Evaluating the Diagnostic Validity of the Ankle-Brachial Index in Diabetic Populations: A Methodological Deconstruction of Pseudo-normality and Its Impact on Health Professional Decision-Making Using a Target Trial Emulation Framework

Target Trial Emulation and Structural Framework

Implementation of a target trial emulation framework allowed for the rigorous assessment of the structural validity of the ABI (Robins & Finkelstein, 2000). The prospective, dual-campus observational design addressed two methodological imperatives: (a) mitigating baseline confounding and (b) neutralizing measurement errors inherent to diabetic hemodynamic assessments. Temporal synchronization was enforced through a strict Time Zero (T0) protocol. T0 was operationalized as the exact date of formal study enrollment and initial referral. This architecture anchored covariate ascertainment and outcome tracking to a unified baseline, effectively preventing immortal time bias.

Institutional Context and Ethical Governance

The evaluation cohort was assembled between October 2024 and August 2025 across two high-volume clinical sites: the University of Malaya Medical Centre (UMMC) and the University Malaya Specialist Centre (UMSC). Recruitment bypassed isolated specialist silos by integrating multidisciplinary pathways, including orthopedic surgery, trauma, and dedicated diabetes care units. Independent institutional review boards validated the ethical protocols for UMMC (MREC Ref: 20241023-14342) and UMSC (MREC Ref: 2025113-14583). Following written informed consent, clinical variables were archived within the ipesakit institutional registry. The surveillance framework adhered strictly to the Declaration of Helsinki and STROBE guidelines to maintain longitudinal fidelity (Von Elm et al., 2007).

Participant Selection and Measurement Fidelity

Clinical Protocol and Metabolic Standardization

Control of extraneous variance in healing trajectories was achieved by enforcing a uniform multidisciplinary care protocol based on International Working Group on the Diabetic Foot (IWGDF) guidelines (Schaper et al., 2024). Standardized interventions encompassed regular sharp debridement, rigorous infection control, and structured mechanical offloading. Antibiotic therapy was governed by an infection-tiered protocol consistent with IWGDF recommendations (Schaper et al., 2024). For all participants presenting with clinical evidence of active wound infection at enrollment, specifically the presence of at least two local inflammatory indicators including erythema, localized warmth, induration, or purulent discharge, empirical broad-spectrum antibiotic therapy was initiated within 24 hours of confirmed eligibility. Transition to culture-directed narrow-spectrum regimens was implemented following microbiological sensitivity reporting, typically within 48 to 72 hours of wound swab processing. Participants requiring intravenous therapy were co-managed by the infectious disease and orthopaedic surgery teams, with oral step-down guided by documented clinical improvement at subsequent surveillance visits. Mechanical offloading was applied universally and stratified by ulcer anatomical location. Total contact casting served as the primary modality for plantar forefoot and midfoot lesions, whereas irremovably modified removable cast walkers were deployed for heel and dorsal wounds where direct casting was contraindicated. Adherence was documented at each scheduled visit, and any deviation from the assigned offloading protocol was recorded as a protocol exception within the institutional registry. Within the evaluation framework, metabolic stabilization functioned as a core parameter. Systemic instability was neutralized through continuous monitoring of glycemic and nutritional indices, specifically HbA1c and serum albumin. Systemic glucocorticoid exposure was tightly restricted and documented to prevent unmeasured pharmacological interference with the outcome metrics.

Hemodynamic Measurement and Metric Operationalization

Certified vascular personnel ascertained the primary evaluative metric, the ABI, at T0 via a strict handheld Doppler protocol (Aboyans et al., 2012). Hemodynamic reproducibility was secured through a mandatory 10-min supine rest period prior to measurement. The analytical framework deliberately bypassed traditional binary categorization by modeling the ABI as a continuous variable. This decision preserved granular dose-response data, enabling the detection of non-linear diagnostic signals routinely obscured by the conventional 0.90 threshold. To explicitly deconstruct measurement distortion, supra-physiologic readings exceeding 1.30 were retained in their raw form rather than artificially truncated. Such operationalization permitted the direct statistical modeling of medial arterial calcification and pseudo-normal hemodynamics. A pre-planned sensitivity analysis subsequently isolated the predictive validity of pure occlusive disease by excluding these non-compressible vessels.

Outcome Operationalization and Censoring Management

Primary and Secondary Endpoints

Baseline target ulcers were identified by the largest surface area or most advanced Wagner grade, quantified via digital planimetry. For participants presenting with concurrent satellite lesions, clinical surveillance and outcome adjudication were strictly restricted to the primary index ulcer. Tracking multiple distinct healing trajectories within a single systemic host violates the independence assumption required for doubly robust estimation. We operationalized the singular index lesion to represent the primary metabolic demand, deliberately excluding satellite wounds from the quantitative data collection architecture to prevent competing healing rates from destabilizing the primary outcome metric. The primary evaluative endpoint was defined as the Percentage Area Reduction (PAR) at 20 weeks. This extended observation window deliberately bypassed the conventional 12-week benchmark to capture the protracted kinetic spectrum of neuroischemic repair. By operationalizing PAR as a continuous outcome rather than utilizing binary closure events, the analytical framework preserves granular physiological data across advanced lesions. This design effectively accommodates the severe chronicity of the cohort and avoids the misclassification of substantial yet incomplete regenerative progress. The PAR was calculated according to the following equation:

P A R = B a s e l i n e A r e a − F i n a l A r e a B a s e l i n e A r e a × 100 %

To fortify the analysis against extreme outliers, PAR values were Winsorized to a strict bounded range of −100% to 100%. The incidence of major ipsilateral limb amputation served as the secondary outcome.

Analytical Framework for Validity Assessment

Composite Weighting for Causal Inference

The evaluation framework implemented a composite weighting strategy to neutralize baseline confounding and potential selection bias resulting from participant attrition. Baseline covariate balance was achieved via stabilized weights derived from the Generalized Propensity Score (GPS) (Austin, 2018). Informative censoring was simultaneously addressed through IPCW. The participant-specific weight ( W f i n a l ) was operationalized as the product of the GPS weight and the IPCW (Cole & Hernan, 2008):

W f i n a l = W G P S × W I P C W

Stability of the estimation was ensured by Winsorizing extreme weights at the 1st and 99th percentiles. Post-weighting balance was evaluated using absolute correlation coefficients between the ABI and baseline covariates. Achieving a threshold of less than 0.1, visualized through Love plots, served as the formal benchmark for validating the pseudo-population suitable for causal inference.

Data Governance and Ascertainment Fidelity

Reporting Standards and Power Considerations

Cohort Assembly and Baseline Structural Dynamics

The analytical cohort comprised 528 participants with active diabetic foot ulcerations. Baseline phenotyping revealed a predominantly elderly demographic with a median age of 61.0 years (IQR: 51.0–71.0). The population was burdened by an overwhelming prevalence of cardiovascular comorbidities (93.2%). The metabolic profile indicated established systemic dysregulation, characterized by a median diabetes duration of 15.0 years and suboptimal glycemic control (median HbA1c 8.40%). Local wound metrics demonstrated significant chronicity and severity; the median ulcer duration was 8.0 weeks, and 68.4% of lesions were categorized as advanced (Wagner Grade 2–3). Concurrent systemic inflammation was evident, reflecting a median C-reactive protein (CRP) concentration of 12.50 mg/L. Comprehensive baseline distributions are detailed in Table 1 (Figure 1).

OPEN IN VIEWER

Table 1

Baseline Structural and Hemodynamic Characteristics of the Evaluated Cohort

Characteristic	Overall (N = 528)
Demographic characteristics
Age, median [IQR], years	61.00 [51.00, 71.00]
Female sex, no. (%)	233 (44.1)
Body mass index (BMI), median [IQR], kg/m2	25.60 [22.20, 27.83]
Vascular risk factors
Smoking history, no. (%)	228 (43.2)
History of CVD, no. (%)	492 (93.2)
Systemic inflammation
C-reactive protein, median [IQR], mg/L	12.50 [7.80, 25.60]
Metabolic and renal profile
Duration of diabetes, median [IQR], years	15.00 [9.14, 22.00]
HbA1c, median [IQR], %	8.40 [7.80, 9.20]
Serum albumin, median [IQR], g/L	34.50 [31.20, 38.20]
eGFR, median [IQR], mL/min/1.73 m2	66.90 [46.59, 85.29]
Ulcer characteristics
Duration of ulcer, median [IQR], weeks	8.00 [4.00, 18.00]
Baseline ulcer area, median [IQR], cm2	5.00 [3.00, 8.50]
Wagner grade 2–3, no. (%)	361 (68.4)
Ulcer anatomical location
Forefoot and digital region, no. (%)	317 (60.0)
Midfoot plantar surface, no. (%)	97 (18.4)
Heel, no. (%)	74 (14.0)
Other (dorsum, perimalleolar), no. (%)	40 (7.6)

OPEN IN VIEWER

Figure 1

Participant Screening and Attrition Trajectory Within the Target Trial Emulation Framework.

Target Trial Emulation Fidelity and Covariate Balance

The composite weighting algorithm successfully neutralized baseline structural confounding. Generalized Propensity Score (GPS) implementation generated stable weight distributions without extreme truncation penalties. The effective sample size (ESS) was maintained at 217.1 after stabilizing weights. Crucially, the weighting architecture achieved strict covariate balance; absolute correlation coefficients for all specified confounding domains were systematically compressed below the rigorous 0.10 threshold across the entire continuous ABI spectrum. This diagnostic confirmation validates the simulated pseudo-population, ensuring that subsequent outcome models are robust against measured baseline imbalances (Supplemental Table S1; Supplemental Figure S1).

Evaluation of Linear Hemodynamic-Healing Trajectories

Initial unadjusted modeling suggested a nominal positive correlation between baseline ABI and 20-week PAR (β = 11.34; p < 0.001). However, following the application of the doubly robust estimation framework, this association was profoundly attenuated. The fully adjusted linear model failed to detect a statistically significant structural relationship between ABI and wound repair (β = 2.48; 95% CI -1.31 to 6.27; p = 0.20). This statistical collapse strongly indicates that a traditional linear framework is structurally inadequate to capture the complex hemodynamic influence of the ABI on tissue repair within this diabetic cohort (Table 2).

OPEN IN VIEWER

Table 2

Divergent Hemodynamic Associations With Wound Repair and Limb Survival Across Successive Methodological Adjustments

Primary outcome: Percentage area reduction	Coefficient (95% CI)	p value
Model 1: Unadjusted	11.338 (7.168 to 15.507)	<0.001
Model 2: Weighted (IPTW/IPCW)	3.409 (-1.596 to 8.413)	0.181
Model 3: Doubly robust	2.479 (-1.313 to 6.271)	0.20
Secondary outcome: Major amputation	Odds ratio (OR)	p value
Model 1: Unadjusted	0.661 (0.558 to 0.783)	<0.001
Model 2: Weighted (IPTW/IPCW)	0.784 (0.674 to 0.912)	0.002
Model 3: Doubly robust	0.765 (0.644 to 0.908)	0.002

Hemodynamic Impact Divergence and Major Amputation Risk

In stark contrast to the healing outcomes, the ABI maintained robust structural validity as a linear predictor of limb survival. Across all model specifications, higher ABI values consistently conferred protection against critical ischemic events. Within the fully adjusted doubly robust framework, each 0.1-unit incremental increase in baseline ABI translated to a highly significant 23.5% reduction in the odds of major ipsilateral amputation (OR 0.77; 95% CI 0.64–0.91; p = 0.002). This divergence confirms that while ABI fails to linearly predict the metabolic demands of wound closure, it remains a mathematically stable proxy for the macro-circulatory flow required to prevent tissue necrosis (Table 2).

Non-Linear Hemodynamic Threshold Modeling

Restricted cubic spline (RCS) modeling delineated a significant inverted U-shaped relationship between the continuous ABI and 20-week PAR (p for non-linearity <0.001; Figure 2). The structural dose-response relationship bifurcated into three discrete hemodynamic phases. In the primary ischemic phase (ABI <0.62), predicted PAR values were negative, indicating a state where perfusion is insufficient to meet basal metabolic requirements, resulting in progressive wound expansion or clinical deterioration (Figure 2).OPEN IN VIEWER

The therapeutic acceleration phase (0.62 ≤ ABI ≤0.90) demonstrated that healing efficacy increased with higher perfusion pressure, reaching a maximum predicted PAR of 27.9% at the 0.90 inflection point. Conversely, in the pseudo-normal phase (ABI >0.90), the healing benefit deviated from linear expectations and entered a sustained declining trajectory. This divergence suggests that beyond the 0.90 threshold, the prognostic fidelity of the ABI is systematically compromised by competing vascular pathologies, primarily medial arterial calcification.

Sensitivity Matrices for Robustness Validation

Multiple sensitivity analyses confirmed the stability of the primary findings across varying analytical scenarios (Supplemental Table S2). Notably, following the planned exclusion of non-compressible vessels (ABI ≥1.30), a positive linear association between the ABI and PAR was successfully restored (β = 6.67; 95% CI 0.40–12.94; p = 0.038). This restoration provides empirical evidence that the downward trajectory observed in the full cohort is a diagnostic artifact driven heavily by extreme arterial stiffening (≥1.30). Without the negative leverage of these non-compressible vessels, the 0.62 to 1.29 range functions as a net positive continuous spectrum.

Furthermore, threshold validation using traditional categorical assignments revealed no significant difference in healing efficacy between the sub-optimal (0.62–0.90) and assumed optimal (≥0.90) groups (p = 0.742). Rather than contradicting the non-linear model, this statistical parity confirms the “ceiling effect” of the ABI. It demonstrates that once the 0.62 ischemic threshold is surpassed, further elevations into the pseudo-normal range (≥0.90) do not confer superior tissue repair, thereby invalidating the traditional reliance on 0.90 as a definitive therapeutic benchmark. Severe healing impairment was specifically isolated to the high-risk ischemic group (ABI <0.62; β = −40.66; 95% CI -82.25 to 0.93; p = 0.056).

To assess the potential impact of unmeasured confounding, E-value analysis yielded point estimates of 1.87 for major amputation and 1.66 for the primary healing outcome. These values indicate that substantial unmeasured confounding would be required to negate the observed associations, thereby reinforcing the structural robustness of the doubly robust estimates.

Subgroup Heterogeneity Assessment

Effect modification analysis revealed that age and sex significantly modified the association between the ABI and tissue repair (p for interaction <0.001; Figure 3). Hemodynamic benefits were predominantly localized to participants aged ≥65 years (n = 215) and female cohorts (n = 233). This localization suggests that tissue repair in these specific demographics may possess a higher degree of macro-perfusion dependence, potentially due to age-related microvascular rarefaction.OPEN IN VIEWER

No significant interaction effects were detected for smoking status (p = 0.72) or Wagner grade (p = 0.67). The observed non-linear patterns and hemodynamic thresholds remained stable across varying levels of wound severity and tobacco exposure. This consistency demonstrates that the identified inflection points (0.62 and 0.90) function as robust structural benchmarks within the broader diabetic foot population, regardless of baseline clinical heterogeneity.

Robustness of Estimates Against Informative Censoring

The evaluation framework rigorously quantified the potential impact of participant attrition on the longitudinal outcome estimates. Of the 528 participants initially enrolled at Time Zero (T0), 54 individuals (10.2%) were lost to follow-up or withdrawn prior to the 20-week primary endpoint (Figure 1). To neutralize potential selection bias arising from non-random loss to follow-up, IPCW was integrated into the doubly robust analytical framework.

A comparative analysis demonstrated that the primary hemodynamic associations remained invariant to participant attrition. Specifically, the effect estimate for ABI in the IPCW-adjusted model (β = 2.48; 95% CI -1.31 to 6.27) was virtually identical to the estimates derived from a restricted complete-case analysis (β = 2.48; 95% CI -1.30 to 6.26; p = 0.20; Supplemental Table S2). This numerical stability across weighting specifications confirms that the observed relationships were not driven by informative censoring or selective attrition, further securing the internal consistency of the target trial emulation.

Limitation

This evaluation is subject to specific methodological boundary conditions. The emphasis on a dual-campus setting ensures high internal validity through standardized multidisciplinary care, yet it necessitates a cautious extrapolation of these thresholds to global cohorts with divergent metabolic archetypes. Hence, replicating this target trial emulation framework across diverse international healthcare infrastructures remains an essential future step to validate the broad applicability of these specific hemodynamic inflection points. Regarding the hemodynamic assessment, the absence of concurrent toe pressure measurements represents a recognized constraint of this evaluation. Toe-Brachial Index (TBI) data were not collected during the study period for two operational reasons. Handheld photoplethysmography equipment required for standardized digital pressure acquisition was not uniformly available across both clinical sites throughout the recruitment window, and the volume and acuity of the patient population made the addition of a second vascular measurement protocol at each enrollment visit operationally unfeasible within the existing MDT workflow. We acknowledge that this limitation is consequential. The TBI is substantially less susceptible to medial arterial calcification than the ABI, as digital vessels are typically spared from the calcification process that artificially elevates ankle pressure readings. In the pseudo-normal ABI range of 0.90 to 1.30, where the prognostic fidelity of the ABI collapses most severely in this cohort, concurrent TBI values would have provided a pressure-independent reference point to distinguish true perfusion adequacy from calcification-driven artifact. The absence of this reference constrains the granularity of the micro-perfusion characterization within that specific hemodynamic stratum. The sensitivity analysis excluding ABI ≥1.30 provides a partial structural substitute for this gap, but it cannot replicate the individual-level discriminatory capacity of a directly measured toe pressure. Integrating TBI alongside ABI at baseline is a clear priority for future iterations of this work.

Simultaneously, the observational design presents an inherent potential for residual confounding from unmeasured factors. Chief among these is offloading adherence during unsupervised daily activity, which, despite being documented at each scheduled surveillance visit, was not formally quantified at the patient level between visits, introducing a degree of unmeasured variability in the primary healing estimates. Furthermore, the methodological rigors of causal inference introduced specific statistical trade-offs. We acknowledge that the implementation of Generalized Propensity Score weighting resulted in a reduction of the effective sample size (ESS) to 217.1. This attrition of statistical information is an expected consequence of achieving strict structural covariate balance across a continuous exposure spectrum in a highly heterogeneous clinical population. To mitigate potential instability, weights were strictly truncated at the 1st and 99th percentiles. These constraints are ultimately contextualized by the E-value point estimates (1.87 for amputation; 1.66 for PAR), which suggest that substantial unmeasured confounding would be required to negate the observed effects. An additional boundary condition involves the explicit focus on proximal physiological confounders, which precluded the formal assessment of effect measure modification across different racial or ethnic strata. Understanding how socio-cultural determinants interact with medial arterial calcification burden remains a distinct evaluative target, necessitating future prospective registries designed to capture comprehensive demographic gradients alongside metabolic metrics. Finally, our protocol restricted hemodynamic evaluation strictly to the ipsilateral limb of the index ulcer. While this maintained focus on the primary target lesion, exploring the contralateral hemodynamic profile in patients with bilateral presentations represents a compelling avenue for future investigation to better characterize systemic microvascular reserve. Consequently, the identified hemodynamic inflection points and structural conclusions remain highly reliable within the target trial emulation framework. The occurrence of unscheduled revascularization events in 23 participants during follow-up, while managed through composite failure adjudication and IPCW correction, represents an additional boundary condition. Although these cases were systematically handled to prevent outcome contamination, the absence of a dedicated post-revascularization subgroup analysis limits the granularity with which perfusion recovery trajectories can be characterized in this cohort.