What Is the ClinVar Database and How Is It Used?

ClinVar is a free, publicly accessible database run by the National Center for Biotechnology Information (NCBI) that collects and organizes information about human genetic variants and their relationship to disease. It holds over three million submitted variants from more than 2,800 organizations worldwide, making it one of the most widely consulted resources in clinical genetics.¹ If you or a family member has had genetic testing, ClinVar is almost certainly one of the databases the lab consulted to figure out what your results mean.

What ClinVar Actually Holds

At its core, ClinVar is an archive of interpretations. When a genetic testing lab analyzes a patient’s DNA and finds a variant, the lab decides whether that variant is connected to a disease. ClinVar collects those decisions, along with the evidence behind them, so other labs and clinicians can see what has been reported before. The database covers inherited (germline) variants and variants acquired in tumors (somatic), and it accepts variants of any size, type, or location in the genome.²

Submissions come from a broad range of contributors: clinical testing laboratories, research labs, locus-specific databases, expert panels, and curated resources like OMIM and GeneReviews.³ This means ClinVar is not a single authority delivering one verdict on each variant. It is more like a curated bulletin board where many organizations post their independent assessments, and those assessments sometimes agree and sometimes do not.

The Five-Tier Classification System

Most inherited variants in ClinVar are classified using a standardized five-category system developed jointly by the American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP). Those five categories are: pathogenic, likely pathogenic, uncertain significance, likely benign, and benign.⁴ In plain terms, a “pathogenic” variant is one that is considered to cause disease, while a “benign” variant is considered harmless. The “likely” labels indicate strong but not quite definitive evidence in either direction. “Uncertain significance,” often shortened to VUS, means there is not yet enough evidence to call the variant harmful or harmless.

This classification process draws on multiple types of evidence: how common the variant is in the general population, whether computational tools predict it damages protein function, whether lab experiments have tested its biological effects, and whether the variant tracks with disease in families. No single piece of evidence is usually enough on its own; labs weigh them in combination. The result is a judgment call, and that is exactly why the same variant can end up with different labels from different labs.

Star Ratings and Review Status

Not all entries in ClinVar carry equal weight, and the database uses a star-rating system to signal how thoroughly a variant’s classification has been reviewed. The scale runs from zero stars to four. A zero-star entry might be a submission with no assertion criteria provided and no documentation of the method used to reach the classification. A one-star entry means the submitter provided their interpretation criteria. Two stars indicate that multiple submitters agree on the classification without conflict. Three stars are reserved for variants reviewed by an expert panel, and four stars go to variants backed by a practice guideline.

ClinVar encourages submissions of variants reviewed by expert panels because expert consensus represents a high standard of evidence.² For everyday use, the star rating gives clinicians and genetic counselors a quick way to gauge how much confidence to place in a given entry. A four-star variant backed by practice guidelines carries far more clinical weight than a zero-star entry from a single lab with no explanation of how they reached their conclusion.

How Clinicians and Genetic Counselors Use ClinVar

ClinVar has become the single most consulted resource for genetic counselors doing variant interpretation. In a survey of genetic counselors across multiple specialties, about 73% reported using ClinVar “often,” and roughly 61% said they were “very comfortable” with it, making it the top-ranked tool by both frequency of use and user confidence.⁵ Clinicians who more frequently returned genetic test results to patients tended to use genetics-specific resources like ClinVar and population databases more often than those who returned results less frequently.⁶

In practice, a genetic counselor might pull up a patient’s variant in ClinVar to see how many labs have submitted interpretations, whether those interpretations agree, and what evidence was cited. If a patient received a VUS result from their testing lab, the counselor might check ClinVar periodically to see if new submissions have pushed the variant toward pathogenic or benign. This kind of ongoing surveillance is one of ClinVar’s most valuable features: it is not a static reference book but a living record that changes as evidence accumulates.

When Labs Disagree

Because ClinVar aggregates interpretations from hundreds of independent organizations, conflicting classifications are inevitable. In a study of cancer-gene panel testing at one institution, about 3.8% of variant calls had a discrepancy at the time of the original genetic test report.⁷ A separate study looking at multiplex cancer testing across commercial laboratories found that interpretations differed for about a quarter of all variants, and roughly 11% of participants had a variant where the disagreement spanned the critical boundary between pathogenic or likely pathogenic on one hand and VUS on the other, a gap that could change medical decisions.⁸

These conflicts matter because a variant labeled pathogenic might lead to preventive surgery, intensified cancer screening, or testing of family members, while a VUS typically triggers a wait-and-see approach. The genes most prone to conflicting interpretations in cancer testing include CHEK2, ATM, RAD51C, and PALB2, genes that are clinically important but where evidence is still accumulating.⁸

Identifying discrepancies between laboratories’ assertions in ClinVar is actually one of the most common triggers for genetic counselors to question or investigate a lab result further.⁵ In that sense, ClinVar’s transparency about disagreements is a feature, not a bug. Without a shared database, clinicians would have no easy way to know that another lab across the country reached a different conclusion about the same variant.

How Conflicts Get Resolved

Among variants where labs disagree, a detailed analysis found that only about 78% reached majority consensus on whether the variant was actionable, and just under half reached consensus on pathogenicity. These rates were far lower than the overall concordance rates across ClinVar, meaning the contested variants are a distinct and disproportionately troublesome subset.⁹

To address this, the Clinical Genome Resource (ClinGen), a large NIH-funded initiative, works closely with ClinVar. ClinGen organizes Variant Curation Expert Panels (VCEPs) that bring together specialists in specific genes or diseases to systematically evaluate disputed variants. A pilot study involving three of these expert panels found that for roughly half of conflicting variants, the source of the conflict could be addressed through quick flagging rather than the hours of deep expert review that a full curation would require.¹⁰ Some conflicts turn out to stem from outdated submissions, others from differences in how labs applied the ACMG criteria, and still others from one lab having access to data the other did not. When an expert panel issues a ruling, it overrides lower-level submissions in ClinVar’s display hierarchy, giving the variant a three-star review status.

Reclassification Over Time

Variant classifications are not permanent. As population databases grow, as new functional studies are published, and as more patients are tested, the evidence base shifts, and variants get reclassified. ClinVar tracks these changes, making it possible to look back at how a variant’s classification has evolved. An analysis comparing ClinVar and another major variant database (HGMD) found that variants labeled pathogenic or likely pathogenic in ClinVar are reclassified about six times more often than their counterparts in HGMD, and this higher rate of reclassification has likely contributed to ClinVar having a lower false-positive rate over time.¹¹

A study that reinterpreted common pathogenic variants in ClinVar found that older classifications, submitted before 2014, were significantly more likely to need revision than newer ones (67% versus 23%). Of variants submitted with zero stars and no documented criteria, about 71% should have been downgraded, compared to about 31% for one-star variants that at least provided their interpretation criteria.¹² The pattern is clear: variants submitted in ClinVar’s early years, often by a single lab with limited evidence, have been steadily corrected as more data becomes available.

A concrete example illustrates how this works. Among variants associated with inherited metabolic diseases, researchers found 11 variants that were classified as pathogenic but appeared in genotypes that should have caused detectable disease in healthy individuals from the 1000 Genomes Project. By late 2020, 10 of those 11 had been reclassified to non-pathogenic categories in ClinVar. Many had originally been submitted as pathogenic after being seen in a single affected patient, before the star rating system even existed. As population frequency data and additional submissions accumulated, the early calls were corrected.¹¹

Somatic Variants and Cancer Applications

ClinVar was originally built around inherited variants, but cancer genomics has pushed it to expand. Tumors acquire mutations that are not present in a patient’s germline DNA, and oncologists increasingly need to know whether those somatic mutations affect treatment decisions. ClinVar now supports three distinct types of classification: germline classifications for inherited disease, oncogenicity classifications that assess whether a somatic variant gives cancer cells a growth or survival advantage, and clinical impact classifications that indicate whether a somatic variant affects therapy, diagnosis, or prognosis for a particular cancer type.¹³

The oncogenicity terms follow recommendations from a joint effort by ClinGen, the Cancer Genomics Consortium, and the Variant Interpretation for Cancer Consortium, while the clinical impact terms draw from guidelines by the Association for Molecular Pathology, the American Society of Clinical Oncology, and the College of American Pathologists. When a submission indicates that a variant has therapeutic impact, it also specifies the relevant drug or drug class, which makes ClinVar useful for precision oncology workflows where treatment depends on the specific mutations in a patient’s tumor.¹³

This expansion is relatively new. All classifications submitted before January 29, 2024 were treated as germline by default, even if the submitter had originally indicated a somatic origin. ClinVar has been contacting those submitters to request updates, but the transition means some older somatic data is still filed under the germline category.

Programmatic Access and Machine Learning

ClinVar is not just a website you browse. It is designed for large-scale computational use. Clinical bioinformatics pipelines routinely pull ClinVar data to annotate variants found in sequencing runs, and the database supports this through APIs and bulk data downloads in structured formats like XML.¹⁴ When a sequencing lab processes hundreds of variants from a single patient’s exome, automated tools check each one against ClinVar to flag known pathogenic variants before a human analyst ever looks at the data.

ClinVar has also become a foundational training dataset for machine-learning tools that predict whether a variant is harmful. These tools use ClinVar’s labeled variants as their ground truth: if ClinVar says a variant is pathogenic, the algorithm learns what features pathogenic variants tend to share. Researchers building these models do have to be careful, though. One group developing an AI system for pathogenicity estimation found that because ClinVar data was used to generate the training labels, using ClinVar-derived features during training could inflate accuracy measurements. They solved this by preparing separate models for performance measurement and for generating explanations.¹⁵ This kind of circularity risk is something the field is aware of and actively managing.

The Population Diversity Problem

One of ClinVar’s most significant limitations is not unique to ClinVar, but it shows up clearly there. The population databases used to interpret variants, and the clinical testing that feeds ClinVar, skew heavily toward people of European ancestry. An analysis of gnomAD data at loci with expert-reviewed pathogenic and likely pathogenic variants in ClinVar found that European ancestral populations made up about 56% of observations and nearly 60% of allele counts.¹⁶

This matters in a practical way. Population frequency is one of the key inputs in variant classification: if a variant is common in healthy people, it is unlikely to cause a serious genetic disease. But “common in healthy people” is only useful if you have frequency data from the patient’s ancestral population. For underrepresented groups, frequency data is thinner, which means more variants end up classified as VUS rather than benign, leaving patients and clinicians in limbo. A VUS result can be deeply unsettling for a patient who wants a clear answer, and for clinicians it means fewer actionable findings and more uncertainty. Increasing the diversity of genomic sequencing programs is an active priority in the field, but the gap has not yet been closed.

Privacy and the Ethics of Sharing Variant Data

ClinVar is an open-access database, which raises the question of how patient privacy is handled. The data in ClinVar is variant-level, not patient-level. A submission says “this variant was classified as pathogenic for this condition by this lab,” but it does not include the patient’s name, medical record, or identifiable information. A position statement from a group of international genomics experts argued that the link between a genetic variant and a specific condition is “fundamental biological knowledge, not personal information,” and therefore should not require individual consent to share.¹⁷

The picture gets more complicated when additional case-level detail is involved. Sometimes, interpreting a variant accurately requires knowing more about the patient’s phenotype, family history, or broader genomic context. That kind of information edges closer to identifiable personal data. The same position statement recommended that more detailed case-level or extensive genomic data be shared through controlled-access models, with appropriate consent, while still pushing toward making as much data open and de-identified as possible.¹⁷ In practice, ClinVar threads this needle by keeping submissions at the variant-interpretation level and leaving the richer clinical data to other, more restricted repositories.

Legacy Data and the Zero-Star Problem

If there is one persistent concern about ClinVar, it is the quality of its oldest entries. The database launched in 2013, and many early submissions predate the standardized ACMG/AMP classification guidelines published in 2015. These legacy entries often carry zero stars, meaning the submitter provided no assertion criteria and no documented method. The reinterpretation study mentioned earlier found that about 28% of the variants examined had not been updated since those structured guidelines were proposed, and the zero-star entries were far more likely to need downgrading than entries with at least some documented criteria.¹²

For users of ClinVar, the practical lesson is straightforward: always check the star rating before acting on an entry. A zero-star “pathogenic” classification from 2013 with no supporting criteria is not the same as a three-star classification from an expert panel that applied current guidelines. The database makes this distinction visible, but you have to look for it. Genetic counselors are generally trained to weigh these factors, but patients or non-specialist physicians browsing ClinVar on their own could easily miss the difference, especially if they focus on the classification label and skip the review status.