How Accurate is 23andMe?
If you are wondering how accurate 23andMe is before you buy or after you’ve gotten your results, there are definitely some things you need to know. DNA testing kits are not as simple as a pregnancy test (or any other test looking for a single substance) and they can be much more complex than other types of DNA tests (such as a paternity test).
Considering how many traits that a test kit covers, it’s also important to understand the difference in accuracy and predictability between the BRCA breast cancer test and the variants in the genes that cause red hair. It’s one thing if your 23andMe test predicts red hair and you don’t have it, it’s quite another thing to assume that you will be cancer-free just because your 23andMe test said you don’t carry the BRCA gene mutations. If you really want to understand how accurate a 23andMe test is, you need to understand how these tests work, what they are actually measuring, and how their algorithms predict your traits.
In this article, the editors here at Genomelink have broken down everything you need to know about the different aspects of a 23andMe test and DNA testing in general that will help you better understand how accurate they are. Whether you are curious about your eye color, genetic health risks, an ancestry test, or other genetic information, you should definitely understand the following information. (You should also note that most of the information below applies to all genetic testing companies and consumer genetic tests!)
There are Different Levels of Accuracy
The first thing to know is that the word “accuracy” can describe many different aspects of a typical DNA test that measures single-nucleotide polymorphisms (SNPs). Each SNP is a specific location in your DNA that can have 1 of 4 different variants. An SNP test measures around 700,000 of these variants across all of your chromosomes. Combined, your chromosomes contain 3 billion nucleotides. So, a 23andMe genotyping is only testing around .023% of the human genome.
The good news is that the specific variants a 23andMe test looks at have already been correlated to various diseases and traits and many have been approved by the Food and Drug Administration (FDA). So, measuring these SNPs can be very predictive of actual traits and diseases. For example, genetic diseases (such as Cystic Fibrosis) are caused by mutations in the DNA and you can find these types of mutations in your Carrier Status reports. The bad news is that genomics science is incredibly complex and we don’t fully understand the effect of SNPs that we are not measuring. With that in mind, let’s look at the different levels of accuracy in a 23andMe test.
Sample Data Accuracy in DNA Sequencing
When you spit in a tube, you are giving a DNA sample to 23andMe. At the laboratory, lab technicians extract this DNA and run it through a machine that searches for each of the 700,000 SNPs that 23andMe is looking at. At this level, results from this test are 99%+ accurate. In other words, if 23andMe says you carry a variant on a specific chromosome, you likely do.
Since you can download this raw data from 23andMe, you can easily have this information analyzed by a wide variety of other sites to gain more valuable insights into your data.
Statistical Analysis in DNA Testing
At this level and above is where DNA testing services start to get complicated. The “statistical analysis” is where data scientists use large amounts of SNP data to correlate certain conditions and diseases with specific SNPs that you carry. Though we won’t get into the complex statistical methods here, scientists use Genome-Wide Association Studies (GWAS) and also Polygenic Risk Score (PGS) to measure a trait in many different people and then try to see which genetic variants are also found in people who carry the trait.
So, even though this level is complex and you would need a Ph.D. in Genetics to fully understand it - you should rest assured that the methodology behind this sequencing analysis is sound. If you are simply using 23andMe to learn more about genetic markers, find family members, or get ancestry reports, the accuracy and predictability of genetic tests are more than enough to be very helpful.
DNA Data Volume
While GWAS can be very accurate and predictive if the dataset is large enough, there is a limitation. Since scientists are only measuring around 700,000 SNPs, there is an enormous amount of “the unknown” that is not being measured. For instance, scientists are not measuring the whole genome or measuring how the genome interacts with the environment.
In other words, while some SNP tests can predict certain diseases and produce great ancestry reports, the best advice to avoid most diseases is still a well-balanced diet and exercise. While statistical methods can prove beyond a doubt that two things are correlated, genetic tests alone are not yet advanced enough to prove that most genes actually cause a certain disease. They may increase your risk, but the actual cause of most diseases and traits is most likely complex interactions between your DNA and your lifestyle. See the section below on health results for more information on this topic.
Limitation on the Reference Dataset
The accuracy of any prediction is limited to the data used to make the prediction. The predictions made by a DNA test are no different. While the accuracy of the SNP measured may be near-perfect, if you compare the SNP you carry to a skewed set of data you can get some very inaccurate results.
This is best explained with an ancestry DNA testing example. If a small company with a small DNA database were to measure your DNA and provide you with an ethnicity estimate, it may come back and contradict everything you know about your family tree. For example, if there are only Irish and British people in the reference population a company uses to compare your DNA with, they will inevitably provide you with an ethnicity estimate that includes high percentages of Irish and British. However, this can also be true of health traits and other traits. If a company only correlates a disease to a reference population made of Scandinavian people, the test results may not correlate to a global population.
The good news is that 23andMe has the second-largest DNA database in the world (behind Ancestry.com) with over 10 million users. This also means that they have one of the best-developed set of reference populations on the market. So, it is very unlikely that any of their traits will suffer from a bias in a reference population. If a 23andMe test tells you that you carry an increased risk for Alzheimer’s disease in your health test (or any other health condition), you should likely see a doctor or genetic counselor for more information.
Operator Error in DNA Tests
While very rare, it should be no surprise that sometimes lab technicians and statisticians screw up. So technically, human error can be present in almost any step of the DNA testing process. However, 23andMe typically processes samples multiple times and does a Quality Review of portions of your data before it is uploaded to your account. This almost always ensures that any sources of human error (such as mixing your sample with another person) are caught before your data is released.
Health Results are Accurate, But That Doesn’t Mean What You Think!
As far as the editors here at Genomelink can tell, most users that claim “inaccuracies” in their 23andMe data have simply failed to grasp what their results are actually telling them. Let’s start with a simple physical trait to see how this can work.
A user with red hair receives the result that they are “not likely to have red hair.” So, the user claims that their results are inaccurate and disavows the rest of the results. However, 23andMe never told the user that they could predict red hair with 100% accuracy. The company monitors many different variants in the genome that can cause red hair, but not all of them. This user simply has a very rare mutation that has not been correlated to red hair - yet, at least.
Using similar logic, a user may get health reports back that says they do not carry a rare BRCA mutation that has been correlated to higher rates of breast cancer. Confident this result is as good as a diagnosis, the user foregoes annual breast cancer testing and normal health examinations. What this user fails to realize is that the BRCA mutations being measured only account for about 5% of all breast cancers and are mostly present in the Ashkenazi Jewish population. Though the user does not carry a known mutation, they still carry essentially the same predisposition as the general population of developing a non-BRCA form of breast cancer.
This user should have read the fine print that 23andMe must include with most of their health-related and wellness tests to receive FDA approval. While a positive test can be a good reason to see a doctor or geneticist, a negative result does not mean you are in the clear. In other words, a 23andMe test is a relatively cheap way to look for a few known variants that can affect your health. It is not a diagnostic test, and it cannot replace regular doctor visits, a family history of health, a good diet, or regular exercise.
TL; DR: 23andMe tests are extremely accurate and in most cases predictive, but you must read the fine print. A 23andMe test is not a diagnostic test and should not be used as an excuse to forego normal healthcare checkups or engage in a major lifestyle change without consulting a doctor.