant conditions was further supported by our observation that pre-transplant clinical covariates predicted late cardiomyopathy relatively poorly compared with prediction of early cardiomyopathy.

In conclusion, conventional cardiovascular conditions among HCT survivors remain important in influencing subsequent risk of both early and later cardiomyopathy. While certain SNPs may influence the risk of cardiomyopathy, especially early-onset disease, further work is needed to evaluate previously published findings and determine how best to incorporate such information into any potential risk prediction models relevant for HCT survivors. Even if replicated SNPs do not enhance overall risk prediction as measured by AUC, consideration of genetic factors may still be important in re-classifying certain individuals from lower to higher risk categories.⁴⁴ Given that many genetic associations also appear to be relatively modest, newer methods that can better identify and incorporate multiple weak biomarkers may be more approrpriate.⁴⁵ Finally, given the relatively small numbers of HCT survivors seen at any institution, multi-institutional collaborations will be important in order to advance this area of research.

HIGHLIGHTS.

• Anthracycline exposure, hypertension, and diabetes increase cardiomyopathy risk

• Polymorphisms in CELF4, HAS3, and UGT1A6 may also be associated

• Cardiomyopathy risk may be predicted using readily available clinical covariates

• Incorporating genetic risk factors to enhance prediction requires further research