Is this the holy GRAIL for cancer detection?
GRAIL is rolling out Galleri, their multi-cancer early detection test
Last month a study funded by GRAIL as part of the Circulating Cell-free Genome Atlas (CCGA) reported validation results for GRAIL’s Galleri test. Galleri is a multi-cancer early detection test, a long sought after blood test that can reportedly screen for over 50 cancer types from a single blood draw.
Galleri works by detecting circulating tumour DNA (ctDNA) in the blood, a byproduct of the rapid turnover of tumour cells. Galleri as a product is now live and available in the USA for a list price of $949.
For a screening test such as Galleri, two metrics tell us almost everything we need to know about the performance of the test: specificity (true negative rate) and sensitivity (true positive rate). In the validation study, the authors screened 4077 people and reported sensitivity and specificity for each of the cancer types that Galleri can screen for.
Let’s start with the good news, specificity is high across all cancer types and the mean specificity across all of them is 99.5%. Specificity is a measure for the true negative rate, that is: given that you do not have cancer, how likely is the test to return a negative result? This measure is important for a screening method as a low specificity leads to a higher number of false positives. False positives in cancer screening can lead to unnecessary follow-up scans and biopsies which are costly for the healthcare system and harmful for the patient.
Interpreting the sensitivity of Galleri is more complicated. Sensitivity tells us the test’s ability to find cancer. In aggregate across all cancer types and stages, Galleri has a sensitivity of 51.5%, meaning that roughly half of all cancers are missed. On the surface, that doesn’t sound great, although given the lack of weapons in the multi-cancer detection arsenal, this is an acceptable level of sensitivity.
Breaking down the sensitivity by cancer type, it is clear that not all cancers are equal when screened for by Galleri:
Eleven cancer types including ovary, pancreas and colon cancer have an excellent specificity above 80%. Notably, Galleri also achieved a high sensitivity for lung cancer (74.8%), particularly impactful as this is the leading cause of death amongst all cancers. On the other end of the scale, Galleri performs poorly for some cancers, notably prostate and breast cancer, which both have large sample sizes.
Segmenting by cancer type does not tell the full story. Galleri also performs differently depending on the cancer stage. As a rule of thumb in cancer detection, earlier is better. Detecting cancer at an earlier stage leads to a higher chance of survival. The table below, truncated from the study, shows the sensitivity segmented by cancer stage (Roman numerals one to four, I-IV):
To interpret the sensitivity results, it’s good to revisit how a ctDNA-based screening test works. When cells within a tumour die, either by natural turnover or through other means (immune system targeting or treatment), fragments of the cancer cell’s genome are released into the bloodstream and circulate. There are many factors that can affect the amount of ctDNA found in blood, one of these factors is the size (volume) of the originating tumour. Larger tumours will generally contain more cells and hence have a higher turnover in number of cells.
Returning to the results table, Galleri achieves lower sensitivity for early stage cancer (stage I or II), where we expect the tumours to be smaller in volume and hence we also expect these patients to have a lower amount of ctDNA in their blood. In other words, although disappointing, these results are to be expected! For later stage cancer (stage III or IV), Galleri achieves a sensitivity of approximately 70-90%.
GRAIL are very keen to push Galleri as early cancer detection. However, the analysis shows that sensitivity for early stage cancer (stage I) is poor, only detecting 16.8% of these cancers. Plainly, Galleri is not great at early stage cancer detection. On the other hand, since sensitivity for late stage cancer is excellent, it can be argued that Galleri will help to (sensitively) find late stage cancer earlier. Without a doubt, ctDNA-based cancer screening will save lives, but this first version of Galleri is not the magic bullet for detecting cancer when it is easiest to treat.
Galleri is certainly a leap forward in screening, there are no other screening methods for many of the cancer types covered. However, the low sensitivity for early stage cancers casts a shadow on the results. Further trials and longer follow-up periods are required to understand whether widespread use of Galleri can result in an improvement in the most important metric of all: mortality.
Nice piece! Two nits:
"Specificity is a measure for the true negative rate, that is: given that the test result returns negative, how likely are you to be truly cancer-free?"
This is a common mistake - you've actually given the definition of a different measure here, "negative predictive value" (NPV). Specificity is different: given that you don't have the disease, how often do you get a negative result.
In other words:
Specificity = P( negative test result | no disease )
NPV = P( no disease | negative test result )
"Eleven cancer types including ovary, pancreas and colon cancer have an excellent specificity above 80%. Notably, Galleri also achieved a high specificity for lung cancer (74.8%), particularly impactful as this is the leading cause of death amongst all cancers. On the other end of the scale, Galleri performs poorly for some cancers, notably prostate and breast cancer, which both have large sample sizes."
In this paragraph, where you wrote "specificity", I think you meant "sensitivity".