I work with this data regularly (in fact am probably a coauthor on at least one of the cited publications). For the vast majority of people who do not have specific contraindications, an occasional diagnostic scan adds imperceptible cancer risk. The data are extremely noisy and the amount of radiation is very small.
That said, older radiation exposures (say before 1960) were much more likely to be unshielded and cause extremely high doses, which *did* have a measurable effect.
Also just wanted to say that I really enjoy your writing and think that you should have a wider audience.
Wow, what a coincidence! I wonder if some of the public impressions regarding these things are a hangover from earlier scans. Personally, I was pretty surprised to see these numbers—I had the uninformed impression that the risks were orders of magnitude higher. Always nice to have a new thing to not worry about!
This isn't meant to be criticism, just things I thought of while reading your post.
1. If I lay on the beach for 2 minutes in strong sunlight, I won't get a sunburn. If I stay for 25 minutes, I will get a sunburn. As far as I know, we don't know if ionizing radiation has a similar threshold effect (on the low end).
2. I thought this was going to be a post about confounding factors. The naive way to study this would be to look at the number of CT scans a person gets and how long they live. There is an obvious confounding factor - why did did they get a CT scan in the first place? Humans are made of confounding factors.
3. Many people die WITH cancer and not OF cancer. Cancer can end life, and we tend to concentrate on that at the population level. It's relatively easy to compile statistics about cause of death. Treatments may delay death, but it cannot prevent it. Humans die. There will always be a leading cause of death.
Metrics and incentives are hard. Along with the number of people who die, we could measure "years of life added by treatment". Otherwise, a clueless genie would just shoot cancer patients in the head to reduce "deaths due to cancer" when we rub on that lamp.
4. Cancer also destroys quality of life. That is much harder to measure at scale. So some people, including health care professionals may choose to stop treatment. Metrics and incentives are hard. So we could add "quality years of life added by treatment".
Re #4: I would definitely rather do this calculation with DALYs or QALYs rather than years of life lost. I wasn't able to find any paper that estimated the average number of DALYs that would be saved if all cancer were cured. However, it just occurred to me that we could get a rough estimate: This paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8719276/) looks at the total years of life (YLL) and DALYs lost due to cancer. The DALYs were overwhelmingly due to lost life expectancy: In low-income countries it was 98.3% (17.7/18.0), in middle-income countries it was 97.6% (74.5/76.3) and in high-income countries it was 94.5% (48.1/50.9). So, I think we can be pretty confident that the calculation wouldn't change at least for DALYs.
> "The DALYs were overwhelmingly due to lost life expectancy"
I'm trying to parse this.
1. If a person starts treatment for cancer at 65, has two more 'good years' (67), then survives for one more bad year, but is bedridden and then passes on (68), then they would have 67 DALYs and 68 total years, for a ratio of 98.5. (I misunderstood the terms so this is wrong, see comment below)
2. If a child began treatment for leukemia at age 4, and then survived to age 88 with no loss of ability, then the ratio would be 100. (I think this example is still right)
In case one, I'm more interested in the ratio of post detection quality of life, ie 2 good years to 3 years total = 67%
In case two, I'm more interested in life expectancy with and without treatment. A child of 4 may live another 70 years if treated. An octogenarian will not live another 70 years with treatment.
In other words, I think we should default to aggressively treat cancer in kids, and we should let adults make their own informed decision about treatment.
Generally speaking parents, doctors and society want kids to survive and it is worth the brief pain and trauma if the prognosis has even a moderate chance of success. A 20% success rate is worth it for treating childhood cancer (.2*70=14 average added years), but a 20% success rate may not be worth it if you are 80 - If you are 80 you should make an informed choice.
If you are 80, your life expectancy is under 10 years ( https://pubmed.ncbi.nlm.nih.gov/7565998/ ) - a 20% survival rate implies an average of 2 years of life added, but on average MOST of those added years will have low quality of life.
2. Years of life lived with disability. (The weights depend on the disability but, e.g., a year lived with schizophrenia count as around 0.5 DALY)
The statement "The DALYs were overwhelmingly due to lost life expectancy" means that the vast majority of costs come from years of life lost rather than from years lived with disability.
In this example:
> 1. If a person starts treatment for cancer at 65, has two more 'good years' (67), then survives for one more bad year, but is bedridden and then passes on (68), then they would have 67 DALYs and 68 YLL, for a ratio of 98.5.
This isn't the right way to count DALY or YLL. In order to do that we would need to know when the person would have died without cancer. If you imagine that they would have lived without disability until age 70 without cancer then in your example
YLL = 2
DALY = 2 + 0.5 = 2.5 (assuming a disability weighting of 0.5)
Your table doesn't match the calculation in the text -- you said that it was 1.16 hours per mSv, but the table lists coronary angiogram as 1 mSv but 4 hours of life expectancy cost. What's the deal?
(Especially curious since I once had a very long heart-related fluoroscopy procedure.)
OK, I looked into this, and it looks like I just completely dropped the factor of 2.66 years in my calculations, meaning ALL the numbers were off by a factor of 2.66 🤦. After seeing that, I went and triple-checked the rest of the numbers. Everything else looked OK except coronary angiography, which I mistakenly wrote down as 1 mSv rather than the correct 10 mSv.
I think this is fixed now—all the numbers will be larger by a factor of 2.66 (up to unit rounding) except for coronary angiography which is larger by a factor of 26.6.
Thanks! That's very interesting... with those numbers, I actually wonder if the heart procedure I had was in fact worth it (it was to fix a low-risk condition).
Wish I'd had more info like this when I was making the decision about it. Thanks for posting.
If you were interested in going deeper on the per organ risk, there are ICRP standard weighting factors for each organ that can be used to calculate the “effective dose” in millisieverts for the whole body.
Interesting idea. Actually doing the calculation is probably above my pay-grade, but do you have any reference? I had trouble finding a recent and reliable one with the factors. I'm interested to see how much it varies between different parts of the body. (I also have a suspicion that this calculation is so tricky that if you want to go beyond "Fermi golf" levels of accuracy you need to be reeaaaallly careful.)
Actually, if I recall correctly, the calculation is simple enough that it was covered in an undergraduate class of mine for an aerospace engineering degree. It's the actual determination of the weighting factors themselves which is difficult, but that is left to the ICRP.
The equation for calculating effective dose is simply the sum(Wt*Ht), where Wt is the dimensionless tissue weighting factor for all tissues t and H is the equivalent dose absorbed by each tissue. The equivalent dose is similarly calculated as sum(Wr*Dr), where Dr is the absorbed dose of each kind of radiation and Wr the weighting factor for said radiation type (e.g. photon, electron, alpha particle, etc.).
The most recent values for tissue weighting factor can be found in the linked document, under ICRP 103.
This website also seems to have all of the relevant calculations explained at least as well, if not in as much detail, as Pisacane's "The Space Environment and its Effect on Space Systems," which I referenced.
It also highlights the important point that these values are for stochastic effects. Above a certain radiation threshold, you're not looking at long-term risk, but immediate mortality. This data is not intended to include that type of risk.
One last addendum: because the units of absorbed dose are in Gray, which has SI units of Joules per kilogram, there are some calculations you might like to perform which are slightly nonintuitive. That is, you cannot simply say that a person exposed to 0.5 Gy of proton radiation across their entire body--which creates a 1 Sievert equivalent dose (the proton radiation weighting factor is 2) and 1 Sievert effective dose (over the whole body is 1 by definition)--would have 1 Gy of radiation were the same amount of energy focused on their lungs. 1 Gy of radiation to the lungs is 0.12 Sv (the tissue weighting factor for lungs is 0.12), but the same amount of energy to the lungs, divided by the weight of the lungs (which I here assume to be 1 kg after a quick Google, and the average body to be 71 kg) is actually 71 Gy, which gives an effective dose of 8.52 Sv. If this seemed trivial I apologize--it greatly confused me when I learned it. Hopefully it also served as an example.
I'm not sure this gets you much beyond "Fermi Golf" levels of accuracy, but engineers, businesses, and mission planners also don't do much more than this when it comes to calculating safety for astronauts and earthbound radiation workers, so perhaps it's good enough.
Isn't there was lots of evidence that linear model of radiation damage is wrong at low-dosage levels?
For example, living at high altitudes is not associated with more cancer in general but high altitudes get more radiation than low. Similarly, places with more naturally occurring radiation in the soil don't show more cancers in the population. In fact often they show less.
I work with this data regularly (in fact am probably a coauthor on at least one of the cited publications). For the vast majority of people who do not have specific contraindications, an occasional diagnostic scan adds imperceptible cancer risk. The data are extremely noisy and the amount of radiation is very small.
That said, older radiation exposures (say before 1960) were much more likely to be unshielded and cause extremely high doses, which *did* have a measurable effect.
Also just wanted to say that I really enjoy your writing and think that you should have a wider audience.
Wow, what a coincidence! I wonder if some of the public impressions regarding these things are a hangover from earlier scans. Personally, I was pretty surprised to see these numbers—I had the uninformed impression that the risks were orders of magnitude higher. Always nice to have a new thing to not worry about!
This isn't meant to be criticism, just things I thought of while reading your post.
1. If I lay on the beach for 2 minutes in strong sunlight, I won't get a sunburn. If I stay for 25 minutes, I will get a sunburn. As far as I know, we don't know if ionizing radiation has a similar threshold effect (on the low end).
2. I thought this was going to be a post about confounding factors. The naive way to study this would be to look at the number of CT scans a person gets and how long they live. There is an obvious confounding factor - why did did they get a CT scan in the first place? Humans are made of confounding factors.
3. Many people die WITH cancer and not OF cancer. Cancer can end life, and we tend to concentrate on that at the population level. It's relatively easy to compile statistics about cause of death. Treatments may delay death, but it cannot prevent it. Humans die. There will always be a leading cause of death.
Metrics and incentives are hard. Along with the number of people who die, we could measure "years of life added by treatment". Otherwise, a clueless genie would just shoot cancer patients in the head to reduce "deaths due to cancer" when we rub on that lamp.
4. Cancer also destroys quality of life. That is much harder to measure at scale. So some people, including health care professionals may choose to stop treatment. Metrics and incentives are hard. So we could add "quality years of life added by treatment".
Re #4: I would definitely rather do this calculation with DALYs or QALYs rather than years of life lost. I wasn't able to find any paper that estimated the average number of DALYs that would be saved if all cancer were cured. However, it just occurred to me that we could get a rough estimate: This paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8719276/) looks at the total years of life (YLL) and DALYs lost due to cancer. The DALYs were overwhelmingly due to lost life expectancy: In low-income countries it was 98.3% (17.7/18.0), in middle-income countries it was 97.6% (74.5/76.3) and in high-income countries it was 94.5% (48.1/50.9). So, I think we can be pretty confident that the calculation wouldn't change at least for DALYs.
> "The DALYs were overwhelmingly due to lost life expectancy"
I'm trying to parse this.
1. If a person starts treatment for cancer at 65, has two more 'good years' (67), then survives for one more bad year, but is bedridden and then passes on (68), then they would have 67 DALYs and 68 total years, for a ratio of 98.5. (I misunderstood the terms so this is wrong, see comment below)
2. If a child began treatment for leukemia at age 4, and then survived to age 88 with no loss of ability, then the ratio would be 100. (I think this example is still right)
In case one, I'm more interested in the ratio of post detection quality of life, ie 2 good years to 3 years total = 67%
In case two, I'm more interested in life expectancy with and without treatment. A child of 4 may live another 70 years if treated. An octogenarian will not live another 70 years with treatment.
In other words, I think we should default to aggressively treat cancer in kids, and we should let adults make their own informed decision about treatment.
Generally speaking parents, doctors and society want kids to survive and it is worth the brief pain and trauma if the prognosis has even a moderate chance of success. A 20% success rate is worth it for treating childhood cancer (.2*70=14 average added years), but a 20% success rate may not be worth it if you are 80 - If you are 80 you should make an informed choice.
If you are 80, your life expectancy is under 10 years ( https://pubmed.ncbi.nlm.nih.gov/7565998/ ) - a 20% survival rate implies an average of 2 years of life added, but on average MOST of those added years will have low quality of life.
In general DALYs count two things:
1. Years of life lost. (1 year = 1 DALY)
2. Years of life lived with disability. (The weights depend on the disability but, e.g., a year lived with schizophrenia count as around 0.5 DALY)
The statement "The DALYs were overwhelmingly due to lost life expectancy" means that the vast majority of costs come from years of life lost rather than from years lived with disability.
In this example:
> 1. If a person starts treatment for cancer at 65, has two more 'good years' (67), then survives for one more bad year, but is bedridden and then passes on (68), then they would have 67 DALYs and 68 YLL, for a ratio of 98.5.
This isn't the right way to count DALY or YLL. In order to do that we would need to know when the person would have died without cancer. If you imagine that they would have lived without disability until age 70 without cancer then in your example
YLL = 2
DALY = 2 + 0.5 = 2.5 (assuming a disability weighting of 0.5)
So the ratio would be 2/2.5 = 80%.
Ah, I got YLL backwards. I'll edit my post
Your table doesn't match the calculation in the text -- you said that it was 1.16 hours per mSv, but the table lists coronary angiogram as 1 mSv but 4 hours of life expectancy cost. What's the deal?
(Especially curious since I once had a very long heart-related fluoroscopy procedure.)
Eeek, yes! Let me look into that!
OK, I looked into this, and it looks like I just completely dropped the factor of 2.66 years in my calculations, meaning ALL the numbers were off by a factor of 2.66 🤦. After seeing that, I went and triple-checked the rest of the numbers. Everything else looked OK except coronary angiography, which I mistakenly wrote down as 1 mSv rather than the correct 10 mSv.
I think this is fixed now—all the numbers will be larger by a factor of 2.66 (up to unit rounding) except for coronary angiography which is larger by a factor of 26.6.
Thanks! That's very interesting... with those numbers, I actually wonder if the heart procedure I had was in fact worth it (it was to fix a low-risk condition).
Wish I'd had more info like this when I was making the decision about it. Thanks for posting.
If you were interested in going deeper on the per organ risk, there are ICRP standard weighting factors for each organ that can be used to calculate the “effective dose” in millisieverts for the whole body.
Interesting idea. Actually doing the calculation is probably above my pay-grade, but do you have any reference? I had trouble finding a recent and reliable one with the factors. I'm interested to see how much it varies between different parts of the body. (I also have a suspicion that this calculation is so tricky that if you want to go beyond "Fermi golf" levels of accuracy you need to be reeaaaallly careful.)
Actually, if I recall correctly, the calculation is simple enough that it was covered in an undergraduate class of mine for an aerospace engineering degree. It's the actual determination of the weighting factors themselves which is difficult, but that is left to the ICRP.
The equation for calculating effective dose is simply the sum(Wt*Ht), where Wt is the dimensionless tissue weighting factor for all tissues t and H is the equivalent dose absorbed by each tissue. The equivalent dose is similarly calculated as sum(Wr*Dr), where Dr is the absorbed dose of each kind of radiation and Wr the weighting factor for said radiation type (e.g. photon, electron, alpha particle, etc.).
The most recent values for tissue weighting factor can be found in the linked document, under ICRP 103.
https://www.ncbi.nlm.nih.gov/books/NBK158810/table/T50/
This website also seems to have all of the relevant calculations explained at least as well, if not in as much detail, as Pisacane's "The Space Environment and its Effect on Space Systems," which I referenced.
https://www.nuclear-power.com/nuclear-engineering/radiation-protection/equivalent-dose/radiation-weighting-factor/
It also highlights the important point that these values are for stochastic effects. Above a certain radiation threshold, you're not looking at long-term risk, but immediate mortality. This data is not intended to include that type of risk.
One last addendum: because the units of absorbed dose are in Gray, which has SI units of Joules per kilogram, there are some calculations you might like to perform which are slightly nonintuitive. That is, you cannot simply say that a person exposed to 0.5 Gy of proton radiation across their entire body--which creates a 1 Sievert equivalent dose (the proton radiation weighting factor is 2) and 1 Sievert effective dose (over the whole body is 1 by definition)--would have 1 Gy of radiation were the same amount of energy focused on their lungs. 1 Gy of radiation to the lungs is 0.12 Sv (the tissue weighting factor for lungs is 0.12), but the same amount of energy to the lungs, divided by the weight of the lungs (which I here assume to be 1 kg after a quick Google, and the average body to be 71 kg) is actually 71 Gy, which gives an effective dose of 8.52 Sv. If this seemed trivial I apologize--it greatly confused me when I learned it. Hopefully it also served as an example.
I'm not sure this gets you much beyond "Fermi Golf" levels of accuracy, but engineers, businesses, and mission planners also don't do much more than this when it comes to calculating safety for astronauts and earthbound radiation workers, so perhaps it's good enough.
I wanted to add the obligatory XKCD link. (average yearly dose ~4mSv)
https://xkcd.com/radiation/
Isn't there was lots of evidence that linear model of radiation damage is wrong at low-dosage levels?
For example, living at high altitudes is not associated with more cancer in general but high altitudes get more radiation than low. Similarly, places with more naturally occurring radiation in the soil don't show more cancers in the population. In fact often they show less.