# Close to Peak COVID?

## An analysis of confirmed cases and estimated infections in Ireland in September 2020

At the time of writing the 7-day rolling average of confirmed cases in Ireland was just over 200 cases per day and growing. Given that our peak was 903 cases per day back in April, this suggests we are now exceeding 22% of this peak. Some people have questioned whether this is a fair comparison, and there are reasons to believe that it may not be:

- We are much better at finding cases now than we were in April, because our testing infrastructure is a lot better; tests have reached approximately 10,000 tests per day recently, compared with about 5,000 tests during the peaks of spring.
- Test positivity rates are much lower today than they were in spring, which means we should be missing fewer true infections now; test positivity peaked at more than 20% during April, compared with current rates of less than 2%.

It turns out that although current cases do exceed 22% of their past peak, the current number of estimated *true infections *is just 14% of its past *infection* *peak*, which is likely to be a more realistic relative assessment of our current outbreak.

The rest of this article explains why this is the case, but let’s be clear: whether its 22% or 14%, the point is that cases are growing too quickly — *they are doubling every 18 days at the moment and we know this doubling time will reduce if we don’t act decisively* — which means we may only have one shot to bring them under control.

# Estimating True Infections

We all know that confirmed cases are a subset of the true number of infections because even if we ignore testing issues, 40% of infections are asymptomatic and mild cases account for 80% of all infections. So, even if everyone with symptoms reported for testing we would still miss at least 40% of true infections.

Can we estimate the true number of infections? There are several models that do just this. Most of them work by backtracking from fatalities to estimate how many past infections would be needed to produce an observed number of deaths. They do this by evaluating lots of different combinations of infection fatality rates and different durations from infection to death, to find a combination that provides the best fit to the observed data. Some models will include test positivity information as well. Others also incorporate population age demographics.

The graph below shows the actual number of confirmed cases for Ireland, along with the 7-day rolling average of test positivity rates. To make it easier to read the lows of the summer against the highs of spring, the y-axis uses a log scale. The graph also shows the corresponding estimates from the following 4 commonly used infection prediction models:

- The Institute for Health Metrics and Evaluation (IHME)
- Imperial College London (ICL)
- Youyang Gu (YYG)
- The London School of Hygiene & Tropical Medicine (LSHTM)

We can see how, during most of the period represented, the models predict new infections that are a multiple of confirmed cases, peaking in April. Recent rises in the number of confirmed cases have been accompanied by a corresponding increase in predicted infections. However, somewhat surprisingly, the IHME and ICL models predict *fewer* infections than there were confirmed cases in August. The likely explanation for this is that it is due to issues that arise when there are days with zero deaths, which is a key parameter that these models rely on as the basis for their predictions.

This above chart shows the average of the model predictions, but we only include predictions on days when the models predict more infections than there were confirmed cases. One implication of this is that the most recent predictions (from August 12 onwards) are based exclusively on the YYG model. It is clear too that the gap between true infections and confirmed cases is narrowing, even though positivity rates have been rising in recent days. To compensate for this, in the above, from August 12, the predicted number of new infections are scaled from new cases, based on the infection gap as of August 12. This ensures that new infections rise with confirmed cases (the dotted red line) rather than remaining flat.

# Peak Cases vs. Peak Infections

We can now calculate the number of cases/infections per day *relative* to their past peak. The results, shown below, indicate that current case numbers are approximately 22% of their April peak, but current infections are 14% of their April peak.

On the one hand, to those who may be alarmed that we are approaching the highs of April, we still have a distance to go, which means there is still time to put the virus back in its box! On the other hand, to those who claim that current true infection levels are so far below what they were in April that there is no need to worry, this logic is similarly flawed: yes estimated infections are just 14% of April highs, but they are currently on an 18-day doubling trajectory, and if we allow transmission to continue as it has been, then this doubling time will reduce further, leading to the type of exploding numbers we saw earlier in the year. And these numbers will put pressure on the hospital system eventually.

It is worth noting too that the sum of these estimated infections comes to approximately 300,000. This means that just over 6% of the population have been infected so far or, to put it another way, 94% are fair game for the virus and it will be only too eager to oblige if we allow it in. We have no choice but to redouble our efforts to contain transmission, once again. There is still time and the coming days will reveal whether we are able to take full advantage of it. The question is not what to do but whether we will do it.