Materials and methods

This study was registered as a clinical audit and approved by the Clinical Audit and Effectiveness Team at Nottingham University Hospitals NHS Trust (NUH). Formal ethical approval was not required, in accordance with National Health Service (NHS) Health Research Authority guidance, as the project was defined as a clinical audit. Consequently, public clinical trial registration was not sought, as registry protocols are reserved for prospective interventional research.

Study population

This retrospective, single-centre observational study was conducted at NUH. As the centralised hub for vascular services across Nottingham and Nottinghamshire, NUH operates across the Queen’s Medical Centre and City Hospital campuses, serving a diverse regional population.

Clinical data for all patients undergoing vascular surgical procedures over a four-month period (July 2024 to October 2024) were retrospectively retrieved from electronic health records (EHRs) using relevant clinical procedure codes. Initially, 835 patients were identified (N=835).

Eligibility criteria

Inclusion criteria comprised patients of all ages and genders undergoing any vascular surgical intervention during the study period (July 2024 to October 2024). Exclusion criteria included day-case procedures (e.g. isolated angioplasty with or without stenting), incomplete longitudinal perioperative data (defined as the absence of either a preoperative or postoperative Hb value), and lack of documented 30-day follow-up outcomes.

Briefly, patients of all ages and genders undergoing any vascular surgical intervention during the study period were eligible. To ensure predictive model integrity and maintain an inpatient focus, strict exclusion criteria were applied. No patients were excluded based on the underlying aetiology of their anaemia. The study cohort comprehensively included individuals with all forms of pre-existing or baseline anaemia, encompassing megaloblastic and microcytic types.

The resulting high exclusion rate reflects the retrospective nature of EHR data capture in a mixed elective or emergency vascular service, driven primarily by the exclusion of day-case procedures and cases lacking complete longitudinal perioperative data.

Following these exclusions (n=639), a final analytical cohort of 196 patients was established. No missing outcome data were present within the final analytical cohort. The patient selection process and attrition are detailed in the study flowchart (Figure 1).

Figure 1: Cohort Flowchart.

Study outcomes and variable definitions

The primary outcome was a composite of short-term adverse clinical events occurring within 30 days of the index procedure, including all-cause mortality, unplanned hospital readmission, or a prolonged index hospital stay (seven days and above).

The primary predictor variables were categorised according to the WHO anaemia criteria, utilising sex-specific Hb thresholds [5]. To ensure consistency across models, the same four-level severity scale was applied to both pre-operative and post-operative measurements: no anaemia Hb ≥130 g/L (males) or ≥120 g/L (females), mild anaemia Hb 110-129 g/L (males) or 110-119 g/L (females), moderate anaemia Hb 80-109 g/L, and severe anaemia Hb <80 g/L.

Data collection and sample size

Data were retrospectively collected from the Trust’s digital EHR and Nervecentre, including demographic data (age and gender), comorbidities (diabetes mellitus, hypertension), procedures, laboratory findings (preoperative Hb, postoperative Hb), admission durations, and 30-day follow-up outcomes, which included unexpected readmission and mortality. Data collection was conducted by clinicians following a standardised protocol with predefined clinical procedure codes. The entire extraction process was directly supervised by senior consultant vascular surgeons, and any discrepancies in data categorisation were resolved through consensus compliance reviews.

Due to the retrospective nature of this study, a consecutive convenience sampling strategy was utilised. The cohort comprised all patients who were formally coded for an inpatient vascular surgical procedure within the EHRs during the designated four-month study window (July 2024 to October 2024) at Nottingham University Hospitals. By extracting every consecutive patient matching these specific procedural codes within the index period, selection bias was minimised. A post hoc assessment of sample size adequacy confirmed that the resulting 158 composite adverse outcome events across the final analytical cohort (n=196) significantly exceeded the standard 10 events per variable (EPV) guideline for logistic regression, ensuring robust statistical validity and model reproducibility [6].

Statistical analysis

Data were compiled within Microsoft Excel (Microsoft Corp., Redmond, WA, USA) and analysed using JASP (Version 0.16.0; University of Amsterdam, Amsterdam, Netherlands). A descriptive synthesis of the cohort was performed, with continuous variables summarised as means (±SD) or medians (IQR) and categorical parameters expressed as absolute frequencies and percentages. To delineate clinical differences across various patient profiles, exploratory subgroup analyses were conducted based on gender, procedural classifications (minimally invasive, open surgery), and baseline comorbidities (diabetes, hypertension). For these comparisons, Pearson’s chi-square test or Fisher’s exact test was used as appropriate for categorical data, with statistical significance threshold maintained at p<0.05.

The association between the staged categories of anaemia severity and the primary composite 30-day adverse outcome was initially assessed using bivariate analysis with the chi-square test. Subsequently, binomial logistic regression was used to evaluate three separate predictive models to identify the strongest clinical indicator among the preoperative anaemia severity model, the postoperative anaemia severity model, and the persistent perioperative anaemia model.

The preoperative and postoperative anaemia models were designed to evaluate the association between preoperative and postoperative anaemia severity, respectively, and 30-day adverse outcomes. Anaemia severity was stratified according to the WHO anaemia criteria. The persistent perioperative anaemia model was designed to evaluate the association between sustained anaemia at both the preoperative and postoperative time points and 30-day adverse outcomes. This model was defined as a binary variable (Yes/No), representing patients who met the WHO criteria for anaemia at both the preoperative and postoperative time points.

To identify the optimal prognostic framework, models were appraised for clinical parsimony via the Akaike Information Criterion (AIC) and for explanatory power using Nagelkerke R2. The area under the receiver operating characteristic curve (AUC) was calculated for each model to compare their relative discriminative accuracy in predicting 30-day composite adverse outcomes. All metrics were computed directly within JASP's open-source regression framework under standard parametric assumptions, requiring no proprietary software licensure.