Corona in Israel: Why I ignore serious cases

Eliezer (Ilja) Schwarzberg
4 min readAug 22, 2021

Corona in Israel is peaking once again. As I previously pointed out, the current wave deserves as much attention as all of the earlier ones.

This seems to be true despite the vaccines and, presumably, better treatment in the hospitals.

To reach this conclusion I only looked into new positive cases (by age) and total deaths. I knowingly ignored additional data points that are provided by the Ministry of Health, most notably the number of serious cases.

Daily development of serious cases from the government’s official data dashboard

There are some articles, for example, for or against the vaccines, that make their points based only on this data.

I am not saying these articles or the analyses are necessarily wrong. I am just saying I do not have faith in the data on this level. Let me explain.

Definitions

My first an foremost question is:

Who defines when a patient is critically sick?

As far as I understand, it happens in the hospital, where I trust the medical personnel to do their job and save people’s lives. (We cannot thank them enough!!) However, I am less fond of using the labels “serious” or “critical” in analysis.

These labels might change over time, or they may differ from one patient to the next.

Furthermore, one of the assumptions when using this data is that only people who were “seriously ill” will die. I am at all not sure this is the case. If someone dies at home, will she retroactively be defined as seriously ill for statistical purposes?

Built-in biases

One of the best ways to demonstrate my skepticism is the comparison of the daily snapshot vs. the daily new cases:

Daily snapshot of serious cases (top); daily new serious cases (bottom)

In the past month, the total number of active serious cases went up from roughly 100 to 700. On the other hand, the daily numbers in the same time period add up to 2,000. Especially in the last week or two, we have seen ~100 additional serious cases every day, while the total increases on average by 20 per day.

Of course it makes sense! People who get seriously sick do not stay like this forever. They either recover or, God forbid, pass away. The vast majority seems to recover, by the way.

The problem from the data point of view is the following. People defined as seriously sick seem usually to quickly leave this condition, one way or the other.

It means very few people stay in serious condition for weeks; most recover or die within a day or two.

Recovering does not have to mean the patients leave the hospital, but rather that their condition has improved enough, as to not be defined as seriously sick.

In a small dataset where the population changes so quickly, a significant bias is inevitable. However, without low level data there is zero chance of accounting for it. Consider how much better the analysis would be if you could add the following data for each patient in serious condition?

  • Number of days since the positive test / initial symptoms
  • Time since vaccination
  • Age
  • If/when the person had COVID previously

Without more information from the raw data, I would not touch these aggregations.

What to look out for

This is why when I look into the data, I try to find the most clear-cut insights. As such, I rely on the number of positive tests by age and the death count. Both seem not to be subject to definition. It is possible to skew these numbers too, e.g., the number of positive cases would go down by testing fewer people. However, it looks like Israel is keeping the tests up and running, so I am not too worried about these.

On the other hand, data backfill is important to keep in mind.

Do not trust yesterday’s numbers.

Therefore, if you analyze governmental datasets, beware of a sudden change in the past. A good way to see it, is to compare the government data with another source:

worldometers for Israel claim 55 deaths on 2021–08–22
Government website claims only 16 deaths on 2021–08–22

Same date with two completely different numbers: 55 by worldometers (data by Johns Hopkins University) vs. 16 by gov’s data dashboard. The reason for the discrepancy is trivial. Israeli authorities updated a few days at once, while Hopkins only took the difference in total of today vs. yesterday.

A minor backfill is no big deal, but something you want to keep in mind when checking recent numbers.

Stay safe and healthy. Hopefully COVID will be over soon and for good.

--

--

Eliezer (Ilja) Schwarzberg

Data Scientist @Weel. Into Machine Learning, Data analysis using Python, GCP, G Suite. Love to get my hands dirty