WP5 aimed to identify clinical and genetic risk markers in a large cohort of CCS. For eligibility in the PCL WP5 genetic study, the following inclusion criteria were applied: (1) patients were younger than age 18 years at cancer diagnosis; (2) patients had been treated for cancer with cisplatin, carboplatin, or both; (3) patients were off therapy and had at least one pure tone audiometric evaluation available after the end of chemotherapy; and (4) patients had provided biomaterial (saliva or blood) for DNA extraction (Figure 1). Subjects were excluded from the study if they had permanent hearing loss identified before the start of cancer therapy.

WP4b used a subset of research subjects from WP5 and aimed to identify additional genetic risk markers of hearing loss by GWAS with better control of nongenetic risk factors. An additional set of inclusion criteria was imposed to reduce potentially confounding factors and focus on the influence of cisplatin. These were (1) cancer treatment included upfront cisplatin, either cisplatin as the single platinum drug throughout the entire course of treatment or change from cisplatin to carboplatin during treatment, and (2) no radiotherapy administration to the brain or inner ear (Figure 1). Research subjects whose initial treatment included carboplatin were excluded.

The PanCareLIFE study has been approved by the local ethics committees: Kantonale Ethikkommission Bern, 362/2015; Comitate Etico Regionale, 507REG2014; Ethical Committee University Hospital Brno, June 11, 2016; Ethics Committee Fakultni Nemocnice v Motole, Prague; De Videnskabsetiske Komiteer Region Hovedstaden, H-1-2014-125; Ethikkommission Medizinische Universität Graz, 27-015 ex 14/15; Ethikkommission der Universität Ulm, 160/17; Ethikkommission der Universität zu Lübeck, 14/181; Ethik-Kommission der Ärztekammer Westfalen-Lippe und der Westfälischen Wilhelms-Universität Münster, 2014-619; Medische Ethische Toetsings Commissie Erasmus MC; Medisch Ethische Toetsingscommissie, 2015_202. Informed consent was obtained from the patient or their the legal representatives.

CCS were recruited through an institutional network from several countries in Europe. The participating institutions are referred to as data providers. Each data provider collected retrospective demographic, diagnostic, treatment, and audiometric data from their medical record files and registries. Diagnostic data included the International Classification of Disease –coded diagnosis and date of diagnosis (for each tumor). Treatment-related data included information on platinum treatment (eg, platinum compound, dose per cycle, cumulative dose, date of start and stop treatment, and infusion duration) and potentially ototoxic comedication (eg, amikacin, gentamycin, tobramycin, furosemide, vincristine, vancomycin). The data were stripped of all identifiers and assigned a unique PCL-ID number, rendering the data pseudonymous for the investigators of this study. Data providers sent their data to the PCL data center in Mainz (German Childhood Cancer Registry, University Medical Center Mainz, Germany), which also collected and archived the genetic, clinical, and audiological data from the lab and the audiometry center.

The main outcome of this study was hearing function following platinum treatment in pediatric cancer survivors. Diagnosis of hearing loss was based on pure tone audiometry performed at frequencies of 250, 500, 1000, 2000, 4000, 6000, and 8000 Hz [22]. Data providers sent the pseudonymized original audiograms to the audiometry center (Department of Phoniatry and Pediatric Audiology, University Hospital Muenster, Germany) for standardized review. The audiogram assessors were blinded to patient characteristics including their treatment, such as platinum compound, platinum dose, or cranial irradiation. The assessors graded the severity of hearing loss using the Muenster criteria [23,24]. The Muenster criteria considers minimal hearing loss (Muenster grade 1: >10 to ≤20 dB) and allows the detection of early post-cisplatin hearing loss. Clinically relevant hearing loss was defined as Muenster grade ≥2b [24]. The International Society of Pediatric Oncology (SIOP) Boston criteria [25,26] was used as an independent, secondary grading. Clinically relevant hearing loss was defined as SIOP grade ≥2 (Table 1).

Detailed methods for DNA collection and genotyping are described elsewhere [27]. For the first aim of the genotype study, a candidate gene approach was applied to validate 10 previously identified single nucleotide polymorphisms (SNPs) associated with hearing loss in childhood cancer patients and survivors [19,28-40]: ACYP2, LRP2, NFE2L2, OTOS, TPMT, SOD2, SLC22A2, GSTP1, ABCC3, and SLC16A5 [30,32,35,38,40-43]. In WP4b, array-based genotyping data was used. Thereafter, these data were merged with data from TaqMan PCR generated in WP5. Next, novel SNPs that were independently associated with treatment-related hearing loss in childhood cancer patients were explored within WP4b by GWAS. For maximum standardization, array genotyping in the PCL consortium was conducted by one partner (genetic laboratory of the Department of Internal Medicine in the Erasmus Medical Center, Rotterdam, the Netherlands) [44]. The Infinium Global Screening Array (Illumina, Inc), which contains >770,000 SNPs, was used [45].

A stringent quality control protocol was applied where multiple filters were used to ensure the quality of the genetic data prior to either imputations or analysis. The quality control procedure is described elsewhere [27]. To remove poorly genotyped SNPs and individuals from the data, a call rate of 97.5% was applied. In addition, a Hardy-Weinberg equilibrium test (P<1*10^–7) was assessed to identify potential genotyping errors. Samples with gender mismatches, familial relationships, and extreme heterozygosity were removed to ensure sample quality. After the quality control, principal components were calculated in order to adjust for population heterogeneity and technical confounders in all subsequent analyses [46]. Imputations were performed using the Michigan Imputation Server with default settings [47]. The reference panel chosen for imputations was the Haplotype Reference Consortium (HRC r1.1) [48]. This panel has also been used in large-scale population-based studies such as the Rotterdam Study [49] and Generation R [50].

To estimate the number of cases required for the GWAS analyses, a sample size calculation was performed. Assuming a risk allele frequency of 0.2, a case to control ratio of 1:1, and a P value threshold of P<5*10^–8 for the GWAS analysis, a cohort of 574 patients was considered sufficient to detect an odds ratio of at least 2.8 with a statistical power of 80% in the design of the study.

For both candidate gene and GWAS analyses, genetic profiles from children who were treated with cisplatin and have hearing impairment were compared to those of children treated with cisplatin who did not develop hearing impairment. Relationships of categorical data were compared using the chi-square and Fisher exact tests. Comparison of distribution between groups with continuous data was tested with the Mann-Whitney and Kruskal-Wallis tests. Standard logistic regression models adjusting for age at diagnosis, gender, total cumulative cisplatin dose, and principal components were employed to calculate odds ratios with 95% confidence intervals in order to assess the risk of hearing loss. Principal component analysis, a common tool that has been widely used for the combined analysis of correlated phenotypes in genetic linkage and association studies, was used to correct for population stratification by modeling ancestry differences between cases and controls. Bonferroni correction was used in the candidate gene analysis to adjust for multiple testing. In the GWAS, a suggestive significance threshold of P<1*10^–6 was used to identify relevant SNPs that could be important but did not reach genome-wide significance (P<5*10^–8). All statistical analyses were performed by investigators of WP4b and WP5 in close collaboration with the Biostatistical Support Group of UMC Mainz and Ulm.

For both the WP4b candidate gene approach and WP4b GWAS, replication analysis is planned within an independent Canadian cohort from the Canadian Pharmacogenomics Network for Drug Safety (CPNDS).

This paper describes the design of the PCL ototoxicity genetics study aiming to identify clinical and genetic risk markers in a large and heterogeneous cohort of CCS and identify additional genetic risk markers of hearing loss by GWAS with better control of nongenetic risk factors in a more homogenous subcohort of CCS. Data were collected from 1124 CCS from 7 different European countries. Some of the previously identified genetic variants for hearing loss were validated by a candidate gene approach. In addition, the first international GWAS of cisplatin-induced hearing loss sets out to identify novel allelic variants in the largest European cohort assembled for such a genome-wide pharmacogenetics association study so far.

For this study, a subcohort was recruited consisting of patients who were treated with cisplatin and did not receive cranial irradiation, a well-known independent risk factor for sensorineural hearing loss. From 14% to 27% of children who received radiotherapy without ototoxic chemotherapy suffered from high-frequency hearing loss [51,52]. This risk of hearing loss increases in patients who require platinum-based chemotherapy combined with radiation. Whether the same genetic markers are associated with platinum- and radiation-induced hearing loss is unknown. To limit contamination by the presence of confounding factors such as cranial irradiation, a more homogenous subcohort of patients was selected for WP4b.

Appropriately sized cohorts are required to identify genetic determinants of platinum-induced hearing loss. The many associations that are tested in a GWAS require a very low significance threshold to prevent an inflated genome-wide type I error. This reduces the probability of identifying SNPs with small effect size, unless sample sizes are large enough to achieve sufficient power to identify such SNPs. The large combined cohort within the PCL consortium is expected to provide adequate statistical power.

Many classification systems of drug-induced hearing loss have been developed—Brock grading system [53], American Speech-Language-Hearing Association (ASHA) criteria [54], Chang classification [55], Muenster classification [24], and the SIOP Boston Ototoxicity Grading Scale [26]—yet an international standard for ototoxicity reporting is still lacking. Choice of the classification system and definitions of hearing loss may have an impact on the frequency of occurrence in childhood cancer survivors [56], as shown in a recent study that investigated the influence of several classification methods in a large prospective cohort of platinum-treated children and adolescents. Estimates of the overall occurrence of hearing loss (40% to 56%) and severe hearing loss (7% to 22%) cover a wide range [57]. Compared to other methods, Muenster grade 1 is considered a strong predictor for the need of hearing support in CCS, with reported sensitivity and specificity levels of 67% and 87%, respectively [58]. In addition, the SIOP Boston scale might be superior to determine hearing loss compared to the ASHA, Brock, and Common Terminology Criteria for Adverse Events (CTCAE) methods, based on the high number of evaluable assessments, sensitivity, and earliest time to detect hearing loss [57]. In order to strengthen our study, both the Muenster classification and SIOP Boston scale were used for a valid interpretation of the severity of platinum-induced hearing loss in this cohort.

The availability of a large European set of clinical, audiometric, and genetic data provides the PCL consortium excellent opportunities for further collaboration, including replication studies in independent transatlantic cohorts or meta-analyses. In order to validate findings from the initial discovery cohort, it is standard practice to include an independent replication cohort. A collaboration with the CPNDS for replication of results of this study has been initiated. The patients enrolled in our replication cohort were recruited from 11 hospitals and health care centers in Canada. Hearing loss in the CPNDS cohort was originally graded according to the CTCAE [29]. Applying a standardized definition for hearing loss facilitates a combined analysis of the CPNDS and PCL data. For that purpose, end point harmonization was pursued by reevaluating all audiograms (EC) of the CPNDS cohort according to Muenster and SIOP criteria. Additional replication cohorts could be needed for future international collaborations.

The PanCareLIFE ototoxicity studies have some limitations. As a result of missing or unclassifiable audiograms, some patients cannot be included due to a missing phenotype. Because many of the patients with missing audiograms might have good hearing function, they might therefore no longer be followed up for audiometric testing, As a consequence, the risk of ototoxicity based on the results of this study might be an overestimation of the true risk. Currently, the International Late Effects of Childhood Cancer Guideline Harmonization Group is developing recommendations for audiological monitoring in CCS. The guideline unifies existing recommendations and provides optimum follow-up practices, which is important for consensus on the frequency and timing of audiological evaluations after childhood cancer [59].

In summary, our paper described the design of a genetic susceptibility study that addresses an important late effect of cancer therapy (ie, platinum-induced hearing loss in survivors of cancer diagnosed and treated during childhood). Identification of genetic risk factors may assist in the development of more accurate prediction models that can be incorporated in future clinical trials of platinum-based therapies for cancer. Increased knowledge of nongenetic and genetic risk factors of cisplatin-induced hearing loss may contribute to the development of preventive methods to improve quality of life in CCS.