Dear Editor,
We read with great interest the review by Naing et al. [1], outlining how artificial intelligence (AI) might be integrated across the pulmonary hypertension (PH) care pathway, from echocardiographic image acquisition to risk stratification. Framing the discussion around the specific barrier of geographic dispersion and limited tertiary center access gives this roadmap a concrete equity rationale that many AI-in-cardiology reviews lack.
1. A ground-truth limitation disclosed for one model is left unstated for a closely related one
The left heart disease prediction model is explicitly limited by the absence of right heart catheterization data in its development and validation cohorts, as tricuspid regurgitation velocity and echocardiographic surrogates were used instead of the invasive gold standard [2]. The subsequent proof-of-concept deep learning model predicting all-cause pulmonary hypertension was built on the same structure, trained, and tested against a peak tricuspid regurgitation velocity threshold rather than catheterization-confirmed pressure [3]. However, this identical constraint was not mentioned when its area under the curve of 0.84 was presented. Readers comparing the reported accuracy of the two models would reasonably assume that the second carries the same caveat as the first, and stating it explicitly would prevent this omission from reading as though its ground truth were more rigorous than it is.
2. Area under the curve values from studies with different ground truths and populations are presented as a single accumulating body of evidence
The review lists area under the curve figures ranging from 0.70 for the authors' own left heart disease model to 0.988 for a chest radiograph-based model, moving through echocardiographic and electrocardiographic approaches from other groups [4]. These studies differed in whether pulmonary hypertension was confirmed by catheterization or inferred from echocardiographic surrogates, whether the target was all-cause disease or a specific subtype, and whether the population was unselected or already referred for suspected disease and enriched accordingly. Presenting these figures in sequence without distinguishing these differences risks implying a consistent, improving evidentiary trajectory for AI in pulmonary hypertension diagnosis when the studies are not answering the same diagnostic question in comparable populations.
3. Preliminary, unpublished tools and externally validated ones are presented without a maturity distinction
Several models central to the proposed roadmap, including the left heart disease predictor and all-cause pulmonary hypertension model, originate from the authors' group and are described as proof-of-concept or derived from a conference abstract without a peer-reviewed manuscript. These sit alongside externally validated tools from other research groups, discussed with comparable confidence and without a clear signal about which components of the roadmap rest on published, externally replicated evidence versus early stage internal work still awaiting full peer review.
Taken together, these considerations do not diminish the value of proposing a structured pathway for integrating AI into pulmonary hypertension care in settings where specialist access is limited. Applying the ground-truth caveat consistently across both models, distinguishing studies by their reference standard and population when comparing performance figures, and flagging the maturity level of each cited tool would strengthen the evidence basis for this roadmap.
CRediT authorship contribution statement
D. Beaulah Princiba: Conceptualization, Data curation, Formal analysis. Roopa R. Kulkarni: Visualization, Writing – original draft, Writing – review & editing.
Ethical approval
Not applicable (commentary on previously published research).
Declaration of Generative AI and AI-assisted technologies in the writing process
Generative AI tools were used solely for language refinement and formatting assistance. All scientific interpretation, critique, and conceptual analysis were independently developed by the authors.
Funding
None.
Declaration of competing interest
The authors declare no conflicts of interest.
Acknowledgment
None.
Contributor Information
D. Beaulah Princiba, Email: prime999000@outlook.com.
Roopa R. Kulkarni, Email: roopakulkarni-ece@dsatm.edu.in.
Data availability
No new data were generated or analysed in this work.
References
- 1.Naing P., Scalia G.M., Murdoch D., Ranasinghe I., Forrester D.L., Strange G., et al. Artificial intelligence enhanced contemporary pulmonary hypertension care. Am. Heart J. Plus. 2025;60 doi: 10.1016/j.ahjo.2025.100673. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Farina J.M., Scalia I., Pereyra Pietri M., et al. Evaluation of ePLAR, the echocardiographic pulmonary to left atrial ratio, in a large cohort with pulmonary hypertension. Eur. Heart J. Open. 2025 doi: 10.1093/ehjopen/oeaf174. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Gao Y., Kim S., McIntosh C. 2025. EchoingECG: An Electrocardiogram Cross-modal Model for Echocardiogram Tasks. [DOI] [Google Scholar]
- 4.Zepeda-Lugo C., Sanchez-Lizarraga M., Insfran-Rivarola A., Arévalos A. Effectiveness of AI models for cardiovascular disease diagnosis: a systematic review and meta-analysis. Rev. Investig. Innov. Cienc. Salud. 2025 doi: 10.46634/riics.504. [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
No new data were generated or analysed in this work.