CDSS: Revolutionizing Clinical Decision-Making with AI-Driven Support Systems

Limitations in physician expertise, increased human errors, and rising healthcare costs have posed significant challenges to healthcare systems worldwide. In recent years, with the growing technological advancement and digitalization of the medical industry, clinical decision support systems (CDSS) have garnered increasing attention.

As an active branch of medical knowledge engineering and artificial intelligence research, various novel clinical decision support systems are being continuously developed. These systems aim to transform the traditional diagnostic paradigm that relies on physicians’ experience and laboratory indicators, thereby enabling intelligent clinical decision guidance and ensuring healthcare quality.

What isCDSS？

According to statistics, approximately 44,000 to 98,000 people die annually in the United States due to medication errors, resulting in economic losses of approximately $17 billion to $29 billion. According to a 2006 survey by the U.S. National Academy of Sciences, there are approximately 400,000 cases of preventable, medication-related patient harm occurring within hospitals each year, and approximately 530,000 cases occurring outside hospitals (including clinics). Practice has shown that the lack of complete and shared resident health information, leading to duplicate diagnoses, is a significant factor in the rapid growth of healthcare expenditures. Furthermore, insufficient decision support, resulting in inappropriate medication use or improper procedures, is the primary cause of most medical errors and even liability incidents.

With the rapid advancement of technologies such as computing and network communications, clinicians, healthcare institution administrators, and health policy makers are increasingly seeking decision support services enabled by information technology. In particular, the development and application of emerging technologies—including data mining, online analytical processing (OLAP), and artificial intelligence—have provided the technical foundation for the implementation of clinical decision support systems.

Clinical Decision Support Systems (CDSS) are human-computer interactive health information technology applications designed to provide clinical decision support (CDS) to physicians and other healthcare practitioners, facilitating clinical decision-making through data, models, and other aids. The concept of “Clinical Decision Support Systems” continues to evolve. The currently prevailing working definition, proposed by Robert Hayward of the Centre for Health Evidence, is: “linking clinical observations with medical knowledge to influence physicians’ choice of interventions, thereby improving the quality and outcomes of healthcare services.”

Clinical decision support systems (CDSS) are also a significant application of artificial intelligence in medicine. Some believe that, in the future, the diagnosis and treatment of common diseases could be entirely entrusted to clinical decision support systems. Typical examples of CDSS include: physician order entry systems, the MYCIN medical diagnostic expert system, Quick Medical Reference (QMR), Logic Health Assessment System, and drug therapy screening systems.



CDSSApplications of

Early-generation novel clinical decision support systems were typically used solely to assist physicians in making diagnostic and therapeutic decisions. Clinicians would first input information, then await the system’s output of an analyzed “correct” decision, and finally determine the feasibility of the candidate options provided by the system, thereby guiding the decision-making process.

Novel decision support theories emphasize the interaction between physicians and systems, leveraging physicians’ knowledge reserves and extensive clinical experience alongside the system’s management of medical data to facilitate in-depth analysis of patient information, thereby enabling the most appropriate diagnostic and therapeutic decisions. This approach offers significant advantages over relying solely on either physicians or Clinical Decision Support Systems (CDSS) independently. Notably, after physicians input patient data, the CDSS can output relevant information and generate customized plans tailored to individual cases for clinicians’ review. This allows physicians to conveniently select useful information while discarding erroneous recommendations.

CDSS Classification

CDSSBy System Architectureinto two categories：

• Based on the Knowledge Base




• Non-Knowledge Base

Most clinical decision support systems fall into the first category and comprise three components: a knowledge base, an inference engine, and a human-computer interaction interface. The knowledge base stores extensive compiled information, typically managed and stored using IF-THEN rules. For example, regarding drug interactions, a rule may be formulated as: “IF Drug X is administered AND Drug Y is administered, THEN display a warning message.” Advanced users can also customize rules within the knowledge base via a separate editing interface to meet specific needs, such as implementing real-time updates for new medications. The inference engine automatically integrates and analyzes patient data based on the rules contained in the knowledge base. The human-computer interaction interface feeds the analysis results back to the user or serves as the system input.

Non-knowledge-based clinical decision support systems mostly adopt the form of artificial intelligence, mainly throughMachine LearningAutomatically extract rules from existing experience. Common construction methods include support vector machines, artificial neural networks, and genetic algorithms.

CDSSBy Timing of Use:

• Pre-diagnosis: Assisting physicians in preliminary diagnostic preparation.




• Diagnosis: Assisting physicians in reviewing and filtering candidate options to refine final clinical decision-making.




• Post-diagnosis: Uncover the relationships within data from patients' medical histories and clinical records to help predict future health outcomes.

CDSSofFeatures and Functions

ExcellentClinical Decision SupportSystemCharacterized by the following features： 

• Automatically push results without requiring user activation of the system;




• Integrated into clinical workflows, rather than operating independently of them;




• Based on electronic systems, rather than paper-based systems;




• Used at the bedside or during consultation, rather than before or after patient contact;




• Provide recommendations, not assessments.

Clinical Decision Support Systems can alleviate the workload of clinicians, reduce technical risks in clinical practice, and improve patient satisfaction.

Clinical Decision SupportSystemofKey features include:

1. Prompt: Provide clinicians with relevant information to support better decision-making, prevent medical errors, and improve healthcare quality and outcomes.

2. Interventions: Categorized into five types, namely warnings and alerts; information buttons; grouped medical orders (order sets); document management and formatting; and representation of related data.



About CDSSofResearch

In a review of 100 studies involving clinical decision support systems (CDSS), 64% reported improvements in healthcare providers’ performance, while 13% noted enhanced patient outcomes. In another analysis encompassing 70 studies on CDSS, 68% of the clinical trials demonstrated that CDSS can improve clinical workflow.

However, many medical practitioners remain pessimistic about the application of Clinical Decision Support Systems (CDSS). The results of a quinquennial evaluation of CDSS effectiveness, published in 2014, suggested that research on clinical decision support systems still has a long way to go. A significant gap remains between concept and reality, and the advantages of CDSS in terms of digital health technology and cost-effectiveness are still far from sufficient.

The study also found that the integration of clinical decision support systems with electronic health records (EHRs) had no significant effect on reducing mortality risk. Are clinical decision support systems truly effective? Experts remain divided. While there may be certain benefits in other areas, clear evidence is currently lacking.

FacingDisorderwithDevelopmentOutlook

Despite the powerful functionality and ease of use of clinical decision support systems (CDSS), their actual application in clinical practice remains limited. This is primarily due to two reasons: first, the construction of knowledge bases fails to meet the needs of clinicians; second, most systems are disconnected from clinical workflows and lack technical integration with electronic medical records (EMR). As a result, the manner in which these systems deliver decision support does not align with clinicians’ workflow habits, thereby reducing their willingness to adopt such tools.

The Complexity of Medical KnowledgeThis necessitates the consideration of numerous patient-specific factors during system design, such as symptoms, signs, laboratory test results, family history, genetic data, epidemiological information, and existing medical literature. Meanwhile, tens of thousands of new clinical studies are published annually with varying quality, and the sheer volume of data poses significant challenges to system maintenance.

More successful clinical decision support systems are often confined to specific domains, with limited scope. For instance, the Leeds Abdominal Pain Diagnostic System, launched in 1971, achieved a diagnostic accuracy of 91.8%, compared to 79.6% for physicians; however, this system was designed solely for diagnosing abdominal pain.

Complexity of Clinical WorkflowsIt also increases the difficulty of system integration. In particular, many hospitals enforce strict logical isolation or even physical isolation between their internal and external networks, which further restricts the in-hospital application of certain online Clinical Decision Support Systems (CDSS). Currently, most systems remain separate from clinical workflows, requiring physicians to open the CDSS independently and spend time entering patient data. The resulting flood of alert messages leads to alert fatigue among healthcare professionals, thereby reducing work efficiency.

The most successful cases of integration to date are pharmacy systems and billing systems. Since pharmacy workflows are relatively straightforward, CDSS primarily addresses drug-drug interactions, making it easier to design.

Some believe that electronic health records (EHRs) will become the mainstream approach in the healthcare industry, effectively guiding nursing practice and ensuring patient safety. The development of a clinical nursing decision support system begins with the establishment of a knowledge base, followed by the creation of logical reasoning algorithms, and their integration with existing EHR systems. Although EHR systems can acquire, transform, display, and analyze certain information, they fail to meet the demands of complex clinical decision-making if they cannot filter and refine data. In this regard, clinical decision support systems are undergoing further advancement.

By Chen Kun

Editor: Huang Jia