Updated: Sep 8, 2020
In my first article, we laid the ground work for examining the effect of existing medical conditions on COVID outcomes. The article was prompted by a series of tweets and social media that suggested only 9,000 people really died of COVID. The rest died of other conditions. We’re going to build a model that explains how pre-existing conditions impact COVID outcomes and use that to explain the concept of comorbidity.
We have a data set from the CDC with a lot of information. To keep things simple, we’ll start by looking at the relationship between deaths and the existence of medical conditions. This data set doesn’t break out the individual conditions, so we’re looking at a simple yes/no comparison. This will make the first step a little easier to follow. To keep things simple, I won’t go into all of the details of how the model works. Suffice it to say, we input the data and the criteria we’re interested in. Lights flash, hard drives whirr, robots wave their arms yelling, ‘Danger! Danger! Danger!’ and a piece of paper full of holes spits out and provides us with a lot of numbers. With a little work, we can turn those numbers into something that makes sense.
The model provides quite a bit of information. It tells us the odds for those without medical conditions, the odds for those with medical conditions, whether the variable (has a medical condition) is statistically significant, and what the range of outcomes should look like. First, we need to know if the variable even matters. The model provides two indicators. One is called a p-value and one is called a z-value. Generally, if the p-value is less than 0.05, the variable is considered significant. The z-value should typically be less than -3 or greater than 3 for the variable to matter. According to our model, the p value is effectively zero for both conditions. The z-value for no medical conditions is -132 and for medical conditions is 87. When it comes to predicting COVID outcomes, it appears that existing medical conditions matter a lot, since these values are well outside their respective ranges.
Those values tell us that existing conditions matter, but it doesn’t tell us how much they influence the outcome. The model provides numbers to help us understand that as well. For our data set, the model calculated that the odds of death for those without medical conditions was 18 to 1000. For every 18 people that had a positive test, no medical conditions, and died, a thousand people with a positive test and no medical conditions survived. This works out to a probability of dying of 1.75% for those without conditions.
For those with medical conditions, the odds jumped up to 272 to 1000 with a probability of 21.4%. Remember that in the previous post, the general odds were 159 to 1000. So these numbers seem to make sense. Having medical conditions make your odds of dying go up. Not having medical conditions makes your odds go down. The model also gives us a range (called a confidence interval). It tells us that for those with a medical condition, the odds of dying increase 14 to 16 times versus those without a medical condition.
Odds and probabilities of dying from COVID for those with and without medical conditions
Let’s also be clear on what this DOESN’T say. It doesn’t say that individuals died of their pre-existing conditions. EVERY SINGLE PERSON IN THIS DATA SET DIED OF COVID. The pre-existing conditions increased or decreased the likelihood, but they didn’t cause the outcome. COVID is the cause of death. Without COVID, they wouldn’t have died. Understanding the relationships helps us make predictions to guide medical and social decisions. Don’t confuse that with thinking that influencing is the same as causing. The data says that pre-existing medical conditions predict a higher likelihood of dying from COVID. They are still dying from COVID.
Also, before you get too excited, we still have a lot of work to do. Next, we’ll look at how sex, age, and race impact the likelihood of dying. Then we’ll look at everything together. When we look at everything together, these odds will change, because the other conditions influence the pre-existing conditions. Those interactions will change our model, and we will have to consider how they influence each other as well as how they influence the outcome.