A Genome-First Approach to Characterize DICER1 Pathogenic Variant Prevalence, Penetrance, and Phenotype

Key Points

Question  What are the prevalence, risk, and phenotypic spectrum of individuals with a germline putative loss-of-function (pLOF) variant in DICER1 according to a genome-first approach in a population-scale cohort?

Findings  In this cohort study, DICER1 pLOF variants were more than twice as common (even after adjustment for relatedness) than previously observed. Malignant tumors were observed in 16% of participants with a DICER1 pLOF variant, which is comparable to the frequency of neoplasms in the largest phenotype-first DICER1 studies published to date.

Meaning  The genome-first approach complements more traditional approaches to syndrome delineation and may be an efficient approach for risk estimation in monogenic disorders.

Importance  Genetic disorders are historically defined through phenotype-first approaches. However, risk estimates derived from phenotype-linked ascertainment may overestimate severity and penetrance. Pathogenic variants in DICER1 are associated with increased risks of rare and common neoplasms and thyroid disease in adults and children. This study explored how effectively a genome-first approach could characterize the clinical traits associated with germline DICER1 putative loss-of-function (pLOF) variants in an unselected clinical cohort.

Objective  To examine the prevalence, penetrance, and phenotypic characteristics of carriers of germline DICER1 pLOF variants via genome-first ascertainment.

Design, Setting, and Participants  This cohort study classifies DICER1 variants in germline exome sequence data from 92 296 participants of the Geisinger MyCode Community Health Initiative. Data for each MyCode participant were used from the start of the Geisinger electronic health record to February 1, 2018.

Main Outcomes and Measures  Prevalence of germline DICER1 variation; penetrance of malignant tumors and thyroid disease in carriers of germline DICER1 variation; structured, manual review of electronic health records; and DICER1 sequencing of available tumors from an associated cancer registry.

Results  A total of 92 296 adults (mean [SD] age, 59 [18] years; 98% white; 60% female) participated in the study. Germline DICER1 pLOF variants were observed in 1 in 3700 to 1 in 4600 participants, more than double the expected prevalence. Malignant tumors (primarily thyroid carcinoma) were observed in 4 of 25 participants (16%) with DICER1 pLOF variants, which is comparable (by 50 years of age) to the frequency of neoplasms in the largest registry- and clinic-based (phenotype-first) DICER1 studies published to date. DICER1 pLOF variants were significantly associated with risks of thyroidectomy (odds ratio [OR], 6.0; 95% CI, 2.2-16.3; P = .007) and thyroid cancer (OR, 9.2; 95% CI, 2.1-34.7; P = .02) compared with controls, but there was not a significant increase in the risk of goiter (OR, 1.8; 95% CI, 0.7-4.9). A female patient in her 80s who was a carrier of a germline DICER1 hotspot variant was apparently healthy on electronic health record review. The term DICER1 did not appear in any of the medical records of the 25 participants with a pLOF DICER1 variant, even in those affected with a known DICER1-associated tumor or thyroid phenotype.

Conclusions and Relevance  This cohort study was able to ascertain individuals with germline DICER1 variants based on a genome-first approach rather than through a previously established DICER1-related phenotype. Use of the genome-first approach may complement more traditional approaches to syndrome delineation and may be an efficient approach for risk estimation.

The phenotype-first approach, the traditional and proven strategy in clinical cancer genetics, historically has been productive in linking phenotype with germline variation. The Geisinger MyCode Community Health Initiative exemplifies an alternative to this method: the genome-first approach, in which individuals with pathogenic variants are ascertained on the basis of genotype. Specifically, the Geisinger-Regeneron DiscovEHR collaboration links electronic health records (EHRs) to population-scale exome sequencing to create a platform for genomic discovery, drug development, and clinical genomic implementation.¹ The genome-first approach coupled with deep phenotyping demonstrated its utility and accuracy in ascertainment of the predictive value of CFTR [OMIM 602421] screening and cystic fibrosis.² Previous analyses of these data have focused on more common genetic disorders (eg, BRCA1/2 [OMIM 113705] and hypercholesterolemia).^3,4

Heterozygous germline pathogenic variants in DICER1 [OMIM 606421], an essential component of the microRNA (miRNA) processing pathway, underlie an autosomal dominant tumor-predisposition disorder that confers increased risk of a variety of rare and common neoplasms in children and adults.⁵ The neoplasm risks associated with pathogenic DICER1 variants include pleuropulmonary blastoma (PPB, a lung sarcoma); cystic nephroma; Wilms tumor and renal sarcomas; Sertoli-Leydig cell tumor (SLCT); gynandroblastoma; thyroid nodules; thyroid cancer; and nasal, eye, pituitary and pineal tumors.^6-8 Nonneoplastic syndrome manifestations include thyroid disease (especially multinodular goiter⁷), macrocephaly,⁹ retinal changes,¹⁰ and kidney and urinary tract anomalies.¹¹ Although the prevalence of these DICER1-specific neoplasms is low, the prevalence of loss-of-function DICER1 variants in the general population is estimated to be approximately 1 in 10 600 people and is more common than expected.¹²

As with many monogenic disorders, the ascertainment of individuals with pathogenic germline variants in DICER1 is often triggered by symptomatic individuals. Typically, a child or adolescent who unsuspectingly harbors a germline DICER1 loss-of-function variant presents with a lung, kidney, or ovarian mass. Pathologic diagnosis of a PPB, cystic nephroma, or SLCT then leads to genetic counseling, DICER1 testing, and cascade testing in other family members. In this analysis, we apply the genome-first approach to a population-scale (>92 000 individuals), EHR-linked, exome-sequenced cohort to investigate the prevalence, risk, and phenotypic spectrum of individuals with germline putative loss-of-function (pLOF) variants in DICER1.

The study cohort consisted of individuals who consented to participate in the MyCode Community Health Initiative, an institutional review board–approved program to create a biorepository of blood, serum, and DNA samples for broad research use, including genomic analysis. The DICER1 variants in this study were derived from the first 92 296 participants, of whom 1165 (1.26%) were younger than 19 years. Exome sequencing for the DiscovEHR collaboration has been previously described in detail.¹³ Familial relationship was inferred using identity-by-descent analysis (PLINK software, version 1.9), and pedigrees were reconstructed using PRIMUS software.¹⁴ This study was approved by the General Institutional Review Board; participants in MyCode have provided broad consent for research use of their exome and EHR data. Imputed pedigrees such as those in this manuscript are therefore generated from the exome data and not through recontact of the patient. Under the rules of the institutional review board, imputed pedigrees do not require additional consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Individuals with pLOF DICER1 variants identified by germline exome sequencing underwent open record review through a Geisinger Institutional Review Board–approved protocol. We developed an abstraction form to document the EHR imaging and interpretations, hospitalizations, surgical procedures, and medical history. In addition, the data abstraction was guided by thyroid-related clinical traits (unigoiter, multinodule goiters, thyrotoxicosis, and thyroidectomy) and all malignant tumors obtained from a structured EHR database. Previous EHR reviews found that reviewers were able to locate more information when they were guided by the diagnosis or procedure, especially when the dates of diagnosis or procedure were included. Three independent reviewers (K.M., J.H., and Y.H.), including a clinical geneticist (K.M.) and an endocrinologist (Y.H.), performed the EHR review, which included examining clinical laboratory test results, problem lists, practitioner notes, and scanned documents. In addition, the reviewers also performed broad searches in the EHR for the terms DICER1, DICER, DICER syndrome, genetics, pleuropulmonary blastoma, and ppb.

We applied our published scheme with modifications to classify DICER1 variation into 4 categories: pLOF (similar to “pathogenic” by the American College of Medical Genetics and Genomics and the Association for Molecular Pathology [ACMG-AMP] criteria),¹⁵ predicted deleterious (similar to “likely pathogenic” by the ACMG-AMP criteria), variant of uncertain significance (VUS), and likely benign (LB).^12,16 For statistical analyses, individuals with germline DICER1 pLOF, predicted deleterious, VUS, or LB variants (carriers) were matched to noncarriers (individuals with reference DICER1 sequence) by sex, age, race, and smoking status; for analyses of missense variants, smoking was used as a covariate. CADD¹⁷ and REVEL¹⁸ (bioinformatic variant pathogenicity prediction methods) were also used to investigate the utility of these tools in predicting the consequence of germline DICER1 variation. A CADD score of 20 or higher or a REVEL score of 0.5 or higher were considered damaging based on the developer’s recommendation.

We used the International Classification of Diseases, Ninth Revision (ICD-9), International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), and procedure or Current Procedural Terminology codes (eTable 1 in the Supplement) to find diagnoses related to thyroid disease (including thyroidectomy) and malignant tumors in the EHR. For individuals with DICER1 pLOF variants, cancer diagnoses were confirmed using data in the Geisinger Cancer Registry (1943-2017); nonmelanoma skin cancers were excluded. We also searched the cancer registry for known DICER1-associated tumors in the MyCode cohort, regardless of germline DICER1 status.

Archival pathology materials on tumors were obtained and subjected to DNA extraction and DICER1 sequencing, as previously described.¹⁹

Kaplan-Meier analyses with contingency tables were performed for thyroid and malignant tumor phenotypes using the R version 3.6.0 survival package (R Foundation for Statistical Computing).²⁰ Dates of entry of each MyCode participant into the Geisinger registry were used from the start of the Geisinger EHR to February 1, 2018. Statistical difference in Kaplan-Meier curves was determined using Cox proportional hazards analysis and likelihood ratio test²¹; 2 × 2 contingency tables were then used to test for association of variants and clinical traits by odds ratios (ORs) and 95% CIs. The significance of the association was evaluated using the Fisher exact test values. Associations between DICER1 variants and phenotypes were considered significant if Bonferroni-corrected multiple testing resulted in P < .008 (calculated for α = .05 for 6 clinical traits tested).

A total of 92 296 individuals (mean [SD] age, 59 [18] years; 98% white; 60% female) participated in the study. The cohort demographic characteristics were similar to those previously reported in DiscovEHR studies³ and summarized in eTable 2 in the Supplement. Table 1 gives the numbers of unique variants and individual carriers of DICER1 variants (full details in eTable 3 in the Supplement), classified according to our previously published scheme.¹² There were 12 unique DICER1 pLOF variants, including 5 (42%) frameshift, 2 (17%) stop-gain, 3 (25%) canonical splice site, 1 (8%) initiator loss, and 1 (8%) hotspot codon missense variation; all variants were confirmed by Sanger sequencing. The 12 unique pLOF variants were observed in 25 individuals (1 in 3700 people). Pedigrees inferred from identify-by-descent analysis¹⁴ indicated that 9 of the pLOF variant carriers in 4 pedigrees were closely related (eFigure 1 in the Supplement). When calculated using only unrelated individuals (n = 20), the prevalence was 1 in 4600.

The demographic characteristics and thyroid-related clinical features of the 25 participants with DICER1 pLOF variation are given in Table 2 and are similar to the sequenced cohort (eTable 4 in the Supplement). Notably, 2 DICER1 pLOF carriers were deceased. One was a White woman (patient 24) with a history of pineoblastoma (when she was an adolescent) and meningioma (in her 40s) with a germline DICER1 frameshift variant (p.Ser1823Valfs) who died in her 40s; cause of death was not documented in the EHR. The second was a White man (patient 4) with an absent germline DICER1 start site (p.Met1?) and a history of hypothyroidism who had been taking levothyroxine since his 40s and had renal cell carcinoma in his late 40s (and had undergone nephrectomy), diverticular rupture and partial colectomy (in his 50s), ultrasonic evidence of pancreatic cysts (in his 60s), and liver cirrhosis. He died in his 70s from complications of cirrhosis.

A woman in her 80s (patient 23) harbored a germline DICER1 p.Asp1709Glu hotspot variant (variant allele frequency, 23%). This variant has been reported to be somatically mutated in fetal adenocarcinoma of the lung,²² SLCT, and primitive germ cell tumor (yolk sac).²³ In addition, in silico bioinformatic tools (REVEL score, 0.89; metaSVM score, deleterious; CADD score, 25.5) predict more severe consequences on protein function. However, EHR review found a history of liver cysts on abdominal ultrasonography and a history of breast biopsy with no evidence of atypia or malignant tumor. We note that mesenchymal hamartoma, a cystic lesion of the liver, is reported to arise from a DICER1 mutation.²⁴ This patient has had 120 encounters with Geisinger physicians during the last 16 years; thus, her lack of thyroid-related clinical findings is not attributable to a lack of medical care.

Notably, the terms DICER1 or DICER did not appear in any of the EHRs of participants with a DICER1 pLOF variant. Furthermore, EHR review found that most carriers had sufficient data in the EHR, with a median number of encounters of 43 (range, 6-249) and a median length of follow-up of 14 years (range, 2-18 years). These findings are comparable to the matched noncarriers, who had a median number of encounters of 62 (range, 0-974) and a median length of follow-up of 14 years (range, 0-21 years).

Four biopsy-proven malignant tumors (excluding nonmelanoma skin cancers), identified from the Geisinger Cancer Registry, were observed in the 25 individuals with a DICER1 pLOF variants (Table 3). Of these, high-quality DNA was available only from the papillary thyroid carcinoma, which harbored multiple somatic DICER1 hotspot variants, in addition to the germline p.Asn1668Ilefs. In the 25 individuals with a DICER1 pLOF variant compared with 7550 noncarriers matched by age, sex, race, and smoking status, we observed significantly greater associations with thyroid cancer (OR, 9.2; 95% CI, 2.1-34.7; P = .02) and thyroidectomy (OR, 6.0; 95% CI, 2.2-16.3; P = .007); thyroidectomy remained significant after Bonferroni correction (Figure 1; eTable 5 in the Supplement). There was no excess of malignant thyroid tumors in carriers of predicted deleterious, VUS, and LB variants compared with matched noncarriers.

We searched the EHR for known rare DICER1-associated tumors (ovarian SLCT, endometrial rhabdomyosarcoma, nasal chondromesenchymal hamartoma, and PPB). Of 8574 female participants with tumor-biopsy records, there was 1 poorly differentiated ovarian SLCT in a woman in her 40s and an endometrial rhabdomyosarcoma in a female patient in her teens (Table 3). In both tumors, somatic DICER1 pLOF and hotspot variants were identified (Table 3). Neither person carried a germline exonic DICER1 sequence or copy number variant.

To investigate the frequency of malignant tumors in carriers of a germline DICER1 predicted deleterious variant (metaSVM score, deleterious), we examined the number and type of malignant tumors (excluding nonmelanoma skin cancer) recorded in the Geisinger Cancer Registry (eTable 6 in the Supplement). There were 40 DICER1 predicted deleterious variants in 84 individuals. No malignant tumors were observed in individuals with 32 (80%) of these variants. Twelve carriers (20%) of 8 predicted deleterious variants had at least 1 malignant tumor; DNA of suitable quality was obtainable from a few tumors. One germline variant (p.Gly1364Ala) was observed in 2 individuals: a woman in her 30s with thyroid cancer and a man in his 40s with seminoma. Somatic sequencing of the seminoma identified a DICER1 p.Gly1809Arg hotspot variant (as previously reported²⁵) but with a very low variant allele frequency (1.5%). One germline variant (p.Thr806Met) was observed in 4 individuals, each with 1 cancer. Of the 4 tumors in germline DICER1 p.Thr806Met carriers, high-quality DNA was obtainable from 1 (sigmoid colon adenocarcinoma), which harbored no additional DICER1 somatic variation. In 27 VUS carriers, there was one woman in her 20s (germline p.Asn1393_Thr1394insAsn) with a follicular thyroid carcinoma; on somatic sequencing, no DICER1 hotspot variant was detected.

Figure 1 and eTable 5 in the Supplement give the ORs of sex-, race-, and smoking-matched carriers and noncarriers for development of thyroid disease or thyroidectomy for the 4 categories of DICER1 variation. Overall, there was a significant increase in the risks of thyroidectomy and thyroid cancer in pLOF carriers; however, only the risk of thyroidectomy remained significant after Bonferonni correction for multiple testing. There was no significant increase the prevalence of goiters or thyrotoxicosis. In addition, we did not observe an increase in the prevalence of cancers in all pLOF carriers compared with noncarriers. eFigure 2 in the Supplement shows heatmaps of age at onset for the thyroid phenotypes stratified by different DICER1 variation subclasses.

To further investigate the 2 individuals with a DICER1 pLOF variant and a diagnosis of thyrotoxicosis or hyperthyroidism, an endocrinologist performed an EHR review. The first patient, a woman in her 50s (patient 20) (Table 2), had a thyroid cancer treated by thyroidectomy in her 20s and was prescribed levothyroxine. There was no EHR laboratory or pathology documentation of an organic cause underlying her hyperthyroidism. The second patient, a woman in her 40s (patient 9), had EHR documentation of Graves disease, including pathology review of her thyroid and an abnormal radioactive iodine scan.

To more thoroughly explore the consequence of germline DICER1 missense variation, we investigated the probability of thyroid disease (defined as an incidence of goiter, hypothyroidism, cysts, benign thyroid tumors, or thyrotoxicosis) in carriers of DICER1 missense variants predicted to be damaging by 3 bioinformatic tools (using the thresholds recommended by the developers of metaSVM, CADD, and REVEL), compared with sex-, race-, and age-matched noncarriers (Figure 2). No significant differences were observed in these analyses.

This cohort study is, to our knowledge, the first study to use EHR-linked exome sequencing in a large cohort to investigate the DICER1 tumor-predisposition disorder and provides critical experience regarding the strengths and limitations for genome-first investigations of monogenic tumor-predisposition disorders in general. From 92 296 exomes, this study estimated germline DICER1 pLOF variant prevalence to range from 1 in 3700 to 1 in 4600 people. This is more than twice as common (even after adjustment for relatedness) than the previous estimate of 1 in 10 600 people¹² (non-Finnish European: 1 in 9058), which was based on frequency of DICER1 variants in the noncancer cohort of the Exome Aggregation Consortium (n = 53 105) and was approximately 60% of the size of the cohort used in the current analysis. This refined estimate is comparable to the prevalence of germline pathogenic DICER1 variants in The Cancer Genome Atlas (1 in 4600; n = 9173 exomes).¹⁶ It is also comparable to the prevalence of other common genetic disorders, such as fragile X,²⁶ 22q11.2 deletion syndrome,²⁷ and neurofibromatosis type 1.²⁸ The discovery, through genome-first approaches, that pathogenic variant prevalence in DICER1 (and in other important tumor-predisposition genes, such as BRCA1 and BRCA2⁴) is more common than expected has important clinical implications. First, accurate estimates of variant prevalence (and penetrance) are needed to assess whether a variant is too common to be causative for a mendelian disorder of interest²⁹; these determinations are crucial in the development of variant interpretation rules set by ClinGen based on the ACMG-AMP¹⁵ criteria. These standardized, peer-reviewed rule sets are important in the identification of at-risk individuals. Second, a higher prevalence suggests that many more people are at risk than previously recognized, which influences a priori risk estimates in genetic counseling. Third, higher prevalence highlights the importance in risk estimation of accurate penetrance estimates, which in DICER1 carriers is known to depend on age and sex.⁸ However, estimates of penetrance in DICER1 carriers may also need to be tiered based on family history. Family history is important in risk estimation in some monogenic tumor-predisposition disorders, such as those associated with pathogenic variants in BRCA1/2 or PALB2 (OMIM 610355).³⁰

Among the 25 individuals with a germline DICER1 pLOF variant, 4 of 25 (16%) had a history of malignant tumor, which is comparable (by 50 years of age) to the frequency of neoplasms in the largest registry- and clinic-based (phenotype-first) DICER1 studies published to date.^19,31,32 However, the penetrance estimate in the current analysis is conservative and limited by the small number of observations and multiple causes of underascertainment. For example, pediatric and other early-onset aggressive cancers would be undercounted, especially if they lead to death or otherwise make it less likely for an individual to enroll in MyCode. In addition, individuals with goiter and thyroidectomy, especially at an early age, would have abrogated risk of developing thyroid disease.

This study was able to identify less common (pineoblastoma) and more common (thyroid carcinomas) DICER1-associated neoplasms as well as a potentially novel DICER1-associated neoplasm (renal cell carcinoma) that merits additional follow-up. In addition, this study identified 2 known DICER1-associated neoplasms (SLCT and rhabdomyosarcoma) in patients without any germline DICER1 variation, consistent with low-level mosaicism or tumor-only DICER1 variation. Accurate estimates of the prevalence of DICER1 mosaicism and tumor-confined variation are also crucial to providing useful clinical risk estimates and genetic counseling; this study found that the genome-first approach is effective in providing data toward determining those estimates. The data indicate how genome-first ascertainment informs phenotype. For example, multiple publications indicate that germline mosaicism for DICER1 hotspot codon variation is linked to a severe overgrowth phenotype.^19,31,32 Surprisingly, this study found an apparently healthy woman in her 80s who harbored a previously reported, Sanger-confirmed hotspot variant (variant allele frequency, 23%). It was not possible to perform clinical phenotyping in this participant and the variant could have arisen from clonal hematopoiesis, although this has not been reported in DICER1. If germline, this finding suggests that the phenotype arising from DICER1 hotspot variation may not be as severe as previously reported and illustrates an advantage of genome-first ascertainment.

It is notable that on manual review of the EHR, the term DICER1 did not appear in any of the records of the 25 participants with a pLOF DICER1 variant, even in the records of patients with a known DICER1-associated tumor or thyroid phenotype. This finding is perhaps not surprising given the older age of the cohort and the lack of specificity of the phenotypes. However, these participants and their relatives (especially children and female individuals) are at risk for a set of well-characterized DICER1-associated tumors that can be prospectively identified and managed³³ with established DICER1-specific surveillance guidelines.³⁴ Such surveillance is particularly appealing (and, with subsequent interventions, potentially curative) for highly morbid DICER1-associated tumors, such as PPB and ovarian sex-cord stromal tumors. However, given the reduced penetrance of DICER1 pathogenic variants, many at-risk individuals will never develop problems and will thus not benefit; rather, they will be harmed by false-positive results, follow-up biopsy, and costs. These issues will become only more magnified and urgent for DICER1 and other genes, with widespread population-scale sequencing, obligatory return of secondary findings,³⁵ and germline results in tumor sequencing.³⁶ The solution is to optimize the risk-benefit ratios in an age-, sex-, and race-specific way through the development of accurate penetrance estimates using additional biomarkers, better imaging, and surveillance modalities. The influence (if any) of family history, modifier genes, DICER1 genotype and phenotype correlates, polygenic risk, and environmental exposures needs to be identified. To do this, both genome-first and phenotype-first approaches are likely needed.

This study has limitations. The cohort, though large and clinically unselected, is 98% White and required enrollment in a health care system. Even in a cohort of 92 296 individuals, the number of DICER1 pLOF variants was small. The study focused on germline and somatic sequence–based alterations in DICER1 and did not evaluate methylation or copy-number variation (except small [<10 base pairs] deletions) at this locus. The phenotyping was limited to the accuracy of the EHR. Cancer or thyroid diagnoses made outside of the Geisinger registry may have been missed, although this is less likely given the EHR review performed for pLOF carriers. Pediatric cancer diagnoses may be incompletely ascertained, especially for participants enrolling in the health care system as adults. Last, individuals with the most damaging germline DICER1 variation may have died before having an opportunity to enroll in the MyCode cohort.

To our knowledge, this is the first study to ascertain individuals with DICER1 variation (as an archetypal monogenic tumor-predisposition disorder) based on a genome-first approach rather than through a previously established DICER1-related phenotype. Use of the genome-first approach not only complements more traditional approaches to syndrome delineation but also may be an efficient approach for risk estimation in monogenic disorders.

Accepted for Publication: January 4, 2021.

Published: February 25, 2021. doi:10.1001/jamanetworkopen.2021.0112

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Mirshahi UL et al. JAMA Network Open.

Corresponding Authors: Douglas R. Stewart, MD, Clinical Genetics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, 9609 Medical Center Dr, Room 6E450, Rockville, MD 20850 (drstewart@mail.nih.gov); David J. Carey, PhD, Department of Molecular and Functional Genomics, Weis Center for Research, Geisinger Clinic, Geisinger, 100 N Academy Ave, Danville, PA 17822 (djcarey@geisinger.edu).

Author Contributions: Drs Mirshahi and Kim had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Mirshahi and Kim contributed equally to this work.

Concept and design: Mirshahi, Kim, Manickam, Bauer, Hill, Murray, Carey, Stewart.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Kim, Manickam, Schultz, Bauer, Rosenberg, Murray, Stewart.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Mirshahi, Kim, Best, Rosenberg.

Obtained funding: Stewart.

Administrative, technical, or material support: Chen, Hu, Haley, Golden, Stahl, Carr, Harney, Field, Schultz, Hill, Murray, Carey, Stewart.

Supervision: Best, Bauer, Carey, Stewart.

Conflict of Interest Disclosures: Dr Field reported receiving a salary from ResourcePath Salary outside the submitted work. Dr Hill reported receiving grants from the National Cancer Institute during the conduct of the study and being an owner of ResourcePath LLC, a company developing liquid biopsy diagnosis of DICER1-related cancers outside the submitted work. Dr Murray reported receiving grants from Regeneron during the conduct of the study. Dr Carey reported receiving grants from the National Institutes of Health outside the submitted work. Dr Stewart reported receiving personal fees from Genome Medical Inc outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics of the National Cancer Institute, Bethesda, Maryland, and Geisinger, Danville, Pennsylvania. This work used the computational resources of the National Institutes of Health High Performance Computing Biowulf cluster.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The content of this publication does not necessarily reflect the views or policies of the US Department of Health and Human Services, and mention of trade names, commercial products, or organizations does not imply endorsement by the US government.

Additional Contributions: We thank the participants of the MyCode Community Health Initiative for use of their genomic and electronic health information, without whom this study would not be possible. Enrollment of MyCode participants and exome sequencing was supported in part by the Regeneron Genetics Center. We thank the Geisinger-Regeneron DiscovEHR collaboration for making the genotype and phenotype data available for this project.