Wednesday, July 11, 2012

The Numbers Don't Add Up - Selectively Using Statistics

The Gist:  Patient safety is of the utmost importance in medicine, yet it lies in a very delicate balance with physician and student training.  Recently, it seems as though the literature suggests we should be doing fewer procedures, or at least - safer procedures.  These efforts to perform clinically significant interventions (aka smart medicine) on patients could mean fewer procedures (and thus, less practice for trainees).  Simulation is an excellent and crucial part of training, as evidenced in lumbar puncture in this article from Neurology.  However, as these articles by Tun, et al and Meguerdichian et al in July 2012 Annals of EM show, clinical situations often deviate from the more "clean" simulated endeavors.

Recently, I had a patient who presented just over 6 hours after the onset of  a severe headache.  The patient's blood pressure was elevated to a systolic of 180 mmHg and the headache, though gradual in onset, was different in character than prior headaches.  After only partial response to a migraine cocktail, we decided to send the patient for a head CT, though subarachnoid hemorrhage (SAH) seemed unlikely.  After a clean scan, I talked with my attending.  Every ounce of clinical gestalt I have (or the sole ounce, as is likely the case) prompted me to plan to amp up the analgesia from acetaminophen, reassess, and then discharge.  This seemed like a classic migraine and the patient's neuro exam continued to be perfect.  In my plan, I mentioned that one could consider a lumbar puncture (LP) but the odds that this would result in a clinically important outcome were minimal.

Fortunately, my attending realized the grave blunder I had made as a medical student and offered to stay late with me and do the LP, if I wanted to do one.  I had nearly talked myself and my attending out of a procedure - unthinkable for a procedure junky!  When presenting the options to the patient, I couldn't help but feel we were slightly paternalistic.  We presented solid evidence and advised the patient that we doubted that the LP would yield anything; however, we didn't quote the magnitude to which to which it was unlikely we would find something (probably <0.2%).  However, I had a bout of inner turmoil, knowing that the test would likely yield nothing besides a pair of slightly more experienced young hands and a very happy medical student.  As it turns out, the tap was clean, the patient tolerated the procedure excellently, and I felt gratitude towards the patient and more confident in my abilities

What's the deal with lumbar punctures?   We still get to stick the febrile infants, the encephalopathic (see anecdote below*), and the visibly sick headache patients with a higher pre-test probability of serious pathology; however, there's presently some controversy regarding LP in CT negative headaches.  
  • This paper, published in the BMJ, kicked off much of this controversy last August.  It demonstrated that third generation CT scanners had a sensitivity of 100% in the cohort scanned within 6 hours of headache onset, although not all negative CT scans were followed up with an LP (rather, they used 6 month clinical follow up).  The sensitivity was less in the group with negative CT scans greater than 6 hours after onset, as the blood diffuses ans and hemolyzes, obscuring the difference between CSF and blood.  
  • Drs. David Newman and Ashley Shreves dive into the world of subarachnoid hemorrhage in this podcast.  In analyzing the data, they end up determining that, in patients presenting to the ED with a headache and a negative head CT, LP will likely benefit 1 in 625..probably not worth it for most patients.  Sinai EM Media Site has a lovely cheat sheet of the basic derivations (helpful while following along with the podcast).
  • Note:  AHA guidelines still recommend LP if CT scan negative.
* A month or so ago, I was working up a patient for altered mental status.  CT was fairly clean and the patient struck the attending as a likely drug abuser, based on dentition and other factors.  I proposed an LP, which my attending agreed to when the patient's drug screen came back negative.  Finding?  Meningitis.  Moral of the story:  Use your clinical gestalt and the evidence to guide decision making.  In the words of Dr. Mark Crislip of Gobbets o' Pus fame, "the plural of anecdote is not data."  


  1. great post! but you lost me at drug screen... :)

    1. haha. funny thing is - I pulled out your blog post on UDS (saved in my Evernote app) when the attending challenged my plan to not include a UDS in the workup