Covid19 US death rate “code” cracked, up to 18 days forward
Intro
In this short work you’ll learn that the covid19 death rate going forward for the next up-to-18 days in the US is already pre-determined based on new cases that have already occurred. In short, the death rate “code” has been cracked, I believe and I can demonstrate quantitative forecast of near term deaths. Policy implications are discussed — we can start to bend the death rate curve within a month if we act today.
I have updated with the latest data to observe how the data in Fig 1 tracks the prediction. Latest update includes data up to Nov 25.
Summary
It seems possible to quantitatively predict the US covid19 death rate on a daily basis, up to 18 days forward, based on cases already contracted. This is because the daily death rate in the US lags new cases, both 7 day averaged, by approximately 18 days. You can tell this specific lag from new case vs. daily death inflection points, as demonstrated in my recent work from Nov 15. Accordingly, I have reverse engineered the dynamic mortality rate to a stable value since mid-July of just under 2% of contracted cases. Based on this, on Nov 15, I predicted the orange dots in Fig 1 leading to my prediction of a 2x increase to the death rate by Dec 3, and have tried to make this graph clearer to the reader by focusing on the dotted elliptical area where we are in time at present. Data prior to Nov 15 already existed, and we can observe mortality rate moving upwards from Oct 17. The darker blue since Nov15 has actually occurred since I released that paper — the data matches the prediction so far quite well. Prior to this, it was not obvious to enough people that the death rate is on a real uptick. Also, that uptick will get worse over the next 2wks. I maintain my prediction that covid19 will be the #1 killer in the US within approximately 2wks of my Nov 15 prediction.
In short, I believe the death rate “code” has been cracked, in this work, for how fast death rates will rise up to 18 days in the future.
Policy implications
- We probably cannot bend the death rate curve over the next up to 18 days. Accordingly, hospitals and medical staff need to be braced for the onslaught. This methodology can and should also be extended to local regions to forecast hospitalization requirements. But we can influence beyond the 18 day time period.
- We can definitely bend the curve beyond 18 days by reducing new case counts today. Masks and physical distancing can have an effect within 7–14 days (lots of references, and also discussed somewhat in my previous work. If new cases begin to decline in 10 days, the death rate will begin to decline in 28 days. So if we start today, deaths can begin to decline in roughly 1 month. If we stay the course and do not make positive change, however, deaths will continue to escalate further.
Discussion and sources
A lot of discussion surrounds new case count. These are staggering numbers (>12M as of today — approaching 4% of the total US population), often quoted, but don’t directly translate, for most people, into actual mortality. For the first time, I believe, a definite proportionality constant between new cases and near-term mortality has been derived and demonstrated. That proportionality constant for this time is just under 2% of cases 18 days ago (1.73% for Oct, 1.62% for Nov through Nov 20). Similarly, the deaths 18 days from now are just under 2% of the new cases today. So for 1.6% mortality rate of positive cases, if you read 200,000 new cases on a particular day, this translates to approximately 3,200 deaths in a day approximately 18 days later.
Fig 1 shows the aggregate death rate per 100K population, calculated by myself for 7 day average sourced from primary open source data from onset until Nov 20. A few features are clear. First, covid19 is already on the death rate map next to the US three top killers: heart disease, cancer and accidents (2018 data from was the CDC was used) The death rate in April was horrible, and represented an even higher local percentage of the population at that time because deaths were confined to the northeast US. This also meant that in much of the remainder of the US, covid19 was not taken seriously, and here we are today. The light blue curve is the historical covid19 death rate prior to my Medium article of Nov 15. Now look at the area within the ellipse. The darker blue curve is the nearly 1 week of data since that article, and also explicitly shown are the orange dots which predict death rate already from the Nov 15 paper out to Dec 3. In this case, I use the average dynamic mortality rate over the last 18 days, but that’s a small detail. It is clear that since I released my recent paper, the data follows the prediction with uncanny correlation. I intend to update this in another week or so and to further predict. What is clear is that the orange dots (prediction) soon eclipse cancer and heart rate to make covid19 the US #1 killer. If my hypothesis is correct, nothing can stop that.
At that time, I noted the inflection points in the new case rate and the death rate, and determined from those inflection points that there was at the present time an 18 day lag. Oct 17 in Fig 1 in this work corresponds to the same feature in new cases on Sept 29 in Fig 1 of my most recent work. This is the origin, quantitatively, of the 18 day lag I refer to.
Fig 2 is the dynamic mortality rate, and the calculation was described as Fig 4 in this very recent work. From the assumption that inflection points are true indicators of death vs new case lag, I tested this prediction by calculating the mortality rate as if this were true. What I get is a stable mortality rate from mid July, which means two interdependent things: (1) we neither significantly better nor worse since mid-July at treating covid19, but prior to then there as a lot of learning and (2) the stability of this curve is consistent with that this curve is in fact representative of reality. If (1) isn’t true, then (2) might also not be true, but since the curve is stable and also predicted from death vs new case lag, Occam’s razor suggests this is, to first order, a correct causation/correlation argument. This is what enables the predictive calculations above on a going forward basis. If this rate hasn’t changed in 4 months, it is unlikely to change over the next 18 days.
Fig 3 shows the mortality rate vs time, directly copied from Worldometers, a source of primary data, and it is this measure which has been historically available until now. This measures the mortality rate for known and resolved cases, not the population as a whole, and was discussed in more detail as Fig 2 in this recent work. While this publicly available data is an important metric, my issue is that the value at any time is an average, and not necessarily representative of the mortality rate going forward. A more useful metric for our purposes would be the dynamic mortality rate. Unfortunately, there is no way to know this from the primary data as such, because it is not possible to know when the cases were contracted which led to today’s deaths. It is only in a place of stable mortality rate (not changing) that we can simplify the requirement — we don’t need to know exactly when the case was contracted. The methodology in Fig 2 would be far less meaningful if the dynamic mortality curve were not stable. This new way of sensing dynamic mortality rate gives predictive power that I didn’t realize possible before.
Conclusion
The method of prediction I made on Nov 15 for US covid19 death rate appears to be matching the data since then. We can predict the US Covid19 deaths up to 18 days forward. Deaths are escalating and are predicted to make covid19 the US #1 killer imminently, based on cases already contracted and logged. We cannot change this.
We need a national mask mandate today. We need physical distancing today. If we do this today, deaths will start to decline in approximately 1 month.
