Healthcare organizations are currently flying blind. While 91% of AI deployment studies report technical performance, only 7% report on the patient experience (Liu et al., 2023). We know if the AI is accurate; we don't know if it’s human.

The Efficiency-Flourishing Gap

AI enables a dangerous new phenomenon: the ability to improve operational metrics while simultaneously degrading human welfare. An algorithm can flag a sepsis risk 15 minutes faster, but if the resulting workflow makes the patient feel like an object rather than a person, have we actually "improved" care?

The Flourishing Metric is a governance framework designed to bridge this gap. It operationalizes sovereignty by turning philosophical concepts into measurable data points.

The Five Pillars of Clinical Sovereignty

To move from theory to practice, organizations must audit AI based on these five dimensions:

• Pillar 1: Patient Autonomy. We must use modified "Perceived Control" scales. It’s not enough for the AI to be right; the patient must feel they have the agency to participate in the decision.
• Pillar 2: Preservation of Clinical Judgment. We must track "Automation Bias." Are clinicians simply rubber-stamping AI suggestions? We need to measure override patterns to ensure professional expertise isn't being hollowed out.
• Pillar 3: Relationship Quality. Using adapted "Trust in Physician" scales, we must ask: Is the screen a bridge or a barrier?
• Pillar 4: Algorithmic Equity. We need disaggregated data. If an algorithm is 90% accurate overall but only 60% accurate for a specific demographic, it isn't "working."
• Pillar 5: Transparency. We must move past "Black Box" medicine. Sovereignty requires that both the clinician and the patient understand the why behind a recommendation.

A Call for "Sovereign Governance"

Implementation requires more than just new surveys. It requires a structural shift in how we buy and deploy technology:

1. Procurement: Demand that vendors provide data on override patterns and demographic performance.
2. Triangulation: Don't just trust surveys. Look at behavioral data (e.g., how often is a clinician actually looking at the patient vs. the AI interface?).
3. External Validation: Just as we audit financials, we must audit flourishing to prevent "metric gaming."

Conclusion

The difference between a "Sovereign Clinician" and a "Data Entry Clerk" is the ability to exercise judgment within a relationship of trust. If our metrics don't protect that space, our technology will eventually destroy it.

The Flourishing Metric ensures that AI serves the people, not just the institutions.

Pillar 2 Isn't Theoretical — We Already Have the Data on Automation Bias

I wrote that Pillar 2 requires tracking automation bias so clinicians don't simply rubber-stamp AI suggestions. That risk isn't speculative. A 2024 empirical study of 210 participants using an AI-enabled wound-care decision support tool found that agreement with incorrect AI recommendations — the textbook definition of automation bias — was significantly reduced by diagnostic training, certified wound-care experience, and physician status, while a higher perceived benefit of the system actually increased false agreement. Automation Bias in AI-Decision Support, 2024 empirical study Read that last finding twice: the more a clinician trusts the tool's value, the more likely they are to accept a wrong answer from it. Non-specialists were the most susceptible. Automation bias findings on non-specialist susceptibility

That's exactly the mechanism the Flourishing Metric's Pillar 2 is designed to catch, and it tells you something concrete about how to operationalize it: don't just log override rates in aggregate. Segment them by specialty, training level, and tenure, because the research says junior and non-specialist staff carry disproportionate risk. If your override dashboard shows a flat 8% override rate hospital-wide, that number is hiding the units where automation bias is actually concentrated. Comprehensive diagnostic training before rollout isn't a nice-to-have — the study identifies it as the single most effective lever you have against this failure mode. Training as automation bias mitigation

Pillar 4's Warning Already Happened at Scale — Here's the Receipt

Pillar 4, Algorithmic Equity, isn't a hypothetical either. The most cited case study in healthcare AI bias involves a commercial risk-prediction algorithm used across the U.S. health system to guide care management decisions for millions of patients. Researchers found that at identical algorithm-assigned risk scores, Black patients were considerably sicker than White patients — measured by objective markers of uncontrolled illness — because the algorithm used healthcare cost as a proxy for health need. Obermeyer et al., "Dissecting racial bias in an algorithm used to manage the health of populations," Science Because less money is historically spent on Black patients with the same level of need, the algorithm learned to systematically underestimate their risk. The result: only 17.7% of Black patients who needed extra care were being flagged for it under the existing algorithm. Correcting the bias by removing cost as the target variable would have raised that to 46.5%. Bias correction results, Science

That's the disaggregated data Pillar 4 demands — not an aggregate accuracy number, but a demographic breakdown of who the algorithm actually helps. The uncomfortable finding in this study is that the algorithm looked accurate by conventional metrics while still being badly biased, because the metric being optimized (cost) was itself a flawed proxy for the outcome that mattered (health need). Cost as a flawed proxy for health need Any hospital procurement process that asks a vendor “what's your model's accuracy?” without asking “accuracy at predicting what, for whom?” is exposed to the exact same failure mode.

Pillar 5 Now Has a Regulatory Floor — Use It

Transparency, Pillar 5, stopped being purely aspirational in 2024. ONC's HTI-1 Final Rule established the first federal transparency requirements for AI and predictive algorithms embedded in certified health IT, giving clinical users a consistent, baseline set of information to assess an algorithm's fairness, appropriateness, validity, effectiveness, and safety. ONC HTI-1 Final Rule overview This matters because certified health IT underpins care at more than 96% of hospitals and 78% of office-based physicians nationwide — meaning this rule already touches nearly every clinical AI deployment most of you will encounter. Certified health IT coverage statistics

If you're building a Sovereign Governance procurement checklist, HTI-1 gives you regulatory language to demand, not just governance philosophy. Ask vendors for the same baseline information ONC now requires certified health IT to disclose about how their algorithm was developed, validated, and evaluated for bias — even for tools that fall outside HTI-1's technical scope. Regulatory minimums make a useful floor. The Flourishing Metric's job is to make sure your organization doesn't treat that floor as the ceiling.

Continue the Conversation

If this resonated, here is where to go next: Join the AI-in-Healthcare Workshop · Get the Books · Contact Dan