IJISSS Editorial Board Editor in Chief:
John Wang, Montclair State U., USA
Associate Editors:
M. Brian Blake, U. of Notre Dame, USA Schahram Dustdar, Vienna U. of Technology, Austria Juergen Gnoth, U. of Otago, New Zealand Patrick C. K. Hung, U. of Ontario Institute of Technology (UOIT), Canada Nicole Immorlica, Northwestern U., USA Kamal Jain, Microsoft Research Center at Redmond, USA Jan Marco Leimeister, Technische Universitaet Muenchen, Germany Yinsheng Li, Fudan U., China Marcin Matuszewski, Nokia Research Center, Finland Sanjay Mehrotra, Northwestern U., USA Luiz Moutinho, U. of Glasgow - Scotland, UK Ronald L. Rardin, U. of Arkansas, USA Rainer Schmidt, Aalen U., Germany Cathy H. Xia, IBM T. J. Watson Research Center, USA Justin Zhan, Carnegie Mellon U., USA
International Advisory Board: James C. Spohrer, IBM Almaden Research Center, USA IGI Editorial:
Heather A. Probst, Director of Journal Publications Elizabeth C Duke-Wilson, Managing Editor, Production
International Editorial Review Board: Laurence Alpay, Leiden U. Medical Centre, The Netherlands Parthasarathi Banerjee, NISTADS, India Ravi Behara, Florida Atlantic U., USA Joseph Cazier, Appalachain State U., USA Ferhan Cebi, Istanbul Technical U., Turkey Narciso Cerpa, Universidad de Talca, Chile Chun Hung Cheng, The Chinese U. of Hong Kong, China Jagpreet Chhatwal, Merck Research Laboratories, USA Lawrence Cunningham, U. of Colorado at Denver and Health Sciences Center, USA Tugrul U. Daim, Portland State U., USA Tony Diana, Federal Aviation Administration, USA Rong Duan, AT&T Research Lab, USA Hamidreza Eskandari, Red Lambda Inc., USA Larbi Esmahi, Athabasca U., Canada Jose Espiritu, The U. of Texas at El Paso, USA Richard Feinberg, Purdue U., USA Xiuzhen (Susan) Feng, Beijing U. of Technology, China Jan Frick, U. of Stavanger, Norway Lars Friedrich, Commerzbank AG, Germany James H. Gerlach, U. of Colorado at Denver and Health Sciences Center, USA Jorge Gomes, ISCTE - Lisbon U. Institute, Portugal Patrizia Grifoni, CNR-IRPPS, Italy Evangelos Grigoroudis, Technical U. of Crete, Greece Aparna Gupta, Rensselaer Polytechnic Institute, USA Dianne Hall, Auburn U., USA Roslin Hauck, Illinois State U., USA Bill Hefley, Carnegie Mellon U., USA Keith M. Heston, Accenture, USA Ashish Jain, Lehman Brothers, USA Marijn Janssen, Delft U. of Technology, The Netherlands Shantha Jayalal, U. of Kelaniya, Sri Lanka Kristina Jespersen, Aarhus U., Denmark Arun Jotshi, AT&T Inc., USA Oualid Jouini, Ecole Centrale Paris, France Tauno Kekale, U. of Vaasa, Finland Markus Kirchberg, Institute for Infocomm Researc - A*STAR, Singapore SC Lenny Koh, The U. of Sheffield, UK Abdullah Konak, The Pennsylvania State U. Berks, USA Christopher Koroneos, Lhtee/Aritotle U., Greece Joseph Lai, Hong Kong Polytechnic U., Hong Kong Thomas Lee, U. of Illinois, USA Jan Marco Leimeister, Universität Kassel, Germany Yair Levy, Nova Southeastern U., USA Xiaoqing Li, U. of Illinois at Springfield, USA Yang Li, British Telecom Plc., UK Yinsheng Li, Fudan U., China Janet Light, U. of New Brunswick Saint John, Canada Fuhua Lin, Athabasca U., Canada Peng Liu, Penn State U., USA Francisco Javier Lloréns-Montes, U. of Granada, Spain Sathasivam Mathiyalakan, Winston State Salem U., USA K. Jo Min, Iowa State U., USA Aleksandra Mojsilovic, IBM Research, USA
IGIP
Manuel Mora, Autonomous U. of Aguascalientes, Mexico Clive Morley, RMIT U., Australia Alamelu Nachiappan, Pondicherry U., India Karthik Namasivayam, The Pennsylvania State U., USA Juan Luis Nicolau, U. of Alicante, Spain Selmin Nurcan, U. Paris 1 Pantheon Sorbonne, France Liam O'Neill, UNT – HSC, USA Takayuki Osogami, IBM Tokyo Research Laboratory, Japan Bernhard Ostheimer, U. of Giessen, Germany Thomas Owens, Brunel U., UK Teemu Paavola, LifeIT Plc, Finland Mark C. Paulk, Carnegie Mellon U., USA Aysegul S. Peker, North Carolina State U., USA Luca Quadrifoglio, Texas A&M U., USA Reetta Raitoharju, Turku School of Economics, Finland Avnish Rastogi, Providence Health & Services, USA Vadlamani Ravi, Institute for Development & Research in Banking Technology, India Pekka Ruotsalainen, National Research & Development Centre for Welfare and Health, Finland Fjodor Ruzic, Institute for Informatics, Croatia Vamsi Salaka, Ariba Inc., USA Lauri Salmivalli, Turku School of Economics, Finland Chris Schlueter Langdon, U. of Southern California, USA Marc Schniederjans, U. of Nebraska-Lincoln, USA Shital Shah, Rush U. Medical Center, USA Chung-Chi Shen, National Dong-Hwa U., Taiwan Victor (Shengli) Sheng, New York U., USA Soheil Sibdari, U. of Massachusetts, USA Marianna Sigala, U. of the Aegean, Greece Ronggong Song, National Research Council of Canada, Canada Martin Steinert, U. of Fribourg/Switzerland, Switzerland Fay Sudweeks, Murdoch U., Australia Albert Wee Kwan Tan, National U. of Singapore, Singapore Wui-Gee Tan, U. of Southern Queensland, Australia John Thompson, Buffalo State College, USA Wendy van der Valk, Eindhoven U. of Technology, The Netherlands Clemens Van Dinther, FZI Research Center for Information Technologies, Germany Dharmaraj Veeramani, U. of Wisconsin-Madison, USA Harry Jiannan Wang, U. of Delaware, USA Shengyong Wang, State U. of New York at Binghamton, USA Shouhong Wang, U. of Massachusetts Dartmouth, USA Wen-Li Wang, Penn State Erie, USA Yun Wang, U. of Houston - Victoria, USA Mudasser F. Wyne, SOET - National U., USA Wei Xie, American Airlines, USA Dana Yagil, U. of Haifa, Israel Venkata Yanamandram, U. of Wollongong, Australia Jian Yu, The U. of Adelaide, Australia Qi Yu, Rochester Institute of Technology, USA Ruoyi Zhou, IBM Almaden Research Center, USA Yilu Zhou, George Washington U., USA Ling Zhu, Long Island U., USA
IGI Publishing www.igi-global.com
Call for Articles International Journal of Information Systems in the Service Sector An official publication of the Information Resources Management Association! The Editor-in-Chief of the International Journal of Information Systems in the Service Sector (IJISSS) would like to invite you to consider submitting a manuscript for inclusion in this scholarly journal. The following describes the mission, the coverage and the guidelines for submission to IJISSS. MISSION: The International Journal of Information Systems in the Service Sector (IJISSS) provides an outstanding channel for practitioners and researchers (from both public and private areas of the service sector), software developers, and vendors, to contribute and circulate ground-breaking work and shape future directions for research. IJISSS will also assist industrial professionals in applying various advanced information technologies. It details the complimentary relationship between the advancement of the service sector and the evolution of information systems. COVERAGE/MAJOR TOPICS: Recommended topics include, but are not limited to, the following: IJISSS topics include, but are not limited to: • • • • • • • • • • • • •
Business services Creative problem solving Decision making under uncertainty Decision-support systems Forecasting, planning, scheduling, and control Green service and sustainability Logistics network configuration Matching supply with demand Performance measures and quality control Public service management Self-service systems Service business models Service information systems
ISSN 1935-5688 eISSN 1935-5696 Published quarterly
All submissions should be emailed to: John Wang, Editor-in-Chief International Journal of Information Systems in the Service Sector Email:
[email protected]
For Full Submission Guidelines, please turn to the back of this journal or visit the IGI Global website at www.igi-global.com. Ideas for Special Theme Issues may be submitted to the Editor-in-Chief.
Please recommend this publication to your librarian. For a convenient easyto-use library recommendation form, please visit: http://www.igi-global.com/ IJISSS and click on the "Library Recommendation Form" link along the left margin.
International Journal of Information Systems in the Service Sector January-March 2009, Vol. 2, No. 1
Table of Contents
1 11
Research Articles
Predicting Ambulance Diverson Abey Kuruvilla, University of Wisconsin Parkside, USA Suraj M. Alexander, University of Louisville, USA Hybrid Value Creation in the Sports Industry: The Case of a Mobile Sports Companion as IT-Supported Product-Service-Bundle Jan Marco Leimeister, Kassel University, Germany Uta Knebel, Technische Universitaet Muenchen, Germany Helmut Krcmar, Technische Universitaet Muenchen, Germany
26 Connect Time Limits and Performance Measures in a Dial-Up Modem Pool System Paul F. Schikora, Indiana State University, USA Michael R. Godfrey, University of Wisconsin Oshkosh, USA Brian D. Neureuther, State University of New York, College at Plattsburgh, USA 49 ICT Usage by Greek Accountants Efstratios C. Emmanouilidis, University of Macedonia, Greece Anastasios A. Economides, University of Macedonia, Greece 71
Exploring the Adoption of Technology Driven Services in the Healthcare Industry Umit Topacan, Bogazici University, Turkey A. Nuri Basoglu, Bogazici University, Turkey Tugrul U. Daim, Portland State University, USA
International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010 1
Predicting Ambulance Diverson Abey Kuruvilla, University of Wisconsin Parkside, USA Suraj M. Alexander, University of Louisville, USA
ABSTRACT The high utilization level of emergency departments in hospitals across the United States has resulted in the serious and persistent problem of ambulance diversion. This problem is magnified by the cascading effect it has on neighboring hospitals, delays in emergency care, and the potential for patients’ clinical deterioration. We provide a predictive tool that would give advance warning to hospitals of the impending likelihood of diversion. We hope that with a predictive instrument, such as the one described in this article, hospitals can take preventive or mitigating actions. The proposed model, which uses logistic and multinomial regression, is evaluated using real data from the Emergency Management System (EM Systems) and 911 call data from Firstwatch® for the Metropolitan Ambulance Services Trust (MAST) of Kansas City, Missouri. The information in these systems that was significant in predicting diversion includes recent 911 calls, season, day of the week, and time of day. The model illustrates the feasibility of predicting the probability of impending diversion using available information. We strongly recommend that other locations, nationwide and abroad, develop and use similar models for predicting diversion. Keywords:
911 Calls, Ambulance Diversion, Emergency Medical Systems, Logistic Regression
BACKGROUND A majority of Emergency Departments (EDs) across the United States perceive they are at or over capacity (Lewin Group, 2002). As ED visits have been on the rise, the number of hospital EDs and beds available at hospitals has decreased (Nawar, Niska, & Xu, 2007; U.S. General Accounting Office [GAO], 2003) In literature, several authors discuss factors contributing to ED saturation, ranging from high patient acuity and bed shortages (Derlet, Richards, & Kravitz, 2001) to lab delays and nursing shortages (Richards, Navarro, & Derlet, 2000). When EDs reach their capacity, ED staff is unable to promptly care for new arrivals, and services within the hospitals are unable to accommodate the specific needs of new ambulance arrivals; hence ambulances must be diverted to other facilities that can provide critical care. This situation, referred to as “Ambulance Diversion,” not only results in delays in emergency care (Redelmeier, Blair, & Collins, 1994), but could also contribute to patients’ clinical deterioration (Glushak, Delbridge, & Garrison, 1997). We attempt to develop a mathematical model whereby
DOI: 10.4018/jisss.2010093001 Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
2 International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010
hospitals/EMS agencies in a region can use 911 calls and diversion status of hospitals to predict the likelihood of the occurrence of diversion.
LITERATURE REVIEW A study of the literature shows that the rising trend in ambulance diversions started causing concern during the late 1980s (Richardson, Asplin, & Lowe, 2002), resulting in reports, position papers and task forces studying this problem from the early 1990s (Frank, 2001; Vilke, Simmons, Brown, Skogland, & Guss, 2001; Pham, Patel, Millin, Kirsch, & Chanmugam, 2006). However, owing to the elevated utilization level of EDs, ambulance diversion continues to be an issue today and is a common and increasing event that delays emergency medical care (Redelmeier et al., 1994). A wide range of literature exists, discussing the problem and various solutions have been suggested. A U.S. General Accounting Office survey (2003) found that while about two of every three EDs reported going on diversion at some point in fiscal year 2001, a much smaller portion— nearly 1 of every 10 hospitals—was on diversion more than 20 percent of the time. A cohort of twenty-two master’s degree candidates from the University of Virginia (2001) did a detailed study on diversion at Richmond hospitals, and outlined problems and solutions, analyzed via a simulation model. A government study (U.S. House of Representatives, 2001), quoting instances of diversion from the local press in all states, reported that ambulance diversions have impeded access to emergency services in the metropolitan areas of 22 states. Vilke et al., (2001) tested the hypothesis that, if one hospital could avoid ED diversion status, need for bypass could be averted in the neighboring facility. They concluded that reciprocating effects can be decreased with one institution’s commitment to avoid diversion, thus decreasing the need for diversion at a neighboring facility. Neely, Norton, and Young (1994) found that ambulance diversions increase transport times and distances. One community served by four hospitals reduced ambulance diversion during a year, by 34% (Lagoe, Kohlbrenner, Hall, Roizen, Nadle, & Hunt, 2003). This was accomplished by sending daily diversion statistics to hospital chief executive officers and ED directors and managers, along with each hospital individually implementing its own measures to reduce diversion hours. Schull, Mamdani, and Fang (2004) found that there was an increase of diversion hours during the months of November and December and correlated it to the effect of flu on diversion. Only two papers in medical literature referred to 911 calls being used in a transport decision. Anderson, Manoguerra, and Haynes (1998) explored the effect of diverting poison calls to a poison center and Neely et al. (1994) found that diversion of 911 patients correlated strongly with unavailability of specific categories of beds. Several communities have also produced policies to honor patient requests regardless of diversion status and to limit the total time of diversion for each hospital. Some large metropolitan areas have established oversight task forces to study and track the diversion issue in their communities (GAO, 2003). Some communities have addressed this issue with political mandates by non-medical personnel banning the use of diversion (Anderson, 2003). One city’s solution (Lagoe & Jastremski, 1990) was the installation of an ambulance diversion system, whereby ambulances carrying patients with relatively minor injuries were diverted, when necessary, from the city’s busy emergency departments to less crowded ones in neighboring counties. A model that is able to predict the likelihood of a hospital or a combination of hospitals going on ambulance diversion would give advance warning to hospitals and allow them to take proactive steps to prevent it. The region itself would be better served by, for example, rerouting ambulances to alternate medical establishments. This would reduce transportation time and free
Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010 3
up ambulances quicker for the next emergency. A survey of the literature yielded only one model for predicting an impending diversion. This model used the “work score” of an ED to predict ambulance diversion (Epstein & Tian, 2006). However, the model was developed for individual hospitals and does not consider the diversion status of other hospitals in the region; also, the workscore is not an easily available independent variable. The model presented in this article defines the joint probability of the diversion status of a collection of hospitals, based on readily available data, such as 911 calls and current status of hospitals. The objective of this article is to propose a methodology and evaluate its feasibility for predicting the likelihood of ambulance diversion using readily available data, such as 911 calls and the diversion status of hospitals. Its purpose is to encourage metropolitan areas to develop similar models to predict ambulance diversion.
SELECTION AND EVALUATION OF METHODOLOGY Since the purpose of this article is to encourage metropolitan areas to develop similar models to predict the likelihood of ED diversion, we used multinomial logistic regression, a widely accepted statistical methodology that relates independent variables to the likelihood of a dependent event. The model developed uses data such as 911 calls and the diversion status of hospitals in the region, which is readily available from health/EMS agencies. 911 calls have been used for transport decisions (Anderson et al., 1998), and there is a strong association between the number of ambulance patients and diversion (Schull, Lazier, Vermeulan, Mauhenney, & Morrison, 2003). Since the first call made by, or for, a potential ED patient is to “911,” the number of calls for transport to hospitals in the region is an obvious leading indicator of the patient load at an ED. Other variables that could affect the number of patients at an ED include the time of year or season (such as flu season), the day of week, and time of day; hence, these variables are also used as independent variables in the model. The proposed methodology was evaluated through a retrospective examination of real ambulance diversion and emergency 911 call data for the Kansas City metropolitan area. 911 call data was obtained for the Kansas City metropolitan area for the year 2003 and first 6 months of 2004. Twenty-nine of the 36 hospitals in the region were on diversion at some point during that period. The emergency 911 call data for the Kansas City metropolitan area was obtained from the Metropolitan Ambulance Services Trust (MAST), the ambulance service authority for the area. The data included the time a call was received, the type of emergency, and the hospital to which the patient was initially transported. The diversion data was obtained from “EMSystems,” a program that tracks diversion status of hospitals. The system helps emergency departments report their status and thus coordinate and communicate with the other hospitals/EMS agencies in the region. EMSystems includes the time at which a hospital went on diversion, the reasons for going on diversion, and how long it stays on diversion. Three hospitals with the most instances of diversion from the region were selected for analysis. The number of variables used in the model increases exponentially as the number of hospitals increases. Also, preliminary experiments indicated that a logistic model was useful in determining the probability of diversion at each of the hospitals based on the 911 calls, particularly for hospitals where diversion was most frequent. For these reasons, this study focuses on the three hospitals with the highest incidence of diversion. The preliminary logistic model was modified to account for the correlations among the diversion statuses of hospitals within a region, because the state of any one hospital affects the impending state of other hospitals. For instance, consider a region with two hospitals A and B, and Hospital A is similar to Hospital B with respect to any
Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
4 International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010
specialized types of patients it accepts, such as trauma patients. The current state of Hospital A clearly affects the future state of Hospital B because Hospital B has responsibility for a larger patient population when Hospital A is on diversion and vice-versa (Vilke et al., 2001). In order to simultaneously consider the current ED state at each of the hospitals to predict the probability of diversion at any one of the hospitals, a multivariate multinomial model was constructed. While the individual logistic models provided valuable information on their own, the collective valid are coded as 0, 1,…, K - 1; the vector x denotes the p covariates. The logit functions are represented as: æ P (Y = k | x )ö÷ ÷ = x ' b, k = 1..., K - 1. gk (x ) = ln ççç çè P (Y = 0 | x )÷÷ø
where β is a vector of unknown coefficients Thus, the conditional probability that a hospital is in a specific state, that is on diversion or not, is given by the following equations: P (Y = 0 | x ) = P (Y = k | x ) =
1 K -1 g (x ) i
1 + å i =1 e e
gk (x )
K -1 g (x ) i
1 + å i =1 e
, k = 1,..., K - 1.
One of the key assumptions of the multinomial model is that the observations are independent. Thus for each of the models, only one out of every d observed values of the response variable is used, so that no variable representing 911 calls in a period is used twice. If all the observations of the response variable were used, nearly all observed 911 data would be used d times (t-d+1 to period t), in establishing the model, which would make the assumption of independence less plausible. This approach also significantly decreased the computation time and improved the overall fit of the models. The data processing and statistical analysis were accomplished using Microsoft Excel and the statistical packages R and SAS. The data on ambulance diversion and 911 calls is partitioned into bins of length 1 hour. For each bin, the response variable gives the conditional probability of the future state of the three hospitals. There are eight possible outcomes that could occur in the next period which can be denoted by O, A, B, C, AB, AC, BC, and ABC. These are coded as k = 0, 1, …, 7, respectively. The letters denote which hospitals are on diversion. For example, AC represents the state where hospitals A and C are on diversion, but hospital B is not. The state O represents the situation where none of the hospitals are on diversion. There are several explanatory variables that are represented in the vector x. This includes the state of diversion during the current hour, which is a categorical variable with the same eight possible states as the response variable. Also, included is another categorical variable representing the period of the day, which is defined by partitioning the day into early morning (midnight–6 a.m.), morning (6 a.m.–noon), afternoon (noon–6 p.m.), and evening (6 p.m.–midnight). Additionally, indicator variables enable distinguishing between different years, weekdays and weekends, and seasons, such as flu season (which we have specified as being the months of November and December). In addition, three variables ct, ct-1, and ct-2 that represent the number of 911 calls
Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010 5
during the current hour, the previous hour, and the hour before the previous hour, are included. The explanatory variables included in the model are shown in Table 1. The multinomial logistic regression model is defined using the explanatory variables and the historical data on the state of the hospital EDs. The statistical significance of the variables is shown in the analysis of effects table, Table 2. As indicated in the table, only the variable for weekend is not statistically significant at a 5% level; however, it is a natural variable to include in the model. The Deviance and Pearson goodness-of-fit statistics also give no evidence against the model. The odds ratios are shown in Tables 3, 4, and 5.
DISCUSSION The results indicate that the most recent 3 hours of the overall number of 911 calls are significant in predicting whether hospitals will go on diversion. Beyond that, the calls do not seem to be significant in predicting diversion. Also, other factors like the current state of diversion, indicator variables for flu, and quarter of day and the year were highly significant. The logistic model that has been developed illustrates the feasibility of predicting the probability of diversion. Since a number of EMS agencies are currently using online technologies with the data necessary for prediction readily available, this type of analysis might be timely. An early warning of impending diversion provided by this model could enable hospitals and EMS agencies to take appropriate Table 1. Codes for the data set Variable Name
Description
Codes
Current
The current state of diversion.
O = no hospitals on diversion A = only A is on diversion B = only B is on diversion C = only C is on diversion AB = A and B are on diversion AC = A and C are on diversion BC = B and C are on diversion ABC = all hospitals are on diversion
Period
The period of the day.
0 = early morning (midnight–6am) 1 = morning (6am–noon) 2 = afternoon (noon–6pm) 3 = evening (6pm–midnight)
Weekend
An indicator variable for the weekend.
0 = weekday 1 = weekend
Year
An indicator variable for the year 2003.
0 = year 2004 1 = year 2003
Flu
An indicator variable for the flu season.
0=not flu season 1=flu season
The number of 911 calls during the current hour.
Non-negative integer
The number of 911 calls during the previous hour.
Non-negative integer
The number of 911 calls during the hour before the previous hour.
Non-negative integer
Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
6 International Journal of Information Systems in the Service Sector, 2(1), 1-10, January-March 2010
Table 2. Effect table for factors considered Effect
df
Chi-Square
P-Value
CURRENT
49
13590.9257
