Read on Twitter
Hi #epitwitter our state-level analysis is up on medRxiv. We look at 11 data streams (cases, hospitalizations, ICU, vent numbers, deaths etc) available from DOHs in RI, MA, PA. Data assembly not automatable, unfortunately (help welcome) https://www.medrxiv.org/content/10.1101/2020.11.17.20232918v1 Key points below 1/
1. **Vulnerable populations are not always able to isolate effectively during a lockdown**. Very clear signals, visually in the data and in our inference, that contact rates for >80 individuals were the highest (among age groups) during the lockdown. 2/
This means that a strategy of returning to normalcy and protecting the vulnerable will not work. 3/
2. **Hospitalization numbers can be underreported**. They are underrepoted in PA and MA in the 'new hospitalizations' data stream but not in the 'current hospitalizations' data stream. Plz be careful summing hosp numbers from dashboards unless you abs know what you're doing. 4/
When you see 100 new hospitalizations reported somewhere, keep in mind that there may be another 30% that are unreported (depending on that state's DOH reporting) & there may be another XX% w severe disease in congregate care settings that are not classified as hospitalized. 5/
What is this XX% ? It depends on what phase of the epidemic we're in, and whether congregate care facilities/nursing homes are experiencing a big wave right now. 6/
3. **Improved clinical case management**. Some good news. It appears that either severity or mortality (or both) dropped from spring to summer, likely due to improved clinical management of hospitalized patients. Posteriors for ICU admission below, for RI/MA/PA left-to-right. 7/
4. **Infection Fatality Rate (IFR)**. Our IFRs are on the higher end of published studies, >2% age-adjusted. This is probably due to the RI/MA epidemics running through the most vulnerable sub-populations in Mar-May. Higher CFRs in these states are consistent with this. 8/
5. **Live attack rate estimation**. We can provide live attack-rate estimates, already posted on http://mol.ax/covid but need more validation from seroprevalence surveys. No inconsistencies so far w 4 published studies, but some gut feeling that our estimates are a bit low 9/
Thanks @RIHEALTH @MassDPH @PAHealthDept @nathanwikle @athutran @lmadoff @BillHanage @ciddpsu @HaiderInam @KarelBrinda @JustinPritcha19 and many others 10/
