This is one of the most commonly asked questions in IT World, and my answer has always been “CPOE first” – largely because that has always been David Bates’s (the world’s leading IT/safety researcher) answer. But I’ve changed my mind. Here’s why.

Before I start, I promised that I’d let you know if I ever blogged on a topic in which I have a financial conflict of interest. On this, I do: I serve as a paid member of the Scientific Advisory Board of IntelliDOT, a company that makes a stand-alone bar coding system. If that freaks you out, stop reading. But recognize that if you had asked me the “bar coding or computerized provider order entry?” question last week, I would have answered “CPOE”.

That’s because the evidence supporting CPOE is substantially stronger than the evidence for bar coding. For example, a search of “CPOE” on AHRQ Patient Safety Network (AHRQ PSNet) turns up several studies (for example, here and here) demonstrating substantial reductions in serious medication errors, and one documenting an impressive return-on-investment for CPOE. Although we and others have noted that this evidence can be challenged for its lack of generalizability (most of it came from studying the Brigham’s home-grown CPOE system, not the commercial systems that the rest of us are buying) and its impact on proxy outcomes (error rates) rather than on true patient harm, these are mostly quibbles: a good CPOE system does appear to decrease errors at the prescribing phase. Which makes sense.

The literature supporting bar coding is wimpier. The only rigorous study of bar coding I can find is one showing impressive safety gains after implementation of a bar coding system – in the clinical pharmacy, not at the point of care. In fact, a search of PSNet looking at the efficacy of bar coding turns up more studies showing unintended consequences and mischief – including a wonderful paper written for our Quality Grand Round series involving a case in which two patients (one a diabetic, the other not) ended up with each others’ wrist bands through an ED mix-up. Patient B would have gotten a (probably fatal) slug of insulin for a sky-high blood sugar (drawn from patient A, who was wearing B’s bracelet) had an alert doc not overheard a nurse telling a colleague that she was going into Patient B’s room to give him his insulin. Other papers have made the point that clunky bar coding systems can lead to nurse workarounds that subvert the safety features. Clearly, these systems are not panaceas, and – like CPOE (see here and here) – they can easily be undermined by bad design and inattention to human factors.

This week, I had the pleasure of giving a keynote speech at a national bar coding conference, a quirky affair called the “unSummit”, run by two charming bar coding evangelists, Mark Neuenschwander and Jamie Kelly. Mark, in particular, is an unusual guy: a former minister – no Reverend Wright jokes, please – who has a preacher’s passion for bar coding. At the unSummit, I gave a big picture patient safety talk to the 400 bar coding groupies, and then focused on why the uptake of bar coding by American hospitals has been so painfully slow. “If you’re a hospital, there are certain things you have to do, like passing JCAHO and doing anything CMS tells you to do,” I explained. “So you’re competing for limited bandwidth against a bunch of other non-mandatory safety-oriented interventions: teamwork training, simulation, rapid response teams, preventing diagnostic errors, and CPOE. Right now, bar coding lacks the evidence to win that competition.”

But on my flight home, I started thinking about the big, high profile errors I’ve heard about in the last year or two, both at UCSF and nationally. And I had an epiphany. Or maybe it was the turbulence. But here goes.

At UCSF Medical Center (which has a very good paperless chart but neither CPOE nor bar coding; the former is slated to launch soon, with bar coding to follow), virtually every terrible medication error case I can recall in the past couple of years involved a nurse administering a medicine. And talk about “the business case for safety”: these days, really nasty errors get reported to the state health department, whose dour investigators come swooping in, turning over every rock in the building and threatening to shut you down if they find one glitch too many. Believe me, this is not fun. Nor cheap.

Then I began thinking about the People Magazine/60 Minutes-type errors over the past few years – Linda McDougal's unnecessary mastectomy because of a path lab mix-up, Dennis Quaid’s twins' massive heparin overdose, the fatal error in Madison in which a nurse infused an epidural anesthetic intravenously into a pregnant woman. It hadn’t dawned on me previously, but all of these cases represent identification errors that probably would have been prevented by a decent bar coding system.

Thinking about this drumbeat of tragedies, I tried to recall a major medication error in the last few years that would have been prevented by CPOE... and I couldn’t. Not that there aren’t any, but it does seem like today’s Oh-My-God-How-Could-This-Happen med errors are now disproportionately administration, not prescribing, mistakes.

What is going on? I suspect that some of the prescribing errors that CPOE can prevent are now avoided because so many docs are using handheld prescribing aids like Epocrates, and because Joint Commission regs, such as banning high risk abbreviations like “10U insulin” and “qd”, are eliminating some of the worst offenders. Moreover, with everybody now on their toes about medication safety, an errant prescription has many downstream opportunities (pharmacist, nurse, even patient or family) to be caught before it kills.

On the other hand, there is generally nothing that stands between the busy nurse who makes a dose calculation error or confuses a vial of heparin for insulin – and tragedy. The nurse has only one chance to get it right, and no safety net if she gets it wrong. Add to this the effects of the nursing shortage (busier nurses, more temps, more young grads), patients on more and more complex meds, fuller hospitals… and you inch ever closer to disaster.

Why did CPOE gain so much more momentum than bar coding over the past decade? Here’s my theory: because it involves physicians. Think back to the early days of clinical IT. Many of the movers and shakers were physician-informaticists, and they had to sell the case for change (and considerable investment) to their fellow physicians if there was to be any hope of their hospital taking the IT leap. It is logical that they would have deemed prescribing errors to be the main culprit: those are the ones that they themselves had committed and witnessed. As for public demand, doctor’s handwriting has been fodder for Jay Leno jokes for decades. Have you ever heard a stand-up comic prattle about the nurse who gave a patient the wrong med?

In other words, medication administration errors (and laboratory/pathology specimen errors) tend to be out-of-sight, out-of-mind to physicians and the public. Moreover, they involve assembly-line processes and simple execution (no pun), all kinda boring. As for bar coding, how exciting could something be when they've had it at the checkout counter at Safeway for decades?. And so, despite their importance, administration errors (which represented more than one-third of all med errors in Bates’s seminal study) were largely ignored… by researchers, by early IT adopter healthcare systems (the VA is an exception but for some reason didn’t focus on studying this intervention), by physicians, and by the public. And nurses, I think, have been ambivalent about bar coding – hopeful that it might prevent mistakes but worried that it would create workflow hell.  

So CPOE became the darling of the healthcare IT set, winning all the accolades and getting most of the push. And since bar coding is much less expensive than CPOE, there wasn’t as much corporate energy put into developing systems and promoting them.

Now, I could be wrong about this. After all, whatever the reasons, the fact remains that bar coding has not been researched very much or very well. But, with all the medication administration errors I’m hearing about, this is now an area in which I am willing to relax my evidence standards a bit – it is beginning to seem like the equivalent of barricading the cockpit doors after 9/11, a “relatively” low cost, low complexity (at least when compared with CPOE) and commonsensical intervention that can potentially save a lot of victims – both patients and nurses.

A few weeks ago on NBC’s Today Show, Dennis Quaid discussed (the first half of this clip is on his kids’ medication error, the last half on his new movie; it is morning television, after all, and they have to sell soap) his new foundation, set up to prevent the kinds of errors that nearly killed his twins. “We’re going to concentrate on one thing at a time: bedside bar coding… A lot of times patients end up getting the next door neighbor’s medicine… nurses are so overworked… and mistakes occur.” Quaid noted that he was suing the heparin drug manufacturer but had not yet decided whether to sue the hospital, Cedars-Sinai. “Isn’t the hospital going to institute [bar coding]?” asked host Meredith Vieira.  “They have not said they’re going to as of yet,” said Quaid, clearly implying that he might soon remove his high-priced lawyers’ muzzles.

I’m the last person to argue for health policy by Hollywood heartthrobs, but I think that similar cases are occurring all-too-often out of the klieg lights, and that many of them can probably be prevented. The question is whether we wait for better evidence and better systems. I’m pretty sure that our friends at Cedars-Sinai wish they hadn’t.

Ultimately, of course, we need both bar coding and CPOE, and we need rigorous studies looking at what works and what doesn’t. But you have to start somewhere. Even though the evidence continues to trail, based on what I know today, if I was a hospital ready to get into the IT game, I’d go with bar coding first.