Laboratory Considerations for Releasing Next-Generation Sequencing Data to Patients Open Access

NGS testing includes preanalytic, analytic, and interpretation phases. From an accreditation standpoint, NGS test validation encompasses the entire analytic phase, which includes both the wet bench portion and the data processing (bioinformatics pipeline or dry bench) portion, the latter referred to as a bioinformatics pipeline.^5,6  According to the College of American Pathologists (CAP) Molecular Pathology Checklist item MOL.36015 (version 2022), “[t]o determine acceptable beginning-to-end test performance, validation/verification of the NGS analytical wet bench component must be integrated with the bioinformatics component for the intended test (see MOL.36151), including all laboratories participating in a distributive testing process.” Furthermore, variant classification, interpretation, and reporting are typically performed together as part of the ordered service (or test), as reflected by the existing Current Procedural Terminology codes for NGS (Figure 1). For multigene panels, these steps are also known as genomic sequencing procedures. However, variants may also be separately reclassified or reinterpreted later. Furthermore, particularly in the case of exome or genome sequencing in the setting of undiagnosed hereditary disorders, the laboratory that performed the original test may be asked to reanalyze or reinterpret the exome or genome in the context of evolving medical knowledge. Validation of a sequencing test from start to finish, including the entire analytic phase, limits how intermediary files preceding the final variant call file (VCF) can be used for clinical purposes by a laboratory or clinician independent of the laboratory that originally validated and performed the analytic phase.

The rationale for the current regulatory definition that considers the wet bench and bioinformatics portions together as the validated test is that changes in either can affect the final output. Processes in the wet bench portion may require different technical approaches or settings in the bioinformatics pipeline. The wet bench portion includes the sample type used, extraction of nucleic acid, library preparation, enrichment (if performed), sample pooling (if performed), and sequencing. Library preparation readies the sample for sequencing and may include adding sequencing adapters, sample-specific bar codes, or other bar codes. Various quality control (QC) steps are also performed throughout library preparation and may include measuring concentrations of nucleic acids at different steps, as well as assessment and/or selection of nucleic acid fragment size. In addition, for somatic testing, whether a laboratory uses a comparison between both tumor and normal tissue or only sequences tissue from the tumor may affect sequencing results. Different technical approaches to the sample type, extraction, enrichment, library preparation, and QC steps can also affect the final output. For example, different extraction methods may lead to differences in guanine cytosine bias and fragment lengths (University of Minnesota, Minneapolis, internal data). Copy number alteration detection is affected by the number of probes or amplicons across the region of interest.⁵  Several sequencing steps affect the structure and settings of the bioinformatics pipeline, including pooling of samples, read length options, single versus paired-end reads, sequencing chemistry, and QC metrics. This may not seem intuitive for those used to using blocks or slides from another institution for downstream testing; however, many molecular laboratories will not accept DNA extracted from another laboratory for testing because even small changes in testing can affect the final result. Changes in any step may require revalidation of the entire test process, and limitations on the performance characteristics of the assays may result from the analytic approaches.

The data processing portion of the test starts with the base call files from the sequencer (Figure 2). If samples have been pooled for sequencing, which is commonly done, these files will contain data from multiple patients and require de-multiplexing to obtain individual FASTQ files, often referred to as the raw data. A FASTQ file may be modified; for example, if unique molecular indexes are used, all reads with a particular index may be condensed to a single read giving a secondary FASTQ. The individual reads from a FASTQ file are aligned to a reference sequence to obtain a BAM file, which is often used in practice to visualize data. The BAM file is analyzed for differences from the reference to create a VCF. A VCF may also be filtered to exclude certain variants, such as common benign variants, poor quality variants, sequencing artifacts, variants outside the area of interest, and germline variants in the setting of paired tumor–normal sequencing. The final VCF, which may be filtered or nonfiltered, contains the variants that are reviewed. The final VCF may also include: (1) information about how the variants are classified as to their potential pathogenicity; (2) clinically significant reported variants; and (3) nonreported variants. Particularly when testing for hereditary disorders, laboratories may use a broad chemistry, such as an exome or genome, but only interrogate a portion of the sequenced regions. Each test has a defined region of interest (ROI) that describes the genes to be analyzed as well as any noncoding regions, including introns. Variants outside the ROI may or may not be optimally sequenced and may or may not be excluded from the final VCF.

Which files from these bioinformatics processes are considered part of the designated record set must be determined by the health care entity. A VCF or equivalent that includes variants that were reported, and potentially any variants classified as nonreportable (eg, benign), that are in the ROI is likely a part of the designated record set because this file contains the variants that are classified and interpreted. In contrast, the base call files from the sequencer cannot be considered part of the designated record set, because these files often contain data from multiple patients. Files generated between the base call file and the final VCF, which are referred to as intermediary files (Figure 2) and include FASTQ files, BAM files, and any preliminary VCFs, may be a part of the designated record set for the time that the institution retains these files. The FASTQ is the first patient-specific file available and fits the broad definition of “other information used … to make decisions about individuals.”²  The FASTQ might be considered analogous to a gross tissue specimen in anatomic pathology, which in either case requires additional steps (for example, selecting areas of the gross specimen to sample, fixing the specimen, cutting slides, and ordering immunohistochemical or special stains as needed to obtain the end product, such as a set of slides, that undergoes interpretation). The BAM file is also an intermediary file but is more directly used during the interpretive process, usually to visualize the data, and more closely fits the definition of “other information used … to make decisions about individuals.”²  However, both the FASTQ and BAM files may contain nonvalidated base calls from regions of the genome whose detection was not optimized and/or validated and are not part of the ordered test and therefore will contain information that was not used to make decisions about the patient (Figure 3). Likewise, FASTQ and BAM files for a targeted or smaller panel run on exome or genome chemistry will contain genetic information that is not part of the ordered clinical test and not used to make decisions about the patient (Figure 3). The latter situation is similar to that of hemoglobin being run on a complete blood count analyzer, where the entire complete blood count is tested but only 1 component is reported. The rest of the data were not used to make decisions about the patient and are not available. The intermediary files are not, in and of themselves, the output or result of a validated test, but rather are used in the context of the originating laboratory’s total wet bench and dry bench processes to generate a result. Only the validated subportions of those files are used to generate results and make decisions about the patient. This limits how these files can be used and reanalyzed for clinical care, especially when the intermediary files contain a larger ROI than the final VCF.

In some cases, it may be challenging to include intermediary files, such as the FASTQ and BAM, as part of the designated record set. For example, some commercial bioinformatics options are proprietary, and the laboratory cannot access all intermediary files. Additionally, some laboratories use a distributive testing model for NGS where the bioinformatics component is performed by a separate entity. In this case, the intermediary files may not be readily available to the primary laboratory. This has been a challenge for the CAP for developing in silico NGS proficiency testing that challenges the bioinformatics portion of the test. Laboratories in these situations need to have a process worked out to meet the federal requirement to provide patients with their health care data. If a separate commercial entity performs the bioinformatics component and retains the data, the laboratory should direct the patient to that entity. All laboratories (and/or affiliated companies) need to provide the designated record set data within the 30-day time limit, although a single 30-day extension is possible. A suggestion has been made to decrease the time limit to 15 days in the future (with a 15-day extension), so laboratories should streamline this process to ensure they can comply with the regulation.⁷ 

Policies for data retention will vary across laboratories, and not all sequencing files will be retained indefinitely. According to the CAP checklist requirement MOL.35870 (version 2021), NGS data necessary to support primary results and reanalysis must be retained for a minimum of 2 years or longer based on local, national, or state regulations, but there is no universal requirement for indefinite storage. Small files (VCF) are more likely to be archived than large files (FASTQ) because of the cost of data storage, which can be significant. For example, a paired-end read genome sequence will generate 2 FASTQ files, each between 100 and 200 GB, depending on the coverage depth. Even compressed, these files consume 25 to 50 GB. A BAM file is 50 to 100 GB, whereas a VCF is around 0.1 GB (University of Minnesota, internal data). The designated record set will therefore vary based on a laboratory’s data retention policies and the timing of the patient’s request for data.

When responding to a patient who requests their NGS data, the health care entity is required to provide NGS data that are part of a designated record set and that exist within their records. The health care entity is not required to summarize, reanalyze, or specially prepare the data, but it may do so for a fee if requested.¹  Outside of this, the health care entity may only charge for the labor for copying, supplies needed, and postage; it may not charge for the labor required for locating data.²  The health care entity must transfer the data in the format and method requested by the patient. If it is not able to transfer the data in the requested format or if the method would compromise the health care entity’s data, the entity must work with the patient to find an alternative method to transfer the data. If the patient is informed and understands the security risks but still requests that their data be sent by a nonsecure method, the health care entity is not responsible for the insecurity of those data; otherwise, the health care entity is responsible for the security of the data. Depending on the size of the files, some options for transfer could include an external storage device, straight-through processing transfer, email, and cloud storage transfers (eg, Box [Box Inc, Redwood City, California] and Google Drive [Google, Mountain View, California]). Some of these options are not secured, but they can be used at the patient’s request if they understand and accept the security risks. The health care entity’s responsibility ends with providing the patient (or representative) with an acceptable copy of the designated data set in a way acceptable to the patient (or representative).

Laboratories should not discourage patients from requesting their NGS data, but rather should help individuals understand what files are available, the content of the files, uses of the files (Table 2), and potential limitations. Patients may just want to possess a copy of their data, but often they will want the data for a specific purpose, and a discussion or provision of an educational document about how the different file types could be used may be helpful to ensure that the patient gets the appropriate data for their needs. Laboratories may also provide information on alternatives to the raw data that would better meet the patient’s needs, such as an updated or different test that the patient may wish to consider discussing with their health care provider. In addition, laboratories may want to provide a disclaimer with the release of NGS data to educate patients about the limitations of using NGS data outside of a validated process. Most patients and clinicians will not be aware of the test validation process and the limitations of NGS data that do not go through a fully validated process. The disclaimer may contain language that the data are provided in accordance with the Privacy Rule and that only the original test report is intended for clinical use. General information about NGS limitations (as discussed in the discordant results section, below), as well as assay-specific information about the limitations of the data set, could also be provided. For example, if a laboratory uses exome chemistry but has only validated specific regions of interest, a disclaimer that other regions have not been assessed for quality or accuracy could be included.

Analysis of data using a different bioinformatics pipeline introduces the possibility of getting results different from those originally reported. Any difference should be confirmed using a clinically validated test. Assuming the discordance is confirmed clinically, there are multiple reasons this may occur. First, the 2 bioinformatics algorithms may not interrogate the same areas of the genome. Each NGS test defines an area of the genome that will be interpreted and excludes regions that are not of known interest for the clinical scenario and that do not give reliable information (often due to high guanine cytosine content, homology elsewhere in the genome, or repetitiveness, among other reasons). Available clinical information helps to define the areas of the genome that will be analyzed, particularly for exome and genome analysis, and if different clinical data are available, this also may affect results. Second, the validated limit of detection (LOD) may differ among NGS tests. The LOD is an important consideration in oncology testing but is also important in detecting mosaicism or somatic variants causing congenital disorders (for example, somatic overgrowth syndromes). Third, different bioinformatics approaches are needed to detect different types of variants. The approach to detect single-nucleotide variants is different from the approach to detect copy number variants, and the pipelines (or assay) may not be validated to detect the same types of variants. The NGS report should list the classes of variants that are detectable, the LOD, other limitations, and the area of the genome covered, which will allow recognition of these causes of discordant results. Lastly, clinical knowledge and performance of bioinformatics tools advance rapidly. Therefore, laboratories should maintain information about the Human Phenotype Ontology (HPO) terms used (if applicable), the databases used (along with versioning), and any additional information required to recreate the analysis.

Discrepant results should not be immediately considered errors because they frequently represent an advancement of medical and scientific knowledge, intrinsic differences in complex tests, and/or differences in data analysis. This is analogous to an Internet search whereby results generated or prioritized using 1 search engine may be different than those generated by another. Furthermore, different search results may appear compared with those that were generated a year ago. None of these results are wrong; the tools are just structured differently to give different information as knowledge changes, leading to different results over time. Before deeming a discrepancy an error, it is necessary to exclude the possibility of differences in ROI, the use of a pipeline that does not cover the variant type in question, and a different LOD, among other factors. Additionally, the person evaluating the discrepant results needs to be sufficiently qualified to understand and evaluate the limitations. Rarely, there may be a discordance that is not attributable to the reasons listed above and that should have been detected based on the assay performance and knowledge at the time. No test is perfect, and despite our best efforts, errors occur.⁸  In this instance, the discrepancy should be investigated.

How NGS data can be used depends greatly on what data are requested and released. The final report can be reviewed and acted on clinically. Some electronic medical record systems can accommodate variant-level data from a final VCF report. The final VCF, as the product of the validated test when limited to the ROI, can be reinterpreted and used clinically. However, intermediary files are not the end product of a validated test and therefore must not be used clinically. If a laboratory receives an intermediary file and reprocesses the data, the results would not be clinically validated, even if both laboratories are Clinical Laboratory Improvement Amendments certified and/or CAP accredited. The second laboratory would need to cooperatively validate this process of receiving intermediary data from the primary laboratory or confirm the data with a clinically validated assay before using any of the data for clinical purposes. However, if the primary laboratory sends another laboratory a final VCF that only includes data from the ROI for which the test was validated, those data are derived from a complete (clinically validated) analytic process, and in this situation the final VCF may be used for reinterpretation and clinical decision-making. The following clinical scenarios are examples laboratorians may encounter when data are requested from them or potentially provided to them.

A child underwent genetic testing by germline exome sequencing at laboratory A. The child’s parents request that laboratory A send the FASTQ file to an unaffiliated laboratory, laboratory B, to perform an analysis of the data using their bioinformatics platform. laboratories A and B have not validated this process. Therefore, this scenario is different from a distributive model where 2 laboratories have validated that the separate wet bench and bioinformatics platform together give acceptable performance. In this scenario, laboratory A is responsible for securely sending the FASTQ file to the patient or laboratory B, and laboratory B may process the FASTQ file through their bioinformatics algorithm. The results from this algorithm are not clinically validated and could be used for research but not for clinical care. The patient would require testing with a fully validated test to confirm any results obtained at laboratory B and used for clinical care. The same would be true for a request of any intermediary file generated before the final interpreted VCF.

Laboratory A and laboratory B may have different findings, despite using the same raw data. These differences may be due to different HPO terms, the use of different databases, other aspects of data filtering, the use of alternate bioinformatic tools, and/or the interrogation of nonvalidated genomic regions within the data. Therefore, along with the raw data that are saved to recreate the analysis, according to the CAP checklist, the HPO terms used for filtering, databases queried along with versioning, and any other parameters that may have affected the results should be retained.

A limited/small NGS panel was reported on a patient’s tumor sample. Laboratory A ran exome chemistry, but method validation, data analysis, and interpretation were limited to the genes included in the panel. The patient asks laboratory A to send the raw data to unaffiliated laboratory B, which plans to reanalyze the full exome data for both clinical and research purposes. Laboratories A and B have not validated this process for clinical use. Laboratory B may reinterpret the variants in the final VCF containing data on genes included in the small panel for clinical use. However, data from the remaining genes that were not part of the original test (small panel) may not be interpreted and used for clinical purposes. If a research finding from the additional genes/regions is of clinical interest, it must be confirmed and reported using a clinical test that has undergone an end-to-end validation of both the wet bench and bioinformatics portions of the test.

A patient asks laboratory A to send a copy of their genomic test result and the list of variants that underwent interpretation (final VCF) to laboratory B for reinterpretation. Laboratory A is responsible for the secure transmission of the information, and laboratory B may reinterpret the data and report their analysis for use in clinical care because the files are from a fully validated laboratory test. However, not all laboratories would be willing to accept data for reinterpretation because they may have concerns about the possibility of missed variants due to assay design, limitations on detected variants, or even challenges with billing and reimbursement.

A patient asks the laboratory to reformat an electronic copy of their FASTQ file and send it to them via email so they can upload it to a web-based program that provides ancestry information. The laboratory is not required to reformat the FASTQ file that it maintains. The laboratory may provide the patient temporary access to download the unaltered FASTQ file from a secure server, but it could also choose to manipulate the FASTQ according to the patient’s request and charge a fee for the reformatting. Although patients often request files to be sent by email, this is not practical for large files, such as FASTQ files. A VCF from a limited panel may be sent by email if the patient fully understands the security risks and still desires this mode of transmission. The laboratory should not interfere with the patient’s request for data, but it may choose to inform the patient about the limitations of using the requested data for purposes beyond the scope of the original validated test.

It has been 5 years since a patient had their sequencing test, and they seek to obtain the FASTQ from the testing laboratory, but the institution only keeps the FASTQ for 2 years. In this scenario, the FASTQ would no longer be part of the designated record set because it is no longer maintained by the health care entity. The laboratory is not required (and is not able) to provide the file, nor is it required to maintain intermediary files indefinitely.

The parents of a child who had exome sequencing performed as part of a diagnostic workup request the raw data from testing and intend to share the data with their child’s physician. The testing was performed 3 years ago, and the laboratory has the files. The laboratory shares the files (along with the HPO terms used) and the databases (along with versioning) from the original analysis. The parents and physician realize that the shared data from the original analysis were not reanalyzed before sharing. They contact the laboratory to request an updated analysis. The laboratory is not required, under the regulations addressing access of individuals to PHI, to update or reinterpret the data before sharing it. However, they may perform data reanalysis and bill for this service.

In response to a patient’s requests to send raw genomic data to themselves, health care providers, and/or other laboratories, the designated record set for the genomic data is legally required to be provided (45 CFR §164.524). The US Department of Health and Human Services has given guidance regarding the designated record set² ; however, health care entities are ultimately responsible for determining which specific files are part of their designated record set. Regardless of which NGS files are included in the designated record set, it is important to recognize that released data may include data that are not part of a complete validated test and to recognize that nonvalidated data should not be used for clinical purposes without confirmation by a clinically validated method.

The NGS testing process can be divided into an analytic phase and an interpretation phase. Current regulatory standards require the entire analytic phase to be validated, from handling the clinical sample/specimen through developing a final VCF; the interpretation and the reporting are not part of the analytic phase but together with the analytic phase constitute the ordered service. If the primary laboratory sends any raw genomic data from steps preceding the final VCF (ie, FASTQ, BAM, or preliminary VCF), those data should not be used by the receiving laboratory for clinical purposes, unless this process has been specifically validated between the 2 laboratories as part of a distributive testing model. Because those data are developed before the completion of the clinically validated, total analytic process, a second laboratory may not use them to complete the analytic process unless the second laboratory has jointly validated the total analytic process with the primary laboratory. However, a final VCF limited to the ROI, sent from the primary laboratory to another laboratory, is derived from a complete (clinically validated) analytical process. The filtered VCF may be used, in this situation, for reinterpretation and clinical decision-making.

Laboratory staff must understand the requirements for providing genomic data to patients on request, as well as be able to explain the potential limitations of the data and their clinical use. In response to a patient’s request, laboratory staff may discuss these aspects with the individual requesting data and recommend reanalysis or retesting and educate them about the contents and limitations of the genomic data requested. Finally, laboratory staff should support patients’ decision to access their medical information and provide NGS data from the designated record set in a way that assures the security of the data and their appropriate use.

The authors wish to thank Ellen Lazarus, MD, for providing medical editing support. Additionally, the authors thank the College of American Pathologists Cross Council Molecular Working Group for the discussion surrounding this topic that was the basis for this manuscript.