Development and validation of a prognostic model of postnatal circulation in fetuses with pulmonary atresia or critical pulmonary stenosis with intact ventricular septum: protocol for an observational cohort study with prospective validation

Abstract

Background

Pulmonary atresia or critical pulmonary stenosis with intact ventricular septum (PA/CS-IVS) is a rare and heterogeneous congenital heart disease with highly variable postnatal outcomes. Existing prenatal prognostic models are limited by small sample sizes, heterogeneity in echocardiographic variables, and suboptimal validation. Robust prenatal prediction of postnatal circulation is essential for parental counselling, perinatal planning, and selection of candidates for potential fetal intervention.

Methods

This international observational cohort study will include non-intervened fetuses with a prenatal diagnosis of PA/CS-IVS diagnosed between 16 + 0 and 28 + 6 weeks of gestation. Retrospective cases (from 2013 onwards) will be used for model development, and prospectively recruited cases for external validation. Standardized prenatal and postnatal echocardiographic data will be collected using predefined protocols and reviewed by a central core laboratory. Multivariable logistic regression models will be developed following TRIPOD recommendations, with internal validation by bootstrapping. Secondary analyses will explore longitudinal prenatal data and machine-learning approaches.

Discussion

This study aims to develop and validate a clinically applicable prognostic model for PA/CS-IVS using standardised, multicentre data. The results are expected to improve prenatal counselling and perinatal management, and will support the design of future interventional trials.

Trial registration

ClinicalTrials.gov, NCT07095829.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12884-026-08947-z.

Background

Pulmonary atresia or critical pulmonary stenosis with intact ventricular septum (PA/CS-IVS) is a rare congenital heart disease characterised by severe obstruction of the right ventricular outflow tract and a wide spectrum of right ventricular underdevelopment [1, 2]. The reported incidence is low, but the condition carries substantial morbidity and mortality, with outcomes that vary markedly according to postnatal circulation strategy [3, 4]. Affected neonates may achieve biventricular or one-and-a-half ventricular circulation, or may require univentricular palliation or cardiac transplantation, pathways that are associated with profoundly different long-term survival, quality of life, and healthcare burden [5, 6].

From an obstetric perspective, accurate prenatal prognostication is essential. Parents must be counselled regarding expected postnatal treatment pathways, survival, and long-term outcomes. Moreover, risk stratification informs the intensity and frequency of fetal surveillance, timing and location of delivery, and coordination of multidisciplinary perinatal care in specialised centres [5]. In selected cases, prenatal identification of fetuses at very high risk of univentricular outcomes has also raised interest in fetal cardiac intervention, such as fetal pulmonary valvuloplasty to improve right heart growth in utero and promote biventricular circulation, a procedure that carries potential benefits but also non-negligible maternal and fetal risks [7].

Despite its clinical importance, prenatal prediction of postnatal circulation in PA/CS-IVS remains challenging. Several multiparametric scores and prediction models have been proposed; however, most are derived from small, single-centre cohorts, span long recruitment periods with evolving management strategies, and use heterogeneous definitions and measurement techniques [8]. As a result, external validation has been limited, and existing models tend to underestimate the risk of univentricular circulation. Recent systematic reviews and validation studies have highlighted these limitations and emphasised the need for standardised data collection, harmonised echocardiographic protocols, and robust multicentre cohorts [9, 10]. In particular, there is a lack of prospectively collected data with centralised image review that would allow reliable assessment of fetal cardiac growth and haemodynamic evolution over time. Addressing these gaps is critical to improving prenatal counselling and perinatal decision-making in this complex condition.

To respond to these unmet needs, we designed a large, international, observational cohort study with standardised echocardiographic assessment and centralised review to develop and validate a robust prognostic model for PA/CS-IVS.

Hypothesis and aims

Hypothesis

A prognostic model based on standardized prenatal echocardiographic parameters obtained between 16 and 28 weeks of gestation can accurately predict transplantation-free survival with biventricular or one-and-a-half ventricular circulation at 2 years of postnatal age in fetuses with PA/CS-IVS. Incorporation of longitudinal prenatal echocardiographic data reflecting cardiac growth and haemodynamic evolution will improve prognostic performance. Specific prenatal features are hypothesised to be associated with the development of right ventricle-dependent coronary circulation.

Primary aim

To develop and externally validate a prognostic model for the prediction of transplantation-free survival with biventricular or one-and-a-half ventricular circulation at 2 years of postnatal age in non-intervened fetuses with PA/CS-IVS diagnosed between 16 + 0 and 28 + 6 weeks of gestation.

Secondary aims

1. To develop and validate a prognostic model predicting transplantation-free survival with biventricular or one-and-a-half ventricular circulation at 1 year of postnatal age.

2. To develop a prognostic model incorporating longitudinal prenatal echocardiographic measurements to assess the impact of fetal cardiac growth and haemodynamic changes on 2-year postnatal outcomes.

3. To describe prenatal and postnatal outcomes in non-intervened fetuses with postnatally confirmed PA/CS-IVS, including fetal and infant mortality, number and type of postnatal interventions, oxygen saturation at 2 years of age, and need for cardiac transplantation.

Methods

Study design

This is an international observational cohort study. Retrospective cases collected from 2013 onwards will be used for model development, and prospectively recruited cases for model validation. Only fetuses not undergoing prenatal cardiac intervention will be included in the analysis. Participation in the study will not modify routine prenatal, perinatal, or postnatal management. To minimise bias related to temporal changes in management, retrospective cases are restricted to specialised centres with established expertise and identical variable definitions to those used prospectively.

Study setting and participants

Pregnant women carrying fetuses with suspected PA/CS-IVS will be recruited for the prospective cohort from tertiary referral centres with expertise in fetal cardiology. Eligible participants will be identified during routine clinical care and invited to participate after confirmation of eligibility. Retrospective cases will be identified by applying the same eligibility criteria through a systematic review of local databases.

Eligibility criteria

Inclusion criteria

Prenatal echocardiographic diagnosis between 16 + 0 and 28 + 6 weeks of gestation of:

1. Absence of antegrade pulmonary valve flow or severely stenotic pulmonary valve with minimal flow

2. Ductal-dependent pulmonary circulation

3. Intact ventricular septum

Exclusion criteria

1. Elective termination of pregnancy

2. Prenatal pulmonary valvuloplasty

3. Unconfirmed diagnosis of PA/CS-IVS after birth

4. Functional PA/CS-IVS (e.g., Ebstein anomaly or monochorionic twin pregnancy)

5. Associated major cardiac defects, except persistent left superior vena cava, aberrant right subclavian artery, or tiny muscular VSD

6. Significant extracardiac anomalies or known genetic syndromes

7. Inadequate image quality or incomplete echocardiographic data

8. Loss to follow-up

Echocardiographic assessment

All included cases will undergo standardised fetal echocardiographic evaluation according to a predefined protocol. The following examinations will be included:

1. Baseline fetal echocardiography at diagnosis (16–28 weeks of gestation)

2. Follow-up fetal echocardiography performed at least 8 weeks after the baseline scan (preferably between 32 and 36 weeks of gestation)

3. First complete postnatal echocardiogram prior to any surgical or catheter-based intervention

A summary of the workflow of prospective cases is presented in Fig. 1.

Fig. 1.

Workflow of prospectively recruited cases

Measurements will include ventricular dimensions, valve diameters, Doppler-derived haemodynamic parameters, and qualitative assessments of ventricular morphology and function. Fetal measurements will be converted to gestational age–adjusted z-scores, and postnatal measurements to body surface area–adjusted z-scores (https://zscore.chboston.org/). A summary of the main collected variables is presented in Table 1 and the full case report form is presented as Supplemental Material 1.

Table 1.

Key variables collected in the study

Category	Variable	Timing	Description/purpose
Maternal and pregnancy characteristics	Maternal age	At inclusion	Descriptive
	Gestational age at diagnosis	At inclusion	Descriptive
	Pregnancy outcome	End of pregnancy	Descriptive
Prenatal fetal echocardiographic predictors	Tricuspid valve z-score	Baseline prenatal scan (16–28 weeks)	Candidate predictor
	Presence of ventriculo-coronary connections	Baseline prenatal scan	Candidate predictor
	Right ventricle/left ventricle length ratio	Baseline prenatal scan	Candidate predictor
	Tricuspid valve/mitral valve inlet ratio	Baseline prenatal scan	Candidate predictor
	Pulmonary annulus/aortic valve ratio	Baseline prenatal scan	Candidate predictor
	Tricuspid regurgitation (none, mild, moderate, severe)	Baseline prenatal scan	Candidate predictor
	Right ventricular inflow duration/cardiac cycle duration (%)	Baseline prenatal scan	Candidate predictor
	Pulmonary atresia vs critical pulmonary stenosis	Baseline prenatal scan	Candidate predictor
Longitudinal fetal variables	Change in ventricular and valve dimensions	Follow-up prenatal scan (≥ 8 weeks later)	Secondary/exploratory
	Change in haemodynamic parameters	Follow-up prenatal scan	Secondary/exploratory
Neonatal echocardiographic assessment	Postnatal ventricular morphology and function	At birth	Descriptive
	Presence of ventriculo-coronary connections	At birth	Descriptive
Primary infant outcome	Transplantation-free survival with biventricular or one-and-a-half ventricular circulation	2 years postnatal	Primary outcome
Secondary infant outcomes	Transplantation-free survival	1 year postnatal	Secondary outcome
	Type and number of postnatal interventions	Up to 2 years	Secondary/descriptive
	Oxygen saturation	2 years postnatal	Secondary outcome
	Cardiac transplantation	Up to 2 years	Secondary outcome

Data collection

Clinical and echocardiographic data will be collected using electronic case report forms within a secure REDCap [11] database (Supplemental Material 1). All data will be anonymised prior to analysis and managed in accordance with applicable data protection regulations.

Core laboratory

In prospectively recruited cases, echocardiographic studies will be uploaded in anonymised DICOM format to a central core laboratory. Measurements will be repeated by core laboratory reviewers blinded to local measurements and clinical outcomes. Discrepancies exceeding 20% will trigger re-evaluation by a second reviewer, the core lab principal investigator. If there is continued disagreement, the core lab will contact the submitting centre for a review of the original measurement, and if this was in error the submitted value will be replaced by the core lab value provided that the image quality is acceptable. If that is not the case, the value will be excluded from the analysis. A similar approach will be used for variables based on qualitative or semiquantitative measurements such as the magnitude of regurgitant jets on Colour Doppler and ventricular function.

Outcome measures

Primary outcome

Transplantation-free survival with biventricular or one-and-a-half ventricular circulation at 2 years of postnatal age.

Secondary outcomes

Transplantation-free survival at 1 year; postnatal diagnosis of right ventricle-dependent coronary circulation; fetal and infant mortality; number and type of postnatal interventions; oxygen saturation at 2 years; need for cardiac transplantation.

Statistical analysis

Sample size

Based on feasibility considerations and prior literature, 8 predictor variables are considered [9]. Therefore, a total sample size of 200 cases is planned, with approximately 160 cases used for model development and 40 cases for validation. This sample size is considered sufficient [12] to develop a multivariable prognostic model with an acceptable number of events per predictor parameter.

Model development

The criteria for transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) [13] will be followed for model development (Supplemental Material 2). A prediction model using logistic regression will be performed, with the outcome “transplantation-free survival with a biventricular or a one-and-a-half repair at two years postnatal age”. After careful literature review, eight preselected variables will be evaluated in a univariate analysis comparing both groups (Tricuspid valve z-score, presence of ventriculo coronary connections, right ventricle/left ventricle length ratio; tricuspid valve/mitral valve inlet ratio; pulmonary annulus/aortic valve ratio; tricuspid regurgitation (none, mild, moderate, severe); right ventricle inflow duration/cardiac cycle duration (%); pulmonary atresia vs severe stenosis), and those with p < 0.10 will be entered into a logistic regression model. Predictor selection was guided primarily by clinical relevance and prior evidence, with univariate analyses used to inform, but not solely determine, inclusion in the multivariable model. Following primary model development, models will be internally validated to decrease optimism through bootstrapping with 1000 iterations.

Machine-learning analyses are exploratory and hypothesis-generating and will not replace the primary regression-based prognostic model. Following primary model development, internal validation, and assessing model performance, four machine learning models (Random Forest, Support Vector Machine Regression, Gradient Boosting Machine, and Artificial Neural Networks) [14] will be developed to predict the primary outcome. All machine learning models will incorporate all the same predictors used to develop the logistic regression model. An iterative matrix process will be performed to assess optimal tuning hyperparameters for all machine learning models, with root mean square error utilized to determine optimal tuning. All models will then be internally validated to reduce overfitting with tenfold cross- validation. Machine learning models will also be used to determine the importance and weight of predictors to facilitate explainability.

Model performance

Model performance will initially be investigated by assessing overall performance, classification, and discrimination. Overall performance will be evaluated through Nagelkerke R, in which the amount of variability in outcomes is explained by the prediction model, and the Brier score, a performance measure for the distance between observed and predicted outcome. Classification will be assessed by sensitivity, specificity, positive predictive value, and negative predictive value. Discrimination, which describes the ability of the prognostic model to distinguish between patients with and without the outcome, will be estimated with the area under the receiver operating characteristic curve (AUROC) and graphed the sensitivity against 1-specificity for consecutive cutoffs for the predicted probability of an outcome.

Model validation

The predictive ability of the models will be subsequently tested based on discrimination, calibration, and measures of overall performance. The measures of overall performance and discrimination will be the same as those described in the development phase. Calibration refers to the agreement between observed and predicted probability. Models will be assessed graphically and estimated with the intercept and slope calibration. In case of perfect fit of the model to the data, calibration intercept is equal to 0 and slope equal to 1. Calibration will be quantified by the calibration slope (with 95% confidence intervals) and graphically by plotting the predicted risk against the observed outcome using a loess smoother.

Additional analyses

Machine learning

Following primary model development, internal validation, and assessing model performance, four machine learning models (Random Forest, Support Vector Machine Regression, Gradient Boosting Machine, and Artificial Neural Networks) will be developed to predict the primary outcome. All machine learning models will incorporate all the same predictors used to develop the logistic regression model. An iterative matrix process will be performed to assess optimal tuning hyperparameters for all machine learning models, with root mean square error utilized to determine optimal tuning. All models will then be internally validated to reduce overfitting with tenfold cross- validation. Machine learning models will also be used to determine the importance and weight of predictors to help inform clinical practice and decision making.

Missing data

Missing data will be handled using multiple imputation when appropriate. All data will be tested for missing values prior to analyses. Missing data will be categorized as missing completely at random, missing at random, or missing not at random. If missing data is below 5%, a complete case analysis will be performed. If missing data is above 5% and < 20%, multiple imputation will be performed. Imputation iterations will be at a minimum of 20 iterations and will proportionally increase with the percentage of greatest missing data. Missing outcomes will be marked prior to imputation, with outcomes imputed. However, imputed outcomes will be excluded from analyses.

Ethics and dissemination

The study has received approval from the institutional review board of the coordinating centre, and local ethical approval will be obtained at each participating site as required. Study findings will be disseminated through peer-reviewed publications and presentations at scientific meetings.

Patient and public involvement

Patients and the public were not involved in the design of this study due to its methodological nature.

Discussion

This study addresses an important unmet need in the prenatal management of pulmonary atresia or critical pulmonary stenosis with intact ventricular septum (PA/CS-IVS), a condition characterised by marked prenatal heterogeneity and uncertainty regarding postnatal circulation. Existing prognostic models tend to underestimate the risk of univentricular outcomes and are frequently limited by small sample sizes, heterogeneous management strategies, and a high risk of overfitting [9, 15].

By providing a standardised, multicentre framework that combines prospective data collection, centralised echocardiographic review, and rigorous statistical methodology, this study aims to overcome several of these limitations. The resulting prognostic model is expected to improve prenatal counselling and support perinatal decision-making including decisions regarding fetal interventions.

Some limitations should be acknowledged. Owing to the rarity of PA/CS-IVS, the sample size is constrained by feasibility considerations, which may limit model complexity and precision. In addition, although management is not dictated by the protocol, variations in postnatal treatment between centres may influence outcomes. Finally, while exploratory machine learning approaches are included, these analyses are hypothesis-generating and will not replace the primary regression-based model.

Accurate risk stratification is particularly relevant in the context of prenatal interventions such as fetal pulmonary valvuloplasty, where potential benefits must be weighed against procedural risks and appropriate candidate selection remains essential. A robust and externally validated prognostic model may therefore represent a key step towards harmonising clinical practice and informing the design of future interventional studies in this population.

Trial registration

The study was registered to promote transparency and prospective documentation of study objectives and methods. ClinicalTrials.gov, NCT07095829.

Abbreviations

AUROC

Area under the receiver operating characteristic curve

BSA

Body surface area

CS

Critical pulmonary stenosis

DICOM

Digital Imaging and Communications in Medicine

eCRF

Electronic case report form

GA

Gestational age

IVS

Intact ventricular septum

ML

Machine learning

NICU

Neonatal intensive care unit

PA

Pulmonary atresia

PA/CS-IVS

Pulmonary atresia or critical pulmonary stenosis with intact ventricular septum

PPV

Positive predictive value

NPV

Negative predictive value

REDCap

Research Electronic Data Capture

ROC

Receiver operating characteristic

RV

Right ventricle

LV

Left ventricle

TAPSE

Tricuspid annular plane systolic excursion

TRIPOD

Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis

Authors’ contributions

CV conceived and designed the study and drafted the initial manuscript. AMG, MM, SAM, and SP contributed to the study design and provided critical intellectual input. LA and EGM contributed to protocol development, data acquisition procedures, and coordination of participating centres. AG supervised the study design and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This study is funded by the Instituto de Salud Carlos III (Spanish Ministry of Economy, Industry and Competitiveness), grant number PI23/00776.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study was approved by the Institutional Review Board of Hospital Universitario 12 de Octubre, Madrid, Spain. Each participating centre will obtain local ethical approval in accordance with national regulations. Written informed consent will be obtained from all prospectively recruited participants prior to inclusion. For retrospective cases, the requirement for informed consent was waived in accordance with local regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.