The optimal annual case volume for acute type A aortic dissection surgery in relation to long-term outcomes

Abstract

OBJECTIVES

Previous analyses of the volume–outcome relationship have focused on short-term outcomes such as early mortality. The current study aims to update a novel statistical methodology, facilitating the evaluation of the relation between procedural volume and time-to-event outcomes such as long-term survival, using surgery for acute type A aortic dissection as an illustrative example.

METHODS

This study employed an existing dataset of type A dissection outcomes, retrieved from literature. Studies were included when reporting on annual case load and long-term survival, which served as the primary outcome of interest. Individual patient data were reconstructed from the included studies, and a hazard ratio was determined per study in relation to overall survival, after which the calculated hazard ratios were incorporated in a restricted cubic-spline model, facilitating the application of the elbow method.

RESULTS

Fifty-two studies were included (n = 14 878 patients), with a median follow-up of 5 years. One-, 3-, 5- and 10-year survival of the overall cohort were 82% [95% confidence interval (CI) 82–83%], 79% (95% CI 78–80%), 74% (95% CI 74–75%) and 60% (95% CI 59–62%), respectively. A significant non-linear volume–outcome relation for long-term survival was observed in both the unadjusted and adjusted analyses (P = 0.030 and P = 0.002), with an optimal annual case load of 32 cases/year (95% CI 31–33).

CONCLUSIONS

Based on the available data, these findings imply that the annual case volume to achieve optimal long-term survival is located near a procedural volume of 32 cases/year. After accrual of more annual procedures, long-term survival may no longer significantly improve any further.

Open in new tabDownload slide

INTRODUCTION

Centralization of health care services is an increasingly important theme in health care policy and may particularly apply to high-risk and less frequently performed procedures, such as aortic surgery in general, and surgery for acute type A aortic dissection (ATAAD) in particular [1]. Recently, a new methodology to evaluate the relation between procedural volume and surgical outcome was introduced, with subsequent determination of the optimal annual case volume by use of a new statistical method [2–4]. This primary analysis was aimed at early mortality as the outcome of interest and lacked the ability to evaluate the volume–outcome relation for time-to-event end-points such as long-term survival.

However, in ATAAD and other procedures, the extent and adequacy of the index operation determine prognosis [5, 6]. Inherently, there is a trade-off between a higher risk of a more extensive index procedure—resulting in increased early mortality—but favourable long-term survival [5]. As such, an extension of the proposed method to evaluate time-to-event outcomes such as survival during long-term follow-up is warranted, particularly for ATAAD surgery. Consequently, in this updated report, we present an additional method to determine the optimal annual case volume in relation to time-to-event outcomes using a similar dataset summarizing all available published evidence of patients undergoing surgery for ATAAD.

MATERIALS AND METHODS

Protocol and registration

Our report was designed as an extension of an existing systematic review and meta-analysis and was registered in PROSPERO (initial registration date: 18 July 2022 [CRD42022345024], updated with the current report on 11 November 2023) and complied with the 2020 PRISMA statement (see Supplementary checklist). The dataset of the original study was re-used for the current study, employing the new statistical method to determine the volume–outcome relation for time-to-event outcomes.

Eligibility criteria

The eligibility criteria were previously described [3]. In short, studies were initially included when reporting on (i) number of consecutive ATAAD patients (both DeBakey type I and II), (ii) years of inclusion and (iii) the primary outcome (in-hospital and/or 30-day mortality). For the current study, focusing on time-to-event outcomes, the previously defined inclusion criteria are now limited to the sole inclusion of studies reporting on long-term survival, with adequately reported Kaplan–Meier (K-M) curves (including numbers at risk at various timepoints).

Information sources and search strategy

A systematic search was applied to three electronic databases (PubMed, EMBASE and the Cochrane Library), and the last search was performed on 29 March 2023 for all databases (M.J.K. and S.H.). The full search strategy can be found in Supplementary Material 1.

Selection process and data collection

Studies were primarily screened based on title and abstract, after which a second evaluation comprised the assessment of full texts. This process was performed by M.J.K. and S.H. Afterwards, data were collected in a predefined worksheet (M.J.K. and S.H.).

Outcomes and effect measures

Long-term survival is the primary outcome of the current study, and freedom from reoperation will serve as a secondary outcome. Given the nature of the data derived from literature, a competing risk analysis for reoperation-free survival could not be performed.

Risk of bias assessment

A quality assessment of studies was performed by use of the Newcastle Ottawa Scale, with a modification for single-arm studies [3, 7], summarizing studies in having a ‘low’ risk of bias, ‘some concerns’ of risk of bias or ‘high risk’ of bias.

Data synthesis

Pooled baseline study, patient and procedural characteristics were presented in tables, including pooled mean values with corresponding 95% confidence intervals (CIs) or standard deviations (SDs). All continuous variables were analysed in a meta-analysis of means using inverse-variance weighting within a random-effects model. Data on the prevalence of comorbidities were included in a meta-analysis of proportions, also using inverse-variance weighting within a random-effects model, to calculate mean values for the overall cohort.

Using the K-M-derived individual patient data (IPD) meta-analysis approach [8], unadjusted long-term survival was evaluated. One- 3-, 5- and 10-year survival were reported based on K-M estimates for the overall cohort. Survival data were incorporated in a Cox frailty model with study-level adjustments for mean age of the population, percentage of male patients, geographical continent where the patients were included, median inclusion year of the patients in the study, and the frailty parameter (centre) as a random-effects term [9]. Based on results from multivariable Cox frailty models, adjusted predicted survival curves were computed with the conditional balancing method.

Next, a hazard ratio (HR) was calculated per centre, using the unadjusted and adjusted survival of the overall (aggregated) sample as reference, excluding the study of interest, per calculation (for one specific centre). The HRs are reported with corresponding 95% CIs. The HR per centre is then plotted on the y-axis, with the centre’s corresponding annual case volume on the x-axis. The potential non-linear relation between annual case volume and the HR per centre is investigated using a restricted cubic-spline (RCS) model for meta-analysis (three knots) and presented graphically (non-linear mixed effects model). The elicitation of three knots was predetermined to enhance the clinical interpretability of the data by improving model smoothing and facilitating the approximation of an optimal annual case volume, in agreement with the description by Harrell [10]. Statistical heterogeneity among studies is objectively assessed using the I²-metric and τ². To explore potential heterogeneity, sensitivity analyses were performed by stratification for geographical continents.

As previously described [3], the optimal annual case volume threshold is determined by the ‘elbow of the curve’ [2], a mathematical concept derived from optimization statistics. In short, the mathematical principle lies within the connection of a straight line between the extremes of the curve, after which the maximal distance from the straight line to the curve is calculated (Supplementary Material 2 demonstrates a visual representation of the elbow method). At this location in the curve, an increase in case volume does not result in a further improvement of the survival rate (i.e. ‘when relative costs to increase a specific parameter are no longer worth the corresponding performance benefit’ [8]).

For the current analysis, we used R Statistics Version 3.6.0 (R Foundation, Vienna, Austria), using ‘metafor’, ‘survival’, ‘ggplot2’, ‘pathviewr’ (for the elbow analysis) and ‘discfrail’ (frailty Cox model).

Data availability

The data repository and generated R codes will be shared openly through GitHub upon publication of this study (https://github.com/samuelheuts/R-codes-and-manual-for-analyses-update).

RESULTS

Study inclusion

Based on the initial search [3], 140 studies were included, originating from 140 individual centres (5 continents, 26 countries). Of these 140 centres, 52 centres reported on long-term survival. Consequently, these 52 studies were the primary source of data for the current updated analysis (n = 14 878 patients, Supplementary Material 3 presents the individual included studies/centres; please see the references in the Supplementary Material for an overview of all included studies/centres).

Patient population

The mean age of the sample was 60.1 years (SD 4.8 years) and 34.4% (95% CI 33.6–35.2%) of patients were females. DeBakey type I dissection was most prevalent (80.3%, 95% CI 79.2–81.3%), and 3.8% (95% CI 3.5–4.2%) of patients had heritable thoracic aortic disease. Aortic root surgery was performed in 24.4% of cases (95% CI 23.6–25.2%), and total arch replacement was performed in 26.6% of patients (95% CI 25.9–27.4%, Table 1).

Risk of bias

A study quality assessment can be appreciated in Supplementary Material 4. Overall, risk of bias ranged from ‘low’ (n = 40) to ‘high’ (n = 12). As all studies reported on the primary outcome of interest, all studies were included for final analysis.

Long-term survival

The median follow-up of the entire cohort was 5.0 years (interquartile range: 2.2–8.6 years). The unadjusted and adjusted survival curves are presented in Fig. 1A and B (adjusted for age, sex, geographical continent and year). One-, 3-, 5- and 10-year survival of the overall cohort were 82% (95% CI 82–83%), 79% (95% CI 78–80%), 74% (95% CI 74–75%) and 60% (95% CI 59–62%, Table 2).

The volume–outcome relation

The HR was calculated per centre in relation to the survival of the overall cohort, excluding that centre. Then, these unadjusted and adjusted HRs were plotted on the y-axis, and the annual case load was on the x-axis (Fig. 2A and B).

The RCS analyses of these HRs in relation to the volume revealed a significant non-linear volume–outcome relation for long-term survival (Fig. 2A and B, P = 0.030 and P = 0.002 for the unadjusted and adjusted HRs; I² = 85%, τ² = 0.073, P = 0.001, and I² = 80%, τ² = 0.060, P = 0.003, respectively).

The optimal annual case load based on long-term survival

Using the RCS model, the elbow method could be applied to determine the optimal annual case volume in relation to the HR. In the unadjusted analysis, the optimal annual case volume was 32 cases/year (95% CI 31–33 cases/year, Fig. 3A), which was confirmed in the adjusted analysis (32 cases/year, 95% CI 31–33, Fig. 3B). Based on the available data, these findings imply that the annual case volume to achieve optimal long-term survival is located near a procedural volume of 32 cases/year. After accrual of more annual procedures, long-term survival does not seem to significantly improve further.

Sensitivity analyses

As can be appreciated in Supplementary Material 5, there was a notable variability in procedural volumes between the continents (Europe volume range 10–34 cases/year, North America range 6–47 cases/year, Asia range 5–241 cases/year). For Europe and Asia, we did not observe a significant (non-linear) relation between volume and long-term survival (P = 0.663 and P = 0.507, respectively). For North America, such a significant relation was present (P = 0.021), albeit in a linear model.

Freedom from reoperation

Data on freedom from reoperation were unfortunately only available for 10 studies. Figure 4A presents the pooled freedom from reoperation of these studies, resulting in a 1-, 3-, 5- and 10-year freedom from reoperation rate of 95.5% (95% CI 94.7–96.2%), 90.0% (95% CI 88.8–91.3%), 86.4% (95% CI 84.9–88.0%) and 73.8% (95% CI 70.8–76.8%, Table 2). Figure 4B presents the adjusted rates.

As for long-term survival, an RCS model was fitted to HRs (Fig. 4C and D present the unadjusted and adjusted RCS model). Although we observed a progressive decrease in the HR when case volume increased, this relation was not statistically significant (P = 0.295 and P = 0.330 for the unadjusted and adjusted analyses, respectively). Due to the low number of studies, an optimal annual case volume could not be determined.

DISCUSSION

In this updated report, the optimal annual case volume for ATAAD surgery to achieve optimal long-term survival was determined using an innovative statistical approach. Based on these new analyses, the optimal annual case volume was established at 32 cases/year (unadjusted), for which the robustness was confirmed in the adjusted analysis. This optimal case load closely corresponds to the previous analysis based on early mortality (38 cases) [3].

Indeed, the primary outcome of the original study was early mortality in relation to annual hospital case volume. At that time, the secondary outcome was long-term survival and its relation to annual hospital case load. In our previous report, studies were divided into quartiles based on annual case volumes, and the highest volume quartile was associated with superior long-term survival (>29 cases/year) [3]. Still, this categorization into volume quartiles lacked the ability to appreciate the non-linear relation between volume and outcome and the subsequent determination of an actual optimal case volume for this end-point.

Of note, the definition of high-volume centres is a sensitive concept, and has been subject to many previous studies, also in surgery for ATAAD. For example, Chikwe et al. found improved early mortality when performing >13 cases/year at the centre level, in an analysis of the Nationwide Inpatients Sample (NIS) database [11]. Furthermore, Merlo et al. found a lower mortality when centres performed >11 annual procedures, based on data from the same registry [12]. Although these studies have provided valuable insights into the relation between volume and outcome in ATAAD surgery, they based their thresholds on categorical groups such as quartiles [11], which do not capture the intuitively present non-linear relation between volume and outcome. Furthermore, these studies were primarily focused on short-term outcomes such as early mortality.

As described elegantly by Vonlanthen et al. [13], there is an inherent relation between experience and outcome. Moreover, institutional volume is a viable surrogate to determine expertise. Although the relation between procedural volume and outcome is intuitive, it cannot be absolute (nor linear), as outcome is dependent on other factors as well (such as patient risk profile and clinical presentation). Consequently, outcomes such as mortality (or survival) reach a ‘plateau phase’ at a certain point. The point at which these volume–outcome curves start plateauing can be considered the ‘optimum’, as results will not (significantly) increase further with the accrual of more cases. In turn, this optimum may constitute the threshold for expertise and can define a ‘high-volume’ centre.

Long-term survival is arguably the most important end-point from both the patient and the societal perspective. In the current study, encompassing the totality of evidence derived from published literature, we observed a 10-year survival rate of 60% (95% CI 59–62%), which closely corresponds to a multi-institutional analysis by Biancari et al. [14] (65.3% in the European registry of type A aortic dissection [ERTAAD], comprising 18 higher volume European centres). In surgery for ATAAD, the extent of the index procedure largely determines prognosis. Indeed, more extensive replacement of dissected or aneurysmal aortic segments may protect against future rupture, re-dissection or further aneurysmal growth. Although concomitant root and/or (total) arch surgery may increase the risk of the initial operation in type A dissection, this risk might be mitigated by fewer future aortic events and re-interventions in the longer term, potentially resulting in superior long-term survival [5, 6]. Therefore, recently released guidelines suggest that an individualized and tailored strategy seems indicated, preferentially in high-volume centres [15]. Based on the presented analyses, objective institutional volume thresholds for ATAAD surgery can be formulated. Importantly, this approach is not confined to surgery for ATAAD and can be applied to several other (cardiovascular) interventions in need of centralization [4, 16].

Several factors should be taken into account when centralizing ATAAD care. First, dedicated aortic teams play an increasingly important role in the care for patients with aortic disease, both for elective and emergency surgery. Andersen et al. have previously described the effect of the introduction of such a team, drastically improving outcomes of ATAAD patients [17]. In addition, the feasibility of a transfer of ATAAD patients should be considered, as ATAAD patients may deteriorate during transportation, negatively influencing the eventual net outcome. Interestingly, the achievability of such transfers was validated previously, based on modelled data and clinical experience in the USA [1, 18]. In their commendable study, Goldstone et al. demonstrated that operative and long-term mortality were significantly reduced in patients rerouted to high-volume centres, with acceptable transportation distances from a low- to high-volume centre of 50 miles (80 kilometres). However, it is questionable whether these findings are generalizable to larger countries with increased transportation times, and Goldstone’s (and our) findings should therefore be interpreted with care in other geographical locations. Finally, the index procedure is not the only factor influencing long-term outcomes. Indeed, distal aortic disease often persists, particularly in DeBakey type I dissections. Intensive clinical and imaging follow-up is therefore indicated [19] to timely recognize impending aortic events and prevent those through elective (distal) re-interventions.

The current study complements many of these previous studies by providing an evidence-based method to objectively define a high-volume centre. Nevertheless, the patient’s presenting status dictates the urgency of intervention, sometimes rendering a transfer infeasible. Still, in relatively stable patients, these previous studies—in conjunction with the current analysis—support the notion that a transfer to a high-volume centre could result in a clinically relevant prognostic benefit [1, 3], which is also advocated by the most recent guidelines [15]. Based on these results, we have a more complete understanding of the volume–outcome effect in surgery for ATAAD, which seems to range between 32 and 38 cases per year. However, as patient characteristics and geographical locations seem to play an important role as well, future studies should focus on the concepts to establish volume norms for specific regions.

Limitations

A limitation of the current approach is the potential of publication bias, which is intuitively present. Still, we hypothesize that the volume–outcome relation would only be strengthened further with incorporation of potential non-published data (i.e. data from small centres with suboptimal results). Furthermore, the tests for heterogeneity had a statistically significant result, but this is actually expected as I² is an expression of the proportion of observed variance and therefore reflects the volume–outcome effect [3]. In addition, we observed notable differences in volumes between geographical continents, which complicates the generalizability of the observed results. However, we hypothesize that similar volume–outcome relations would be observed in the separate continents if more and higher-volume centres were present in these stratified areas. Also, our study could not investigate the effect of individual surgeon volume or the potential presence of dedicated aortic teams. Finally, there are still other important factors we were unable to analyse, such as detailed technical and procedural characteristics. Therefore, the estimated threshold between 32 (survival) and 38 (early mortality) annual procedures has to be interpreted in this nuanced context.

CONCLUSION

Through this updated method, the relation between procedural volume and time-to-event outcomes such as long-term survival can be assessed. In addition, it allows for the determination of the optimal annual case load with respect to these outcomes as well. Using an existing dataset summarizing the totality of published evidence on early mortality and long-term survival following surgery for ATAAD, we conclude that the optimal annual case load varies between 32 and 38 institutional procedures per year for this procedure.

SUPPLEMENTARY MATERIAL

Supplementary material is available at EJCTS online.

FUNDING

None received.

CONFLICT OF INTEREST

Nothing to declare.

ACKNOWLEDGEMENTS

None.

DATA AVAILABILITY

All data are openly available, and coding is shared on GitHub upon publication of this report through https://github.com/samuelheuts/R-codes-and-manual-for-analyses-update.

AUTHOR CONTRIBUTIONS

Michal J. Kawczynski: Conceptualization; Data curation; Formal analysis; Methodology; Visualization; Writing—original draft. Sander M.J. van Kuijk: Methodology; Supervision; Validation; Writing—review & editing. Jules R. Olsthoorn: Data curation; Software; Validation; Writing—review & editing. Jos G. Maessen: Conceptualization; Supervision; Validation; Writing—review & editing. Suzanne Kats: Conceptualization; Supervision; Validation; Writing—review & editing. Elham Bidar: Conceptualization; Supervision; Validation; Writing—review & editing. Samuel Heuts: Conceptualization; Data curation; Formal analysis; Methodology; Visualization; Writing—original draft.

Reviewer information

Reviewer information European Journal of Cardio-Thoracic Surgery thanks Sven Peterss, Tim Berger, and the other anonymous reviewers for their contribution to the peer review process of this article.

ETHICAL APPROVAL

Not applicable, as the current study comprises data derived from a systematic review of literature.

REGISTRATION

This study was preregistered and updated in PROSPERO (initial registration date 18 July 2022 [CRD42022345024], updated with the current study on 11 November 2023).

REFERENCES

ABBREVIATIONS

ABBREVIATIONS
 
• ATAAD

  Acute type A aortic dissection

 
• CI

  Confidence interval

 
• HR

  Hazard ratio

 
• IPD

  Individual patient data

 
• K-M

  Kaplan–Meier

 
• RCS

  Restricted cubic splines

 
• SD

  Standard deviation

Author notes