Share with your friends










Submit

Analytics Magazine

Intelligent decision support in healthcare

January/February 2012

Runki Basu, Norm Archer and Basudeb MukherjeeBy (left to right) Runki Basu, Norm Archer and Basudeb Mukherjee

Decision support is a crucial function for decision makers in many industries. Typically, decision support systems help decision-makers to gather and interpret information and build a foundation for decision-making. Such systems may range from simple software systems to complex knowledge-based and artificial intelligence systems. Decision support systems can be database-oriented, spreadsheet-oriented or text-oriented in nature.

In healthcare, clinical decision support systems (CDSS) can play a significant role. Clinical decisions that are routinely taken by healthcare service providers are often based on clinical guidance and evidence-based rules derived from medical science. However, intelligent decision support systems (IDSS), through the interpretive analysis of large-scale patient data with intelligent and knowledge-based methods, “allow doctors and nurses to quickly gather information and process it in various ways in order to assist with making diagnosis and treatment decision” [1]. IDSS can be applied in healthcare in diverse areas such as the examination of real-time data from diverse monitoring devices, analyses of patient and family history for the purpose of diagnosis, reviews of common characteristics and trends in medical record databases and many more areas [1].

This article demonstrates how a hybrid architecture combining the concepts of data mining (DM) and artificial neural networks (ANN) can be applied to patient data for intelligent decision support in healthcare [2][3]. An IDSS in healthcare gathers and incorporates healthcare-specific domain knowledge and performs intelligent actions, including learning and reasoning while recommending clinical steps to take and justifying the outcomes [2][3].

Understanding Intelligent Decision Support: Some Definitions

To comprehend intelligent clinical decision support, it is important to define some relevant terms and related decisions. Several approaches are available for aiding decision support in healthcare, while artificial intelligence support remains “one of the many methodological domains from which good and necessary ideas can be derived” [4].

Clinical Decision Support Systems (CDSS): Computer applications that support and assist clinicians in improved decision-making by providing evidence-based knowledge with respect to patient data. This type of computer-based system consists of three components: a language system, a knowledge system and a problem processing system [1][3].

Intelligent Decision Support System (IDSS): Intelligent decision support is provided by a system that helps in decision-making through a display of intelligent behavior that may include learning and reasoning. Such learning and reasoning can be achieved through implementing rule-based expert systems, knowledge-based systems or neural network systems [2].

Artificial intelligence (AI): AI refers to the art of empowering computers with intelligence similar to that of humans. This is achieved by combining hardware and software systems so they can perform tasks that are rule-based and require decision-making [3].

Artificial Neural Network (ANN): A mathematical model that simulates the structure and functional aspects of biological neural networks. It mimics in a simplified way how the human brain processes information. Composed of a number of highly connected processing elements (neurons/nodes), ANNs, like people, learn by example. ANNs have the ability to identify meaning from complicated data and to extract patterns and trends that are too complex to be noticed by either humans or other computer-based techniques. A trained ANN can be an “expert” in the category of information it has been given to analyze. This “expert” can subsequently be used to predict and answer “what if” questions in a new situation of interest [3].

Differences Between IDSS and DSS

An IDSS induces specific domain knowledge from raw data by identifying and extracting strategically useful information patterns from this data, thus making the extracted patterns understandable and usable for decision-making. IDSS, unlike DSS, “allows for supporting a wider range of decisions including those with uncertainty” [1]. IDSS, in addition to giving recommendations, may also contribute estimates of the level of confidence in the recommendations it gives.

IDSS can handle complex problems, applying domain-specific expertise to assess the consequences of executing its recommendations. Decisions supported by IDSS also tend to be more consistent, timely and better managed in terms of managing uncertainty in the outcomes. The justification of outcomes provided by an IDSS is particularly significant if it allows clinical experts to validate the explanations provided by the IDSS [1].

How Can Intelligent Decision Support Help?

Managing knowledge in healthcare organizations to aid clinical decision-making requires transforming information into actionable intelligence that can be interpreted by different functional workgroups within the organization. This is demonstrated in Figure 1, a representation of the healthcare knowledge cycle from Patel et al [4], which shows how artificial intelligence can be used to analyze healthcare data and generate a representation of knowledge that can in turn be used for information and process modeling.

Figure 1: The knowledge cycle implemented with AI methods and tools (adapted from [4]).
Figure 1: The knowledge cycle implemented with AI methods and tools (adapted from [4]).

Intelligent decision support systems can help in multiple ways in clinical decision-making at both the individual patient level and the population level. For example:

  • Diagnose by regularly interpreting and monitoring patient data. An IDSS can implement rules and patterns for individual patients, based on clinical parameters, and raise warning flags when such rules are violated. These flags can lead to clinical interventions that save lives.
  • Help chronic disease management through establishing benchmarks and alerts. For chronically ill patients, a deviation noticed by an IDSS in, say, a blood test reading from a diabetic patient could result in an intervention before the patient gets into difficulty.
  • Help public health surveillance by detecting pandemic diseases or in surveillance of chronic diseases. In case of a pandemic, an IDSS can interpret data and predict possible future spread of the disease.
  • Additionally, IDSS can perform regular clinical decision support functions like preventing drug-drug interactions. Even if not noticed by the prescribing physician, an IDSS can spot incompatibilities between prescribed medications and/or dosages for the patient.

Case Study

In order to understand the efficacy of IDSS in clinical setting, consider the case of Jane Doe, a 41-year-old female diabetic patient who takes Metformin, Lipitor and aspirin. She has an average blood pressure (BP) of 108/55 mmHg and heart rate of 60 beats per minute (BPM) and no prior history of cardio-vascular ailments. She has been consulting with Dr. Smith, a family physician, for the last eight years. Dr. Smith uses an Electronic Health Record (EHR) system. Jane self-monitors her Blood Glucose Level, BP and BPM about four times a day and stores the data in an online Patient Health Record (PHR) system. Her PHR system is integrated with Dr. Smith’s EHR system so that data can be uploaded from the PHR system to the EHR system. Over time, monitored data can generate significant sets of data that can be used, in conjunction with medical care guidelines, for the care of patients with specific chronic diseases. The data can be mined continuously to derive and update intelligent decision rules that are focused on specific patients and that can adapt over time to patient status [3].

On a given day, Dr. Smith finds Jane’s BP to be 120/70 with a heart rate of 80 BPM, but he notices a mild systolic murmur. Normally in such a situation, Dr. Smith would prescribe medium intensity Warfarin oral anti-coagulation therapy. But Jane, who is also taking aspirin, runs the risk of developing an increased risk of uncontrolled bleeding if she takes Warfarin concurrently with aspirin [3].

At this point, if Dr. Smith’s EMR were integrated with an IDSS, a prescription for Warfarin would trigger two alerts and associated recommendations.

The IDSS creates a patient profile over time through the process of machine learning, using previous data collected from the patient, and triggers an alert as soon as one or more values within the profile get out of range. In Jane’s case, the system detects that both her BP and heart rate are out of range for Jane’s profile, even though the same readings might be within range for other patients, and therefore triggers an alert for immediate intervention [3]. (See Table 1.)

Table 1: IDSS intervention in patient specific scenario.
Table 1: IDSS intervention in patient specific scenario.

The IDSS also detects that the combined prescribed dosage of both aspirin and Warfarin enhances the risk of bleeding and signals an alert of a drug-to-drug reaction, including a recommendation of a corrective dose of 100 mg aspirin per day and also to monitor the international normalized ratio (INR) for Warfarin treatment [3]. An INR is a laboratory test that measures the time for blood to clot and then compares it with an average time. A higher INR indicates a longer time for blood to clot, thereby preventing formation of clots that may cause stroke. INR is a useful test to monitor the impact of anticoagulant medicines such as Warfarin. If INR is too high then uncontrolled bleeding may occur [5].

Note that if Dr. Smith were using an EMR with an installed DSS, a drug-drug interaction alert would have triggered but not a heart rate or blood pressure alert, which is based on patient profiling and pattern recognition.

INR = International Normalized Ratio Table 2: Potentially clinically significant drug-drug interactions [3].
INR = International Normalized Ratio
Table 2: Potentially clinically significant drug-drug interactions [3].

Architecture of an IDSS Model

The IDSS model described here combines an association rule-generator algorithm based on data mining of a knowledge base (KB) with an artificial neural network (ANN) system [2]. The system is capable of building domain knowledge from existing datasets and applying this knowledge to solve clinical problems. As data mining extracts domain-specific knowledge from organizational databases, it also enhances the process of knowledge acquisition. Neural networks can learn patterns from large volumes of data and use the knowledge thus extracted to help solve problems [2].

The solution discussed here has been adapted from a similar architecture presented by Viademonte et al. [2]. In the context of the primary healthcare system, the system creates and tracks patient profiles and then uses the patterns it recognizes from the data to identify unusual test readings and trigger alerts for possible intervention. It can also assist in diagnosing certain diseases based on a set of observed symptoms and suggest recommendations. The IDSS retrieves information learned in the past, creates domain knowledge from recalled information and translates it into “new” domain knowledge, to serve as a predictive tool. It is basically a hybrid system for applying descriptive and predictive models for intelligent decision-making [2][3].

This system allows raw data to be retrieved from data bases and processed into data models. These data models support descriptive methods that are stored in knowledge bases. Subsequently, predictive methods (based on neural network models) are generated that produce predictions [2].

The IDSS architecture under discussion can operate either through data mining for knowledge acquisition or through a neural network-based system operating as an advisory system. While the data mining technology offers expertise, the ANN-based system acquires knowledge through learning and reasoning as well at the intuitive user interface level [2]. At the data level, the system depends on a master data warehouse that combines relevant data repositories, case bases and knowledge bases. The elements of the architecture include [2]:

  • a decision-oriented data repository, such as a data warehouse
  • case bases
  • inductive algorithms for data mining (descriptive method)
  • knowledge bases
  • an intelligent advisory system (predictive method)

Through a training process, data mining algorithms can be applied to multiple data bases to build multiple descriptive models that are stored in knowledge bases. As neural networks are applied to these descriptive models, predictive models are generated. Domain specific cases or case bases and their corresponding data models are created by extracting data from the data warehouse. The process ensures data consistency within that domain [3]. A “case” in this discussion represents an instance of a problem within the specific domain and falls into a specific well-defined class consisting of attributes and values [2][3].

Once data mining has been successfully employed to extract relevant relationships from the case bases, association rules are applied to produce general knowledge that is stored in the knowledge base. In the medical field, specific clinical cases or practice guidelines can be used as case bases from which data can be mined to produce clinical knowledge for generating descriptive clinical features or for decision support functions [2][3].

The dashed lines in Figure 2 symbolize processes and the solid lines with arrowheads symbolize data flows between components. The data warehouse consists of pre-processed historical data that are mined. Subsequently, cases are selected, extracted and stored in case bases [2].

Figure 2: Main building blocks of the IDSS Model (adapted from [2]).
Figure 2: Main building blocks of the IDSS Model (adapted from [2]).

Conclusion

The IDSS model described in this article is capable of learning, generalizing and self-organizing in order to recognize complex patterns and assist in decision support [2]. The case study of Jane, the 41-year-old diabetic patient, indicates that when self-monitoring data and test data at each patient visit are available to a physician using an EMR with IDSS support, the physician would be able to make better decisions. In the future, uploading self-management monitoring data automatically from patient personal health records to family physician electronic medical records may become the norm for chronically ill patients, and the intelligent decision support system discussed here would be able to play an important role in providing improved patient care.

In future research, integrating interactions between different components will be implemented through a manager component for coordinating neural network functions and data mining. Continuous learning of the neural network can also be implemented and knowledge inter-operability among different systems can be expanded through the application of standard XML terminology [2].
Runki Basu is vice president of Synergy Tech, an IT company specializing in healthcare. Basu has more than 15 years of experience providing business and technology solutions to corporations in India, the United States and Canada. Based in Toronto, Canada, her current focus is in health informatics. Her areas of interest and expertise include medication management using health information technology, secure and private communication of patient data and application of data mining and artificial intelligence in clinical decision support.

Dr. Norm Archer is a professor emeritus in the DeGroote School of Business at McMaster University in Hamilton, Ontario, Canada. He is active in the study of organizational problems relating to the implementation of eBusiness approaches in health, business and government organizations.


Basudeb Mukherjee (mukherjee.basu@gmail.com) is the president and CEO of Synergy Tech. Mukherjee has spent more than 20 years providing business and technology solutions to Fortune 500 Companies including Lehman Brothers, Philip Morris, Priceline.com, Marsh & McLennan, Prudential Securities and Toronto Transit Corporation, as well as several agencies of the provincial government of Ontario, Canada. His current focus is in healthcare where he conducts research on advanced adoption of technology in healthcare and provides business solution to healthcare service providers and pharmaceutical companies across North America.

References

  1. D. Foster, C. McGregor, S. El-Masri, 2005, “A survey of agent-based intelligent decision support systems to support clinical management and research.
  2. S. Viademonte and F. Burstein, 2006, “From knowledge discovery to computational intelligence: A framework for intelligent decision support systems,” Chapter 4 in “Intelligent Decision-making Support Sytems,” Springer-Verlag London Limited, pp. 57-78.
  3. Basu, R., Fevrier-Thomas, U., Sartipi, K., 2011, “Incorporating hybrid CDSS in primary care practice management,” McMaster eBusiness Research Centre, November 2011.
  4. V. L. Patel, E. H. Shortliffe, M. Stefanelli, P. Szolovits, M. R. Berthold, R. Bellazzi and A. Abu-Hanna, 2009, “The coming of age of artificial intelligence in medicine,” Artificial Intelligence in Medicine, Vol. 46, No. 1, pp. 5-17.
  5. Heart Rhythm Society: International normalized ratio (INR).

business analytics news and articles

Related Posts

  • 41
    Analytics isn’t a new subject, even for me. Like most professors, I did a lot of analytics in graduate school, and even had a real job as head of statistical computing at Harvard for a while. Then I lost interest for 20 years outside of an occasional correlation or regression…
    Tags: decision, support, knowledge, intelligence, data
  • 37
    July/August 2015 By Sagar Anisingaraju and Mo Kaushal (l-r) The healthcare industry in the United States is going through a major transformation and is becoming much more consumer friendly. This process of “consumerization” will drive healthcare executives to pay close attention to what patients feel and think about healthcare organizations…
    Tags: data, patient, healthcare
  • 33
    Fall 2009 CLICK HERE TO GO TO THE DIGITAL VERSION OF THIS ARTICLE Integrating often-competing methodologies helps analyze and solve particularly tough problems. By Jeremy F. Shapiro The information revolution has reached a stage where business managers are seeking to extract intelligence from their large and growing transactional databases. After…
    Tags: data, system, systems, decision, knowledge, intelligence
  • 32
    March/April 2015 Welcome to the ‘We Economy’: Finding the business in your data By Arnab Chakraborty, Michael Svilar and Prith Banerjee (l-r) The Internet of Things (IoT) and the data it is producing are becoming main catalysts for change and transformation for businesses and consumers alike. Today, with every object,…
    Tags: data, healthcare, patient
  • 30
    July/August 2015 How neuro-dynamic programming enables smart machines to think ahead. By Scott Zoldi The media and watercooler chatter alike increasingly focus on how advances in machine learning and artificial intelligence (AI) are boosting the ability of predictive analytics to benefit businesses’ bottom lines. Some of that talk ponders the…
    Tags: data, decision, artificial, intelligence

Headlines

Grocery store data speeds early detection of food-borne illness

Foodborne illnesses, like salmonella, E. Coli and norovirus infections, are a major public health concern affecting more than one out of six Americans each year, according to the Centers for Disease Control and Prevention (CDC). During a foodborne illness outbreak, rapidly identifying the contaminated food source is vital to minimizing illness, loss and impact on society. Read more →

Gartner: Top 25 North American Supply Chain Undergraduate Programs

Gartner, Inc. recently released the findings from its biennial Top 25 North American Supply Chain Undergraduate Programs report. Fifty-one institutions participated in the report that is intended to support chief supply chain officers (CSCOs), heads of supply chain strategy and supply chain HR partners in building a strong portfolio of university recruiting and internship partners. Read more →

Study: Tweets better than Google Trends at forecasting TV ratings

How well does the emotional and instantaneous content in tweets perform relative to the more deliberate searches recorded in Google Trends in forecasting future TV ratings? In a massive big data analysis using data from Twitter, Google Trends and other widely used websites for entertainment information, a forthcoming article in the INFORMS journal Marketing Science finds that mining Twitter content is significantly more effective than Google Trends in its ability to predict future TV ratings. Read more →

UPCOMING ANALYTICS EVENTS

INFORMS-SPONSORED EVENTS

INFORMS Annual Meeting
Nov. 13-16, 2016, Nashville, Tenn.

CONTINUING EDUCATION

Foundations of Modern Predictive Analytics
Nov. 17-18, 2016, Nashville, Tenn. 

WORKSHOPS

The Arithmetic of Uncertainty: A Cure for the Flaw of Averages
A one-day (9 a.m.-5 p.m.) workshop with optional hands-on computer lab
Sept. 14: Washington, D.C. (Crystal City)
Sept. 20: Houston
Sept. 21: Dallas
For more information and to register, click here.

CAP® EXAM SCHEDULE

CAP® Exam computer-based testing sites are available in 700 locations worldwide. Take the exam close to home and on your schedule:


 
For more information, go to 
https://www.certifiedanalytics.org.