AI for healthcare holds immense value for improving the patient outcomes. One key factor in this is the Clinical Decision Support Systems (CDSS). But how does one integrate the AI tools into complex healthcare workflows?

The CDSS adoption in healthcare happens through a series of progresses levels incrementally. Each level signifies an increase in the level of automation and decision-making support offered to clinicians. Each level builds upon the previous, enhancing capabilities and ensuring a safe transition from human-driven to AI-assisted and ultimately autonomous clinical support.

For AI enthusiasts, EHR system builders, and healthcare professionals, understanding these levels is crucial for developing, implementing, and leveraging CDSS effectively. Let’s look into what these levels are.

Navigating the Levels of CDSS

Level 0: No Automation

At the foundational level, there is no automation. Healthcare providers rely entirely on traditional methods, such as textbooks, guidelines, and personal experience, to make clinical decisions. This level serves as a baseline, highlighting the absence of CDSS and the full dependency on human expertise and manual processes.

Examples:

• • A doctor manually checks a patient’s medical history and current medication list by flipping through paper records.
  • A nurse follows a standard printed protocol for wound care without any electronic reminders or alerts.
  • A clinician consults a pharmacology textbook to verify potential drug interactions.

Level 1: Rule-Based Assistance

The first step towards automation involves rule-based assistance. These systems use predefined if-then rules to offer basic guidance and alerts. For example, they can identify potential drug interactions or contraindications based on a set of established guidelines. This level introduces consistency and efficiency, reducing human error by providing straightforward, rule-based support.

At this level, CDSS provides basic rule-based assistance. These systems utilize predefined if-then rules to offer simple guidance, such as identifying drug interactions or contraindications.

Examples:

• • A clinician receives an alert from the CDSS when prescribing a new medication that could interact with an existing one.
  • The system highlights potential allergic reactions based on a patient’s known allergies.
  • Basic reminders for routine vaccinations or screenings based on patient age and gender.

Level 2: Advanced Decision Support

Building on rule-based assistance, Level 2 systems provide more sophisticated support by integrating patient-specific data and offering contextual alerts and reminders. They also incorporate evidence-based recommendations into their functionality.

Advanced decision support systems integrate evidence-based recommendations into their functionality. By leveraging up-to-date clinical guidelines and research, these systems offer more sophisticated advice on diagnostic tests, treatment options, and care plans. They merge patient-specific data with clinical evidence to enhance decision-making processes.

Examples:

• • A provider receives a reminder to order a mammogram for a patient based on her age, medical history, and family history of breast cancer.
  • The system suggests appropriate diagnostic tests and treatments based on current clinical guidelines and the patient’s specific symptoms.
  • Alerts for preventive care measures tailored to individual patient profiles, such as reminders for diabetes screening in high-risk patients.

Level 3: Predictive and Prescriptive Analytics

At this level, CDSS leverages predictive and prescriptive analytics to provide deeper insights. These systems can predict patient outcomes and suggest optimal interventions, dynamically adjusting recommendations based on real-time data and patient feedback. These systems predict patient outcomes for a given treatment plan and prescribe specific actions to optimize care. They analyze vast datasets to identify patterns and make proactive recommendations, significantly enhancing clinical foresight and decision-making accuracy.

Examples:

• • The system predicts a patient’s risk of developing sepsis based on real-time monitoring of vital signs and lab results, prompting early intervention.
  • Personalized medication dosing recommendations based on genetic data and patient response history.
  • Real-time integration with wearable devices to monitor chronic conditions and adjust treatment plans dynamically.

Level 4: Full Automation with Human Oversight

Systems at this level operate autonomously for most routine clinical scenarios, while complex or ambiguous cases still require human oversight. These CDSS can independently manage straightforward cases, allowing healthcare professionals to focus on more intricate and critical situations. Human validation ensures safety and reliability.

Level 4 systems are capable of managing routine clinical cases autonomously while requiring human oversight for complex or ambiguous situations. These systems significantly enhance efficiency by allowing clinicians to focus on more intricate cases.

Examples:

• • The CDSS autonomously manages routine follow-ups for chronic disease patients, including ordering necessary tests and adjusting medications, with clinicians reviewing and approving recommendations.
  • Automatic scheduling of preventive care appointments based on patient risk profiles and historical data.
  • Management of post-operative care, with the system autonomously monitoring recovery progress and flagging any deviations for clinician review.

Level 5: Complete Autonomy

At the highest level, CDSS achieve complete autonomy, managing all clinical decisions without human intervention. These systems are designed to handle routine, complex, and rare conditions independently, ensuring consistent and efficient care delivery across all scenarios.

Complete autonomy is pinnacle of CDSS AI capability, where the AI system manages all clinical decisions without needing human intervention. However, this level requires rigorous validation and acceptance to ensure trust and reliability in fully autonomous clinical decision-making.

Examples:

• • The system autonomously diagnoses and treats a wide range of conditions, continuously updating treatment plans based on patient progress and emerging medical knowledge.
  • Emergency care automation where the CDSS handles triage, diagnostics, and initial treatment without human input, ensuring rapid and standardized response.
  • Comprehensive management of inpatient care, from admission to discharge, including real-time adjustments to treatment plans and automatic documentation of all clinical decisions and actions.

The progression of Clinical Decision Support Systems from basic rule-based assistance to complete AI autonomy represents a transformative journey in healthcare. Each level builds upon the previous, enhancing capabilities and ensuring a gradual, safe transition from manual processes to sophisticated, AI-driven support.

The feature matrix of CDSS software levels

The below table presents the feature matrix displaying the progression of AI capabilities in the CDSS software.

Feature	Level 0: No Automation	Level 1: Rule-Based Assistance	Level 2: Advanced Decision Support	Level 3: Predictive and Prescriptive Analytics	Level 4: Full Automation with Human Oversight	Level 5: Complete Autonomy
Basic Rule-Based Guidance	✗	✓	✓	✓	✓	✓
Contextual Alerts and Reminders	✗	✗	✓	✓	✓	✓
Patient-Specific Data Integration	✗	✗	✓	✓	✓	✓
Evidence-Based Recommendations	✗	✗	✓	✓	✓	✓
Predictive Analytics	✗	✗	✗	✓	✓	✓
Prescriptive Analytics	✗	✗	✗	✓	✓	✓
Real-Time Data Integration	✗	✗	✗	✓	✓	✓
Dynamic Adjustment Based on Feedback	✗	✗	✗	✓	✓	✓
Autonomous Routine Case Management	✗	✗	✗	✗	✓	✓
Complex Case Management with Human Oversight	✗	✗	✗	✗	✓	✓
Full Autonomous Management (All Scenarios)	✗	✗	✗	✗	✗	✓

• Level 0: No Automation

  • Features Present: None.
  • Features Absent: All features are absent as this level represents no automation.

• Level 1: Rule-Based Assistance

  • Features Present: Basic Rule-Based Guidance.
  • Features Absent: Contextual alerts, patient-specific data integration, evidence-based recommendations, predictive and prescriptive analytics, real-time data integration, dynamic adjustment, autonomous management.

• Level 2: Advanced Decision Support

  • Features Present: Basic Rule-Based Guidance, Contextual Alerts and Reminders, Patient-Specific Data Integration, Evidence-Based Recommendations.
  • Features Absent: Predictive and prescriptive analytics, real-time data integration, dynamic adjustment, autonomous management.

• Level 3: Predictive and Prescriptive Analytics

  • Features Present: Basic Rule-Based Guidance, Contextual Alerts and Reminders, Patient-Specific Data Integration, Evidence-Based Recommendations, Predictive Analytics, Prescriptive Analytics, Real-Time Data Integration, Dynamic Adjustment Based on Feedback.
  • Features Absent: Autonomous management.

• Level 4: Full Automation with Human Oversight

  • Features Present: Basic Rule-Based Guidance, Contextual Alerts and Reminders, Patient-Specific Data Integration, Evidence-Based Recommendations, Predictive Analytics, Prescriptive Analytics, Real-Time Data Integration, Dynamic Adjustment Based on Feedback, Autonomous Routine Case Management, Complex Case Management with Human Oversight.
  • Features Absent: Full autonomous management in all scenarios.

• Level 5: Complete Autonomy

  • Features Present: All features.
  • Features Absent: None. This level represents complete autonomy in all clinical scenarios.

Defining the user-stories for CDSS levels as Product Manager

From a product manager perspective who is trying to develop a CDSS software how does the personas and user-stories look like? Below I present a rough overview of how one can define the AI capabilities for an EHR system at each level of complexity.

Level 1: Rule-Based Assistance

Feature: Basic Rule-Based Guidance

Persona: Clinicians
User Story: As a clinician, I want the system to provide basic rule-based guidance to help me quickly identify drug interactions and contraindications, ensuring patient safety.

Level 2: Advanced Decision Support

Feature: Contextual Alerts and Reminders

Persona: Providers
User Story: As a provider, I want the system to provide contextual alerts and reminders based on patient data to assist me in preventive care and early intervention for chronic conditions.

Feature: Patient-Specific Data Integration

Persona: Clinicians
User Story: As a clinician, I want the system to integrate patient-specific data from electronic health records (EHR) to provide personalized treatment recommendations and care plans.

Feature: Evidence-Based Recommendations

Persona: Providers
User Story: As a provider, I want the system to provide evidence-based recommendations for diagnostic tests and treatment options, improving clinical decision-making and patient outcomes.

Level 3: Predictive and Prescriptive Analytics

Feature: Predictive Analytics

Persona: Payers
User Story: As a payer, I want the system to use predictive analytics to identify high-risk patients for certain conditions, allowing for proactive care management and cost-effective interventions.

Feature: Prescriptive Analytics

Persona: Clinicians
User Story: As a clinician, I want the system to use prescriptive analytics to optimize medication dosages and suggest personalized treatment plans, improving patient adherence and outcomes.

Feature: Real-Time Data Integration

Persona: Providers
User Story: As a provider, I want the system to integrate real-time patient data from monitoring devices to adjust treatment plans and interventions dynamically, ensuring timely and effective care.

Feature: Dynamic Adjustment Based on Feedback

Persona: Clinicians
User Story: As a clinician, I want the system to dynamically adjust treatment plans based on real-time feedback and patient responses, optimizing care delivery and patient satisfaction.

Level 4: Full Automation with Human Oversight

Feature: Autonomous Routine Case Management

Persona: Providers
User Story: As a provider, I want the system to autonomously manage routine cases, allowing me to focus on complex or high-priority patients that require my expertise and intervention.

Feature: Complex Case Management with Human Oversight

Persona: Clinicians
User Story: As a clinician, I want the system to assist me in managing complex cases by providing recommendations and insights, while still allowing me to review and validate decisions for patient safety.

Level 5: Complete Autonomy

Feature: Full Autonomous Management (All Scenarios)

Persona: Providers
User Story: As a provider, I want the system to operate autonomously across all clinical scenarios, managing routine and complex cases without human intervention, ensuring consistent and efficient care delivery.

Conclusions

By adopting CDSS at progressive levels, healthcare can harness the power of AI. However, there are few ethical and safe implementation considerations that need to be noted for improved patient care.

• Ethical considerations: As CDSS becomes more autonomous, ensuring patient privacy, data security, and fairness in algorithms is paramount.
• Human-AI collaboration: Clinician training and a focus on human-AI collaboration are crucial to optimize the benefits of CDSS.

The CDSS AI progression levels presented above strikes a balance between capturing key milestones in CDSS evolution and avoiding unnecessary complexity. Levels 1-3 represent a gradual increase in the sophistication of decision support. Levels 4 and 5 highlight the critical shift towards shared and then autonomous decision-making.

If you are looking to adapt AI for your business processes or services, I can help you with the product planning, roadmap, architecture, development and end-to-end delivery. If you would like to know more would be happy to start with a free consultation session. Leave a message or connect on LinkedIn.