📄 medicine

Updated Testing Sensitivities and Variants of Uncertain Significance in CanRisk

This paper introduces updated default testing sensitivities and variant handling protocols in the CanRisk platform, derived from large population studies, which result in lower predicted cancer risks for individuals with negative genetic test results while cautioning against their use for counseling those with potentially pathogenic variants of uncertain significance.

Original authors: Douglas Frederick Easton, Tim Carver, Lorenzo Ficorella, Helen Hanson, Katie Snape, Olga Tsoulaki, Juliet A Usher-Smith, Stephanie Archer, Marc Tischkowitz, Antonis Antoniou

Published 2026-07-23

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Douglas Frederick Easton, Tim Carver, Lorenzo Ficorella, Helen Hanson, Katie Snape, Olga Tsoulaki, Juliet A Usher-Smith, Stephanie Archer, Marc Tischkowitz, Antonis Antoniou

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the weather for a specific town. You have a super-smart computer model that looks at history, the current wind, and the clouds to tell you if you need an umbrella. But there's a catch: the model relies on how well your weather station can see the clouds. If your station has a blind spot and misses a storm cloud, the model might think the sky is clear when it's actually pouring. This is exactly the kind of puzzle scientists face when trying to predict a person's risk of getting cancer.

In the world of genetics, scientists use powerful tools to look at a person's DNA—their biological instruction manual—to find "typos" called pathogenic variants. These typos can make it more likely for someone to develop breast, ovarian, or prostate cancer. One of the most famous tools for this is called CanRisk. It acts like a crystal ball, calculating a person's future risk based on their family history and their genetic test results. But for this crystal ball to work, it has to know one crucial thing: how good is the test at finding the typos? If the test misses a typo, the tool might give a false sense of security. The big question isn't just "Do you have a typo?" but "If the test says 'no typos found,' how sure can we really be that there aren't any hiding in the shadows?"

This paper is about sharpening the focus of that crystal ball. The authors, a team of geneticists and computer scientists, realized that the settings in the CanRisk tool were a bit outdated. They decided to update the "sensitivity" settings—essentially, the tool's ability to say, "I have looked everywhere, and I am confident there are no dangerous typos here." By using new data from large studies and talking to real-world lab experts, they tweaked the math to reflect how modern genetic testing actually works today.

Here is the story of their update, told with a little more color.

The Old Map vs. The New GPS

Think of the old version of the CanRisk tool as an explorer using a map from the 1950s. The map was good, but it had some foggy areas. Specifically, the tool assumed that if a genetic test came back "negative" (meaning no bad typos were found), there was still a decent chance that a dangerous typo was hiding, just out of sight.

Why did the tool think this? It was worried about two things:

The "Uncertain" Clouds: Sometimes, the test finds a weird letter in the DNA, but scientists aren't sure if it's a dangerous typo or just a harmless quirk. These are called Variants of Uncertain Significance (VUS). In the old system, if a test found one of these "uncertain" clouds, the tool treated the whole test as "negative" but kept the risk high, just in case that cloud was actually a storm.
The "Hidden" Rearrangement: The tool also worried that the test might miss huge chunks of DNA that had been shuffled around (large rearrangements), especially in the famous BRCA1 and BRCA2 genes.

The authors suggest that this old way of thinking was too cautious. They argue that in modern labs, the technology is so advanced that it's like upgrading from a 1950s map to a high-definition GPS. Today's tests are incredibly good at spotting the "uncertain" clouds. If a lab says "negative," they usually mean, "We looked, and we didn't find any dangerous typos, and we didn't even find any potentially pathogenic uncertain variants that might be dangerous."

The Big Update: Raising the Bar

The team proposed a new set of rules for the tool. Instead of assuming the test might miss a lot of things, they updated the tool to assume that modern tests are very, very good. They set the "technical sensitivity" (how good the machine is at reading the DNA) to 99% for small typos and 95% for the big shuffled chunks.

They also made a crucial change in how they interpret a "negative" result. In the past, a negative result meant "No Pathogenic Variant found." Now, the tool treats a negative result as "No Pathogenic Variant AND No Potentially Pathogenic Uncertain Variant found."

This might sound like a tiny tweak, but it's like changing the rules of a game. If you tell the tool, "We are sure there are no potentially pathogenic uncertain variants," the tool stops worrying about the "what ifs."

The Result: A Clearer Sky

When the authors ran the numbers with these new settings, the results were dramatic. For a person who gets a negative test result, the tool now says their risk is much lower than it used to.

Let's look at a specific example from the paper. Imagine a 40-year-old woman with a family history of breast and ovarian cancer.

Before the update: If she took a genetic test and it came back negative, the tool said there was still about a 6.5% chance she was carrying a hidden dangerous typo. Her lifetime risk of ovarian cancer was estimated at 3.1%.
After the update: With the new, sharper settings, that chance of a hidden typo drops to just 0.34%. Her estimated lifetime risk of ovarian cancer drops to 0.3%.

The authors explain that this happens because the new settings trust the test more. If the test says "nothing found," and the test is 99% good at finding things, then the odds that something is hiding are tiny. The "fog" has lifted.

What the Tool Can't Do Yet

However, the authors are careful not to say this solves everything. They point out a few important limits.

First, the new settings only work if the test result is truly "clean." If a person does have a suspicious VUS that might be pathogenic, the tool currently does not make any risk predictions. It's like a weather app that says, "I can't tell you if it will rain because there's a weird cloud I don't understand." The authors admit they need to figure out how to handle these "suspicious clouds" in the future, but for now, the tool stays silent on them.

Second, the tool still treats all dangerous typos in a gene as having the same level of risk. It doesn't yet know that some typos might be slightly less dangerous than others, or that some might be super-dangerous. The authors say they plan to add this nuance later, but for now, the tool keeps things simple.

Why This Matters

The main takeaway is that the CanRisk tool is now more in sync with how real doctors and labs work today. By updating the settings to reflect that modern tests are highly accurate, the tool gives people who test negative a clearer, more reassuring picture of their future. It reduces the anxiety of "what if I missed something?" for the vast majority of people.

But the authors also warn that this clarity comes with a condition: you have to trust that the test was done correctly and that no potentially pathogenic uncertain variants were found. If there are any suspicious VUS that might be pathogenic left in the DNA, the tool needs more information before it can make a prediction. It's a step forward, making the genetic crystal ball a bit clearer, but the scientists are still working on the fine-tuning to make it perfect.

Technical Summary: Updated Testing Sensitivities and Variants of Uncertain Significance in CanRisk

Problem Statement
The CanRisk platform utilizes the BOADICEA and CanRisk-Prostate algorithms to predict breast, ovarian, and prostate cancer risks based on multiple factors, including germline pathogenic variants (PVs) in nine susceptibility genes (ATM, BARD1, BRIP1, BRCA1, BRCA2, CHEK2, RAD51C, RAD51D, PALB2). A critical limitation in previous iterations of the model was the handling of genetic testing sensitivities and Variants of Uncertain Significance (VUS).

The existing framework treated a "negative" test result as the absence of a detected PV, but it did not distinguish between a test that missed a PV due to technical failure and a test that detected a variant but classified it as a VUS. Consequently, the overall sensitivity of testing was modeled as the product of technical sensitivity and the probability that a PV is not misclassified as a VUS. This approach resulted in lower effective sensitivities, particularly for BRCA1 and BRCA2, leading to higher residual carrier probabilities and inflated cancer risk estimates for individuals with negative test results, even when no potentially pathogenic VUS were present.

Methodology
The authors propose a revised framework for defining testing sensitivities that aligns with current clinical practices and large population study data. The methodology involves:

Reclassification of Negative Results: The authors argue that in current clinical practice, a "negative" test result implies the absence of both pathogenic variants and potentially pathogenic VUS. Therefore, the model should calculate the residual carrier probability based on the observation "No PV or potentially pathogenic VUS" rather than just "No PV."
Mathematical Adjustment:
- The original overall sensitivity ( $s$ ) was defined as $s = (1 - c)(1 - v)s_T$ , where $s_T$ is technical sensitivity, $v$ is the probability a variant is detected but not recognized as a PV (VUS), and $c$ is the probability of a "cryptic" variant not targeted by the test.
- The revised approach removes the $v$ term from the sensitivity calculation for negative results. The effective sensitivity becomes $(1 - c)s_T$ . This assumes that if a potentially pathogenic VUS were present, it would be reported, and the test would not be considered "negative" in the context of risk calculation.
Updated Technical Assumptions:
- Protocol Uniformity: A single testing protocol is assumed for all nine genes, covering single nucleotide variants (SNVs), indels, and large rearrangements.
- Technical Sensitivity Values: Based on feedback from UK Genomics Laboratories, the authors assume a technical sensitivity of 0.99 for SNVs and indels, and 0.95 for large rearrangements across all genes.
- Cryptic Variants: The assumption of "no cryptic PVs" ( $c=0$ ) is maintained, with the authors noting that any undetected variants are effectively absorbed into the polygenic risk component of the model.

Key Contributions

Revised Default Sensitivities: The paper provides updated default sensitivity values for all nine genes, which are now significantly higher than previous defaults due to the exclusion of the VUS misclassification factor from the "negative" test scenario.
Implementation in CanRisk: These new sensitivities have been incorporated into the CanRisk platform (version v.4.1 onwards) as the default settings.
Clarification of VUS Handling: The update clarifies that the tool does not currently generate risk predictions for individuals with a "negative" test result if a potentially pathogenic VUS is present. The documentation has been updated to reflect that a negative result implies the absence of both PVs and suspicious VUS.

Results
The application of revised sensitivities leads to a marked reduction in predicted cancer risks for individuals who test negative:

Residual Carrier Probability: In a case study of a 40-year-old woman with a family history of breast and ovarian cancer, the residual PV carrier probability after a negative test dropped from ~6.5% (using current sensitivities) to ~0.34% (using revised sensitivities).
Risk Estimates: Corresponding cancer risk estimates decreased significantly. For the same case study, the lifetime ovarian cancer risk was reduced from 3.1% (current sensitivities) to 0.3% (revised sensitivities) following a negative test.
Gene-Specific Impact: While the most significant reduction is observed in BRCA1 and BRCA2 due to higher sensitivity to large rearrangements and SNVs, all other genes show reduced residual carrier probabilities due to the inclusion of large rearrangement detection in the assumed protocol.

Significance
The authors state that these updated settings better reflect current clinical usage and the interpretation of "negative test" results in genetic counseling. By increasing the effective sensitivity of testing, the model reduces the residual risk estimates for individuals with negative results, potentially alleviating unnecessary anxiety and clinical uncertainty. However, the paper emphasizes a necessary caveat: the tool is not designed to provide risk predictions for individuals carrying potentially pathogenic VUS, as the probability of such variants being truly pathogenic is not yet fully integrated into the model. The authors note that future updates aim to address atypical risk variants and the specific handling of VUS probabilities.