Background
Coronary vasomotor disorders, occurring both at microvascular and epicardial level, have been
demonstrated as responsible for myocardial ischemia in a sizeable group of patients
undergoing coronary angiography (CAG), with clinical manifestations ranging from ischemia
with non-obstructive coronary arteries (INOCA) to myocardial infarction with non-obstructive
coronary arteries (MINOCA), along with life-threatening arrhythmias and sudden cardiac death.
Intracoronary provocative testing with administration of acetylcholine (ACh) at the time of
CAG may elicit epicardial coronary spasm or microvascular spasm in susceptible individuals,
and therefore is assuming paramount importance for the diagnosis of functional coronary
alterations in patients with suspected myocardial ischemia and non-obstructive coronary
artery disease (CAD). However, previous studies mainly focused on patients with INOCA, whilst
MINOCA patients were often underrepresented . In addition, intracoronary provocative testing
is still largely underused in clinical practice, probably because of concerns regarding the
risk of complications, especially in the acute clinical setting. Of note, the landmark
"Coronary Microvascular Angina" (CorMicA) trial demonstrated that a strategy of adjunctive
invasive testing for disorders of coronary function in patients with non-obstructive CAD
linked with stratified medical therapy is superior to usual care in improving patients'
outcomes, including reduction in angina severity and better quality of life. Therefore,
assessing the presence of coronary vasomotor disorders is of mainstay importance in order to
implement the optimal management and improve clinical outcomes.
Of interest, the investigators recently demonstrated that performing an ACh provocative test
in patients with myocardial ischemia and non-obstructive coronary arteries is safe with a low
rate of complications, without differences between patients presenting with INOCA or MINOCA.
In particular, a previous history of paroxysmal atrial fibrillation (AF), a
moderate-to-severe left ventricle (LV) diastolic dysfunction and a higher corrected QT (QTc)
dispersion at baseline electrocardiogram (ECG) were independent predictors for the occurrence
of complications during the test and, therefore, patients with these characteristics may be
those requiring particular attention during the test. Moreover, the investigators
demonstrated that performing an ACh provocative test has relevant prognostic implications, as
patients with a positive test have a higher risk of major adverse cerebrovascular and
cardiovascular events (MACCE) at follow-up, and, therefore, performing an ACh test can help
in stratifying the prognosis, especially in MINOCA patients, suggesting the presence of a net
clinical benefit deriving from its use. Furthermore, the investigators recently demonstrated
that some clinical (MINOCA as clinical presentation and elevated circulating levels of
C-reactive protein) and angiographic (presence of myocardial bridging) features are
independent predictors for a positive response to ACh test.
Of interest, the identification of clinical predictors for a positive ACh test could allow
the development of predictive models for a positive or negative response based on clinical
and/or angiographic features readily available in the catheterization laboratories, thus
helping clinicians in the diagnosis of coronary vasomotor disorders even in patients at high
risk of complications (e.g.: history of AF, LV diastolic dysfunction, long QTc interval or
QTc dispersion at baseline ECG, bradyarrhythmia). Moreover, the implementation of such
predictive models in clinical practice could avoid the need of performing a provocative test,
significantly reducing the duration of invasive procedures as well as the associated risks
and allowing a fast determination of the most appropriate treatments and clinical paths, an
efficient planning, and a parsimonious use of medical resources. In addition, developing
predictive models for the risk of future cardiovascular events could help clinicians in the
prognostic stratification and the choice of therapeutic strategies in the post-discharge
management, possibly identifying those patients that may need a more aggressive therapy and a
closer follow-up.
Therefore, the investigators hypothesize that:
Clinical predictors for a positive ACh test response could be identified, allowing the
development of predictive models and/or clinical risk scores that could help clinicians
in the diagnosis of coronary vasomotor disorders and the implementation of the most
appropriate management.
A positive ACh test could be associated with a higher rate of adverse cardiovascular
events at follow-up, thus helping in the prognostic stratification of INOCA and MINOCA
patients and identifying those that may need a more aggressive therapy and a closer
follow-up.
Primary objective
To derive and validate predictive models/clinical risk scores able to predict a positive ACh
test response in INOCA and MINOCA patients basing on clinical and/or angiographic features.
Secondary objective
To derive and validate predictive models/clinical risk scores able to predict a worse
clinical outcome in terms of major adverse cardiovascular and cerebrovascular events (MACCE),
defined as the composite of cardiovascular death, nonfatal myocardial infarction (MI),
hospitalization due to unstable angina (UA), and stroke/transient ischemic attack (TIA) in
INOCA and MINOCA patients basing on clinical and/or angiographic features.
Study design
Observational study.
Sample size calculation
Up to our knowledge no study has investigated the creation of a potential score for Ach test
positivity. Hence this would represent the first and, as such, is includable among pilot
studies and, therefore, no formal sample size calculation is needed, but all of patients
satisfying inclusion criteria can be included. Based on the study design, which pertains the
creation and validation of a score, which would require a training and validation cohort, and
will alongside imply the use of regression methods, the investigators plan to include 600
patients. Such a sample size would allow for the stratification in two cohort and the
analysis of the approximately 50 covariates included in the study. In fact, according to van
Smeden, events per variable (EPV) may go beyond the common rule of EPV≥10.
Statistical analysis
Descriptive analysis and between-groups comparisons
The sample will be described in its demographic, anthropometric, clinical, instrumental,
variables through descriptive statistical techniques. In depth, qualitative variables will be
expressed by absolute and relative percentage frequencies. Quantitative variables, indeed,
will be reported either as mean and standard deviation (SD) or median and interquartile range
(IQR), respectively in the case they were normally or not normally distributed. Their
distribution will be previously assessed by the Shapiro Wilk test. Between groups differences
in the demographic, clinical, laboratory and pathologic features will be assessed by the Chi
Square or the Fisher's exact test as for qualitative variables (with Freeman- Halton's
extension when appropriate), whilst quantitative variables will be evaluated either by the
Student's t test or the Mann- Withney U test, according to their distribution.
Derivation and validation of clinical risk scores
Data used for score development will be derived from a prospectively enrolled sample of 550
NOCAD patients, consecutively admitted to the Department of Cardiovascular Sciences of
Fondazione Policlinico Universitario A. Gemelli IRCCS, Rome (Italy). There is no generally
accepted approach for the estimation of the sample size for derivation of score prediction
models. Hence, the investigators based for the derivation of the score to include in the
multivariable model a number of covariates consistent with the most recent rules on the
minimum number of events per variable needed. The investigators will randomly allocate the
participants to two cohorts, one cohort will be used to develop the score model (derivation
cohort), and the other to validate and assess the diagnostic abilities of the score
(validation cohort). Multiple imputation will be applied to handle missing data, by "imputeR"
R package. Univariable and multivariable regression models will be performed on the
derivation cohort to identify independent predictors of a positive ACh test to be included in
the scoring system. In depth, the investigators will compute Odd Ratios (ORs) and 95%
Confidence Intervals (CIs) of the predictor candidates for the outcome (i.e., positive Ach
test) by univariable logistic regression models. Predictors to be included in the
multivariable model will be selected based on univariable analysis (p<0.05 or suggestive,
i.e. 0.05<p<0.10) and expert opinion. The multivariable logistic regression model will
produce β coefficient and Standard Errors (SE) for each variable. The performance of the
model will be assessed based on diverse methods, such as Somers' Dxy rank correlation,
C-index, Nagelkerke R2 value, calibration intercept and slope, and Brier score. Finally, the
Hosmer-Lemeshow goodness-of-fit test will allow for the calibration in the derivation cohort.
Calibration plots will further provide a graphic representation of the association between
the predicted and observed outcome by locally weighted scatterplot smoothing. "rms",
"predtools" and "magrittr" R packages will be used for the whole analyses set. Internal
validation of the model will be performed based on a bootstrap procedure. The investigators
will then pass to develop a scoring system to predict the outcome providing an integer value
to each predictor included in the scoring system based on each variable's β coefficient in
the derivation cohort. Appropriate cutoff values will be set for a rule-in and rule-out
approach to help in decision-making. In depth, quantitative independent predictors will be
further transformed into either ordinal or nominal qualitative variables. Transformation will
be performed seeking for one or more optimal cut-points by appropriate selection methods
based on ROC curves analysis, by mean of "pROC" R package. Particularly, "OptimalCutpoints" R
package will be used applying the SpEqualSe selection method, which returns the highest
accuracy. The diagnostic abilities (i.e., sensitivity, specificity, positive likelihood
ratio, and negative likelihood ratio) of each score will be then calculated, and the patients
will be divided into 3 groups (positive, negative, suspect). For validation, the developed
score will be applied to the validation cohort, and the discrimination and calibration
performances will be described, as aforementioned. The overall performance will be described
in terms of Sensitivity, Specificity, Accuracy (or Positive Predictive Value, PPV), F1 Score,
Accuracy, False Positive Rate (FPR), False Discovery Rate (FDR) and False Negative Rate
(FNR). Statistical analyses will be carried out using R software, version 4.2.0 (CRAN ®, R
Core 2022).
Derivation and validation of predictive models using artificial intelligence/machine learning
models (second phase)
In the second phase of this study, the investigators further plan to develop a predictive
model of MACCE in the studied population, as well as to extend the study out of our clinical
facility to potentially validate the models also externally. In this context, the
investigators foresee to further enhance the analysis by adding Machine Learning methods,
such as XGBoost, Random Forest, Neural Networks, which will be chosen based on the type of
data and question to be answered. ML methods will be applied in Python software.