University of Southern California Data Evaluation Discussion
Please write a discussionDiscussion prompt:Prompt : In the healthcare industry, everything is measured. From drip rates to dosages, it seems as though numbers are everywhere. But how does this reflect the care given to the patients? After reviewing chapter 5, reflect on the following:Which type of data (qualitative or quantitative) do you think yields the most information? Why?Which type of data (qualitative or quantitative) is most effective for forecasting future trends? Why?Which practices can a healthcare manager use to ensure that the data used for forecasting trends is accurate?Provide examples of data evaluation in different types of facilities. Provide examples of data evaluation for different types of issues, such as billing or loss of revenue. ALL citations and references needs to be APA 7th edition format. THANK YOUtextbook may be used as a reference. The APA format for your text is as follows:Langabeer, J. R., & Helton, J. (2016). Health care operations management: A systems perspective (2nd ed.). Burlington, MA: Jones & Bartlett LearningCHAPTER5Please write a discussionDiscussion prompt:Operations Research MethodsGOALS OF THIS CHAPTER1. Describe the application of operations research methods in health care.2. Understand how to identify and eliminate bottlenecks.3. Use forecasting methods to estimate patient volumes and demand.4. Understand the concept of capacity and its relationship to demand.5. Explain why tracking systems can improve process flows.6. Describe bar codes and radio frequency identification and their roles in operations management.Health care facilities are busy places with hundreds of people constantly coming and going. To maintain efficient operations, organizations must optimize patient and other process flows. This entails: Understanding patient demand. Aligning capacity and resources with demand. Using de-bottlenecking approaches to improve throughput. Managing patient and asset flows through tracking systems.The use of tools and techniques such as operations research help to incorporate quantitative methods that can improve decision making. Techniques such as wait time minimization models and forecasting algorithms help to support improvements in process and patient flows. To make informed decisions about changing processes, decisions must rely on data, not just subjective gut feel. This chapter discusses these concepts in detail.OPERATIONS RESEARCHThroughout the years, operations research (OR) has been defined in many ways, often using different terms to describe the same body of knowledge and methods. In England and Europe, operations research is commonly called operational research, although the terms are synonymous. Similarly, the term management science (MS) has become popular in some schools of business, though usage is mixed. Operations research is still used primarily in industrial engineering departments and other schools outside of business, but for the purposes of this research, all of these terms (i.e., operational research, management science, and operations research) are considered identical and interchangeable.The simplest definition is what the Institute for Operations Research and the Management Sciences (INFORMS, 2014b) uses today: “a discipline that deals with the application of advanced analytical methods to help make better decisions.”Generally, the operations management and management sciences can be combined using the term “OR/MS” and describe using a scientific view and quantitative methods to support managerial decision making (Hillier & Hillier, 2008; Anderson, Sweeney, Williams, & Loucks, 1999). The Operational Research Society (n.d.) defines OR as “the discipline of applying advanced analytical methods to help make better decisions”; it posits that “by using techniques such as problem structuring methods … and mathematical modelling [sic] to analyse [sic] complex situations, operational research gives executives power to make effective decisions and build more productive systems.”The three key terms used or implied in most definitions are structured, decision making, and improvements. Structured implies that techniques will focus on using rigor and sophistication. Many times this also requires a reliance on data and a mathematical or quantitative basis, although this is not always the case. Traditional methods can be classified as “hard” (i.e., relatively mathematically intense) and “soft” (i.e., rigorous but qualitative, which stresses structured problem solving for complex and messy problems that cannot be solved by traditional math models). Advanced quantitative methods, such as simulations, optimization, and mathematical models incorporating probabilities and other variables, are often tools used in this scientific process.The focus of OR relies on improving the outcomes of decision makers through use of better methods and techniques that comprehensively and systematically produce options, scenarios, and better results (Trick, 2003). Exploring data in new ways, using new techniques, or building models that can help determine the effects of decisions so that managers and other decision makers can improve the quality of their decisions is a fundamental goal of OR.Finally, OR is about making improvements in performance (Ackoff & Sasieni, 1968). Scientific rigor and better quality decisions should result in improved operating, financial, or strategic performance. OR is not supposed to be arbitrary or exploratory for its own sake; the results need to be better through the OR if the discipline is to grow and thrive. Thus, the new slogan for INFORMS and other OR organizations is the “science of better,” focused on improving outcomes and results.Based on its focus and intent, it is important to evaluate the scope of OR for the health care industry, both currently and its future potential.MANAGEMENT DECISION MAKINGIn a completely rational model explaining how managers “do” (descriptive models) or “should” (normative models) behave in organizations, the emphasis is placed on maximizing outcomes of the decision process. Management of any organization would identify the goals of a specific problem or situation, generate alternatives, and select the one that is optimal. In this environment, OR methods would appear to be highly complementary. OR techniques allow managers to seek alternatives; evaluate these choices using probabilities, risks, and other variables as key criteria; and then model potential outcomes. Unfortunately, managers in organizations do not always behave rationally, which has opened the decision sciences field to a much less rational approach to decision making. Due to behaviors, politics, and other potential influences, the rational model is not the norm.OR methods play a vital role in the management decision-making process. For these purposes, decisions are defined as a choice between two or more alternatives, and management decision making is the process in an organization by which decisions are made.Because managerial decision making occurs at higher levels of an organization and typically involves major commitments of resources or changes in strategic direction, this research seeks to understand how decision processes work in health care organizations. Understanding the unique aspects of this industry is important because they have been described as service intensive and goal ambiguous in many respects. Management theorists, such as Harrison (1987), have suggested that as the organization’s environment becomes more complex, there is a higher use of “judgment” in decision making and less procedural computation, as in a rational model of decision making. Better understanding of the health care industry’s organizational environment and the specifics of the decision-making process can offer greater insight into how decisions are made, which criteria are used, how the search for alternatives occurs, and the role analytical or quantitative methods can play in the evaluation of alternatives in decision making.A BRIEF HISTORY OF OPERATIONS RESEARCHOR seeks to apply structured analytical techniques to improve decisions made by managers. These can come in the form of qualitative (i.e., soft) techniques or the more commonly cited quantitative techniques. For this reason, it is typically described today by management theorists as being its own “school” but as a derivative of the scientific or classical school of management thought (George, 1968; Salveson, 2003), which evolved from the work of Frank and Lillian Gilbreth, Frederick Taylor, and others.Based on most accounts, the OR discipline can be traced back to the pre–World War II 1930s and 1940s. The British government brought together several interdisciplinary teams to apply science to investigate military tactics. OR groups were used to develop the first radar system around 1941 to help the British military track and identify aircraft. This led to the use of OR for improving other communication systems, and it became instrumental in the Royal Air Force, Army, and Navy (McCloskey & Trefethen, 1954). It was due to these efforts to incorporate scientific and mathematical information into military activities that OR found its niche. Subsequently, operations researchers were deployed to numerous projects throughout all of the British armed forces. With success in England, OR began to move into U.S. military operations during the early 1940s.During the latter part of that decade, the Massachusetts Institute of Technology developed courses in OR, and in the early 1950s a complete curriculum was developed in OR/MS by Columbia University, Case Western Reserve University, and others. Many universities in England followed suit and developed OR short courses during this time frame as well. The Operational Research Society of the UK (previously the OR Club) was formed in 1950 and is considered to be the world’s oldest OR society (Symonds, 1962). The Operational Research Quarterly began publication in 1950, and the journal Management Science was launched in the United States in 1952, both providing avenues for OR in which to publish and expand. Annual conferences soon began uniting academic researchers worldwide, and since this time the OR discipline has continued to thrive (Schrady, 2001).Based on its military beginnings, OR quickly became known for incorporating scientific processes into decision making, and it is sometimes called a systems approach (Ackoff, 1971; Riggs & Inoue, 1975). A systems approach refers to how OR attempts to study the underlying behavior and structure of the systems—or interrelated set of processes, events, and activities—that define most problems and decision realms.This systems approach recognizes that forces and relationships exist between the environment and the internal processes, and that they can be analyzed closely, modeled, and then used for predicting or simulating results. Systems can be defined formally as the “collection of activities that share in their transformation to achieve a defined purpose” (Riggs & Inoue, 1975, p. 70). When systems are modeled, they then can be manipulated in various ways to estimate the effects of changing policies or decisions. Therefore, when applied to management, OR has shown that through a variety of methods (e.g., linear programming, optimization) better or improved results can be identified.OPERATIONS RESEARCH APPLIED TO HEALTH CAREOR was applied to health care as early as the 1950s, with one of the first OR articles related to medicine published in the Operational Research Quarterly (Bailey, 1952). This early work was sponsored by a trust of the British National Health Service and led to a small collection of articles. Around the same time in the United States, the Johns Hopkins Hospital assigned a contract position (joint with the Army Operations Research) for a full-time director to assist in hospital management decisions (Flagle, 2002). From the 1960s through the early 1970s, there appeared to be a growing interest in OR, with the field gaining significant momentum around 1970.It was then that the Operations Research Society of America (now part of INFORMS) held its first symposium on health services delivery (Young, 1969). The Health Applications Section of INFORMS was created in the early 1970s and currently has more than 500 members (INFORMS, 2014b). Subsequently, in 1975, the European Working Group on Operational Research Applied to Health Services was formed and now claims 242 members in more than 30 countries (Operational Research Applied to Health Services, 2014). The result of these societies is a much broader, global effort to apply OR to health care delivery processes. Both of these groups have conducted annual meetings and conferences to continue to encourage innovation in and research on OR topics in health care. In addition, in the late 1970s, the Society for Medical Decision Making was formed to help introduce more quantitative and sophisticated methods into health care decision processes.Prior to this time, there were several articles published on decision methods and quantitative techniques in health care administration, but they were less focused on the unifying themes, which center on building quantitative models of systems that are stochastic in nature and ultimately patient focused.Since this time OR has developed some momentum, although not as much as might be expected considering the size and complexity of the health delivery system. Carter (2002) described the lack of OR focus in health care when he stated in his research that only two members of the entire INFORMS membership community were professionals working in hospitals or health organizations and that fewer than 2% of the entire membership body was involved in the Health Applications Section. Figure 5–1 shows a brief time line of the significant early events in health care OR.About 15 books have been written that focus exclusively on health care and OR. The most significant include Operational Research Applied to Health Services (Boldy, 1981), Application of Operations Research to Health Care Delivery Systems (Fries, 1981), and Operations Research in Health Care (Shuman, Speas, & Young, 1975). More recently, the edited collection from Brandeau, Sainfort, and Pierskalla (2004), Operations Research and Health Care, provides detailed application of OR methods to health operations and clinical processes. Health Operations Management, edited by Vissers and Beech (2005), focuses on using OR and operations management to improve patient flows and logistics in health care organizations. It was the first book to concentrate specifically on this area and discusses basic concepts and frameworks for classifying processes in health, provides methods for analyzing supply and value chains, and offers multiple case studies on outpatient clinic scheduling, master planning, and admissions planning, among others. Other works include Blumenfield (1985), Kessler (1981), and Koza (1973). Nearly all of these books include a summary of key applications and methods used in health care, and most have focused on either clinical decisions or patient logistics.FIGURE 5–1 Time Line of Significant OR EventsPierskalla and Brailer (1994) developed a bibliographic survey of OR applications in health care and describe numerous applications for OR methods. Carter (2002) maintains a database of OR research articles focused on health care, and there were more than 800 in 2002. A simple Google search shows about two million hits for the combination of “operations research” and “health care,” although most of these are likely related to clinical or patient care uses of OR. It does appear, however, that OR has started to penetrate at least part of the health care field in certain areas.OPERATIONS RESEARCH APPLICATIONSGiven the political and community concerns about health care access and costs, it is critical to use more sophisticated tools for solving problems involving variability, uncertainty, and risk. One of the key areas where OR methods can contribute is in the modeling of patient volumes and flow through organizations and health systems. Patient flow means the movement of patients from initial point of entry or service to the point when the patient exits the system. This entails understanding the key processes and transactions that patients must experience in multiple departments (such as admissions, triage, treatment room, laboratory, pharmacy, and finance) and through the network of providers. This process perspective in health care management modeling is extremely important.Linear programming has been somewhat widely used to minimize labor costs in health care settings. Linear programming is a mathematical technique designed to make decisions that optimize the trade-offs necessary for resource allocation. Linear programming problems focus on maximizing (usually revenue) or minimizing (usually costs). This represents the objective function of the problem. Constraints are the restrictions that are inherent in the problem that limit the degree of change. For example, if a hospital chooses to minimize nurse labor costs but must ensure that at least one nurse is on shift at all times, this represents a constraint.Simulation models have also been applied to labor staffing problems. A simulation model is a computer application that predicts the behavior or performance of a process or how something may perform in the real world. Discrete event simulation models allow for changes in resources and inputs. For instance, a model of the emergency department can show patient flow and movement if resources are changed, tasks are modified or realigned, or variability in demand occurs. Commercial software for simulation is widely available.Revenue Cycle ManagementGiven today’s reimbursement models, operations management in the United States is largely focused on maximizing revenues (and not just minimizing resources or expenses). Financial decisions arise from a contracting perspective with third-party payers and insurers, and it is necessary to ensure that the reimbursement from payers exceeds the operational cost in each service line. This process is called revenue management, or revenue cycle management. Revenue cycle management is the process of managing claims processing, setting payment practices, and generating revenue. It should be an analytical method for determining prices to achieve specific objectives, such as greater demand, higher utilization, or maximizing margins. Price (payer reimbursement) optimization models can be built that minimize risk (the variance in net profitability of a payer contract), which results in a formula such as:where pj is equal to the price of an input or patient service j, dj is the demand, and cj is the cost for service j. Several constraints are used (such as an equation to define minimal net margin requirements) as well as a variety of other parameters.Risk and Financial Simulation ModelsFinancial simulation models were described in the early 1970s as potential OR tools for improving planning outcomes. Many large Fortune 500 corporations constructed formal models that used mathematical programming to dynamically explore changing financial policies, debt leverage, or changes in operational conditions. In essence, these tools help to create pro forma financial statements given certain assumptions and historical relationships. The models range from simple, deterministic, and top down to more complex stochastic, multivariable simulation models. Simulation models allow managers to play “what if” using many different assumptions and scenarios.Most simulations in health care utilize Monte Carlo simulation analysis, which combines probability theory with random number generation and defined distribution patterns to iteratively simulate outcomes. Monte Carlo methods have been incorporated into spreadsheet solution solvers and programs such as @RISK, RiskAMP, and Crystal Ball. Software tools that incorporate Monte Carlo’s statistical powers allow managers to simulate budgets and plans.DE-BOTTLENECKINGAssume that a hospital admissions department has two full-time employees who admit patients into the hospital during the 8-hour day shift. Each employee has a computer and monitor with access to the admission system, which takes approximately 30 minutes to complete for an average new patient admission. Therefore, the maximum capacity of this process is 32 new patient admissions daily (2 employees × 8 hours × 2 patients per hour). This 400-bed hospital has a 72% occupancy rate and frees up approximately 40 rooms daily. The challenge for this hospital has always been to get more patients into the process earlier.As described in this example, only 32 patients can be admitted based on current capacity at the entry point of the process, even though 40 is the actual demand or theoretical capacity further downstream in the process. Therefore, if more than 32 patients arrive, a bottleneck would exist (Demand > Capacity). A bottleneck is a choke point, or a point in a process where demand exceeds available capacity. In other words, a bottleneck can occur at any point where capacity is insufficient to meet demand due to physical or logical constraints. A bottleneck can also be a person, role, or any other barrier or obstacle to cooperation and work performance among departments.One of the keys to increasing throughput or capacity is to remove these obstacles or bottlenecks, which is called de-bottlenecking. In the preceding example, potential solutions for reducing the bottleneck might be to add labor (recruit additional employees), reduce the process time below 30 minutes (invest in systems and procedures that allow for faster processing), or remove forms or tasks that are redundant. All of these should be considered. Figure 5–2 provides an example of a bottleneck, shown visually as a funnel. In a funnel, the neck of the funnel limits volume throughput. In other words, the narrowest part of the funnel determines how quickly volume can be moved through the process, thus creating a bottleneck.FIGURE 5–2 Process De-BottleneckingThe key to being able to de-bottleneck is to thoroughly analyze both demand and capacity to determine where the bottleneck exists. To be successful in improving processes, it is important to determine if the bottleneck is the result of an inability to handle demand at all times, or just at a specific point in time, as well as to discover if other barriers to throughput exist.Bottlenecks can occur at any point in the process: where a patient enters the hospital, at registration, during transition time of equipment, and at time of discharge. The earlier the bottleneck exists in the process, the fewer the number of patients (or throughput) that can be pushed through the system. Alternatively, a bottleneck at the end of the process typically results in wait times and inefficiency that can eventually affect the entire system. Eliminating a bottleneck at the beginning, only to discover that more exist in the middle or end of the system, will not help increase throughput. That is why it is important to study all processes systematically and to identify those obstacles that really limit capacity.FORECASTING PATIENT DEMAND AND VOLUMESForecasting patient demand is the first step to thoroughly understanding changes in activity levels over time. Comprehensively defining patient logistic flow involves tracking volumes intraday, as well as throughout the week, using time-series data. If a hospital does not exhaustively know patient volumes and traffic levels, it cannot project volumes for individual departments and services throughout the day. Without understanding demand, it is nearly impossible to align resources and capacity with demand.Forecasting is a collaborative process that estimates the volume of patients who will be served over a specific time period. More precisely, it is a projection of demand that will occur along three dimensions: service type, location, and time dimensions. Service type includes the specific procedures performed or the staff involved in the effort. Location includes the specific department, unit, floor, or other geographical location that performs the service types. Time refers to the hour, day, week, and month that the demand was met. Forecasts are based on time-series data. Time series refers to a set of values or observations at successive points in time.Forecasting, by definition, is the practice of making a prediction or estimation about the future (Makridakis, 1996). It involves modeling the past to define the future. Demand forecasting, then, is the practice of predicting future demand to accomplish specific business goals, such as more accurately planning how many beds or clinics are needed or how much staff to hire. Performing forecasting really well allows managers to minimize unproductive wait time, maximize customer service, and in general improve operational efficiencies—the goal of operations management.There are two major types of forecasts: qualitative and quantitative (Armstrong, 2001). Qualitative methods include mainly market research, executive opinion, or Delphi methods to make subjective or judgmental decisions about the future without relating demand to historical performance quantitatively. Qualitative methods for demand forecasting may be useful for gauging potential demand of entirely new products that have no relationship with other products and cannot be reasonably estimated statistically. Qualitative forecasts of new products that a surgeon or specialty area requires may be the best use of these types of forecasts.In health care, forecasting should primarily be based on quantitative methods. Quantitative forecasts can be broken down into two major types: univariate and multivariate methods. Univariate can be defined as dependence on a single variable; univariate methods attempt to forecast demand by exploring historical data relative to a single variable, such as number of patients, procedures, or items. In standard hospital environments, all of the transactional details for patient volume are captured in the clinical scheduling or information system, such as the number of admissions or the number of surgeries. In addition to this, clinical systems also capture the date patients are admitted and discharged, which procedures were given, the drugs and supplies administered, and prices charged. Reliance on any one of these transactional data elements is a univariate method, which reflects the single variable that will be analyzed to assess historical usage levels and then, based on this analysis, used to make a projection about future values.With univariate forecasting, there are several different statistical models that are often called on to assess patterns in the data. These include such methods as Box-Jenkins, linear trend analysis, exponential smoothing, moving averages, least squares, and many others. These models all have specific advantages and disadvantages that make them useful for single-variable forecasts. Some of these methods will be discussed in the rest of this section.Moving Average ForecastA moving average calculates an average historical figure for a specific time period, such as the last 3 rolling months, and then extrapolates this average forward. This is a very imprecise type of forecast because it actually lags the relevant time period. In a constantly growing environment, moving average can be too conservative, and it is underbiased in its predictions. The mathematical calculation of a moving average forecast is:The term moving indicates that as a new data point becomes available, the oldest data value drops off and is replaced. In other words, if you were calculating a 3-month moving average, the calculation would sum the last 3 months’ actual historical data values and divide the total by 3. For example, if historical data values were 10, 20, and 30, the moving average forecast would be 20, calculated as follows:Trend ForecastingAnother type of forecasting algorithm is based on simple trend analysis. Trend analysis looks for linear upward or downward movements in data and then extrapolates them going forward. Trend models are effective when demand for a product exhibits fairly consistent demand over time. The basic formula for calculating trend forecasts uses the initial starting point or intercept and adjusts for slope (or angle of the trend) over time. This is often called rise over run, and it is mathematically calculated as follows, where y is the forecasted value, a is the y-axis intercept, b is the slope of the regression line, and x is the independent variable.Other MethodsSmoothing methods in demand forecasting are useful because they use a factor to weight the most recent demand observations more than in previous periods, and they help account for errors in previous periods. Smoothing, whether it is exponential (i.e., discounts previous periods with a higher magnitude as the observations age), double exponential, or third order, focuses on improving forecast accuracy by giving more weight to the most relevant historical periods.Box-Jenkins is a slightly more complex model that uses regression or curve-fitting techniques at predefined time intervals for the single variable being analyzed. It combines single-variable linear regression with a moving average technique to achieve good results from univariate methods.A much more comprehensive set of forecasting methods falls within the category called multivariate. Multivariate methods attempt to use more than one variable to help better explain or model the past to make more accurate forward projections about the future. Although factors such as seasonality and cyclicality (i.e., business cycles that repeat similar patterns over time) can be detected and modeled using advanced univariate methods, they are much more common in multivariate methods. Using multiple variables to help make predictions about the item being forecasted allows seasons and cycles to be combined with other causal factors (e.g., pricing, promotions, events) to model relationships with other variables and improve forecast accuracy.The most common form of multivariate demand forecasting in large-scale causal forecasting is multiple regression. Multiple regressions use other contributing factors to help better explain the past and predict the future. For example, when forecasting demand for a downstream department (e.g., radiology), we might find a causal relationship with number of admissions, number of square feet in the hospital, patient acuity levels, case mix index, or other variables.Excel and other spreadsheet packages can be used to create both univariate and multivariate forecasts. The Excel functions—trend, forecast, growth—and many others allow users to create forecasts with time-series data for linear trends, exponential curves, and moving averages. They are fairly simple and straightforward. The transactional data can be organized to show the time dimension, or periods, and the corresponding item usage. Then use of Excel’s “=forecast” or similar function can be implemented to point to the known dependent and independent variables, which will then plot the forecasted value. This can be shown in spreadsheet or graphical views, as Figure 5–3 illustrates.Similarly, analysts can use Excel to simulate multiple regressions, using the data analysis add-in package. These regressions are slightly more sophisticated than simply using linear trends because regressions attempt to fit or model the historical transaction data to predict more probable future estimates.FIGURE 5–3 Forecasting Volumes in ExcelThe Forecasting ProcessThe process of forecasting demand involves four key steps:These steps are typically performed in a wide range of time intervals, from short range (next day or week), intermediate (next month), or long term (next year or two). For demand forecasting as it relates to patient volumes in health care, forecasting is typically done in short and intermediate time intervals. Longer-term forecasting is usually done for strategic planning purposes, such as for adding bed capacity or capital investment in new space or equipment.The process starts with an analyst, operations manager, or planner identifying or isolating what is to be forecast; patient admissions, appointments, visits, clinic registrations, research protocols, supply usage, and pharmaceutical sales are common forecasting applications. Typically, forecasting is used to make specific business decisions, such as how many of each type of pharmaceutical to order next week or how many outpatients to expect next month. Most health care forecasts tend to focus on univariate methods, where time-series data are forecasted.Once identified, the planner must gather all historical data for this variable. Data collection may come from a variety of systems, depending on the time-series data selected. For example: Appointment data reside in the organization’s scheduling system. Admissions data come from the admission discharge transfer system. Pharmaceutical or supply information is stored in an enterprise resource planning or other purchasing system.Once the system has been selected, either an interface or a download of historical data will have to be requested from the information systems group, unless the data are available for export directly. A choice of any attribute or other characteristic that describes the data values might also be collected. Time-series data, which represent values over time, are necessary for most mathematical forecasts to predict the future.Once these data are in place, they should be incorporated into either a spreadsheet solution (for simple forecasts) or a sophisticated forecasting package. There are many excellent software solutions that can inexpensively and simply model and analyze the historical demand patterns to help understand the past and make accurate projections for the future.The planner then needs to analyze the data to make sense of the forecast and ensure that the results seem appropriate. Closely examining the forecast and history will ensure that there were no issues with the data and that the forecast is reasonable. Finally, the analyst must continually monitor and adapt the forecast to ensure that forecast accuracy increases over time (or, alternatively, that the error rate decreases). This can be accomplished using tracking signals or by monitoring forecast errors such as mean absolute percent error. Error rates should be used in the monitoring process to adapt or refine the model to obtain better projections the next time.It is important to focus on the data variation, whether it is random or predictable. One of the goals of demand forecasting is to reduce the uncertainty or variability that inherently exists. Ways to do this include looking at the source of the data, examining the frequency of the process, searching for patterns in volumes or demand behaviors (e.g., spikes due to purchasing increases to draw down operating budgets at year end by departments), and identifying the best level at which to forecast.BASIC PRINCIPLES OF FORECASTINGThere are some principles of forecasting that should be kept in mind to improve results. First, forecasts are always inaccurate. There is no process that will repeatedly match forecast to actual. That is why it is important to quantify the error and use it to adapt the forecasts for the future. Forecasts made at high levels (e.g., total number of inpatients weekly) are always more accurate than at the lowest levels (e.g., outpatient appointments in a specific location at a certain time). The more granular the forecast, the less precise it will be, but that is typically where the value of forecasting really can be found. Creating forecasts at the lowest levels and then grouping them accordingly for planning purposes is vital to a healthy process. Finally, it must be remembered that forecasts are only the starting point for the planning process—forecasts help provide a basis for further refinements and the selection of a most likely scenario for the future. Here are some additional guidelines and principles.Level of HierarchyDecide on the level at which you wish to forecast. Forecasting at the lowest levels (typically, a patient procedure at an individual location in the hospital) provides significant levels of detail, but if this detail is not necessary it should not be used. Aggregation of the data allows for more strategic viewing, but some of the richness of the underlying data is lost. Thus, a trade-off exists between the details gained and the additional level of effort required. Forecasting attributes allow a different perspective, which may be useful during negotiations with suppliers. As much as practical, use downstream transactional data. The best source of demand is actual customer requisitions or items that have been directly issued or charged to patients, not warehouse orders or inventory movements.Decompose the ForecastUnderstand the real demand-forecasting problem first; then break it down into smaller, less complex parts. This is the principle of decomposition, which uses a general approach to drill down into more specific, narrower areas.Time HorizonDecide on a realistic forecasting horizon. Although the business process should dictate the forecasting horizon, shorter time horizons provide more reliable results. For most demand forecasts, forecasting out more than 3 to 6 months is not optimal.Apply Quantitative TechniquesUtilize a mathematical or statistical forecasting application if at all possible, preferably one that is integrated with the hospital’s existing information systems. More advanced tools can help to automatically isolate the effects from seasonality, pricing, operating cycles, or other causal factors and apply appropriate algorithms without significant manual intervention. Also, use combination approaches if possible. Weighting of specific statistical models based on their historical standard errors, such as the Bayesian approach, tends to generate significantly better forecasts than single-forecast methods. Some excellent solutions that are widely used in various industries include Forecast Pro (www.forecastpro.com), SAS (www.sas.com), and Logility (www.logility.com).Simplicity FirstTry forecasting in the simplest fashion possible, and add complexity only if necessary. If multiple demand patterns generate poor forecasts due to complexity or scale, look for causal relationships and better statistical models to build a more robust solution. Be careful to not “overfit” the forecasting models. In many cases, too many variables are used in multivariant forecasting. Adding this complexity does not always result in improved forecasting accuracy, so be mindful of challenging the concept that more is always better by validating each variable used in the model.Reliable Data SourcesUtilize reliable data sources. Data coming from hospital resource planning or other clinical systems tend to be the most accurate. It is important not to use any systems or data points that are incomplete or have errors or missing data. Look for alternative sources of data that can reliably feed the demand forecasting system to generate the most valid, reliable results.Cleanse the DataCleanse or scrub the data using business rules. Data coming from most hospital systems or business warehouses today tend to be inaccurate in some manner. Cleansing or scrubbing the data by applying logic and business rules (such as “do not import any history that has negative values”) results in higher-quality forecasts.Causal RelationshipsAvoid making predictions on predictions. Causal relationships that are highly judgmental about the future (e.g., expected changes in interest rates or weather) tend to serve as poor causal factors because their forecast is usually inaccurate and unpredictable. Basing your product’s demand forecast on these forecasts often yields unreliable results.Exception ReportingMake use of exception reporting to flag problem areas. Specific forecast combinations that may be problematic should be flagged based on specific business rules (e.g., where forecast error is greater than 15%).Graphical Analysis of TrendsView forecasts graphically—visual representation of data allows users to better interpret results and identify inconsistencies. Graphical analyses allow patterns to emerge more readily than in straight tabular forms.Apply Insight and IntuitionNever use statistical results without applying business intelligence. We know that forecasts are always wrong, so it is important to apply human business intelligence to ensure validity within the current context. For example, a statistical forecast might generate specific values, but if the models applied did not know that a clinic is closed on Mondays, the demand will be overstated.Use Unconstrained DataDo not forecast based on constraints. For example, if historical patient visits were down last month because of a major snowstorm that limited patient volumes, this constrained or reduced demand is artificial and biases the forecasts. Forecasting based on these artificially low figures should be explained through a causal event, by adding “pseudo” sales to account for an unrealistic month or by eliminating that period as an outlier.Measure Errors and Accuracy LevelsMeasure forecast accuracy in multiple ways. Use multiple measures of forecast accuracy or error to help remove the distortion that occurs when firms become fixated on a single measure. Use the forecasting error to improve the next forecast so that the errors generated in the last forecast are fed back into the next one to improve the quality of the forecast. Typical forecasting software or spreadsheet solutions will provide at least the mean square error rates, which is a simple statistical calculation that squares the difference between the forecast and the actual values. Another similar calculation is the mean absolute deviation (MAD), which is the sum of the absolute difference between the average of the actual values and the forecast, divided by the number of observations. Mathematically, this is calculated as follows:For example, if the time-series forecasted values were 10 in August and 8 in September, and actual values observed for those months, respectively, were 9 and 7, the MAD would be 1. The first step is to calculate the mean value of the actual data, which would be 8 in this case ([9 + 7] ÷ 2). Second, subtract the mean from the forecast value for each observation. Third, take the absolute value (the value regardless of the positive or negative sign) of the difference. In this case, that is 2. Fourth, divide this by the number of observations (2). Therefore, the MAD is 1.0, calculated as follows:Tracking the mean absolute deviation or the mean square error allows forecasters to compare how accurate their forecasts are over time to continue to refine and improve the calculations and methodologies.CAPACITY ANALYSISOnce demand is known, it is extremely important to understand how much capacity exists. Capacity refers to the amount of resources or assets that exist to serve the demand. In health care, capacity can be measured in terms of multiple resources, including: The number of available beds, treatment or examination rooms, and clinics. Labor availability of physicians, nurses, and other providers. Availability of key medical technologies and equipment (e.g., diagnostic imaging, x-ray). Supplies and other resources. Elevators, hallways, and other facility space. Cafeteria, parking, and other support services.Capacity analysis requires detailed understanding of the organization’s resources, including labor, technology, and facilities. Documentation of this capacity should be done using time-series data, similar to how demand-series data were treated, to track capacity changes over time.For example, if a hospital has a magnetic resonance imaging (MRI) machine, the assumption may be that it could operate 24 hours per day, 7 days per week. This is called the design capacity, which is the maximum stated or theoretical output for a resource. However, when closely analyzing the equipment over a period of time, it would be discovered that there is necessary downtime for maintenance or repairs or other reductions to stated capacity. Therefore, the more important capacity term is effective capacity. Effective capacity adjusts the design capacity with average expected utilization rates. For example, if average operating efficiency or utilization is 75% on the MRI machine, then the effective capacity is 18 hours, calculated using the following equation, where Ce is effective capacity, Cd is design capacity, and U represents utilization rates:Consider this example. A hospital clinic has two treatment rooms and offers services that typically require 30-minute appointments. Therefore, approximately two patients can be seen each hour in each room. The daily design capacity of this system, based on an 8-hour day, is therefore 32 (2 × 8 × 2). This is the design capacity given “average” procedure types for the clinic and standard cycle times (the process for calculating normal times will be discussed later in this chapter as part of time and motion studies). However, these averages do not take into account any deviations, such as scheduling problems, patient delays, or transition times in between patients. Historically, the average clinic room utilization is 72%. Therefore, the effective capacity is really only 23 patients per day.CAPACITY PLANNING: ALIGNING CAPACITY WITH DEMANDCapacity planning refers to the planning process for aligning capacity with demand, analyzing whether resource constraints (shortages) or surplus (excess) exist at all points in time. If 100 hours per week of physician labor is available to a specific clinic, yet demand forecasts suggest 1,400 procedures and 120 hours of potential patient demand, there is a mismatch or lack of alignment between capacity and demand. This is very common in health care, where either demand or capacity is limited (or both). Creating a strategy for effectively dealing with this takes five key steps:1. Forecast patient demand at detailed levels (by hour, location, etc.).2. Using productivity estimates, translate this demand into capacity requirements (where patient flow exists; which resources will be used).3. Analyze current level of capacity in terms of hours of labor or equipment available or numbers of other resources. Translating capacity into a per-hour basis is the most common measurement (e.g., 11 hours of equipment time available on an MRI daily, or 362 hours of nursing labor).4. Estimate the delta (or change) between capacity and demand on a per-hour or other basis.5. Develop a strategy for aligning capacity with demand.Typically, this involves mapping supply and demand over time, graphically analyzing the data, and then developing plans for adding or removing capacity. The most common strategies for dealing with capacity constraints are as follows:1. Increase capacity, where capital or operational dollars allow. Adding capacity suggests purchasing new capital equipment that could allow the facility to perform more procedures or operate longer hours. Organizations also add capacity by hiring more labor, adding swing beds, or increasing total square footage for new clinics or rooms. Other options include contracting with other facilities to provide additional capacity or subcontracting certain service lines. The use of return on investment models should be utilized to ensure that the benefits of adding capacity are greater than the marginal costs to invest in the capacity expansions.2. De-bottleneck, which may free capacity. The use of process engineering tools can identify bottlenecks, and targeted improvement methods can eliminate them.3. Reduce demand, where possible and profitable. This may include reducing the services or procedures provided or redirecting patients to other competitor or partner facilities.4. Transfer capacity from other areas (i.e., sometimes capacity exists in certain areas or departments that is often not needed, which can be used to fund capacity expansions in other areas). For example, if facilities or space is the issue, square footage can be reduced in one department and provided to another.MINIMIZING WAIT TIMESTypically, one of the biggest bottlenecks in health care involves the issue of wait times. Wait time is defined as the time interval during which there is a temporary cessation of service. Alternatively, it is the amount of time that has elapsed or has been delayed from the start point until some action occurs or until service is provided. Most of us experience wait times everywhere in our daily lives, even if they are brief—at a gas station, restaurant, convenience store, or coffee shop.In health care, wait times are frequently a source of poor patient satisfaction and process inefficiency. In emergency rooms, for example, wait times of up to several hours are common. Some waits are more acceptable than others. Another common example of wait time is when patients arrive at a clinic but spend time waiting to get registered or checked in.Wait lines occur in all areas of the hospital—such as patient admissions, financial services, physicians’ lobbies—and are generally considered to be routine and a part of everyday business in health care. This is inaccurate. Understanding wait times is a required step to modeling process and staffing changes to improve service. Wait times are generally one of the most controllable and significant variables driving waste and inefficiency.Wait lines form because people are seeking service faster than they can be served. There are several situations where queues typically form in health care:1. Point of admission (entry).2. Financial services.3. Point of discharge (exit).4. In the front lobby.5. Treatment or exam rooms.6. High-volume departments, such as emergency departments or operating rooms.7. Point-of-use for key clinical technologies (e.g., MRI, computed tomography, position emission tomography).8. In common clinical ancillary services (laboratory, pharmacy, blood bank).9. Elevators, hallways, or other common spaces.10. In the individual physician’s office.11. At supporting services (cafeteria, gift shops, social work).Wait lines can be minimized using advanced quantitative tools. They can be modeled to improve service, align staffing with projected volumes, and control the service levels (or minutes spent in a queue). Wait line simulation models can be built around all aspects of an organization to improve service and process efficiency.There are three key components of wait line simulation models: arrival rate, service rate, and queue structure. The speed at which patients arrive is called the arrival rate. Arrival rate is represented by the Greek letter lambda (l) and is always defined as X per unit of measure (e.g., 12 patients per hour). The speed at which employees can serve them is called the service rate. Service rate is represented in most equations by the Greek letter mu (m). The queue structure is defined by a few subvariables, including number of simultaneous servers or channels, which represent the employees who offer assistance to the guest or patient represented by the symbol (c), and the number of phases in the process (p). Most health care wait lines are considered to be a finite problem. Therefore, finite wait time minimization models can be defined generically as:There is a lot of complexity that can be built around queuing models, but for purposes of this text one primary model is discussed—that of multiple channels (or multiple servers) providing service through a single-phase process. For instance, at a clinic waiting room, there are two employees at the front desk who check in patients, register them, ensure that updated medical insurance is on file, and confirm that all other forms for registration are completed. This is represented in Figure 5–4.In the example in Figure 5–4, there are currently three servers, or channels, who can provide service to the customers. All three of them are on the phone, and only one person is currently providing service to one of the waiting guests. There is a buildup of four customers in the waiting line. There is only one phase, in that the next step after receiving service is to visit the physician. In many processes, however, there are multiple waiting rooms, or phases.FIGURE 5–4 Wait Time Simulation ModelsA key indicator for managing customer service is the number of minutes that a patient has to wait in the queue. This can be modeled using the following equation, where W = wait time, L = the number of customers in the system or queue, and l represents the arrival rate, or the speed at which new patients arrive in the clinics:For example, assume that there are currently 5 people in the system, and they arrive every 2 minutes (or 30 per hour). The average wait time would be 10 minutes, solved as follows:However, in practice, the number of people in the system is a complex calculation and solving for L requires several calculations that are best done in a spreadsheet solution. The formula that follows shows how to solve for L when it is not given as an assumption. In this calculation, L = total number of customers in the system, Po = the probability that no customers are in the system, and all other variables are as defined earlier (Anderson, Sweeney, & Williams, 1997).To calculate Po, the mathematical calculation is also quite complicated:Average number of customers waiting in line is calculated as:Finally, another important calculation is to define how long it takes for a patient to wait in line versus the total time spent in the system (Wq; both receiving service and waiting in the queue). This can be calculated as follows, which basically subtracts the inverse of the service rate from the total waiting time:Wait Time ExampleA patient arrives at the Solder County Hospital emergency department (ED) and finds a waiting line that is currently 60 patients long. Several negative comments are passed on to the front desk employees, which are then communicated to the ED director. When she looks out in the waiting area, she too becomes annoyed with this situation, and she makes up her mind that something must be done to help improve the situation. She decides to engage the hospital’s management engineering department to study the situation and recommend possible solutions. The director wants to comprehensively understand current waiting times and determine if staffing levels are appropriate to meet these stated service levels or analyze what changes might be made.The process has only one phase—patients are registered and then transferred back to a primary treatment or exam room (this is a simplification of course, for illustrative purposes only). The potential population or number of patients is finite and, most important, there are three employees at the front desk to handle all admissions and registration, so it is considered multichannel. The ED director has defined a service-level policy of 45 minutes, suggesting that each patient should have to wait no more than this time prior to being moved to an exam room before being seen by a triage nurse or other provider, although admittedly total cycle time or wait time has never been comprehensively monitored.After careful analysis over a 1-week period, the management engineer assigned to the project conducted several detailed cycle time studies. He discovered that on average during the morning shift there are approximately 50 patients arriving every hour and that the front desk personnel can register a patient in approximately 3.5 minutes, or 17 patients per hour.Using the formulas provided earlier, the probability that there are patients in the system is very high, and the Po (or probability of the waiting queues being completely cleared) is less than one-tenth of 1% (0.004). Therefore L (average number of customers in the system) is around 51, which is similar to the 60 that the ED director found on the day this project began. The total wait in the system is found to be a little more than 1 hour (61 minutes). Because registration time is only 3.5 minutes, the total time spent waiting in line is nearly 58 minutes (i.e., 61 – 3.5). This is significantly higher than the 45-minute service level that the director expected.How can this situation be improved? There are some key options:1. Streamline, or reduce, the number of checks or steps that the front-desk personnel are required to perform to increase throughput and shorten the registration time to fewer than 3.5 minutes. For example, if the process can be shortened by just 5% (to have a service rate of 18 patients per hour, or 3.33 minutes per check in), the total waiting time would fall to just 13 minutes in line!2. Add another employee (additional capacity). Recruiting one more employee (or channel) would cause the total waiting time in the line to fall to just 2 minutes. Of course, the costs of that additional employee must be evaluated relative to the benefits of reducing the queue.Wait Time Decision MakingDepending on the system, it may be necessary to use different optimization algorithms. The algorithms are different for each of the four types of systems:1. Single channel, single phase.2. Single channel, multiple phase.3. Multiple channels, single phase.4. Multiple channels, multiple phases.This text covered only the third type of system. For a more comprehensive discussion of the optimization models for all four systems, consult Introduction to Queuing Theory (Cooper, 1981).Wait times create poor service levels and are bottlenecks for system throughput. As much as possible, and as long as total benefits exceed costs, they should be minimized. In reality, however, there is no such thing as an optimal solution with wait lines. They can be minimized, but the total cost of adding new channels must be carefully weighed against those gains. Similarly, if a bottleneck in registration is eliminated, it may just move that bottleneck to the physician’s or nurse’s treatment rooms. Moving a choke point back one step in the process does not create any system benefits, so it is important that the total system wait times and process be analyzed carefully.TIME AND MOTION STUDIESOne of the best ways to minimize wait time is to increase speed o
