The accuracy of clinical symptoms and signs for the diagnosis of serious bacterial infection in young febrile children: prospective cohort study of 15 781 febrile illnesses BMJ 2010;340:c1594

I read on average about eight published medical studies a week. Mostly from the emergency medicine literature (about 5 or 6 journals).

I listen (rather unpredictably)  to 18 different medical podcasts. Mainly I listen to EMA cause I think it’s the best thing out there even if you do have to pay for it. I have 10 different medical blogs on my reader subscription.

That is a lot of information.

I retain abou 2% but at least it stops me getting bored.

Over the past few years I have got much better at critical appraisal and I have become far more cynical about a lot of what is published and a lot of what people typically think health care provision should consist of.

Every now and then the thought flutters through my head that I should share my cynicism and rambling reviews of papers on the blog. [Those who come here for the pictures of my brothers dog and my quotes of books by dead people will be most disappointed I know… I’ll give the postings titles that aren’t song titles to make it easier for you to tell the difference]

So maybe I’ll start today. And I’ll try and keep it readable to the non-medics too.

The accuracy of clinical symptoms and signs for the diagnosis of serious bacterial infection in young febrile children: prospective cohort study of 15 781 febrile illnesses BMJ 2010;340:c1594

Let me start with a brief comment on accuracy. Most people would think that being accurate means getting it right. Unfortunately, like most things in medicine, and indeed in life, it’s more complicated than that.

We can get it right most of the time but not always.

I’m afraid you’ll have to live with that. Or if you’re one of the one where we get it wrong then you won’t have to live with that cause maybe you’ll be dead.

We can do better, no doubt. But we cannot give you the “accuracy” you want. For lots of reasons that would distract me from the paper.

Kids get fevers a lot. They seek medical attention (or their parents do) fairly often when they get a fever.

The vast majority are fine (in that they are miserable, grumpy, with a fever and poor feeding but will get better all on their own if you give them a few days.). Some of them have nasty things wrong with them and will either do badly or even die if you don’t do something. Incidentally there are lots of the ones who do badly who will do badly no matter what we do but I like to at least try rather than miss it altogether.

If I had a 100 kids in the waiting room with a fever (and sometimes it feels like it) then if I sent them all home then it would only matter (as in some one comes to actual proper harm) in about 2 maybe 3 of them.

So even without a medical degree you’re off to a good start just by saying that everyone is fine.

These guys think that doctors aren’t quite cutting it with their history taking, examination and decision making (or educated guesswork as I call it) so they thought that if they measured all the symptoms and signs (“like it hurts when I pee”, or “you’re breathing is twice that of normal”) in 18000 kids with a fever that they could use all the information gathered and make a computer algorithm that will tell us which ones have the “serious bacterial illnesses”.

That’s the basic summary.

More technically they did something increasingly common:

1) identify a target population – in their case every kid under 5 with fever –  inevitably you start with a low prevalence population.

2) come up with a proforma to prospectively collect lots (more than 40 variables) of data on every kid with fever. All data collected by the Emergency Physician (EP) looking after the kid. My problem here is that they didn’t do interobserver agreement on the proforma – two doctors can look at the same kid and disagree substantially on points of history and examination.

3) let the EP come up with a probability of diagnosis before they do any tests. While this is nice in theory it’s not what we normally do in practice. Often the urinalysis is done before I’ve seen the patient or a GP has sent the child in with a provisional diagnosis. So in practice coming up with a probability (especially in terms of actual percentages) is not everyday practice. I imagine we’re not great at it. A computer model may give you a number, an EP is unlikely too. So it seems unfair.

4) use logistic regression (which I still don’t entirely understand but everyone does it) to come up with a list of variables that predict certain diseases of interest. Bung these altogether into a computer algorithm, identify and enter the relevant data and hey presto out pops your probability of an SBI (serious bacterial illness).

5) compare which was better at picking up SBI, the doc and his “judgement/gestalt” or the computer model

Besides the issues which I take with their methodology the big issue i take is their definition of “serious bacterial illness” (SBI). In fact it’s not really just their definition – all of the literature on this stuff uses the same crazy idea.

Let me use their own numbers to tell the story:

By their definition they had 1140/15781 (7.2%) had a serious bacterial illness by their definition.

A third did not get any antibiotics prescribed. Of these:

“One death occurred in a child with a lethal congenital disorder on end of life management.”

And of the others:

“eight were unwell at follow-up an average of 10.2 days later and none was febrile.” (unwell at follow-up could mean poor feeding, poor form).

So of 363 kids with an SBI who did not get treated (the whole point of the study is that kids with SBI need treated, the word serous implies this don’t you think?) only 2.2% were even considered “unwell” at follow up and no one (it seems) died of anything relevant.

They don’t tell us how the 2 thirds with an SBI who did get treated did – which would have been interesting because serious bacterial illnesses do exist in children and they do kill children.

So, how do they define SBI in kids:

Bacteremia (pretty serious most of the time I think)

Pneumonia – though this meant CXR changes mainly – lots of kids have pneumonia, lots of them just need some antibiotics and can go home fine and do great, lots of them probably have viral pneumonia and if we hadn’t done the CXR they would have been fine with nothing – best example is all the legions of kids with bronchiolitis.

Urinary tract infections – some kids get UTIs, actual significant infection in the urinary tract. Lots of them grow bugs from your urine when you test it – I’m not sure the two are the same thing. I’m pretty sure lots of them aren’t serious by any definition.

All of this is a little silly.

Let me be clear, someone who has a fever and happens to grow a bug from the urine does not necessarily have an SBI, in fact most do not.

So back to the numbers. In this study of the serious bacterial illness?:

Bacteremia – 6%

Pneumonia – 48%

UTI – 48%

[Incidentally some must have had both has the numbers don’t add up]

Sick kids, look… well… sick. Most of the time. Sometimes they can be difficult to pick up. I’m not saying it’s easy but I’m not sure this helps us any.

The drive to identify this definition of SBI (which is what all the published papers are pushing at) makes us switch off our thinking as soon as we “exclude” an SBI missing out on the other bacterial illness that probably are serious – for example they excluded mastoiditis and epiglotitis from being an SBI, because they are rare and didn’t fit the model.

It also doesn’t cover other really important things that cause fever like Kawasaki’s disease.

Advertisements

1 Response to “The accuracy of clinical symptoms and signs for the diagnosis of serious bacterial infection in young febrile children: prospective cohort study of 15 781 febrile illnesses BMJ 2010;340:c1594”


  1. 1 mrsa swabs June 1, 2013 at 4:52 am

    The Wii, and the new Kindle, but it wasn’t for us. O y medio una campa? Apple says they didn’t bactrim for mrsa go LTE just yet
    because first-gen chipsets would force unwanted design decisions, and customers want a Verizon device now.


Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s




About

May 2010
M T W T F S S
« Apr   Jun »
 12
3456789
10111213141516
17181920212223
24252627282930
31  

%d bloggers like this: