Read on Twitter
How much did overdose deaths *actually* increase during the pandemic?
The CDC releases rolling 12-month data. But we want to know the true month-to-month death toll.
Today in @AMJPublicHealth, @samirakre and I reverse engineered the CDC stats.
https://bit.ly/3dnbt9P
The CDC releases rolling 12-month data. But we want to know the true month-to-month death toll.
Today in @AMJPublicHealth, @samirakre and I reverse engineered the CDC stats.
https://bit.ly/3dnbt9P
This paper started with a frustration.
In Dec 2020, CDC released an emergency advisory describing 'unprecedented increases in overdose deaths' during the pandemic.
But this report used a 12-month death count beginning June 2019, pre-COVID.
https://emergency.cdc.gov/han/2020/han00438.asp
In Dec 2020, CDC released an emergency advisory describing 'unprecedented increases in overdose deaths' during the pandemic.
But this report used a 12-month death count beginning June 2019, pre-COVID.
https://emergency.cdc.gov/han/2020/han00438.asp
81,000 people died of overdose in the 12 months ending May 2020, a huge increase over the previous record of about 70K.
Countless media reports cited these numbers, often confusing "12 months ending May 2020" for the entire year of 2020.
Countless media reports cited these numbers, often confusing "12 months ending May 2020" for the entire year of 2020.
The true overdose death total for 2020 is likely to be closer to 100K.
Bigger picture: why use 12-month numbers (June 2019-May 2020) when what we really want to know is 'how much did overdose mortality increase DURING the pandemic (March - May 2020)??
Bigger picture: why use 12-month numbers (June 2019-May 2020) when what we really want to know is 'how much did overdose mortality increase DURING the pandemic (March - May 2020)??
Based on previous research with EMS data we knew there was a big spike starting in March 2020 @LeoBeletsky
https://jamanetwork.com/journals/jamapsychiatry/fullarticle/2773768
But 12-month stats would 'water down' this sharp increase in overdose deaths from (March-May) with 9 months of lower, pre-pandemic numbers.
https://jamanetwork.com/journals/jamapsychiatry/fullarticle/2773768
But 12-month stats would 'water down' this sharp increase in overdose deaths from (March-May) with 9 months of lower, pre-pandemic numbers.
@samirakre and I realized that it's possible to recover the original monthly values that the CDC uses to make the rolling 12-month sums.
We developed an algorithm to do this by cross-referencing rolling, provisional 12-month sums with published monthly data through 2019.
We developed an algorithm to do this by cross-referencing rolling, provisional 12-month sums with published monthly data through 2019.
We also create 95% prediction intervals, because the 'final' and 'provisional' CDC data series we use have some minor differences, mainly due to reporting lags.
So we compare both series for the years they overlap (2015-2019) and quantify the average degree of error.
So we compare both series for the years they overlap (2015-2019) and quantify the average degree of error.
We found that over 9,000 people died from drug overdose in May 2020 --making it the deadliest month on record -- a ~60% increase over May 2019.
Each month from March to September 2020 was substantially higher than anything seen in years prior.
Each month from March to September 2020 was substantially higher than anything seen in years prior.
Virtually every state with a large enough population to assess death rates saw a spike in May 2020.
The biggest increases were seen in West Virginia, Kentucky, and Tennessee.
The biggest increases were seen in West Virginia, Kentucky, and Tennessee.
West Virginia saw overdose death rates nearly *triple* in May 2020 compared to May 2019.
However, the 12-month death count showed only a 24% increase over the prior year, putting it in the middle of the pack...
However, the 12-month death count showed only a 24% increase over the prior year, putting it in the middle of the pack...
Nevada had a -4% decrease in the 12-month CDC data ending May 2020.
But our monthly numbers show a +63% increase.
These two sets of data tell drastically different stories.
But our monthly numbers show a +63% increase.
These two sets of data tell drastically different stories.
Overdose data transparency is a political issue.
For example, West Virginia doesn't publish their early overdose death numbers, as some other states do.
They had the biggest pandemic-related spike in May 2020, yet that wasn't public knowledge.
Meanwhile...
For example, West Virginia doesn't publish their early overdose death numbers, as some other states do.
They had the biggest pandemic-related spike in May 2020, yet that wasn't public knowledge.
Meanwhile...
...as overdose deaths spiked, new legislation in WV was being developed to shut down frontline harm reduction organizations that distribute naloxone and overdose prevention materials
See @Narcocast coverage by @ZachWritesStuff @LaurenMPeace https://narcocast.com/episode-54-west-virginia-americas-worst-overdose-and-hiv-hotspot-with-lauren-peace/
See @Narcocast coverage by @ZachWritesStuff @LaurenMPeace https://narcocast.com/episode-54-west-virginia-americas-worst-overdose-and-hiv-hotspot-with-lauren-peace/
Data innovations during the pandemic have highlighted that we can do better.
We have seen how logistical and political barriers were overcome to report COVID deaths in a *daily* fashion.
We have seen how logistical and political barriers were overcome to report COVID deaths in a *daily* fashion.
Yet for the nation's overdose crisis that has claimed over 600,000 American lives in the past decade, it can take almost 2 years to get precise monthly overdose death numbers for some states.
Yes, overdose tracking is complicated.
Toxicology and autopsy reports can lag by months in the most underfunded places.
We need to invest in the country's public health infrastructure.
Rapid data is used by most wealthy nations to keep their people safe. We deserve the same.
Toxicology and autopsy reports can lag by months in the most underfunded places.
We need to invest in the country's public health infrastructure.
Rapid data is used by most wealthy nations to keep their people safe. We deserve the same.
But being transparent with the data we do have is an important start.
Critically, the provisional data don't include race/ethnicity, even though we know overdoses are increasing the fastest among Black and Indigenous communities. cc: @DrAyanaJordan https://www.npr.org/2021/03/03/970964576/drug-overdose-deaths-surge-among-black-americans-during-pandemic
Critically, the provisional data don't include race/ethnicity, even though we know overdoses are increasing the fastest among Black and Indigenous communities. cc: @DrAyanaJordan https://www.npr.org/2021/03/03/970964576/drug-overdose-deaths-surge-among-black-americans-during-pandemic
To me this is similar to the NYT having to sue the Trump admin's CDC in order to obtain COVID death numbers disaggregated by race/ethnicity.
We knew disparities were likely happening, but the data were not released publicly.
https://www.nytimes.com/2020/10/23/opinion/coronavirus-data-secrecy.html
We knew disparities were likely happening, but the data were not released publicly.
https://www.nytimes.com/2020/10/23/opinion/coronavirus-data-secrecy.html
The researchers at the CDC do great work in a difficult data landscape and political situation.
Let's support them in improving overdose data surveillance and transparency by making it a political priority.
Let's support them in improving overdose data surveillance and transparency by making it a political priority.
@samirakre and I have made the code and data from this framework publicly available:
https://github.com/akre96/cdc_overdose
We will continue to update the results as CDC releases provisional 12-month data each month.
https://github.com/akre96/cdc_overdose
We will continue to update the results as CDC releases provisional 12-month data each month.
Hopefully the CDC will be able to start publishing the underlying monthly values alongside rolling aggregates for each state.
I think that would be a better solution than 2 grad students recovering them and posting them on github each month.
I think that would be a better solution than 2 grad students recovering them and posting them on github each month.
Some important steps have been made in that direction in the new administration's CDC.
For example this 'now-casting' dashboard is very cool
https://www.cdc.gov/nchs/nvss/vsrr/vsrr11-dashboard/index.htm
I would love to see something similar with race/ethnicity data + all the info from the provisional data.
For example this 'now-casting' dashboard is very cool
https://www.cdc.gov/nchs/nvss/vsrr/vsrr11-dashboard/index.htm
I would love to see something similar with race/ethnicity data + all the info from the provisional data.
Importantly, it is possible that 2020 was even worse than these grim figures.
If data reporting lags got worse during COVID, then we would underestimate the true death toll in 2020.
Read more about limitations here:
If data reporting lags got worse during COVID, then we would underestimate the true death toll in 2020.
Read more about limitations here:
Huge thanks to my co-author @samirakre, and many colleagues and mentors who have supported us in this work.
Data and code: https://github.com/akre96/cdc_overdose
Paper: https://ajph.aphapublications.org/doi/10.2105/AJPH.2021.306256
Data and code: https://github.com/akre96/cdc_overdose
Paper: https://ajph.aphapublications.org/doi/10.2105/AJPH.2021.306256
Also, we should finally support widespread adoption of common-sense harm reduction and substance use treatment measures @MorganGodvin https://www.latimes.com/opinion/story/2021-01-05/overdose-epidemic-fentanyl-treatment-buprenorphine-methadone
