Summary: Good news!! If you read this post last year, about 50,000 fewer people are dying each year than you thought.
I’m very sceptical of the WHO estimate of 81,000-138,000 annual fatalities. Following the citations:
That WHO stat (citation 6 in the report linked in footnote 2, note that link automatically downloads a pdf) cites Gutierrez et al. 2017.
Gutierrez et al. have a range of 81,410 to 137,880. This range is not a confidence interval, but “combined upper estimates of mortality” (page 2) — meaning these numbers are each upper bounds. Gutierrez et al. have two citations for the range, Kasturiratne et al. 2008 and Chippaux 1998.
Kasturiratne et al.’s “high estimate” is 93,945 deaths per year. They come to this number by averaging rural and urban incident/fatality rates in a way which I don’t agree with,[1] and applying regional fatality rates to areas without data.[2]
In Chippaux, the largest number I can find is 125,345. Despite the specificity, this paper relies heavily on round numbers.[3] Also, the paper is 26 years old, and the number of people employed in agriculture (top risk for snakebites) has fallen by more than a third.
I have no idea where the 137,880 number in Gutierrez et al. comes from, and therefore where the WHO got that figure, and therefore where the number in this post comes from. It’s widely cited to the WHO report, but I lose track of it before Gutierrez et al.
What numbers seem more reasonable? From the 16-year-old Kasturiratne et al. paper, I find their low estimate (19,886 annual deaths) too low, so I’d accept their range of 19,886 to 93,945 as a acceptable confidence interval (90% for the year 2007). The Chippaux numbers are much too high, and the higher Gutierrez et al. number is entirely unsourced. Further, as noted above, the decline in agricultural labor over the past two decades will have reduced exposure to snakes, and the rapid expansion of medical care in rural India will have further addressed the problem.
Thankfully, the 2019 GBD report[^5] did a much better job (imo) of estimating this. Their value estimate for all deaths from venomous animals is 79,700, and venomous snakes is 63,400.[4] This seems much more reasonable, though GBD figures do have a history of being too high, and I personally revise my median estimate down to 55,000.
[1] They assume that all rural areas have the highest incident and fatality rate, which seems obviously untrue. But to create an upper bound when dealing with lots of missing data, fine.
[2] “Our approach minimised this effect [of ignoring regions with missing data], although in some regions, such as the Caribbean, lack of data still meant that we were forced to use very high rates in our calculations of the high estimate.” (page 1601)
[3] 80% of this high estimate relies on this paragraph:
“In Asia (population, ca. 3500 million) as a whole there may be up to 4 million snake-bites each year, of which almost 50% are envenomed. Approximately half of the victims reach hospital and the annual number of deaths resulting can be estimated at 100000.”
The 100,000 number is not justified any further.
[5] I used the query tool here; you need to make a free account.
[4] This leads to the question, what other other venomous animals are killing 16,000 people per year? I couldn’t find an easy answer; my guesses in order are scorpions, jellyfish, and spiders. I highly recommend Venomous by Christie Wilcox for a cultural and scientific history of venom.
These are the comments I come to the EA Forum for!
It wouldn’t be the first time I’ve seen the WHO reports on snakebites get something blatantly wrong. If I recall correctly, they accidentally flipped a figure for percentage of people who couldn’t afford treatment in the Bangladesh study.
Did you look into the Indian one million death’s study? As far as I remember, that was the study which led WHO to revise their previous global estimate of 50k, as it found 50k annual deaths to snakebites in India alone.
If you have, I’d be curious to get your take on how much weight to put on it. My understanding is that if not for this study, WHO’s estimate would be around what you arrived at.
Great wee analysis here—in future this kind of thing might be better posted on your “quicktakes”, or maybe there as well because otherwise it could well get lost quickly on the forum and few might see it.
Summary: Good news!! If you read this post last year, about 50,000 fewer people are dying each year than you thought.
I’m very sceptical of the WHO estimate of 81,000-138,000 annual fatalities. Following the citations:
That WHO stat (citation 6 in the report linked in footnote 2, note that link automatically downloads a pdf) cites Gutierrez et al. 2017.
Gutierrez et al. have a range of 81,410 to 137,880. This range is not a confidence interval, but “combined upper estimates of mortality” (page 2) — meaning these numbers are each upper bounds. Gutierrez et al. have two citations for the range, Kasturiratne et al. 2008 and Chippaux 1998.
Kasturiratne et al.’s “high estimate” is 93,945 deaths per year. They come to this number by averaging rural and urban incident/fatality rates in a way which I don’t agree with,[1] and applying regional fatality rates to areas without data.[2]
In Chippaux, the largest number I can find is 125,345. Despite the specificity, this paper relies heavily on round numbers.[3] Also, the paper is 26 years old, and the number of people employed in agriculture (top risk for snakebites) has fallen by more than a third.
I have no idea where the 137,880 number in Gutierrez et al. comes from, and therefore where the WHO got that figure, and therefore where the number in this post comes from. It’s widely cited to the WHO report, but I lose track of it before Gutierrez et al.
What numbers seem more reasonable? From the 16-year-old Kasturiratne et al. paper, I find their low estimate (19,886 annual deaths) too low, so I’d accept their range of 19,886 to 93,945 as a acceptable confidence interval (90% for the year 2007). The Chippaux numbers are much too high, and the higher Gutierrez et al. number is entirely unsourced. Further, as noted above, the decline in agricultural labor over the past two decades will have reduced exposure to snakes, and the rapid expansion of medical care in rural India will have further addressed the problem.
Thankfully, the 2019 GBD report[^5] did a much better job (imo) of estimating this. Their value estimate for all deaths from venomous animals is 79,700, and venomous snakes is 63,400.[4] This seems much more reasonable, though GBD figures do have a history of being too high, and I personally revise my median estimate down to 55,000.
[1] They assume that all rural areas have the highest incident and fatality rate, which seems obviously untrue. But to create an upper bound when dealing with lots of missing data, fine.
[2] “Our approach minimised this effect [of ignoring regions with missing data], although in some regions, such as the Caribbean, lack of data still meant that we were forced to use very high rates in our calculations of the high estimate.” (page 1601)
[3] 80% of this high estimate relies on this paragraph:
“In Asia (population, ca. 3500 million) as a whole there may be up to 4 million snake-bites each year, of which almost 50% are envenomed. Approximately half of the victims reach hospital and the annual number of deaths resulting can be estimated at 100000.”
The 100,000 number is not justified any further.
[5] I used the query tool here; you need to make a free account.
[4] This leads to the question, what other other venomous animals are killing 16,000 people per year? I couldn’t find an easy answer; my guesses in order are scorpions, jellyfish, and spiders. I highly recommend Venomous by Christie Wilcox for a cultural and scientific history of venom.
These are the comments I come to the EA Forum for!
It wouldn’t be the first time I’ve seen the WHO reports on snakebites get something blatantly wrong. If I recall correctly, they accidentally flipped a figure for percentage of people who couldn’t afford treatment in the Bangladesh study.
Did you look into the Indian one million death’s study? As far as I remember, that was the study which led WHO to revise their previous global estimate of 50k, as it found 50k annual deaths to snakebites in India alone.
If you have, I’d be curious to get your take on how much weight to put on it. My understanding is that if not for this study, WHO’s estimate would be around what you arrived at.
Great wee analysis here—in future this kind of thing might be better posted on your “quicktakes”, or maybe there as well because otherwise it could well get lost quickly on the forum and few might see it.