Data were obtained from a secure clinical database containing vascular surgery records from public and private tertiary referral centres in Galway, Ireland. The dataset included procedures classified into aortic, carotid, and peripheral vascular disease categories. Data collection began in 2002 and followed a structured, periodically audited process. The two contributing centres conducted formal reviews of the database approximately every four years to ensure accuracy, completeness, and internal consistency.

No statistical outliers were identified during the preliminary inspection, as procedure counts and patient-level variables had already been validated as part of the database's maintenance procedures. All data were re-examined, cleaned, and pre-processed prior to modelling. Data cleaning steps included verifying variable coding, confirming temporal consistency across years, and checking for missing or duplicate entries. Variables were standardised where necessary, and categorical variables were encoded appropriately. Because the dataset represented a complete census of available records for the study period, no imputation procedures were required. Finally, this prepared dataset was used directly for the regression-based supervised ML predictive modelling performed in the study

We used ML-based regression analysis as a statistical method to relate the number of surgeries to one or more independent variables. Several modelling methods were employed, including linear Regression, Polynomial Regression, Power Regression, and Exponential Modelling.

We used observed data from two deliberate practice volume vascular centres in the West of Ireland, both of which use the same triage, management, and technology trends. Examining trends and shifts in surgical volumes will provide a foundation for predictive modelling and future planning.

Our model uses the most recent Irish Census Bureau population projections, published in 2024, to estimate the number of vascular surgeons required to provide health care to the Irish population (11). Irish Medical Council Registered Specialists data for 2023 were used to identify the number of practising vascular surgeons in 2023. We used the data previously described by Go et al. (12), a ratio of 1.4 vascular surgeons per 100,000 population. This data was derived by Richard Cooper and used by Merritt Hawkins in “A market analysis of vascular surgery supply, demand, compensation and recruiting trends: prepared for the Society for Vascular Surgery, November 13, 2017” to estimate vascular surgery workload (12). Although not assumed to be directly transferable to the Irish system, we have used this United States - based benchmark as a pragmatic external reference in the absence of Ireland-specific vascular surgery workforce ratios. In addition, the median productivity of vascular surgeons, estimated at 8,481 work relative value units (wRVU)/year, was derived from the 2017 wRVU data from the “Medical Group Management Association Physician Compensation and Production” reports.

Data on aortic surgeries were segmented into four five-year eras, comparing public and private sectors (Figure 2). The number of surgeries has evolved in both public and private hospitals.

Data on aortic surgeries were segmented into four five-year eras, comparing the public and private sectors (Figure 3). This visualisation highlights the trends in surgical volumes over time in both public and private hospitals.

Data on Carotid surgeries were segmented into four five-year eras, comparing the public and private sectors (Figure 4). This visualization highlights the trends in surgical volumes over time in both public and private hospitals.

Trends in aortic surgeries across different locations, as estimated using polynomial regression —a subset of supervised machine learning —are displayed in Figure 5.

Trends in carotid surgeries across different locations, as estimated using polynomial regression —a subset of supervised machine learning —are displayed in Figure 6.

Trends in peripheral surgeries across different locations, as estimated using polynomial regression —a subset of supervised machine learning —are displayed in Figure 7.

Figure seven illustrates the trends in peripheral surgeries at different locations using exponential modelling, a technique within supervised ML. For the peripheral dataset, the Exponential Model was chosen for public and private hospitals due to its highest R-squared values: 0.52 for public hospitals and 0.18 for private hospitals. The lower R-squared value for private hospitals indicates a less accurate model, likely due to more dynamic, variable trends. Forecasts for 2028 estimate 189 surgeries in PUB and 357 in PRV. This suggests a notable difference in the growth rates and predictability between the two sectors.

Figure eight displays trend modelling techniques to forecast aortic surgeries by year. This chart shows that public hospitals operate under a capped model, indicating limited capacity for growth in surgical volumes. In contrast, private hospitals follow an uncapped model, showing a strong positive trend. This suggests that, assuming current conditions remain constant, private hospitals will continue to meet increasing demand without constraints.

Figure nine employs trend modelling techniques to forecast the number of carotid surgeries by year. This chart shows that public hospitals operate under a cap, suggesting limited capacity for growth in surgical volumes. In contrast, private hospitals follow an uncapped model, indicating a strong positive trend and the ability to meet increasing demand without restrictions.

Figure ten uses trend modelling techniques to forecast peripheral surgeries by year. This chart shows that public hospitals operate under a capped model, suggesting limited capacity for growth in surgical volumes. In contrast, private hospitals follow an uncapped model, indicating a strong positive trend and the ability to meet increasing demand without restrictions.

Currently, we have fewer than 50% of the required vascular surgeons working in Ireland. By 2030, the shortage of vascular surgeons in Ireland is projected to increase the clinical workload by 39%, requiring an additional 647,938 wRVUs, with a 28% workforce shortage. By 2050, shortages are expected to increase the clinical workload by 7%, necessitating an additional 727,585 wRVUs. Our model predicts a decline in surgeon shortages by 17% through 2040, with the workforce size meeting near demand by 2050, resulting in a 7% shortage (Table 1).

We used a power model for PUB and a polynomial model for the PRV for aortic surgeries. The polynomial model at the PRV yielded an R-squared of 0.99, indicating a robust fit. In contrast, the power model for PUB had an R-squared of 0.17. The polynomial trend projects a significant increase in surgeries at the PRV, while the trend at PUB is relatively flat.

Peripheral surgeries at both locations were modelled using an exponential fit. The R-squared value for PUB was 0.53; for the PRV, it was lower but acceptable given the limited data. The trend at PUB shows a decline since 2007, whereas the PRV indicates a slow positive increase with a recent inflexion point, suggesting a shift in the trend.

Carotid surgeries were also modelled using exponential trends. The private hospitals showed a strong positive trend (R-squared = 0.86), while the public hospitals displayed a flatter, decreasing trend (R-squared = 0.50). By 2028, private hospitals are expected to perform significantly more carotid surgeries than public hospitals.

The study predicts a shortage of vascular surgeons in the coming decades, with potential workforce size and demand alignment by 2050. Until then, vascular surgeons must increase their productivity to manage the workload. Burnout, changing practice patterns, geographic maldistribution, and expanding healthcare coverage will impact the future workforce's ability to meet population needs. Addressing future demands and capacity challenges through public-private partnerships, investing in vascular training, and improving wRVU compensation is crucial to ensure an adequate and sustainable vascular surgery workforce.

The rise of endovascular and hybrid interventions reflects the dynamic evolution toward individualized, multidisciplinary care models (16). Maintaining the current level of vascular practice requires a substantial increase in manpower. The growing demand for specialized interventions, such as aortic, carotid, and peripheral surgeries, requires recruiting and training more vascular surgeons with specific areas of expertise. Specialization allows surgeons to develop more profound proficiency and improve patient outcomes in their particular field, which is critical given the increasing complexity of vascular diseases (16).

Moreover, the increasing workloads and the projected need for higher productivity underline the importance of integrating AI and ML systems into vascular practice. These technologies can streamline administrative tasks, enhance diagnostic accuracy, and optimize surgical planning, freeing vascular surgeons to focus on complex interventions. By leveraging AI and ML, we can reduce strain on surgeons and improve efficiency across the healthcare system.

The reliance on wRVUs as a measure of productivity is unsustainable without significant changes in reimbursement policies. Current wRVU compensation models do not adequately reflect the increasing demands placed on vascular surgeons. To sustain the workforce and ensure high-quality care, there must be a 65% increase in reimbursement rates. This adjustment would compensate for the increased workload and help attract and retain talented surgeons (21).

We were limited by the data, as we used only two vascular institutes within a geographical area for the predictive modelling. Furthermore, we used a subset of ML algorithms to improve model accuracy and forecast workforce needs, with accuracy assessed using R-squared values. The inherent limitations of having only four buckets of data over 22 years impact the accuracy of long-term predictions. Thus, our models prioritize the nearest buckets for more reliable forecasts, acknowledging that projections for 2023 are more accurate than those for 2050. Finally, although the two data sets from the public and private institutes exhibit different trends, we applied a single model to identify the model that best predicts future trends.

We did not perform explicit uncertainty quantification, as we used a basic supervised ML regression model in our time-series models, based on four aggregated, stratified time buckets over a 22-year period, across diverse datasets. Because the models are fitted on only four aggregated time periods, formal prediction intervals would not adequately capture structural uncertainty and would give a spurious impression of precision, particularly for forecasts beyond 2040. Our forecasts should therefore be interpreted as scenario-based estimates, with greater confidence in near-term projections and more caution for long-term horizons. Furthermore, R² values should be interpreted appropriately to qualify the predictive strength of the model.

Furthermore, the benchmark of 1.4 vascular surgeons per 100,000 population was derived from U.S. workforce modelling, which may not be directly applicable to Ireland. Differences in population structure, the public–private mix, the distribution of vascular services, and the roles of interventional radiology and cardiology could mean that the optimal surgeon-to-population ratio for Ireland is higher or lower. Our projections should therefore be viewed as indicating the direction and scale of expected workforce strain rather than prescribing precise workforce targets for Ireland.

We used the full available dataset to estimate model parameters, given the limited number of time points and the exploratory nature of the analysis. No cross-validation was performed because there was a risk of “data leakage” in time-series data. Moreover, cross-validation methods might not properly handle non-independent grouped data.