Table 3. Barriers and strategies in the implementing phase of AI implementation.
| Barriers for implementing |
|---|
|
Buy-in • Clinicians’ fatigue, alert fatigue, false positives, wrong assumptions and lack of domain knowledge seen in the model; Lack of technical quality and incorrect physical space of using the system, poor usability; Not seeing value in AI. Workflow • Lack of guidance on how to act in a new workflow–non-uniform usage; Non-interoperable systems. Finance and human resources • Underinvestment and lack of physical infrastructure to bring benefits of AI. |
| Strategies for implementing |
|
Leadership • Establishing follow-up procedures for adherence and utilization during pilot. Buy-in • Clinicians decide on model performance parameters; Using local data and local experiences; Starting with a technically simpler project; Showing progress in solving a clinical problem; Arranging a peer-review of the AI model; Possibility for clinicians to try the model; Conveying value through communication using good-quality evidence, thoughtfully framing messages about AI system, exposing strengths and weaknesses, indications and contraindications, and clinicians’ responsibilities; Assurance for snoozing or turning-off the AI system if not satisfactory. Change management • Informal communication helping to convey value of AI outputs; Building communication channels and strengthening relationships. Engagement • Common meetings to understand problem, product and workflow requirements, to create a shared vision; Regular meetings between healthcare professionals and AI developers; Involving healthcare professionals in training AI model and user experience formation. Workflow • Determining a moment when AI system could be most useful; Designing a prototype of the new workflow and surrounding technical infrastructure–prioritizing less disruption; Estimating resources; Defining new roles and responsibilities; Iterating prototypes of workflow and infrastructure with relevant staff; Testing new workflow and collecting data on performance and usability; Customizing AI system’s interface to match the workflow; Arranging regular meetings with leadership and stakeholders to align on unified vision of the workflow. Finance and human resources • Performing fundraising or arranging public-private partnerships for ensuring possibilities for actual use of the system. Training • Doing training formally and informally; Focusing the training on how to understand the outputs of the AI system and how to act in the new workflows; Addressing ‘black box’ issue; Including different stakeholders in preparing training materials and reviewing them; Training clinicians to take on a trainer’s role from technical to empathy knowledge and skills; Creating social media channels for sharing knowledge. Evaluation and monitoring • Having a trial period; Taking baseline measurements; Collecting quantitative and qualitative data on user experience; Assessing clinicians’ adherence; Identifying most useful components of AI system. |