Time Stamps

  • 03:33 The art of diagnosing the flu
  • 05:20 Bayesian statistics and the diagnostic odds ratio
  • 07:07 The ins and outs of diagnostic testing for the flu
  • 08:40 Changes in influenza guidelines – testing and treatment
  • 11:22 The not-so-rosy origin story of neuraminidase inhibitors (NAIs)
  • 13:48 Neuraminidase inhibitors (NAIs), mortality benefit and limitations of studies
  • 15:53 Take aways

Show Notes

  • There are many influenza symptoms and most of them are not very helpful in confirming a case of the influenza. They have poor diagnostic odds ratios, which is defined as the ratio of the positive likelihood ratio over the negative likelihood ratio.
  • To confirm a case, the IDSA recommends an influenza PCR
  • The influenza PCR can still have false positives. As with any test, apply the test to your pre-test probability – a number you come up with based on your local context – to get your post-test probability – the chance a patient truly has disease.
  •  The new 2018 influenza guidelines recommend:  
    • Treat any inpatient with suspected influenza, even without a confirmatory test, and even after they’ve had symptoms for 48 hours.
    • Consider testing outpatients if it will change management.
    • Consider treating outpatients who are likely to suffer complications from the flu, even without a confirmatory test
  • Neuraminidase inhibitors, such as oseltamivir and zanamavir, were originally thought to decrease symptoms by 24 hours, and reduce complications and hospitalizations
  • Data from the 2009 influenza pandemic, when neuraminidase inhibitors were used broadly, generated new data which actually suggested a mortality benefit
    • However, this data comes from observational studies, rather than randomized control trials, so may be subject to bias.
  • Ultimately, neuraminidase inhibitors are the best and only option we currently have to treat the flu. This is why, despite their known side effects and questionable outcomes, IDSA guidelines have changed to recommend using them more broadly.


In part 2 of our flu series, we explore the ins and outs of influenza testing and the complicated history of influenza treatment, just in time for flu season!

Today we’re going to tell you a dark and scary story.

S: It may be just after the Holidays, but we’re gonna take you back to Halloween.

J: Spooky.

S: It’s a frightening tall tales where misinformation is rife and the truth is as terrifying as an episode of “Are You Afraid of the Dark.”

J: And with as many plot twists as an episode of Goosebumps.

S: Today we’re going to take a deep dive into a white hot viral upper respiratory infection.

J: Influenza.

S: What did you think we were gonna say? Coronavirus?

J: After all, we promised to have a follow up from our last flu episode, and flu season is in full swing.

S: And this one is turning into a bit of a doozie. But if you do wanna hear about Coronavirus stay tuned for our interview with Dr. Fiske.

J: Unlike previous years, Flu B is as common then Flu A this year, making up the majority of cases so far.

S: And CDC data suggests an earlier start to the flu season then previously.

J: So hopefully you got your flu vaccine since listening to our last episode!

S: But now it’s time to get into what do you do once your patient gets the flu.

J: The recommendations are NEW as of last year.

S: Hot off the press!

J: Yes there are some major changes. For those of you holding onto that 48 hour rule about starting oseltamivir, hold your horses – let’s take a deep dive to keep you up to date.

S: Here’s what we’ll cover today: We’ll start off by figuring out whether the person you’re seeing really has the flu. 

J: More specifically, Steve will nerd out about diagnostic odds ratio to help tease apart which symptoms are more predictive of the flu.

S: Then we’ll review how good are the diagnostic tests for the flu and how they can guide your pre-test probability.

J: And once you’ve made the diagnosis, we’ll review what medications we have for the flu and whether to use them.

S: We’ll cover management differences in the inpatient vs. outpatient settings.

J: And since there are new guidelines, we’ll go over how and why they changed. And how reliable is the evidence supporting them?

S: This is going to take us back in time again, but unlike last time, we won’t be going back nearly a century.

J: This time, let’s see how far we’ve come in the past decade.

We’d like to thank Dr. Denise McCullough, esteemed ID fellow at the University of Washington, and Dr. Jen Spicer, an assistant professor of Infectious disease at Emory University, for peer revieweing this episode.

S: The other night I was actually watching Downton Abbey, and it happened to be the Christmas episode where Matthew’s fiancée Lavinia dies of the Spanish Flu.

J: Gotta love that English Melodrama, Steve.

S: Period pieces are my weakness. Well I couldn’t help but notice that the doc on the show goes around diagnosing people left and right with the flu.

J: What’s wrong with that? 

S: How does he really know they have the flu? It’s not like they had PCR back then.

J: Omg Steve, what a stickler. They didn’t have procalcitonins either!

S: To muddy the waters, did you know that procalcitonin may not actually be as good as we once thought for distinguishing between viral and bacterial pneumonias?

J: We digress, Steve! Let’s get back to the flu, how were they diagnosing it?

S: So let’s start with diagnosis and roll back to the 19th century when they relied purely on symptoms to diagnose the flu. Here’s a classic description from our last episode:

J: “The most striking symptoms are: sudden onset with chills, severe headache with pain… in limbs and general malaise… the maximum temperature was 103 to 104… Many cases develop a cough, harsh in nature with a thick sputum.”

S: That sounds like literally any upper respiratory infection I’ve ever had… but it’s not that different from the IDSA guidelines. They list similar symptoms.

J: So do some of these symptoms predict the flu better than others?

S: Well some do, mainly fever and cough.

J: And how good are they?

S: The answer is well, not really. Let me explain…

J: Uh oh, are you about to talk statistics? Like positive predictive values?

S: Haha yes Janine, stats are unavoidable when you’re talking about diagnostics. But today we’re going to skip over predictive values…

J: Why’s that?

S: Because PPVs, as they’re called, are dependent on disease prevalence. And we want to know how predictive symptoms are regardless of prevalence. 

J: Exactly! And remember, a positive likelihood ratio should be greater than 1 to help with your diagnosis — really good LRs are usually 5 or above.

S: And a negative likelihood ratio should be between 0 and 1 – an LR of 0.5 is fine, but 0.2 and below are better.

J: So fever carries a positive likelihood ratio of around 2 for flu and cough has a negative likelihood ratios of around 0.5

S: That’s not great. Is it the best we’ve got? 

J: Unfortunately yes – and this data comes from my favorite series: the JAMA Rational Clinical Exam series.

S: Well, looking at that, if you have fever, cough, AND acute onset, your positive likelihood ratio goes up to 5.4.

J: Still not great! But the highlight of that article is that sneezing has a negative likelihood ratio of 0.47!

S: Amazing, I’m adding that to my review of systems!

J: Ok, well that article also brings up something called a “diagnostic odds ratio”.

S: That’s my favorite! Janine I’m a simple guy. And sometimes it’s nice to have only 1 number to describe the accuracy of a test, rather than trying 2. The diagnostic odds ratio divides the positive LR and the negative LR, y’know a ratio.

J: In layperson language, the diagnostic odds ratio gives you the odds of having the symptom among individuals with disease compared with the odds of having the symptom among those without disease

S: And just like an LR, the higher the number, the better the test.

J: If you end up close to 1, your test isn’t that great. Numbers between 0 and 1 suggest that you’re not asking the right question…

S: Now back to the flu … from the JAMA rational clinical exam article, fever has a diagnostic odds ratio of 4.5. For cough it’s a little lower at 2.8.

J: So what do we do with that? Clearly not everyone with a fever has the flu!

S: True. All the stats we just covered are a way of thinking about the fundamental principles of Bayesian probability.  When we say that we’re trying to rule something in or out, what we actually mean is that using Bayesian logic, we’re trying to either increase or decrease the probability that we think a patient has a diagnosis.

J: Before you apply your diagnostic test, you have to define your pre-test probability

S: And that’s influenced by the prevalence of flu this year, whether the patient was exposed to people with the flu, or whether they’re immunocompromised.

J: Then apply your test – in this case checking a patient’s temperature or asking if they have a cough – and based on your positive likelihood ratio you get a post-test probability.

S: If your test isn’t that good, then your post-test probability may not be much different, so your test hasn’t helped you very much. And that’s what we’re seeing with these flu symptoms.

J: Exactly. If your pre-test probability is high, it may not matter too much, but if your pre-test probability is medium-low, you may want to do additional testing.

S: That’s right. Now what additional testing you may ask? There are three ways to detect influenza, that’s using either DNA amplification (aka PCR), antigen detection, or culture.

J: We rarely use viral cultures because they take a long time.

S: And the antigen test is ok, but PCR is effectively the gold standard for diagnosis of influenza. It has the highest sensitivity and specificity.

J: In fact, the latest IDSA guidelines recommend the PCR over the antigen test in all situations. We’ll explain why:

S: While PCR and antigen testing have the same high specificity, antigen testing is much less sensitive. 

J: So it’s not great in populations where it would be bad to miss the flu, like sick hospitalized patients.

S: Another way to say this is that PCR has a higher positive likelihood ratio compared to antigen testing.

J: So putting all together, even though both tests have the same specificity, the improved sensitivity of PCR doubles the likelihood in favor of a diagnosis.

S: And tying this back to our explanation of diagnostic odds ratios, if we calculate it for PCR it sits nearer to 100.

J: Which is much much higher than fever and cough, which if you remember were more in the 5 range.

S: But again, all these numbers are meant to help you get from a pre-test probability to a post test probability, so you can decide whether to treat your patient for flu or not.

J: Bringing us back to reality, Steve.

S: And that depends on the prevalence of flu in your population.

J: After all, if there’s no flu around you, the PCR test may be giving you a false positive.

S: Likewise, when flu is common

J: Like now

S: A negative result might come from either a false negative, which could be the test or just poor collection technique.

J: So with that caveat in mind, who does the IDSA recommend testing?

S: They say in hospitalized patients, pretty much everyone should be tested.

J: And when we say all patients we mean all patients.

S: This includes obvious cases, like people coming in with acute respiratory illness, with or without fever.

J: To less obvious cases, like people who coming in with worsening dyspnea from COPD or heart failure. Or even just people who are immunosuppressed.

S: And what about the outpatient world? Test everyone with acute respiratory symptoms?

J: Not exactly. Only test if it changes clinic management. There are some cases where it would be reasonable to treat empirically for the flu, like if your pretest probability is really high or your patient is really sick. 

S: And some cases where you wouldn’t treat even if they have it, like if they’re not that sick.

J: And when we say treatment, it’s not just tamiflu, aka oseltamavir, anymore! There are now a number of other neuraminidase inhibitors like peramivir, zanamivir, and possibly hitting pharmacies near you, laninamivir. The formulations are oral, nasal, or IV. In 2018, the FDA approved Baloxavir, which is a cap-dependent endonuclease inhibitor – a novel mechanism for anti-influenza medications. 

S: I totally understood what that means.

J: Another thing that’s changed with the new flu guidelines is when to treat. Remember the 48 hour cutoff?

S: You mean, if their symptoms started more than 48 hours ago, it’s not worth starting treatment?

J: You can still use that rule if someone is outpatient and not that sick.

S: But otherwise the IDSA recommends that all people who are hospitalized with suspected or proven flu should be treated with antiviral treatment as soon as possible, regardless of when symptoms started.

J: And all outpatients with severe or progressive illness, or are at high risk from developing complications should be started on antivirals.

S: We’re talking anyone with a chronic medical problems (like COPD, HF, etc), an immunocompromised state, also folks over the age of 65, infants younger than 2, and pregnant women up until 2 weeks post-partum.

J: This is irrespective of vaccination history. Also consider treating people who will be near other folks that are at high risk for complications.

S: So, basically everyone.

J: LOLs they definitely broadened their treatment recommendations, but it’s still not everyone!

S: Well, the guideline to treat all inpatients with suspected flu is a pretty big change. 

J: Yeah – the old 2009 guidelines used to say that for admitted patients you should only start medication after the 48 hour cut off if it’s proven with testing. So we are a lot more liberal with treatment now.

S: And to explain how that happened, we have to do another historical dive – this one inspired by an amazing collaboration between the British Medical Journal and Cochrane to uncover previously hidden flu treatment data.

J: Sounds nefarious. The title is so good: Multisystem failure: the story of anti-influenza drugs.

S: Yes, (spoken in a spooky voice) this dark story takes us back to 1999, when the first neuraminidase inhibitors were approved – oseltamavir and zanamivir.

J: Initial data suggested that those anti-virals were effective at reducing the duration of symptoms by a whole 24 hours. They also thought these drugs might reduce the risk of complications and hospitalization.

S: A 2005 Cochrane review suggested that use of NAIs decreased the risk of bronchitis and pneumonia, going so far as to say that “In the case of a serious localised confirmed epidemic, NAIs could be used to prevent serious complications.”

J: This data, and other papers, led to oseltamir and zanamavir being stockpiled prior to the 2009 H1N1 pandemic.

S: But just as the 2009 flu season was about to hit, a new Cochrane review came out questioning these results, saying: news flash! “Neuraminidase inhibitors might be regarded as optional!! Paucity of good data has undermined previous findings for oseltamivir’s prevention of complications from influenza.”


Mic dropped bitches.

J: The folks at Cochrane were particularly concerned because some of the most promising data supporting the use of oseltamavir had come under suspicion. Roche, the manufacturer of oseltamavir, had refused to share their study data publicly, as is normally required, and were wishy washy about their results. 

S: To make matters worse, the Cochrane group read up on the FDA files on zanamivir, which suggested that the drug was approved not because of its superior performance but because it offered “an alternative therapeutic approach for an important public health problem” in a market where “current influenza treatment options [were] limited.”

J: Yikes! So it wasn’t actually better, it was just the only option? Thousands on thousands of people were treated with these drugs during the 2009 pandemic…

S: Hashtag Drama. And we’re just scratching the surface. 

J: Check out our show notes for some fascinating reviews on what this whole scandal meant for the pharmaceutical industry.

S: A widely used drug clouded by potential misinformation or at least lack of information from a global pharmaceutical company….It’s almost a bad Hollywood movie. I feel like a conspiracy theorist.

J: Well yes things from here take a turn towards the unexpected.

S: Things get worse?

J: Nope! Following the 2009 pandemic, multiple large meta-analyses actually showed a mortality benefit.

S: Ah my movie.

J: And not just kind of a benefit, there were large trials suggesting effects up to a 75% reduction in mortality with taking an NAI, with reductions in complications and hospitalizations as well.

S: And while the outcomes were better when treatment was started early, the benefits of the antivirals were still present even if the start was after 48 hours of symptoms.

J: So the drugs were better than we thought after all. And this is why the IDSA guidelines changed so much from 2009 to 2018 to recommend broader treatment. 

S: That said, this data is not without its critiques. Most importantly, all the new data is observational – none of it comes from RCTs, especially for sick inpatients. 

J: Remember this new data comes from the heart of the 2009 pandemic where folks were desperate to treat people, so people were not exactly lining up to design randomized controlled trials.

S: And the observational studies were at risk of the usual types of bias present in case control studies, most importantly immortality bias.

J: In the case of flu treatment, this bias highlights the fact that in order to make it into the case group you must have survived until the time where the flu was detected and then it was decided to include you into the study.

S: When this bias is adjusted for, the effects of NAIs may be less profound.

J: We’re not bringing this up to shake your faith in treating folks with NAIs, but rather to highlight that this has been a contentious debate for some time

S: That is kind of the point of this segment: that there are things we accept as true that may not have as firm a grounding as we’d like to say.

J: But remember, the fact that is in spite of imperfect data for NAIs, the recommendation for treating sick and vulnerable patients with them is pretty strong (A-II, A-III), 

S: And that although it’s an imperfect medication, so far it’s the best widely available drug we have for these patients, particularly for groups at high risk of morbidity and mortality from flu.

And with that, it’s a wrap. Let’s summarize:

  1. First of all, there are many flu symptoms and most of them are not very helpful in confirming a case of the flu. They have poor diagnostic odds ratios, which is defined as the ratio of the positive likelihood ratio over the negative likelihood ratio.
  2. Lesson 2: To really confirm a case, the IDSA recommends a flu PCR. Less good is the antigen test – only use that if you have no other options.
  3. The flu PCR can still have false positives. as with any test, you should apply it to your pre-test probabilty – a number you come up with based on your local context – to get your post-test probability – the chance a patient truly has disease.
  4. But new flu guidelines have looser recommendations about who to treat – they basically recommend treating any inpatient with suspected flu, even without a confirmatory test, and even after they’ve had symptoms for 48 hours.
  5. You treat with neuraminidase inhibitors like oseltamivir and zanamavir, which originally were thought to decrease symptoms by a day, and reduce complications and hospitalizations.
  6. But as the big pharmaceutical scandal revealed, this data was not completely accurate, and ultimately not as convincing as originally thought.
  7. But on a positive note, data from the 2009 flu pandemic, when they treated everyone, actually produced data supporting mortality benefit.
  8. The data wasn’t great quality – lots of observational studies and no RCTs – but the results were surprising and impressive.
  9. So the NAIs may actually save lives after all!
  10. But that’s why we’re here – to gossip about scandals, and help you process the data for yourself.



Tags: , , , ,