Evaluating AI in Healthcare: Implementing Approaches

By Sumona Bose

March 7, 2024


Artificial intelligence (AI) holds great potential in healthcare by enhancing clinical decision-making and patient outcomes. However, a significant gap exists between the development of AI models and their successful integration into clinical practice. Despite the proliferation of AI-based clinical decision support systems (AICDSS), only a meager 2% of these models progress beyond the prototyping stage, leaving the actual clinical impact largely unexplored.

Evaluating Clinical Value through Rigorous Trials

The evaluation of AICDSS through randomised controlled trials (RCTs) stands as a critical step in determining their true clinical value. While some RCTs have been conducted, their outcomes paint a nuanced picture. Although these trials showcase promising statistical performance of AI, nearly half of them fail to demonstrate improved patient outcomes. This discrepancy underscores the complexity of assessing AI solely based on quantitative metrics like accuracy. This may not capture the practical utility of these systems in real-world healthcare settings. Table 1 unpacks the definitions associated with interpreting the patient outcomes. This helps clinicians and researchers shift from the arbitrariness that hinders real-world settings.

Reported in N (%)
Implementation outcomeaClinical explanationImplementation stageRCTs (N = 64)Guidelinesb (N = 5)
AppropriatenessIs the AI compatible with the clinical workflow and is it useful?Early5 (8)0 (0)
AcceptabilityIs the AI acceptable, agreeable, or satisfactory for the users?Ongoing10 (16)0 (0)
FeasibilityCan the AI be successfully used as intended by the manufacturer?Early16 (25)0 (0)
AdoptionDo the users express the initial decision, or action to try or employ the AI?Ongoing6 (9)0 (0)
FidelityIs the AI implemented as intended by the manufacturer?Ongoing31 (48)0 (0)
Implementation costWhat is the cost impact of implementing the AI system?Late4 (6)0 (0)
PenetrationHas the AI been adopted by all groups of trained users?Late0 (0)0 (0)
SustainabilityIs the AI maintained within ongoing clinical operations over time?Late1 (2)0 (0)

Table 1: AI in RCTs, Definitions of implementation outcomes were adapted from the taxonomy of implementation outcomes by Proctor et al (2011). 

The Need for a Holistic Evaluation Approach

A comprehensive understanding of AI’s role in clinical practice necessitates a multi-faceted evaluation strategy. Current guidelines like Developmental and Exploratory Clinical Investigations of DEcision support systems driven by Artificial Intelligence (DECIDE-AI) and Consolidated Standards of Reporting Trials–Artificial Intelligence (CONSORT-AI), fall short in providing robust measures for assessing AI implementation success. To address this gap, a mixed-methods evaluation approach,  proves invaluable in dissecting the various dimensions of AICDSS implementation.

Bridging the Gap in Implementation Evaluation

Despite the increasing focus on RCTs evaluating AICDSS in clinical settings, a gap exists in the comprehensive evaluation of implementation outcomes.  While metrics like ‘fidelity’ are commonly reported using quantitative measures, aspects such as ‘acceptability’ and ‘appropriateness’ that demand qualitative scrutiny are often overlooked. This imbalance underscores the need for a more holistic approach towards evaluating the implementation of AICDSS, encompassing factors beyond statistical performance. Figure 1 reiterates the comprehensive value of integrating implementation outcomes in AI in healthcare, revealing an innovative future in the field.

figure 2
Figure 1: a In the current situation, AI-CDSS, are clinically deployed, after going through multiple preclinical validations (e.g., external and temporal algorithm validation) to assess their clinical utility and effectiveness. b To enhance comprehension of factors that contributed to successful implementation or failure at the bedside, implementation outcomes should be systematically integrated in future clinical trials evaluating AICDSS in real-world clinical settings. *Implementation outcomes as described by Proctor et al.


While the efficacy of AICDSS in healthcare settings is crucial, understanding the contextual nuances is imperative. Enhanced systematic reporting of implementation outcomes alongside effectiveness metrics can bridge the existing gap in comprehensively assessing the impact of clinical AI. Embracing an inclusive evaluation framework will not only validate the effectiveness of AICDSS but also shed light on the intricate interplay between AI technology and healthcare delivery.

Reference url

Recent Posts

Epidemic intelligence

How User-Centered Perspectives Foster Innovation in Digital Health Surveillance for Epidemic Intelligence in Europe

🌍🔬European Epidemic Intelligence: Overcoming Challenges in Data Reporting and Surveillance 📊🦠

Effective epidemic intelligence is crucial for managing public health threats. Discover the challenges faced by practitioners and the proposed improvements to enhance data reporting and surveillance systems. Let’s work together to build a healthier future! 💪

#EpidemicIntelligence #PublicHealth #DataReporting #Surveillance #HealthcareInnovation #AIConsulting #LifeSciences #ValueBasedHealthcare #MarketAccess

Join the conversation and share your thoughts on improving epidemic intelligence! 💬

Quality-Adjusted Life Year (QALY)

Quality-Adjusted Life Year (QALY) in Modern Healthcare Decision-Making

🔍 Understanding QALYs in Healthcare Decision-Making 🔍

The recent passage of HR 485 has sparked a debate on the use of Quality-Adjusted Life Years (QALYs) in healthcare. QALYs are crucial for measuring the effectiveness of medical treatments, ensuring fair treatment for all. Learn about their importance and the potential implications of HR 485.

#Healthcare #MedicalResearch #QALYs #HealthPolicy #HealthEconomics #MedicalInnovation


Patient Empowerment in Their Pockets: AI Chatbots in Healthcare

🧠 Unlocking the Power of AI: Chatbots Simplifying Pathology Reports 📊 Explore how artificial intelligence is transforming patient understanding of complex medical data. Learn more about the potential impact on healthcare communication and accessibility. #HealthTech #PatientCare #AIInnovation

When you partner with Syenza, it’s like a Nuclear Fusion.

Our expertise are combined with yours, and we contribute clinical expertise and advanced degrees in health policy, health economics, systems analysis, public finance, business, and project management. You’ll also feel our high-impact global and local perspectives with cultural intelligence.



1950 W. Corporate Way, Suite 95478
Anaheim, CA 92801, USA



© 2024 Syenza™. All rights reserved.