Review the case study in Chapter 5.
After the initial patient misidentification, what systemic reasons can you identify that lead to subsequent caregivers not recognizing that the patient transferred to, prepped for, and undergoing surgery was not the right patient? Use the three components of root cause analysis to evaluate the following.
- How significant was the role of data collection and data analysis in measuring the chronology of events? Explain.
- From your perspective, what steps should have been taken to avoid this latent medical error?
- Evaluate why and how humans produce errors and what precautions should have been taken in this case study.
For additional details, please refer to the Module One Case Study Guidelines and Rubric document.
Single-Event Analysis There are three primary components of an RCA: data collection, data anal- ysis, and corrective or preventive action. RCA has an obligation to iden- tify corrective action and follow-up to ensure that change is undertaken and to determine the impact of the recommended changes. The difference between an analysis of an individual case and one analyzing multiple events is primarily focus. Single-case analysis relies more on data collection while multiple-event analysis is driven by data analysis. For an individual case, the analyst may need to talk to all or the majority of individuals who partici- pated in a patient’s treatment. On the other hand, the analyst who has access to a database of information from multiple events may elect to mine the data to identify the most at-risk parts of a process.
Individual case analysis requires knowledge of the appropriate ques- tions to ask and the interpersonal skill to effectively ask questions. The inter- viewer must be able to elicit trust and gain the cooperation of her subjects while exploring potentially sensitive issues dealing with job performance, teamwork, and other issues. For multiple-event analysis, data analysis skills are required to identify commonalities. Creating Pareto charts and histo- grams is an effective way to produce information to focus quality improve- ment efforts. Data analysis, including the creation of frequency tables, will be demonstrated when multiple-event analysis is discussed. Both types of analysis require the analyst to understand the RCA process and its goals.
It makes intuitive sense to explore how a single event is examined before expanding into the analysis of multiple events, since individual cases provide the raw data for multiple-event analysis. Table 5.3 summarizes the three basic steps in RCA: data collection, data analysis, and initiation of corrective action.
Table 5.3 Single Case Root Cause Analysis Processes 1. Identify event or improvement opportunity 2. Assemble team based on knowledge and authority 3. Build timeline of activities 4. Analyze process and identify potential causes 5. Identify root cause(s) 6. Identify potential corrective or preventive changes 7. Select and implement change 8. Follow-up: Was change implemented, continued, and effective?
Table 5.3 synthesizes the root cause process recommended by four dif- ferent authors (Knudsen et al. 2007; Mills et al. 2005; Perkins et al. 2005; Rooney and Vanden Heuvel 2004). These authors propose processes rang- ing from four to eleven steps. Table 5.3 highlights the transitions from data collection to data analysis to corrective or preventive action. The authors using more steps to describe the process provide the reader with a greater understanding of what is required while the four-step version defines the fundamentals. The reader may again benefit from recognizing the differ- ent terminology: data collection is variously described as flowcharting the current process, describing the event, collecting information, or collecting data. The point is that no matter which process is selected, they all cover the same ground. Comparing and contrasting these approaches will give the reader greater understanding of RCA.
The first task in RCA is to understand what has occurred. This requires describing the outcome and the process leading up to an event in sufficient detail to identify the reasons why the outcome occurred. This is a com- plex task in treatment processes in which a patient presents with multiple conditions, is seen by several care givers, and receives dozens of tests and procedures. Establishing the timeline of activities is an essential first step when analyzing a process. The timeline should cover what was done, where it occurred, who made decisions, who provided care, and who was present during the activity. Timelines are valuable because they place each event in the context of what came before and after, emphasizing that medical care is a process rather than a series of discrete steps. Flowcharting is a tool that increases comprehension of a process, illu- minates discrepant views of the process and serves as a tool to resolve any divergent views, and records the present state of the system. A primary ben- efit of documentation is to identify different treatment protocols that may not be recognized by those working in the system.
Flowcharts should iden- tify all the significant actions and care givers in a process, where actions or decisions were made, the entry of inputs into the system, the outputs produced, and the results achieved. After documenting the process, the essential task is to disaggregate the care episode into its constituent steps to identify where the error arose, the potential causes of the problem, and the points at which the error could have been prevented (before), detected (concurrent), or corrected (after). Unlike timeline construction, where we were concerned with what was tak- ing place, discovery of root causes requires the delicate process of docu- menting every event including who was present when the error arose, when and where it occurred, and why it occurred. After identifying an active cause, the analyst should explore the latent causes that could have contribute to failure.
RCA can be viewed as asking a structured series of questions to deter- mine why an event or outcome occurred and to trace it back to its root causes. Various methodologies have been advanced to direct the search for root causes. Three techniques are described next, beginning with the simplest and most intuitive approach and progressing to a more structured method.
Data Collection Techniques
The Five Whys
The National Health Service (NHS) in England suggests asking why five times when an undesirable event occurs. For example, if a patient receives an incorrect medicine, the sequence of questions could be:
Question 1: Why did the patient receive the wrong medicine? (an obvious question) Response 1: The prescription was wrong. (one of dozens of possible answers)
Question 2: Why was the prescription wrong? (an obvious follow-up ques- tion in response to the first answer) Response 2: Physician made incorrect pharmaceutical decision.
Question 3: Why did the physician make the wrong decision? Response 3: Incomplete patient chart.
Question 4: Why was the patient chart incomplete? Response 4: Physician assistant did not record test results.
Question 5: Why didn’t the physician assistant record the test results? Response 5: Tests results were phoned to secretary and secretary did not tell physician assistant (NHS 2013a).
The first why defines the problem, and the first response establishes the overall direction of the inquiry. Successive whys flow naturally from the received responses. Successive responses limit the potential cause and suggest the follow-up question that should be asked. Analysts must rec- ognize that the initial response, and perhaps even the initial question, may not be appropriate and that multiple attempts using the Five Whys may be required.
Five questions may or may not uncover the root cause. One may be able to discern a cause using fewer than five questions, and at other times more than five questions will be needed (especially when more than one cause is at work). The general rule holds that an analyst will be reasonably close to identifying a root cause after framing five why questions. More important than the general rule is the recognition that identifying a root cause need not be an overwhelming and time-consuming process but rather is based on a straightforward and effective questioning technique. Information- gathering techniques are the first skill an analyst must develop.
In the preceding scenario, a narrow interpretation could be that the secretary made an error in failing to inform the physician assistant—the active error. In many cases, the identified cause may be both an active and a latent error. Is illegible handwriting an active error? Yes. Is an organiza- tional culture that tolerates illegibility a latent error? Yes.
A narrow interpretation could lead to the conclusion that the secretary must be more careful. A broader view would ask: Why does a patient care system depend on such a tenuous transmission of information, a sixth why. Should the telephone be used for relaying clinical information? If so, should a secretary relay clinical information? If so, should there be a defined proce- dure for recording the call, recording the relayed data in the patient chart, and notifying care givers? These questions attempt to uncover latent errors that may contribute to medical error.
System weaknesses are often difficult to identify because people have grown accustomed to doing things in a particular way, and these processes often work. The problem for root cause identification is that accepted prac- tices work the majority of the time but can fail when a confluence of events, issues, or pressures arise. The analyst’s job is to identify unsafe practices, recognize the conditions under which practices may fail, and design and implement safeguards to handle these conditions.
The Why-Why Diagram
A variation of the Five Why technique is the Why-Why diagram. The prob- lem with the Five Whys is that it is linear; a problem is identified and sub- sequent exploration follows a single path. The administration of the wrong medication can arise from a multitude of causes, so a wrong prescription is only one of several explanations. The patient may have forgotten to inform staff of a previous allergic reaction, the prescription could have been mis- read, an unprescribed drug could have been administered, or there may be some other cause.
The primary difference between the Five Whys and the Why-Why dia- gram is the former asks why an event occurred; the latter takes a broader view by exploring how an event could arise. The Why-Why diagram, rather than seeking a single cause, attempts to expand understanding of the many potential causes that could lead to an undesirable event. To construct a Why-Why diagram, the analyst should first ask why an event occurred
(identical to the Five Why approach) but instead of being satisfied with a single answer would continue to inquire into other possible explanations: What else could have caused the event? One can think of this as pursu- ing the most likely causes for an event instead of a one-track, five-question exercise. An analyst could find himself with dozens, if not hundreds, of potential causes. The potential causes are best viewed as branches in a tree diagram where a single event could result from one or more causes.
The Why-Why diagram, Figure 5.4, demonstrates how a wrong medica- tion could arise from various causes. For display purposes, the example has been limited to two responses to each question. Accordingly, the two main causes of a wrong medication are: the prescription was wrong (a rule or knowledge error) and the wrong medication was administered (a skill error or violation). This diagram shows that when only two responses are allowed per question, five levels of questioning produces 32 potential causes. Each question doubles the number of causes; 2 → 4 → 8 → 16 → 32. If each question produces three answers, five levels of questioning produces 243 potential causes. The general rule is that the number of potential causes will equal the number of responses raised to the fifth power (potential causes = number of responses5). The Why-Why diagram is more difficult to manage than the Five Whys but does more justice to clinical situations where more than one cause can produce an undesired outcome.
The procedure for the Five Whys and Why-Why diagram is question and response: an analyst follows each response to one or more possible causes to pursue the root cause. The Why-Why diagram is more time con- suming given its broader focus (what could have occurred versus what did occur); however, the greater upfront investment can lead to higher bene- fits later. By recognizing multiple potential causes, corrective or preven- tive action can be developed to address more than one weakness at a single point in time (rather than waiting for each to occur before action is taken). A second advantage is that if the analysis arrives at a wrong conclusion (X was caused by Y when X was actually the result of Z), the team does not have to restart the data collection process but can return to the Why-Why diagram to assess why the event reoccurred. Did the event result from one of the previously recognized potential causes that was not pursued?
The Why-Why diagram demonstrates that a single event may be the result of multiple causes. The next technique begins with the premise that undesired outcomes are generally the result of a combination of factors and that one factor is insufficient to explain an adverse event. Normally reliable processes may fail when they are placed under extreme pressure. The goal of the third technique is to identify why normally reliable processes fail, to discover what change or changes produced the failure
The Is–Is Not Matrix
Kepner and Tregoe (1965) developed a more structured and general sys- tem of searching for root causes that is not dependent upon the responses offered to prior questions. The Is–Is Not matrix approach to problem spec- ification is based on the assumption that a process has a known standard of performance and that a deviation from the standard has been observed. Deviation provides the impetus for a precise defining of the problem: who or what is affected, where did it occur, when did it occur, and how extensive is the deviation?
The process begins by answering the following series of questions:
• What is the deviation: who or what was affected?
• Where was the deviation observed?
• When did the deviation appear?
• How big is the deviation or how many deviations are observed?
These questions provide half the information required to identify the cause of a deviation. The second step is to ask the opposite questions: Who is not affected, and where and when does the deviation not arise? The analyst must identify where deviations are not observed. This could be the same pro- cess operating at a different point in time or an identical or similar process operating at a different location. The assumption is that there are identifiable characteristics which distinguish deviations from the outcomes of processes that meet performance standards. The Kepner-Tregoe system requires ana- lysts to define the problem (what is the error, where is the error) and contrast it with situations in which the same result could arise but did not.
The two sets of data specifying what is a problem and what is not a prob- lem facilitate data analysis. Can an error be traced to readily identifiable differ- ences between situations in which the problem arose and those in which it is not observed? These factors examine differences in what is occurring (process), who is affected (the patient), the individuals providing care, the location of care, and the time or day of care. The Is–Is Not matrix is a balanced approach to the discovery of active and latent errors as it documents the error and the context in which it arose. Kepner and Tregoe’s approach to cause discovery is more systematic than the Five Whys or the Why-Why diagram in that it stud- ies the specific event, where deviations from expectations occur, and situations in which similar errors could arise but did not. Rather than an ad hoc, follow- the-data-where-it-leads approach, the Is–Is Not Matrix recommends a review of specific factors that could contribute to variation in performance.
Step one in the Is–Is Not procedure is to clearly state the problem so everyone on the analysis team understands what they are attempting to resolve. Step two divides the problem into five parts. A review of the litera- ture shows that specific applications add and subtract elements from the basic five-part framework. The framework is shown in Figure 5.5; column 2 records the basic information on the problem. Similar to newspaper writ- ing, where reporters are supposed to answer who, what, where, when, why, and how, the goal of the Is–Is Not matrix is to assemble the relevant facts on what has occurred (Is).
Data Collection Analysis Figure 5.5 Is–Is Not Template Unlike with a newspaper article, the analyst also reports nonoccur- rences (Is Not). Step three requires analysts to explore situations in which the problem is not observed (column 3, the Is Not column). Undesirable outcomes can result from differences in patients as well as differences in care provided. Severely ill patients of advanced age with multiple comor- bidities should have lower outcomes than younger or less ill patients. So the question arises, is the problem a medical error or a predictable outcome given the patient’s condition? Outcomes also depend on treatment. We must recognize differences in the use of equipment (catheters, ventilators, etc.) and the duration of their use. Step three documents cases in which an undesirable outcome could have arisen but did not.
Step four, which uses the Difference column (column 4), encourages the analyst to contemplate whether the characteristics—who is and is not affected, who was and was not part of care process, where the problem did and did not arise, and when the problem did and did not occur—point to Step 1 Problem Statement: Step 2 Is Step 3 Is Not Step 4 Difference Who (what) is affected? What is occurring? Customers, materials, equipment… How important is the problem or what is the extent of the problem? Frequency, seriousness Who is part of the process? By whom, near whom? Where does the problem occur? Location When does the problem occur? Date, day of week, time of day, proximity to other events the underlying causes of the event. Are differences between occurrences and nonoccurrences trivial, or are they causal?
The analyst must ask, if the factor were changed—if a different patient, provider, time, or location was involved—would the result be the same? An affirmative answer suggests a system problem (the failure would occur independent of who was involved) or suggests an earlier failure (anyone working with the information would reach the same conclusion). If failure can be traced to a particular person (others would not have performed this way) or time (it was due to the fact that this occurred on the overnight shift or on the weekend), then the solution is somewhat easier: correct the broken component.
The Is–Is Not Matrix functions similarly to control groups and experimental groups in random clinical trials, where researchers attempt to hold a number of factors constant, allowing them to attribute any differ- ence in outcomes to one factor that is different between the two groups. For example, we may be able to attribute a reduction in nosocomial infections to the prophylactic use of antibiotics (a what factor) when the who, when, and where factors are similar. The strength of the Is–Is Not technique is that it provides a format to explore problems. Asking who (or what) is affected, who is part of the production process, where the problem arises, and when the problem occurs provides the elemental questions to begin exploring a problem. Kepner and Tregoe note that in many cases it will not be a single factor that explains a deviation but a set of circumstances. For example, the fact that patients are over 65 years of age does not satisfactorily explain a patient falling since many patients over 65 do not fall (and some under 65 fall). Root cause identification requires additional information to define why a particular patient over 65 fell and sustained injury when others did not. Was it due to that patient’s medical situation or other elements (or both)? Additional information—where, when, and who—is needed to explain the adverse result. Does the combination of patient age, other patient-specific factors, time of day, and location provide the critical elements to identify why a particular patient fell and sustained injury? Unlike the Five Whys, which is geared toward identifying a single cause, the Is–Is Not matrix examines a confluence of issues encompassing active and latent causes. Figure 5.6 explores the problem of patient falls with injury.
The Is–Is Not matrix bridges the transition from data collection to data analysis since the Is and Is Not columns record data, and the Difference column provides a head-to-head comparison that challenges analysts to consider what factors could contribute to the problem and what factors could provide appropriate safeguards to prevent the problem from aris- ing. Specifying all the circumstances in which events are not occurring
or being observed could be an extremely time-consuming and unproduc- tive task. If we were examining a single patient fall, we could attempt to compare this fall with all the other patients hospitalized that day who did not fall but obviously some comparisons would yield little information, for example, comparing elderly patients to pediatric or ICU patients. Similarly, it would be possible to identify similar patients (similar age and condition treated by the same personnel in the same location at the same time) who did not fall, thus making it impossible to identify differences. The Is–Is Not matrix requires the analyst’s judgment to direct it toward productive comparisons. Besides bridging the gap between data collection and analysis and cast- ing a wider net encompassing active and latent error, the Is–Is Not matrix is a useful tool for comparing multiple instances of an event and identify- ing commonly appearing elements in the Is and Is Not columns. As stated earlier, because these questioning processes will often identify multiple potential reasons for an event, using the Is-Is Not Matrix to examine mul- tiple events may allow the analyst to focus on substantial differences—those occurring in a majority of cases—and sift out the trivial differences.
Data Analysis and Identifying a Cause After the data has been collected, it must be studied to identify the root cause. Determining the root cause is a matter of judgment. Particularly in medicine a course of treatment can be second-guessed given differences in practice style, but analysts need to inoculate themselves against hindsight bias. Hindsight bias is the tendency, after an adverse outcome has occurred, to attribute predictability to its causes. The question is, without prior knowl- edge of the adverse outcome would one have been able to find fault with the circumstances that preceded it? Expert judgment is a method for identifying root causes and reaching decisions based on the insight of recognized knowledgeable individuals. Expert judgment can be solicited through literature review: who has pub- lished on a topic and what did it say? Expert judgment is often obtained by assembling a panel of leaders in the field; these individuals may have published in the area, performed a high volume of procedures, or may hold office in professional societies. After the group has been assembled, the members are polled to identify root causes.
Consensus is a decision reached by a group as whole. Unlike expert judgment, no external opinions are solicited and the group relies upon itself to reach a decision. Chapter Four covered two techniques, multivoting and Q-sort, to identify solutions, but these two techniques can also be employed to identify a root cause when more than one factor may be the cause of a problem.
Statistical analysis attempts to quantify the relationship between two variables. Typically, statistics establishes statistical significance at 5%: is there less than a 5% chance that we would observe the change in one vari- able given a change in another variable? If the probability of observing the relationship is more than 5%, we say the observed relationship may be due to chance. Chapter Nine examines common statistical tools to determine if variables are related.
Implementing Corrective or Preventive Action The final step is modifying the delivery system to design a new process (with corrective action), document the change, communicate the change to affected parties, and provide the opportunity for feedback. This modifi- cation may include altering clinical practice guidelines, changing policies and procedures, and issuing alerts. The point that Shojania et al. (2001) made is worth repeating: “researchers now believe that most medical errors cannot be prevented by perfecting the technical work of individual doc- tors, nurses, or pharmacists. Improving patient safety often involves the coordinated efforts of multiple members of the health care team, who may adopt strategies from outside health care.” A multilevel approach that rec- ognizes why humans err is needed to develop systems and technology to make medical errors less likely to occur and more likely to be identified when they do arise.