Anthony R. Benedetto, FACHE

ABSTRACT
Six Sigma is a quality improvement methodology that was developed at Motorola to address manufacturing quality issues. General Electric Medical Systems Health Care Solutions (GEMS) adapted Motorola's Six Sigma approach for use in the healthcare environment, which is a mixture of manufacturing-like activities and service activities. GEMS was contracted to provide assistance to the University of Texas M.D. Anderson Cancer Center Diagnostic Imaging Film Library in the use of Six Sigma methodology to improve the film library's performance.

This thesis describes the application of Six Sigma methodology to our film library improvement project. A radiology film library performs service activities almost exclusively; thus, we hypothesized that manufacturing-based Six Sigma methods might need significant modifications for our film library project. Although several Six Sigma tools were found to be of little or no value for our project, those Six Sigma tools that we found applicable did not need any significant modification.
The most important findings were:

• Six Sigma dramatically improves service activities,
• valid quantitative data about services can be very difficult and expensive to gather,
• personnel with limited education and work experience need extensive coaching to encourage creative and big-picture thinking, and
• change management must be an integral component to achieve and sustain dramatic changes.

PREFACE
Overview of Quality Improvement Methodologies
Many quality improvement (QI) methodologies have become popular during the past few decades. Some have been formulated to provide modest, incremental improvements; these could be categorized as "evolutionary." A few examples of evolutionary QI methodologies are continuous quality improvement (CQI), total quality management (TQM), and quality circles.

Another category of QI methodologies is the "revolutionary"; these methodologies support the idea that only drastic change with challenging goals is truly effective. Examples of revolutionary QI methodologies are reengineering and Six Sigma [see Kissler (1996) for a more detailed overview].

Introduction to Six Sigma Methodology
Six Sigma has gained popularity because of major successes achieved by Motorola and General Electric, among other corporations. The Six Sigma methodology will be described in much greater detail later in this thesis. Briefly, Six Sigma involves a stringent examination of quantitative data in conjunction with work simplification strategies that focus on designing exceptionally high quality into the individual steps of a process. By concentrating on high quality throughout the process, the expense and hassle of performing extensive inspection and rework can be avoided.

Six Sigma was developed by Mikel Harry while he was at Motorola. At that time, Motorola was spending almost a billion dollars per year to correct poor manufacturing quality (Harry and Schroeder 2000). Six Sigma methodology breaks down the manufacturing process into its component steps, measures data on quality for each step, and focuses on improving the quality of each step to such a high degree that the overall product quality at the end of the manufacturing process is extraordinarily high-that is, with minimal defects. In the first four years after adoption of the Six Sigma methodology across its product line, Motorola had realized $2.2 billion in savings.

Manufacturing Versus Service Processes
By its nature, the manufacturing process is readily amenable to quantitative analysis. The dimensions of parts can be measured with great accuracy. The number of parts passing or failing inspection at any step can be quantified. The amount of rework and waste material can be measured, and reasons why each occurred can be assigned.

In the service industries, processes dealing with customers are much more difficult to measure, even qualitatively. As examples, the quality of an employee's face-to-face interaction with a customer, an employee's use of good decision-making skills, and customer satisfaction are not easy to evaluate on a subjective high-to-low scale, much less on an objective numerical scale. Additionally, customer interactions are rarely exactly identical from one customer to the next. For example, a high rating for an encounter with a difficult customer may be much more meaningful and valued than a high rating for an encounter with an easy-to-please customer.

The Radiology Film Library
When a patient undergoes a radiology examination, the resulting films are placed into a film jacket and sent to the film library for storage. The film jacket is checked out to authorized borrowers (primarily physicians) when the patient undergoes another examination, is treated as an inpatient in a hospital, or is seen during an outpatient clinic visit. When the borrower is finished with the film jacket, it is returned to the film library and reshelved.

Modern healthcare practice in large medical centers is multidisciplinary, focused on patient satisfaction, and attentive to cost reductions. A cancer center, such as UT M.D. Anderson Cancer Center, exhibits these characteristics, with special emphasis on the multidisciplinary approach to diagnosis and treatment of cancer. The goals of increased patient satisfaction and reduced cost are addressed, among other ways, by drastically reducing the time between the patient's first visit and the eventual development of a definitive plan of care. At M.D. Anderson, the goal for this initial phase is four days, during which time the patient undergoes many tests and has clinic visits with many physicians representing the various oncology specialties (medical, surgical, radiation, etc). The need to have the film jacket in multiple locations each day presents a major challenge to a film library.

Performing a radiology examination is quite similar to a manufacturing process in regard to the ability to break the process into component steps and to measure times and other quality metrics for each step. Moving the film jacket around the campus, on the other hand, is not a rigorously defined process in the sense of rigid standardization, inputs, outputs, and so forth. The film jacket could be anywhere at any given moment, and it might need to go almost anywhere on the campus. The first user may need the film jacket long enough to delay or prevent the use of the film jacket by the next user. A physician may take the film jacket to a consultant without informing the film library, thereby making it extremely difficult to track down the film jacket for a subsequent requestor.

The UT M.D. Anderson Cancer Center contracted with General Electric Medical Systems Health Care Solutions (GEMS) to assist us in using Six Sigma methodology to dramatically improve the performance of our film library. We had previously experienced great success with GEMS in improving throughput in our computed tomography (CT) section using Six Sigma. We were concerned that we might not be able to duplicate the success seen with the CT section (a manufacturing-like, linear process) because most film library processes are nonlinear, unpredictable, and difficult to quantify. Furthermore, the CT section employs personnel with associate degree or higher education who consider themselves healthcare professionals, whereas most of our film library personnel have high school or General Educational Development (GED)-level education and see their duties as a job rather than as a calling.

Purpose of this Thesis
The purpose of this thesis is to describe the problems we encountered and the adaptations we made in applying Six Sigma methodology to our film library services. Overall, the problems were related primarily to the inherent difficulties in acquiring quantitative data about our processes, rather than shortcomings in Six Sigma theory or practice. We ended up choosing those elements of Six Sigma methodology that worked for our environment and simply disregarding the remainder. Consequently, the adaptations necessary to use Six Sigma techniques were minor, but the challenges of developing metrics and acquiring meaningful, reliable quantitative data were significant.

EVOLUTIONARY VERSUS REVOLUTIONARY CHANGE

Evolutionary Change
In the beginning, there was the individual worker who was skilled in producing baked goods, swords, baskets, and other goods. He individually procured his raw materials, made everything with his own hands, and sold his wares from a stall in the local town market. He gradually learned the best way to make his products and made them the same way every day.

As our sole proprietor became more successful, he found the need to hire helpers to satisfy the increased demand. He realized that all of the information he carried in his head and scribbled on scraps of parchment needed to be conveyed to his new employees. Otherwise, his successful products would become inconsistent in quality and he would lose customers. The first standard operating procedure (SOP) was born.

Over time, he and his employees made changes that improved the products. These changes were incorporated in the SOPs on a casual basis as they arose. Because everything seemed to be going smoothly and business was good, our budding entrepreneur saw no need to systematically study his processes to see if improvements could be made that had not yet been stumbled on. Unfortunately for him, one of his competitors had discovered that occasionally stepping back and taking a more holistic look at her processes yielded insights that contributed to higher quality and improved margins. Being clever with words, she decided to give this new approach a catchy name. However, being relatively indecisive, she could never decide between "total quality management" or "continuous quality improvement." This legacy remains today.

Total quality management (TQM) and continuous quality improvement (CQI) live on today as the most highly developed examples of evolutionary process improvement. At their heart, TQM and CQI hold that every process contains elements that can be improved, no matter how long the process has been in use or how successful the end product. Rather than being content with "if it's not broken, don't fix it," TQM and CQI proponents are more likely to say "although it's not broken, we can make it better if we look closely enough." Typical goals for a TQM project might be a 10 percent reduction in accounts receivable or a 15 percent increase in outpatient surgery cases with no staffing increase.

Evolutionary techniques have their place in any business environment. There is no such thing as the perfectly effective and perfectly efficient process-some element of improvement can always be realized by a clever person or group of people. However, highly successful processes may have been fine-tuned so well over time that major changes simply are not needed. In these situations, the techniques of evolutionary change work well-for example, workers' quality circles and team meetings to discuss incremental change.

Revolutionary Change
Healthcare processes are far from perfect. The time from initial presentation to cure may be long and expensive because of poorly designed diagnostic work-up algorithms and outdated knowledge of modern drugs. Readmission rates after surgery, or emergency admission rates for acute illnesses, may be too high. Medical administration errors and adverse drug reactions during inpatient stays may increase the length of stay and, consequently, the cost of the episode. In the era of cost-based reimbursement, these inefficiencies lingered far longer than necessary because little financial pressure or incentive existed to improve performance. Today, in our cost-conscious environment, we have recognized that high-quality healthcare also tends to be low-cost healthcare, and we have become much more aggressive about addressing quality and cost issues.

TQM and CQI techniques are effective in producing needed change, but inherently they are focused on incremental, piecemeal change. A team will work on improving one segment of a process, then next year another team will work on another segment, and so forth, until eventually the entire process has been addressed. However, when pressure to implement broad-based and/or major changes rapidly is placed on a process, TQM and CQI work less well because they do not stress the importance of looking at the entire process holistically.

When a process is identified as being significantly flawed, tweaking and fine-tuning bits and pieces of the process over an extended time may not be acceptable approaches to bring the process up to the desired level of performance. The two most popular tools for addressing revolutionary change are reengineering and Six Sigma. They have many similarities, but they also have significant differences.

Reengineering is defined (Hammer and Champy 1993) as "the fundamental rethinking and radical redesign of business processes to achieve dramatic improvements in critical, contemporary measures of performance" [emphasis added to reflect the four key elements of reengineering]. The personnel at Motorola who crafted the Six Sigma philosophy "focused on making improvements in all operations within a process, producing results far more rapidly and effectively" (Harry and Schroeder 2000). Thus, reengineering and Six Sigma recognize that:

• processes are key to quality,
• most processes are poorly designed and implemented,
• overall success rates are very sensitive to individual subprocess success rates, and
• rapid, dramatic change requires looking at the entire process, not just a few individual subprocesses.

Six Sigma Versus Reengineering
The most important difference between reengineering and Six Sigma is the highly structured, quantitative basis on which Six Sigma is built. Reengineering recognizes the role of statistics and quantitative analysis, but Six Sigma tends to place extremely heavy reliance on data to drive decisions. "Business sense" always comes into play, but only after the data have been used to narrow the range

Quoting from a General Electric Six Sigma training manual (General Electric 2000), data are important because:

• "We don't know what we don't know." [Note: our knowledge is always incomplete.]
• "If we can't express what we know in the form of numbers, we really don't know much about it."
• "If we don't know much about it, we can't control it."
• "If we can't control it, we are at the mercy of chance."

In healthcare, data may be especially important. Physicians are such powerful figures within the healthcare industry that they are sometimes elevated to positions of unchallenged authority in their specialty area. They may come to believe that their own experience, or at least their remembrance of their experience, is representative of the experience of all physicians in the same specialty. They fail to realize that their memory may be less than perfect, that some of their outcomes may have been less than ideal, and that their patient population may be different in important ways from the patient populations of their peers. Thus, the way they practice their specialty is based on personal, anecdotal experience and is rarely enriched by retrospective, quantitative study of their career-long medical practice.

If we allow the anecdotal experience of one or a few influential physicians to drive our clinical pathways and other approaches to providing optimized healthcare to all patients, we are likely to find that our real-world outcomes fall far short of our targets. If, on the other hand, we quantitatively analyze the results of many physicians in treating a wide variety of patients for a carefully defined clinical condition, we are much more likely to derive a treatment plan that will provide the best possible outcome at the lowest cost and with the least likelihood of failing. Data are good, but good data are better.

THE SIX SIGMA METHODOLOGY
The Six Sigma Breakthrough Strategy
Six Sigma was developed at Motorola to reduce high costs associated with poor final product quality at the end of an assembly line (Harry and Schroeder 2000). Motorola calculated that it was spending up to 20 percent of its total revenues (almost $1 billion per year) on correcting poor quality discovered at the time of final inspection. The engineers charged with reducing these costs came to the conclusion that the only sensible way was to eliminate as many steps in each process as possible, then to design near-perfect quality into each step. Only then would they have a chance of low-defect final product that required essentially zero rework to correct defects.
The eight-step Six Sigma Breakthrough Process is described by Harry and Schroeder (2000) as:

1. Recognize-realize that a business problem exists for which Six Sigma tools may be appropriate.
2. Define-clearly express the problem and the desired final state in written form.
3. Measure-use historical or prospectively obtained data to adequately characterize current quality, costs, and other factors.
4. Analyze-apply statistical tools and common business tools to the data to characterize the key drivers of the problem areas identified in the Measure phase.
5. Improve-develop alternative solutions, evaluate them, and test them on a pilot basis to determine if the desired outcome is being achieved.
6. Control-determine the best methods for ensuring that any new solution is actually working.
7. Standardize-write a new standard operating procedure and provide training to all persons involved in the process to ensure that the process is carried out according to the SOP.
8. Integrate-implement the new SOP and persistently use the appropriate control measures to ensure that the process is performed in compliance with the SOP over the long term.
   

Harry and Schroeder further categorize these eight steps into four stages presented in Table 1.

Table 1. Six Sigma, as Described by Harry and Schroeder (2000)

The GE Twist on Six Sigma
The General Electric Company (GE) has been one of the highly visible success stories for Six Sigma (Harry and Schroeder 2000). GE recognizes the same elements as given in Table 1, but they label the stages somewhat differently (General Electric 2000). My understanding of the GE philosophy is presented in Table 2.

GEMS uses the acronym DMAIC (de-may'-ik) to encapsulate the philosophy for its own employees and for clients.

Six Sigma as a Solution to Manufacturing Problems
Six Sigma was developed at Motorola to address a manufacturing problem. For the purpose of this thesis, a manufacturing process need not be the assembly of a widget on a traditional factory assembly line. Rather, a manufacturing process can be thought of as any process whereby resources are applied in a sequential manner to produce a single product or service, with relatively high predictability in any individual step of the process. Our CT throughput project is described later in this article to further demonstrate how healthcare processes fit this manufacturing process model.

1. The weakest link in the chain: In a manufacturing process, when the first step is completed, the second step is always the same; when the second step is completed, the third step is always the same; and so on. At the end of the day, all full cycles of the process have gone through all of the steps in the same order. If the process is well designed, each step will have been performed with predictable expenditure of resources and with predictable quality.
   In a manufacturing process, the quality of the final service or product can be calculated from the quality at the end of each step. For example, if the "success rate" of Step 1 is 90 percent and the success rates of Steps 2 and 3 are 95 percent and 90 percent, respectively, the overall success rate of the process will be 0.90 x 0.95 x 0.90 = 0.77, or 77 percent (see Table 3). In Six Sigma terminology, this is called "rolled throughput yield."
   
   Table 3. Rolled Throughput Yield
   

   Example	Step 1	Step 2	Step 3	Overall Success Rate
   1	0.900	0.950	0.900	0.77 (77%)
   2	0.983	0.983	0.983	0.95 (95%)
   3	0.950	1.000	1.000	0.95 (95%)

   
   If the goal is to have an overall success rate of 95 percent (Example 2 in Table 3), the success rates of each individual step will need to be higher than 95 percent, because the overall success rate is the product of the individual rates. Note also that the success rate of each individual step must be no lower than the overall desired success rate; that is, the overall success rate cannot be higher than the success rate of the least successful intermediate step. If each of the individual success rates were equal, each would need to be 98.3 percent to achieve an overall success rate of 95 percent. In the Example 3 in Table 3, if we want the overall success rate to be 95 percent and if Step 1 has a success rate of 0.95, Steps 2 and 3 must be perfect (success rates of 1.0). As will be mentioned later, rolled throughput yield does not fit the film library because almost every interaction is different. However, the concept that the worst-performing step limits performance is still valid and important.
2. Introduction to Six Sigma statistical foundation: The "six sigma" in Six Sigma is derived from the statistics of a Gaussian distribution. The Gaussian (also called "normal") distribution is the familiar bell-shaped curve that describes many biological and nonbiological processes.
   
   The Gaussian distribution is characterized by the mean and the standard deviation. The mean, •, is the average value of all of the individual values making up the data set. The standard deviation, • (Greek symbol for sigma), is a measure of how widely spread the data are around the mean; for example, a large value of • represents a large spread of the data. The data are divided equally on both sides of the mean, so that half of the data are above the mean and half are below the mean. Mathematically, the area under the Gaussian curve between any two points on the x-axis represents the total number of data points whose x-values lie in that range (see Figure 1). (Call (312) 424-9473 to obtain a copy of Figure 1.).
   
   For example, suppose we wish to analyze the radiology report turnaround time (TAT), which can be defined as the time from the end of the patient's examination to the time the radiologist signs the final edited report. We would measure the TAT for each report and arrange the data in a table that lists each TAT range of interest (say, 30-minute increments) and gives the number of TATs that fall in that time range. The resulting distribution could be plotted on rectangular graph paper, with the y-axis being the number of reports in each TAT range and the x-axis being the report turnaround times. The Gaussian distribution would predict that 68.3 percent of all of the reports would have TATs that fall in the range µ ±1•, 95.4 percent would fall in the range µ ± 2•, 99.7 percent in the range µ± + 3•, and so on.
   
   Another way of using the ranges described in the previous paragraph is to predict how many data points would not fall within any given range. In the case of µ ± 2• for example, 4.6 percent of the reports would have turnaround times that were either shorter than the bottom of the range or longer than the top of the range. Because we are only interested in the long TATs, and because the Gaussian distribution predicts an equal number above the range and below the range, we would expect 2.3 percent to be longer than µ± 2•. If we now set µ± 2• as our target, that is, we do not want TATs to be longer than this value, we see that 2.3 percent of all reports would have TATs longer than our target. In Six Sigma terminology, these reports would be called "defects."
   
   Six Sigma proposes that each step in a multistep process should have fewer than µ± 6• defects, which corresponds to 0.0000002 percent defects, or two defects in every one billion opportunities. When reading the Six Sigma literature, however, the figure cited most frequently is 3.4 defects per one million opportunities. This less stringent figure is based on the observation that processes inherently vary slightly from day to day and that this variation is unavoidable and acceptable statistically. [Note for the statistically sophisticated reader: Long-term drift in processes is accounted for by introducing a ±1.5• shift. Thus, when looking up 3.4 defects per million in a Z table, you will find that this value falls at ±4.5•, not ± 6•. See Chapter 8 of Harry and Schroeder (2000) for a more complete discussion. Also, sample sizes are always chosen to be large enough to minimize the differences between population and sample descriptors.]
   
   For a concrete example, suppose the average number of patient bills processed per day by a single clerk is ten, with a standard deviation of three. If we look at an individual clerk's performance over the next 100 days, we would expect the daily output during 95 of those 100 days (95% of the data) to be 10± 2s•˜± (2)(3) = 10 ± 6, or as few as 4 bills and as many as 16 bills. (Statistically, the µ± 2• range is called the 95% confidence interval.) During the other five days, the clerk is likely to process fewer than 4 bills or more than 16 bills.
   
   A major element of Six Sigma philosophy is to first reduce variability in a process before trying to shift the average performance. (Statistically, the standard deviation is reduced first, then the mean is shifted.) The rationale is that a process showing a great deal of variability is likely to have multiple major and minor factors driving the variability, making it difficult to focus on the "vital few" truly important factors. For example, failure to have an SOP and to have properly trained the personnel on the SOP are likely to be the primary causes of high variability in the beginning stages of a quality improvement project. Once these remedies have been put into place, any residual variability and any variation from average targeted performance is likely to be due to important factors in the process itself.
   
   
   Table 4. Reduce Variability First (example of patient bill generation)
   

   Mean	S.D.	Mean ± 2 SD	95% Confidence Interval
   10	7	10 ± 14	0-24
   10	4	10 ± 8	2-18
   10	1	10 ± 2	8-12

   
   In the billing clerk example, we might want the clerk to eventually process a mean of 15 bills per day. Suppose that the clerk's standard deviation is 7 bills per day, yielding a 95 percent confidence interval of 0 to 24 bills per day (negative output is not possible, so the lower end of the interval is 0). Before we try to edge the mean output up to 15, we would want to get the clerk working predictably with very few days at the lower daily output levels, which translates into a need to reduce the standard deviation to a much smaller number. Note in Table 4 that increasingly smaller standard deviations result in fewer low-output days.
3. CT throughput example: Due to growth in our CT workload, our institution found that we were not able to meet the demands of our referring physicians. The usual reaction to such a dilemma is to consider adding shifts or purchasing additional CT scanners. Although initially plausible, such solutions can be quite expensive, with no guarantee of success. For example, running the scanners 24 hours per day will not increase throughput if patients are not willing to come to the scanner at 4:00 in the morning! We decided that an opportunity to increase throughput might be available without such drastic measures.
   
   The multistep process of performing an outpatient CT exam is an example of a healthcare process that fits the manufacturing model for which Six Sigma was originally developed. The following steps are performed exactly in this sequence for every patient:
   1. The patient arrives.
   2. The patient receives any necessary preparation.
   3. The examination is performed.
   4. The films are printed.
   5. The radiologist interprets the images and dictates a report.
   6. The transcriptionist prepares the preliminary report.
   7. The radiologist edits and signs the report.
   8. The report is sent to the requesting physician and to the patient's chart.
      

   
   If operations are standardized, the length of time for each step can be calculated from historical data, including an estimate of each step's variability (its standard deviation). Based on these calculated times, the number of patients per eight-hour shift can readily be calculated.
   
   When we employed Six Sigma techniques to increase our CT throughput, we quickly found that each step had not been optimized. For example, patient preparation times that initially were 45 minutes have been reduced to less than 5 minutes in many cases. We also found that we needed to design our CT processes to take advantage of the fact that we had nine essentially identical CT scanners, which cushions us against cascading delays when one patient experiences a delay in his or her examination process. For example, if a patient develops a contrast reaction during an examination, that patient's examination would be protracted, and the next patient scheduled for that scanner would be delayed. With multiple identical scanners, we were able to redesign the process to make sure that patients were reassigned to the next available scanner. Although this seems obvious when seen on this page, it is not straightforward or easy to accomplish when the scanners are spread out geographically across a large institution.
   
   Our Six Sigma CT throughput project ran formally for about two years. The initial improvement in throughput, with no additional machines or shifts, was a 45 percent increase in examinations (Elsberry 2000). The project is still ongoing, although less formally, since that time, because CT scanner technology continually changes and because the geographical distribution of our scanners frequently changes. Every change presents an opportunity for conditions to worsen; fortunately, every change also represents a potential opportunity for conditions to improve nonlinearly.

Radiology Film Library Does Not Fit the Manufacturing Model

1. Film library overview: The UT M.D. Anderson Cancer Center Film Library performs four important major functions: storing films, receiving and in processing films from other institutions, lending films to on-campus users, and lending films to off-campus users. Some film libraries have responsibility for report distribution and other clerical functions. However, all film libraries perform the four core functions mentioned above. Figure 2 is a schematic representation of these functions at M.D. Anderson. (Call (312) 424-9473 to obtain a copy of Figure 2.)
2. Film library compared to CT: Key criteria mentioned earlier for the manufacturing model were that the individual steps are predictable and that they are rigidly sequential. The handling of film jackets is the antithesis of these criteria. A film jacket already could be checked out to any one of 1,000 or more borrowers (physicians, clinics, radiology sections, etc) when a requestor asks for it. Patients go to clinic visits and radiology exams in patterns that vary from clinic to clinic, and sometimes even among physicians within a given clinic. Something unexpected that shows up on a CT scan may trigger an immediate add-on magnetic resonance imaging (MRI) scan, with the film jacket needed in both locations when the radiologists interpret the two exams.
   
   When a film library comes under criticism, the criticism tends to be global: "fix the film library!" After only superficial scrutiny, it becomes quite clear that "the film library" is not a clear-cut single process that follows the manufacturing model, such as a CT scan does. Rather, the film library is actually a large suite of processes that are related and interdependent, but not describable as a collection of steps that are predictable and easy to quantify. Clearly, the concept of rolled throughput yield makes no sense for "the film library." Does this make the concept wrong? No. Rolled throughput yield simply is a tool that is inapplicable and remains in the toolbox unused. This was one of the important lessons learned in our film library project.

DIFFICULTIES ENCOUNTERED USING SIX SIGMA IN FILM LIBRARY PROCESSES
Data: Generation, Validity, and Feedback

1. Information systems: As was shown in Figure 2, the radiology information system (RIS) is at the center of a film library's operations. The RIS is a computer system that tracks patients and films, generates bills, captures reports, and so forth. Each patient's film jacket is bar-coded for easy, reliable check-in and check-out. Every new exam is entered automatically into the RIS. At our institution, movement of the patient through the radiology department is also tracked in the RIS, which is of value to the film library when a requestor asks to see films before the radiologist has rendered the interpretation of an exam.
   
   The selection of a radiology information system is always a compromise among less-than-optimal options. No vendor provides a product that exactly fits an institution. In some cases, the fit is not very good at all, but an RIS must be purchased anyway, because a large radiology department cannot be run manually. If the persons evaluating the fit are biased toward patient tracking and billing, it is possible for an RIS to have good performance for these functions but less than adequate performance for film library functions. Rarely can an RIS be customized by the purchaser to optimally handle local peculiarities.
   
   Even with the best RIS, problems arise because of the usual lack of full integration of the RIS with other institutional systems. For example, we retrieve roughly 1,500 to 1,800 film jackets every weekday night and deliver them to a clinic or radiology section for the next day's appointments. Our RIS does not provide for importing of appointment information from the scheduling module of the institution's healthcare information system (HIS) to the RIS. Although the HIS can produce a listing of all appointments, it does not know which patients have film jackets. Thus, in the past the film librarians spent many hours looking for film jackets that did not exist. Fortunately, our radiology informatics group wrote an interface program that takes a nightly dump of information from the HIS scheduling module and compares it against the film jacket listings in the RIS, then prints a pull list for our film librarians. Unfortunately, this program has been less reliable than we would prefer, so the film librarians too frequently end up devising impromptu workarounds when the program misbehaves. Also, because of limited availability of programming support, changes to the program are practically impossible to obtain when clinical operations change.
   
   An example of a film library metric whose data can be readily gathered from an information system with minimal manual data collection is the percentage of film jackets requested by potential borrowers that are delivered each morning. The program mentioned above provides a count of all of the appointment labels printed (denominator). The number of appointment labels remaining after all of the jackets have been placed on the delivery carts is hand counted (numerator). This percentage can be calculated daily with little effort.
   
   Our RIS focuses on the end points of processes and provides little information about intermediate steps. For example, suppose we wish to assess how promptly we retrieve film jackets when a requestor is at our loan window or places a request telephonically. When a requestor asks about the availability of a film jacket, it would be helpful if the time of the initial RIS inquiry were logged, because this is usually within a few seconds after the arrival of the requestor at the loan window or after making the request telephonically. Our RIS captures the final checkout time, but not the time of the initial inquiry. To assess our performance in delivering film jackets rapidly to requestors, we must capture the time of the initial inquiry manually.
2. The challenges of manual data collection: Because of the paucity of managerially useful data available from most information systems, much of the data needed for a Six Sigma analysis must be collected manually. Manual data collection is time consuming and is a hassle for the person(s) who collect the data. When the frontline employee is the one who must collect the data, the employee must be educated about the importance of the data and about the exact data that are needed. Then, the employee must actually collect the data as prescribed. For some employees, a supervisor must closely monitor the data collection to ensure that the data are collected accurately and completely.
   
   A film library example is the metric for the timeliness of fulfilling a customer's request. Our metric specifies target fulfillment times for film jackets requested in person at the loan window and for telephone/fax/e-mail requests. To calculate these times, the request submittal time and the check-out time must be captured. Determining one or both of these times may also require the patient's medical record number so that the time can be retrieved from the radiology information system. Each e-mail and fax request automatically receives a time stamp, provided that the internal clocks are set accurately. However, the request submitted in person at the loan window and the telephone request require the film librarian to note the submittal time on a log sheet. The first step that the film librarian takes upon receiving an in-person or telephone request is to look up the films in the RIS; then he or she pulls the jacket from the shelves if it is available in the film library. For an in-person request, the jacket is quickly checked out to the requestor, and the film librarian prepares to work with the next customer. Thus, for both the in-person request and the telephone request, recording the original submittal time is not part of the normal workflow. The film librarian is likely to guess at the submittal time in retrospect. The longer the lapse in time between the transaction and the film librarian remembering that they need to enter the submittal time, the less accurate the guess. As the target fulfillment time for a loan window transaction is about five minutes, even small inaccuracies in the submittal time can seriously distort the data.
3. Data that are impossible to obtain: The fundamental metric of Six Sigma is defects per million opportunities, which translates in common language as the percentage error rate. Error rates require knowledge of the number of errors associated with a population of transactions (the numerator) and the total number of transactions (the denominator). Techniques can be devised to measure the number of errors, although some errors can be very difficult to identify. The bigger challenge can be the denominator, that is, the total number of transactions. Some important performance measures are impossible to obtain, so substitute (usually inferior) measures must be used.
   
   A film library example of an impossible-to-obtain metric is the number of outside films that arrive in the institution that are not reported to the film library for entry into the RIS. Outside films are films that were generated at other institutions and that are brought or sent to our institution for review by the patient's physicians and for comparison to radiology exams performed here. A high reporting rate helps when other potential borrowers request these outside films. In this example, both the numerator and the denominator are essentially impossible to obtain. Some clinics may fail to notify the film library that outside films have arrived. If clinic personnel do not send the films to the film library for filing in the patient's M.D. Anderson film jacket, but instead hold the films in the clinic or send the films back themselves, the film library never knows that the films are on the campus. Thus, we cannot know about their failure to report the films (numerator), and we cannot know that the films were on the campus (denominator). With 20,000+ new patients per year, each of whom is likely to bring or send outside films, it is logistically impossible to design a system for ensuring that the arrival of every outside film is captured.
4. Tracking film librarian errors: Film librarians undergo training in basic customer service skills, such as telephone etiquette, and in specialized film library procedures. As they go about their duties, they interact with colleagues and customers face to face. They interact with the RIS, the computer for the off-site warehouse, and other information systems. Finally, they perform repetitive manual tasks, such as filing film jackets in the shelves, which are not observable by the supervisor on a continual basis and are not captured in an information system.
   
   No enduring record exists of face-to-face interactions. Unless a supervisor happened to personally witness the interaction, no independent evaluation is made of the quality of the interaction. Even when the supervisor witnesses an interaction, the evaluation is inherently subjective and difficult to quantify. Finally, during a face-to-face interaction, there are many factors to evaluate, so making a judgment about the quality of the interaction requires assessment of multiple factors in a very short time, then making a record of that assessment in a quick, reliable, and retrievable manner.
   
   For a machine interaction, there are two types of error. The first type is an error of commission; that is, the film librarian performs an improper action. In many cases, these errors will be noticed and can be tracked back to the person who committed them. However, this tracking ability relies on the information system's built-in error capture mechanisms. If the RIS does not capture errors, does not capture the identity of an operator, or does not readily allow reports tailored to the investigation of errors, the RIS is of no help in identifying errors of commission. RIS vendors seem to have the misperception that all users will always comply with all published policies and procedures.
   
   The second type of machine interaction error is an error of omission; that is, the film librarian fails to perform a required action. When and if these errors are detected, they can be extremely difficult to track, because there is no machine record of the action. Sometimes an action could have been performed only by a particular individual, but such occurrences are uncommon in a large, busy film library where similar duties are performed by multiple people on all three shifts and during the weekend.
   
   Irrespective of whether interactions are face to face or with the RIS, quantifying these interactions is surprisingly difficult. A fundamental metric of Six Sigma is the ratio of defects to opportunities, that is, the number of defects observed divided by the total number of opportunities for those defects to occur. Even though every check-out and check-in of a jacket may be tracked using the RIS, the total number of opportunities, that is, the total number of handlings, is rarely available.
   
   As a more concrete example, suppose that 300 film jackets were sent to the CT section today. When we check the RIS, however, we find that ten film jackets still show that they are in the file room, meaning they are physically in the CT section but had not been properly checked out in the RIS. This is a defect rate of 3 percent (10 defects out of 300 opportunities). While each of the ten improper check-outs could be detected by querying the RIS, all of the data necessary to calculate this error rate cannot be extracted from the RIS. For example, can we ask the RIS to give us a report of all jackets checked out to the CT section on this date? Can we ask the RIS to compare a list of the patients whose film jackets were requested against a list of the jackets checked out to the CT section? Very few commercial RIS products offer this level of functionality.
   
   In our case, we must walk to the CT section, write down the medical record numbers on all of the film jackets that we find there, then promptly look up the film jackets manually in the RIS to determine if they were properly checked out. Although doable, this certainly makes the acquisition and analysis of data unattractive and expensive. If we let the obstacles discourage us from collecting the data, the resulting lack of data means that the process can be out of control without our knowledge.
   
   As difficult as it can be to dig data out of the bowels of the RIS, it is even more difficult to obtain quantitative data on face-to-face interactions that do not involve an entry in the RIS. For example, a physician will come to the loan window and ask to see a patient's films. The film librarian retrieves the film jacket and hands it to the physician, who reviews the films briefly in a nearby viewing area and promptly returns them. In this transaction, the film librarian should come to the loan window promptly and should greet the physician courteously. The film librarian must obtain the necessary information, search the RIS, and retrieve the film jacket. Upon the return of the films, the film librarian must reshelve the films in their original location. If the physician walks off with the films without properly signing them out, the film librarian must try to remember the physician's name and check out the films to the physician in the RIS. In a busy environment, the physician's name may be lost forever.
   
   All of these important pieces of the transaction occur with no RIS record of the transaction. When attempting to assess the quality of this service, essentially no historical data are available. Is the physician having to wait too long for an initial greeting? Are the films available a high enough percentage of the time? Is the physician having to wait too long for the films to be retrieved? Are "borrowers" walking off with films too frequently? All of these important operational questions, the answers to which have an impact on many other film library operations, are impossible to retrieve from the RIS or from any other information system. Obtaining them requires manual observation of multiple transactions by a trained observer over an extended period of time. Once again, this can be done, but it is so time consuming and tedious that it is likely to be sampled infrequently, perhaps too infrequently to rapidly alert us to unsatisfactory levels of performance.
   
   Another example of a manual task that is neither observable on a continual basis nor tracked in the RIS is the accuracy with which film jackets are filed in the shelves. Inaccurately filed film jackets lead to extended search times, overlooked film jackets, and frustrated film librarians and requestors. Figure 3 is a data collection sheet used by a supervisor or other designated auditor. (Call (312) 424-9473 to obtain a copy of Figure 3.) To collect these data, the auditor must manually inspect 100 percent of the film jackets in a shelving section, checking for each of the listed errors. In a large film library, this is a tedious, mind-numbing, and time-consuming task. Fortunately, the data are powerful. Figure 4 is a summary of three data collection sessions, showing highly variable performance between different film librarians. (Call (312) 424-9473 to obtain a copy of Figure 4.) We were able to use these data to focus on those film librarians whose performance was substandard by providing remedial education or closer supervision, as appropriate.
5. Six Sigma reporting tools are difficult for frontline employees to grasp: Film library employees tend to have little education beyond high school. Even supervisory personnel may have joined the film library soon after high school and may not have pursued additional college-level education. Their foundation for understanding percentages, statistics, and important Six Sigma concepts is weak. Data that Six Sigma leaders find compelling may be lost on the frontline employee, making it difficult to rouse the employee to corrective action.
   
   Percentage error rates are fairly well understood by most frontline employees, because they have been exposed to such calculations in high school grade calculations, as what constitutes an A, a B, or an F. To most of us, a grade of A represents superior performance. A grade of A might be awarded for a 90 percent success rate. When we measure film library performance and find that a particular metric is 94 percent, the reaction of most frontline employees is that they have done well. However, in Six Sigma, we are aiming for a 99.9999 percent success rate. Explaining why 94 percent is good enough for an A in school, but not nearly good enough at work, is a leadership challenge.

Developing Solutions

1. Baseline data require that a decent process already exist: During the Measure and Analyze phases, baseline data are gathered to help pinpoint factors that are driving substandard performance. A performance standard may need to be established de novo during the Define phase, because of the discovery of an important area of substandard performance or due to lack of an existing performance standard. During the Measure phase, baseline data would be collected to define the degree of substandard performance. The most likely explanation of substandard performance in some cases may be that a formal process addressing this area has never been in place. The Analyze phase is predictably going to demonstrate very poor performance.
   
   A film library example is the timeliness of return of films that we lend to other institutions. Although we had always specified a loan period when we lend the films, we rarely monitored the return of the films on a systematic basis. Rather, we chased down the films when a new requestor asked for them. As we cannot promptly fulfill the new request if the films are not under our control, it became clear that we needed to implement a formal method of systematically following up on delinquent films. Our RIS has not been helpful to date, because its dunning functionality cannot distinguish between on-campus and off-campus delinquencies. When we measured the return rate, we found it (unsurprisingly) to be dismal. The major factor driving the poor return rate was determined to be failure to have a process for following up with borrowers to get the films returned promptly. We did not need data to tell us this, but we had to establish a baseline so that we could observe improvements as we implement the new process. The baseline also helped us narrow the list of factors likely to cause problems so that we could address them in the new SOP.
2. Frontline employees have difficulty dealing with multistep processes:
   Most high school graduates have not been exposed to problems that have dozens of steps and multiple conditional branches. Film library problems are of this type, frequently generating process maps several pages long. Following a complicated process from start to finish, especially when one conditional branch leads immediately to another (or two, three, or more), is a skill that is difficult to acquire in the short time span of a Six Sigma project team. The Six Sigma leader is faced with the challenge of getting the team members to understand the existing complicated multistep process. Then the team must cope with the even more difficult task of considering alternative subprocesses and their effects on the process. These challenges are surmountable, but upper management must understand that project times will be lengthened when working with employees whose educational and work experience backgrounds are limited.
3. Involving the evening, night, and weekend shifts: A fundamental tenet of Six Sigma, indeed any process improvement effort, is that all stakeholders must be included, especially those at the front line who deal with the customers every day and who know the problems most intimately. A film library in a large institution operates 365 days per year, 24 hours per day. The weekend shifts may be staffed with personnel who work only during the weekend. In contrast, weekday day shift is the usual schedule for film library leaders, information systems representatives, clinic personnel, and other persons who are included as project team members. Scheduling meetings to accommodate day-shift personnel is already a challenge; scheduling meetings that require the attendance of people from all three weekday shifts, plus the weekend shift, is extremely challenging. The usual compromise is to have a series of meetings: Monday morning meetings (for weekend night-shift employees and weekday day-shift employees), Friday afternoon meetings (for weekend evening-shift employees and weekday day-shift employees), midweek morning meetings (for weekday night shift and day-shift employees), and midweek afternoon meetings (for weekday evening-shift and day-shift employees). While this gets everybody into a meeting (except weekend day-shift employees), it leads to an inefficient extended series of meetings because each team must spend time at each meeting catching up with what was done at the other teams' meetings. The meetings must reiterate problems, rather than homing in quickly on a potential one-meeting solution if everyone were in the same room at the same time. These challenges are surmountable, but upper management must understand that project times will be lengthened because of the need for multiple iterative meetings.

Change Management Must Be Integral and Intensive
An evolutionary process improvement project requires attention to change management, as very few people welcome change. A Six Sigma project requires exceptional attention to change management, because the desired change is revolutionary, not just evolutionary. The way a person performs his or her job is likely to change in many respects, which is uncomfortable and, sometimes, threatening. Involving all stakeholders from the very beginning of the Six Sigma project is necessary, because they repeatedly see and hear the reasons why change is needed, but stakeholder involvement alone is not sufficient.

Implementation of major changes can be very threatening, especially for employees in the lower pay grades typical of film library positions. Additionally, a dramatic change in a process may be overwhelming to those who have had little experience with making sudden changes, either at work or in their private life. Even if they are motivated to change, the extent and rapidity of change may be more than they can handle. Thus, not only must employees be educated about the need for change, but their ability to cope with change must be assessed and implementation plans must be tailored to introduce change at an acceptable rate.

For any one project, this tailoring is fairly straightforward. However, when multiple Six Sigma projects make recommendations for process changes (we had nine mini-project teams within our overall Six Sigma project), the film library leadership team must consider the entire film library's ability to absorb change. The film library leadership must also balance upper management's desire for rapid change against the ability of the film library staff to cope with it. An individual project team's recommendations may be easy to implement quickly, but implementing multiple teams' recommendations may require an extended implementation period to ensure success, especially if any other recommendations require coordination or are in conflict with each other.

We commissioned a Six Sigma project team to improve the cycle time for mailing out M.D. Anderson films on loan. When we began, two mail-out film librarians were functioning virtually unsupervised and with no written SOP. Requests were not logged in to ensure that all requests were processed, so some requests had cycle times of weeks, because the paperwork was misplaced and not discovered until someone called to complain. Additionally, there was no in-process documentation of what work had been accomplished, so that an unexpected absence of one film librarian meant that the other film librarian could not easily tell what requests were extant and what remained to be done on each request. The top graph in Figure 5 shows the baseline data. (Call (312) 424-9473 to obtain a copy of Figure 5.)

The project team recommended a new SOP, the most important feature of which was the creation of a tracking form so that each step of the process was documented. The second graph in Figure 5 shows that the new SOP was quite effective in eliminating outliers and in reducing the average cycle time. During this period of time, our GEMS consultant was working very closely with the mail-out film librarians to answer questions and guide them through use of the new tools.

The third graph in Figure 5 demonstrates a return to the baseline state after the consultant moved on to other mini-project teams and the mail-out film librarians failed to comply with the new SOP. In response to this regression to the old performance, a log-in sheet was created, a change to strict first in, first out (FIFO) processing was implemented, a new mail-out film librarian has joined the team, and a new supervisor has begun monitoring the mail-out team's performance more closely. The fourth graph in Figure 5 shows that this refocused attention to the SOP has reduced the mean and median cycle times, but unfortunately has not yet reduced the long-cycle-time outliers to an acceptable level. We are gathering detailed data about the outliers to better understand the reason for their existence and to develop methods to minimize their occurrence. (See Kotter [1996] for an excellent discussion of change management.)

CULTURE, POWER DIFFERENTIALS, AND OTHER FACTORS THAT CAN COMPROMISE THE SUCCESS OF A SIX SIGMA PROJECT IN A MEDICAL ENVIRONMENT
The factors discussed in this section can make or break a Six Sigma performance improvement effort. None of them is specifically a Six Sigma issue, but, rather, each is an issue that any performance improvement project may face. They rise to a higher level of importance in a Six Sigma project, because Six Sigma projects have high expectations and require heavy investment of institutional resources. Failure of a Six Sigma project can be significantly more painful than failure of a CQI project.

The STAT Trump Card
Processes and staffing levels in a film library always address the various levels of urgency expected. When a single STAT request arrives in isolation, it is usually handled well. When multiple STAT requests arrive within a short time period, the film librarians triage the requests to the best of their ability and make sure that all such requests are completed as quickly as possible. Some requestors inevitably will feel that their requests routinely take too long, so they will begin labeling all of their requests as STAT. If too many requestors adopt this approach, the SOP that was so carefully crafted to handle STAT and routine requests in a timely, orderly manner begins to unravel.

Most Customers "Outrank" Film Librarians
The customers of a film library include patients, physicians (including radiologists), nurses, and clinic managers and their representatives. Each of these types of customers "outranks" the frontline film librarian. The institutional customers are usually in a higher pay grade position. Even many patients have higher educational levels than the film librarians. When conflicts arise, the film librarian is in an inferior position to defend and enforce film library policies. For many film librarians, the response is to accede to the insistent or irate customer. The result is a violated SOP and the occurrence of the downstream negative consequences the SOP was developed to avoid in the first place.

A more insidious result is that the film librarians begin to feel that their skills are undervalued and that no one respects them. By this point, SOPs are either routinely ignored or the enforcement of SOPs has led to ugly confrontations. A major task of the film library leadership team is to educate institutional leaders about the importance of compliance with film library SOPs. Institutional leaders, in turn, must demonstrate a firm, public commitment to enforcement of film library SOPs. If certain physicians or clinics are permitted to thumb their noses at the SOPs, word will get around to the entire institution and all of the hard work invested in the Six Sigma project will be wasted. Additionally, turnover of the truly excellent film librarians is likely to be high, because good workers will not endure such unsatisfactory working conditions.

Changes Are Difficult to Communicate and Sustain in a 24/7 Operation
As mentioned earlier, around-the-clock operation presents the challenge of establishing a two-way dialogue with workers on the shifts other than the weekday day-shift. During the early phases of a Six Sigma project, it was seen to be difficult to include all shifts in project team meetings. During the Implement and Control phases, it is equally difficult to ensure that new processes are implemented properly, then sustained, on all shifts. Special efforts must be devoted to monitoring off-shift performance and following up quickly when the data indicate that performance is not up to expectations.

Non-Film Library Stakeholders can Refuse to Change Behaviors
The film library affects and is affected by many elements of the institution outside of the radiology department, yet the film library leadership team has no control, and very little influence, over them. Upper management must recognize the important role of film library SOP compliance in accomplishing the objectives of the institution and must be prepared to be firm with physicians, nurses, and clinic personnel who fail to comply. For example, a film jacket loan period of 24 hours is established for on-campus borrowers. A physician may check out a film jacket and fail to return it promptly. When the film library receives another request for that film jacket, the physician may refuse to release it, thereby jeopardizing the patient's care by delaying access of the second requestor. A very clear message must be sent from the medical staff office that such behavior is unacceptable. If the physician happens to be the chief of surgery or the biggest revenue generator in the institution, the situation can become quite uncomfortable. Nevertheless, favored treatment of any person eventually leads to everyone expecting the same favored treatment.

A related aspect is the belief by some radiologists that, because the film library is part of radiology, radiologists should have higher priority when competing requests occur. For example, a patient may be scheduled for a clinic visit in the morning and a radiology procedure in the afternoon. Because the radiologist fears that the clinic will not return the films in time for the afternoon procedure, causing the radiology department to have a protracted cycle time, the radiologist may instruct the film library to never send films to a clinic when there is a radiology procedure the same day. If this is allowed to continue, the underlying issue of getting the films from the clinic to the radiology department in a timely manner is ignored, whereas it may be solvable with little effort. In the meantime, however, the physician who did not have the films available for clinic visits is going to bad-mouth the film library. Thus, the film library will get the rap for something that is turf-related and outside of the influence of the film library leadership team.

Film Librarian Job Is Rarely a Career Choice
In too many institutions, the film librarian position is barely above the housekeeping and dietary aide positions. The film library is viewed by employees as a stepping stone from the minimum wage jobs up to the "good" jobs. Their motivation is acquiring work experience and finding a better job, not becoming an excellent film librarian. When faced with the necessity to be self-starting or to go out of their way to satisfy an urgent or unusual request, it is unwise to expect them to rise to the occasion.

Tight Labor Markets
At the time this thesis was written, unemployment was at very low levels. A good film librarian quickly learns that he or she has acquired computer, time management, and customer service skills that may be much more richly rewarded in higher paying positions. Turnover will be high unless film librarian pay rates are upgraded to remain competitive.

Tight labor markets make it more difficult to attract good candidates for film library positions. Film libraries "offer" a high pressure, high volume, physically taxing, low pay environment. Even when candidates are relatively plentiful, it is difficult to attract and retain good people. When job candidates have lots of options, film libraries have a tough time winning them over.

A less obvious concern about tight labor markets is that managers are more likely to tolerate an employee who exhibits substandard performance, reasoning that substandard performance is better than having a vacancy. The reader of this thesis readily sees the fallacy of this argument, but the manager who is in the thick of the battle and who will feel the pain of the vacancy has a more difficult time coming to terms with it.

Job Security Issues as Radiology Moves to Filmless Operation
Radiology is making a transition to filmless operation. In some large institutions, the transition is well underway, even to the point where no films are being generated for routine clinical exams. As the film librarians (and candidates for vacant positions) begin to notice this transition, they quickly realize that the need for their jobs is going to disappear within a few years. This realization does not necessarily cause employee motivation to rise (Johnson 1998), and it certainly contributes to turnover among the good employees, who are able to find employment elsewhere.

A Six Sigma Project Will Be Expensive
A Six Sigma project is commissioned to solve a major problem, so the stakes on the outcome side are high. It should not be a surprise that a major payoff requires a major investment of institutional resources. A major Six Sigma project will require participation of key people from multiple components of the institution, all of whose time will represent a loss from their usual duties. Upper management must understand that performance in these usual duties will suffer, and the leaders of these areas must be shown tolerance for any reduced performance.

Not only must all stakeholders be involved, but only the best of the stakeholders should be involved. The project leader must be an experienced, accomplished employee. When this person is diverted from his or her normal duties for an extended time, the employee's supervisor is likely to feel considerable pain due to the employee's absence. Likewise, the best technologists, the best nurses, and the best clerks will be selected, again causing pain during their team meeting absences.

A Six Sigma project is data driven. Most institutional information systems are not set up to allow non-IS personnel to extract data that are meaningful to the project team. Thus, one or more IS personnel will need to be assigned in support of the team, because a Six Sigma project without data may be worse than doing nothing.

Obtaining Cooperation from Borrowers
A film library has many stakeholders. A partial list includes referring physicians, radiologists, patients, film librarians, clinics, radiology sections, and outside institutions. The film library manager has operational authority over only one of these groups. The other groups are not only not under the control of the film library manager, but they "outrank" the film library manager when disputes arise. If a clinic using a film jacket transfers the jacket to another clinic for a consultation but fails to tell the film library, the RIS will show the film jacket to be in the original clinic. A subsequent requestor will have to wait while the film library calls the original clinic and eventually tracks down the film jacket.

What impact does this lack of cooperation from borrowers have on the film library's Six Sigma performance? If one of the Six Sigma metrics is "time from request of a film jacket until film jacket delivered to requestor," any failure to release a film jacket, or any undocumented transfer of a film jacket, is going to show up as a higher-than-specification transaction. If one of the Six Sigma metrics is a question on a satisfaction survey that asks the physicians' agreement with the statement, "The film library always retrieves film jackets promptly for me," the metric will suffer.

Another way that lack of cooperation appears is what Harry and Schroeder (2000) call "the hidden factory." The hidden factory is the many actions performed by film librarians to compensate for failures in earlier handling of a film jacket. In the improper jacket transfer described in the preceding paragraphs, the film jacket may have been checked out to Clinic A. A physician in Clinic A took the film jacket to Clinic B and gave it to another physician for an informal consultation. In all likelihood, the Clinic A physician did not tell anyone else in Clinic A what he was doing, and the physician in Clinic B did not tell anyone in Clinic B that she had the film jacket. When a third physician comes to the film library loan window looking for the film jacket, the film librarian will look up the jacket in the RIS. Seeing that the jacket is checked out to Clinic A, the film librarian will call Clinic A and talk to the front desk receptionist. The receptionist will ask if anyone knows anything about the film jacket. After an extended hold on the telephone, the receptionist will deny that Clinic A has the film jacket.

The film librarian will then look up the patient's appointment schedule in the HIS, figure out the name of the physician who actually saw the patient, and call the clinic asking to speak with the physician. With luck, the physician's nurse will tell the film librarian that the physician mentioned taking the film jacket to Clinic B for a consultation; the nurse will not remember that a physician's name was mentioned. The film librarian will call Clinic B and talk to the front desk receptionist. The receptionist will look on their film jacket cart and will not see the patient's film jacket, and he will not see the patient on their appointment list, so he will deny any knowledge of the film jacket. If the film librarian is persistent enough, more phone calls and/or visits to Clinic A and Clinic B searching for the film jacket will occur, and the film jacket will eventually be found and checked out to the new borrower.

If the Clinic A physician or the Clinic B physician had made a 30-second call to the film library to report the transfer, this entire saga could have been avoided. As it was, the new requestor was frustrated, the film librarian was frustrated, the personnel in the clinics were frustrated, and valuable time was lost by a lot of highly paid people. But all of this frustration and waste is undocumented and not visible-it is the hidden factory that eventually turns a poor-quality situation into an acceptable situation, but at great cost.

CONCLUSIONS
When we set out to "fix the film library," we knew that Six Sigma tools had worked very well for our CT throughput problem. We also felt that the fundamental Six Sigma philosophy could be used in a film library environment. But we were less sure about the degree of adaptation we would need to make, because many of the Six Sigma tools are best suited for a manufacturing model.

The basic concept that any process could be subjected to the Define-Measure-Analyze-Improve-Control (DMAIC) analysis proved to be true. The DMAIC technique appears to be equally applicable to the manufacturing model and to the services model.

Having said that, however, it is much more difficult to acquire data for a service-model process than for a manufacturing-model process. The Measure and Control phases are a much bigger challenge for the service-model process, because the subprocesses themselves may be harder to quantify and the data may be difficult to gather once a metric is determined.

A Six Sigma project requires a sizable investment of institutional resources, primarily in terms of the time spent away from their regular jobs by the best people in each stakeholder category while participating in Six Sigma project team meetings and in doing any project "homework." Also, most Six Sigma projects would benefit from having a full-time leader assigned to the project, rather than asking someone to try to fulfill Six Sigma responsibilities in addition to his or her regular duties. Insti