The more I dig into Ohio’s 4,000 missing deaths, the more I’m blown away. This shouldn’t have happened, period. It should have easily been caught if they had the right people and processes in place (1/12)

This is my new hospitalization model that I display a few times a week. It’s based off of the currently hospitalized count and assumptions around the rate of discharges and deaths (2/12)

It worked extremely well predicting the 7-day MA from 5/31 to around 10/31 with an r2 of 0.9406 on hospitalizations and 0.8791 on discharges + deaths (3/12)

As you can see in the chart below, my modeled values tracked actual values extremely well until November, then it abruptly broke. I was still predicting the trend well, but I was way high on the level of hospitalizations (4/12)

I knew something was going on and assumed multiple things could be happening. Was remdesivir driving longer hospital stays, throwing off my model? Were we miscounting hospitalizations? Were lag times from admission date / DoD extending? (5/12)

I don’t have access to @govmikedewine or @ohdeptofhealth to ask questions. Tweeting to @ohdeptofhealth is like sending messages into a black hole. If they hired strong analysts and allowed outside analysts to question the data, this problem gets caught much earlier (6/12)

Since I didn’t have access to the right people to ask questions, I rebuilt my model. I never dreamed they’d be undercounting deaths by 4,000. Here’s what changed... (7/12)

One of the key assumptions in my model is discharges + deaths as a % of the beginning hospital census. From 6/1 to 10/31, it had daily volatility as expected (consistent with days of the week that have higher / lower releases), but the average was consistent around 9.7% (8/12)

Then in November the bottom fell out. The avg. discharges + deaths as a % of beginning census dropped from 9.7% to 6.7% since 10/31. In short, this means fewer people as a percentage of the beginning hospital balance were getting discharged or died each day. Red flag. (9/12)

Further, @ohdeptofhealth has access to more information than you or I do. They should have easily been able to see deaths were running low. An error of this magnitude should have been caught much sooner than this– I could see something was off in mid to late November (10/12)

At the end of the day, @ohdeptofhealth is compiling data that our elected officials are using to make important decisions. Our leaders need access to high quality data sets that they can have confidence in (11/12)

@govmikedewine / @ohdeptofhealth – get this fixed, now. Cut down on the lag time between date of death & reporting. Hire the right people. Clean up your processes. Finally, let analysts who understand how to interpret data ask questions to hold you accountable (12/12)