Health Systems (2014), 1–7 © 2014 Operational Research Society Ltd. All rights reserved 2047-6965/14 www.palgrave-journals.com/hs/
ORIGINAL ARTICLE
A data mining approach for estimating patient demand for mental health services Stephan Kudyba1 and Thad Perry2 1 New Jersey Institute of Technology, Newark, U.S.A.; 2Tennessee Tech University, Cookeville, Tennessee, U.S.A.
Correspondence: Stephan Kudyba, New Jersey Institute of Technology, University Heights, Newark, NJ 07102, U.S.A.
Abstract The ability to better estimate future demand for health services is a critical element to maintaining a stable quality of care. With greater knowledge of how particular events can impact demand, health-care service providers can better allocate available resources to more effectively treat patients’ needs. The incorporation of data mining analytics can leverage available data to identify recurring patterns among relevant variables, and these patterns provide actionable information to corresponding decision markers at health-care organizations. The demand for mental health services can be subject to variation from time of year (seasonality) and economic factors. This study illustrates the effectiveness of data mining analytics in identifying seasonality and economic factors as measured by time that affect patient demand for mental health services. It incorporates a neural network analytic method that is applied to patient demand data at a U.S. medical center. The results indicate that day of week, month of year, and a yearly trend significantly impact the demand for patient services. Health Systems advance online publication, 11 July 2014; doi:10.1057/hs.2014.12 Keywords: data mining; seasonal demand; neural networks; mental health demand; decision support systems
An introduction to identifying information through data mining
Received: 4 June 2013 Revised: 8 January 2014 Accepted: 23 May 2014
There is rising recognition of the value of information that exists in data resources in organizations across industry sectors. Trends, relationships among variables, and recurring patterns all may exist in data bases and provide insightful descriptions of various processes and enhance the ability to forecast and generate quantitative models that facilitate decision support for practitioners. An essential element to identifying patterns and relationships and generating models that facilitate simulations are multivariate techniques. A prominent field in the multivariate arena involves data mining methods that incorporate mathematical functions and algorithms that process data resources in order to extract actionable information for decision makers. This process of information extraction through data mining is often referred to as knowledge discovery (Fayyad et al, 1996; Kudyba, 2010) or, in other words, the identification of valuable information that enhances knowledge for those who make decisions. The notion of leveraging data resources with mining methods to enhance decision making through knowledge discovery is becoming a critical component of organizational efficiency given the evolving era of big and new data resources. Data resources are growing every year given the introduction of new technologies across industry sectors. The health-care industry is experiencing a significant increase in data resources owing to the ongoing evolution of the digital age. The creation of
2
A data mining approach for estimating patient demand
electronic medical records, the process of e-prescribing, medical devices that automatically download patient physiological elements, and the increased utilization of information systems at the private practitioner and hospital and health systems level are facilitating the creation of vast resources that can provide information to increase efficiency in a number of applications. Data mining analytics are being applied in the health-care industry across a variety of areas. Some of these include the analysis of workflow activities of large health-care provider organizations that include studies that investigate the drivers of patient length of stay, patient demand and bottlenecks in emergency room throughput, and patient satisfaction rates. Other areas involve risk stratification applications or the better identification of patient populations at risk of developing chronic illnesses. Finally, semantic mining applications are being applied to electronic health records to better understand treatment and outcomes and patient diagnosis. One particular data mining technique involves the use of neural networks, which is an approach that incorporates algorithms that processes historical data to identify both linear and non-linear patterns. The resulting models can then be used to conduct ‘what if’ simulations on out of sample data, new data, and forecast data. Neural networks have been utilized in an array of industry applications that include the forecasting of bank failures, traffic patterns, and even rainfall (Tam & Kiang, 1992; French et al, 1992; Yasdi, 1999). Neural networks and multivariate techniques are also being incorporated in the health-care industry to aid in knowledge discovery in a number of applications that include treatment effectiveness (Lisboa et al, 2008) and general operations of health-care organizations (Batal et al, 2001) and patient throughput in emergency rooms (cite).
Advanced analytics and mental health services A major factor in achieving increased efficiency in providing services in any industry is the ability to better understand what drives demand for these services. With this information, decision makers can more accurately apply the optimal number of resources that are required to meet varying amounts of demand. Predictive analytic methods can improve the accuracy of estimating patient demand for organizations that provide mental/behavioral health services. Through analyzing patient services data of a psychiatry and behavioral health center of a major U.S. medical center with multivariate analytic methods, it was determined that seasonality factors along with general macroeconomic trends had noteworthy effects on the demand for mental health services. With this information, decision makers can more accurately apply available resources to meet patient demand, and better manage costs while providing a more consistent service. In seeking to increase efficiencies, organizations identify a process or functional area that can be improved regarding resource allocations that perform some type of task.
Health Systems
Stephan Kudyba and Thad Perry
Decision makers study the situation at hand and adjust employee pools, technological infrastructure, and complementary operations.
Demand for mental health services These efficiency ventures can take on a different level of intensity, however, when firms that rely on external funding (e.g. government programs) face shortfalls in the form of budget-tightening policies (Nauert, 2010). Many mental health-care facilities rely on governmental funding to hire staff, operate clinics, and provide a host of services for a multitude of ailments. Given the deteriorating fiscal situation in the United States, these health-care providers are facing funding cuts to operate facilities. To make matters more difficult, the demand for mental health services has increased dramatically over the past few years. Factors such as increased demand from returning military from foreign endeavors and the recent deterioration of the macro economy (e.g. increased unemployment and stagnant wage growth) along with an aging population have produced dramatic increases in the number of individuals who suffer from mental health-related ailments (Honberg et al, 2011; DiGilio & American Psychological Association, 2012). In order to maintain adequate care in a situation of tightening budgets and increased demand for services, these organizations must enhance their allocation of available resources. In the case of health-care services, providers must better equate existing clinical staff with patient demand. Advanced analytics can be used to leverage data resources to extract information regarding the demand for health services, where predictive models can be generated to provide decision support in estimating future demand and thus enable providers to better allocate resources to meet patient needs (Kudyba, 2012).
Data mining and predictive modeling in patientcentered decision support The Patient Protection and Affordable Care Act (PPACA) changes the way health-care providers use data. New delivery models such as patient-centered medical homes and accountable care organizations require providers to understand their patients’ needs in a more efficient and effective manner. Challenged with optimizing health-care resources, ensuring the right services are delivered at the right time, and increasing overall quality of care demand the thoughtful use of health-care data. Enhancing medical decision support activities, improving diagnosis and categorization of chronic conditions, and accurately predicting hospital and emergency department utilization are essential components of health-care delivery under PPACA. Health-care informatics techniques are critical in understanding and supporting these health-care delivery components. Data mining and predictive modeling techniques are central to this because of substantial improvements in information technology as well as data collection and aggregation of disparate data sources.
A data mining approach for estimating patient demand
Medical diagnostic decision support (MDDS) systems have been used for decades (Miller, 1994). These systems have been developed because it is acknowledged that health-care providers are often asked to make critical clinical judgments based on imprecise and/or incomplete patient information. Inadequate information leads to mistakes, which can dramatically affect quality of care. For example, treatment for patients suffering from clinical depression is complicated, often exacerbated by the presence of comorbid conditions, social support challenges, and poor medication adherence. Recognition of this has led to web-based MDDS systems for depression care management (Fortney et al, 2010). This type of electronic decision support system ensures the appropriate implementation of evidence-based chronic care models. Data mining methods have been used to identify the socio-demographic, physical, and psychological factors most important to the early detection and treatment of serious health-care conditions. Penny & Smith (2012) explored data mining techniques to improve the quality of life of patients suffering from irritable bowel syndrome (IBS). This longitudinal cohort study examined logistic regression, classification, and neural network models. These models demonstrated that IBS severity, psychological morbidity, marital status, and employment status significantly influenced a patient’s health-related quality of life. These results provide the best information to afford better assessment and management of patients with IBS. Other studies have examined data mining methods to improve the accuracy of diagnostic systems based on information derived from multiple, disparate data sources (Sridhar, 2013), as well as recognition of the uniqueness of health-care data mining methods and techniques (Cios & More, 2002). In addition, with advances in information technology, it is now possible to combine data from electronic medical records with human knowledge (i.e., expert information) to optimize the accuracy of diagnostic systems. Prediction of the onset of liver cancer (Kuo et al, 2012), classification of malignant colorectal tumors and abnormal livers (Gao et al, 2012; Acharya et al, 2012), and prediction of mortality of patients with cardiovascular disease (Austin et al, 2012) are now commonplace. Other important applications of health-care data mining and predictive modeling techniques are resource allocation and demand management in the emergency department and hospital setting. Sun et al (2011) developed predictive models to determine the likelihood of a hospital admission based on information collected at the point of emergency department triage. Examining 2 years of hospital data collected by nurses from emergency department patients at the point of triage, regression models were developed to determine the strongest factors in accurately predicting a patient’s immediate inpatient admission from the emergency department. Outside of the obvious admission criteria (e.g., heart attack, lifethreatening trauma), it is not always clear that a patient will be admitted at the point of emergency department triage for conditions such as respiratory infections,
Stephan Kudyba and Thad Perry
3
pleurisy, or orthopedic concerns. The results from this study demonstrated that age, patient acuity category, and emergency department arrival mode were the strongest predictors for hospitalization. These predictive models, if used at the point of triage, could be used for early admission planning and resource challenges faced by inpatient and acute care facilities. Similarly, there have been several studies that demonstrate the effectiveness of data mining techniques in forecasting hospital admissions, returns to the emergency department, demand for specific illnesses, same-day admissions, and emergency department demand (Peck et al, 2012; LaMantia et al, 2010; Kudyba, 2012; Jones et al, 2009).
Seasonality and estimating the demand for mental health The concept of identifying repetitive or cyclical trends in time for demand of particular processes is often referred to as identifying seasonal patterns of demand. Traditionally, organizations apply analytics to determine two main sources of information regarding seasonality: whether demand for their products or services has seasonal patterns (e.g. do their sales increase or decrease according to a particular point in time on a repetitive basis) and, if seasonality exists, what is the magnitude of the change in demand according to a particular point in time. Prior research has shown that seasonality effects for patient demand for mental healthcare exist. For example, the ‘winter effect’ has been cited as being associated with increases in depression-related ailments (Lurie et al, 2006; Fullerton & Crawford, 1999). Studies have also concluded that other factors such as general economic distress (e.g. unemployment, financial stress) drive demand for mental health services (Catalano, 1991; Dooley et al, 2000). Analytic techniques have been used to model the affects of seasonality on patient demand for health services in order to better forecast future demand and allocate resources accordingly. Existing research incorporating calendar-based data has concluded that seasonality provides valuable decision support for the estimation of patient demand for urgent care clinics and emergency room facilities. Step-wise linear regression was applied to daily patient volume, which was matched with calendar data (e.g. day of week and month of year) and weather data to forecast the number of patients seeking urgent care (Batal et al, 2001). The results indicated that regression models incorporating calendar data were useful in estimating future patient demand while weather data only provided marginal improvement to the analysis. Time series methods, linear regression, and neural network methods incorporating daily patient visits and calendar data have been utilized to study patient visits to emergency room facilities (Holleman et al, 1996; Jones et al, 2008). This application is seen as particularly useful in helping alleviate overcrowding and enhance staffing and patient throughput by providing predictive information of patient demand (Wargon et al, 2009).
Health Systems
4
A data mining approach for estimating patient demand
The aim of this work is to leverage and extend existing research to address another area of healthcare that is experiencing increased patient demand (e.g. mental healthcare). The research in this study incorporates calendar data on patient visits to a mental health-care facility and extends the analysis by introducing a yearly time effect to daily and monthly components of demand utilizing a neural network approach. The ‘year’ variable may provide unique explanatory value to this sector of care by introducing longer-term events (e.g. economic distress factors) that may impact the demand for mental health services.
Mental and behavioral health center background Data was gathered from a department of psychiatry/behavioral health center at a major U.S. medical center that provides various programs to treat health issues in the community. One particular program targets consumers who have histories of substantial transition failures that result in higher recidivism, adherence issues, and shorter tenure in the community, especially individuals with cooccurring disorders as defined by the Department of Human Services (e.g. individuals who have at least one mental disorder as well as alcohol or drug use disorder). Illnesses can involve various types of dementia, psychotic disorders related to drug and alcohol addiction, and anxiety and mood disorders. The program provides services that consist of a set of counseling interventions provided by trained clinicians to clients living in the community. In order for the health-care provider to better meet the needs of existing patients, advanced analytics can be utilized to better understand patient demand on a calendar basis. Data resources that described this process were gathered and analyzed with neural network multivariate analytics to identify the effects of particular days of a week and month of a year on demand for services. Data that also addressed longer-term effects on demand was also incorporated to determine whether macroeconomic indicators played a role in driving the need for services. This entire model would provide a robust scenario of both long- and short-term patterns in patient demand.
Stephan Kudyba and Thad Perry
Table 1
Multivariate inputs
Dependent variable
Total minutes of clinician time with patients (per day) (Patient Demand (Individual patient session duration was for Services) aggregated to sum total patient demand per day) Independent (Time variables) Day of week (Monday through Friday/Categorical) Month of year (January through December/Categorical) Year (2008, 2009, 2010/Categorical) n = 740 DF = 574
Table 2
Descriptive statistics of independent variables Mean Maximum Minimum
Standard Deviation
Number of Observations
Day Monday Tuesday Wednesday Thursday Friday
205 154 135 206 215
608 784 368 1376 570
15 15 15 15 15
107 110 66 211 121
147 151 149 146 147
Month January February March April May June July August September October November December
202 229 203 170 164 157 176 182 240 151 173 163
608 1280 570 416 400 420 570 832 1376 528 896 432
15 15 15 15 15 15 15 15 15 15 15 15
128 194 114 87 101 106 114 144 210 107 142 88
62 61 62 60 61 61 60 62 63 62 61 65
Year 2008 2009 2010
214 212 141
570 1376 400
15 15 15
106 189 74
253 232 255
Data and analytic methodology Data resources that recorded patient demand over the period in question involved total minutes of clinician time of providing care to patients with mental health disorders each day of the week for each month of a given year. Health-care clinics were available to patients 5 days a week, beginning on Monday and ending on Friday. Three categorical independent variables were incorporated that measured different time aspects to analyze their impacts on patient demand for health-care services. The first variable included the yearly measure of services rendered to patients. The period analyzed is noteworthy given the economic fallout that transpired in the U.S. economy. The period 2008 through 2009 included an increase in unemployment to 9.3%, doubling that seen in 2007.
Health Systems
Housing prices declined by roughly 18% in 2008 and the major U.S. stock indices declined in value by roughly 35%. The remaining two independent variable categories were month of year and day of week. The dependent variable was defined as the total minutes spent with patients on a given transaction day for each day over the 3-year period. Descriptions of corresponding variables are provided in Tables 1 and 2.
Neural network analysis and results The neural network methodology in this case refers to the utilization of complex computer algorithms that identify existing patterns and relationships within historical data.
A data mining approach for estimating patient demand
The neural network architecture used in this analysis incorporates a multilayered perceptron (Rumelhart, 1986) with a feedforward backpropagation testing function (Hinton, 1992). The neural network modeling process begins with an input layer that includes nodes that correspond to each independent (driver) variable. The driver variables in this case include days of the week, months of a year, and the years 2008 through 2010. Driver variables are assigned weights by the algorithm, where the weighted sum of these inputs is passed into a squashing function in the hidden layer where non-linear calculations are performed on the variables relative to the dependent variable. The combined results in the input and hidden layers are passed to an output layer and compared with the historical dependent variable. Weights for variables are estimated by the backpropagation training method. The backpropagation process simply attempts to reduce the error from the output generated in the neural net architecture for a given row of data and the value of the target/dependent variable in the training set (historical data). This process is conducted for each row of data in the training set and an epoch is defined as when the entire data set has been processed. The action involving the initial assignment of weights to driver variables as they pass through the input node and then the weighted sum of inputs passed to the squashing function and values calculated in the output layer is the feedforward action. The backpropagation process estimates the error of the model output relative to the historical output for each row. This error is used by the algorithm to adjust the variable weights in the attempt to reduce the model output error during the next iteration of data processing. The process is repeated on the training set until the error is no longer reduced. The final model is a set of code that involves a weighting scheme for independent/driver variables. Neural networks can be compared with regression analysis, with a major differentiator being that the n-net approach is based in algorithmic processing that incorporates a dynamic weighting mechanism. These advanced analytic methods go beyond mere identification of retrospective capabilities of basic reporting and provide decision markers with quantitative models that describe relationships between variables underpinning processes. These models provide simulation capabilities to project potential outcomes given adjustments to process variable inputs. More simply put, analytic methods such as neural networks (algorithms in the data mining spectrum) process historical data and determine whether there are reliable consistencies in the frequency and magnitude of occurrences in that data (Chu & Zhang, 2003). In this case do Mondays or Tuesdays of every week or particular months over the 3 years entail significant trends/patterns regarding demand for patient services? The output for the multivariate approach could yield actionable information for hospital staffing departments. The results could identify whether a particular day of the week consistently experiences above/below average demand, and would also provide an estimation of the
5
Stephan Kudyba and Thad Perry
Table 3
Impact statement for independent variable significance
Variable
Impact
Year Month Day
4.83* 2.76* 1.06**
*Significant at 95%; **Significant at 90%.
detailed level of the demand (e.g. number of minutes of patient demand). With this information, health-care staffing operations can better maintain adequate clinicians on a daily basis with greater accuracy to facilitate consistent care for patients. In the case at hand, significant seasonal patterns were identified. The analytic results illustrated that the independent variables (year, month of year, and day of week) all registered statistically significant effects on the demand for mental health services. Table 3 provides impact metrics, which are equivalent to the more traditional F-Statistic for categorical variables in a least squared approach to illustrate the significance of the independent variables (calendar data) in affecting the dependent variable of patient demand for services. The interpretation of the variable relationships in the demand for mental health services follows. The independent variable of year illustrates that early 2008, which featured the greatest economic distress, yields the greatest demand for patient services. The trend for the entire period (2008–2010) depicts a declining effect for mental health services as the U.S. economy began to stabilize (see Figure 1). This supports results in early studies that linked economic distress to increased demand for mental health services. The month of year variable also supports findings in previous studies regarding a ‘winter effect’ for such ailments as depression. The months from January through April depict increases in patient demand, with February illustrating the highest demand (roughly 30% above nonwinter months, except for September). The month of September also depicted increased demand for services, matching that experienced in February (see Figure 2). The end of summer effect and preparation for return to work or depression from lack of work were possible explanations. Finally, day of week also provided noteworthy results and depicts a surge in patient activity on Mondays, which illustrates an approximate 20% increase over other days of the week (see Figure 3). The reasoning for this finding may be largely attributed to the fact that clinics in this case were closed on Saturdays and Sundays and Monday depicted pent-up demand.
Closing comments on analytic models for knowledge generation and decision support This work extends the scope of prior research by incorporating an event element in the form of macroeconomic
Health Systems
6
A data mining approach for estimating patient demand
Stephan Kudyba and Thad Perry
Yearly Effect on Patient Demand
Day of Week Effect on Patient Demand
210
240
190
230
180
Minutes of Service
Minutes of Service
200
170 160 150 140 130
220
210
200
120 2008
2009
2010
190
Year
Figure 1
Independent variable year relationship on demand.
180 Monday
Tuesday
Wednesday
Thursday
Friday
Day of Week
Figure 3 Independent variable day of week relationship on demand for services.
Monthly Effect on Patient Demand 220 210
Minutes of Service
200 190 180 170 160 150 140 130 Jan
Feb Mrch April May June July Aug Sep
Oct
Nov Dec
Months
Figure 2 Independent variable month relationship on demand for services.
factors into a seasonality approach to better estimate future demand for mental health services. The independent variable of yearly demand for mental health services depicts how noteworthy changes in the state of the economy impact the demand for mental health services. This information in conjunction with the more seasonal results of the monthly independent variable and operational results of daily demand provides a strategic benefit to decision makers. Data mining techniques applied to data resources provide a robust decision support mechanism that can help reduce uncertainties in
consumer/patient demand and therefore enhance their ability to optimize resource allocations. Resource optimization takes the form of better matching available clinician time on a calendar basis with the expected demand for their services. Model results can be applied to the operational framework of the staffing characteristic of the medical center as a decision support mechanism. Total clinicians on staff on particular days along with clinicians’ time availability to treat patients needs to be estimated. Factors including preparation activities prior to patient visits and post patient visit activities such as reviews and time for administrative paper work need to be taken into consideration when estimating available clinician time during the day to treat patients. This estimation has to be applied to model output of daily expectations of patient demand. This enables health-care providers to maintain a more stable and improved quality of care, which may enhance the operational state of organizations (e.g cost reductions through reduced idle clinician time or reduced liabilities resulting from malpractice from understaffed operations). The importance of enhancing the quality of care for mental health services cannot be underestimated. Applying the appropriate amount of care, as measured by time of care and adequately trained care provider, provides a benefit not only to the individual being treated but society as a whole as it mitigates adverse events that can result from inappropriately or untreated illnesses.
References ACHARYA UR et al (2012) Data mining framework for fatty liver disease classification in ultrasound: a hybrid feature extraction paradigm. Medical Physics 39(7), 4255–4264. AUSTIN PC, LEE DS, STEYERBERG EW and TU JV (2012) Regression trees for predicting mortality in patients with cardiovascular disease: what improvement is achieved by using ensemble-based methods? Biometrical Journal 54: 657–673.
Health Systems
BATAL H, TENCH J, MCMILLAN S, ADAMS J and MEHLER PS (2001) Predicting patient visits to an urgent care clinic using calendar variables. Academic Emergency Medicine 8(1), 48–53. CATALANO R (1991) The health effects of economic insecurity. American Journal of Public Health 18(9), 1448–1152. CIOS KJ and MORE GW (2002) Uniqueness of medical data mining. Artificial Intelligence in Medicine 26(1–2), 1–24.
A data mining approach for estimating patient demand
CHU C and ZHANG G (2003) A comparative study of linear and nonlinear models for aggregate retail sales forecasting. International Journal of Production Economics 86(3), 217–231. DIGILIO D and AMERICAN PSYCHOLOGICAL ASSOCIATION. Mental and behavioral health and older Americans. Fact Sheet, [WWW document] http://www.apa.org/about/gr/issues/aging/mental-health.aspx. DOOLEY D, PRAUSE J and HAM-ROWBOTTOM K (2000) Underemployment and depression: longitudinal relationships. Journal of Health and Social Behavior 41(4), 421–436. FAYYAD U, SHAPIRO G and SMYTH P (1996) From data mining to knowledge discovery in databases. Artificial Intelligence 13(3), 37–54. FORTNEY JC et al (2010) A web-based clinical decision support system for depression care management. American Journal of Managed Care 16(11), 849–854. FRENCH M, KRAJEWSKI R and CUYKENDALL R (1992) Rainfall forecasting in space and time using a neural network. Journal of Hydrology 137(1–4), 1–31. FULLERTON KJ and CRAWFORD V (1999) The winter bed crisis – quantifying seasonal affects on hospital bed usage. QJM 92(4), 199–206. GAO P et al (2012) Which is a more accurate predictor in colorectal survival analysis? Nine data mining algorithms vs. the TNM staging system. PloS One 7(7), e42015. Epub 25 July. HINTON G (1992) How Neural Networks Learn from Experience. Scientific American 267(3), 144–151. HOLLEMAN DR, BOWLING RL and GATHY C (1996) Predicting daily visits to a walk-in clinic and emergency department using calendar and weather data. Journal of General Internal Medicine 15(4), 237–239. HONBERG R, KIMBALL A, DIEHL S, USHER L and FITZPATRICK M (2011) State Mental Health Cuts: A Continuing Crisis. NAMI [WWW document] http:// www.NAMI.org (accessed 24 June 2014). JONES SS, THOMAS A, EVANS RS, WELCH SJ, HAUG PJ and SNOW GL (2008) Forecasting daily patient volumes in the emergency department. Academic Emergency Medicine 15(2), 159–170. JONES SS et al (2009) A multivariate time series approach to modeling and forecasting demand in the emergency department. Journal of Biomedical Informatics 42(1), 123–139. KUDYBA S (2010) Healthcare Informatics: Increasing Efficiency and Productivity. Taylor Francis, New York. KUDYBA S (2012) Utilizing multivariate analytics for decision support to enhance patient demand forecasting. International Institute for Analytics October, 1–7.
Stephan Kudyba and Thad Perry
7
KUO MH, HUNG CM, BARNETT J and PINHEIRO F (2012) Assessing the feasibility of data mining techniques for early liver cancer detection. Studies in Health Technology and Informatics 180: 584–588. LAMANTIA MA et al (2010) Predicting hospital admission and returns to the emergency department for elderly patients. Academic Emergency Medicine 17(3), 252–259. LISBOA P et al (2008) Time-to-event analysis with artificial neural networks: an integrated analytical and rule-based study for breast cancer. Neural Networks 21(2–3), 414–426. LURIE SJ, GAWINSKI B, PIERCE D and ROUSSEAU SJ (2006) Seasonal affective disorder. American Family Physician 74(9), 1521–1524. MILLER RA (1994) Medical diagnostic decision support systems – past, present, and future: a threaded bibliography and brief commentary. Journal of the American Medical Informatics Association 1(1), 8–27. NAUERT R (2010) Demand up yet funds drop for mental health care, Psych Central. [WWW document] http://psychcentral.com/news/2010/07/ 19/demand-up-yet-funds-drop-for-mental-health-care/15740.html. PECK JS, BENNEYAN JC, NIGHTINGALE DJ and GAEHDE SA (2012) Predicting emergency department inpatient admissions to improve same-day patient flow. Academic Emergency Medicine 19(9), E1045–E1054. PENNY KI and SMITH GD (2012) The use of data-mining to identify indicators of health-related quality of life in patients with irritable bowel syndrome. Journal of Clinical Nursing 21: 2761–2771. RUMELHART D, HINTON G and WILLIAMS RJ (1986) Learning internal representations by error propagation. In Parallel Distributed Processing: Explorations in the Microstructure of Cognition. (RUMELHART DE, MCCLELLAND JL and CORPORATE PDP RESEARCH GROUP, Eds) MIT Press. SRIDHAR S (2013) Improving diagnostic accuracy using agent-based distributed data mining systems. Informatics for Health and Social Care 38(6), 172–195. SUN Y, HENG BH, TAY SY and SEOW E (2011) Predicting hospital admissions at emergency department triage using routine administrative data. Academic Emergency Medicine 18(8), 844–850. TAM K and KIANG M (1992) Managerial applications of neural networks: the case of bank failure predictions. Management Science 38(7), 926–947. WARGON M, GUIDET B, HOANG TD and HEJBLUM G (2009) A systematic review of models for forecasting the number of emergency department visits. Emergency Medicine Journal 26(6), 395–399. YASDI R (1999) Prediction of road traffic using a neural network approach. Neural Computing & Applications 8(2), 135–142.
Health Systems