Sunday, June 14, 2009

Patient Tracking

First generation patient tracking systems did a lousy job of actually keeping the time line clear, events between admit and discharge. We had to fight with one vendor because the date stamp was supposed to be unique to the cath procedure, yet two caths in one day is not unheard of. Plus where did the patient go in between events A and B? That would be hard to reconstruct, if not impossible.

For statistical purposes, only a few useful intervals were recorded, such as EMT response time (e.g. from dispatch to ER).

Once in the cath lab, we had detailed timelogs going. The doc didn't always show up right away (how could she, this is reality, not television), whereas other times it's more elective and slow moving (not urgent).

My account of research hospital work isn't a criticism so much as a chronicling. Hospitals haven't had the luxury of affordable digital computer equipment for that many years. Insurance companies are actually better endowed in some ways, although those scales might be tipping. The question is more how do we move forward, capitalizing on know-how, like with RFID on the wrist bands, on the gurneys, on both, or on neither.

"Depends on the hospital" would be the short answer, and the kind of job vendors do, persuading us they're really sharing rightly (an allusion to "right sharing", a Quaker concept). You want efficacious use of precious time/energy as there's little room for waste where saving lives is concerned.

A schemaless system isn't thereby unstructured (JSON is malleable but not structureless) and might be used to "grab everything potentially useful" in a "jumbled bag" kind of sense. There's a new cath model getting tried, a new valve, a new stent. Docs will think of things they might like to know about, but aren't sure, might be barking up a wrong tree.

With SQL (RDBMS), it's a total PITA to keep wonking on tables, why the Registries got imposed. Docs had to discipline themselves to keep the Dilberts from stressing out too much, and charging too much money, making administrators angry.

"Make it top-down" was the IT voice from above, "don't let them use Access and keep messing it up." Some docs became shy about nursing pet theories at the local level because every little request seemed to upset someone's applecart. Other docs rolled up their sleeves and dove into computer science, trying to figure out why the hold up (like what's up with this open source business, and what wheels do we keep needlessly reinventing?).

Anyway, now we're in a new era. A patient entering a hospital potentially starts a data stream like on Facebook, though not as public and chatty, lots of technicians making notes. There's a structure here (in the LMRs), but they're not "fill in the blanks" uniform, as you'll find if you create two or three of them randomly, express different patient histories. Medicine is too ramified to think in terms of everything hitting the same dots, even with lots of overlap and family resemblance.

Then later, safely removed from the point of care, not interfering in any way, we write MapReduce scripts (in JavaScript or whatever) to comb through these "has everything" grab bags ("has lots of holes" is also still true) and populate relatively stable SQL models after the fact.

In statistical research, you're looking for samples, so if patient X gets skipped over, it's not necessarily an omission in treatment. Sensors will randomly sample, right from the get go. Clinical research records (CRRs) aren't what a doctor uses on an individualized basis, but for outcomes research. Even if the legal medical record (LMR) is likewise electronic, these two kinds of record are conceptually different, in terms of workflow.

Do patients spend more time in hallway A or hallway B, when routing to cath lab, and does it matter if there's a coffee shop open upstream from C? Hospitals with time for those kinds of question would be in the small minority.

Traffic analysis in urban settings gives a sense of it though: you don't need to track every car in intimate detail to get an overall picture, but where a patient's hospital experience is concerned, more data is better, as thoughtful analysis doesn't occur in a vacuum.

Empirical measures are the life blood of medical science. So in terms of "erring on the side of caution", we're likely to record too much rather than too little (or wish we could -- again, bioinformatics is still in its infancy).

For further reading:
Charting the Future
(for sysadmins) in my 4D Solutions presentations folder.