Patient-specific ascending aortic intervention criteria

Abstract

OBJECTIVES

Ascending aortic aneurysms pose a different risk to each patient. We aim to provide personalized risk stratification for such patients based on sex, age, body surface area and aneurysm location (root versus ascending).

METHODS

Root and ascending diameters, and adverse aortic events (dissection, rupture, death) of ascending thoracic aortic aneurysm patients were analysed. Aortic diameter was placed in context vis-a-vis the normal distribution in the general population with similar sex, age and body surface area, by conversion to z scores. These were correlated of major adverse aortic events, producing risk curves with ‘hinge points’ of steep risk, constructed separately for the aortic root and mid-ascending aorta.

RESULTS

A total of 1162 patients were included. Risk curves unveiled generalized thresholds of z = 4 for the aortic root and z = 5 for the mid-ascending aorta. These correspond to individualized thresholds of less than the standard criterion of 5.5 cm in the vast majority of patients. Indicative results include a 75-year-old typical male with 2.1 m² body surface area, who was found to be at increased risk of adverse events if root diameter exceeds 5.15 cm, or mid ascending exceeds 5.27 cm. An automated calculator is presented, which identifies patients at high risk of adverse events based on sex, age, height, weight, and root and ascending size.

CONCLUSIONS

This analysis exploits a large sample of aneurysmal patients, demographic features of the general population, pre-dissection diameter, discrimination of root and supracoronary segments, and statistical tools to extract thresholds of increased risk tailor-made for each patient.

Open in new tabDownload slide

INTRODUCTION

Acute Stanford type A aortic dissection is a catastrophic medical emergency with a high attendant morbidity and mortality. In ascending thoracic aortic aneurysm patients, size-triggered elective repair to prevent dissection is the mainstay of clinical care. International guidelines have long recommend prophylactic surgery at a diameter of 5.5 cm for non-syndromic ascending thoracic aortic aneurysms, a threshold derived from seminal retrospective [1] and prospective [2, 3] studies.

However, this size criterion is imperfect; according to data from the International Registry of Acute Aortic Dissection (IRAD), 60% of dissected ascending aortas measure <5.5 cm, and 40% <5.0 cm [4]. Recent studies [5–7] have unveiled an instantaneous increase in ascending aortic diameter (7–8 mm) at the moment of dissection, highlighting the fact that the true aortic size at which dissection actually occurred was even smaller. It has thus been argued that the recommendation of elective surgical repair at 5.5 cm has been set too ‘rightward’ [4, 5, 8], and a ‘left-shift’ is warranted for earlier repair at ∼5 cm.

Another problematic aspect of a ‘universal’ 5.5 cm threshold for all non-syndromic ascending aortic aneurysm patients is the lack of individualization. Lower surgical thresholds may be considered for patients of small stature [9], resonating the abstract impression among clinicians that males, older and taller individuals are entitled to larger aortas. Individualization attempts have been made, pivoting around 3 basic characteristics; Sex, Age, and Body Surface Area (‘SAS’). Koechlin et al. pointed out that the mean aortic root diameter at the time of dissection is significantly smaller in females [10]. Pape et al. underlined that younger age is an independent predictor of dissection at an aortic diameter of <5.5 cm [11]. Davies et al. shed light on body surface area (BSA) corrected risk [12].

In our effort to refine this 5.5-cm criterion for the ascending aorta, we have recently elaborated the unique natural behaviour of aortic root (‘root’) and mid-ascending aortic (‘mid’) aneurysms finding root dilatation more dangerous [13]. The present study aims to add further granularity by individualization based on the SAS variables.

PATIENTS AND METHODS

This study was approved by the Human Investigation Committee of Yale University School of Medicine (IRB # 1609018416, date of approval: 23 September 2018), and it adheres to the TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) statement [14]. Patients’ informed consent was waived due to the study being a medical record review.

Patients

The Aortic Institute at Yale-New Haven database was interrogated for patients with ascending aortic diameter of at least 3.5 cm, recorded from 1990 to 2019. One retrievable computed tomography or magnetic resonance imaging scan was required. For patients with 2 or more scans, only the last one was considered. As the root and mid diameters were re-measured, echocardiograms were not included. Neither transoesophageal nor transthoracic echocardiography permits multiplanar reconstruction of the root, and the mid and upper portions of the ascending aorta can evade adequate visualization by echocardiographic means. Patients with chronic or type B aortic dissection, traumatic aortic injury, past aortic root or ascending aortic surgery, or congenital aortic malformations were excluded (Supplementary Material, Fig. S1).

Aortic imaging and dissection-related diameter correction

The mids (mid-ascending diameters) of our patient cohort were measured perpendicularly to the centreline, approximately at the level of the pulmonary artery bifurcation or at the point of maximal dilation. Root measurement is a matter of ongoing debate [15]. In this study, it was defined for patients with trileaflet aortic valves as the average of 3 distances, from each sinus to the opposite commissure; or patient with bicuspid valves, it was defined as the average of 2 distances, parallel and perpendicular to the long axis of the valve opening. These valves were labelled bicuspid via echocardiographic or direct intraoperative visualization. Marfan syndrome was diagnosed genetically or clinically by the senior author (John A. Elefteriades) based on the constellation of signs. Aortic measurements were performed by Paris D. Kalogerakos and confirmed by Mohammad A. Zafar or Bulat A. Ziganshin. Radiology reports were consulted, and any emerging measurement discrepancies were resolved in a core meeting.

In line with data from recent studies [5–7], pre-dissection aortic size in patients presenting with acute type A dissection was estimated using the formula published by our group [5]. Regardless of antegrade or retrograde [6] propagation, the mid enlarges by roughly 0.8 cm [5, 13] at the moment of dissection. Interestingly, the root diameter remains unaffected [6].

Endpoints

State-issued death certificates were obtained and analysed to ascertain, as best possible, the precise cause of death for each patient. The end-points for this study were acute type A aortic dissection, ascending aorta rupture, pre-emptive surgery and death (Table 1). After reaching an end-point, or at the end of data collection, patients were censored. The time elapsed from the CT scan to censoring varied from practically zero to a few months; hence, it was considered negligible. Given the extremely high mortality of rupture and untreated dissection, such undiagnosed cases would be registered as deaths.

Two composite end-points (CEs) were calculated: (i) acute type A dissection, rupture and ascending aorta-related death; and (ii) acute type A dissection, rupture and all cause death. The 2nd CE includes non-aorta-related deaths. A handful of these deaths, however, are likely to be undiagnosed or misclassified aorta-related deaths. Therefore, the 2nd CE includes all aorta-related deaths but, by including non-aorta-related deaths, it overestimates the true aortic risk (an ‘upper limit’ of sorts). Contrarily, the 1st CE includes only aorta-related deaths, but not their entirety, thus underestimating the true aortic risk (a ‘lower limit’ of sorts). Positive family history was defined as at least 1 relative with a known or suspected aneurysm in any vascular distribution or a dissection confirmed on imaging or autopsy.

Aneurysmal patients are outliers in the normal distribution describing the aortic diameter of the general population

Any cohort containing patients with aortic pathology does not represent the general population. Aortic diameter values of the cohort, the hallmark of aortic health, are skewed, while for the general population, they are normally distributed [16–26]. To overcome this selection bias, the general population must be included in the study, so as to put the ‘diseased’ diameter into proper context (Supplementary Material, Fig. S2). Each aneurysmal patient should be regarded as an outlier of the general population’s normal distribution (ND).

General population and subgroups of common traits

The SAS-based distributions are normal [16–26] (Supplementary Material, Table S1) and describe more accurately the dispersion of aortic sizes within subgroups of similarly aged and body-sized males and females.

Classification of aortic size based on Sex, Age, and Body Surface Area-based normal distributions

Regarding the severity of the enlargement, the aorta was defined as ‘normal’ for up to 2 standard deviations (SDs), as ‘dilated’ for 2–4 SDs and as ‘aneurysmal’ over 4 SDs. Regarding the location of the enlargement, it was defined as ‘marfanoid’ (root—Table 2), ‘supracoronary’ (mid—Table 3) or ‘tubular’ (root and mid—Supplementary Material, Table S3).

Aneurysmal patients are outliers of the Sex, Age, and Body Surface Area-based normal distributions

The degree of aneurysmal disease could be better appreciated if each patient is compared to all persons sharing the same SAS variables. Each aneurysmal patient should be regarded as an outlier of the respective ND. The SAS variables determine the exact ND, while the actual aortic measurement determines the exact place on that ND.

Multiple normal distributions and z scores

The general population is easy to study but jumbles all patients into a single ND, concealing the impact of the SAS variables. In order to reveal this impact, the study is oriented towards the SAS-based NDs, which are tricky to compare because each one has a different mean and SD. However, they can all be described by a common metric, the z score, which is a useful and accurate tool. Even though each ND has a different mean (in cm), all these means can be expressed as z score = 0. Likewise, every z score locates the same corresponding point of relative size across all NDs. In this way, all healthy and diseased people are normalized, according to the SAS variables, into a family of NDs. Expressing aortic size with the z scores allows us to appreciate the relative degree of aortic enlargement, a feat that cannot be achieved by looking at the crude size measurement.

Risk estimation based on z scores

The next step is correlating the z scores with the observed risk of Major Adverse Aortic Events (MAAEs). The goodness of fit of logistic regression and polynomial regression were estimated to choose how to convey the risk, eventually favouring the latter. The term risk is used interchangeably with the term probability, which is defined as the ratio of the total number of events to the total number of aneurysms for each size group. The presented methodology explores the correlation of z score and risk, regardless of MAAE’s timing or follow-up time. Analyses that focus on time, or on each variable separately, are, inherently, not a part of this specific approach. Risk, namely the percentage of eventful aortas among size groups of 0.5 z score, was plotted against the z scores, and the resultant curve is an ordinary least square regression, i.e. a single, high-order polynomial trend line that best fits the data. This curve (risk of MAAEs versus aortic z score) returns the numerical value of the observed risk and can be inspected for possible ‘hinge points’, i.e. points of steep risk increase. Against this background, the presented model does not predict the exact size that an aorta will dissect, neither claims to estimate the exact, true risk. Its worth rests on notifying over which size an aorta poses a rapidly growing threat. The coefficient of determination (R²) represents the proportion of the variance for adverse events that can be explained by aortic size in the regression model. The Risk Calculator estimates risk as the ordinate of the above-mentioned curve. The abscissa is the z score, which the Calculator computes based on the SAS variables and the actual aortic measurements, using formulas that are analysed later.

Individualized hinge point estimation via conversion of z scores to actual aortic sizes

Appraisal of the z score-based hinge points

In the analysis below, the z score-based hinge point is defined as that point in the risk curve where the curve’s almost zero slope turns into a substantial positive number. Visually, the curve runs almost horizontally and, at this ‘hinge point’, it turns upwards. These points will be critically appraised, by converting it to aortic diameter measured in cm, for various exemplary patients,—for example, a typical male (75 years old with 2.1 m² BSA), a petite female (75 years old with 1.7 m² BSA) and a tall male in his youth and in his senescence (50 or 80 years old with 2.3 m² BSA). These individualized, z score-based, hinge points for the 4 exemplary patients will represent a SAS adjusted estimate of the aortic sizes that carry an increased risk of adverse events. These aortic sizes are thresholds of increased risk, and they will be compared.

Aortic root and mid-ascending aorta discrimination

Recently, we have demonstrated that aortic root and mid-ascending aorta have unique natural histories and deserve individualized management [13]. In this context, in the present analysis, the study of z score-based individualized hinge points is undertaken separately for the mid and for the root.

Comprehensive mathematical formulas that describe the sex, age and body surface area-based normal distributions

The present study requires mathematical formulas that describe the SAS-based ND to which each patient belongs. Despite the plethora of meritorious studies that focus on the normal aortic diameter [16–26], only 1 [23] satisfied the strict specifications set during study design. Most authors chose among the following: reporting the means and/or the Upper Limit of Normal (ULoN) without presenting the underlying linear correlation, withholding parts of the formulas or the SDs, reporting about the descending aorta, reporting normative tables or nomograms, rounding up numbers to the 1st decimal place thus initiating a deviation that would inflate in subsequent steps and relying on simple linear formulas with low R² even though exponential formulas are arguably better in describing growth in biological systems [27]. Our methodology required 8 complete SAS-based mathematical formulas that provide precise estimates for the mean and the SD, for the root and for the mid, and for men and for women (Supplementary Material, Table S2), as detailed in the study by Campens et al. [23]. They reported exponential formulas that estimate with high R² (67–84%) the ULoN given the SAS, and the z score given the SAS and actual aortic measurement. A disadvantage was grounding their work on echocardiograms, which are subject to interobserver bias and inferior to electrocardiogram-gated CT scans in visualizing dilatation. By manipulating the above-mentioned formulas, we get the equations for mean and SD (Supplementary Material, Table S2 with Formula Derivation). Of note, there is no actual aortic measurement in this subprocess. The formulas in Supplementary Material, Table S2 integrate sex, age, BSA and location (root or mid) to return the patient-specific mean and SD of the unique ND to which each patient belongs.

At this point, we have 2 different z scores for each one of the 1162 patients: 1 for the root and 1 for the mid. The next step is correlating the risk of adverse events with these z scores, separately for the root and the mid. It is noteworthy that the correlation of risk with root size ignores mid size, even though the risk increasing effect of the latter permeates this association. The upshot is confined to the, arguably benevolent, roots with low z scores, which are expected to exhibit pseudo-high risk because of the aneurysmal mids that dissect. A similar fallacy holds true for the mids of low z scores.

RESULTS

Patient characteristics and fit of regression models

The patients’ characteristics and the end points are listed in Table 1. Based on the SAS variables and the actual aortic measurements, each patient was placed in the respective ND by deriving the z scores with formula z.

Even though the Hosmer–Lemeshow tests did not rule out logistic regression (root: significance = 0.702, mid: significance = 0.109), the Nagelkerke R squared tests were very small (root: Nagelkerke R² = 0.06, mid: Nagelkerke R² = 0.05), and the receiver operating characteristic curves showed classifiers with random performance (Supplementary Material, Fig. S3), therefore logistic regression was rejected. The polynomial regression was a much better fit with R² values of 86% and 68% for the 1st CE (Figs 1 and 2), therefore it was preferred.

Risk estimation based on z scores

These expressions are generalized forms, thus true for everyone. When mean and SD are substituted with the numerical values provided from the respective formulas (Supplementary Material, Table S2), plugged in with the SAS variables of each patient, the equations R and M return the patient-specific hinge points for the root and the mid-ascending diameter, respectively. These hinge points can be compared to the actual aortic measurement performed for each patient. If the latter is greater than the former, pre-emptive surgery should be considered. Note in Figs 1 and 2 that the danger hinge point for the root is smaller (SD 4) than for the mid (SD 5).

Conversion of z score-based hinge points to actual aortic sizes for the 4 exemplary patients, and the calculator

The definition of ULoN at approximately $mean+2×SD$ was known a priori. The predictions of this study are the individualized hinge points (Table 4). A user-friendly calculator is provided online. A single patient print-out is provided in Fig. 3.

The user inputs patient sex, age, height, weight and maximal aortic diameter at 2 locations: the root and the mid. The calculator will output the patient-specific hinge point, z score and observed risk along with its increase rate, separately for the root and for the mid. These outputs are to be interpreted as follows:

The hinge point (mm) indicates the aortic diameter beyond which the associated risk of MAAEs increases steeply, and it is printed in red colour for emphasis.

The z score (unitless) indicates how many SDs away from the mean lies the aortic diameter of the patient under scrutiny, and it turns red if that diameter has reached or surpassed the respective hinge point.

The observed risk (%) is indicative of the associated risk of MAAEs, and its increase rate (%) connotes the trend of this risk to increase even further. These are printed in red colour to connote that the patient is in enhanced jeopardy, past the respective hinge point.

DISCUSSION

Overview

The revelation that the aorta enlarges abruptly at the moment of dissection [5–7], combined with the recent observation that the root is more malevolent than the mid [13, 28], adds more nuances to the surgical decision-making process. To further prevent dissection, earlier surgical repair must be cautiously implemented. During this ‘left shift’, patients’ SAS should be factored in to achieve further personalization.

Numerous studies have focused on the SAS variables and elucidated their impact on the incidence, related morbidity and surgical results after type A aortic dissection [10, 12, 29]. This study attempts to incorporate available ascending thoracic aortic aneurysm clinical data, aneurysmal aortic size when placed in the context of ‘normal’ aortic size, the emerging understanding of unique behaviour of root versus mid aneurysms and statistical tools into a comprehensive, personalized risk estimation system (graphical abstract). Individualized thresholds of increased risk of MAAEs can be deduced based on SAS variables. Furthermore, an intuitive classification system that refines the definitions of ‘dilated’ and ‘aneurysmal’ is presented.

Individualized classification and hinge points

The numerical results exhibit intuitive hinge points that are tailor-made for each patient (Table 4) and adjust surgical thresholds based on patient body size, as proposed in the European guidelines [9]. These are intended to aid clinicians by informing one side of the scale in the surgical decision-making process, the aneurysm-related risk. To clarify the complementary role of the equations R and M, and the formula z in this decision-making, Table 5 provides an algorithm with 2 alternative paths that clearly suggest or discourage surgery based on risk increase. The other side of the balance scale is the surgical risk. Clinical judgement dictates the best balance of aneurysm risk versus surgical risk for each individual patient.

Limitations

The retrospective design of this study is the main source of its limitations. Even though pre-emptive surgery confounds the results by obviating further aortic enlargement and its ensuing adverse events, surgery cannot ethically be withheld. Conducting randomized trials with arms of aneurysmal patients deprived of surgery raises strong ethical concerns, considering the fatality of acute type A dissection. Thus, retrospective observation emerges as an important tool for exploring the natural history of aortic dilatation. Misclassification bias is probable for some patients with unknown or other cause of death, as in both cases, death might actually be aortic. Despite this pitfall, the true, unknown occurrence of death related to the ascending aorta lies between known aortic death and all-cause death. Under this light, the 1st and the 2nd CEs could be perceived as the lower and upper boundaries of the true occurrence of MAAEs. Given that death misclassifications are possible but not common, the 2nd CE likely overestimates the risk by a considerable margin. Our focus is concentrated on the 1st CE, which underestimates the risk; however, it appears to be acceptably close to the true incidence. Regardless, exact risk prediction is difficult because pre-emptive surgery confounds the results and precludes calibration. The tangible offering is unveiling over which size an aorta poses a rapidly growing threat.

Sampling bias is a major worry in retrospective research. The representativeness of the included patients has been confirmed in an earlier work of our group [13]. Healthy individuals are less likely to seek medical attention, thus individuals with comorbidities, such as Marfan syndrome, and a normal sized aorta, up to ∼4.0 cm, are more likely to present with MAAEs, causing an overestimation of risk in this size group. This fallacious increase is reflected in the left portion of Figs 1 and 2. The other reason of risk overestimation in aortas with small roots (Fig. 1) is that several of these aortas dissect due to their aneurysmal mids. Similarly, Fig. 2 does not convey information about the roots, thus the aortas with small mids exhibit pseudo-high risk due to aneurysmal roots. The study sample becomes representative over 4.5 cm. This study does not include a test set and a separate validation set, as might be done via an Artificial Intelligence approach. We do not feel that the ‘n’ of this study, while very large for a clinical investigation, would support such test set/validation set analysis.

It could be argued that this study might provide skewed hinge points due to a basic limitation, that the NDs are derived from a single source [23]. Apart from the strength of this study and the intuitive numerical results, the current analysis can easily be fine-tuned to include other estimates of ND. Other future developments such as adding more variables to the SAS approach, substituting BSA with height [29], or even changing the size metric from aortic diameter to aortic length [30], can also be accommodated by this adaptable methodology. In the case of length, which is a relatively neglected aortic dimension, its inclusion might catapult the predictive value of this model. The proposed threshold length of 11 cm might be more reliable than the current diameter criterion [30]; however, this metric lacks individualization. A similar analysis using aortic length instead of diameter could have been achieved, but the normal length of the ascending aorta has not been studied thoroughly, thus it is technically infeasible to incorporate it in the model at the present time. The inclusion of pathological factors, such as hypertension and connective tissue disease, in the z score-based risk estimation warrants a careful statistical approach and further studies.

The ‘universal’ surgical threshold for ascending aortic aneurysm patients should be replaced with individualized thresholds, which factor in sex, age, BSA and aneurysm location; root or mid. An evidence-based methodology for unveiling these tailor-made thresholds has been presented. For the vast majority of patients, their tailor-made mid thresholds are lower than the overall 5.5 cm surgical threshold in society guidelines, with the root thresholds being even lower. The approach presented herein can be expected to enhance medical knowledge, patient care and procedural skills in the assessment and care of patients with ascending thoracic aortic aneurysm. Advanced statistical and epidemiological techniques are translated directly into clinical implications and guidelines. An advanced automated risk estimating on-line tool is developed.

SUPPLEMENTARY MATERIAL

Supplementary material is available at EJCTS online.

FUNDING

This study was not funded.

Conflict of interest: Dr John A. Elefteriades is a Principal of CoolSpine. The other authors have no disclosures.

DATA AVAILABILITY

The full dataset cannot be shared publicly as it contains protected health information. Data can be provided upon reasonable request and ethical approval from the Human Investigation Committee of Yale University School of Medicine.

Author contributions

Paris D. Kalogerakos: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Visualization; Writing—original draft. Mohammad A. Zafar: Data curation; Visualization; Writing—original draft. Yupeng Li: Formal analysis; Investigation; Visualization; Methodology. Hesham Ellauzi: Data curation; Resources. Sandip K. Mukherjee: Investigation; Resources. Bulat A. Ziganshin: Data curation; Supervision. John A. Rizzo: Supervision; Methodology. John A. Elefteriades: Supervision; Writing—review and editing.

Reviewer information

European Journal of Cardio-Thoracic Surgery thanks Gabriele Piffaretti, Guido Gelpi, Yutaka Okita and the other anonymous reviewers for their contribution to the peer review process of this article.

REFERENCES

ABBREVIATIONS

ABBREVIATIONS
 
• BSA

  body surface area

 
• CE

  Composite end-points

 
• IRAD

  International Registry of Acute Aortic Dissection

 
• ND

  Normal distribution

 
• SD

  Standard deviation

 
• TRIPOD

  Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis