Covid-19: Comorbidity

We’ve all seen statistics that most people who die of Covid-19 have at least one comorbidity. They also almost all have the particular comorbidity of age. The biggest risk, by far, is being old. The question that I don’t see being properly asked anywhere (I’d love for this post to be unnecessary because there’s a better one) is: What is your chance of death from Covid-19 if infected, conditional on which if any comorbidities you have? Which ones matter and how much? If you don’t have any, how much better off are you than your age group in general?

Thus, most people are looking at the age chart, without adjusting for their health status, unless they have an obvious big issue, in which case they adjust up. Which leads to an incorrect overall answer. That isn’t obviously a bad thing in terms of resulting behavior in practice, but that doesn’t mean we shouldn’t attempt to figure out the answer.

I used New York State’s information on deaths to get the rates of most of the major comorbidity candidates by age, and various Google-fu combined with wild mass approximation and fitting to different age groupings (not guessing, but not fully not guessing either) to get approximate population prevalence data. What matters?

A key question will always be, is this a proxy for something else, such as poverty, general poor health or obesity? Or is it the real problem? Here, we need to use some common sense and physical intuition. This isn’t attempting to be super rigorous, but rather to get an approximation.

Then, once we’ve looked at all of them, I’ll attempt to put them all together, and solve for the risk of someone in good overall health.

Spreadsheet is here, you can look at the numbers in somewhat more detail, and see some of the sources I used, on the Comorbidity tab.

In each case, the “Population X” column attempts to guess the rate at which the population has it. The morbidity column is the rate at which those in NY state that died of Covid-19 had it.

For the first few graphs I fit NY’s data to the groupings in the prevalence data I found. Later I started always using NY’s ranges instead.

Hypertension

Morbidity Population
Age Hypertension Hypertension
18-39 21% 8%
40-59 43% 33%
60+ 61% 63%

Hypertension in the young seems to matter. If you are 18-39, your relative chance of dying more than doubles. In your 40s and 50s combined, it’s a jump of about a third. Above age of 60, it does not matter. Should we be suspicious that this is a proxy for poor health, given that? Somewhat, definitely. It will be a recurring pattern, which we’ll need to get to make sense.

Diabetes

Morbidity Population
Age Diabetes Diabetes
18-44 28% 4%
45-64 40% 17%
65-74 44% 25%
75+ 34% 25%

Clearly this is a huge deal at young ages. I still don’t really believe the 4% number for the general population, but multiple sources are around there. That 4% of the population is more than a quarter of all deaths under 45 in New York. That’s a huge deal. Diabetes is a huge deal the entire way. This makes some sense, as it seems likely to correlate with slash cause direct physical problems for those with Covid-19 that can kill them. But this amount of effect is still surprisingly extreme.

Hyperlipidemia

Morbidity Population
Age Hyperlipidemia Hyperlipidemia
0-4 0% 7%?
5-14 0% could
15-29 4% be
30-39 5% up
40-49 9% to
50-59 16% 45%
60-69 22% in
70-79 25% adult
80-89 24% pop?
90+ 20% so
Unknown 6% meaningless

I could not get population numbers, because no one can agree on what Hyperlipidemia actually is and is not. By some definitions, almost half the adult population has it, because people like to say that we need medications and are “at risk” and turn everything into a disease. By others, it’s single digit percentages. So these morbidity numbers could be anything from scary high to same as the population, depending on the definition used, and I don’t know what that was. If anyone does know, please tell me.

But given what this physically is, any link seems more like correlation than causation, and the numbers listed seem plausibly like general population rates anyway, so I’m going to say this likely doesn’t matter.

Coronary Artery Disease

Morbidity Population
Age C. Artery D. C. Artery D.
0-4 0% 0%
5-14 0% 0%
15-29 0% 0%
30-39 0% 1%
40-49 3% 4%
50-59 6% 8%
60-69 11% 13%
70-79 14% 18%
80-89 16% 25%
90+ 13% 26%

The population rates I approximated are modestly higher across the board. Probably there’s no effect and this is a measurement error.

Dementia

Morbidity Population
Age Dementia Dementia
0-4 0% 0%
5-14 0% 0%
15-29 0% 0%
30-39 0% 0%
40-49 0% 0%
50-59 1% 0%
60-69 4% 1%
70-79 10% 5%
80-89 18% 24%
90+ 28% 37%

It’s weird that this reverses at older ages, probably because of measurement, but perhaps because people with dementia have overall better health at that age due to the ones with poor health having died more often? Whereas in younger people, if you have dementia things are much more likely to be generally terrible in other ways instead?

From the 50-79 year old data I’d have said this might matter, but the prevalence rate is low there, and where the rates are high, the numbers are reversed. I don’t think dementia is doing work here.

Renal Failure

Morbidity Population
Age Renal Failure Renal Failure
18-44 6% 7%
45-64 11% 12%
65+ 11% 37%

I suppose that those with renal failure die soon thereafter, thus the 37% population rate is not going to be reflected in an alive group. It’s certainly not going to be protective. That implies there might be some real effect in the younger groups if you squint hard enough, but seems much more likely this is not a risk factor.

COPD

Morbidity Population
Age COPD COPD
0-4 0% 0%
5-14 0% 0%
15-29 1% 2%
30-39 1% 3%
40-49 2% 5%
50-59 5% 7%
60-69 8% 8%
70-79 10% 10%
80-89 10% 9%
90+ 8% 8%

Very clear no effect.

Atrial Fibrulation

Morbidity Population
Age Atrial Fibrulation Atrial Fibrulation
0-4 0% 0%
5-14 0% 0%
15-29 0% 0%
30-39 1% 1%
40-49 1% 1%
50-59 2% 1%
60-69 4% 2%
70-79 8% 5%
80-89 12% 9%
90+ 14% 11%

Some effect, with minimal impact on numbers for those without the condition, and likely the effect is correlational given the physical conditions involved.

Cancer

Morbidity
Age Cancer
0-4 0%
5-14 0%
15-29 3%
30-39 2%
40-49 2%
50-59 4%
60-69 7%
70-79 8%
80-89 9%
90+ 8%

About 0.5% of New Yorkers get newly diagnosed with Cancer every year. But translating that into a background cancer rate by age proved very difficult. It certainly can’t be what they mean by cancer here, since that means that at older ages cancer would be highly protective, so they’re clearly only counting current conditions or some other similar thing.

It seems obvious that being actively sick from cancer treatment would make Covid-19 much worse to get, but it’s not at all obvious the condition itself would matter much for many cancers, and they’re also different conditions, so it’s all very confusing. Not sure what to do here.

My guess is this is mostly ‘general poor health caused by cancer or treatment’ effects, to extent it matters.

Stroke

Morbidity Population
Age Stroke Stroke
0-4 0% 0%
5-14 0% 0%
15-29 1% 0%
30-39 1% 1%
40-49 3% 1%
50-59 4% 2%
60-69 7% 5%
70-79 8% 10%
80-89 8% 15%
90+ 6% 15%

Again, those numbers on the right are largely guesses. It does seem clear that stroke is a risk factor when you are young. We once again see strangely low morbidity rates for the older age groups, pointing to a likely general difference in methodology. Probably selection effects.

That’s everything listed by New York.

Obesity

When you are young, it seems like it matters a lot whether there is something relevantly seriously wrong. But it has to be something that matters. Having a health problem in an area that Covid-19 doesn’t attack does not seem to matter.

Two of the problems measured matter a lot. Hypertension and Diabetes together are about 12% of the population and constitute ~50% of the deaths up to age 40-45. The rest don’t seem to matter much at all, and you could safely ignore them.

It is safe to assume that anyone with serious trouble breathing for other reasons is going to have similar problems. Thus, asthma, obesity and so on are also (probably) serious risk factors.

One source I found was this one, which notes that 35.8% of hospital patients with Covid-19 were obese. In the city, 22% of the population is obese (and a majority are at least overweight). 43% of the invasive treatment group, which was presumably in far worse shape, was obese versus 31% for the non-invasive group, so it seems that the extra risk carries over to outcomes after hospitalization.

This source found an inverse correlation in Covid-19 patients between age and BMI, which would also make sense.

From another source at ScienceNews:

For instance, of 180 patients hospitalized from March 1 to March 30, the most prevalent underlying condition for adults ages 18 to 49 was obesity. Of 39 patients in that age range, 23, or 59 percent, were obese, researchers report in the April 17 Morbidity and Mortality Weekly Report.

Lighter and her colleagues found that patients under 60 with a BMI over 35 were at least twice as likely to be admitted to the ICU for coronavirus than patients with healthy BMIs, the researchers report April 9 in Clinical Infectious Diseases. Those same patients were three times more likely to die from the infection than those with a lower BMI, she says.

The team tracked 3,615 people who tested positive for SARS-CoV-2, the virus that causes COVID-19, at a New York City hospital from March 4 to April 4. Of those, 1,370, or 38 percent, were obese. In patients over 60, weight did not appear to be a factor in hospital admission or the need for intensive care, she says.

Again, it seems like ‘nothing matters much if you’re old.’ But if you’re young, things do matter.

Let’s say for the time being that obesity triples your risk if you’re under 60. Obviously, this doesn’t go away the moment you turn 60, so we’ll want to do more than triple for those under 40, much less than triple for those in their 50s, and some effect probably in your early 60s.

Continuing to do approximations, if we accept the figure that 52% of Type II Diabetics were obese back in 2006, whereas Type I is actually lower than the population rate, which is now more like 40%. New York City’s seemingly terrible 22% looks positively amazing by comparison if actually the same measure.

This accounts for substantial increased risk for diabetics, but the majority clearly remains unexplained by weight. Obesity alone would have gotten us from 4% to about 10% in the young group, and we ended up at 28%.

Hypertension, on the other hand, seems mostly to be a proxy for obesity, so we can mostly ignore it.

Overall, though, obesity seems by far the most important consideration other than age, since it’s so common and has such a huge impact.

Age Alone

To adjust from a baseline we need a baseline. For New York the relative risks look like this:

%Of Deaths Population Relative Risk
0-4 0.02% 7% 0.2%
5-14 0.04% 14% 0.3%
15-29 0.3% 22% 2%
30-39 1% 16% 9%
40-49 4% 14% 26%
50-59 10% 9% 116%
60-69 20% 7% 287%
70-79 27% 5% 525%
80-89 25% 2% 1066%
90+ 12% 1%? (cuts off at 85) 1500% or so?

 

Then we must guess the true IFR (infection fatality rate), and adjust for the three comorbidities that we’ve found matter: Obesity and diabetes. They still correlate.

We also have to adjust for the likelihood of infection in the first place, since that changes your risk conditional on infection. This is a relatively small effect according to antibody test results, and should at least sort of be cancelled out in some ways for practical purposes, so I’m not going to worry much about it.

Most coronavirus cases are definitely not being detected by positive tests. The antibody tests show that. So the IFR is much lower than the CFR. Given death rates and antibody tests, the plausible range for IFRs is about 0.5% to 1.5%. It will also depend on conditions on the ground in various ways, of course. But as a baseline, I’m going to continue to say 1% death rate for the state. If you disagree, multiply all the numbers I get as appropriate.

Obesity gets slightly more common with age, which I’ll adjust for.

For the younger groupings, we have 4% Diabetes and (in New York) 25% Obese. Obesity dominates that group by size, but diabetes still matters. Together that’s about 27% of the population. The 4% of that that are diabetic account for 28% of the cases. The other 23% that are obese become 49% of the remaining cases, or 35% of all cases. Add that together, and that’s 63% of cases from this 27%, with the remaining 37% coming from the other 73% of the population. So if you’re healthy, with healthy defined as ‘not obese and not having diabetes’ your risk is cut roughly in half when young. There’s too many error bars all over the place, so I don’t want to try and be more exact than that.

Other conditions doubtless also matter somewhat. 10% of New York has asthma, which presumably makes a big difference, but all I could find was “may be at higher risk” repeated over and over, rather than any numbers.

This all these effects decay as you age. By age 70, there’s little or no difference.

But roughly you end up with a chart that looks something vaguely like this:

Reminder: This assumes overall infection fatality rate of ~1% and is full of guesses and approximations as all hell:

Age Risk (Healthy) Risk (Diabetes) Risk (Obesity) Risk (All Pop)
0-4 0.001% 0.013% 0.005% 0.002%
5-14 0.002% 0.016% 0.006% 0.003%
15-29 0.008% 0.078% 0.031% 0.015%
30-39 0.043% 0.26% 0.17% 0.09%
40-49 0.13% 0.53% 0.43% 0.26%
50-59 0.72% 1.8% 1.4% 1.2%
60-69 2.0% 4.0% 3.0% 2.9%
70-79 4.3% 6.5% 5.4% 5.2%
80-89 11% 11% 11% 11%
90+ 15% 15% 15% 15%

 

This entry was posted in Uncategorized. Bookmark the permalink.

16 Responses to Covid-19: Comorbidity

  1. myst_05 says:

    Is it possible to calculate hospitalisation rates per each age group too? That should be a good proxy for how many people could be expected to develop long term side effects, in the absence of follow up studies on recovered patients.

    It would also be interesting to calculate the effectiveness of hospital treatment per each age group. E.g. what percentage of 80 year olds are saved if they get admitted to the hospital.

    • TheZvi says:

      The problem with hospitalization rates is that they are a function of decisions made by humans that vary greatly from place to place with local conditions and philosophy. But they do track this in various places. Also, obviously hospitalization is not one thing, especially when there are shortages / overwhleming.

      The whole philosophy here is that one should figure it out anyway, of course.

      Whether or not one can do it with comorbidities will depend on the data reporting.

  2. sniffnoy says:

    Note that COVID-19 may be causing strokes…

  3. Erasmus says:

    I liked the observation that some patients with dementia may have “died more often”. That is the population we need to study. 🙂
    But seriously, are these co-morbidities similar to those who die of flu? Is the age distribution similar for flu? (Not the 1918 flu, but the last 30 years of USA seasonal flu.)
    It would also be interesting to look at, if possible, the percentage of people with co-morbidities of either dementia and/or of “being over 80 years old”, who were ALSO nursing home residents. Indeed, being
    resident in a nursing home appears to be the most significant co-morbidity of all.
    Finally, we should ask “Of what do patients actually die? The presence of the virus in the body is not a cause of death. Rather, die from Covid through several different pathways. You die from (a) a cytokine “storm”, i.e., an overreacting immune response; or
    (b) ARDS (acute respiratory disease syndrome, i.e., pulmonary failure) and (c) sometimes from the side effects of intubation, which, as you might imagine is very hard on the bodies of those over 70.
    Without breaking down the data for the actual proximate cause of death, the investigation of age cohorts and co-morbidities can only be partially illuminating. I don’t think the data exists for studying this, at least at present.

    (Btw, amazing blog, just discovered it.)

    • Ben Albahari says:

      So potentially the disproportionate number of deaths for those with preconditions are due to contagion within nursing homes. Where the correlation between people with preconditions and people who are in nursing homes cancels out the correlation between preconditions and mortality.

      I did a quick search on http://biomed-sanity.com/ but couldn’t find anything to support this hypothesis, but it seems plausible.

  4. Ben Albahari says:

    Hi Zvi,

    A couple of months ago I wrote a covid-19 risk calculator that’s gotten some press, and even translated into Spanish. Here’s the link:

    https://www.solenya.org/coronavirus

    I’ve updated the calculations to leverage your table for age & preconditions, which were better than what I had. Having said this, the fact preconditions now have no bearing on the outcome of 80+ year olds seem to me a regression on what I had.

    Here’s Swedish data, which shows death rates for 80+ year olds double for those with 1 precondition, then double again for 2 or more preconditions:
    https://www.socialstyrelsen.se/statistik-och-data/statistik/statistik-om-covid-19/statistik-over-antal-avlidna-i-covid-19/

    Any idea how to reconcile this?

    Thanks,
    Ben

    • TheZvi says:

      Thanks!

      Best idea I’ve heard on this was that comorbidities correlate with nursing homes, and nursing homes are super dangerous versus being at a regular home, so the samples were skewed, but not clear that this explains anything. Certainly can’t explain double and then double again, nothing like that is anywhere in my data and it would imply *huge* death rates for those groups. It also would have a huge effect on those without one, since at that age the comorbidities are very common.

      • Ben Albahari says:

        Let’s assume then that your current model needs revision.

        * Can you use the Swedish data to update your analysis?
        * Is there something in your model you think might have caused this discrepancy?
        * Is there something skewed in the data you were using?
        * What key information would you need to resolve this – if so, let me know and I’ll try to find relevant articles.

        It’s strange that neither of us have been able to find an article by an epidemiologist that directly answers the risk profile question. I wonder if this indicative of genuine difficulty in the analysis, a lack of perceived value in the question, or our inability to search journals in the field. Or maybe some mixture of those factors.

      • TheZvi says:

        I think it’s that the idea of doing something useful to people in practice, letting people know where they stand, is seen as not useful, therefore no one tries? Actual guess as to why no one’s doing it. It’s not that hard.

        Certainly my model can be improved. But the ‘comorbidities stop mattering as people age’ thing in NY seems very strong. So we’d need to figure out what was up with that. We’d also want a second similarly strong data set. The Swedish data might be a good next place to go, but do you have a place to access it in English? The Google Translate button isn’t working for some reason and I can’t parse it well enough to use it.

      • Ben Albahari says:

        Here’s the direct link to the spreadsheet – some names in Swedish are quite close to English counterparts so it’s not that hard to figure out the meaning:
        https://www.socialstyrelsen.se/globalassets/sharepoint-dokument/dokument-webb/statistik/statistik-covid19-avlidna.xlsx

        Interestingly, their detailed data on sheet #4 has a strange age granularity: “Under 70”, “70-74″,”75-79″,80-84″,”85+”. The breakdown is useful in terms of covering gender and varying number of comorbidities, but is weak in terms of the range of comborbidities covered, but maybe that’s ok. Let me know if it’s suitable/usable.

      • Ben Albahari says:

        Possibly useful Mexican data (don’t worry it’s not in Spanish):

        Click to access 2020.05.03.20089813v1.full.pdf

  5. Ryan Ferris says:

    Asthma is separated out from COPD in this MI study: https://www.medrxiv.org/content/10.1101/2020.05.04.20074609v1 Is it not separated out in the NY data? Or is Asthma just not a comorbidity to Covid-19 in NY? (That would be weird…) 25M Americans have Asthma, a debilitating upper respiratory disease. Yet, asthma itself doesn’t seem to be a very strong comorbidity to Covid-19 which apparently starts as an upper respiratory infection.

  6. TheZvi says:

    There was no distinct listing for Asthma. Certainly people who have it are assuming they are at higher risk and I wouldn’t tell them not to without futher analysis.

  7. Pingback: Covid-19: My Current Model | Don't Worry About the Vase

  8. Erasmus says:

    My conclusion is that the one strong correlation — age — is simply a reflection of the fact that many many Americans over the age of 85 are…in fact…nursing home residents. Or hospitalized patients who came from nursing homes. Last I saw, 40% of the deaths in the US are nursing home patients. In some states like Minnesota its like, 80 % or more. In short, the
    the “co-morbidity” is being a nursing home resident. I don’t know if any of you have visited a nursing home but if you have….you should not be surprised. I would be very interested to see if the data can be broken down and we can what the age-85 plus mortality rate is among nursing home patients versus non-residents. Given the politics around whether some governors missed the ball on this, I doubt these statistics are available. Can we all agree they would be interesting if they were?

  9. Pingback: Covid 7/9: Lies, Damn Lies and Death Rates | Don't Worry About the Vase

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s