Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 200
Patient Safety: Achieving a New Standard for Care 6 Adverse Event Analysis CHAPTER SUMMARY Iatrogenic injury often arises from the poor design and fragmentary nature of the health care delivery system. The detection and analysis of adverse events, both individually and in the aggregate, can reveal organizational, systemic, and environmental problems. This chapter examines the functional requirements for the two fundamental components of adverse event systems—methods for detecting adverse events and methods for analyzing such events—and the implications for data standards. The primary method of adverse event detection is voluntary reporting, and as result, most adverse events in health care today are not detected. Even if larger numbers of adverse events were detected, the information would be of limited value because of differing definitions of adverse events and varying data collection and analysis methods. There are many ways to detect adverse events—through reporting systems, document review, automated surveillance of clinical data, and monitoring of patient progress. These approaches are ultimately complementary and require a broad range of data elements covering demographic information, signs and symptoms, medications, test results, diagnoses, therapies, and outcomes. While all the available methods are complementary and each has its strengths and weak-
OCR for page 201
Patient Safety: Achieving a New Standard for Care nesses, automated surveillance is likely to become the most important source of adverse event data. More research is needed to improve the effectiveness of these detection systems and to broaden the types of adverse events that can be detected through automated triggers. Ultimately, an integrated approach, using patient safety data standards, will evolve, with electronic health record systems providing decision support at the point of care, preventing adverse events to the extent possible and facilitating the collection of reporting data when adverse events do occur. Use of adverse event systems is also aimed at identifying improved health care processes through the analysis of adverse event data. This process involves selecting and defining the adverse events to survey, defining the analysis population, collecting surveillance data, analyzing surveillance findings (identifying causal factors), and using the findings to develop interventions. The process requires standard definitions of adverse events, minimum datasets for describing the events, standard definitions of dataset variables, and standard approaches for collecting and integrating the data. INTRODUCTION An adverse event is defined as an event that results in unintended harm to the patient by an act of commission or omission rather than by the underlying disease or condition of the patient. The understanding that adverse events are common and often result from the poor design of health care delivery systems (Institute of Medicine, 2000) has led to the development of institutional adverse event systems. These systems are used to collect data on adverse events that make it possible to learn from such events and identify trends that may reveal organizational, systemic, and environmental problems. Despite these developments, most adverse events are undetected. The reason is that most health care organizations rely on voluntary reporting for the detection of adverse events (Bates et al., 2003; Cullen et al., 1995), and spontaneous reporting has been demonstrated to be a minimally effective way of detecting such events (Classen et al., 1991; Cullen et al., 1995; Jha et al., 1998). Even if larger numbers of adverse events were detected, the value of the information would be limited because existing adverse event systems use widely differing definitions, characterizations, and classification approaches.
OCR for page 202
Patient Safety: Achieving a New Standard for Care In this emerging field of study, many different definitions of adverse events are used, and a common terminology has yet to emerge. One of the more difficult problems in discussing patient safety is imprecise taxonomy, since the choice of terms has implications for how the problems related to patient safety are addressed. This imprecision makes comparison of different studies and reporting systems difficult. With few exceptions, existing studies each report data for different populations, and they frequently differ in the way they define, count, and track adverse events. Major variations in nomenclature with no fixed and accepted consensus hamper further research and application. Adverse event systems have two fundamental components—methods for detecting adverse events and methods for analyzing such events. The remainder of this chapter explores in turn the requirements for each of these components and the implications for data standards. The final section presents a future vision for the use of adverse event systems. DETECTION OF ADVERSE EVENTS: MULTIPLE APPROACHES REQUIRING A BROAD SET OF DATA ELEMENTS Sources of Adverse Event Data There are many sources of adverse event data. These include the following: Voluntary and mandatory reporting from internal hospital systems, state and federal systems, and patients themselves and their relatives. Document review, including patient charts, medical–legal documents, death certificates, coroners’ reports, complaint data, and media reports. Automated surveillance of patient treatment data, including clinical patient records, hospital discharge summaries, and Medicare claims data that may be a response to a patient injury. Monitoring of the progress of patients to anticipate conditions that could lead to adverse events or to identify adverse events and implement corrective actions. Reporting and chart review approaches identify adverse events that have already occurred. The focus is on the analysis of a subset of adverse events to determine root causes and identify improvements in care processes, ultimately improving patient safety.
OCR for page 203
Patient Safety: Achieving a New Standard for Care Automated surveillance of data and monitoring of patient progress, referred to as concurrent surveillance methods, are prospective in that they start with a clinical care process and seek to identify critical points in that process at which failures are likely to occur (e.g., when medication is prescribed). These approaches aim to prevent adverse events from happening in the first place or to quickly identify an adverse event once it has happened. For example, a concurrent surveillance system might monitor pharmacy orders for the use of antidote medications, then quickly send a trained professional to review any such case detected. The reviewer determines whether an injury or near miss has occurred and then investigates and classifies the event. More important, because such a review occurs in real time, a clinician can often intervene to prevent or ameliorate resulting harm. While prospective surveillance systems can be created and operated effectively using solely manual methods, automated methods offer more cost-effective and elegant solutions when automated clinical data systems are available. The committee believes increased attention should be devoted to concurrent surveillance methods since many common causes of adverse events are already known (Agency for Healthcare Research and Quality, 2001). Comparison of the Various Approaches for Adverse Event Detection Broad-based studies of the relative effectiveness of the detection methods outlined above have not yet been carried out. However, a number of epidemiological studies have examined the relative strengths and weaknesses of voluntary reporting, retrospective chart review, and automated surveillance for detection of adverse drug events (ADEs). Using inpatient data, Classen et al. (1991) established that automated surveillance could effectively detect ADEs at a much higher rate than voluntary reporting. Cullen et al. (1995), again using inpatient data, demonstrated that voluntary reporting uncovered only a small fraction of the ADEs identified by a nurse investigator reviewing charts daily. Jha et al. (1998) compared automated surveillance with chart review and voluntary reporting using inpatient data. They found that automated surveillance and chart review each identified many more ADEs than did voluntary reporting. They also found that automated surveillance and chart review identified different types of events. Automated surveillance was more effective at identifying events associated with changes in laboratory results, such as renal failure. Chart review was more effective at identifying events manifested primarily through symptoms, such as changes in mental state.
OCR for page 204
Patient Safety: Achieving a New Standard for Care In addition to ADEs, automated surveillance approaches have proven to be effective in identifying nosocomial infections and falls (Bates et al., 2003), although these approaches are not used routinely. Evans et al. (1986) demonstrated that computerized surveillance is at least as effective as traditional surveillance methods used by infection control practitioners for identifying hospital-acquired infections. Natural language processing was used to search radiology reports for indications that a patient fall after the second day of hospitalization was a reason for the radiological examination (Bates et al., 2003). Although there has been considerable success in using automated surveillance techniques for detecting certain types of adverse events, there will continue to be many problems that will make automated detection without manual over-read challenging. For example, in searching anesthesia records for problems arising from the management of diabetes in the peri-operative period, large amounts of redundant information might be picked up as a result of patients having iatrogenic diabetes in the peri-operative period. In conclusion, for ADEs, and probably for other types of adverse events as well, the three approaches to event detection reviewed—automated surveillance, chart review, and voluntary reporting—complement each other, with voluntary reporting being most effective at identifying potential adverse events or near misses (see Chapter 7). It is also likely that these three methods complement patient monitoring systems. Any patient safety data standards developed must be supportive of all the above detection methods. Data Requirements for Adverse Event Detection Voluntary and Mandatory Reporting To encourage people to report, voluntary reporting tends not to be prescriptive about what types of events are to be reported or what information should be supplied. Generally, just a short description of what happened is required. The recipient of the report is then tasked with creating a report for analysis purposes. Mandatory reporting systems usually specify in some detail the types of adverse events that must be reported and analyzed. For example, in New York State all hospitals (inpatient and outpatient) and freestanding clinics must report a wide range of adverse events to the New York Patient Occurrence Reporting and Tracking System (see Appendix C).
OCR for page 205
Patient Safety: Achieving a New Standard for Care Chart Review Chart review to identify possible adverse events involves reading physician and nurse progress notes and carefully examining the chart if certain indicators are present. For ADEs, these indicators might include an unexpected need for blood transfusion, the transfer of the patient to an intensive care unit, falls, explicit comments in the chart about a drug reaction, abnormal laboratory values, unexpected hypotension, and recent changes in mental state (Cullen et al., 1995). More recently, chart review has begun to use the rules incorporated in automated surveillance techniques. The Institute for Healthcare Improvement and Premier, Inc., have modified the automated surveillance methodology (Classen et al., 1991) created at LDS Hospital, Salt Lake City, to develop an ADE trigger that does not require computerized technology. The tool has about 20 triggers, outlined in Box 6-1, BOX 6-1 Triggers for Chart Review to Detect Adverse Drug Events Receiving diphenhydramine Receiving vitamin K Receiving Flumazenil Receiving Droperidol or Ondanestron Promethazine or Hydoxyzine or Trimethobenzamide or Prochlorperazine or Metoclopramine Receiving naloxone Receiving Diphenoxylate or Loperamide or Kaopectate of Pepto-Bismol Receiving sodium polystyrene Partial thromboplastin time >100 seconds International normalized ratio >6 White blood count <3,000 Serum glucose <50 Rising serum creatine Clostridium difficile positive stool Digoxin level >2 Lidocaine level >2 Gentamicin or Tobramycin levels: peak >10, trough <2 Vancomycin level >26 Theophylline level >20 Oversedation, lethargy, fall, hypotension Rash Abrupt cessation of medication Transfer to a higher level of care SOURCE: Rozich et al., 2003.
OCR for page 206
Patient Safety: Achieving a New Standard for Care relating to medications, laboratory results, and signs and symptoms. Using this tool, it takes a reviewer about 20 minutes to review an average inpatient chart. This low-tech tool has produced consistent, reliable, and relevant data, although the cost of its use is not low (Rozich et al., 2003); indeed, relative to computer screening, the cost per event is very high. Automated Surveillance of Clinical Data An epidemiological study at Brigham and Women’s Hospital using primary care data collected in 1995–1996 exemplifies some of the different approaches to automated surveillance. This study demonstrated the feasibility of identifying ADEs using automated surveillance of outpatient electronic medical records (Honigman et al., 2001). The study used four different approaches for identifying ADEs: International Classification of Diseases (ICD)-9 codes—Each patient record is scanned for ICD-9 codes that are often associated with the presence of possible ADEs. New allergies—An ADE may be present when a patient has a known allergy or a medication is listed as a new allergy. This approach requires knowing the patient’s medications, including dose, interval, and quantity. Computer detection rules—These are Boolean combinations of medical events, for example, new medication orders or laboratory results outside certain limits that suggest an ADE might be present. One such rule is “If patient is receiving phytonadione (vitamin K) AND on Coumadin, then an ADE may be present.” A list of such rules is given in Box 6-2. Data mining—Free-text searching of the electronic medical record is used to identify for each medication taken an indication of its known adverse reactions. For the drug type “diuretic,” fatigue is a potential adverse reaction and “drowsiness,” “drowsy,” and “lassitude” are some of the synonyms used instead of the word “fatigue.” Box 6-3 lists some potential adverse reactions (plus synonyms) for the diuretic drug group. Monitoring of the Progress of Patients The progress of patients can be monitored as they pass through the care process both to anticipate and protect against circumstances that could lead to adverse events and to implement corrective actions based on analysis of patient injuries discovered in the past. Monitoring systems are particularly important when addressing potential injuries of omission. One example of
OCR for page 207
Patient Safety: Achieving a New Standard for Care BOX 6-2 Rules for Detecting Possible Adverse Drug Events Using Automated Surveillance Receiving new diphenhydramine AND no diphenhydramine within last 7 days AND patient not on paclitaxel AND no blood transfusion in last 1 day AND no diphenhydramine at bedtime Receiving oral vancomycin Blood alkaline phosphatase >350 units/liter (L) Receiving phytonadione (vitamin K) AND on Coumadin Receiving ranitidine AND platelet count has fallen to less than 50 percent of previous value or below 100,000 Serum carbamazepine >12.0 micrograms/milliliter (µg/mL) Serum digoxin >1.7 nanograms (ng)/mL Serum bilirubin >10 milligrams/deciliter Serum cyclosporine >500 µg/L Serum potassium >6.5 millimoles/L Blood eosinophils >6 percent Receiving kaopectate Receiving loperamide Serum n-acetyl procainamide >20 µg/mL Serum phenytoin results >20 µg/mL Serum phenobarbital results >45 µg/mL Receiving prednisone AND diphenhydramine Serum procainamide >10 µg/mL Serum aspartate amino transferase >150 U/L AND no prior result >150 U/L Serum theophylline >20 µg/mL Serum valproate >120 µg/mL Serum quinidine >5 µg/mL Serum alanine aminotransferase >150 U/L AND no result >150 U/L in last 7 days SOURCE: Honigman et al., 2001. this approach is monitoring the progress of individual patients and groups of patients with the same condition as they pass through the care process using measures for assessing the quality of care given, such as those of the Diabetes Quality Improvement Project (DQIP). Another example is monitoring all patients at a particular point in the care continuum, such as through use of a validation system for medical prescribing based on computerized physician order entry. This section examines the general data requirements of these two examples.
OCR for page 208
Patient Safety: Achieving a New Standard for Care BOX 6-3 Sample of Triggers for Outpatient Adverse Drug Events In the case of an outpatient taking a diuretic, the following adverse reactions (and their synonyms) would serve as triggers for detection of a potential ADE: Dizziness (also syncope, lightheaded, vertigo, “wooziness”) Fainting (also blackout, loss of consciousness, syncope or near syncope, vagal reaction, vasometer collapse, vasovagal reaction, “swooning”) Fall(s) Fatigue (also drowsiness, drowsy, lassitude, lethargic, lethargy, listless, listlessness, malaise, tired) Hypokolemia (also low potassium, muscle cramps, potassium decreased, potassium deficiency) Hyponatremia (also low serum sodium) Hypotension (also arterial blood pressure decreased, low blood pressure, postural hypotension) Renal failure (also kidney shutdown, chronic renal insufficiency) Weakness (also decreased muscle strength, lack of strength) SOURCE: Bates, 2002. DQIP has developed a core set of evidence-based measures1 for assessing the quality of adult diabetes care. These measures are used to monitor the progress of individual patients and groups of patients with diabetes as they pass through the care process. The measures include those used for external accountability and internal quality improvement. The core set for accountability encompasses measures in seven areas of outpatient care: hemoglobin A1C (HbA1C) management, lipid management, urine protein testing, eye examination, foot examination, blood pressure management, and smoking cessation. The set for quality improvement includes measures in these seven areas and in two additional areas—influenza immunization and aspirin use. 1 These performance measures were initially developed by the Centers for Medicare and Medicaid Services, the Foundation for Accountability, the American Diabetes Association, and the National Committee for Quality Assurance. In 2002, DQIP merged with a performance collaboration of the American Medical Association, the Joint Commission on Accreditation of Healthcare Organizations, and the National Committee for Quality Assurance to form the National Diabetes Quality Improvement Alliance.
OCR for page 209
Patient Safety: Achieving a New Standard for Care Making the determination that a particular patient is a diabetic and applying the DQIP measures requires a number of data elements: Presence of the following data elements: insulin medication, oral hypoglycemic medication, date of ambulatory encounter, diagnosis of ambulatory encounter, medication prescribed at ambulatory encounter, date of inpatient encounter, diagnosis at inpatient encounter, date of emergency room (ER) encounter, and diagnosis at ER encounter (see Table 6-1). The two annual accountability measures for hemoglobin management—percent of patients receiving one or more HbA1C tests and percent of patients with most recent HbA1C level >9.0 percent—require the data elements HbA1C test, date of HbA1C test, and HbA1C level (see Table 6-2). Computerized physician order entry systems accept physician orders (e.g., for medications and for laboratory/diagnostic tests) electronically in lieu of the physician’s handwritten orders on a prescription pad or an order sheet. Order entry systems offer the potential to reduce medication errors through a number of validation procedures. One procedure is to determine the extent of therapeutic duplication between the newly prescribed medica- TABLE 6-1 Data Requirements for the Definition of an Adult Diabetes Patient Definition Data Requirements Those who were dispensed insulin and/or oral hypoglycemics/antihypoglycemics Insulin medication Oral hypoglycemic medication Date of ambulatory encounter Diagnosis of ambulatory encounter Medication prescribed at ambulatory encounter Date of inpatient encounter Diagnosis at inpatient encounter Date of ER encounter Patient age OR Those who had two face-to-face encounters in an ambulatory setting or non-acute inpatient setting or one face-to-face encounter in an inpatient or emergency room setting with a diagnosis of diabetes NOTE: Patients with gestational diabetes excluded. SOURCE: American Medical Association, Joint Commission on Accreditation of Healthcare Organizations, National Committee for Quality Assurance, 2001.
OCR for page 210
Patient Safety: Achieving a New Standard for Care TABLE 6-2 Data Requirements for Diabetes Quality Improvement Project Measures Performance Measure Quality Improvement Measures (per year) HbA1C management Per patient: Number of HbA1C tests received Trend of HbA1C values Across all patients: Percent of patients receiving one or more HbA1C test(s) Distribution of number of tests done (0, 1, 2, 3, or more) Distribution of most recent HbA1C value by range Lipid management Per patient: Trend of values of each test Across all patients: Percent of patients receiving at least one lipid profile (or all component tests) Distributions of most recent test values for total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, and triglycerides by range Urine protein testing Per patient: Any test for microalbuminuria received If no urinalysis or urinalysis with negative or trace urine protein, a microalbumin test received Across all patients: Percent of patients receiving any test for microalbuminuria Percent of patients with no urinalysis or urinalysis with negative or trace urine protein who received a test for microalbumin Eye examination Per patient: Dilated retinal eye exam performed by an ophthalmologist or optometrist Funduscopic photo with interpretation by an ophthalmologist or optometrist Across all patients: Percent of patients receiving a dilated retinal eye exam by an ophthalmologist or optometrist Percent of patients receiving funduscopic photo with interpretation by an ophthalmologist or optometrist
OCR for page 215
Patient Safety: Achieving a New Standard for Care and (3) use decision support tools (e.g., computerized physician order entry). Thus, comprehensive clinical and patient safety data are necessary for adverse event detection and monitoring. ANALYSIS OF ADVERSE EVENT SYSTEMS Functional Requirements Understanding an Adverse Event An outside physician calls hospital administration after one of her patients develops a near-fatal adverse reaction thought to be secondary to a drug–drug reaction to a medication prescribed in an emergency department 2 days previously. The patient safety team is assembled and after some careful detective work determines that the cause of the problem was that house staff rotating into the hospital from outside institutions were trained inadequately in use of the hospital electronic health record. Determining which of many interwoven processes should be implicated in a typical case of error is a critical step in eliminating sources of risk in the health care system. Making this determination involves asking four main questions.2 First, what is the event we are trying to eliminate? In this case we are trying to prevent patients from receiving an inappropriate drug. Second, which roles or processes must occur for this event to happen? Here, steps include recognizing a patient’s need for a medication, prescribing, filling the prescription, delivering it to the patient, and so on. Next, when did the event occur, and were there co-occurring events that could be related? Here, the fact that this reaction occurred in close proximity to the initiation of a new medication is helpful. Finally, where did the event or associated processes take place? In this case, characteristics of emergency departments, outpatient pharmacies, and homes are important. In the parlance of public health professionals, adverse event surveillance should characterize a latent3 problem within a complex system, placing the event in context rather than characterizing it as primarily the failing of a single upstream process, such as a hospital, patient, or provider. The 2 A similar approach is adopted for the analysis of a near miss; see the next chapter. 3 James Reason distinguishes two types of errors—active and latent (Reason, 1990). Active errors are associated with the performance of front-line operators, such as doctors and nurses. Latent errors result from underlying system failures.
OCR for page 216
Patient Safety: Achieving a New Standard for Care focus should be more on enhancing knowledge about risks and how to prevent them rather than on blaming, shaming, and punishing individuals. This process of understanding is the lynchpin of an effective safety culture, and its importance points to the main deficiencies in existing standards for representing potential or actual adverse medical events. One hallmark of effective analysis of adverse events is that it leads to system changes that inherently make it easier for those working in a health care delivery environment to do the job right, as opposed to a constant emphasis on more education or closer oversight—both second-hand markers for blame. Since much of health care is organized around the convenience of clinicians, however, it is important to note that interventions that alter the sequence of work flow are more challenging to implement. Addressing Errors of Omission Efforts such as those of ORC Macro4 and DQIP extend the breadth of the nomenclature needed for adverse event systems by including errors of omission. In the latter cases, in addition to characterizing errors and near-miss events by specifying what, which, when, and where, there is a need for additional elements or classification. For example, an error of commission, such as the ICD-9, Clinical Modification (CM) measure of foreign body left in during a procedure, may be adequately characterized by knowing the probable cause for leaving the foreign body in (why), the conditions under which it occurred (when and where), and the people present for the procedure (who). On the other hand, analysis of an error of omission (e.g., DQIP measure for HbA1C count), could benefit from more data about the patient. The DQIP measures indicate the specific patient data required to confirm a diagnosis of diabetes (prescription or dispensing of insulin and/or oral hypoglycemics/ antihyperglycemics during the reporting year, exclusion of women with gestational diabetes). To assess errors of omission, the dataset to compare HbA1C test rates should be expanded to include data about how the diagnosis was established, in addition to data for risk stratification or covariate analysis. 4 ORC Macro is a research, management consulting, and information technology firm based in Calverton, Maryland.
OCR for page 217
Patient Safety: Achieving a New Standard for Care Implications for Data Standards The above examples and requirements of adverse event analysis point to the need to enhance existing data standards to support adverse event reporting. The most usable of standards will include clear and unambiguous event definitions, minimum datasets that characterize the population and setting, explicit data collection processes, and methods for integrating data across systems and settings. Definitions of Terms An examination of the literature on patient safety raises many questions. Paramount among these is the problem of definitions of terms, with differing definitions of errors, adverse events, and near misses being used from one publication to another. Often, the addition of a single word creates ambiguity across the entire spectrum of reporting. For example, are potential adverse events synonymous with near misses? Do nonpreventable adverse events stem from errors? Will medication errors include actions taken by a family member who, for example, might administer insulin injections in an area with poor absorption of the medication? As with data collected for clinical trials, strict definitions of terms, including processes by which the data may be obtained, are critical to acquiring information on adverse events in a reproducible fashion. For example, each type of adverse event must be precisely defined, including examples and events that are outside the definition. Unfortunately, few standard terminologies include such definitions. DQIP represents a model for both the use of terms and the standardization of data collection. Each measure encompasses inclusion and exclusion definitions, confounding patient demographic or other data, the rationale for the importance of the measure, and a process by which the measure should be obtained. In contrast, many clinical terminologies contain terms that do not have precise definitions or conditions of use. For example, the ICD code for diabetes without ketoacidosis could refer to a patient with either Type II diabetes or well-controlled Type I diabetes. Moreover, it is not clear for many terms whether they are used to describe a point in time or a chronic condition. The ICD code for diabetes with ketoacidosis, for instance, should be applied only to a single encounter because the ketoacidosis will resolve, while the underlying diabetes will remain. In the case of a patient with Alzheimer’s disease, however, the presence of any encounter with that diagnosis passes forward to all subsequent encounters.
OCR for page 218
Patient Safety: Achieving a New Standard for Care Term definitions should also include a precise description of the groups of patients who are included and excluded. As noted above, for example, DQIP excludes patients with diagnoses such as gestational diabetes, a condition that generally resolves once the pregnancy is over. Precise definitions facilitate the prospective collection of data on similar events and an understanding of how the rate of these events is altered by interventions. They also allow large numbers of near misses and minor incidents to be analyzed (see Chapter 7). From a practical standpoint, health care workers need assistance when collecting data using such detailed definitions. For example, the appropriate definitions might appear on a computer screen when the data are being collected. Minimum Datasets To specify definitions and potential uses of terms to be included in an adverse event system, it is necessary to have minimum data requirements for the system. These minimum requirements should be stated and defined explicitly. Regardless of how an adverse event is detected, the process for reporting and analyzing is essentially the same. Data are collected on each adverse event. Using these data, a subset of adverse events (as well as near misses; see Chapter 7) is analyzed to determine their root causes and recovery procedures. Improvements in the delivery of care are then devised and implemented. Aggregate analyses of adverse events for which the more detailed analyses were not carried out may also lead to improvements in the delivery of care. Once an adverse event has been validated, the committee believes a report of the event should be a combination of narrative and coded elements. Coding is essential if large numbers of events are to be analyzed efficiently. However, any given coding system reflects a particular understanding of the key features of adverse events. Additional research can lead to new perspectives on what constitutes such features, and the availability of a narrative enables the adverse event to be recoded based on this new understanding. At a minimum, the narrative text should give a brief description of what happened and the reporter’s view of why it happened. Using the narrative and further information from the medical record, and possibly from the reporter, an adverse event record should be coded along the following dimensions (at a minimum):
OCR for page 219
Patient Safety: Achieving a New Standard for Care The discovery Who discovered/reported the event—role, not names How it was discovered The event itself What happened—the type of adverse event Where in the care process the event was discovered and/or occurred When it occurred Who was involved—functions, not names Why—the most dominant cause based on a preliminary analysis Likelihood of recurrence of similar adverse events Severity of the event Preventability of the event Ancillary information Product information (blood, devices, drugs) if involved in the adverse event Patient information, including age, gender, ethnicity, diagnoses, procedures, and comorbid conditions Detailed analysis On the basis of the above information, a decision should be made as to whether a formal root-cause analysis should be carried out (a similar decision is required to investigate a near miss; see Chapter 7). Using automated surveillance together with other detection methods will lead to the detection of a much greater number of adverse events that might warrant such an analysis than would otherwise be possible. All such events cannot feasibly be investigated. Thus if root-cause analyses are not focused on a critical subset, then (1) useless analyses will be carried out because there is no time to do them properly, and (2) effort will be devoted to performing root-cause analyses at the expense of testing and implementing real system changes that can reduce injury rates. The decision to carry out a root-cause analysis will normally depend on the following factors: The likelihood of recurrence of similar adverse events—the assessment is facilitated by access to a database of adverse events. If a similar case has recently been investigated, full root-cause analysis will have only marginal utility. The severity of the adverse event—can be assessed by direct observation.
OCR for page 220
Patient Safety: Achieving a New Standard for Care Whether the adverse event potentially represents a previously unknown problem—a judgment call drawing on the collective expertise of the patient safety team. Access to a database of adverse events also helps here. The resources available to carry out such analyses—another judgment call for the patient safety team. The potential for correction—depends on the expertise of the patient safety team. A number of risk assessment indices have been developed to help in making the decision as to whether a root-cause analysis should be carried out. Chapter 9 provides further discussion on risk assessment as well as on methods for classifying root cause data. Results Once a root-cause analysis has been completed, its results, including the following, should be fully documented and acted upon: Failed (and successful) defenses and recoveries for the patient Outcome for the patient Lessons learned and ways to improve patient safety Here there is an important difference between adverse events and near misses. Adverse events require the formal instigation of defenses (for example, a medication is discontinued, a prescription for diphenhydramine is written), whereas near misses involve built-in defenses (for example, automatic compensation through stand-by equipment; see Chapter 7). An examination of public health surveillance systems reveals the importance of refining these datasets, while health services research reminds us that collecting less structured data early in the process will reduce respondent burden and potentially remove inherent biases in the types of data collected. Therefore, it may be important to define an outcome of interest precisely and then allow knowledge gained from the reporting process (both accountability and learning) to inform system developers about data whose collection in the aggregate will be useful. As knowledge about these outcomes and known or suspected causes accrues, the inclusion of elements in a minimum dataset will evolve.
OCR for page 221
Patient Safety: Achieving a New Standard for Care Explicit Data Collection Processes Whenever possible, but especially if data are to be compared across institutions, standards for data in an adverse event system should describe how the data elements should be collected. Descriptions of patient populations that should be included or excluded and specification of whether a patient may be included multiple times during the same encounter will help clarify the group to be investigated. Data sources—including reports from health care providers’ hospital discharge summaries, emergency department notes, computer triggers, electronic clinic notes, and administrative incident reports—should be described. Uniformity of systems and applications for collecting the data (such as surveys, interviews, or claims data) will ensure that the data are comparable across time and location. As noted for the DQIP initial measure set, articulating the collection process and environment exposes cultural or other barriers to data collection (or sharing), facilitates auditing, and improves the data’s external validity. Integrating Data Across Systems and Settings Clearly, one goal of adverse event systems is to allow aggregate reporting of events for purposes of both assessing known problems before and after interventions and detecting new problems. Attention to other requirements will allow appropriate comparisons of events. Standards such as Health Level Seven (HL7) (discussed in Chapter 4) and specifications such as extensible markup language (XML) may help improve data sharing but only if the contents of these shared items are based on the same terminology—for both items and responses. Ensuring that responses are easily combined is often beyond the realm of data standards but must be considered if large datasets will be generated. For example, different systems may allow a male to be represented as “Male,” “1,” “0,” or “M.” Integrating these terms will be a challenge. FUTURE VISION Increasing Importance of Automated Triggers Looking to the future, it is likely that spontaneous reporting will be important indefinitely, especially for near misses; however, use of automated triggers is likely to grow as more computerized information becomes avail-
OCR for page 222
Patient Safety: Achieving a New Standard for Care able and automated detection becomes increasingly feasible (Bates et al., 2003). The result will be the detection of a much higher proportion of adverse events than are found today. Events may be detected through sending signals to quality personnel who can evaluate them, yet increasingly, electronic records will prompt providers to assess in real time whether signals represent an adverse event. For example, when one medication is discontinued and a prescription for diphenhydramine is written, the clinician should be asked whether the patient is allergic to the first medication. Note that it will be important to determine how much data point-of-care providers can handle, since warnings and messages may be ignored if they are too numerous, especially if their relevance is not immediately apparent. Therefore, although automated triggers have enormous potential and have been shown to be highly valuable, the committee recognizes that in the end they will be suitable for certain types of problems but not others. Definitions of Core Constructs As noted above, a fundamental and nettlesome issue has been defining the key concepts relating to patient safety—adverse events and near misses. The failure to use standard definitions for these core concepts has made comparisons among institutions challenging at best. Broad adoption of the patient data safety standards recommended by this committee (and, where necessary, further refinement of the individual constructs) would represent a major step forward in enabling meaningful aggregation and comparison of rates of such incidents from different settings. Detection of Adverse Events Using Claims Data Another approach to detecting adverse events involves using claims data (Iezzoni et al., 1994). While this approach has been fairly effective for surgical patients, it has not worked well for medical patients. However, a recent tool developed by the Agency for Healthcare Research and Quality has demonstrated excellent specificity, although its sensitivity is still quite low (Zhan and Miller, 2003). Improving the coding sets for patient safety–related conditions and events used in claims data (i.e., ICD-9) and employing incentives more broadly could represent an extremely attractive approach, especially if combined with the collection of clinical data (Classen, 2003). For example, codes to distinguish between preexisting conditions prior to a hospital admission and those predating the performance of a procedure would assist in auto-
OCR for page 223
Patient Safety: Achieving a New Standard for Care matically detecting complications of medical management. In addition, diagnoses that arise from complications due to a procedure (e.g., surgical procedure, medication order, absent safety procedure) should be associated with the procedure on claims submissions as a condition of participation. Integrated Approach to Detecting and Preventing Adverse Events If patient safety systems were integrated with electronic health record (EHR) systems, the EHR could prompt the provider to enter certain information when it appeared that an adverse event might have occurred. In the longer term, adverse event systems need to be embedded within the broader proactive hazard analysis framework—an approach to identifying and minimizing/eliminating hazards. Use of hazard analysis techniques would bring us closer to the ultimate goal of eliminating latent system defects and increasing the chances of preventing medical errors and adverse events. Such techniques have proven useful in manufacturing (failure modes and effects analysis) and the food sector (hazard analysis and critical control points) (McDonough, 2002; also see Appendix D). Proactive hazard analysis involves the following cycle: Analyzing the care process to identify for each step of the process known failure points and high-risk events. Identifying the reports/data needed to monitor the key clinical performance variables and patient outcomes and to collect information on failures and near failures (adverse events and near misses). Redesigning the care process to improve patient safety following analysis of the data collected and root-cause analyses of the more serious adverse events and near misses. Analyzing the redesigned process to identify known failure points and high-risk events for each step, paying particular attention to the hazards that may have been introduced at points where the redesigned portions of the care process intersect with the original portions. Identifying for the redesigned care process the reports/data needed to monitor the key clinical performance variables and patient outcomes and to collect information on failures and near failures (adverse events and near misses), then returning to process redesign, and so on. Detailed investigations will doubtless remain the province of patient safety officers, but detection of adverse events could be made much more
OCR for page 224
Patient Safety: Achieving a New Standard for Care efficient if front-line providers became more involved in the detection and reporting of events and the dissemination of preventive measures. Such involvement cannot, of course, be allowed to substantially delay providers in delivering care, but it nonetheless could have a major impact. Early trials of this sort of approach (Bates et al., 1998) have already demonstrated that it can be efficacious, though it has rarely been used to date. REFERENCES Agency for Healthcare Research and Quality. 2001. Making Health Care Safer: A Critical Analysis of Patient Safety Practices. Washington, DC: Agency for Healthcare Research and Quality. American Medical Association, Joint Commission on Accreditation of Healthcare Organizations, National Committee for Quality Assurance. 2001. Coordinated Performance Measurement for the Management of Adult Diabetes. Online. Available: http://diabetes-mellitus.org/diabetes.pdf [accessed January 27, 2004]. Bates, D. W. 2002. Diuretics: Adverse Reactions and Their Synonyms. Personal communication to Institute of Medicine’s Committee on Data Standards for Patient Safety. Bates, D. W., R. S. Evans, H. Murff, P. D. Stetson, L. Pizziferri, and G. Hripcsak. 2003. Detecting adverse events using information technology. J Am Med Inform Assoc 10 (2):115–128. Bates, D. W., M. A. Makary, J. M. Teich, L. Pedraza, N. M. Ma’luf, H. Burstin, and T. A. Brennan. 1998. Asking residents about adverse events in a computer dialogue: How accurate are they? Jt Comm J Qual Improv 24 (4):197–202. Classen, D. C. 2003. Medication safety: Moving from illusion to reality. JAMA 289:1154–1156. Classen, D. C., S. L. Pestotnik, R. S. Evans, and J. P. Burke. 1991. Computerized surveillance of adverse drug events in hospital patients. JAMA 266 (20):2847–2851. Cullen, D. J., D. W. Bates, S. D. Small, J. B. Cooper, A. R. Nemeskal, and L. L. Leape. 1995. The incident reporting system does not detect adverse drug events: A problem for quality improvement. Jt Comm J Qual Improv 21 (10):541–548. Evans, R. S., R. A. Larsen, J. P. Burke, R. M. Gardner, F. A. Meier, J. A. Jacobson, M. T. Conti, J. T. Jacobson, and R. K. Hulse. 1986. Computer surveillance of hospital-acquired infections and antibiotic use. JAMA 256 (8):1007–1011. Honigman, B., P. Light, R. M. Pulling, and D. W. Bates. 2001. A computerized method for identifying incidents associated with adverse drug events in outpatients. Int J Med Inf 61 (1):21–32. Iezzoni, L. I., J. Daley, T. Heeren, S. M. Foley, E. S. Fisher, C. Duncan, J. S. Hughes, and G. A. Coffman. 1994. Identifying complications of care using administrative data. Med Care 32 (7):700–715. Institute of Medicine. 2000. To Err Is Human: Building a Safer Health System. Washington, DC: The National Academies Press. Jha, A. K., G. J. Kuperman, J. M. Teich, L. Leape, B. Shea, E. Rittenberg, E. Burdick, D. L. Seger, M. Vander Vliet, and D. W. Bates. 1998. Identifying adverse drug events:
OCR for page 225
Patient Safety: Achieving a New Standard for Care Development of a computer-based monitor and comparison with chart review and stimulated voluntary report. J Am Med Inform Assoc 5 (3):305–314. Kilbridge, P., E. Welebob, and D. Classen. 2001. Overview of the Leapfrog Group Evaluation Tool for Computerized Physician Order Entry. Washington, DC: The Business Roundtable. McDonough, J. E. 2002. Proactive Hazard Analysis and Health Care Policy. New York, NY: MilBank Memorial Fund/ECRI. National Diabetes Quality Improvement Alliance. 2003. Performance Measurement Set for Adult Diabetes. Online. Available: http://www.nationaldiabetesalliance.org [accessed Jan 27, 2004]. Reason, J. 1990. Human Error. Cambridge, UK: Cambridge University Press. Rozich, J. D., C. R. Haraden, and R. K. Resar. 2003. Adverse drug trigger tool: A practical methodology for measuring medication-related harm. Qual Saf Health Care 12:194–200. Zhan, C., and R. M. Miller. 2003. Excess length of stay, charges, and mortality attributable to medical injuries during hospitalization. JAMA 290 (14):1868–1874.
Representative terms from entire chapter: