Bad Data on Prescribing and Overdose Data Dashboard
Just wanted to give you all a heads-up about a problem I stumbled on a few weeks back. Unfortunately it still hasn't been addressed so now I'm worried that people might be using bad data from this dashboard (the link to the dashboard is here). If you or anyone you work with is using it, you should probably shoot them a warning. I haven't looked at other tabs, but the one for "Overdose Deaths - Vital Records" is way off. Here's what I sent Dr. Dagan Wright (one of the data scientists listed on the contact information tab of the dashboard) back on May 2nd:
"It appears as though the count of deaths for each county, and for the aggregated regions, are doubled. The issue doesn’t seem to impact the statewide number. Description below:
"It appears as though the count of deaths for each county, and for the aggregated regions, are doubled. The issue doesn’t seem to impact the statewide number. Description below:
I’ll use my county as an example, but I checked around 10 counties (and our counties’ region 5 with Marion and Polk).
- Both map and graph show that the 5-year period of 2014-2018 for Yamhill County and “any opioid” gives a count of 70 deaths and a rate of 13.42 deaths per 100k population
- When clicking on a county on the map, or on a data point on the graph (after selecting a county or region), a little bar with an option to look at a table pops up above the mouse-over textbox summary.
- When clicking on the table, and then choosing “Full Data”, 2 distinct rows with identical information appear, and the count and rate columns are exactly half of summary total (35 deaths and rate of 6.71 deaths per 100k population).
- The reason why there would be two distinct rows is not clear, but looking at a bunch of different geographies it’s clear the same issue is happening. I also checked a different category (heroin instead of “all opioids”), 3-year instead of 5-year, and a 2009-2013 value and they’re all consistent. I’m unclear why the statewide summary values are not behaving the same way.
- There isn’t an “apples-for-apples” selection in OPHAT for verifying, but accidental poisonings (which would include more than just Opioids) values were more in line with the 35 deaths than the 70."
So everything on that tab is doubled, including counts and rates, except for the full state-wide number. Dagan responded a few days later saying he appreciated the email and that they're rebuilding the dashboard but having staffing challenges and they'd look into it. I was reminded of this issue this morning so I pinged him again (the other epi on the contact information tab has already told me he hasn't been with the group for a couple years).
Personally, an error of that magnitude on a state-wide public dashboard feels like it should rise on the priority list, especially since opioid deaths are a front-and-center topic right now. But, I felt you all should be made aware in case you or anyone you're working with is currently using these data.
Personally, an error of that magnitude on a state-wide public dashboard feels like it should rise on the priority list, especially since opioid deaths are a front-and-center topic right now. But, I felt you all should be made aware in case you or anyone you're working with is currently using these data.
You can get multiple cause of death data for newer time periods (and some provisional data) using CDC WONDER, which maybe would be a better choice? I haven't used the OHA dashboard in some time (because it's so outdated) and typically use either our own vital records data or ME data to get overdose counts/rates.
You are right for for Yamhill-using CDC WONDER I get 34 total "any opioid" deaths for 2014-2018 combined.
Molly
I've been a CDC WONDER user for many years and have taught myself about extracting overdose deaths by re-creating nationally published data as well as comparing my local vital records data to what is in WONDER. I get periodic questions about how to use the multiple cause of death files so I thought I would share a guide I created to walk you though how I do it. This example uses Linn County to find "any opioid" overdoses by year from 2014 to 2020. Let me know if you find it helpful (or find any issues that don't work with your local data). Note that there are varying case definitions for drug overdoses. This is