David Bates, MD, MSc

Medical Director of Clinical and Quality Analysis, Partners Healthcare

Chief, Division of General Medicine

Brigham and Women's Hospital

75 Francis Street

Boston, MA 02115

USA

I shall start by saying a few words about the nature and scope of the problem of medication errors in the US, followed by something about costs and then move rapidly into the potential effect of information technology for improving medication safety. My focus on medication errors is mainly on three areas:

1. Computer acquisition.

2. Order-entry.

3. Bar-coding.

I will also touch on 'Smart' pumps then very briefly tell you about our Centre of Excellence in Patient Safety. This is a five million dollar project funded by the Agency for Healthcare Research & Quality. Also worthy of mention is the Signature Initiatives Partners. Partners is the delivery system within which I work whose C.E.O. has elected to do six things through our network (one task focuses on medication safety).

The Report from the Institute of Medicine talked about the chasm between where we are today and where we could be. This followed 'To err is Human' and I think, in many ways, is the more important report. An example of how we illustrated the present level of patient safety is our study of two general pre-units and an I.C.U. We examined 10,070 orders and found 530 medication errors, 35 potential adverse drug events and five preventable adverse drug events. These data suggest that about one in a 100 medication errors results in an adverse drug event while seven percent had the potential to do or were "near misses." We mounted the Adverse Drug Event Prevention Study to assess the epidemiology of medication safety and this took place in two large hospitals in the Boston area. The key findings were that there were 6.5 adverse drug events per 100 admissions, one-third of which were preventable and a surprising 62% of the errors were in ordering and transcription. This proportion was surprising, because much of the emphasis in research had been on administration because in part of Ken Barker and Betsy Flynn's work. Another more recent and provocative study was of patients one week after they returned home. When we telephoned them we found an adverse event rate of 19%! Six percent were preventable and 6% were ameliorable; a new term to describe events whose impact could have been lessened had they been detected earlier. Of the adverse events two thirds were adverse drug events. Here is an example of patients with congestive heart failure, started on spironolactone and already on potassium. This is an example of evidence-based medicine gone wrong because there was a large randomised trial suggesting that this was the thing to do but in the real world the patients are not always adequately monitored. This patient became very weak and anorexic, and when they came in, the potassium was over 7.5. This is something that could have been readily prevented with a little monitoring.

Another study we did focussed on adverse drug events among 2,000 outpatients; among these, 18% reported problems with their medications and in 35% of the occasions when they reported the problem the medication was not changed. 20% of patients had symptoms that lasted longer than three months. When we reviewed the charts, only 3% of patients had adverse drug events documented but of those, 5% had to be hospitalised (15% were preventable). It is clear that there was a discrepancy between the patient-reported and chart-documented events. Another later study showed that the true rate was closer to 18%. Patients who had had an adverse drug event were dissatisfied and about half sought medical attention reporting discomfort. There were a number of correlates of patient-reported drug complications including several medical problems, unexplained side effects and primary language that was neither English nor Spanish.

While there has not been that much work done recently on admissions for adverse drug events, this is another interesting area. The literature suggests that between a half a percent and 21% of admissions classified as suffering from adverse drug events. One of our studies showed 1.4% of our admissions were classifiable as adverse drug events and all occurred in outpatient clinics. A very high proportion was severe and about a third were preventable.

Next, I am going to talk about how IT can improve patient safety and there are three key ways of doing this:

1. To prevent errors & adverse events.

2. To facilitate a more rapid response because there are lots of things that cannot be anticipated but if you can move fast then you can keep things from getting out of control.

3. To track & provide feedback about adverse events which is something that we don't do very well today in most of our systems.

We suggested a framework in our paper for the New England Journal of Medicine for improving safety using information technology. One key strategy for preventing adverse events is improving communication tools. Our argument was that medication is communication-intensive and although the impact of improved communication is hard to measure, it is almost certain that if you make improvements and keep people on the same page care will be safer. A small example is an application for our pharmacists that we built - that was relatively easy to do and proved to be very useful. This gives a snapshot on one screen of all the clinical information that they want for inpatients. Another strategy is to make knowledge more readily accessible. That may sound obvious but in our first adverse-drug-event prevention-study, the single biggest correlate of an adverse drug event was not having the necessary knowledge at hand (instead of being buried six or seven screens away). The computer can also perform drug calculations better than people and undertake real-time checks. Assistance with monitoring will be increasingly useful and is not as well worked out as it should be and of course computers can provide decision-support for clinicians.

So now onto CPOE The evidence suggests that this is the single most powerful information source for improving medication safety that we have had so far. Some studies have shown over 80% reduction in medication error rate but you must have associated decision support if you want to achieve this level of benefit.

So how does computerised ordering improve safety? There are several keys. First, people choose doses from menus and just doing that helps to do things right because a tenfold overdose does not appear as one of the options. You can essentially eliminate transcription errors and you can require complete orders (I had no idea how big a problem that was for pharmacists). When I started working in this area I was shocked to find how many orders did not include the dose, the route, the frequency or sometimes even the drug name. My favourite order is '40 mg IV now'. You can also give information when people need it, you can show relevant labs, and guidelines and then you can do all those checks in the background. When we started in 1993 I thought it would take a year or so to do all this list down here in the bottom of the slide and then we would move on. However we are still working out how best to do some of these things. For instance drug allergies turned out to be vastly more complex than I had anticipated. Recently we carried out a controlled trial using concurrent and time-series comparisons with a relatively primitive version of CPOE and achieved 55% decrease in the rate of serious errors. This application only had the top ten drug interactions and the penicillin- sulphur allergy check but nothing else, minimised transcription and required complete orders. A greater reduction in the error rate would have been achieved with more sophisticated decision support.

The NEPHROS (renal dosing) study shows the value of sophistication. This is an application that takes patient data including age and gender and then we make the provider give us the weight which is the hardest thing because we do not have that much of the time. We then calculate a suggested birth dose and most clinicians do not remember that there are hundreds of drugs for which you should adjust the dose. We then suggest the right dose. There were 18,000 patients in this study and fully 42% had some renal deficiency which is far more than I expected. The control group got the appropriate dose and frequency of medication 55 and 34% of the time, respectively, but with decision support that went up to 67% and 59% of the time. We have done a cost-benefit analysis of CPOE and prioritised all the decision support functions and this was one that saved most money and increased care the most. So back to allergies: we looked at drug allergy alerts over three months at Brigham and found they were overwritten 80% of the time, also the override rate had crept up over the years. Furthermore only 6% of the alerts were triggered by the match between the drug order and the allergy list but there were lots of alerts triggered by cross allergies. So we tried to make the warnings for the serious allergies (like anaphylaxis) look different from the non-serious ones (like a stomach upset) because people were ignoring the serious alerts as much as the non-serious alerts. The theory of warning response suggests that this kind of sign will elicit a response particularly if not shown all the time.

I suggest that the key areas of decision support in CPOE, in descending order are the following:

1. Complete orders - relatively straightforward but has a lot of value.

2. Default doses and believe it or not there are some computerised prescribing applications that do not suggest the default dose.

3. Drug allergy checking.

4. Renal dosing.

5. Paediatric dosing.

6. Drug interaction checking is relatively low on my list in terms of value for reasons already discussed.

7. Drug lab checking and dose calculations.

We are studying when to alert and more importantly how to alert so that providers do recognise critical warnings. We need studies to see how different presentation of alerts can change override frequencies. Other important factors are cognition and work flow. Another area of investigation is the means to persuade people to document better. Overall it should be easy to do the right thing, we already require people to say why they override and pharmacists find these reasons very useful. The challenge is to effectively display the information that providers find most useful without overwhelming them. Work has already started on the further challenge: processing data in the background, understanding the context of the users so as to anticipate their needs for decision support. I have to say that more complex decision support has not really been worked out and so most trials of this have failed.

Bar-coding is an inexpensive technology used in every grocery store. If it were widely used in healthcare it would help to answer the questions:

· What? · How much? · Who? · When?

However there is very little good data published about the effectiveness of bar-coding in reduction of medication errors. We are now doing a large study at the Brigham and there is another underway at Wisconsin. Studies like these should be done in a variety of healthcare settings.

IV drugs are another interesting area and intravenous therapy is a vulnerable spot in healthcare. Errors in delivering IV drugs are hard to detect, the errors are almost certainly higher than people realise and have a very high severity level because of the nature of the drugs. Without changing the drug delivery system it will be very hard to intercept these errors. Effective interfaces and integration are going to be critical and it will also be important to engineer out errors and to build approaches that help clinicians do what they want to do without reliance on training. We have been studying one device (built by Alaris) - a smart pump that retains information about the drug that has been given and warns clinicians if there is too high an intended dose. A real example is a Heparin bolus of 4,000 units followed by an infusion of 890 units per hour. The first of which the nurse gives perfectly but then follows this with an infusion rate of 4,000 units per hour. Now the pump alerts the nurse and data from Alaris suggests that around two errors like that happen daily in a typical 400-bed hospital. We have just finished a controlled trial of the impact of these pumps (yet to be published) however some of the early qualitative findings are that administering these drugs is quite complex. I found this trial much more complicated than I had anticipated; there are frequent and tricky handovers during the pump use, for example, a patient typically go from the OR to the Recovery Room to the ICU with handovers at each transition. We had wanted to do this study as a randomised control trial but could not do so because at the beginning of each case we could not found out the end location of the patient because the hospital was completely full. So instead we had to conduct an interrupted time series study. Physicians and pharmacists have little understanding of the issues but nurses do and had we been we known this at the start, the design would have been easier. Another finding was much of the focus had been on the initial programming of the device, but the initial infusion rates were changed so often during the therapy that this represented the bulk of the programming. One of the great things about these pumps that is extraordinary valuable for error research is that these pumps are the equivalent of the black box in an aircraft. They track every single thing that is done with them and thus represent a treasure trove of data. One can ask questions like: whether a particular individual is likely to override and whether some of the rate limits should be removed. We found that some of the limits did not make much sense; there were low-end warnings in place that did not appear to be useful.

However I.T. is not the only way to improve safety - there are lots of other ways - for example we found a 66% reduction in preventable errors in ordering when pharmacist began to participate regularly in ICU rounds. Building a culture of safety is essential for safety to improve.

Next I will briefly mention another exciting area, which is access to health records by patients. We have recently completed a study in which we allowed patients to see a portion of their electronic health records through a secure portal that lets them interact with their practice. There is also a practice gateway to allow patients to practice with the information and a public gateway. Now we are evaluating the popularity of this facility. Our outpatient work recently has suggested that one of the greatest problems is that when outpatients begin to experience a problem with one of their drugs they often do not tell their provider. So this project is a new and very exciting area that may offer new opportunities to improve patient-provider communication. Next is our Centre of Excellence, which focuses on improving medication safety across clinical settings. Our goals are to:

1. Extend research to various populations in different clinical settings.

2. Try to close some of the gaps in current medication error reduction research.

3. Develop new prevention and safety strategies in targeted specific settings.

4. Develop and translate findings to an array of different settings.

There are six different projects. The first is to build a web-based surveillance and reporting tool, which will not be only concerned with medication errors but with all types of errors and adverse events. We are putting that in place in all our Partners institutions to see if the tool increases reporting rates and generates useful data. The second study looks at the high frequency of medication errors in outpatient paediatrics and how often these result in adverse drug events. The third study, that has yet to start, is about medication errors in hospital psychiatric units. Prior investigations in a small number of psychiatric units have revealed very high error rates - hence this much larger project. I have already mentioned the investigations of errors in I.V. therapy. The fifth study is about the use of warfarin in nursing homes. A very large proportion of patients (20%) in US nursing homes are on warfarin and, as you might imagine, there are lots of problems - including interactions - around prescribing warfarin. The final project focuses on developing a tool for measuring organisational culture. Our idea is to use the tool to measure the extent and nature of the safety culture in a particular site - and hopefully improve the local safety culture. All projects at the Centre of Excellence are intended to be synergistic, we try to translate them into different settings and to eliminate some of the research gaps.

Another exciting area is the six Signature Initiatives The first is putting CPOE into six hospitals and rolling out our electronic health record out to all 7,000 physicians (and not just the primary care providers). The second, which I led, is a safety-oriented initiative and this only focuses on medication safety not because this is the only safety issue but because it is the area where interventions are best defined. There are four aspects:

1. Putting all our applications about medication decision support across our network. For this to work successfully there must be consistency. As an illustration, we found that we treated allergies differently in every single application and there were important differences along the lines of being able to warn people about important interactions.

2. We want to prioritise implementation of interventions to improve administration and dispensing. This will be done particularly by barcoding and smart pumps. We cannot afford to do everything at once are considering how to do this in a rational way.

3. Routine monitoring of adverse drug events at all sites; we have already built tools that computerise adverse drug event monitoring. The way that works is the program sits over the database and identifies the signals and the pharmacists follow them up (with an estimated annual savings at Brigham are $900,000).

4. Standardising the information exchange in clinical transitions including drug administration because much evidence suggests that transitions are vulnerable to errors.

So to wrap up I think the inpatient medication system of the future is going to look something like the following. Computerised systems will give feedback to orders that will then go to the pharmacy for review by pharmacists; simple orders will be filled using automation. Point-of-care delivery devices that are linked to order entry systems will actually dispense medication. As mentioned earlier, there is not much evidence that these improve safety at all but they probably will eventually. Patients, drugs and personnel will be barcoded or the equivalent. Administration will be by smart devices, systems will collect data about all warnings & their frequencies. On the outpatients side doctors will write computerised orders that will be screened at the time they are written. Orders will go electronically to pharmacies. However at present that does not happen at all in the US and is prevented by statute in many states. Pharmacists will review orders, counsel patients and simple orders will be dealt with by automation. A.T.M. devices that provide simple fills have been very positively received by patients and seen as very threatening by pharmacists. There will be patient websites with patient medication information that will let people track their progress and report problems. There might be an option of home recording devices to monitor when people are taking their medications. This may seem Big-Brotherish but it would be very nice to know when some medications are being taken. I'll stop with a quote from Yogi Berra who said 'I don't want to make the wrong mistake'. Thank you.

Questions

NB Can I start by saying that one of the issues that this raises is the differences in the human systems and their problems between countries. For instance filling out the prescription properly and transcription errors gave you a 65% reduction in serious adverse events. In the UK neither of these are particular problems. We have very little transcription in the process and that is normally done by pharmacists. What we do not know is the relative importance of the human system and the technology but the money tends to go to solving the latter and the people aspect of the organisation is overlooked. So my question is how transferable is the technology to different human systems in organisations?

DB I think that is a great point. The main benefit of CPOE is that you put in place more sophisticated decision support. You cannot do that with paper systems but the paper systems can solve some of the problems. There is a lot of work about implementing CPOE in small community hospitals and we are doing one such study in San Diego. The issues in these institutions are different in many ways from those in academic centers, but there are also many similarities. We do need work in a variety of places because it is difficult to anticipate these differences.

AJ A very prosaic point about decision support and your renal example. One of the things that concern me is required data entry. Do people check and look up the data properly or do they guess? So in my hospital it is really hard to get the patient's weight recorded. This is usually 70 Kg and bears no relation to the actual weight.

The second point on which I would like your comments is penicillin allergies. Last year we had 15 patients who were given penicillin but who had documented penicillin allergy. Trying to get a proper debate on when is an allergy an allergy or something else is really difficult. So is getting people to carry out proper checks. The most junior doctor working on instructions from somebody else has no idea of the severity of the allergy. One of our anxieties is when we put in penicillin allergy we then have an inflation of antibiotic use because of false warnings that do not matter. Then we have resistance and expenditure problems. So it is really a big dilemma about putting all that information in and we are glad we do not have decision support. Do you recognise those issues?

DB Yes, these are both things we've struggled with. Taking weight first - we had a Weight Committee which sat for a full year and could not agree how to get weight into the computer. We finally mandated it, which has worked. We do allow people to say that they don't know and it is better to plug in 70 Kg than nothing and 25% of entries were that figure in the beginning. As you begin to use a piece of information that people increasingly value then the information gets better. We are not aware of overdosing people by misrecording weight.

Allergies and penicillin is an interesting and difficult problem and we have written two recent papers on this, with one in the Journal of Bioinformational Informatics and the second looking at all the issues around allergies. One big problem was that when a patient experiences a reaction in the hospital, only 7 - 10% of reactions were documented. The explanation was that the providers knew there had been a reaction so they were not going to give the drug again. We think it important to code the type of reaction and to get a pharmacist to go back and find out what the reaction was if listed as unknown. This is an important problem which is not so frequent as to be an overwhelming task but which does only have to be done for a few key drugs like penicillin.

MB As a relative outsider the more I listen the more I worry about the focus on errors. It seems to me that errors do not matter but the adverse drug events that follow from them do. Are there other ways of getting at severe errors? You gave me a clue when talking about ameliorating errors. It seems to me that we must not to try to stop errors but to ensure that the few errors that turn into adverse events are stopped before they really become really serious adverse events. Is this being looked at and looked at quantitatively?

DB We have looked at this quantitatively. I think it is a very important area and an area about whose importance people do not fully appreciate. This is because most errors have little potential for harm and those that do are quite different from the vast majority.

MB But isn't it the case that the more you look at the problem, set up systems, the more errors you define and increasingly the errors become less significant?

DB Some people have suggested that you should not pay any attention to the non-serious errors at all. I think that is going too far because even these cause much waste and many are avoidable. That aside, I think most attention should focus on those that can harm or have potential to do so. It is nice to focus on the latter because they are a lot more frequent so you can afford to do this. The near-misses are probably different from the errors that harm.

MB Using the analogy of driving errors we would not go out and try to eliminate driving errors. I made a dozen on the way here but what we hope is that the system, as a whole does not turn an error into an accident.

DB The auto industry has come up with some forcing functions so for example e.g. you cannot engage reverse gear without applying the brake.

TA Does anyone want to follow up on this interesting debate?

BF Several years ago a study examined if it was possible to prevent the serious errors and the problem is that you cannot predict these serious errors. So our approach is to study all types and try to take them out of the system on the basis that if we reduce all types we will reduce those with most serious consequences.

MB If we applied that across the board we would screen everyone for everything at birth instead we say that this is not worth it and just deal with the problems as they arise.

TA What David is saying is that there is a spectrum of errors but you cannot know in advance which patient is going to have a gastrointestinal bleed from NSAIDS, for example. Obviously in this case the risks are increased by the factors of age, a peptic ulcer and steroids.

RD I do think Martin is saying something important here that needs to be said clearly in this context. There is a cost to error prevention and it is a false prospectus to suppose that we could eliminate all errors and that may not be something that our politicians, health authority representatives and C.E.O.'s can stand up and say but it is something about which this community has to be very clear. We are in the business of error reduction not error elimination and I think Martin is raising the important point that at the margin how hard do we chase these things.

TA Just coming back to Martin do you see a point on that spectrum where a decision would be triggered by X number of patient deaths and Y million pounds of hospital costs a year?

MB I am sure that there are subsets of these errors that would make good economic sense (including patient benefit) & would be sensibly avoided. It seems to me that there are only a few errors where there is an immediate effect of that error. Many of the adverse effects are either because the error is repeated over time or because the sequelae of the errors are allowed to progress and I am not sure that you can go right back and avoid all errors.

TA Can we just continue this part of the debate for a short time? Martin was your point relevant here?

MP A couple of years ago when the O.W.A.M. report was published (Organisation with a Memory) there was a case highlighted where a boy received Vincristine intrathecally. In the analysis there were 13 stages in the clinical procedure where the red light flashed up, so everything happened in sequence, the boy received the drug in his spine and died a horrible death. Suppose we set a lower limit of say, one in a million and in each of the 13 stages there us an equal probability of something going wrong. Using a very crude model it turns out that there is a 30 - 40% chance of any one stage being carried out incorrectly but the collective chance of all 13 stages being simultaneously wrong is minute. What that says to me is that when there are long chains of events/steps in a procedure which result (hopefully) in a very low frequency of something extremely serious then the problem is not the very rare fatality or serious injury it is the quality and consistency of healthcare. I have worked out that there is an approximate probability in this example that 40% of all procedures will have one step wrong, another 20% will have between two & six steps wrong which still doesn't result in a serious injury. In this crude model only 40% of patients will have a complete 13 stage sequence that is done completely by-the-book. It seems to me that in addition to lowering risk we have to concentrate on the quality and consistency of healthcare and in a Utopian world every patient receiving the same procedure would have all steps in that procedure done in the same way.

DB Commercial aviation has achieved a very high safety level by having a very high degree of standardisation. It may not be possible to achieve the same degree of safety in many parts of healthcare because people are different and it is not like flying from A to B but basically I support what you are saying.

MP If I may use the analogy of safe driving using the advanced method of driving? The way to make driving safer is learning a standardised and systematised driving method and monitor the road conditions continuously. So every time you turn a corner you do so in exactly the same way but taking into account local circumstances. So becoming a safer driver can be done with enough educational effort.

AJ I think what Martin has raised is intellectually very interesting. It is too easy when you are in the field to spend time working out how to lower the error rate. There are three types of error in the UK for which you will go to prison for manslaughter - Vincristine administered intrathecally, giving potassium too fast and giving Metrexylate daily. These are all well known and understood and I do not know of any electronic prescribing system that will solve them. You do need I.T. to deal with several risk factors in the more complex risk assessments. In my view there are real healthcare risks that we do not know how to systematise such as I.V. administration using infusion pumps. We are light years away from having solutions to theoretically simple problems and everything we do with electronic prescribing will not make any difference at all.

MW James Reason, who has studied errors for years, has said that drawing an analogy between errors in healthcare and errors in industrial processes very imprecise. This is because patients do not have standardised predictable reactions and so processes in healthcare cannot be standardised and systematised as they can be in, say, aviation.

MP Isn't the point that you can kill the patient but you do not kill yourself at the same time? However a pilot who makes a mistake will also perish along with the passengers.

LF This may be stating the obvious but in all these analogies and all these examples what seems to be glaringly obvious is that we do not know enough about the way healthcare is delivered, the way the health provider interacts with the patient in order to be able to systematise any bit of it. I don't think we have done enough time and motion studies, descriptions of how care is delivered. Sometimes what happens is that you put in solutions, which create more errors. For example accidental extubations were fixed by nurses but when tubes never fall out that created lung problems. The reason was that the decision-making procedure for determining whether or not the patient could breathe unaided was inadequate. So we have to have a better understanding of what is going on before we try to reduce the errors.

TA This is an important point because documents coming out of government assume that we know enough about these errors and now is the time to fix them. The issues that Martin is raising about the relationship between error rates and adverse incidents have created much discussion. Could I pickup the last two points before Maria gives her talk?

RD Taking Linda's remarks first - there is a lot of ethnographic evidence now about how hospitals work. Although sometimes imperfect, the question is how to inject this evidence into these debates. Systems need to be designed to fit this evidence about the real organisation as opposed to the idealised or formal organisation. I was listening to Anne and David talking about the importance of physician-entry. Now my students' ethnographic studies tell me that most prescriptions are given to nurses by phone because the junior doctor is not there and there is an understanding between the junior doctor and the nurse that the next time the doctor is on the ward this will all be written on the chart. So concentrating on physician-entry starts to undermine the working practices that make the organisation tick. So if changing medications requires the presence of the junior doctor then problems may arise elsewhere.

DB The biggest perceived advantage by physicians of CPOE is that you can write orders from remote locations so we stick the terminals everywhere (including in the bathrooms).

AJ Thankyou very much. David's talk has generated a lot of very useful discussion.