Projecting future deaths from COVID-19 cases

As the number of COVID-19 cases increases in the US and Europe over the last few weeks, some people have claimed that deaths have not increased. Indeed, Donald Trump Jr has claimed* that the number of deaths has declined to “almost nothing”.

Daily number of deaths in the US attributed to COVID-19. The black line is a weekly average, which smooths variations and anomalies such as the lower reporting rate associated with weekends.

Firstly, this claim is flatly false – deaths in the US are increasing, and have been doing so for a fortnight. But will the new large spikes in cases lead to large spikes in deaths? The TL;DR is “yes because deaths will always lag behind cases, and we can forecast the number of future deaths based on current cases”. Let me explain…

The time lag between cases and deaths will depend on when people are typically tested (before or after emergence of symptoms) but regardless, a lag will exist. It can take a week or more for the disease to progress to a dangerous condition, and longer for death to ensue for the worst affected people.

Daily number of new COVID-19 cases reported in the US. The black line is a weekly average which increased rapidly in October.

So now that we are seeing record numbers of new cases, what fraction of those are likely to lead to deaths? In the US, there appears to be an approximate 20-day lag from cases to deaths. Plotting weekly deaths as a proportion of cases 20 days previously shows that the proportion of positive test results that translate to deaths has been relatively stable over the last few months (since July). Approximately 1.5-2% of reported cases are leading to deaths 20 days later.

Number of deaths as a proportion of COVID-19 cases 20 days previously in the US.

While relatively stable now, that percentage has declined since the start of the epidemic. There are at least three reasons for that. Firstly, the level of testing has increased, such that more of those who are infected are being detected than at the beginning of the epidemic; as the denominator increases, the ratio of deaths to cases declines if the numerator (deaths) remains constant.

Secondly, it is possible that less vulnerable people (typically younger people) are making up a larger fraction of the infected population, leading to a lower proportion of people dying. Thirdly, treatment of hospitalised patients has improved. But any signs in the plot above of better treatment in the last couple of months requires rather heroic optimism.

The pattern seen at the national level in the US is similar within states. Again, the proportion of reported cases translating to deaths 20 days later is relatively stable over time for most states, although there is more variation. The variation arises at least in part because some states (e.g. Vermont) have had comparatively few cases and deaths. Some of the variation will be reporting anomalies as newly identified deaths are reported in clusters. Other spikes will occur in response to outbreaks in vulnerable communities.

And for similar reasons to the variation over time, the proportion of cases that translate to deaths will vary among states. Some states will have identified a greater fraction of infected patients, some states will have a higher fraction of their infected citizens in vulnerable groups, and health care might differ among the states.

Number of deaths as a proportion of COVID-19 cases 20 days previously in each of the lower 48 states of the US.

All these (and other) factors mean that the ratio of deaths to cases 20 days previously will vary. Across the US, an optimistic scenario is that the number of deaths will be 1.5% of the number of cases 20 days earlier. A figure of 1.75% might be more realistic, and 2% is not out of the question. With the US now having reported over half a million new CORID-19 cases in the last week, deaths are likely to be averaging over 1000 per day within 20 days (up from ~800 currently).

We see a similar pattern in the EU and UK. Again, the ratio of deaths to cases that were reported 20 days previously is relatively stable over time in recent months but varies among countries. But it tells us that the current increases in cases in Europe will almost certainly lead to increasing rates of death. However, several countries are imposing restrictions that will reduce the number of cases (and subsequent deaths and strain on the health system).

Number of deaths as a proportion of COVID-19 cases 20 days previously in each of the countries of the European Union (including the UK).

Meanwhile, the White House is pursuing the line of not controlling the pandemic in the US. Sixty thousand more COVID-19 deaths in the US before the end of year seems like an overly-optimistic lower bound based on the current trajectory of cases. And it will seemingly be another 3 weeks into 2021 before there is even the chance of a change in national leadership, depending on the outcome of next week’s election. If cases contine to increase in the US through January 2021, the number of deaths is unlikely to decline until February at the earliest. The trajectory of COVID-19 in the US looks exceedingly poor.

* I apologise for drawing your attention to this ridiculous claim. Edit: He used data from this CDC site (https://www.cdc.gov/nchs/nvss/vsrr/covid19/index.htm), which tallies the number of deaths on a weekly basis. However, the site notes that more recent data are incomplete (delays of up to 8 weeks), so it unsurprisingly shows reduced deaths in recent weeks.


About Michael McCarthy

I conduct research on environmental decision making and quantitative ecology. My teaching is mainly at post-grad level at The University of Melbourne.
This entry was posted in Uncategorized and tagged , . Bookmark the permalink.