TMCnet News
ASSIGNMENT: Inmate classifcation [Industrial Engineer](Industrial Engineer Via Acquire Media NewsEdge) Decision support tool gives help to Pennsylvania Department of Corrections A major responsibility of any department of corrections (DOC) is assigning and transferring inmates to facilities based on security considerations as well as medical, mental health, behav- ioral and program needs. Assignment decisions are complicated because, while inmates have diverse requirements, state institutions have a heterogeneous nature. These planning and assignment logistics can be improved immensely by applying industrial engineering, as the Pennsylva- nia Department of Corrections found out when a decision support tool was devel- oped to assist staff members in the process of assigning inmates to various locations. The interactive platform integrates analyti- cal models and traditional database access with the business rules the DOC must follow. The Pennsylvania Department of Corrections is responsible for overseeing the operation of 25 state correctional institutions, a motivational boot camp, halfway houses and private vendors that provide specialized treatment and super- vision services. A major responsibility of the correctional staff is its classification and assignment of inmates to specific institutions. An initial assignment is made when an inmate enters the system. Changing inmate profiles or institutional loading can trigger various reassignment decisions. Challenge The initial assignment focused on the current stage. Assignment decisions are complex as there are many inmate-specific factors and infrastructure-specific factors that need to be considered for each case. Inmate factors belong to several needs categories, such as medical, psychological, behavior level, sentence condition, custody level and educational. Infrastructure factors primarily involve the nonuniform nature of the correctional institutions with respect to their physical facilities and specific treatment programs they can support, the distribution of the population, waiting lists for special units and bed capacity. Because of capacity constraints or because of a unique set of inmate factors, it is common for none of the institutions to satisfy all the needs and requirements for a specific inmate. In such cases, one or more of the inmate requirements must be relaxed. The choice as to which require- ment (or requirements) to relax is part of the assignment decision. For example, if an inmate is required to attend multiple programs but none of the available institu- tions can provide them all, corrections staff must make a judgment about program priorities and make the initial assignment accordingly. The inmate would be reas- signed at a later date as appropriate. Not all requirements can be relaxed. For example, a hearing-impaired inmate can defer attend- ing some recommended programs, but this inmate will need a hearing aid all the time. Historically, the inmate assignment process has been largely manual, where a staff member assembles inmate and infrastructure information from DOC databases and then evaluates inmate requirements and current infrastructure realities in the context of DOC busi- ness rules. While general guidelines for inmate assignments are known, the nearly uncountable combinations of factors and the qualitative nature of many of them make the efficiency and optimality of the assignment decisions greatly dependent on experience and subjective judgment of the staff. Decision support system It is well-known that decision processes with a large subjective component are more vulnerable to bias and have more variability than processes rooted in objec- tive facts. It is also generally accepted that a decision support system, when created with the needs of decision-makers in mind, can reduce bias and improve adherence to guidelines. Decision support systems have been shown to be particularly effec- tive in scenarios where, while judgment is involved, there is enough structure for analytical aids to be of value. In an effort to improve the efficiency of the inmate assignment decision-making process and to optimize institutional assignments, the Pennsylvania Depart- ment of Corrections launched the Inmate Institutional Assignment Project, which developed a decision support tool for the process of assigning inmates. The tool emphasizes ease of use by noncomputer experts in an interactive environment while being flexible enough to accommodate changes in system infrastructure and the decision-making approach of the user. The tool provides the staff member making an inmate assignment with a ranked order of institutions from which the member chooses the assignment. Thus the tool eliminates much of the tedium of evalu- ating many combinations, freeing staff members to use their subjective judgment to differentiate from among a small subset of facilities. The process of decision analysis is based on the decision tree technique, which is commonly used in operations research. A decision tree is a flow chart type model or structure that helps iden- tify a strategy most likely to reach a goal. Figure 1 illustrates this particular decision support system. There are three major types of nodes in a decision tree. A chance node or judge node, which is denoted as a diamond, shows different chances that correspond to certain conditions. An activity node, which is denoted as a rectangular box, presents the current decision pool. An end node, which is denoted as a rectangular box with rounded corners, indicates the final deci- sion. Using a decision tree to describe the decision process makes it transparent and easy to understand. Corrections officials deemed it impor- tant that the tool operate as close as possible to how the staff has managed assignments manually. To identify the factors and rules involved in inmate assign- ment, extensive interviews were done with staff members from the Diagnostic and Classification Center, Office of Popula- tion Management, Bureau of Treatment Services, Bureau of Health Care Services and IT departments. These interviews helped identify the factors and require- ments that the tool design team needed to consider for an inmate assignment. They included the categories of medical needs, psychological needs, program require- ments, behavior level, sentence condition, custody level, educational requirements, transportation, population distribution, waiting list for special units and capacity of programs. On the order of 60 factors were identified as being important in the assignment decision. These factors and rules were used to develop a decision tree that implements the basic logic DOC staff members used when making inmate assignments. The tool uses the logic of the decision tree to evaluate and subsequently rank each institution with respect to the inmate being assigned. The institution evaluation process assigns a weight for each factor based on the institutionÕs ability to satisfy the requirements for that factor. Ranking is based on the summation of the weights for each institution. Factor weights are critical for the rank- ing system and decisive for the final assignment recommendations. The proce- dure of institutional assignment involves staff from various departments. Differ- ent backgrounds and duties lead to the fact that they have a particular emphasis on different factors when they consider assignments. This kind of consideration depends on personal experience and pref- erence. In order to get comprehensive and unprejudiced weighting data, surveys were conducted with relevant staff members. The survey requested that each staff member rate each of the previously deter- mined factors on a scale of 1 to 100 in terms of their importance. The baseline/ default factor weights incorporated into the tool were the weighted average values depending on the role of the respective staff members. Implementation and features The major merit of this decision support tool is twofold: It has a friendly user interface and can be operated easily. All important information for institutions and inmates is integrated and transparently displayed on the main inmate summary page shown in Figure 2. At the same time, most interactions with the tool take place on this page. All the possible interactions are indicated by buttons that can be viewed and clicked easily and quickly. When the tool is started, the deci- sion-maker chooses the relevant inmate assignment database by either enter- ing the database path or by pressing the Òchoose databaseÓ button, which initiates a standard file selection dialog. The inmate assignment database contains all inmate and infrastructure information. The data- base is created and updated periodically by the IT department via separate online processes. Once the database is entered, the tool queries and retains information on the institutions. The available bed information is displayed on the main page, with negative numbers indicating over- crowded facilities and programs. To begin the assignment process, the inmate number is entered. The tool then pulls inmate information from the data- base and displays it in the appropriate sections. The decision-maker is able to edit selections in the special information panels of "separations," "medical limits" and "mental health." Other sections are read-only. At any time, the decision-maker can click the "Calculate Scores" button to obtain a ranking score (determined by the tool) for each of the correctional institu- tions. The scores and associated institution name abbreviations are shown in ascend- ing order in the assignment panel. The letters displayed after the institution name are codes indicating major defi- ciencies at the institution based on the current inmate requirements. For exam- ple, code A implies that the inmate does not meet the age limits for the institution, while code C implies that the institution does not support the recommended custody level for the inmate. By making changes to the editable fields, recalculat- ing and observing the ranked scores, staff members can evaluate "what-if" scenar- ios easily. To initiate the inmate assignment dialog, the decision-maker clicks the "Assign Inmate to Facility" button. This opens the assignment dialog shown in Figure 3. The dialog provides a more focused display of the information from the "assignment panel" on the main panel and also includes the bed capacity infor- mation. In particular, a longer version of which conditions each facility doesn't satisfy is displayed. After review, the appro- priate facility radio button is chosen and the "Assign Inmate to Facility" button is clicked to assign the inmate. The local database will be updated to reflect the assignment, and the dialog closes. The assignment dialog may be aborted at any time without affecting the database by clicking the "ABORT Assign- ment" button. The dialog also is used to change any inmate assignments. The other great strength of this tool is its flexibility. In order to accommodate changes in system infrastructure and the decision-making approach of the user, the tool provides three dialog boxes. One, shown in Figure 4, allows the user to modify the weights for inmate factors and programs. The user can change the factors temporarily for the current session or save the changes as new system defaults. The inmate factors dialog (Figure 5) and the programs dialog (Figure 6) are similar. The appropriate institution is selected via the "Choose a Facility" box. Once selected, the factors/programs can be added or removed as appropriate. Future work This decision support tool can assist the Pennsylvania Department of Correc- tions in the inmate assignment process and improve the efficiency of this deci- sion-making procedure. However, the current inmate assignment process still leaves a lot of room for improvement. The present process assigns inmates one after another, even if a batch of inmates is waiting to be assigned. Under this sequential mode, the decision for each inmate only focuses on a single inmate without considering the whole institution system from a global view. One drawback of this kind of sequen- tial assignment is that the sequence of inmates is critical and could affect the succeeding assignments. It is well-known that for a system, the summation of local optimality does not guarantee a global optimality. Therefore, reaching a global optimal- ity is the target for the next stage, and applying a multiple-objective integer optimization model is the key. This model will treat all the inmates and facilities from a system perspec- tive. It could consider a group of inmates at once to eliminate the influ- ence of processing sequence. It also could deal with multiple objectives, such as providing the best assignment recommendation for each inmate, reducing the transportation cost, length of the waiting list and a host of other factors. The database used by the current version of this decision support tool needs preprocessing. It is required to contain all inmate and infrastructure information. Thus, the database is created and updated periodically by the IT department via separate online processes. To make this tool more effective and efficient, reducing these redundant data transformations and integrating with the original DOC system is necessary. The expectation is that the tool could pull all essential data from the current database and pass the results to the existing assign- ment system automatically by specific interfaces. d supporting lower cholesterol Decision support tools could help pave the way toward more personalized medicine by leveraging big data, reported the Richmond Times-Dispatch. AlgorithmRx LLC in Chesterfeld County, Virginia, is testing an algorithm that aims to give physicians a starting point when choosing drugs and dosages to combat high levels of LDL cholesterol, something suffered by more than 70 million Americans in the United States. Doctors can pick various doses from seven commercially available statins, but they don't have a point-of-care decision support tool to determine where to start, according to company offcials. This creates a trial-and- error process that leads to more trips to the doctor, more tests and more healthcare expenses. The company's algorithm, aRx Statin Advisor, uses 40 variables drawn from a patient's electronic medical record, along with population data, statistics on similar patients who have taken statins and the results of those treatments, according to the newspaper. "We are leveraging big data to develop mathematical models that help physicians make better decisions and personalize the medication they are prescribing," said Lenn Murrelle, an epidemiologist and a managing partner for AlgorithmRx. Murrelle said the technology could be available commercially in 12 to 18 months. Dan Li is a doctoral candidate in the Depart- ment of Industrial and Systems Engineering at Lehigh University. Her main research areas include linear optimization, complexity theory and parallel computing. She has led various industry projects in the areas of mathemati- cal modeling in healthcare systems, revenue analysis and management, decision-making, air traffic control, and system development and improvement. She received her bachelor's degree in control science and master's degree in enterprise information and systems engi- neering from Tsinghua University. She is a student member of SIAM and INFORMS. Louis J. Plebani is an associate professor in the industrial and systems engineering depart- ment at Lehigh University, where he earned his Ph.D. in industrial engineering. His research interests include computational operations research and automation and process control. He has been the principal investigator on a number of research contracts for both federal and local governments and for numerous industrial firms. He is a registered professional engineer in Pennsylvania. Tamás Terlaky is the George N. and Soteria Kledaras '87 Endowed Chair Professor and chair of the industrial and systems engi- neering department at Lehigh University. His research area is optimization methods and applications. He has published several books, authored more than 175 research papers, is founding editor in chief of the journal Optimization and Engineering and a member of several editorial boards. He is a senior member of IIE and a fellow of the Fields Institute. Terlaky previously was a professor at McMaster University, Delft University of Technology and Eötvös University, which is where he earned his M.S. and Ph.D. degrees. George R . Wilson is an associate professor and associate chairman of the industrial and systems engineering department at Lehigh University. He received his B.S. and M.S. in industrial engineering and a Ph.D. in indus- trial engineering and operations research from the Pennsylvania State University. His research interests include logistics, service supply chains, resource allocation and model- ing and analysis of large-scale systems. Kristofer "Bret" Bucklen is the director of planning, research and statistics for the Penn- sylvania Department of Corrections. Bucklen received his M.S. in public policy and manage- ment from Carnegie Mellon University's Heinz School of Public Policy. He is currently finishing a Ph.D. in criminology and criminal justice at the University of Maryland. (c) 2014 Institute of Industrial Engineers-Publisher |