Chapter 12: The Clinical Bond Index and Structural Healthcare Injustice

From the General Bond Index to the Clinical Instantiation

The Bond Index was introduced in Geometric Ethics (Bond, 2026, Ch. 16) as a general measure of the moral cost imposed by institutional policy on a population. The clinical Bond Index is the domain-specific instantiation for healthcare.

Definition 12.1 (Clinical Bond Index). The clinical Bond Index \text{BI}(P, S) of an institutional policy P with respect to a patient population S is the expected deviation between the clinical geodesic on the full manifold and the path mandated by the policy:

\text{BI}(P, S) = \mathbb{E}_{s \in S}\left[\text{CF}(\gamma_P^*(s)) - \text{CF}(\gamma_{\mathcal{C}}^*(s))\right]

where: - \gamma_P^*(s) is the path mandated by policy P for patient s — the treatment trajectory that institutional protocols, insurance requirements, and standard operating procedures prescribe. - \gamma_{\mathcal{C}}^*(s) is the clinical geodesic for patient s on the full nine-dimensional manifold — the minimum-cost path that a perfectly calibrated clinician with unlimited resources would follow. - \text{CF}(\gamma) is the clinical friction along path \gamma — the total edge-weight cost including both Mahalanobis distance and boundary penalties. - The expectation is over the patient population S.

Properties of the Bond Index

Proposition 12.1 (Properties of the Clinical Bond Index).

1. Non-negativity: \text{BI}(P, S) \geq 0 for all policies P and populations S. The clinical geodesic is, by definition, the minimum-cost path; any policy that deviates from it incurs additional cost.

2. Zero iff policy is manifold-optimal: \text{BI}(P, S) = 0 if and only if the policy-mandated path coincides with the clinical geodesic for every patient in S.

3. Population decomposability: \text{BI}(P, S) = \sum_{S_k} \pi_k \cdot \text{BI}(P, S_k), where \{S_k\} is a partition of S and \pi_k = |S_k| / |S| is the proportion of population S in subgroup S_k. This allows decomposition of the total Bond Index into contributions from demographic subgroups.

4. Policy monotonicity: If policy P_1 is a refinement of policy P_2 (every path mandated by P_1 is also available under P_2, but P_1 additionally mandates paths that P_2 does not), then \text{BI}(P_1, S) \geq \text{BI}(P_2, S). More restrictive policies have higher Bond Indices.

The Disparity Ratio

The Bond Index for a single population tells us how far the policy deviates from the geodesic. But structural injustice is comparative: what matters is not just how far a policy deviates from the geodesic for one population but how much more it deviates for one population than another.

Definition 12.2 (Bond Index Disparity Ratio). The disparity ratio of policy P between populations S_1 and S_2 is:

\text{DR}(P, S_1, S_2) = \frac{\text{BI}(P, S_1)}{\text{BI}(P, S_2)}

A disparity ratio of 1.0 indicates equal manifold cost. A disparity ratio of 2.3 indicates that population S_1 bears 2.3 times the manifold cost of population S_2 under policy P.

The disparity ratio is the clinical Bond Index’s primary diagnostic output. It transforms “this policy may be unjust” into “this policy imposes 2.3 times the manifold cost on Black patients compared to white patients” — a specific, testable, actionable claim.

Racial Disparities

The most documented form of structural healthcare injustice in the United States is racial disparity. Black Americans experience worse health outcomes than white Americans across virtually every disease category, even after controlling for income, education, and insurance status. The causes are multifactorial — historical institutional betrayal, residential segregation, implicit clinician bias, differential access to care — but the geometric framework unifies them under a single measure.

How racial disparities appear on the manifold:

Black patients in the United States have, on average, systematically different attribute vectors on several dimensions:

• d_5 (Trust): Lower, due to the documented legacy of medical experimentation without consent (Tuskegee syphilis study, Henrietta Lacks, J. Marion Sims), discriminatory treatment in healthcare settings, and ongoing experiences of bias. This is not a personal trait — it is a population-level consequence of institutional history.

• d_9 (Epistemic status): Lower for specific subpopulations due to health literacy disparities, language barriers, and healthcare communication that does not account for cultural context.

• d_6 (Social and relational impact): Different (not necessarily lower) due to different family structures, different community health resources, and different social determinants of health.

When a healthcare policy assumes uniform d_5 across all populations (as most policies do), it systematically underestimates the manifold cost for populations with lower trust. The clinical geodesic for a Black patient with low d_5 is different from the geodesic for a white patient with high d_5 — it requires more trust-building, more transparent communication, more culturally concordant care. A policy that mandates the same path for both populations is manifold-suboptimal for the low-trust population and approximately optimal for the high-trust population.

\text{BI}(P, S_{\text{Black}}) - \text{BI}(P, S_{\text{White}}) = \mathbb{E}\left[\text{CF}(\gamma_P^* \mid d_5^{\text{low}}) - \text{CF}(\gamma_P^* \mid d_5^{\text{high}})\right] + \mathbb{E}\left[\text{CF}(\gamma_{\mathcal{C}}^* \mid d_5^{\text{high}}) - \text{CF}(\gamma_{\mathcal{C}}^* \mid d_5^{\text{low}})\right]

The first term captures the additional cost of the policy-mandated path for low-trust patients. The second term captures the fact that the geodesic for low-trust patients is already more expensive (trust-building requires effort). The sum is always positive: the policy imposes more additional cost on the population that already faces more cost.

Disability Discrimination

QALY-based allocation assigns lower quality weights to disabled patients. The geometric framework explains exactly why this is discriminatory and how to measure the discrimination.

A disabled patient’s attribute vector may have lower d_1 (functional status) but intact or even elevated d_4 (autonomy — many disabled people have fought hard for self-determination), d_7 (dignity — disabled identity is a positive identity for many), and d_5 (trust — disability communities often have strong peer support networks).

A QALY-based policy weights d_1 heavily and d_4, d_7 negligibly. The policy-mandated path for a disabled patient — which deprioritizes them in allocation decisions because their “quality of life” is lower — deviates from the geodesic on the full manifold, which would weight d_4 and d_7 according to the patient’s own dimensional weights.

\text{BI}(P_{\text{QALY}}, S_{\text{disabled}}) > \text{BI}(P_{\text{QALY}}, S_{\text{non-disabled}})

The disability discrimination is mathematical, not moral (Corollary 1 in the GCE paper). It follows from the dimensional collapse: any scalar projection that weights d_1 most heavily will disadvantage populations whose needs are concentrated on other dimensions.

Age Discrimination

Age-based triage — prioritizing younger patients because they have more expected life-years — is standard practice in utilitarian allocation frameworks. The geometric framework identifies the d_3 (justice) cost that the utilitarian framework ignores.

An elderly patient has, by hypothesis, fewer remaining life-years (d_1). But the elderly patient has equal moral worth (d_3), intact dignity (d_7), and specific vulnerabilities (lower d_5 if they have experienced ageist dismissal in healthcare, lower d_9 if cognitive decline affects understanding).

The Bond Index for age-based triage:

\text{BI}(P_{\text{age}}, S_{\text{elderly}}) = \mathbb{E}\left[\beta_{\text{justice}} \cdot \mathbf{1}[\text{age used as deprioritization criterion}] + \Delta d_7 + \Delta d_5\right]

This is nonzero whenever age is used as a triage criterion, because the d_3 boundary is crossed. The QALY framework does not see this cost — the QALY assigns lower value to elderly patients by construction (fewer life-years remaining). The Bond Index measures the manifold cost of that construction.

The Measurement Challenge

The Bond Index is defined in terms of clinical friction along paths on the nine-dimensional manifold. Computing it requires:

1. The policy-mandated path \gamma_P^* for each patient — derivable from institutional protocols, insurance requirements, and standard operating procedures.
2. The clinical geodesic \gamma_{\mathcal{C}}^* for each patient — the minimum-cost path on the full manifold, which requires the 9 \times 9 covariance matrix \Sigma and the boundary penalties \beta_k.
3. The edge weights w(v_i, v_j) along both paths — computed from the Mahalanobis distance and boundary penalties.

The policy-mandated path is observable: it is what actually happens to patients under the policy. The clinical geodesic is a counterfactual: it is what would happen if the policy were manifold-optimal. The difference between the observed path and the counterfactual geodesic is the Bond Index.

Which Dimensions Are Already Measured?

Six of the nine clinical-moral dimensions are routinely captured in electronic health records and clinical quality registries:

Validated Proxies for the Missing Dimensions

Three dimensions require validated proxies — existing instruments that measure constructs corresponding to the manifold dimensions:

d_5 (Trust): - Wake Forest Trust in Physician Scale (Hall et al., 2002): A 10-item instrument measuring fidelity, competence, honesty, confidentiality, and global trust. Validated across demographic groups. - HCAHPS Communication Composite (CMS): “How often did doctors explain things in a way you could understand?” and related items. Available for all US hospitals as part of Medicare quality reporting. - Interpersonal Processes of Care Survey (Stewart et al., 2011): Captures patient perceptions of discrimination, disrespect, and communication quality. Directly relevant to d_5 for minority populations.

d_7 (Dignity): - Patient Dignity Inventory (Chochinov et al., 2008): A 25-item instrument measuring dignity-related distress, originally validated in palliative care populations. - HCAHPS “Treated with Respect” Item (CMS): “During this hospital stay, how often did doctors treat you with courtesy and respect?” Available for all US hospitals. - Person-Centered Care Assessment Tool (Edvardsson et al., 2010): Measures the degree to which care respects patient personhood, autonomy, and identity.

d_9 (Epistemic Status): - REALM (Rapid Estimate of Adult Literacy in Medicine; Davis et al., 1993): A word-recognition test that estimates health literacy in under three minutes. - NVS (Newest Vital Sign; Weiss et al., 2005): A nutrition-label-based assessment of health literacy and numeracy. - Gauge-invariance test (from Chapter 8): Present the same clinical information in two framings (e.g., “90% survival” vs. “10% mortality”) and measure decision consistency. Framing-variant patients have inadequate d_9 calibration.

The Estimation Procedure

Step 1: Retrospective path reconstruction. For a sample of patients in population S, reconstruct the clinical path from EHR data: diagnosis, treatments received, consultations, outcomes. Code each path as a sequence of vertices and edges on the clinical decision complex.

Step 2: Compute policy-mandated path cost. For each patient, compute the clinical friction along the observed path:

\text{CF}(\gamma_P^*(s)) = \sum_{(v_i, v_{i+1}) \in \gamma_P^*(s)} w(v_i, v_{i+1})

using edge weights estimated from the nine-dimensional attribute vectors and the covariance matrix \Sigma.

Step 3: Estimate the counterfactual geodesic. For each patient, compute the clinical geodesic using A* search on the full manifold — the minimum-cost path that accounts for all nine dimensions and all boundary constraints.

Step 4: Compute the Bond Index. Average the difference across the population:

\text{BI}(P, S) = \frac{1}{|S|} \sum_{s \in S} \left[\text{CF}(\gamma_P^*(s)) - \text{CF}(\gamma_{\mathcal{C}}^*(s))\right]

Step 5: Stratify by demographics. Compute \text{BI}(P, S_k) for each demographic subgroup S_k and compute the disparity ratio \text{DR}(P, S_k, S_{\text{ref}}) relative to a reference population.

The Policy

A large academic medical center’s COVID-19 crisis standards of care, implemented during the pandemic surge. The policy uses a modified SOFA score (Sequential Organ Failure Assessment) supplemented by comorbidity adjustment and life-expectancy estimation. Patients are ranked by expected clinical benefit (d_1) and allocated ICU beds in rank order.

The Bond Index Computation

Population: 847 patients evaluated for ICU admission during the surge period.

Demographic stratification: White (n = 412), Black (n = 198), Hispanic (n = 156), Other (n = 81).

Results:

Dimensional decomposition of the Black-White disparity:

The Bond Index disparity between Black and White populations decomposes by dimension:

The largest single contributor to the disparity is d_5 (trust): 1.9 of the 4.2-unit disparity. The policy assumes that all patients have equal trust in the institution, but Black patients — due to documented historical and ongoing experiences of medical racism — have systematically lower trust. The policy-mandated path does not account for this, resulting in higher manifold cost for Black patients.

What the Bond Index Reveals

The triage policy is not explicitly racist. It does not use race as a triage criterion. It uses clinical prognosis (d_1) supplemented by comorbidity adjustment. But the Bond Index reveals that the policy is structurally unjust: it imposes 2.31 times the manifold cost on Black patients.

The injustice is structural because it arises from dimensional omission — the policy operates on d_1 and ignores d_5, d_7, and d_9. The omission is not intentional; it is the inevitable consequence of scalar optimization on a multi-dimensional manifold.

The Bond Index does not prescribe a specific remedy. It diagnoses the problem and quantifies its magnitude. Possible remedies include:

1. Trust-building interventions for high-BI populations: Culturally concordant communication, community health worker support, transparent explanation of the triage process.
2. Modified triage criteria: Supplement SOFA scoring with trust-adjusted communication protocols that ensure all patients understand the triage process equally.
3. Retrospective audit: Compute the Bond Index quarterly and track trends over time. A declining disparity ratio indicates progress; a stable or increasing ratio indicates persistent structural injustice.

Insurance Policy Evaluation

Health insurance policies mandate specific clinical paths — coverage rules determine which treatments are available, which require prior authorization, and which are denied. The Bond Index can evaluate insurance policies for structural injustice:

\text{BI}(P_{\text{insurance}}, S) = \mathbb{E}\left[\text{CF}(\gamma_{\text{covered}}^*(s)) - \text{CF}(\gamma_{\mathcal{C}}^*(s))\right]

If the insurance policy’s coverage limitations force patients in population S onto paths that deviate substantially from the clinical geodesic — because the geodesic includes treatments that are not covered — the Bond Index quantifies the deviation.

Example: A Medicaid policy that does not cover certain medications forces patients onto alternative paths (different medications, delayed treatment, emergency department visits instead of primary care). If the alternative path has higher manifold cost, the Bond Index captures the disparity between Medicaid patients and privately insured patients following the geodesic.

Clinical Guideline Evaluation

Clinical practice guidelines — evidence-based recommendations for the treatment of specific conditions — can be evaluated by the Bond Index. Most guidelines are developed from clinical trial populations that are disproportionately white, male, and middle-class. The guidelines’ recommendations may be manifold-optimal for the trial population but not for populations with different attribute vectors on d_5, d_6, d_7, or d_9.

\text{BI}(P_{\text{guideline}}, S) > \text{BI}(P_{\text{guideline}}, S_{\text{trial}})

when the guideline population differs from the clinical population on dimensions that the guideline does not account for.

Algorithmic Bias Detection

Medical AI systems — clinical decision support, risk prediction, diagnostic algorithms — can be evaluated by the Bond Index for algorithmic bias. An AI system that produces different recommendations based on race, zip code, or insurance status (when these features correlate with but do not cause the relevant clinical differences) is violating the medical Bond Invariance Principle.

\text{BI}(P_{\text{AI}}, S_k) \text{ for all relevant populations } S_k

identifies demographic subgroups for whom the algorithm imposes systematically higher manifold cost. The Bond Index provides a quantitative measure of algorithmic injustice that goes beyond simple disparity statistics (which measure d_1 disparities but not manifold disparities).

The Bond Index as a Time Series

The Bond Index computed at a single point in time is a snapshot. Its real power is longitudinal: computed quarterly or annually, the Bond Index becomes a time series that tracks institutional equity over time.

Definition 12.3 (Bond Index Trajectory). The Bond Index trajectory \text{BI}(P, S, t) is the Bond Index computed at time t for policy P and population S. The trajectory reveals whether institutional equity is improving, stable, or declining.

\frac{d}{dt}\text{BI}(P, S, t) < 0 \quad \Rightarrow \quad \text{Equity improving (manifold cost decreasing)} \frac{d}{dt}\text{BI}(P, S, t) = 0 \quad \Rightarrow \quad \text{Equity stable} \frac{d}{dt}\text{BI}(P, S, t) > 0 \quad \Rightarrow \quad \text{Equity declining (manifold cost increasing)}

The trajectory has diagnostic value: it reveals whether a policy intervention (e.g., implementation of a trust-building program) actually reduces the Bond Index for the target population.

The Counterfactual Problem

Computing the Bond Index requires estimating the clinical geodesic \gamma_{\mathcal{C}}^*(s) — the path that a perfectly calibrated clinician with unlimited resources would follow. This is a counterfactual: it describes what would happen under conditions that do not exist.

The counterfactual can be estimated in several ways:

1. Expert panel consensus. A panel of clinician-ethicists reviews each case and specifies the manifold-optimal path. Labor-intensive but clinically grounded.

2. Best-practice benchmarking. Identify institutions or clinical units with the lowest Bond Index and use their clinical paths as approximations of the geodesic.

3. Computational geodesic. Use A* search on the estimated clinical decision complex to compute the minimum-cost path. Requires the covariance matrix \Sigma and boundary penalties \beta_k.

Each method has limitations. Expert panels are subjective (different panels may disagree). Best-practice benchmarking assumes that the best existing practice approximates the geodesic (which may not be true if all institutions share the same systematic biases). Computational geodesics depend on the accuracy of \Sigma and \beta_k estimates.

The framework recommends using multiple methods and triangulating: if all three approaches produce similar Bond Index estimates, confidence in the result is high. If they diverge, the divergence itself is informative — it reveals where the estimation procedure is most uncertain.

Actionable Thresholds

For the Bond Index to drive institutional action, thresholds for “acceptable” and “unacceptable” disparity must be established.

Proposition 12.2 (Threshold Setting for the Bond Index). The Bond Index disparity ratio \text{DR}(P, S_1, S_2) should be evaluated against community-determined thresholds:

• \text{DR} < 1.2: Within acceptable variation. Monitor but no immediate action required.
• 1.2 \leq \text{DR} < 1.5: Yellow zone. Investigate contributing dimensions. Implement targeted interventions.
• 1.5 \leq \text{DR} < 2.0: Red zone. Significant structural disparity. Policy redesign recommended.
• \text{DR} \geq 2.0: Critical disparity. Immediate policy review and institutional response required.

These thresholds are illustrative, not prescriptive. Each institution and community must establish thresholds that reflect their values — which is itself a metric choice on the manifold.

What It Does Not Measure

The Bond Index measures structural injustice — the systematic deviation between policy-mandated paths and clinical geodesics across populations. It does not measure:

• Individual injustice: The Bond Index is a population-level measure. An individual patient may experience injustice that the population average does not capture.
• Historical injustice: The Bond Index measures current policy effects, not historical wrongs. The legacy of Tuskegee affects current d_5 values, but the Bond Index does not measure the historical wrong itself — only its ongoing consequences.
• Non-healthcare injustice: Social determinants of health (poverty, housing, education) affect the clinical manifold from outside the healthcare system. The Bond Index measures the healthcare system’s contribution to manifold cost, not the contribution of external social structures.

What It Cannot Fix

The Bond Index diagnoses structural injustice and quantifies its magnitude. It does not, by itself, generate remedies. A hospital that computes \text{BI}(P, S_{\text{Black}}) = 2.3 \times \text{BI}(P, S_{\text{White}}) knows the magnitude of the problem but must still design interventions to address it.

The framework suggests the general strategy: identify the dimensions contributing most to the disparity (dimensional decomposition), design interventions that address those specific dimensions, and re-measure the Bond Index after implementation to assess impact. But the specific interventions — culturally concordant care, trust-building programs, modified allocation criteria — require clinical expertise, community engagement, and institutional will that the mathematics alone cannot provide.