Skip to content

And the band played on: Low quality studies being published on Covid19 prediction.

According to Laure Wynants et al Systematic review and critical appraisal of prediction models for diagnosis and prognosis of COVID-19 infection  most of the recent published studies on prediction of Covid19 are of rather low quality.

Information is desperately needed but not misleading information :-(

Conclusion: COVID-19 related prediction models for diagnosis and prognosis are quickly
entering the academic literature through publications and preprint reports, aiming to support
medical decision making in a time where this is needed urgently. Many models were poorly
reported and all appraised as high risk of bias. We call for immediate sharing of the individual
participant data from COVID-19 studies worldwide to support collaborative efforts in
building more rigorously developed and validated COVID-19 related prediction models. The
predictors identified in current studies should be considered for potential inclusion in new
models. We also stress the need to adhere to methodological standards when developing and
evaluating COVID-19 related predictions models, as unreliable predictions may cause more
harm than benefit when used to guide clinical decisions about COVID-19 in the current


  1. jim says:

    Yeah and whoever you are at whatever level of government:

    We have enough freakin’ maps already! What we need are simple tables of quality, up-to-date data.

    Focus on the data, *please*

    • Keith O’Rourke says:

      Apparently everyone (even directors of Biostats schools) is (are) starving for data and unable to get access.

      • jim says:

        Here in WA, we had a handy table updated daily, until suddenly it was replaced by a Tableau-type map of the state, which under other circumstances would have been hilarious because it has two dark blue counties (Seattle) and basically nothing else. DUH.

        Mean time, after the map appeared (Friday) there were no updates until yesterday, and even today it still says “Last updated 11:59pm 3/28/20” which apparently now means the site not the data? whatever, the old table is still there but not updated since Friday. What a mess. If they would have stuck to just posting data we’d all be better off, but the state’s tech department I guess wasn’t going to be outshined by Johns Hopkins! (or they didn’t want to pass on the OT)

  2. Sid says:

    It seems the white house has announced that the model its going to use is . Most importantly uses # of deaths as the only reliable source of data to estimate resource utilization as the uncertainty in that is less than in other sources because of still a lack of testing.

  3. Terry says:

    This guy (Michael Osterholm) seems to know what he is talking about. He thinks it is impossible to do forecasts right now because we just don’t know enough (1:08:00).

    • I was relieved that Michael Osterholm conveyed that forecasting was precarious. Peter Attia’s several podcasts are excellent.

      Michael seemed to echo a few of John Ioannidis’ viewpoints It’s been two weeks perhaps since John’s video came out where he bluntly said that the evidence was exaggerated and unreliable. I wonder how John would construe the current situation.

      I’ve asked Mike Magee, author of ‘Code Blue to share his viewpoints on the health care system here on Andrew’s blog. A timely subject. I steer others to this blog too.

  4. Jordan Anaya says:

    Ironically my boss just asked me if I thought we should publish an analysis using publicly available COVID-19 data.

  5. Rahul says:

    Overall I see too many models. Too little actionable intelligence.

    And worst of all, even where there exists actionable intelligence, even then too little actual novel action.

    I think we are trapped in the paralysis of fear of doing more wrong.

    But unless we try out of the box strategies, even at the risk of doing damage locally, we will never hit the really great solutions!

Leave a Reply