Read on Twitter
As we all know COVID-19 hospitalizations have taken a heavy toll on many healthcare systems. In our new work we try to quantify how heavier patient loads might lead to higher in-hospital mortality of COVID-19 patients.
PREPRINT: https://www.medrxiv.org/content/10.1101/2021.01.11.21249526v1
1/15
PREPRINT: https://www.medrxiv.org/content/10.1101/2021.01.11.21249526v1
1/15
We used data of the day-by-day clinical state of all 19,336 COVID-19 patients hospitalized in Israel from mid-July 2020 to early Jan 2021.
2/15
2/15
We estimate mortality by fitting a multistate survival model for predicting patient mortality.
The model takes as input patients’ age, sex, and clinical state on first day of hospitalization.
3/15
The model takes as input patients’ age, sex, and clinical state on first day of hospitalization.
3/15
We previously showed that just these few covariates lead to very accurate and well-calibrated mortality predictions. Model details are here: https://twitter.com/ShalitUri/status/1309215391160446978?s=20 (paper now accepted for publication at Journal of @AMIAinformatics)
4/15
4/15
We fit the model to patient data from mid-July to mid-Sept 2020. During this period hospitalization load was relatively low: 250-500 severe+critical patients. We then use the model to predict mortality from mid-Sept 2020 until early Jan 2021.
5/15
5/15
Notably, this later period saw large swings in hospitalization load, ranging from 250 severe+critical to more than 900. We see that during the peaks of patient load (Sept - Oct and then again Dec-Jan) 14-day mortality exceeds the model’s predictions by more than 20%.
6/15
6/15
In order to further validate the model we took a deeper dive, and created cumulative mortality graphs for patients hospitalized in every consecutive week.
7/15
7/15
Predictions are spot on for the training data, then underestimate mortality in weeks 0-4 (high hospital load), then again spot on for several weeks of lower load, and finally again underestimate mortality in the last weeks when hospital load is high again.
8/15
8/15
Since our model controls for age, sex, and initial clinical state, it’s less likely increased Sept-Oct and Dec-Jan mortality rates are due to changed patient population.
9/15
9/15
We also believe the fact the model is accurate during Oct-Dec (even though those months are farther out from the time data was fitted on) hints against a model-fitting problem or seasonality issue.
10/15
10/15
Taken together, we think this is evidence that high patient load might be a cause for increased mortality rates.
11/15
11/15
Notably, the Israeli government marked 800 severe+critical hospitalized patients as the point at which standard quality of care cannot be assured. We see that even at 500 severe+critical hospitalized patients mortality rates seem to increase significantly.
12/15
12/15
Many physicians in Israel have been warning us for months that quality of care cannot be assured when patient admissions soar and entire wards are cleared and remade to take in COVID patients. Unfortunately it seems their warnings have indeed been borne out.
13/15
13/15
I am afraid the conclusion we have is quite dire. Hopefully vaccinations and control measures will help stem the tide of infection and illness sooner rather than later.
14/15
14/15
Finally, this is joint work by a fantastic team, led by @H_Rossman and @tomer1812, supervised by @GorfineMalka, @segal_eran and myself, with important contributions from @smadarshilo, @RomGutman, @jonathansomer1 and Asaf Ben Arie. Here’s to a better 2021.
15/15
15/15
