Grand Rounds 9.22.11

Dr. Stoller was kind enough to give us a presentation on hyperbaric oxygen therapy this morning.  Click this link to view a recording of the presentation.

Grand Rounds 9/16/11

Dr. Edward W. Holt, senior GI fellow at CPMC, gave a cogent and engaging presentation this morning on acute liver failure.  Click this link to see a recording of his talk.

Rinne, Weber and Likelihood Ratios

Click here to listen to a recording of the session. Apologies for some minor irregularities in this week's recording. 

Nomogram for using likelihood ratios. 

For HIPPA-related reasons, we are no longer going to be posting Resident Report publically; however, you should soon be able to get a recording of Dr. Schafhalther-Zoppoth's comprehensive introduction to transverse myelitis (and future Resident Reports, noon conferences, etc.) from the Attendings Folder.  We will still be recording the "5 minutes of EBM" sessions, and after a significant outpouring of feedback I (NN) am going to try really hard to remember to talk more slowly.

This week we talked a little about the diagnostic accuracy of physical findings.  Specifically, we talked about likelihood ratios as they apply to diagnostically useful physical signs.

Likelihood ratios (LRs) can be generated from the same information you use to calculate sensitivity and specificity.  What they specifically evaluate is the ratio between the likelihood of a particular diagnostic finding occurring in somebody with the target disorder and the likelihood of the same finding occurring in somebody without the target disorder.  Follow this link for a more comprehensive (and surprisingly readable) discussion of LRs.

Heinrich Adolph Rinne

Likelihood ratios are particularly useful because you can use them to directly translate pre-test probability into post-test probability.  Because the likelihood ratio is an expression of odds, not percent probability, it's easiest to use a nomogram to do this so that you can avoid converting back and forth. To use the nomogram to the upper right (developed by EBM pioneer David Sackett,) connect the pre-test probability to the likelihood ratio of your test using a straightedge.  The number intersected by the straigtedge on the scale furthest to the right is the post-test probability conferred by the likelihood ratio.

We talked specifically about the likelihood ratios associated with Weber's and Rinne's tests for conductive vs. neurosensory hearing loss.  Weber mystified a generation of neurology patients by sticking tuning forks in the middle of their foreheads.  His test is supposed to lateralize to the bad ear in conductive hearing loss, and the good ear in neurosensory hearing loss.  Heinrich Rinne's test for conductive hearing loss involves testing hearing by direct conduction through the mastoid versus air conduction through the tympanum and auditory ossicles. 

Ernst Heinrich Weber

We all learn these tests in medical school, but most people have little sense of their quantitative significance (which is under-emphasized in physical examination generally.) However, if you go to the McGee's magesterial Evidence Based Physical Diagnosis, you will find that Rinne's test has a published positive likelihood ratio of 16.8, and a negative likelihood ratio of 0.2.  For the sake of example, this means that if you take somebody who has a pre-test probability of conductive hearing loss of 20% and Rinne's test reveals that bone conduction is superior to air conduction, their post-test probability rises to around 80%.  Conversely, a normal result would reduce the probability to about 5%.  These are both pretty significant diagnostic yields for a test which is easy to perform and requires nothing more complicated than a vibrating bit of aluminum.

Weber's, on the other hand, does not fair so well.  The finding of lateralization to the good ear as a predictor of neurosensory hearing loss is associated with a modest LR of 2.7, and the negative likelihood ratio for this finding is nonsignificant.  The positive likelihood ratio of lateralization to the bad ear for detecting conductive hearing loss is nonsignificant, and the negative likelihood ratio is only 0.5.  Application of these values to the nomogram is left as an exercise to the reader, but they changes in probability they confer are nothing to write home about.

So the take-home points are that:

A) Many studies of diagnostic strategies report their findings in terms of likelihood ratios, which are not hard to understand and quite useful, and
B) It's worth looking into the published test characteristics associated with your physical exam maneuvers, because some of them (like Rinne's) are very reliable and some (like Weber's) don't change things all that much.

Grand Rounds 9/9/11

Grand Rounds today was given by Dr. Caroline Hastings, who was kind enough to take time out of her busy schedule as director of the pediatric hematology-oncology fellowship at Children's Hospital to talk to us about idiopathic thrombocytopenic purpura.

Click this link to view a recording of Dr. Hasting's presentation.

Doctors, Patients and Framing Bias

Which line is longer?

[Click here for a recording of the noon conference talk on framing bias.]
[Click here for a recording of Dr. Kaur's talk on bacterial meningitis.]

The last couple of posts have been about measures of treatment effect, specifically about absolute and relative differences in risk and numbers needed to treat or harm.

Something to remember about these measures is that, while they're calculated from the same data, they often sound quite different.  You can significantly prejudice somebody's answer to a given question by the way you frame it.  For instance, consider these two questions:

1. Would you like to spend the next three years of your life constantly underslept, trying to do seven things at once, eating mainly hospital food and struggling against insurmountable barriers to do a merely adequate job?
2. Would you like to spend the next three years of your life at an excellent community teaching hospital working with inspiring mentors to join the next generation of excellent internists, all while working for people who really need and appreciate you?

Because you're an internal medicine resident, you know that these are in fact the same question, but you can see how the way you put it makes a difference.  In statistics, this is called "framing bias," and is the subject of a study recently published in the Journal of General Internal Medicine. Perneger and Agoritsas did a large survey of Swiss doctors and patients' understanding of a hypothetical treatment.  The subjects were randomized to several groups: some of them only got information about the treatment's relative benefit, some of them only got information about absolute data in terms of survival or mortality, some of them only got information about the number needed to treat, and some got all the available data. 

They observed that how the information was presented had a substantial and significant effect on what people thought about the treatment.  For instance, while only 51.8% of doctors who only saw the data presented as absolute survival thought the treatment was likely to be better than its predecessor, that proportion rose to 93.8% among doctors who only saw the relative risk reduction.

But what was arguably most fascinating was that there was pretty much no difference between doctors and the patients responses. These doctors' extensive training in understanding quantitative analyses of treatment effects did not appear to reduce their liability to framing bias at all.  This study (which I recommend reading in full) suggests that when we're only given relative measures of effect size, we are very likely to overrate the effect relative to what we would say if we could also see absolute effect sizes.

Fidaxomicin, ARR and NNT

Clostridium difficile endospores

The NEJM recently published a prospective, multicenter trial comparing fidaxomicin and oral vancomycin for the treatment of C. difficile diarrhea.  Fidaxomicin is a newly approved bacteriocidal antibiotic which works by interfering with RNA synthesis.  The hypothetical advantages of fidamoxicin are that it has a narrower spectrum and may reduce the incidence of recurrent C. diff. diarrhea.  The trial analyzed rates of clinical cure and recurrence of C. diff., and met the pre-established criteria for non-inferiority.  It also seemed to show a significant difference in rates of recurrent infection between the fidaxomicin and vancomycin groups, with an absolute risk reduction in the fidaxomicin group of 9.9% in the modified intention-to-treat analysis and 10.7% in the per-protocol analysis.

The absolute risk reduction (ARR) is simply the experimental event rate (EER) minus the control event rate (CER), which in this case is the incidence of recurrent C. diff. infection in the fidaxomicin group minus the incidence of recurrent C. diff. infection in the vancomycin group.  The ARR is a really useful number, because it allows you to calculate the number needed to treat (NNT), which is just 1/ARR.  In this case, if we split the difference between the intention-to-treat and the per-protocol analyses and guess that the ARR is really about 10%, that means the NNT to prevent one case of recurrent C. diff diarrhea is equal to 1/0.1, or 10.  This is significant, but enthusiasm should be tempered by the fact that this is a new drug, and by the fact that whereas a 10 day course of oral vancomycin costs $100 (if you use IV vancomycin given orally instead of the more expensive Vancocin™ pulvules) a ten-day course of fidaxomicin costs about $2200.  Additionally, the study was heavily funded by Optimer pharmaceuticals, which makes the drug, and many of the investigators had received honoraria and other payments from Optimer.