Sometimes I feel like the old geezer at case conferences at my institution. I perceive increasing focus on expensive technology (MRIs and PCRs, among many other acronyms) at the expense of simple history and physical examination. Likely because I was hoping this article supported my old geezer approach, I chose it for review this month. Unfortunately, all I gained from a close look is yet another example of how not to design and report a research study.
Source: Hurst D, Hirsch D, Oster M, et al. Syncope in the pediatric emergency department-can we predict cardiac disease based on history alone? J Emerg Med.2015;49(1):1-7; doi:10.1016/j.jemermed.2014.12.068. See AAP Grand Rounds commentary by Dr. Jeffrey Anderson (subscription required).
PICO Question: Among pediatric patients presenting to the emergency department with syncope, what historical features predict an underlying cardiac etiology?
Question type: Descriptive
Study design: Retrospective review
The study started out interesting, a retrospective review over a 46 month period (more on this later) at 2 tertiary children's emergency departments of children discharged with an ICD-9 code of syncope or near syncope. This turned out to be 3382 individuals without a prior cardiac diagnosis. Of those, 3 had a cardiac cause for their syncope (another 12 had an incidental unrelated cardiac diagnosis). These numbers are in line with prior reports both of the rate of syncope among pediatric ED visits (0.7% here) as well as the rarity of cardiac causes in such cases (less than 0.1% here).
Fine so far, but here's where things started to get confusing and, ultimately, not that helpful to clinicians. The authors chose 4 historical features to review, comparing the 3 cardiac-related syncope patients to 100 children selected randomly from the remaining 3000 or so. Those 4 features were 1) syncope during exertion; 2) exertion accompanied by chest pain; 3) onset of palpitations immediately before the episode; and 4) absence of prodrome. All of these historical features have been suggested to be associated with cardiac syncope, and in fact are included (in the midst of a detailed discussion of historical factors in syncope) in the 2006 American Heart Association guidelines for syncope evaluation.
Beyond that simple analysis, however, the science is stretched pretty thin. First, I wanted to know how the authors chose just those 4 questions to look for. Also, the authors don't tell us much about how this information was gleaned from medical records review. Remember, this was a retrospective chart review. It doesn't appear that these EDs had a standardized history form for evaluation of syncope, and I can't imagine that every chart had unequivocal evidence of evaluation on these 4 historical features. The report doesn't go into detail about how the chart data abstraction was performed; normally they would have strict criteria for interpretation, not to mention verification by more than 1 data extractor. Maybe that was done, but I can't tell.
PICO Question: Among pediatric patients presenting to the emergency department with syncope, what historical features predict an underlying cardiac etiology?The authors give us confidence intervals for sensitivity and specificity for each of the 4 questions. In this clinical circumstance, you can think of each questions as a diagnostic test. Since it could be potentially devastating to miss a cardiac cause of syncope, we would want a highly sensitive test (preferably 100% sensitivity) to use as screening, and we'd be willing to accept some false positives in return. The problem with all this is that we only have 3 cases of the disorder of interest, meaning our confidence intervals are going to be very wide. So, no matter how good the questions might be as screening, we still won't know how they would perform on a much larger scale. For example, all 3 patients had palpitations associated with their syncopal events, resulting in 100% sensitivity. That's great, but the 95% confidence interval is actually 30 - 100%. I wouldn't want a screening test with only 30% sensitivity. The authors also inform us that using any combination of 2 of the questions would result in 100% sensitivity, though they don't give us the confidence interval (which clearly would again be very wide).
Just a word about the chart review period of 46 months. That's an unusual number, and I'm always suspicious when a study doesn't have some number of full years (e.g. 48 months). The reason could be innocent: maybe the medical record system changed and it was no longer possible to do the chart review in the same fashion. However, a real no-no would be if the authors chose the start and end of the period as when a cardiac syncope patient was identified, which would very much skew the incidence data. Presumably they didn't do this, but they should tell us why they chose this study period.
The authors couldn't have done much with their data no matter how careful they were with the study design, because one needs a large number of cases to narrow the confidence intervals around the sensitivity estimates. Still, I'm not convinced that these are the 4 questions most important to ask of our syncopal patients.
And just one final note on old geezerdom. I moved this discussion up to the first week of the month because my local paper, the Washington Post, published an opinion piece suggesting that the stethoscope is becoming obsolete, to be replaced by handheld sonographic devices. So much for my love of the history and physical exam. Of course, that argument presupposes we only use stethoscopes for cardiovascular evaluation, and not for pulmonary and gastrointestinal auscultation. I love gadgets, but I'll need to see more objective data for the reliability and validity of this approach before I hang up my stethoscope.
