Clinical decision support system

Clinical (or Diagnostic) Decision Support Systems (CDSS) are interactive computer programs that directly assist physicians and other health professionals with decision making tasks. These fall under the class of Decision support systems within medical informatics. CDSS offer a powerful medical informatics tool that can promote evidence based medicine.

Why do we need such systems at all? If diagnosis becomes the job of the computer program, what will the doctors do? Such questions overlook the key term support - i.e., they are intended to support the clinician, rather than replace them. Computers are not error or fatigue prone like human beings; however, computers may cause harm through other faults.

In medical diagnosis, there is scope for ambiguity in the inputs, such as the history (patient’s description of the diseased condition), physical examinations (especially in cases of uncooperative or less intelligent patients), and laboratory tests (faulty methods or equipment). Moreover, in treatment, there are the chance of drug reactions and specific allergies, and also of patients' non-compliance of the therapy due to cost or time or adverse reactions. All these factors may not be taken into account by a busy clinician attending hundreds of patients daily.

In all these areas, computers can help the clinician to reach an accurate diagnosis faster. Another new branch of medicine pharmacogenomics is the product of breeding between information technology and biology, leading to individualized treatment.

The basic components of a CDSS include a dynamic (medical) knowledge base and an inference mechanism (usually a set of rules derived from the experts and from the data provided by evidence-based medicine). It may be based on Expert systems or artificial neural networks or both (Connectionist expert systems) - all of which can be loosely termed as Artificial intelligence - AI - techniques.

CDSS may be linked to Electronic health records, for decision making and can also be used for assisting the practice of Evidence-based medicine (EBM) in a more structured way.

The role of even an apparently simple search by Google in making diagnoses has been quantified.

Methods of Decision Support

Various tools applied for Artificial intelligence have been used as decision support systems and data mining.

Knowledge based systems / expert systems

Knowledge-based expert systems are created by having experts use the biomedical literature to identify relationships between independent variables (such as signs and symptoms) and dependent variables (such as likely underlying diseases). Sometimes, local hospital information, such as rates of surgical complications, may be incorporate. These relationships become predefined rules in the form of {if-else if-then} to guide the decision making. Rules may also be obtained by various forms of decision trees (e.g., Iterative Dichotomizer or ID variants) or Bayesian networks.

Rule-based CDSS are the ones that are mostly found in the commercially available clinical informatics applications. Alerts for allergies and possible drug interactions, prompts for drug doses corrected for weight, height, sex, age and underlying clinical condition are the ones that are most commonly touted as CDSS. Arden Syntax was designed and developed as a product-independent syntax for codifying these alerts. Examples of rule-based systems are:

1. Logical or deductive: branching logic (e.g., ID3 or iterative dichotomizer 3); e.g., MYCIN, NEOMYCIN
2. Probabilistic: Bayesian systems weight probabilities rather than use binary options as done be logical systems. An example is DXplain.
3. Hybrid: heuristic reasoning e.g., QMR (Quick Medical Reference), ILIAD, ISABEL. Hueristic methods may perform better for uncommon diseases.

The critical difficulty in constructing rule-based expert systems for medical decision support has been in recruiting experts with domain knowledge to create the knowledge base and train the system. Training expert systems is time-consuming, and has only produced usable results in narrowly-scoped projects. However, due to issues related to its acceptance (a doctor may need to 20-40 minutes to enter a case) and standardization, rule-based CDSS has not delivered its immense promise.

Evidence-adaptive systems

Evidence-adaptive systems are proposed that are based on current evidence and have a mechanism for routine updating of recommendations with new research findings.

Machine learning

In these systems, the relationships between independent variables (such as signs and symptoms) and dependent variables (such as likely underlying diseases) is created by having the system be trained on a "large collection of previously classified examples during a period of supervised learning". A classic example is automated electrocardiogram interpretation.

Artificial Neural Networks

Artificial Neural Networks ANNs (also referred to as connectionist architectures, parallel distributed processing, and neuromorphic systems), are information-processing paradigm inspired by the way the densely interconnected, massively parallel structure of the mammalian brain processes information. ANNs are collections of mathematical models that emulate some of the observed properties of biological nervous systems and draw on the analogies of adaptive biological learning. ANN can perform supervised or unsupervised machine learning, depending on the categorical information maps available at the time of rule inferencing.

Bayesian Belief Network

A Bayesian Belief Network (BBN) is a mathematical model for using AI (probabilistic) networks for predictive purposes. It is used in the CDSS module as an AI engine to help make probabilistic predictions based upon the observations made by the clinician. These could be used for diagnostics, treatment planning, and even predicting outcomes. An example is Pathfinder for surgical pathology.

Connectionist expert systems

Connectionist expert system is a type of Artificial Neural Networks (ANNs) in which humans can help the system revise weights. Although humans help revise the system based on the empiric data in the system, literature based knowledge is not directly used to modify weight. Connectionist expert system are the "inferencing" methods of the ANNs can be backtracked and "rules generation" is possible. This might actually lead to the enhancement and enrichment of the medical knowledge base itself. This system architecture corresponds to the concept of "unsupervised" machine learning, where the algorithm looks for patterns in data without being instructed about the actual categories of information. This is also known as data mining. An example of a connectionist system is HYCONES.

Hybrid systems

An example of a hybrid system has been developed for diagnosing congenital heart disease. This example combines case-based reasoning with a neural network.

Effectiveness

A systematic review of studies published through September, 2004 concluded "many CDSSs improve practitioner performance. To date, the effects on patient outcomes remain understudied and, when studied, inconsistent".

A more recent trial found that a clinical decision support system could improve antimicrobial prescribing.

In internal medicine, a study of four diagnostic support systems diagnosed a collection of tests cases with accuracy ranging from 0.52 to 0.71. Another study of ISABEL found it included the correct diagnosis in its list of 30 proposed diagnoses for a collection of tests cases in 96% of the cases. A study of QMR found that it could correctly list the final diagnosis among the first five diagnoses it suggested for 36% to 40% of 1144 consecutive inpatients.

In pediatrics, a systematic review by the Cochrane Collaboration concluded "there are very limited data from randomised trials on which to assess the effects of clinical decision support systems in neonatal care."

Open source initiatives

OpenCLIPS [1] is an open source release of the CLIPS expert system shell. It has been based on the inference engine of EGADSS [2]. Their goal is to build an active community focused around the support and development of CLIPS [3].

One of the earlier evaluations on the efficacy of Internet-based CDSS has been favorable [4].

A CDSS using open source software and delivered through wireless hand-held device has been found to be effective in stopping smoking in primary care [5].

PubMed Central [6] has nearly 2500 full-text articles on CDSS built with the help of open source software.

Market surveys [7] have predicted that development of more robust CDSS is likely to increase its adaptation.

Challenges

Sim et all have identified five challenges to CDDS:

1. "Capture of both literature-based and practice-based research evidence into machine-interpretable formats suitable for CDSS use"
2. "Establishment of a technical and methodological foundation for applying research evidence to individual patients at the point of care"
3. "Evaluation of the clinical effects and costs of CDSSs, as well as how CDSSs affect and are affected by professional and organizational practices"
4. "Promotion of the effective implementation and use of CDSSs that have been shown to improve clinical performance or outcomes"
5. "Establishment of public policies that provide incentives for implementing CDSSs to improve health care quality"

Further reading

1. Bergman LG, Fors UG, Computer-aided DSM-IV-diagnostics – acceptance, use and perceived usefulness in relation to users’ learning styles. BMC Med Inform Decis Mak. 2005; 5:1
2. Coiera E. (2003) Chapter 25 - Clinical Decision Support Systems. In Guide to health informatics. London: Arnold. ISBN 0-340-76425-2.
3. Kavitha S, Sarbadhikari SN, Rao Ananth N, Implementation of Decision Tree Classifier Using Classification Algorithm for Some Inborn Errors of Metabolism, Proc. Global Convention and Exposition on Telemedicine and eHealth, New Delhi 17-22 August, 2006.
4. Kunnamo I, et al, 2005, National Decision support database based on computer-readable guidelines and using structured data from electronic patient records http://www.terveysportti.fi/pls/kotisivut/docs/f1917377807/gin_poster_decision_support_v2.pdf (Accessed July 2006)
5. Mendelson D, Carino TV, Evidence-Based Medicine In The United States—De Rigueur Or Dream Deferred? Health Affairs, 2005, 24: 133 – 136. DOI 10.1377/hlthaff.24.1.133
6. Pradhan M, The Crystal Ball - The Future of Informatics and Decision Making, http://www.informatics.adelaide.edu.au/topics/DS/MP-CrystalBallTalk.html 2001 (Accessed July 2006)
7. Sarbadhikari SN, A CDSS for diagnosing amenorrhea http://www.geocities.com/drsupten 2006
8. Sarbadhikari, SN Automated diagnostic systems. Indian Journal of Medical Informatics, 2004b, 1: 25-28. (Accessible at http://openmed.nic.in/218/])
9. Smith S, The Classification Algorithm http://www.cs.mdx.ac.uk/staffpages/serengul/The.Classification.algorithm.htm (Accessed July 2006).
10. Trowbridge et al. (2001) Chapter 53: Clinical Decision Support Systems. In Making Health Care Safer: A Critical Analysis of Patient Safety Practices. Evidence Report/Technology Assessment: Number 43. AHRQ Publication No. 01-E058, July 2001. Agency for Healthcare Research and Quality, Rockville, MD. http://www.ahrq.gov/clinic/ptsafety/

External links

• EGADSS is an Evidence-based guidelines and decision support system using Open Source approach.
• OpenClinical maintains an extensive archive of Artificial Intelligence systems in routine clinical use.
• OpenClinica is a professional open source for clinical research.
• Some useful links
• Prerequisites
• A slide show
• Improving Outcomes with Clinical Decision Support: An Implementer’s Guide is a new resource designed to help healthcare organizations use clinical decision support (CDS) to measurably improve key healthcare outcomes such as the quality, safety, and cost-effectiveness of care delivery.