Read on Twitter
Do non-steroidal anti-inflammatory drugs (NSAIDs) really increase the risk of adverse outcomes and death among patients who test positive for SARS-CoV-2?
A new observational, registry-based study in @PLOSMedicine suggests that it doesn't.
[Thread]
![Do non-steroidal anti-inflammatory drugs (NSAIDs) really increase the risk of adverse outcomes and death among patients who test positive for SARS-CoV-2?A new observational, registry-based study in @PLOSMedicine suggests that it doesn't.[Thread]](https://pbs.twimg.com/media/EhdpwrUXcAILM0I.jpg "Do non-steroidal anti-inflammatory drugs (NSAIDs) really increase the risk of adverse outcomes and death among patients who test positive for SARS-CoV-2?A new observational, registry-based study in @PLOSMedicine suggests that it doesn't.[Thread]")
A new observational, registry-based study in @PLOSMedicine suggests that it doesn't.
[Thread]
First things first, here is the article:
Lund LC, et al. (2020) Adverse outcomes and mortality in users of non-steroidal anti-inflammatory drugs who tested positive for SARS-CoV-2: A Danish nationwide cohort study. PLoS Med 17(9): e1003308 https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1003308
Lund LC, et al. (2020) Adverse outcomes and mortality in users of non-steroidal anti-inflammatory drugs who tested positive for SARS-CoV-2: A Danish nationwide cohort study. PLoS Med 17(9): e1003308 https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1003308
And before we dig into the paper, a quick and painless disclaimer: I know the last author (Anton Pottegård) and I like his work.
Side note: transparency is not only about financial ties to the industry...
Side note: transparency is not only about financial ties to the industry...
What was the question exactly?
Is the use of NSAIDs (e.g. ibuprofen) associated with adverse outcomes and mortality among people who test positive for the infectious agent of COVID-19 (SARS-CoV-2)?
Is the use of NSAIDs (e.g. ibuprofen) associated with adverse outcomes and mortality among people who test positive for the infectious agent of COVID-19 (SARS-CoV-2)?
Why on earth would researchers ask that question?
Well, I'm glad you asked. Because at the beginning of the pandemic, some concerns arose regarding the safety of NSAIDs and the possibility that these widely used drugs may negatively influence the prognosis of COVID-19 patients.
Well, I'm glad you asked. Because at the beginning of the pandemic, some concerns arose regarding the safety of NSAIDs and the possibility that these widely used drugs may negatively influence the prognosis of COVID-19 patients.
How did they try to verify this?
They conducted a cohort study in which they followed all Danish residents who tested positive for SARS-CoV-2 between 27 Feb–29 April 2020, and they compared two groups: those who had received NSAIDs in the 30 days before, and those who had not.
They conducted a cohort study in which they followed all Danish residents who tested positive for SARS-CoV-2 between 27 Feb–29 April 2020, and they compared two groups: those who had received NSAIDs in the 30 days before, and those who had not.
To do this, they used nationwide administrative and healthcare registries linked together.
The key is the the Danish Microbiology Database, in which every person who receives a PCR test for SARS-CoV-2 is registered.
https://www.dovepress.com/existing-data-sources-in-clinical-epidemiology-the-danish-covid-19-coh-peer-reviewed-article-CLEP
https://www.eurosurveillance.org/content/10.2807/1560-7917.ES2014.19.1.20667
The key is the the Danish Microbiology Database, in which every person who receives a PCR test for SARS-CoV-2 is registered.
https://www.dovepress.com/existing-data-sources-in-clinical-epidemiology-the-danish-covid-19-coh-peer-reviewed-article-CLEP https://www.eurosurveillance.org/content/10.2807/1560-7917.ES2014.19.1.20667
How did they differentiate NSAIDs users (treatment group) and non-users (control group)?
They looked at whether patients had filled a prescription for NSAIDs between 30 and 1 day before the PCR test. In Denmark, ibuprofen is available OTC but it accounts for only ~15% of sales.
They looked at whether patients had filled a prescription for NSAIDs between 30 and 1 day before the PCR test. In Denmark, ibuprofen is available OTC but it accounts for only ~15% of sales.
Got it, now what kind of outcomes did they look at?
Death within 30 days of a positive SARS-CoV-2 test.
Hospitalization, ICU admission, mechanical ventilation,
and acute renal replacement therapy within 14 days
So, pretty serious stuff.
Death within 30 days of a positive SARS-CoV-2 test.Hospitalization, ICU admission, mechanical ventilation,and acute renal replacement therapy within 14 days
So, pretty serious stuff.
Statistical analysis
People who had received NSAIDs before they took the SARS-CoV-2 test were not exactly comparable to those who had not. It's often the case.
To correct for this, the researchers used propensity score matching including a fairly broad array of confounders.
People who had received NSAIDs before they took the SARS-CoV-2 test were not exactly comparable to those who had not. It's often the case.
To correct for this, the researchers used propensity score matching including a fairly broad array of confounders.
Technical note:
Individuals in the tails of the propensity score distribution were trimmed, and matching was done 1:k (up to 4 controls) with a nearest neighbor approach.
Matching was forced on calendar week to account for the increasing probability of testing over time.
Individuals in the tails of the propensity score distribution were trimmed, and matching was done 1:k (up to 4 controls) with a nearest neighbor approach.
Matching was forced on calendar week to account for the increasing probability of testing over time.
I'm not gonna get into the geeky details, but let's say that the researchers really tried to cover the potential blindspots:
- NSAIDs exposure window extended to 60 days
- NSAIDs exposure window from 60 to 14 days before
- Outcome hazard period extended to 60 days after...
- NSAIDs exposure window extended to 60 days
- NSAIDs exposure window from 60 to 14 days before
- Outcome hazard period extended to 60 days after...
This is getting long, it's time to go to the exciting part: the results!
Entire cohort: 9,236 people who tested positive for SARS-CoV-2
– NSAIDs user: 248 (2.7%) | 224 after matching
– NSAIDs non-users: 8,988 (97.3%) | 896 after matching
For more, look at Table 1
Entire cohort: 9,236 people who tested positive for SARS-CoV-2
– NSAIDs user: 248 (2.7%) | 224 after matching
– NSAIDs non-users: 8,988 (97.3%) | 896 after matching
For more, look at Table 1
The main outcome: 30-day mortality
After matching:
– NSAID users: 6.3% (95% CI 3.1 to 9.4)
– NSAID non-users: 6.1% (95% CI 4.4 to 7.8)
– Risk ratio: 1.02 (95% CI 0.57 to 1.82)
– Risk difference: 0.1 (95% CI −3.5 to 3.7)
So, absolutely no difference.
After matching:– NSAID users: 6.3% (95% CI 3.1 to 9.4)
– NSAID non-users: 6.1% (95% CI 4.4 to 7.8)
– Risk ratio: 1.02 (95% CI 0.57 to 1.82)
– Risk difference: 0.1 (95% CI −3.5 to 3.7)
So, absolutely no difference.
Secondary outcomes (after matching):
Risk Ratio (95% CI)
Hospitalisation: 1.16 (0.87–1.53)
ICU admission: 1.04 (0.54–2.02)
Mechanical ventilation: 1.14 (0.56–2.30)
Renal replacement therapy: 0.86 (0.24–3.09)
Nada, zero, zilch. Nothing. At all.
Risk Ratio (95% CI)
Hospitalisation: 1.16 (0.87–1.53) ICU admission: 1.04 (0.54–2.02) Mechanical ventilation: 1.14 (0.56–2.30) Renal replacement therapy: 0.86 (0.24–3.09)Nada, zero, zilch. Nothing. At all.
Did these results change when the parameters (NSAIDs exposure window, follow-up period...) were changed?
Nope. It's all in Appendix Tables S2, S3 and S4, go and see for yourselves.
Nope. It's all in Appendix Tables S2, S3 and S4, go and see for yourselves.
The 1-million-dollar question: can we trust these results? 
I think we can. With caution and a healthy dose of scepticism, as always, but I think can.
Let's make a list of pros and cons...
I think we can. With caution and a healthy dose of scepticism, as always, but I think can.
Let's make a list of pros and cons...
The main pros:
1. Nationwide cohort including virtually every person living in Denmark who was tested positive
2. High-quality data
3. The propensity-score matching procedure did a pretty good job at balancing important confounders (with one exception)
1. Nationwide cohort including virtually every person living in Denmark who was tested positive
2. High-quality data
3. The propensity-score matching procedure did a pretty good job at balancing important confounders (with one exception)
Main cons:
1. Observational data, aka non-randomized allocation to the treatment groups. It is what it is...
2. Related #1: BMI was not available, yet it is a major risk factor for deteriorating COVID-19. Obesity and overweight were probably underestimated.
[...]
1. Observational data, aka non-randomized allocation to the treatment groups. It is what it is...
2. Related #1: BMI was not available, yet it is a major risk factor for deteriorating COVID-19. Obesity and overweight were probably underestimated.
[...]
3. Filling a prescription for NSAIDs is not the same as actually ingesting NSAIDs. If many of the 248 people who were considered as "NSAIDs users" were in fact not using the drug, the true negative effect of NSAIDs would be obscured.
4. The fact that concerns were raised early on (and widely publicized) opens the door for a pretty thorny risk of bias: physicians may have been even more selective than usual when prescribing NSAIDs, i.e. giving it only the fittest and healthiest people.
5. It is possible that NSAIDs were in fact prescribed to alleviate severe symptoms of COVID-19 *before* the patients were tested and known to have the disease. However, if that was true the researchers would have found a stronger association between NSAIDs and adverse outcomes
All these methodological considerations are valid, but it is—in my opinion—unlikely that they substantially change the results.
For instance, the sensitivity analyses show that moving the drug exposure window to 60–14 days before the positive test has absolutely no influence.
For instance, the sensitivity analyses show that moving the drug exposure window to 60–14 days before the positive test has absolutely no influence.
The authors have been extremely transparent, and that definitely adds to the credibility of their work.
The study protocol was pre-registered and is publicly available, deviations are explained, and post hoc analyses are clearly reported as such.
http://www.encepp.eu/encepp/viewResource.htm?id=34735
The study protocol was pre-registered and is publicly available, deviations are explained, and post hoc analyses are clearly reported as such.
http://www.encepp.eu/encepp/viewResource.htm?id=34735
What does this mean from a clinical point of view, then?
There is no reason not to prescribe NSAIDs for appropriate indications during the pandemic.
In the absence of data from randomized controlled trials, this study represented the highest quality of evidence that we have.
There is no reason not to prescribe NSAIDs for appropriate indications during the pandemic.In the absence of data from randomized controlled trials, this study represented the highest quality of evidence that we have.
Regardless of COVID-19, NSAIDs do have serious side effects and should thus be used with caution, at the lowest possible dose and for the shortest possible duration.But again, this has nothing to do with COVID-19.
So, let's close this thread with a one-sentence recap:
In an observational cohort study of 9,236 Danish residents who tested positive for SARS-CoV-2, the use of NSAIDs was not associated with an increased risk of being hospitalized, admitted to the intensive care unit, or dying.
In an observational cohort study of 9,236 Danish residents who tested positive for SARS-CoV-2, the use of NSAIDs was not associated with an increased risk of being hospitalized, admitted to the intensive care unit, or dying.
One last point, which I should have mentioned at the very beginning: here is the disclosure of competing interests declared by the authors.
Again, transparency matters.
Again, transparency matters.
Afterthought #1:
Some of you have questions about the relatively low prevalence of NSAIDs use in the study population (only 2.7% of individuals who tested positive; i.e. 248 / 9236).
That actually fits the prescription patterns and sales data pretty well
https://accpjournals.onlinelibrary.wiley.com/doi/abs/10.1002/phar.2217
Some of you have questions about the relatively low prevalence of NSAIDs use in the study population (only 2.7% of individuals who tested positive; i.e. 248 / 9236).
That actually fits the prescription patterns and sales data pretty well
https://accpjournals.onlinelibrary.wiley.com/doi/abs/10.1002/phar.2217
