4 ms·

More: > It was important to understand the timing of the outbreak as part of this investigation. Health officials wanted to know when exactly the problems firs

More: > It was important to understand the timing of the outbreak as part of this investigation. Health officials wanted to know when exactly the problems first cropped up, and whether they had missed many cases before those initial reports. That's why they turned to the surveillance data, collected from more than 3,200 emergency rooms from most states. >They found a gradual increase in emergency-room visits among people who vaped or used e-cigarettes, starting in January of 2017. > The surveillance system doesn't automatically flag health concerns – often scientists dive into the data once their suspicion is raised, as it was in this case. The outbreak only became apparent in the data once researchers defined more closely what they were looking for and focused on one age group.

nerdponx 7y ago
> The surveillance system doesn't automatically flag health concerns – often scientists dive into the data once their suspicion is raised, as it was in this case. The outbreak only became apparent in the data once researchers defined more closely what they were looking for and focused on one age group. I'd be surprised if there wasn't some kind of outlier detection method that could automatically flag a sudden increase in unusual cases.
- prepend 7y ago
  The web site has a lot of papers talking about that. They do detect spikes in defined syndromes. But for stuff like this that isn’t a defined syndrome and isn’t really captured well at the hospital it’s not an outlier until scientists do a lot of analysis.
  - AmericanChopper 7y ago
    The ETL for 3200 emergency rooms worth of data must be a nightmare. It would have to be largely unstructured upon ingestion, and I can’t imagine normalising it enough for this sort of automated outlier detection would be possible.
    - prepend 7y ago
      The guide [0] shows like 20 defined fields and then free text notes. It would be cool, but seems kind of impossible to tell a significant outlier (vape lung) from an insignificant outlier (misspelling). The way the article describes it, epidemiologists define a new condition (vape lung) and then look for that. [0] https://www.cdc.gov/nssp/documents/guides/syndrsurvmessagguide2_messagingguide_phn.pdf https://www.cdc.gov/nssp/documents/guides/syndrsurvmessaggui...

nerdponx 7y ago

> The surveillance system doesn't automatically flag health concerns – often scientists dive into the data once their suspicion is raised, as it was in this case. The outbreak only became apparent in the data once researchers defined more closely what they were looking for and focused on one age group. I'd be surprised if there wasn't some kind of outlier detection method that could automatically flag a sudden increase in unusual cases.

prepend 7y ago
The web site has a lot of papers talking about that. They do detect spikes in defined syndromes. But for stuff like this that isn’t a defined syndrome and isn’t really captured well at the hospital it’s not an outlier until scientists do a lot of analysis.
- AmericanChopper 7y ago
  The ETL for 3200 emergency rooms worth of data must be a nightmare. It would have to be largely unstructured upon ingestion, and I can’t imagine normalising it enough for this sort of automated outlier detection would be possible.
  - prepend 7y ago
    The guide [0] shows like 20 defined fields and then free text notes. It would be cool, but seems kind of impossible to tell a significant outlier (vape lung) from an insignificant outlier (misspelling). The way the article describes it, epidemiologists define a new condition (vape lung) and then look for that. [0] https://www.cdc.gov/nssp/documents/guides/syndrsurvmessagguide2_messagingguide_phn.pdf https://www.cdc.gov/nssp/documents/guides/syndrsurvmessaggui...