This is a nice informatics paper that deals mostly with problem lists. These are meticulously maintained (in theory) by inpatient and ambulatory physicians to accurately reflect a patient’s current medical issues. Then, when they arrive in the ED, you do your quick chart biopsy from the EMR, and you can rapidly learn about your patient. However, these lists are invariably inaccurate – studies show they’ll appropriately be updated with breast cancer 78% of the time, but as low as 4% of the time for renal insufficiency. This is bad because, supposedly, accurate problem lists lead to higher-quality care – more CHF patients receiving ACE or ARBs if it was on their diagnosis list, etc.
These authors created a natural language processing engine, as well as a set of inference rules based on medications, lab results, and billing codes for 17 diagnoses, and implemented an alert prompt to encourage clinicians to update the problem list as necessary. Overall, 17,043 alerts were fired during the study period, and clinicians accepted the recommendations of 41% – which could be better, but it’s really quite good for an alert. As you might expect, the study group with the alerts generated 3 times greater additions to the patient problem lists. These authors think this is a good thing – although, I have seen some incredible problem list bloat.
What’s interesting is that a follow-up audit of alerts to evaluate their accuracy based on clinical reading of the patient’s chart estimated the alerts were 91% accurate – which means all those ignored alerts were actually mostly correct. So, there’s clearly still a lot of important work that needs to go into finding better ways to integrate this sort of clinical feedback into the workflow.
So, in theory, better problem lists, better outcomes. However, updating your wife’s problem list can probably wait until after Valentine’s Day.
“Improving completeness of electronic problem lists through clinical decision support: a randomized, controlled trial.”
www.ncbi.nlm.nih.gov/pubmed/22215056