Clinical decision support system (CDSS) assists healthcare professionals by converting real-time medical-related data, documents, and knowledge based into a set of sophisticated algorithms, applying techniques such as machine learning, knowledge graphs, natural language processing, and computer vision to help healthcare providers improve diagnosis, treatment, and prognosis.
CDSS empowers doctors, nurses, patients, caregivers, pharmacists and others to make more informed decisions to deliver effective care.
- Diagnostic support
- Informed decision making
- Medication therapy
- Actionable insights using real-time data and analytics
- Better self-management and early identification of adverse events
- Improved information acquisition on treatment and diagnosis
Use Case Examples
There are many AI-powered CDSS that serve as a guide to healthcare professionals.
• Analyse lifestyle, patient history, clinical and laboratory data to identify patients at risk of cardiovascular disease.
• Analyse patient clinical data to determine which over-the-counter side-effect, counter-interactions and allergy symptoms.
• Analyse heart rhythms to detect atrial fibrillation and other abnormalities and alert caregivers and clinicians, whenever appropriate.
• Google Brain AI CDSS analyzes images on the back of the eye to diagnose diabetic retinopathy and diabetic macular edema, which is the leading cause of blindness.
• The clinical decision support resource shows impact in reducing medical errors, the third-highest cause of U.S. deaths (Wolters Kluwer Health)
Use case development template
Here is the guidance to help you develop business and a high-value use case for clinical decision support systems using Artificial Intelligence and Machine Learning.
This CDSS use case framework guidance describes Esdha’s current research on the topic and should be viewed only as recommendations, unless specific regulatory or statutory requirements are cited.
Challenges & Opportunities
Data: CDSS relies on centralised, clinical data and real-time data sources leading to lack of inadequate supplies in hospital and real-time clinical decision support to healthcare professionals. For example, a study has shown that when pneumococcal vaccine inventories run out, it is not updated in CDSS.
Operational Impact: Poor data quality can affect the quality of decision support provided. There is a need for information standards such as ICD, SNOMED, and other sources.
Transportability and interoperability: with the diversity of of clinical data sources, system exists as stand-alone imposing greater challenges to implementation. Cloud infrastructure helps to reduce the interoperability issues.
System monitoring & maintenance: CDSS rely on real-time data and knowledge base. Healthcare institutions have reported difficulty in monitoring and maintaining the knowledge base, algorithms, rules and data.
Lack validity and human decision making: as users can become more reliant on CDSS without questioning the accuracy of the recommendation provided.
Disruptive alerts: CDSS alerts patients and healthcare professionals using alerts. Studies have found that up to 95% of alerts are inconsequential, leading to fatigue from alerts or distrust. For example, alert on duplicate medication for inflammatory bowel disease can be found inappropriate as the same drug can be applied through different administration routes for treatment.
Knowledge base: overall knowledge creation with the clear evidence base for incorporating CDSS is a challenge and requires specialist input from various care professionals.
Interdisciplinary team: We need an interdisciplinary team consisting of computer scientists, patients, nurses, caregivers and clinicians to align goals, requirements and clinical trial outcomes.
Accountability: CDSS gives rise to structures in which agency is shared – used by different healthcare professionals (nurses, pharmacists, clinicians, physicians) with reliance on CDSS for mutually intertwined and interdependent decisions. This rises the question as to ‘who is accountable or morally and legally answerable’ to adverse outcomes. There is a need for frameworks on medical malpractice liability for AI CDSS.
Cost: due to lack of standardised metrics, it is hard to do cost benefit assessment as cost-effectiveness depends on a range of socio-economic factors including environment, political and technological.
Potential risks & mitigation
Trustworthiness: different stakeholders have distinctive expectations which needs adequate risk-benefit analysis for building rules and outcome measures.
Explainability: might not be required or can be misleading as users may not follow CDSS recommendation. There is also risks associated with misunderstanding recommendations or wrongfully assume causality as explanations are correlation-based, they can be susceptible to error due to random factors. Explainability is useful as long as the outputs are sufficiently accurate, validated and required by the user.
Wrong or misleading recommendation: can result in loss of trust or serious consequences.
Privacy & quality: adherence to data protection and privacy requirements26 27 such as the general data protection regulation (GDPR) will be essential. A standardised approach to data collection can help to address this risk.
Bias, overfitting and validity: build a rigorous criterion to evaluate for biases (such as statistical misrepresentation to the general population), overfitting, and validity.