Dr. Erkan Hassan

Monthly Writings

Evaluations and reviews of the latest in the field.

5 Strategies for Successful Artificial Intelligence Implementation

SUMMARY:

  • The status of most Artificial Intelligence (AI) implementations are promising but immature

  • Integration of AI systems into clinical settings is challenging.

  • These strategies are designed to improve the chances of AI implementation

REVIEW

  • AI offers the ability to evaluate large amounts of heterogeneous real world data developing clinically relevant models with the goal of providing more effective care.

  • AI implementations have been difficult due to numerous challenges including:

    • The complexity of the healthcare system;

    • Developing appropriate AI training models;

    • Lack of algorithm transparency;

    • Dissatisfaction among clinical users;

    • Etc.

  • The following principles are not meant to be exhaustive, but to give the reader the critically needed areas to be addressed

Strategies for Successful AI Implementation

Strategy #1:  Identify the Problem

  • What clinical problem are you trying to solve with AI?

    • Define the clinical benefit desired

  • Who are you building the system for?

  • Do validated AI tools exist that address the problem?

 

Strategy #2: Rigorous Model Validation

  • Outcomes should be clinically meaningful

  • Good predictive performance does not prove a beneficial decision making impact.

  • Evaluate Data Quality:

    • Accuracy

    • Completeness

    • Consistency

    • Reliability

    • Up-to-date

 

Strategy #3: Evaluate AI Tools

  • Use of Data

    • AI tools should identify specific risks and be used to provide tailored intervention(s).

    • Eliminate healthcare disparities among groups.

    • Mitigate system biases.

  • Transparency

    • Sufficient sample size for development and validation .

    • Transparent reproducibility of results.

    • Internal logic content disclosed

  • Performance

    • Interpretability of results.

    • Evaluate accuracy, fairness, risk of overdiagnosis or overtreatment, auditability.

  • Infrastructure requirements:

    • What technology is necessary to use the AI tool

    • Does the technology integrate with existing systems?

  • Financial Implications

    • Cost of tool

    • Impact on revenue and cost of care

    • Avoid directing resources from high priority areas.

 

Strategy #4: Implementation of AI Tools

  • Training

    • Promote a continuous learning environment

    • All technology will need some degree of education

    • All interventions should be designed with regular evaluation

  • Dissemination via Change Management strategies

  • Seamless workflow integration (clinical and administrative workflows)

    • How are issues/errors identified and reported?

  • Technology support

  • Bias Risk

    • AI tools should be easily tailored to the local population

    • Will need to perform local validation

    • Does the tool include social and environmental determinants of health?

 

Strategy #5: Manage AI Tools

  • How will tool be maintained (updates and data corrections)?

  • How will AI tool be monitored for success and clinical outcomes

  • Return on investment

  • Monitor performance continuously

CONCLUSIONS:

  • A thoughtful and integrated approach to AI implementation must occur in order to gain  maximum value.

  • The basic strategies outlined will aid in standardizing AI benefits and help clinicians work more effectively.

Erkan Hassan