Search Results

The 23andMe Personal Genome Service (PGS) uses qualitative genotyping to detect select clinically relevant variants in genomic DNA isolated from human saliva collected from individuals ≥18 years with the Oragene Dx model OGD500.001 for the purpose of reporting and interpreting genetic health risks, including the 23andMe PGS Genetic Health Risk Report for BRCA1/BRCA2 (Selected Variants).

The 23andMe Personal Genome Service (PGS) Risk Report for BRCA1/BRCA2 (Selected Variants) is indicated for the reporting of the following 44 variants in the BRCA1 and BRCA2 genes.

BRCA1: c.68 69del, c.213-11T>G, c.427G>T, c.815 824dup, c.1556del, c.1687C>T, c.1961del, c.2681 2682del, c.2864C>A, c.3481 3491del, c.3598C>T, c.3627dup, c.3756 3759del, c.3770 3771del, c.4035del, c.4065 4068del.c.4327C>T.c.4357+1G>A.c.4964 4982del.c.4986+6T>G.c.5123C>A.c.5177 5180del.c.5266dup

BRCA2: c.658 659del, c.771 775del, c.2808 2811del, c.2957 2958insG, c.3170 3174del, c.3264dup, c.3545 3546del, c.3847 3848del, c.4471 4474del, c.5542del, c.5576 5579del, c.5682C>G, c.5946del, c.6037A>T, c.6275 6276del, c.7024C>T, c.7480C>T, c.7934del, c.8904del

The report describes if a person's genetic result is associated with an increased risk of developing breast cancer and ovarian cancer and may be associated with an increased risk for prostate cancer, and potentially other cancers. The variants included in this report do not represent the majority of the BRCA1/BRCA2 variants in people of most ethnicities. The test report does not describe a person's overall risk of developing any type of cancer, and the absence of a variant tested does not rule out the presence of other variants that may be cancer-related. This report is for over-the-counter use by adults over the age of 18, and provides genetic information to inform discussions with a healthcare professional. This test is not a substitute for visits to a healthcare provider for recommended screenings or appropriate follow-up and should not be used to determine any treatments.

Customer saliva specimens are self-collected using the Oragene-Dx® Device manufactured by DNA Genotek, Inc. cleared by FDA for use with the PGS device under K141410, which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to one of two Clinical Laboratory Improvement Amendments (CLIA) certified laboratories for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents and instrumentation manufactured by Illumina.

The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe for analysis and interpretation. The raw data received is analyzed using 23andMe's proprietary Coregen software, where a genotype is determined for each tested SNP. The results for certain of these SNPs are used to generate personalized reports for the customer that provide information about the detected genotype.

Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which genetic health risk variant(s) have been detected in their sample and provide information about the disease associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically and clinically valid information about the risks associated with the presence of a particular variant. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results.

Here's a breakdown of the acceptance criteria and the study that proves the device meets them, based on the provided FDA 510(k) summary for the 23andMe Personal Genome Service (PGS) Genetic Health Risk Report for BRCA1/BRCA2 (Selected Variants):

Acceptance Criteria and Device Performance

1. Table of Acceptance Criteria and Reported Device Performance

• Negative Percent Agreement (NPA) ≥ 99% for each SNP (variant) compared to bidirectional Sanger sequencing. | - Achieved: Greater than 99% agreement for both PPA and NPA across all 41 tested variants. The study concluded that the 23andMe assay is comparable to bidirectional Sanger sequencing. |
  | Precision (Reproducibility) | - Minimum of 99% correct genotype calls at each of two laboratory sites. | - Achieved: 100% correct genotype calls for all samples across multiple days, operator teams, instruments, and reagent lots at 2 independent laboratory sites.
• Greater than 99% reproducibility and greater than 99% repeatability. |
  | Minimum DNA Input (Sensitivity) | - At least 95% of samples yielding the correct call at the minimum DNA concentration. | - Achieved: 100% correct genotype calls for all samples and reagent lots tested at DNA concentrations of 5, 15, and 50 ng/µL.
• Valid for samples with a DNA concentration range of 5 ng/µL to 50 ng/µL (well within the SOP requirement of 15 ng/µL to 50 ng/µL). |

2. Sample Size and Data Provenance

• Test Set Sample Size: The exact number of individual samples used for the "Method Comparison" (Accuracy), "Precision" (Reproducibility), and "Minimum DNA Input" studies is not explicitly stated as a numerical count. However, the document mentions that samples were "randomly selected from the 23andMe customer database" and that "all 41 variants were included in this study" for each of these performance tests.
• Data Provenance: The document states that samples were identified from the "23andMe customer database." This implies real-world, likely retrospective, data from individuals who have used 23andMe's services. The country of origin is not explicitly stated for individual samples, but 23andMe is a US-based company, suggesting the data is primarily from the US.

3. Number of Experts and Qualifications for Ground Truth

• This device is a genetic test, and the ground truth for performance testing (accuracy) was established by bidirectional Sanger sequencing, which is a widely accepted and highly accurate method for DNA sequencing.
• The document does not mention the use of human experts (e.g., radiologists) in establishing the ground truth for the analytical validity studies. The "truth" in this context is the confirmed genetic sequence obtained through an orthogonal, highly accurate laboratory method (Sanger sequencing), not human interpretation of images or clinical outcomes.

4. Adjudication Method for the Test Set

• Adjudication methods (e.g., 2+1, 3+1) are typically used in studies involving human interpretation or subjective assessments. Since the performance studies for this genetic test rely on objective laboratory methods (genotyping by the device vs. Sanger sequencing as ground truth), no human adjudication method was needed or applied for the analytical performance tests. The comparison was directly between the device's genotype call and the Sanger sequencing result.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

• No, an MRMC comparative effectiveness study was not done. MRMC studies are relevant for imaging devices or other medical technologies where human readers interpret output, and the goal is to assess how a device affects human performance. This is a direct-to-consumer genetic test (an in vitro diagnostic device) that directly provides genotype information, not an interpretation aid for human experts. Therefore, the concept of human readers improving with AI assistance is not applicable here.

6. Standalone (Algorithm Only) Performance

• Yes, standalone performance was done for the analytical validation. The performance criteria (PPA, NPA, correct genotype calls) were measured for the 23andMe PGS assay (the "algorithm only" in this context of a genetic test) against the ground truth established by Sanger sequencing. The results presented in the acceptance criteria table (e.g., ">99% agreement," "100% correct calls") represent the standalone performance of the device. There isn't a "human-in-the-loop" aspect to the genotyping process itself; the device generates the genotype call automatically.

7. Type of Ground Truth Used

• The ground truth used for the analytical validation test set was bidirectional Sanger sequencing, considered the gold standard for confirming specific DNA sequences and variants.

8. Sample Size for the Training Set

• The document does not specify the sample size for the training set. This is typical for in vitro diagnostic devices like genetic tests where the underlying technology (genotyping method) is well-established. The performance is primarily evaluated through robust analytical validation on a test set, rather than machine learning model training. The device leverages a "customized genotyping beadchip" and "proprietary Coregen software," indicating a deterministic or rule-based approach for genotype calling, or a machine learning model that was developed and validated internally but whose training set details are not part of this 510(k) summary.

9. How Ground Truth for the Training Set Was Established

• Given that the document does not explicitly mention a training set or machine learning model development in the traditional sense, it does not describe how ground truth for a training set was established. For molecular genetic tests, the principles of operation are typically based on molecular biology and chemistry, with performance validated through analytical studies against highly accurate reference methods like Sanger sequencing, rather than training on labeled datasets. If machine learning is involved in the "Coregen software" for genotype determination, the ground truth for any such training would logically have been established through a similar process of comparing assay results to known, validated reference genotypes (e.g., from Sanger sequencing or other established methods).

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) is a qualitative genotyping assessment system applied to genomic DNA isolated from human saliva collected using the Oragene Dx OGD-500.001 to simultaneously detect, report, and interpret genetic variants in a broad multigene test. The assessment system is intended to enable users to access information about their genetics that could aid discussions with a healthcare professional. The 23andMe Personal Genome Service Pharmacogenetic Reports are indicated for reporting of the following variants:

Gene: CYP2C19 Variant(s): *2, *3, *17
Gene: CYP2C9 Variant(s): *2, *3, *5, *6, rs7089580
Gene: CYP3A5 Variant(s): *3
Gene: UGT1A1 Variant(s): *6, *28
Gene: DPYD Variant(s):*2A, rs67376798
Gene: TPMT Variant(s): *2, *3C
Gene: SLCO1B1 Variant(s): c.521T>C (rs4149056)
Gene: CYP2D6 Variant(s): *2, *3, *4, *5, *6, *7, *8, *9, *10, *11, *15, *17, *20, *29, *35, *40, *41

This report is for over-the-counter use by adults over the age of 18 and provides genetic inform discussions with a healthcare professional about metabolism of therapeutics.

The 23andMe Personal Genome Service pharmacogenetic reports for CYP2C9, CYP3A5, UGT1A1, DPYD, TPMT and CYP2D6 describe if a person has variants associated with metabolism of some therapeutics but does not describe if a person will or will not respond to a particular therapeutic and does not describe the association between detected variants and any specific therapeutic.

23andMe Personal Genome Service pharmacogenetics report for CYP2C19 describes if a person has variants associated with metabolism of some therapeutics and provides interpretive drug information regarding the potential effect of the identified metabolizer phenotype on citalopram and clopidogrel therapy.

23andMe Personal Genome Service pharmacogenetics report for SLCO1B1 describes if a person has variants associated with the processing of some therapeutics and provides interpretive drug information regarding the potential effect of the identified transport function phenotype on simvastatin therapy.

The PGS Pharmacogenetic Reports are not a substitute for visits to a healthcare professional. The information provided by this report should not be used to start, stop, or change any course of treatment.

The 23andMe Personal Genome Service (PGS) is a direct-to-consumer/over-the-counter, DNA testing service that provides information and tools for consumers to learn about and explore their DNA.

The PGS is a currently marketed, non-invasive genetic information service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies.

Customer saliva specimens are self-collected using the Oragene-Dx® Device manufactured by DNA Genotek, Inc. cleared by FDA for use with the PGS device (K141410, DEN140044, DEN160026. DEN170046. DEN180028. K182784. K193492. and K211499), which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to a Clinical Laboratory Improvement Amendments (CLIA) certified laboratory for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents and instrumentation manufactured by Illumina. The device simultaneously tests for more than 600,000 variants, including those reported under the previously authorized PGS test indications.

The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe for analysis and interpretation. The raw data received is analyzed using 23andMe's proprietary Coregen software, where a genotype is determined for each tested SNP. The results for certain of these SNPs are used to generate personalized reports for the customer that provide information about the detected genotype.

Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on the metabolizer or transporter profile associated with the variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the metabolizer or transport function phenotype associated with the presence of a particular variant, or a combination of variants.

In the pharmacogenetic report for SLCO1B1. information regarding interpretive drug information to certain medications will be provided to the user in a medication "mini report", which is accessed via a link in the pharmacogenetic report for SLCO1B1. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results.

As noted in Table 5.2, the PGS assay components for the SLCO1B1 Drug Transport report such as the custom beadchip, reagents, and instrumentation are the same as the predicate devices. No new reagents were needed and the beadchip was unchanged to test for the c.521T>C (rs4149056) variant. The probes to detect c.521T>C (rs4149056) already existed on the beadchip.

The novel components in this Traditional 510(k) submission are to provide interpretive drug information to one specific medication (simvastatin), and to remove the limitation language requiring confirmatory testing in the 23andMe pharmacogenetics report for SLCO1B1. Pharmacogenetic reports for other genes authorized in DEN180028 will not be modified to remove the confirmatory testing limitation, include interpretive drug information, or add a prescription indication.

Engineering drawings, schematics, etc. of the 23andMe Personal Genome Service (PGS) Pharmacogenetic Reports are not applicable to this device.

The provided document, a 510(k) Summary for the 23andMe Personal Genome Service (PGS) Pharmacogenetic Reports (K221885), focuses on the modifications to the SLCO1B1 pharmacogenetic report. The primary changes are the addition of interpretive drug information for simvastatin and the removal of the confirmatory testing requirement for SLCO1B1. The summary details analytical performance studies conducted to support these changes.

Here's a breakdown of the requested information based on the provided document:

1. Table of acceptance criteria and the reported device performance

The document specifies acceptance criteria for analytical performance studies. These are primarily related to accuracy (method comparison) and precision (reproducibility) for the c.521T>C (rs4149056) variant in the SLCO1B1 gene.

2. Sample size used for the test set and the data provenance

• Test Set Sample Size: The document does not explicitly state a total numerical sample size for the method comparison and precision studies. It mentions that for the method comparison, samples were selected from 23andMe's database to enrich for prevalent variants in specific ethnicities, and allele and diplotype frequencies were used to inform the number of samples selected. For the precision study and MDI, "intended use (saliva) samples were selected from the 23andMe customer biobank based on their putative genotype" and "obtained for each of the c.521C>T genotype combinations." While specific numbers aren't given, the language suggests a controlled selection of samples representing relevant genotype combinations.
• Data Provenance: The samples for performance testing were obtained from the "23andMe customer biobank." This suggests that the data is from real-world customer samples. The document does not specify the country of origin of these customers, but 23andMe is a US-based company, implying a significant US customer base. The data would be considered retrospective as it's from pre-existing biobank samples.

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts

• The ground truth for the analytical performance studies (method comparison, precision, MDI) was established using bidirectional Sanger sequencing, which is a widely accepted laboratory gold standard for genetic variant confirmation.
• The document does not mention the use of human experts (like radiologists or genetic counselors) to establish the ground truth for these analytical performance tests, as the ground truth is a direct genetic sequencing result, not an interpretation of an image or a clinical diagnosis.

4. Adjudication method (e.g., 2+1, 3+1, none) for the test set

• Since the ground truth for the analytical performance studies was established by bidirectional Sanger sequencing, there was no human adjudication method employed for these specific tests. The comparison was directly between the device's genotyping result and the sequencing result.

5. If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

• No MRMC comparative effectiveness study was performed or is relevant for this type of device. The 23andMe Personal Genome Service is a direct-to-consumer genetic testing product that provides raw genetic data and interpretive reports; it is not an AI-assisted diagnostic imaging tool or a system where human readers interpret data with or without AI assistance. The "human readers" in this context are the end-users (consumers) who read the reports, and their "improvement" is not measured in the traditional MRMC sense.
• The document does mention "robust user comprehension testing, previously reviewed and authorized under DEN180028 and K193492," indicating that the readability and understanding of the reports by users were assessed. However, this is distinct from an MRMC study and the specific results of this comprehension testing (e.g., effect size) are not detailed in this 510(k) summary, only referenced as having been sufficient for prior clearances.

6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done

• Yes, standalone performance was evaluated. The method comparison, precision, and minimum DNA input studies directly assess the accuracy and reliability of the 23andMe BeadChip assay and its associated software (Genome Studio, Coregen) in determining genotypes, without direct human intervention or interpretation during the genotyping process itself. The reported performance (e.g., 100% agreement with Sanger sequencing) is the standalone performance of the genotyping system.

7. The type of ground truth used (expert consensus, pathology, outcomes data, etc.)

• The ground truth used for the analytical performance studies (method comparison, precision, MDI) was bidirectional Sanger sequencing, which is a highly accurate and widely accepted method for determining the true genetic sequence/variant status.

8. The sample size for the training set

• The document does not explicitly state the sample size for the training set for the device's algorithms. As the device involves a genotyping array and software for data analysis (Coregen), it's likely that a substantial amount of genetic data was used in the development and initial training/optimization of these systems. However, this 510(k) summary focuses on the validation of modifications to a previously cleared device rather than the initial development of the core genotyping platform.

9. How the ground truth for the training set was established

• The document does not detail how the ground truth for the training set (if any specific to training was used for this modification) was established. For a genotyping platform like this, the "training" (or more accurately, the development and verification) would typically involve using samples with known genetic variations confirmed by gold-standard methods like Sanger sequencing or whole-genome sequencing. The 510(k) focuses on the validation of the current device's performance against a gold standard (Sanger sequencing).

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) uses qualitative genotyping to detect select clinically relevant variants in genomic DNA isolated from human saliva collected from individuals ≥18 years for the purpose of reporting and interpreting genetic health risks, including the 23andMe PGS Genetic Health Risk Report for Hereditary Prostate Cancer (HOXB13-Related). The 23andMe PGS Genetic Health Risk Report for Hereditary Prostate Cancer (HOXB13- Related) is indicated for reporting of the G84E variant in the HOXB13 gene. The report describes if a person has the G84E variant and if a male is at increased risk for prostate cancer. The variant included in this report is most common in people of European descent. The test report does not describe a person's overall risk of developing any type of cancer, and the absence of a variant tested does not rule out the presence of other variants that may be cancer-related. This test is not a substitute for visits to a healthcare provider for recommended screenings or appropriate follow-up and should not be used for diagnosis, to determine any treatments or medical interventions.

The 23andMe Personal Genome Service (PGS) is an over-the-counter (direct-to-consumer), DNA testing service that provides information and tools for consumers to learn about and explore their DNA.

The 23andMe Personal Genome Service (PGS) is a currently marketed, non-invasive genetic information service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies. It is a home use, over-thecounter (direct-to-consumer) DNA testing service intended to provide information and tools for consumers to learn about and explore their DNA.

Customer saliva is self-collected using the Oragene-Dx® Device manufactured by DNA Genotek, Inc. (previously cleared for carrier screening indications under K141410, and the same collection kit used to generate performance data for DEN140044, DEN160026, DEN170046, K182784, DEN180028, and K193492, which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to one of our Clinical Laboratory Improvement Amendments (CLIA) certified laboratories for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, and off the shelf reagents and instrumentation manufactured by Illumina. The multiplex assay simultaneously tests for more than 500,000 variants, including those for the previously authorized indications, as well as for the indications proposed herein.

Raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe. The data is then analyzed using 23andMe's proprietary Coregen software, where a genotype is determined for each tested SNP. The results for certain of these SNPs are used to generate personalized reports for the customer that provide information about the detected genotype.

Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which genetic health risk variant(s) have been detected in their sample and provide information about the disease associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the risks associated with the presence of a particular variant. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results.

The modified components of the Personal Genome Service included in this 510(k) submission are new labeling to include (a) one new variant to be reported, and (b) the qualitative reporting of one's Genetic Health Risk for Hereditary Prostate Cancer (HOXB13-Related).

Engineering drawings, schematics, etc. of Genetic Health Risk Report for Hereditary Prostate Cancer (HOXB13-Related) are not applicable to this device.

The provided document describes the acceptance criteria and study proving the device meets these criteria for the 23andMe PGS Genetic Risk Report for Hereditary Prostate Cancer (HOXB13-Related).

Here's the breakdown of the information requested:

1. Table of Acceptance Criteria and Reported Device Performance

2. Sample Sizes Used for the Test Set and Data Provenance

• Accuracy/Method Comparison Study:
  • Sample Size: Not explicitly stated as a number, but "Saliva samples were selected from the 23andMe customer biobank, based on their predetermined genotype and minimum volume required for testing." This implies a varied sample size based on the availability of specific genotypes.
  • Data Provenance: From the "23andMe customer biobank" and "approved contract laboratory sites." The origin of the customers is not specified beyond "23andMe customer" which is a US-based company, suggesting primarily US data. The study was retrospective, using pre-existing samples from the biobank.
• Precision Study:
  • Sample Size: "DNA samples were selected based on their confirmed genotypes, and were obtained from the 23andMe biobank." Not an explicit number.
  • Data Provenance: From the "23andMe biobank." Implies primarily US data, retrospective.
• DNA Input Study:
  • Sample Size: "DNA samples were obtained from the 23andMe biobank based on their listed genotypes." Not an explicit number.
  • Data Provenance: From the "23andMe biobank." Implies primarily US data, retrospective.
• Interfering Substance Study (referenced from DEN140044):
  • Sample Size: Over 35,000 sample replicates.
  • Data Provenance: Not explicitly stated for this particular study, but given it's for a US regulatory submission by a US company, it's highly likely to be US data, retrospective.
• Labeling Comprehension Study (referenced from DEN160026):
  • Sample Size: Not explicitly stated.
  • Data Provenance: Not explicitly stated, but also likely US data.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications of Those Experts

• Ground Truth Method: For the analytical studies (Method Comparison, Precision, DNA Input), the ground truth for genotyping was established by bi-directional Sanger sequencing.
• Number/Qualifications of Experts: The document does not specify the number or qualifications of experts involved in performing or interpreting the Sanger sequencing results to establish the "truth." It only states that sequencing was performed "by an approved supplier" and that the sequencing results were "considered to be 'truth.'"

4. Adjudication Method for the Test Set (e.g., 2+1, 3+1, none)

• The document does not describe any human adjudication method for establishing the ground truth from Sanger sequencing. It implies that the sequencing results themselves were directly taken as ground truth without further expert consensus or adjudication.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

• No MRMC or comparative effectiveness study involving human readers (e.g., radiologists) with or without AI assistance was performed or described. This device is a direct-to-consumer genetic test, not an imaging-based AI diagnostic tool.
• The closest concept is the "Labeling Comprehension" study, which assesses how well consumers understand the report. It indicates that the report and educational materials were effective in communicating relevant concepts for safe use. This is a measure of user comprehension, not human reader improvement with AI assistance.

6. If a Standalone (i.e. algorithm only without human-in-the-loop performance) was done

• Yes, the performance studies (Accuracy, Precision, DNA Input, Interfering Substance) represent a standalone evaluation of the genotyping assay, which is essentially the "algorithm" or technical process of the device. The accuracy and precision figures are "algorithm only" performance metrics, as they compare the device's genotype calls directly against Sanger sequencing as the ground truth.

7. The Type of Ground Truth Used (expert consensus, pathology, outcomes data, etc.)

• For the analytical performance studies (Accuracy, Precision, DNA Input), the ground truth for specific genetic variants (genotype) was established by bi-directional Sanger sequencing.
• For the clinical performance, the document refers to "published studies of variant frequencies in various populations and the results of analytical studies" and "allele frequencies in the 23andMe customer database." This relies on established scientific literature and aggregated anonymized real-world data rather than individual outcomes or pathology reports.

8. The Sample Size for the Training Set

• The document primarily describes validation studies (test sets) for the analytical performance of the device. It does not provide information about a separate "training set" sample size for developing the genotyping assay or the underlying "Coregen software." The genotyping method described relies on physical beadchip arrays and established principles of DNA analysis, not on a machine learning model that would typically have a distinct training phase with a dedicated dataset.
• The "Customer biobank" is used for selecting samples for the performance studies, which may implicitly reflect data used in the development or refinement of their overall genotyping process, but it's not explicitly defined as a separate 'training set' for an AI model.

9. How the Ground Truth for the Training Set Was Established

• As mentioned above, the document does not elaborate on a distinct "training set" with established ground truth in the context of an AI/ML model for this genetic test. The "Coregen software" analyzes raw data from the beadchip, and its accuracy is validated against Sanger sequencing. The development process of this proprietary software, and any data used to "train" it (if it involves statistical modeling beyond simple rule-based interpretation of genotyping signals), is not detailed in terms of ground truth establishment.

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) is a qualitative genotyping assessment system applied to genomic DNA isolated from human saliva collected using the Oragene Dx OGD-500.001 to simultaneously detect, report, and interpret genetic variants in a broad multigene test. The assessment system is intended to enable users to access information about their genetics that could aid discussions with a healthcare professional. The 23andMe Pharmacogenetic Reports are indicated for reporting of the following variants:

Gene: CYP2C19 Variant(s): *2, *3, *17
Gene: CYP2C9 Variant(s): *2, *3, *5, *6, rs7089580
Gene: CYP3A5 Variant(s): *3
Gene: UGT1A1 Variant(s): *6, *28
Gene: DPYD Variant(s):*2A, rs67376798
Gene: TPMT Variant(s): *2, *3C
Gene: SLC01B1 Variant(s): *5
Gene: CYP2D6 Variant(s): *2, *3, *4, *5, *6, *7, *8, *9, *10, *11, *15, *17, *20, *29, *35, *40, *41

This report is for over-the-counter use by adults over the age of 18 and provides genetic information to inform discussions with a healthcare professional about metabolism of therapeutics.

The 23andMe Personal Genome Service pharmacogenetic reports for CYP2C9, CYP3A5, UGT1A1, DPYD, TPMT, SLC01B1 and CYP2D6 describe if a person has variants associated with metabolism of some therapeutics, but does not describe if a person will or will not respond to a particular therapeutic, and does not describe the association between detected variants and any specific therapeutic.

23andMe Personal Genome Service pharmacogenetic reports for CYP2C19 describes if a person has variants associated with metabolism of some therapeutics and provides interpretive drug information regarding the potential effect of the identified metabolizer phenotype on citalopram and clopidogrel therapy.

The PGS Pharmacogenetic Reports are not a substitute for visits to a healthcare professional. The information provided by this report should not be used to start, stop, or change any course of treatment.

The 23andMe Personal Genome Service (PGS) is an over-the-counter (direct-to-consumer), DNA testing service that provides information and tools for consumers to learn about and explore their DNA.

The PGS is a currently marketed, non-invasive genetic information service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies.

Customer saliva specimens are self-collected using the Oragene-Dx® Device manufactured by DNA Genotek, Inc. cleared by FDA for use with the PGS device (K141410, DEN140044, DEN160026, DEN170046, DEN180028, and K182784), which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to a Clinical Laboratory Improvement Amendments (CLIA) certified laboratory for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents and instrumentation manufactured by Illumina. The device simultaneously tests for more than 600,000 variants, including those reported under the previously authorized PGS test indications.

The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe for analysis and interpretation. The raw data received is analyzed using 23andMe's proprietary Coregen software, where a genotype is determined for each tested SNP. The results for certain of these SNPs are used to generate personalized reports for the customer that provide information about the detected genotype.

Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on metabolizer or transporter profile associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the metabolizer or transporter phenotype associated with the presence of a particular variant, or a combination of variants. In the pharmacogenetic report for CYP2C19, information regarding interpretive drug information to certain medications will be provided to the user in a medication "mini report", which is accessed via a link in the pharmacogenetic report for CYP2C19. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results.

The novel components in this traditional 510(k) submission are to provide interpretive drug information to two specific medications (citalopram and clopidogrel), and to remove the limitation language requiring confirmatory testing in the 23andMe pharmacogenetic report for CYP2C19. Pharmacogenetic reports for the other genes authorized in DEN180028 will not be modified to include interpretive drug information, or remove the confirmatory testing limitation.

Here's an analysis of the acceptance criteria and the supporting study for the 23andMe Personal Genome Service (PGS) Pharmacogenetic Reports, specifically focusing on the CYP2C19 gene, as described in the provided document:

1. Table of Acceptance Criteria and Reported Device Performance

The document focuses on the analytic performance of the genotyping assay for CYP2C19 variants (*2, *3, *17). The primary acceptance criteria provided relate to accuracy (method comparison) and precision (reproducibility).

2. Sample Sizes and Data Provenance for the Test Set

• Method Comparison (Accuracy) & Precision (Reproducibility) Test Sets:

  • "Blinded Study 1" (single lab site, single day): Five 96-well plates of "incoming saliva samples." A specific numerical count of samples is not given, but 5 plates x 96 wells/plate = 480 samples.
  • Coriell Reference Samples: Genomic DNA for the three CYP2C19 variants of interest (*2, *3, *17) was obtained from the Coriell Institute for Medical Research. The exact number of Coriell samples used is not specified.
  • "Blinded Study 2" (East Asian ancestry): Intended use (saliva) samples "randomly selected in an unbiased manner from the 23andMe biobank based on their East Asian genetic ancestry." Numbers not specifically given. The purpose was to increase the likelihood of rare *3 allele combinations.
  • Supplemental Precision Study (saliva): Intended use (saliva) samples selected from the 23andMe customer biobank based on their "putative genotype." Numbers not specifically given but tested over 3 days, with 3 lots of reagents, by unique operator teams, using 3 different serial numbers of each of 2 instruments (Tecan and iScan), at each of 2 laboratory sites.
  • Minimum DNA Input (MDI) Test Set:
    • DNA samples obtained from Coriell and the 23andMe biobank. Exact numbers not specified, but tested at 3 different DNA concentrations with 3 reagent lots.
    • Supplemental MDI Study (saliva): Intended use (saliva) samples selected from the 23andMe customer database based on their "putative genotype." Numbers not specified, but each sample was diluted to 3 different DNA concentrations and genotyped with 3 reagent lots.

• Data Provenance:

  • Retrospective: Samples from the 23andMe biobank and customer database are retrospective, representing previously collected data.
  • Prospective (Implicit): "Incoming saliva samples" in the first blinded study could imply prospective collection for that specific study's initiation if they were fresh samples received for routine processing on that day.
  • External Reference: Coriell Institute for Medical Research samples are an external, well-characterized source.
  • Country of Origin: Not explicitly stated for all samples, but the 23andMe customer base is likely predominantly from the US and other countries where they operate.

3. Number of Experts and Qualifications for Ground Truth Establishment (Test Set)

This device is a genotyping system, and the ground truth for its analytical performance is established through genomic sequencing, not expert interpretation of phenotypic data or images.

• No human experts were used to establish the ground truth in the traditional sense of clinical interpretation.
• The "experts" in this context are the robust and established methods of bi-directional Sanger sequencing and Coriell Institute's characterization of their reference materials. These methods are considered the "source of truth" for genetic sequences and genotypes.

4. Adjudication Method for the Test Set

• No human adjudication method (like 2+1, 3+1) was used as the ground truth is based on objective genetic sequencing data.
• The comparison involved directly comparing the device's genotyping results against the known sequences/genotypes derived from Sanger sequencing or Coriell's reference data. Discrepancies would likely lead to re-sequencing or further investigation to ascertain the true genotype.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

• No, a multi-reader multi-case (MRMC) comparative effectiveness study was not done.
• This device is an automated genotyping system, not one that involves human "readers" interpreting output in a diagnostic setting in the way an imaging AI might. Its primary function is to accurately identify genetic variants. The "interpretive drug information" is a downstream output of the accurate genotyping, not something itself interpreted by multiple human readers in a study of this type.

6. Standalone Performance Study (Algorithm Only)

• Yes, the performance studies described (Method Comparison, Precision, Minimum DNA Input) are all standalone performance studies of the genotyping algorithm and associated laboratory process.
• The studies evaluate the accuracy and reliability of the device itself (the genotyping platform, reagents, and analysis software) in determining genetic variants, separate from how a human might interpret the final report.

7. Type of Ground Truth Used

• Expert Consensus (Genomic): The ground truth for the analytical performance studies was established using:
  • Bi-directional Sanger Sequencing: A gold standard method for DNA sequencing, considered highly accurate for confirming specific genetic variants.
  • Reference Samples from Coriell Institute for Medical Research: These are well-characterized genomic DNA samples with established and verified genotypes, essentially an expert-characterized reference.

8. Sample Size for the Training Set

• The document does not explicitly state a separate "training set" for the genotyping algorithm. Genetic genotyping algorithms often rely on established biochemical principles and mapping to known reference genomes rather than machine learning on a large, labeled training dataset of raw signals to call genotypes.
• However, the "23andMe database" is mentioned, with variant and allele frequencies from "8,004,302 customers" and "16,008,604 alleles." This massive dataset, along with public databases like gnomAD, would likely be used in the development and refinement of allele frequency estimation and potentially for algorithm fine-tuning or quality control, but not as a distinct "training set" in the typical AI sense for classifying novel inputs.

9. How the Ground Truth for the Training Set was Established

• Given that a distinct "training set" as understood in machine learning is not explicitly detailed, the ground truth establishment for algorithm development would implicitly rely on the same principles as the test set: established genetic sequencing methodologies (like Sanger sequencing) and comprehensive genetic reference databases (like the human reference genome and population-specific variant databases). These resources allow for the accurate mapping and identification of genetic variants within the raw genotyping data.

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) uses qualitative genotyping to detect select clinically relevant variants in genomic DNA isolated from human saliva collected from individuals ≥18 years with the Oragene Dx model OGD500.001 for the purpose of reporting and interpreting genetic health risks, including the 23andMe PGS Genetic Health Risk Report for MUTYH-Associated Polyposis. The 23andMe PGS Genetic Health Risk Report for MUTYH-Associated Polyposis is indicated for reporting of the Y179C and the G396D variants in the MUTYH gene. The report describes if a person is at increased risk of developing colorectal cancer. The two variants included in this report are most common and best studied in people of Northern European descent and may not represent the majority of the MUTYH variants found in people of other ethnicities. The test report does not describe a person's overall risk of developing any type of cancer, and the absence of a variant tested does not rule out the presence of other variants that may be cancer-related. This test is not a substitute for visits to a healthcare provider for recommended screenings or appropriate follow-up and should not be used to determine any treatments.

The 23andMe Personal Genome Service (PGS) is an over-the-counter (direct-to-consumer), DNA testing service that provides information and tools for consumers to learn about and explore their DNA. The 23andMe Personal Genome Service (PGS) is a currently marketed, non-invasive genetic information service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies. It is a home use, over-the-counter (direct-to-consumer) DNA testing service intended to provide information and tools for consumers to learn about and explore their DNA. Customer saliva is self-collected using the Oragene-Dx® Device manufactured by DNA Genotek, Inc. Once the sample is collected, it is shipped to one of our Clinical Laboratory Improvement Amendments (CLIA) certified laboratories for testing. DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents and instrumentation manufactured by Illumina. The multiplex assay simultaneously tests for more than 500,000 variants, including those for the previously authorized indications, as well as for the indications proposed herein. The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe. The data is then analyzed using the 23andMe's proprietary Coregen software, where a genotype is determined for each tested SNP. The results for certain of these SNPs are used to generate personalized reports for the customer that provide information about the detected genotype. Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on the risk of disease associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the risks associated with the presence of a particular variant. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results. The novel components in this traditional 510(k) submission are only (a) the variants to be reported, and (b) the qualitative reporting of risk for MAP.

Here's a breakdown of the acceptance criteria and study information for the 23andMe Personal Genome Service (PGS) Genetic Health Risk Report for MUTYH-Associated Polyposis (MAP), based on the provided document:

1. Table of Acceptance Criteria and Reported Device Performance:

2. Sample Size Used for the Test Set and Data Provenance:

• Accuracy (Method Comparison): Saliva samples were selected from the 23andMe customer biobank based on pre-determined BeadChip genotype and minimum volume required. The exact number of samples for the MUTYH variants specifically isn't explicitly stated, but the study compared against Sanger sequencing.
• Precision/Reproducibility: Confirmed genotype DNA samples (commercially available or from 23andMe biobank) for G3696 variants and Y179C common homozygous and heterozygous variants. The Y179C rare homozygous variant was excluded. The exact number of samples for this specific study is not provided, but it involved "multiple days, operator teams, instruments, and reagent lots at two independent laboratory sites."
• DNA Input: DNA samples obtained from commercial sources or 23andMe biobank based on listed genotypes. The exact number of samples for this study is not provided. The study involved diluting each sample to 3 different DNA concentrations and genotyping in a blinded fashion using 3 lots of reagents.
• Interfering Substance: "More than 35,000 sample replicates were tested" across various prior studies (DEN140044).
• Clinical Performance (Allele Frequencies): Frequencies are based on approximately:
  • 872,000 individuals with European ancestry
  • 47,500 individuals with African-American ancestry
  • 38,500 individuals with Ashkenazi Jewish ancestry
  • 35,000 individuals with East Asian ancestry
  • 123,500 individuals with Hispanic/Latino ancestry
  • 10,000 individuals with South Asian ancestry
• Data Provenance: The document generally indicates the use of the 23andMe customer biobank for many of the analytical studies. For allele frequencies, it uses both the 23andMe Database and published literature (e.g., ExAc database, Poulsen and Bisgaard 2008). These would primarily be retrospective data sets from existing samples or databases.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Their Qualifications:

The document does not explicitly state the number of experts or their qualifications for establishing ground truth for the test set specifically.

• Accuracy (Method Comparison): Ground truth was established by Sanger bi-directional DNA sequencing, which is a laboratory gold standard method, not by human expert consensus or interpretation.
• Precision/Reproducibility: Ground truth was established by bi-directional Sanger sequencing for each sample.
• DNA Input: Ground truth was established by bi-directional Sanger sequencing for each sample.
• Labeling Comprehension: Human input was used to assess comprehension rates among typical users, but these are not "experts" establishing a clinical ground truth.

4. Adjudication Method for the Test Set:

Not applicable, as the ground truth for analytical studies was primarily established by Sanger sequencing, a direct molecular method, rather than expert interpretation requiring adjudication.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:

No, a Multi-Reader Multi-Case (MRMC) comparative effectiveness study was not done. The device (23andMe PGS Genetic Health Risk Report for MUTYH-Associated Polyposis) is a standalone genetic test for direct-to-consumer use. The studies performed focused on the analytical performance (accuracy, precision, DNA input, interfering substances) and labeling comprehension of the report itself, not on how human readers (e.g., clinicians) would improve their diagnostic accuracy with or without AI assistance. The report is intended to inform conversations with healthcare providers, not to be directly used for diagnosis or treatment decisions by those providers without confirmatory testing.

6. Standalone (Algorithm Only Without Human-in-the-Loop) Performance:

Yes, a standalone performance study was done. The vast majority of the analytical performance studies (Method Comparison, Precision/Reproducibility, DNA Input, Interfering Substances) evaluate the performance of the "23andMe BeadChip assay" and the "PGS multiplex assay" as an automated genotyping process without human intervention in the result generation. The "Coregen software" analyzes the data to determine genotypes, which are then used to generate personalized reports. This represents the algorithm's standalone performance in genotyping.

7. Type of Ground Truth Used:

• Analytical Studies (Accuracy, Precision, DNA Input): Sanger bi-directional DNA sequencing was consistently used as the "truth" for genotype confirmation against which the BeadChip assay results were compared. This is considered a gold standard molecular method.
• Clinical Performance: Allele frequencies were derived from the 23andMe customer database and published scientific literature/databases (e.g., ExAc database, Poulsen and Bisgaard 2008). This is not a "ground truth" for individual cases, but rather statistical data based on large populations and research findings.
• Labeling Comprehension: Assessed by measuring user understanding of genetic health risk concepts, not by a clinical or pathological ground truth.

8. Sample Size for the Training Set:

The document does not provide information on the sample size used for the training set of the 23andMe PGS system or its Coregen software. The studies described are primarily analytical validation studies of the genotyping platform and the specific variants being reported. The "BeadChip" is a pre-designed array, and while software (Coregen) interprets the data, details about its specific "training" with a dataset are not available in this regulatory submission summary. The system itself (the PGS platform) was previously authorized, and this submission focuses on adding new variants.

9. How the Ground Truth for the Training Set Was Established:

As the document does not detail a specific "training set" for the genotyping algorithm or system, there is no information on how its ground truth was established. The underlying technology (Illumina Infinium BeadChip, Illumina iScan, GenomeStudio software, and 23andMe's Coregen software) is a sophisticated genotyping workflow that would have been developed and validated by the manufacturers. For the specific variants (Y179C and G396D) being reported for MUTYH-Associated Polyposis, their clinical relevance and association with disease are based on established scientific literature and understanding of genetics, rather than a training dataset for the algorithm. The analytical studies confirm the assay's ability to accurately detect these variants, with Sanger sequencing acting as the ground truth for that detection capability.

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) is a qualitative genotyping assessment system applied to genomic DNA isolated from human saliva collected using the Oragene Dx OGD-500.001 to simultaneously detect, report, and interpret genetic variants in a broad multigene test. The assessment system is intended to enable users to access information about their genetics that could aid discussions with a healthcare professional. The 23andMe Personal Genome Service Pharmacogenetic Reports are indicated for reporting of the following variants:

This report is for over-the-counter use by adults over the age of 18 and provides genetic information to inform discussions with a healthcare professional about metabolism of therapeutics. This report describes if a person has variants associated with metabolism of some therapeutics, but does not describe if a person will or will not respond to a particular therapeutic, and does not describe the association between detected variants and any specific therapeutic. The PGS Pharmacogenetic Reports are not a substitute for visits to a healthcare professional. The information provided by this report should not be used to start, stop, or change any course of treatment.

The 23andMe PGS is a non-invasive DNA testing service that uses qualitative genotyping. It is a direct-to-consumer, over-the-counter, DNA genetic test. A user's saliva is self-collected using the Oragene Dx device manufactured by DNA Genotek, Inc. (previously cleared under K141410), which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to one of two Clinical Laboratory Improvement Amendments (CLIA)-certified laboratories for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents, and instrumentation. The multiplex assay simultaneously tests for more than 500,000 variants, including those for the previously authorized indications, as well as for the indication proposed herein.

The raw data is generated by the scanning instrument's software, and then sent to 23andMe (the Manufacturer). The data are analyzed using the Manufacturer's proprietary software, and a genotype is determined for each tested variant. The results for certain of these variants, as noted in the indications for use, are used to generate personalized reports for users that provide information about the predicted metabolic function of the tested variants.

Personalized reports are generated for each user to provide results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on the predicted metabolic function of the specific genetic variants. The genetic variants detected by the test are associated with the metabolism of some therapeutics. If no variant is detected, that information is also provided. If the association between the predicted metabolic function and the combination of detected variants has not been established, the report indicates that the results cannot be determined. The personalized reports are intended to present scientific concepts to users in an easy-to-understand format. The reports provide information about the association between the detected variant and the predicted metabolic function that has been associated with the metabolism of some drugs, further described below. The reports are designed to help users understand the meaning of their results and inform conversations with their doctor or other healthcare professional. The test reports do not provide any information on associations between the detected variants and any specific therapeutic and therefore, the test does not describe if a person will or will not respond to any specific therapeutic.

The 23andMe PGS Pharmacogenetic Reports detect 33 variants in 8 genes: CYP2C19, CYP2C9, CYP2D6, CYP3A5, CYP2D6, DPYD, TPMT, and UGT1A1. The 23andMe PGS Pharmacogenetic Reports provide information on the associated enzyme or protein function and the predicted metabolizer phenotype for variants in drug metabolizing enzymes: CYP2C19, CYP2C9, CYP2D6, CYP3A5, CYP2D6, DPYD, TPMT, and UGT1A1. The predicted metabolizer phenotype is identified according to the number and consequence of each allele where two no-function alleles are associated with being poor metabolizers, one no-function allele is associated with being an intermediate metabolizer, two functional alleles are associated with being normal metabolizers, one gain-of-function allele is associated with being a rapid metabolizer, and two gain-of-function alleles are associated with being an ultrarapid metabolizer. The predicted metabolizer phenotype or protein function is then used to provide information on the potential consequence on metabolism of some medications. For example, poor metabolizers may process some medications slower, internediate metabolizers may process some medications slightly slower than normal, normal metabolizers may process some medication at a normal rate. rapid metabolizers may process some medications slightly faster than normal, and ultrarapid metabolizers may process some medications faster than normal.

The 23andMe PGS Pharmacogenetic Report for SLCO1B1 will indicate that the detected variant is associated with a loss-of-function and slightly decreased transport of some medications.

Here's a breakdown of the acceptance criteria and the study details for the 23andMe Personal Genome Service (PGS) Pharmacogenetic Reports, based on the provided text:

Acceptance Criteria and Device Performance for 23andMe Personal Genome Service (PGS) Pharmacogenetic Reports

1. Table of Acceptance Criteria and Reported Device Performance (Analytical Performance for CYP2C19)

The document primarily details the analytical performance for **CYP2C19 variants (*2, 3, 17) as the representative study for the broader pharmacogenetic reports. Protocols and acceptance criteria for other genes (CYP2C9, CYP2D6, CYP3A5, DPYD, TPMT, UGT1A1, and SLCO1B1) were reviewed and found acceptable, with the plan for the sponsor to perform testing and add them if validation data meet criteria.

Note: The provided text does not explicitly state numerical acceptance criteria for performance metrics like PPA/NPA. However, the reported results demonstrate 100% agreement with the reference method (Sanger sequencing), which can be inferred as meeting implicit high accuracy requirements. The special controls mention the need for "Data appropriate, as determined by FDA, to demonstrate the analytical (i) accuracy and reliability of the device."

2. Sample Size and Data Provenance

• Test Set (Analytical Studies - CYP2C19):

  • Reproducibility/Precision:
    • CYP2C19*2: 4 samples (GG, AG, AA) at Site 1; 4 samples (GG, AG, AA) at Site 2. Each sample replicated 81 times. (Total 81 replicates * 4 samples * 2 sites = 648 technical replicates, plus FQCs/no calls).
    • CYP2C19*3: 4 samples (GG, AG, AA) at Site 1; 4 samples (GG, AG, AA) at Site 2. Each sample replicated 81 times. (Total 648 technical replicates, plus FQCs/no calls).
    • CYP2C19*17: 5 samples (CC, CT, TT) at Site 1; 5 samples (CC, CT, TT) at Site 2. Each sample replicated 81 times. (Total 810 technical replicates, plus FQCs/no calls).
    • Additional precision study: One sample of each genotype for each allele was genotyped in replicates of three at two sites.
    • Data Provenance: DNA samples obtained from an "external vendor." Not explicitly stated, but typically these are commercially available, characterized genomic DNA samples. The study was retrospective in nature as samples were pre-selected.
  • Limit of Detection (LoD): Not explicitly stated, but samples included homozygous common, heterozygous common, and homozygous rare genotypes for each allele (CYP2C19*2, *3, *17). Three replicates of each sample were diluted to three DNA concentrations.
    • Data Provenance: Samples obtained from an "external vendor." Retrospective.
  • Comparison with Sanger Bidirectional Sequencing (Accuracy):
    • CYP2C19*2: 47 (Homozygous Common), 49 (Heterozygous), 48 (Homozygous Rare).
    • CYP2C19*3: 48 (Homozygous Common), 45 (Heterozygous), 39 (Homozygous Rare).
    • CYP2C19*17: 49 (Homozygous Common), 45 (Heterozygous), 47 (Homozygous Rare).
    • Data Provenance: Saliva samples selected from the "23andMe customer biobank" based on predetermined genotypes. This implies a retrospective data provenance, and the potential for selection bias ("samples were chosen from a biobank based on the variants already detected by the candidate assay") is explicitly noted by the FDA. The country of origin of the biobank is not specified, but 23andMe is a US-based company, suggesting primarily US customers.

• Test Set (User Comprehension Study):

  • Sample Size: 602 participants completed the study; 594 subjects included in primary analysis (after excluding 8 careless responders/previous users/technical issues). A second analysis excluded 63 participants who did not scroll/read the reports, resulting in 531 participants.
  • Data Provenance: Participants were "demographically diverse (e.g., age, education) set of users naive to the study subject." The direct-to-consumer nature implies diverse backgrounds, likely from the US, but not explicitly stated. This would be considered a prospective study as participants were recruited specifically for this evaluation.

3. Number of Experts and Qualifications for Ground Truth

• Analytical Studies (Reproducibility, LoD, Comparison with Sanger):

  • Number of Experts: Not applicable in the traditional sense of clinical experts.
  • Qualifications: The ground truth for the DNA samples used in analytical studies was established by bidirectional Sanger sequencing, which is considered a gold standard for genetic sequencing. The "external vendor" supplying samples would also have characterized them.

• User Comprehension Study:

  • Number of Experts: The study mentioned "moderators of the test" for identifying participants who did not scroll/read. Specific qualifications of these moderators are not provided. The "primary endpoint analysis" and "second analysis" indicate a statistical approach rather than expert consensus on individual comprehension.

4. Adjudication Method for the Test Set

• Analytical Studies:

  • Method: For reproducibility, LoD, and accuracy studies, the 23andMe PGS test results were compared to the results obtained from bidirectional Sanger sequencing. Discrepancies (incorrect calls) were counted. This is effectively a direct comparison to a gold standard, not a consensus-based adjudication among multiple interpretations of the same test.
  • Specifics (e.g., 2+1): Not applicable, as it's a comparison to a single reference method.

• User Comprehension Study:

  • Method: Quantitative assessment of participant answers to comprehension questions. Not a traditional adjudication method.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

• Was one done? No, a MRMC comparative effectiveness study involving human readers improving with AI vs. without AI assistance was not mentioned or conducted for this device.
  • This device is a qualitative genotyping microarray intended for direct-to-consumer use, with results used to inform discussions with healthcare professionals. It does not involve a diagnostic image interpretation or similar task where human readers would typically "improve" with AI assistance in the way an MRMC study would measure.
• Effect Size: Not applicable.

6. Standalone Performance Study

• Was one done? Yes, the entire analytical performance section (Reproducibility/Precision, Limit of Detection, Analytical Specificity, Comparison with Sanger bidirectional Sequencing) represents a standalone (algorithm only, without human-in-the-loop performance) evaluation of the 23andMe PGS test's ability to accurately detect genetic variants. The user comprehension study assessed how well users understood the report generated by the algorithm, but the genotyping itself was assessed standalone.

7. Type of Ground Truth Used

• Analytical Studies: Sanger bidirectional sequencing was used as the ground truth for confirming genotypes in the reproducibility, LoD, and comparison studies.

• Clinical Studies (User Comprehension): The ground truth for the comprehension study was defined by pre-defined correct answers to questions about the information presented in the reports. This is a measure of objective understanding of the provided text, not directly related to a medical diagnosis or biological outcome.

8. Sample Size for the Training Set

• Training Set Sample Size: The document does not provide information regarding a specific "training set" sample size for the genotyping algorithm itself.
  • Genotyping microarrays from companies like Illumina (which 23andMe uses/adapts) are developed and validated using extensive reference genomes and known variant databases during their initial design and manufacturing. 23andMe's proprietary software analyzes the raw data, and while such software might undergo internal validation and refinement, the specific training data for the core genotyping calls (determining AG, GG, etc.) is not detailed in this regulatory summary. The focus is on the analytical validation of the test performed by 23andMe using their specific chip and processes.

9. How the Ground Truth for the Training Set Was Established

• Ground Truth for Training Set: As no specific "training set" for the genotyping algorithm's core variant calling function is detailed, the method for establishing its ground truth is not provided in this document.
  • However, for the specific pharmacogenetic associations and predicted metabolizer phenotypes (e.g., "poor metabolizer"), the document states: "The impact of protein or enzyme function for each allele and the predicted metabolizer phenotypes were identified from data in the literature for each allele for each gene." This is ground truth established through expert consensus and published literature.

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) uses qualitative genotyping to detect select clinically relevant variants in genomic DNA isolated from human saliva collected from individuals ≥18 years with the Oragene Dx model OGD500.001 for the purpose of reporting and interpreting genetic health risks, including the 23andMe PGS Genetic Health Risk Report for BRCA1/BRCA2 (Selected Variants). The 23andMe PGS Genetic Health Risk Report for BRCA1/BRCA2 (Selected Variants) is indicated for reporting of the 185delAG and 5382insC variants in the BRCA1 gene and the 6174delT variant in the BRCA2 gene. The report describes if a woman is at increased risk of developing breast and ovarian cancer, and if a man is at increased risk of developing breast cancer or may be at increased risk of developing prostate cancer. The three variants included in this report are most common in people of Ashkenazi Jewish descent and do not represent the majority of the BRCA1/BRCA2 variants in the general population. The test report does not describe a person's overall risk of developing any type of cancer, and the absence of a variant tested does not rule out the presence of other variants that may be cancer-related. This test is not a substitute for visits to a healthcare provider for recommended screenings or appropriate follow-up and should not be used to determine any treatments.

The 23andMe PGS is a non-invasive DNA testing service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies. It is a direct-to-consumer, over-the-counter, DNA genetic test intended to provide information and tools for individual users.

A user's saliva is self-collected using the Oragene Dx device manufactured by DNA Genotek. Inc. (previously cleared under K141410), which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to one of two Clinical Laboratory Improvement Amendments (CLIA)-certified laboratories for testing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping beadchip, reagents and instrumentation manufactured by Illumina. The multiplex assay simultaneously tests for more than 500,000 variants, including those for the previously authorized indications, as well as for the indication proposed herein.

The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe (the Manufacturer). The data are analyzed using the Manufacturer's proprietary Coregen software, and a genotype is determined for each tested variant. The results for certain of these variants, as noted in the indications for use, are used to generate personalized reports for users that provide information about the diseases associated with tested variants.

Personalized reports are generated for each user that provides results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on the risk of disease associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the risks associated with the presence of a particular variant. The reports are designed to help users understand the meaning of their results and inform conversations with their doctor or other healthcare professional. The 23andMe PGS Genetic Health Risk Report for BRCA1/BRCA2 (Selected Variants) reports on three specific variants including the 185de1AG and 5382insC variants in the BRCA1 gene and the 6174delT variant in the BRCA2 gene. The variants included in this report are most common in people of Ashkenazi Jewish descent and do not represent the majority of BRCA2 variants in the general population. Therefore the absence of a variant tested does not rule out the presence of other genetic variants that may be disease-related.

Here's a breakdown of the acceptance criteria and the study that proves the device meets them, based on the provided text:

1. A table of acceptance criteria and the reported device performance

The document specifies acceptance criteria for analytical performance studies, particularly for accuracy.

• Site 1: 0.41% FQC for Homozygous Common, 2.53% FQC for Heterozygous.
• Site 2: 3.85% FQC for Homozygous Common, 8.00% FQC for Heterozygous.
  BRCA1 5382insC (i400378):
• Site 1: 0.41% FQC for Homozygous Common, 1.25% FQC for Heterozygous.
• Site 2: 3.85% FQC for Homozygous Common, 3.18% FQC for Heterozygous.
  BRCA2 6174delT (i400379):
• Site 1: 0.41% FQC for Homozygous Common, 0% FQC for Heterozygous.
• Site 2: 3.84% FQC for Homozygous Common, 2.53% FQC for Heterozygous.
  All valid calls were 100% correct across all conditions. |
  | Detection Limit (LoD) | At least 95% correct calls at the lowest concentration tested (5 ng/uL). | 100% correct calls per genotype for all samples across all reagent lots, at all sample concentrations tested (including 5 ng/uL). |
  | Accuracy (Comparison with Sanger Bidirectional Sequencing) | Minimum of 99% positive percent agreement (PPA) and negative percent agreement (NPA) for each genotype. Uncertainty of the point estimate within an acceptable range, presented using 95% confidence interval. | BRCA1 185delAG:
• Homozygous Common: 100% PPA, 100% NPA (95% CI: 96.6 - 100)
• Heterozygous: 100% PPA, 100% NPA (95% CI: 93.8 - 100)
  BRCA1 5382insC (referring to the first instance of this variant in Table 4 for combined entries):
• Homozygous Common: 100% PPA, 100% NPA (95% CI: 94.0 - 100)
• Heterozygous: 100% PPA, 100% NPA (95% CI: 83.9 - 100)
  BRCA2 6174delT (referring to the combined entries for this variant in Table 4):
• Homozygous Common: 100% PPA, 100% NPA (95% CI: 93.9 - 100)
• Heterozygous: 100% PPA, 100% NPA (95% CI: 92.1 - 100)
  All reported agreements are 100% for correct calls, with some samples failing QC ("No Call" or "FQC") which are accounted for separately. The CIs met the acceptance criteria. |
  | User Comprehension | Minimum of a 90% or greater overall comprehension rate for each comprehension concept, with justification from physician/genetic counselor identifying relevant concepts. | Specific user comprehension studies were not performed for this specific BRCA1/BRCA2 report. Instead, the document refers to studies from DEN160026. The general acceptance criteria for user comprehension studies from the special controls are listed (90% comprehension rate). |

2. Sample sizes used for the test set and the data provenance

• Accuracy Study (Test Set):

  • Sample Size:
    • BRCA1 185delAG Homozygous Common: 109 samples
    • BRCA1 185delAG Heterozygous: 58 samples
    • BRCA1 5382insC Homozygous Common: 60 samples
    • BRCA1 5382insC Heterozygous: 22 samples
    • BRCA2 6174delT Homozygous Common (presumably the combined 59 samples from the table as the 6174delT is mistakenly written as 5382insC for its second entry in Table 4): 60 samples (including 1 FQC)
    • BRCA2 6174delT Heterozygous (presumably the combined 45 samples from the table as the 6174delT is mistakenly written as 5382insC for its second entry in Table 4): 46 samples (including 1 FQC)
  • Data Provenance: Saliva samples were selected from the 23andMe customer biobank. The document doesn't explicitly state the country of origin, but "customer biobank" implies they are from their existing customer base, likely primarily US. The study appears to be retrospective as samples were "selected... based on predetermined genotypes."

• Precision/Reproducibility Study:

  • Sample Size:
    • BRCA1 185delAG: 2 wildtype samples, 1 heterozygous sample. Each replicated multiple times (e.g., 242-321 replicates per site for specific genotypes, across multiple runs).
    • BRCA1 5382insC: 2 wildtype samples, 1 heterozygous sample. Each replicated multiple times (e.g., 321-402 replicates per site for specific genotypes).
    • BRCA2 6174delT: 2 wildtype samples, 1 heterozygous sample. Each replicated multiple times (e.g., 313-323 replicates per site for specific genotypes).
  • Data Provenance: DNA samples were procured and genotyped. No specific country of origin mentioned, likely internal to 23andMe's operations or commercial vendors.

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts

The ground truth for the analytical validation (accuracy study) was established by Sanger bidirectional sequencing performed at an independent laboratory site. There is no mention of human experts defining the ground truth for the analytical studies. The "experts" in the context of user comprehension studies (which were not specifically performed for this report but referenced generally) would be physicians and genetic counselors to determine relevant concepts, but not for establishing a genetic variant ground truth.

4. Adjudication method (e.g. 2+1, 3+1, none) for the test set

No adjudication method for the genetic variant determination is mentioned. The ground truth was established directly by Sanger bidirectional sequencing. This is a direct comparison study, not a human reader study requiring adjudication.

5. If a multi reader multi case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

No such MRMC comparative effectiveness study was done. This device is a qualitative genetic test for detecting specific SNPs, not an imaging AI diagnostic aid for human readers.

6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done

Yes, the analytical performance studies (precision, LoD, and accuracy against Sanger sequencing) represent the standalone performance of the 23andMe PGS test, which is an algorithm interpreting genotyping data. The human element is in the initial sample acquisition and laboratory processing, but the "device" itself (the genotyping and data analysis system) operates in a standalone manner for assigning genotypes.

7. The type of ground truth used (expert consensus, pathology, outcomes data, etc)

The primary ground truth for validating the device's accuracy in identifying genetic variants was Sanger bidirectional sequencing. For analytical studies, this is considered a highly reliable and established gold standard for DNA sequencing.

8. The sample size for the training set

The document describes performance studies on a "test set" (accuracy, precision, LoD), but does not provide details on the training set size for the underlying genotyping and analysis algorithms (e.g., Illumina's GenomeStudio or 23andMe's Coregen software). Genetic tests like this rely on established biochemical and computational methods for genotype calling, which are developed and validated using extensive, but typically not detailed in regulatory submissions in terms of a specific "training set" size in the same way an AI model's training set would be. The focus is on the analytical validation of the test system's performance for the specific variants.

9. How the ground truth for the training set was established

As there's no explicit mention of a "training set" in the context of a machine learning model, the concept of ground truth establishment for it isn't directly addressed. For the underlying genotyping technology and software, ground truth is typically based on known genetic sequences, synthetic constructs, or samples previously characterized by highly accurate sequencing methods like Sanger.

Ask a specific question about this device

The 23andMe Personal Genome Service (PGS) Test uses qualitative genotyping to detect the following clinically relevant variants in genomic DNA isolated from human saliva collected from individuals ≥18 years with the Oragene Dx model OGD-500.001 for the purpose of reporting and interpreting Genetic Health Risks (GHR):

The 23andMe PGS Genetic Health Risk Report for Hereditary Thrombophilia is indicated for reporting of the Factor V Leiden variant in the F5 gene, and the Prothrombin G20210A variant in the F2 gene. This report describes if a person has variants associated with a higher risk of developing harmful blood clots, but it does not describe a person's overall risk of developing harmful blood clots. This test is most relevant for people of European descent.

The 23andMe PGS Genetic Health Risk Report for Alpha-1 Antitrypsin Deficiency is indicated for reporting of the PIZ and PIS variants in the SERPINA1 gene. This report describes if a person has variants associated with AAT deficiency and a higher risk for lung or liver disease, but it does not describe a person's overall risk of developing lung or liver disease. This test is most relevant for people of European descent.

The 23andMe PGS Genetic Health Risk Report for Late-onset Alzheimer's Disease is indicated for reporting of the £4 variant in the APOE gene. The report describes if a person's genetic result is associated with an increased risk of developing Late-onset Alzheimer's Disease, but it does not describe a person's overall risk of developing Alzheimer's Disease. The £4 variant included in this report is found and has been studied in many ethnicities. Detailed risk estimates have been studied the most in people of European descent.

The 23andMe PGS Genetic Health Risk Report for Parkinson's Disease is indicated for reporting of the G2019S variant in the LRRK2 gene and the N370S variant in the GBA gene. The report describes if a person's genetic result is associated with an increased risk of developing Parkinson's disease, but it does not describe a person's overall risk of developing Parkinson's disease. The test is most relevant for people of European, Ashkenazi Jewish, and North African Berber descent.

The 23andMe PGS Genetic Health Risk Report for Gaucher Disease Type 1 is indicated for reporting of the N370S, 84GG, and V394L variants in the GBA gene. This report describes if a person has variants associated with an increased risk for developing symptoms of Gaucher Disease Type 1, but it does not describe a person's overall risk of developing Gaucher Disease Type 1. This test is most relevant for people of Ashkenazi Jewish descent.

The 23andMe PGS Genetic Health Risk Report for Factor XI Deficiency is indicated for reporting of the variants F283L. E117X. IVS14+1G>A in the F11 gene. This report describes if a person has a variant associated with Factor XI deficiency and the potential for a higher risk of excessive bleeding following trauma or surgery, but it does not describe a person's overall risk for excessive bleeding. This test is most relevant for people of Ashkenazi Jewish descent.

The 23andMe PGS Genetic Health Risk Report for Celiac Disease is indicated for reporting of a variant associated with the HLA-DQ2.5 haplotype. The report describes if a person has a haplotype associated with an increased risk of developing celiac disease, but it does not describe a person's overall risk for developing celiac disease. This report is most relevant for people of European descent.

The 23andMe PGS Genetic Health Risk Report for Glucose-6-Phosphate-Dehydrogenase Deficiency is indicated for reporting of the Val68Met variant in the G6PD gene. This report describes if a person has a variant associated with G6PD deficiency and a higher risk for episodes of anemia, but it does not describe a person's overall risk of developing anemia. This test is most relevant for people of African descent.

The 23andMe PGS Genetic Health Risk Report for Hereditary Hemochromatosis is indicated for reporting of the C282Y and H63D variants in the HFE gene. This report describes if a person has variants associated with hereditary hemochromatosis and a higher risk for iron overload, but it does not describe a person's overall risk of developing iron overload. This report is most relevant for people of European descent.

The 23andMe PGS Genetic Health Risk Report for Early-Onset Primary Dystonia (DYT1/TOR1A-Related) is indicated for reporting of the deltaE302/303 variant in the DYT1 gene. This report describes if a person has variants associated with a higher risk for early-onset primary dystonia, but it does not describe a person's overall risk of developing dystonia. This report is most relevant for people of Ashkenazi Jewish descent.

The 23andMe PGS is a currently-marketed, non-invasive genetic information service that combines qualitative genotyping data covering genetic ancestry, traits, and certain heritable health conditions from a single multiplex assay with descriptive information derived from peer reviewed, published genetic research studies. It is a direct-toconsumer, over-the-counter, non-diagnostic, DNA genetic testing service intended to provide information and tools for individual users.

A user's saliva is self-collected using the Oragene-Dx device manufactured by DNA Genotek, Inc. (previously cleared under K141410), which consists of a sealable collection tube containing a stabilizing buffer solution. Once the sample is collected, it is shipped to one of two Clinical Laboratory Improvement Amendments (CLIA)-certified laboratories for processing.

DNA is isolated from the saliva and tested in a multiplex assay using a customized genotyping chip and instrumentation manufactured by Illumina. The multiplex assay simultaneously tests for more than 500,000 variants, including those for the indications proposed herein.

The raw data is generated using Illumina GenomeStudio software, and then sent to 23andMe (the Manufacturer). The data are analyzed using the Manufacturer's proprietary Coregen software, and a genotype is determined for each tested variant. The results for certain of these variants are used to generate personalized reports for users that provide information about the diseases associated with the detected variant.

Personalized reports are generated for each user that provide results of the testing performed. These reports tell the user which variant(s) has/have been detected in their sample and provide information on the risk of disease associated with the variant(s). If no variant was detected, that information is also provided. The personalized reports are designed to present scientific concepts to users in an easy-to-understand format. The reports provide scientifically valid information about the risks associated with the presence of a particular variant. The reports are designed to help users understand the meaning of their results and any appropriate actions that may be taken based on their results.

Acceptance Criteria and Device Performance Study for 23andMe Personal Genome Service (PGS) Genetic Health Risk Test

The 23andMe Personal Genome Service (PGS) Genetic Health Risk Test is a qualitative in vitro molecular diagnostic system for detecting genetic variants associated with various diseases. The regulatory approval details the acceptance criteria and performance study results.

1. Table of Acceptance Criteria and Reported Device Performance

The acceptance criteria for analytical performance (Precision/Reproducibility, Limit of Detection, and Accuracy) are derived from the Special Controls outlined in 21 CFR 866.5950, specifically paragraph (b)(3)(iii)(J).

• PA (CC|CC) = 100% (95% CI: 94.7% to 100%)
• PA (CT|CT) = 100% (95% CI: 94.7% to 100%)
• PA (TT|TT) = 100% (95% CI: 94.6% to 100%)
  Hereditary Thrombophilia (Prothrombin G20210A F2):
• PA (GG|GG) = 100% (95% CI: 94.7% to 100%)
• PA (AG|AG) = 100% (95% CI: 94.6% to 100%)
• PA (AA|AA) = 100% (95% CI: 94.5% to 100%)
  Alpha-1 Antitrypsin Deficiency (PI*Z SERPINA1):
• PA (CC|CC) = 100% (95% CI: 94.8% to 100%)
• PA (CT|CT) = 100% (95% CI: 94.7% to 100%)
• PA (TT|TT) = 100% (95% CI: 94.7% to 100%)
  Alpha-1 Antitrypsin Deficiency (PI*S SERPINA1):
• PA (TT|TT) = 100% (95% CI: 94.4% to 100%)
• PA (AT|AT) = 100% (95% CI: 94.7% to 100%)
• PA (AA|AA) = 100% (95% CI: 94.7% to 100%)
  Late-onset Alzheimer's Disease (APOE s4):
• PA (TT|TT) = 100% (95% CI: 98.8% to 100%)
• PA (CT|CT) = 99.2% (95% CI: 95.7% to 99.9%)
• PA (CC|CC) = 100% (95% CI: 96.3% to 100%)
  Parkinson's Disease (G2019S LRRK2):
• PA (GG|GG) = 100% (95% CI: 86.7% to 100%)
• PA (AG|AG) = 100% (95% CI: 93% to 100%)
• PA (AA|AA) = 100% (95% CI: 51.0% to 100%)
  Parkinson's Disease (N370S GBA):
• PA (TT|TT) = 100% (95% CI: 86.7% to 100%)
• PA (CT|CT) = 100% (95% CI: 94% to 100%)
• PA (CC|CC) = 100% (95% CI: 87.1% to 100%)
  Gaucher Disease Type 1 (N370S GBA):
• PA (TT|TT) = 100% (95% CI: 86.7% to 100%)
• PA (CT|CT) = 100% (95% CI: 94% to 100%)
• PA (CC|CC) = 100% (95% CI: 87.1% to 100%)
  The Manufacturer affirmed that comparison studies for other variants (84GG, V394L in GBA; F283L, E117X, IVS14+1G>A in Factor XI; HLA-DQA1 for Celiac Disease; Val68Met in G6PD; C282Y, H63D in HFE; deltaE302/303 in DYT1) were completed with identical study design, acceptance criteria, and results as the reported variants, including 100% accuracy for all valid calls. |
  | Precision/Reproducibility | The percentage of samples that failed quality control must be indicated (i.e., the total number of sample replicates for which a sequence variant cannot be called (no calls) or that fail sequencing quality control criteria divided by the total number of replicates tested). (b)(3)(iii)(J)(2) | General: All analytical performance studies met the pre-determined acceptance criteria.
  Factor V Leiden: 98.0% correct calls, 2.01% FQCs (0.04% anticipated 2 FQCs).
  Prothrombin G20210A: 98.1% correct calls, 1.85% FQCs (0.03% anticipated 2 FQCs).
  PI*Z SERPINA1: 98.4% correct calls, 1.63% FQCs (0.03% anticipated 2 FQCs).
  PI*S SERPINA1: 98.2% correct calls, 1.76% FQCs (0.03% anticipated 2 FQCs).
  APOE s4: 98.5% correct calls, 1.54% FQCs (0.02% anticipated 2 FQCs).
  G2019S LRRK2: 98.5% correct calls, 1.54% FQCs (0.02% anticipated 2 FQCs).
  N370S GBA (Parkinson's): 96.2% correct calls, 3.81% FQCs (0.15% anticipated 2 FQCs).
  N370S GBA (Gaucher): 97.5% correct calls, 3.12% FQCs (0.10% anticipated 2 FQCs).
  All valid calls in these studies were 100% correct at both sites. The Manufacturer affirmed that reproducibility studies were completed for all other listed variants (84GG, V394L in GBA; F283L, E117X, IVS14+1G>A in Factor XI; HLA-DQA1 for Celiac Disease; Val68Met in G6PD; C282Y, H63D in HFE; deltaE302/303 in DYT1) with identical study design, acceptance criteria, and results. |
  | Limit of Detection (LoD) | Data must be provided demonstrating the minimum amount of DNA that will enable the test to perform correctly in 95 percent of runs. (b)(3)(iii)(J)(5) | The LoD study yielded 100% correct genotype calls for all samples and all reagent lots tested at sample DNA concentrations of 5, 15, and 50 ng/uL. The study passed LoD acceptance criteria at 5 ng/uL. Manufacturer claimed LoD of 15 ng/uL.
  The Manufacturer affirmed that LoD studies were completed for all other listed variants (84GG, V394L in GBA; F283L, E117X, IVS14+1G>A in Factor XI; HLA-DQA1 for Celiac Disease; Val68Met in G6PD; C282Y, H63D in HFE; deltaE302/303 in DYT1) with identical study design, acceptance criteria, and results. |
  | User Comprehension | The user study must meet predefined primary endpoint criteria, including a minimum of a 90 percent or greater overall comprehension rate (i.e., selection of the correct answer) for each comprehension concept. (b)(3)(iii)(M)(4)(vi) | The overall comprehension score for all concepts across all test reports studied was greater than 90%. For the G6PD Deficiency report scenario ("1 Variant Detected"), the report-specific comprehension score was 73.3%, which the Manufacturer mitigated with planned labeling changes in the FAQ section. |

2. Sample Sizes and Data Provenance for Test Set

Test Set for Analytical Accuracy (Comparison to Sanger Bidirectional Sequencing):

• Hereditary Thrombophilia (Factor V Leiden): 68 Homozygous Common (CC), 68 Heterozygous (CT), 67 Homozygous Rare (TT) samples. Total = 203.
• Hereditary Thrombophilia (Prothrombin G20210A): 68 Homozygous Common (GG), 67 Heterozygous (AG), 66 Homozygous Rare (AA) samples. Total = 201.
• Alpha-1 Antitrypsin Deficiency (PI*Z SERPINA1): 70 Homozygous Common (CC), 68 Heterozygous (CT), 69 Homozygous Rare (TT) samples. Total = 207.
• Alpha-1 Antitrypsin Deficiency (PI*S SERPINA1): 65 Homozygous Common (TT), 68 Heterozygous (AT), 69 Homozygous Rare (AA) samples. Total = 202.
• Late-onset Alzheimer's Disease (APOE s4): 316 Homozygous Common (TT), 126 Heterozygous (CT), 101 Homozygous Rare (CC) samples with correct calls. (Total raw samples: 320 TT, 129 CT, 106 CC, including those with "no calls" or "invalid"). Total = 555.
• Parkinson's Disease (G2019S LRRK2): 25 Homozygous Common (GG), 51 Heterozygous (AG), 4 Homozygous Rare (AA) samples. Total = 80.
• Parkinson's Disease (N370S GBA): 25 Homozygous Common (TT), 60 Heterozygous (CT), 26 Homozygous Rare (CC) samples. Total = 111.
• Gaucher Disease Type 1 (N370S GBA): 25 Homozygous Common (TT), 60 Heterozygous (CT), 26 Homozygous Rare (CC) samples. Total = 111.
• Other variants: For Gaucher Disease Type 1 (84GG, V394L), Factor XI Deficiency (F283L, E117X, IVS14+1G>A), Celiac Disease (HLA-DQA1), G6PD Deficiency (Val68Met), Hereditary Hemochromatosis (C282Y, H63D), and Early-Onset Primary Dystonia (deltaE302/303), the manufacturer affirmed that comparison studies were completed with the same design, acceptance criteria, and results as the explicitly detailed studies, testing wild-type, heterozygous, and homozygous samples where applicable, adhering to minimum sample size requirements (e.g., at least 20 unique samples for common genotypes, 3 for rare heterozygous, 3-20 for rare homozygous based on frequency).

Data Provenance: Saliva samples were selected from the 23andMe customer biobank based on their predetermined genotype. The data origin is not explicitly stated geographically, but it is indicated that genotype frequencies are considered for "individuals of European descent," "Ashkenazi Jewish descent," "North African Berber descent," "African descent," "South Asian," and "East Asian" depending on the variant, implying a diverse customer base. The samples are retrospective, as they were chosen from an existing biobank.

3. Number of Experts and Qualifications for Ground Truth

No external experts were explicitly mentioned for establishing the ground truth of the genetic variants in the test set. The ground truth for the test set was established by Sanger bidirectional sequencing, which is considered the "gold standard" for genetic variant determination. This method is an intrinsic, highly accurate laboratory technique, rather than requiring expert consensus interpretation.

4. Adjudication Method for the Test Set

Not applicable. The ground truth was established by Sanger bidirectional sequencing, an objective laboratory method, which means there was no human adjudication process involved for the genetic variant calls.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

No MRMC comparative effectiveness study was performed in the context of human readers improving with or without AI assistance. This device is a direct-to-consumer genetic test for detecting specific variants, not an AI-assisted diagnostic imaging or interpretation tool.

6. Standalone (Algorithm-Only) Performance Study

Yes, the analytical performance studies (Accuracy, Precision/Reproducibility, Limit of Detection) represent the standalone performance of the 23andMe PGS device. The device's genotyping results were compared directly against the gold standard (Sanger bidirectional sequencing) without human intervention in the interpretation of the device's output for genetic calls.

7. Type of Ground Truth Used

The type of ground truth used for the analytical performance studies (accuracy, LoD) was Sanger bidirectional sequencing, which is a highly accurate molecular biology laboratory method for DNA sequencing, serving as the "truth" for genetic variants. This is an objective laboratory measure, not expert consensus, pathology, or outcomes data.

8. Sample Size for the Training Set

The document does not explicitly mention a "training set" in the context of machine learning or AI algorithm development for genetic variant detection. The methodologies described are for analytical validation of a qualitative genetic test. The robustness of the test is demonstrated through extensive reproducibility and accuracy studies using defined sets of samples (both cell lines and customer biobank samples).

For the analytical performance studies (Precision/Reproducibility and Limit of Detection), a range of sample sizes were used:

• Precision/Reproducibility for Hereditary Thrombophilia (Factor V Leiden): 10 DNA samples (3 wild type, 2 heterozygous, 1 homozygous rare, repeated multiple times) used for 1890 replicates.
• Precision/Reproducibility for Hereditary Thrombophilia (Prothrombin G20210A): 10 DNA samples (3 wild type, 1 heterozygous, 1 homozygous rare, repeated multiple times) used for 1620 replicates.
• Precision/Reproducibility for Alpha-1 Antitrypsin Deficiency (PI*Z SERPINA1): 8 DNA samples (2 wild type, 1 heterozygous, 1 homozygous rare, repeated multiple times) used for 1350 replicates.
• Similar numbers of DNA samples and replicates were used for other variants in the reproducibility studies.
• Limit of Detection: Samples were diluted to three different DNA concentrations (5, 15, and 50 ng/uL) and genotyped. The text does not specify the number of individual donor samples used for LoD beyond "each sample was diluted," implying the same initial set of samples used for accuracy/precision.

9. How the Ground Truth for the Training Set Was Established

As mentioned above, a "training set" in the conventional machine learning sense for algorithm development is not explicitly described. The ground truth for the samples used in the analytical validation studies (which could be considered analogous to a "development/test set" for a traditional assay) was established through bidirectional Sanger sequencing, performed by an independent laboratory. This external, highly accurate, and objective method served as the reference standard for confirming the genotypes of the samples used in the performance evaluations.

Ask a specific question about this device