Affiliations of the authors: Harvard Medical School, Boston,
Massachusetts (DWB);
Institute for Safe Medication Practices, Huntingdon Valley,
Pennsylvania (MC);
Harvard School of Public Health, Boston (LLL);
Indiana University of Medicine, Indianapolis, Indiana
(JMO);
University of California–Los Angeles School of Medicine,
Los Angeles, California (MMS);
Massachusetts Institute of Technology, Cambridge, Massachusetts
(TS).
Correspondence and reprints: David W. Bates, MD, MSc, Division of General Medicine and Primary Care, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115; e-mail: ision of General Medicine and Primary Care, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115; e-mail: ision of General Medicine and Primary Care, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115; e-mail: <dbates@partners.org>.
|
Abstract |
|---|
Background: Increasing data suggest that error in medicine isfrequent and results in substantial harm. The recent Instituteof Medicine report (LT Kohn, JM Corrigan, MS Donaldson, eds:To Err Is Human: Building a Safer Health System. Washington,DC: National Academy Press, 1999) described the magnitude ofthe problem, and the public interest in this issue, which wasalready large, has grown.
Goal: The goal of this white paper is to describe how the frequencyand consequences of errors in medical care can be reduced (althoughin some instances they are potentiated) by the use of informationtechnology in the provision of care, and to make general andspecific recommendations regarding error reduction through theuse of information technology.
Results: General recommendations are to implement clinical decisionsupport judiciously; to consider consequent actions when designingsystems; to test existing systems to ensure they actually catcherrors that injure patients; to promote adoption of standardsfor data and systems; to develop systems that communicate witheach other; to use systems in new ways; to measure and preventadverse consequences; to make existing quality structures meaningful;and to improve regulation and remove disincentives for vendorsto provide clinical decision support. Specific recommendationsare to implement provider order entry systems, especially computerizedprescribing; to implement bar-coding for medications, blood,devices, and patients; and to utilize modern electronic systemsto communicate key pieces of asynchronous data such as markedlyabnormal laboratory values.
Conclusions: Appropriate increases in the use of informationtechnology in health care— especially the introduction of clinical decision support and better linkages in and among systems, resulting in process simplification—could result in substantial improvement in patient safety.
|
Introduction |
|---|
Our goal in this manuscript is to describe how information technologycan be used to reduce the frequency and consequences of errorsin health care. We begin by discussing the Institute of Medicinereport and the evidence that errors and iatrogenic injury area problem in medicine, and also briefly mention the issue ofinefficiency. We then define our scope of discussion (in particular,what we are considering an error) and then discuss the theoryof error as it applies to information technology, and the importanceof systems improvement. We then discuss the effects of clinicaldecision support, and errors generated by information technology.That is followed by management issues, the value proposition,barriers, and recent developments on the national front. Weconclude by making a number of evidence-based general and specificrecommendations regarding the use of information technologyfor error prevention in health care.
|
The Institute of Medicine Report and Iatrogenesis |
|---|
Errors in medicine are frequent, as they are in all domains in life. While most errors have little potential for harm, some do result in injury, and the cumulative consequences of error in medicine are huge.
When the Institute of Medicine (IOM) released its report ToErr is Human: Building a Safer Health System in November 1999,1the public response surprised most people in the health carecommunity. Although the report's estimates of more than a millioninjuries and nearly 100,000 deaths attributable to medical errorsannually were based on figures from a study published in 1991,they were news to many. The mortality figures in particularhave been a matter of some public debate2,3 although most agree that whatever the number of deaths is, it is too high.
The report galvanized an enormous reaction from both governmentand health care representatives. Within two weeks, Congressbegan hearings and the President ordered a government-wide feasibilitystudy, followed in February by a directive to governmental agenciesto implement the IOM recommendations. During this time, professionalsocieties and health care organizations have begun to re-assesstheir efforts in patient safety.
The IOM report made four major points—the extent of harm that results from medical errors is great; errors result from system failures, not people failures; achieving acceptable levels of patient safety will require major systems changes; and a concerted national effort is needed to improve patient safety. The national effort recommended by the IOM involves all stakeholders—professionals,health care organizations, regulators, professional societies,and purchasers. Health care organizations are called on to workwith their professionals to implement known safe practices andset up meaningful safety programs in their institutions, includingblame-free reporting and analysis of serious errors. Externalorganizations—regulators, professional societies, and purchasers—are called on to help establish standards and best practice for safety and to hold health care organizations accountable for implementing them.
Some of the best available data on the epidemiology of medicalinjury come from the Harvard Medical Practice Study.4 In that study, drug complications were the most common adverse event (19 percent), followed by wound infections (14 percent) and technical complications (13 percent). Nearly half the eventswere associated with an operation. Most work on prevention todate has focused on adverse drug events and wound infections.Compared with the data on inpatients, relatively few data onerrors and injuries outside the hospital are available, althougherrors in follow-up5 and diagnosis are probably especially important in non-hospital settings.
While the IOM report and Harvard Medical Practice Study deal primarily with injuries associated with errors in health care, the costs of inefficiencies related to errors that do not result in injury are also great. One example is the effort associated with "missed dose" medication errors, when a medication dose is not available for a nurse to administer and a delay of at least two hours occurs or the dose is not given at all.6 Nurses spend a great deal of time tracking down such medications. Although such costs are harder to assess than the costs of injuries,they may be even greater.
|
Scope of Discussion |
|---|
In this paper, we are discussing only clear-cut errors in medicalcare and not suboptimal practice (such as failure to followa guideline). Clearly, this is not a dichotomous distinction,and some examples may be helpful. We would consider a spongeleft in the patient after surgery an error, whereas an inappropriateindication for surgery would be suboptimal practice. We wouldconsider it an error if no postoperative anticoagulation wereused in patients in whom its benefit has clearly been demonstrated(for example, patients who have just had hip surgery). However,we would not consider it an error if a physician failed to followa pneumonia guideline and prescribed a commonly used but suboptimalantibiotic, even though adherence to such guidelines will almostcertainly improve outcomes. Although we believe that informationtechnology can play a major role in both domains, we are notaddressing suboptimal practice in this discussion.
|
Theory of Error |
|---|
Although human error in health care systems has only recently received great attention, human factors engineering has been concerned with error for several decades. Following the accident at Three Mile Island in the late 1970s, the nuclear power industry was particularly interested in human error as part of human factors concerns, and has produced a number of reports on the subject.7 The U.S. commercial aviation sector is also very interested in human error at present, because of massive overhaul of the air traffic control network. A few excellent books on human error generally are available.8–10
While it is easy and common to blame operators for accidents, investigation often indicates that an operator "erred" because the system was poorly designed. Testimony of an operator of the Three Mile Island nuclear power plant in a 1979 Congressional hearing11 makes the point, " If you go beyond what the designers think might happen, then the indications are insufficient, and they may lead you to make the wrong inferences....[H]ardly any of the measurements that we have are direct indications of what is going on in the system."
The consensus among man–machine system engineers is that we should be designing our control rooms, cockpits, intensive care units, and operating rooms so that they are more "transparent"—thatis, so that the operator can more easily "see through" the displaysto the actual working system, or "what is going on." Situationalawareness is the term used in the aviation sector. Often theoperator is locked into the dilemma of selecting and slavishlyfollowing one or another written procedure, each based on ananticipated causality. The operator may not be sure what procedure,if any, fits the current not-yet-understood situation.
Machines can also produce errors. It is commonly appreciated that humans and machines are rather different and that the combinationof both thus has greater potential reliability than either alone.However, it is not commonly understood how best to make thissynthesis. Humans are erratic, and err in surprising and unexpectedways. Yet they are also resourceful and inventive, and theycan recover from both their own and the equipment's errors increative ways. In comparison, machines are more dependable,which means they are dependably stupid when a minor behaviorchange would prevent a failure in a neighboring component frompropagating. The intelligent machine can be made to adjust toan identified variable whose importance and relation to othervariables are sufficiently well understood. The intelligenthuman operator still has usefulness, however, for he or shecan respond to what at the design stage may be termed an "unknownunknown" (a variable which was never anticipated, so that therewas never any basis for equations to predict it or computersand software to control it).
Finally, we seek to reduce the undesirable consequences of error,not error itself. Senders and Moray10 provide some relevant comments that relate to information technology: "The less often errors occur, the less likely we are to expect them, and the more we come to believe that they cannot happen.... It is something of a paradox that the more errors we make, the better we will be able to deal with them." They comment further that, "eliminating errors locally may not improve a system and might cause worse errors elsewhere."
A medical example relating to these issues comes from the workof Macklis et al. in radiation therapy12; this group has used and evaluated the safety record of a record-and-verify linear accelerator system that double-checks radiation treatments.This system has an error rate of only 0.18 percent, with alldetected errors being of low severity. However, 15 percent ofthe errors that did occur related to use of the system, primarilybecause when an error in the checking system occurred, the humanoperators assumed the machine "had to be right," even in theface of important conflicting data. Thus, the Macklis groupexpressed concern that over-reliance on the system could resultin an accident. This example illustrates why it will be vitalto measure to determine how systems changes affect the overallrate of not only errors but accidents.
|
Systems Improvement and Error Prevention |
|---|
Although the traditional approach in medicine has been to identifythe persons making the errors and punish them in some way, ithas become increasingly clear that it is more productive tofocus on the systems by which care is provided.13 If these systems could be set up in ways that would both make errors less likely and catch those that do occur, safety might be substantially improved.
A system analysis of a large series of serious medication errors(those that either might have or did cause harm)13 identified 16 major types of system failures associated with these errors. Of these system failures, all of the top eight could have been addressed by better medical information.
Currently, the clinical systems in routine use in health care in the United States leave a great deal to be desired. The health care industry spends less on information technology than do most other information-intensive industries; in part as a result, the dream of system integration been realized in few organizations. For example, laboratory systems do not communicate directly with pharmacy systems. Even within medication systems, electronic links between parts of the system—prescribing, dispensing, and administering—typically do not exist today. Nonetheless, real and difficult issues are present in the implementation of information technology in health care, and simply writing a large check does not mean that an organization will necessarily get an outstanding information system, as many organizations have learned to their chagrin.
Evaluation is also an important issue. Data on the effects of information technology on error and adverse event rates are remarkably sparse, and many more such studies are needed. Although such evaluations are challenging, tools to assess the frequency of errors and adverse events in a number of domains are now available.14–19 Errors are much more frequent than actual adverse events (for medication errors, for example, the ratio in one study6 was 100:1). As a result, it is attractive from the sample size perspective to track error rates, although it is important to recognize that errors vary substantially in their likelihood of causing injury.20
|
Clinical Decision Support |
|---|
While many errors can be detected and corrected by use of humanknowledge and inspection, these represent weak error reductionstrategies. In 1995, Leape et al.13 demonstrated that almost half of all medication errors were intimately linked with insufficient information about the patient and drug. Similarly, when people are asked to detect errors by inspection, they routinely miss many.21
It has recently been demonstrated that computerized physician order entry systems that incorporate clinical decision support can substantially reduce medication error rates as well as improve the quality and efficiency of medication use. In 1998, Bates et al.20 found in a controlled trial that computerized physician order entry systems resulted in a 55 percent reduction in serious medication errors. In another time series study,22 this group found an 83 percent reduction in the overall medication error rate, and a 64 percent reduction even with a simple system.Evans et al.23 have also demonstrated that clinical decision support can result in major improvements in rates of antibiotic-associated adverse drug events and can decrease costs. Classen et al.24have also demonstrated in a series of studies that nosocomialinfection rates can be reduced using decision support.
Another class of clinical decision support is computerized alertingsystems, which can notify physicians about problems that occurasynchronously. A growing body of evidence suggests that suchsystems may decrease error rates and improve therapy, therebyimproving outcomes, including survival, the length of time patientsspend in dangerous conditions, hospital length of stay, andcosts.25–27 While an increasing number of clinical information systems contain data worthy of generating an alert message, delivering the message to caregivers in a timely way has been problematic. For example, Kuperman et al.28 documented significant delays in treatment even when critical laboratory results were phoned to caregivers. Computer-generated terminal messages, e-mail, and even flashing lights on hospital wards have been tried.29–32 A new system, which transmits real-time alert messages to clinicians carrying alphanumeric pagers or cell phones, promises to eliminate the delivery problem.33,34 It is now possible to integrate laboratory, medication, and physiologic data alerts into a comprehensive real-time wireless alerting system.
Shabot et al.33,34 have developed such a comprehensive system for patients in intensive care units. A software system detect alerts and then sends them to caregivers. The alert detection system monitors data flowing into a clinical information system. The detector contains a rules engine to determine when alerts have occurred.
For some kinds of alert detection, prior or related data are needed.
When the necessary data have been collected, alerting algorithms
are executed and a decision is made as to whether an alert has
occurred (Figure 1
).
The three major forms of critical event detection are critical
laboratory alerts, physiologic "exception condition" alerts,
and medication alerts. When an alert condition is detected,
an application formats a message and transmits it to the alphanumeric
pagers of various recipients, on the basis of a table of recipients
by message type, patient service type, and call schedule. The
message is sent as an e-mail to the coded PIN (personal identification
number) of individual caregivers' pagers or cell phones. The
message then appears on the device's screen and includes appropriate
patient identification information (Figure 2).
|
|
Alerts are a crucial part of a clinical decision support system,35and their value has been demonstrated in controlled trials.27,35In one study, Rind et al.27 alerted physicians via e-mail to increases in serum creatinine in patients receiving nephrotoxic medications or renally excreted drugs. Rind et al. reported that when e-mail alerts were delivered, medications were adjusted or discontinued an average of 21.6 hours earlier than when no e-mail alerts were delivered. In another study, Kuperman et al.35 found that when clinicians were paged about "panic" laboratory values, time to therapy decreased 11 percent and mean time to resolution of an abnormality was 29 percent shorter.
As more and different kinds of clinical data become available electronically,
the ability to perform more sophisticated alerts and other types
of decision support will grow. For example, medication-related,
laboratory, physiologic data can be combined to create a variety
of automated alerts. (Table 1
shows a sample of those currently included in the system used
at Cedars-Sinai Medical Center, Los Angeles, California.) Furthermore,
computerization offers many tools for decision support, but
because of space limitations we have discussed only some of
these; Among the others are algorithms, guidelines, order sets,
trend monitors, and co-sign forcers. Most sophisticated systems
include an array of these tools.
|
|
Errors Generated By Information Technology |
|---|
Although information technology can help reduce error and accidentrates, it can also cause errors. For example, if two medicationsthat are spelled similarly are displayed next to each other,substitution errors can occur. Also, clinicians may write anorder in the wrong patient's record.
In particular, early adopters of vendor-developed order entry have reported significant barriers to successful implementation, new sources of error, and major infrastructure changes that have been necessary to accommodate the technology. The order entry process with many computerized physician order entry systems currently on the market is error-prone and time-consuming. As a result, prescribers may bypass the order entry process totally and encourage nurses, pharmacists, or unit secretaries to enter written or verbal drug orders. Also, most computerized physician order entry systems are separate from the pharmacy system, which requires double entry of all orders. This may result in electronic/computer-generatedmedication administration records (MARs) that are derived fromthe order entry system database, not the pharmacy database,which can result in discrepancies and extra work for nursesand pharmacists.
Furthermore, many computerized physician order entry systems lack
even basic screening capabilities to alert practitioners to
unsafe orders relating to overly high doses, allergies, and drug–drug
interactions. While visiting hospitals in 1998, representatives
of the Institute for Safe Medication Practices (ISMP) tested
pharmacy computers and were alarmed to discover that many failed
to detect unsafe drug orders. Subsequently, ISMP asked directors
of pharmacy in U.S. hospitals to perform a nationwide field
test to assess the capability of their systems to intercept
common or serious prescribing errors.36
To participate, pharmacists set up a test patient in their computer
system, then entered actual physician prescription errors that
had actually led to a patient's death or serious injury during
1998 (Table 2).
Only a small number of even fatal errors were detected by current
detection methods.
|
These anecdotal data suggest that current systems may be inadequateand that simply implementing the current off-the-shelf vendorproducts may not have the same effect on medication errors thathas been reported in research studies. Improvement of vendor-basedsystems and evaluation of their effects is crucial, since theseare the systems that will be implemented industry-wide.
|
Management Issues |
|---|
A major problem in creating the will to reduce errors has beenthat administrators have not been aware of the magnitude ofproblem. For example, one survey showed that, while 92 percentof hospital CEOs reported that they were knowledgeable aboutthe frequency of medication errors in their facility, only 8percent said they had more than 20 per month, when in fact allprobably had more than this.37 Probably in part as a result, the Advisory Board Company found that reducing clinical error and adverse events ranked 133rd when CEOs were asked to rank items on a priority list.38 A number of efforts are currently under way to increase the visibility of the issue. For example, a video about this issue, which was developed by the American Hospital Association and the Institute for Healthcare Improvement, has been sent to all hospital CEOs in the United States, and a number of indicators suggest that such efforts may be working.
|
The Value Proposition |
|---|
For information technology to be implemented, it must be clearthat the return on investment is sufficient, and far too fewdata are available regarding this in health care. Furthermore,there are many horror stories of huge investments in informationtechnology that have come to naught.
Positive examples relate to computer order entry. At one largeacademic hospital, the savings were estimated to be $5 millionto $10 million annually on a $500 million budget.39 Another community hospital predicts even larger savings, with expected annual savings of $21 million to $26 million, representing about a tenth of its budget.40 In addition, in a randomized controlled trial, order entry was found to result in a 12.7 percent decrease in total charges and a 0.9 day decrease in length of stay.41Even without full computerization of ordering, substantial savingscan be realized: data from LDS Hospital23 demonstrated that a program that assisted with antibiotic management resulted in a fivefold decrease in the frequency of excess drug dosages and a tenfold decrease in antibiotic-susceptibility mismatches, with substantially lower total costs and lengths of stay.
|
Barriers |
|---|
Despite these demonstrated benefits, only a handful of organizationshave successfully implemented clinical decision support systems.A number of barriers have prevented implementation. Among theseare the tendency of health care organizations to invest in administrativerather than clinical systems; the issue of "silo accounting,"so that benefits that accrue across a system do not show upin one budget and thus do not get credit; the current financialcrisis in health care, which has been exacerbated by the BalancedBudget Amendment and has made it very hard for hospitals toinvest; the lack, at many sites, of leaders in information technology;and the lack of expertise in implementing systems.
One of the greatest barriers to providing outstanding decisionsupport, however, has been the need for an extensive electronicmedical record system infrastructure. Although much of the datarequired to implement significant clinical decision supportis already available in electronic form at many institutions,the data are either not accessible or cannot be brought togetherto be used in clinical decision support because of format andinterface issues. Existing and evolving standards for exchangeof information (HL7) and coding of this data are simplifyingthis task. Correct and consistent identification of patients,doctors, and locations is another area in which standards areneeded. Approaches to choosing which information should be codedand how to record a mixture of structured coded informationand unstructured text are still immature.
Some organizations have moved ahead with adopting such standardson their own, and this can have great benefits. For example,a technology architecture guide was developed at Cedars-SinaiMedical Center to help ensure that its internal systems anddatabases operate in a coherent manner. This has allowed themto develop what they call their "Web viewing system," whichallows clinicians to see nearly all results on an Internet platform.Many health care organizations are hamstrung, because they haveimplemented so many different technologies and databases thatinformation stays in silos.
A second major hurdle is choosing the appropriate rules or guidelinesto implement. Many organizations have not developed processesfor developing and implementing consensus choices in their physiciangroups. Once the focus has been determined, the organizationmust determine exactly what should be done about the selectedproblem. Regulatory and legal issues have also prevented vendorsfrom providing this type of content. Finally, despite good precedentsfor delivering feedback to clinicians for simple decision support,changing provider behavior for more complex aspects of careremains challenging.
|
The National Picture |
|---|
A national commitment to safer health care is developing. Althoughit is too soon to determine how it will "play out" (the initialfixation on mandatory reporting has been an unwelcome diversion,for example), it seems clear that many stakeholders have a realinterest in improving safety. Doctors and other professionalsare in the interesting position of being expected to be bothleaders in this movement and the recipients of its attention.Already a national coalition involving many of the leading purchasers,the Leapfrog Group, which includes such companies as GeneralMotors and General Electric, have announced their intentionto provide incentives to hospitals and other health care organizationsto implement safe practices.42 One of the first of these practices will be the implementation of computerized physician order entry systems. Similarly, a recent Medicare Patient Advisory Commission report suggested that that the Health Care Financing Administration consider providing financial incentives to hospitals that adopt physician order entry systems.43 The Agency for Healthcare Research and Quality has received $50 million in funding to support error reduction research, including information technology–related strategies. California recently passed a law mandating that non-rural hospitals implement computerized physician order entry or another application like it by 2005.44 Clearly, many look to automation to play a major role in the redesign of our systems.
|
Recommendations |
|---|
Recommendations for using information technology to reduce errorsfall into two categories—general suggestions that are relevant across domains, and very specific recommendations. It is important to recognize that these lists are not exhaustive, but they do contain many of the most important and best-documented precepts. Although many of these relate to the medication domain, this is because the best current evidence is available for this area; we anticipate that information technology will eventually be shown to be important for error reduction across a wide variety of domains, and some evidence is already available for blood products, for example.45,46 The strength of these recommendations is based on a standard set of criteria for levels of evidence.47For therapy and prevention, evidence level 1a represents multiplerandomized trials, level 1b is an individual randomized trial,level 4 is case series, and level 5 represents expert opinion.
General Recommendations
|
Footnotes |
|---|
This work is based on discussions at the AMIA 2000 Spring Congress; May 23–25, 2000; Boston, Massachusetts.
|
References |
|---|
This article has been cited by other articles:
Abstract
Sample of Wireless Alerts Currently in Use at Cedars-Sinai Medical Center,
Los Angeles, California