On Covid-19 in Italy — An updated view at the data from last week
A week ago, on 21 March, I tried to use the Covid-19 stats from Italy to provide a short term prediction for new cases and mortality rate:
It’s fair to say that my prediction was not accurate, although not totally off (more details later) but this is clearly not enough to deter me from doing more analysis.
Here in London we’ve gone through our first week of lockdown, and numbers from the pandemic are increasing fast, tracking Italy’s record with a two-week delay. Looking at Italy’s data is a bit like trying to predict what is going to happen here. I am currently using the latest data from 28 March.
What was incorrect in my prediction from a week ago? The number of deaths didn’t grow as much as the number of new confirmed cases (which is good) and the number of new cases slowed down, but not as fast as I was hoping (which is bad). Combine the two together and the possible outcome might not be much different from what was expected, but perhaps it will be spread over a longer period of time. And this is also bad — it means a longer time locked into our houses, I presume.
But let’s go in order…
First of all, 18 days into the full-Italy lockdown, we have probably started to see some initial signs of slowdown from the virus: the growth of new confirmed cases has gone from over 20% per day of the first week of March, down to around 7–8% of 23–28 March. I was hoping this would now be at around 4%. So, not quite there yet.
Also, let’s not forget that the percentage of growth tells us about the shape of the curve, but it’s not the full story. Absolute numbers are much scarier, due to the perverse effect of “compounding” after a full month of double-digit-percentage growth: so while at the beginning of the month we were seeing a few 100s to just over 1,000 new cases per day, last week averaged at nearly 6,000 new cases per day. No reasons to rejoice yet, really.
My prediction for the slowdown of new cases was more marked (see chart above) in anticipation of the lockdown effect, so it’s still not clear whether the lockdown, as currently implemented, will have the full desired results or not. We need to keep an eye on the numbers for longer, to be able to tell.
Secondly, the number of deaths has followed a path less dreadful than my prediction (which is a good thing!) but directionally it still recorded increasing levels. This means that while the mortality rate has not remained constant (but has reduced), but what did also happen is that while we saw a relative slow down in new cases (down to 7% or so), the new deaths have continued to grow at 10% or more per day.
If we look at the numbers in detail, my prediction was looking at the growth in cases from 9 days before (averaged over 3 days) to be the best predictor for a given day’s growth in mortality. See chart below:
When I presented this model, on 21 March, the total confirmed cases from 9 days before (12 March) were 15,113 and I considered them “responsible” for the total of 4,825 deaths recorded up to 21 March, with a mortality rate of 31.9%.
Move the needle by a week, and you see that the additional 25,922 confirmed cases of 12–19 March (41,035–15,113) would be responsible for the additional 5,198 (10,023–4,825) victims from 21–28 March, implying a much reduced mortality rate of 20% for that week.
So, what might have changed to record this different (and improved) result? There are several possibilities:
- More “mild” cases of illness were identified, which therefore had less probability to succumb to the virus. This could be due to a larger number of tests conducted
- Our hospitals found better ways to treat the serious cases, improving their chances of full recovery or at least keeping them alive for longer
- The method or the timing to calculate and report deaths caused by the virus, or new cases has changed
For the first point, if we look at the number of tests done during this period (chart below) we can see that Italy indeed went from 86,011 tests (which found 15,113 positives) to 182,777 tests (for a total of 41,035 positives) in just a week, between 12–19 March. So we can certainly see that the testing activity increased massively (over 96,000 tests in 1 week) and it’s possible that the span of confirmed cases included more people with mild symptoms who had more chances to recover.
At the same time, however, we can see that the percentage of positives found per test done is actually increasing (from around 17% to 27%) so it’s definitely not the case that Italy started to test “speculatively”. Perhaps more people took the virus threat seriously and called in to be tested? Or simply the testing capabilities (availability of swabs, bandwidth of testing labs) improved and Italy could afford to test many more suspected cases?
Interestingly, a single region, Veneto, is doing many more tests, in proportion, than the rest of the country: 89,380 tests for 7,930 confirmed cases as of 28 March — around 11 tests per confirmed case, when the national average is more like 4 to 1. Lombardy, the most affected region, with 5 times more cases than Veneto (nearly 40,000), has performed around the same tests as Veneto. Probably a sign of an overwhelmed system. As we know averages don’t tell the full story and can actually be misleading.
It’s also important to know that someone who is affected by the virus, then is tested positive and eventually recovers, in Italy is tested four times: twice to confirm the positive results, and twice to confirm the full viral recovery. So all those tests are not done to different people, and therefore the percentage of different people who were tested and found positive (vs. total of different people tested) are even higher than those 17% and 27% figures.
Is Italy testing enough people to have a good picture of the spread of the virus? Despite the over 400,000 tests done so far, the spread of the virus is so wide that this is hard to achieve.
I have no information on the improved ability to treat people with the virus, but I read several times that new medicines have been tried with some success on Covid-19 patients, and I would also expect the availability of intensive care units to be increased over time. So this may well be a factor too.
I also have no information about new ways to account for deaths or new cases, so I will discount this factor completely.
In summary the decreased mortality rate experienced over the past week could well be a combination of more testing done and better treatment and hospital facilities.
Some of the data available is of difficult interpretation, especially when we look at different countries, as there are different methodologies, timings and standards driving the way people are tested and accounted for in the various categories. Sometimes these differ even within the same country, for understandable reasons (e.g. the local health system is overwhelmed) or by design (e.g. the Veneto region in Italy testing much more extensively and proactively than other regions).
Looking at the next few weeks, can we expect new deaths to slow down? Over the past 9 days there have been over 51,000 new confirmed cases (from 41,035 of 19 March to 92,472 of 28 March). If the most recent trend confirms, these could sadly translate to around 10,000 new deaths over the next 9 days (at 20% mortality rate).
But testing has intensified significantly over the same period: there have been around 247,000 tests in 9 days, so the hope is that the recent confirmed new cases are on average less serious than in the past, and also that when hospitalization was required, they found better treatments.
WIth a mortality rate reduced to 10% from this group, we would expect a total of just over 5,000 deaths over the next 9 days.
The chart above shows the mortality rate when calculated on the same day’s total confirmed cases, or when we go back a number of days. These lines will eventually converge, once the virus spread is finally under control. They will converge to a high percentage if we continue to test relatively little, or they will converge to a low figure if we start to test more. Our ability to provide a high quality treatment to serious cases will also affect the outcome.
Something important to also discuss, is the number of people in intensive care, which is also duly reported on a daily basis by Italian authorities. These are the most serious cases, and their number could also be a short term predictor for fatalities. Certainly, if we expect to see a marked decrease in new deaths, we have to first see a decrease in the number of patients in intensive care at any point in time. So far, this number has been steadily increasing, and it is very large at the moment, standing at nearly 4,000.
Ok, I think I’ve finished trying to justify my mistakes with the earlier prediction. I made a couple of big assumptions and they have proved incorrect: one on the right side (less deaths) and one on the wrong side (growth of new cases still fairly high). Are we still going to see 30,000 deaths as per initial simulation, before this first wave of contagion is over, in Italy? Let’s hope not, but it’s still hard to tell.
The situation within Italy is not homogeneous. The northern regions and Lombardy in particular are way ahead of the rest of the country in the contagion, and while we hope the other regions won’t get to the same level of cases and fatalities (as they took the tougher measures at an earlier stage) we might still witness more than one peak, and a prolonged situation with many different battles to fight.
Be safe, stay home!
Un abbraccio, Italia.
