Artificial intelligence systems for predicting chronic ischemic heart disease outcomes in cardiac surgery patients based on presence of anemia: a literature review

Full Text

BACKGROUND

Cardiovascular disease represents a significant and growing burden on healthcare systems [1]. In Russia, the number of people undergoing heart surgery exceeds 600 thousand annually [2, 3]. Of them, 30 to 70% of patients have anemia which is associated with a 4-fold increased risk of one-year mortality, a 5-fold increased risk of stent thrombosis, a 1.3-fold increased risk of recurrent acute coronary syndrome, and a 2-fold increased risk of bleeding [1–3]. More than 10 modifiable and many non-modifiable risk factors have been identified for unfavorable prognosis of ischemic heart disease (IHD), with anemia being a significant one. Artificial intelligence (AI) is increasingly being used to determine poor prognostic factors for IHD [4–41]. Therefore, the aim of this review was to analyze existing AI platforms designed to predict IHD progression and consider anemia.

MATERIALS AND METHODS

This review presents Russian and foreign studies on the use of AI platforms to predict IHD progression considering presence or absence of anemia. The PubMed and Russian Science Citation Index databases were searched from 2000 to early 2024. Using "artificial intelligence", "anemia", "ischemic heart disease", "hemoglobin", "cardiac surgery" as keywords, 906 papers were found and analyzed according to the criteria presented. As the search for papers using all keywords did not yield any results, the above keywords were used in pairs, and papers were found presenting data on the use of digital platforms with AI elements to predict IHD outcomes. As a result, 38 papers were selected and analyzed using parameters included in the interpreted input data in AI systems. The list of outcomes (endpoints) predicted by researchers was also assessed.

RESULTS

Appendix 1 lists the key clinical parameters interpreted using AI platforms described in papers and the outcomes they predict. This review evaluated 38 proprietary AI platforms, interpreted data, and predicted outcome parameters. We classified the interpreted parameters as clinical, instrumental, and laboratory ones. This analysis showed that most platforms included parameters that are rarely determined in routine practice, and only 8 of 38 systems included complete blood count (CBC) as one of prognostic parameters.

We identified two types of outcomes predicted by developers of AI platforms: 1) probability of a clinically defined adverse outcome (acute coronary syndrome, death, etc.); 2) probability of a morphologically defined adverse outcome (e.g., plaque rupture). We believe that such a classification helps understand the practical value of using AI platforms developed by different researchers. On the one hand, the accuracy of system is improved due to increasing the amount of output data and the proportion of objective and quantifiable prognostic features. On the other hand, the need to collect and enter large amounts of information into the system is limited in practice by technical capabilities of a healthcare organization, data collection time, etc. In addition, estimated probability of a morphologically defined adverse outcome (e.g., risk of plaque rupture) is difficult to use in routine practice as a guide for clinicians because the traditional risk-oriented approach is based on assessing probability of clinically adverse conditions, not morphological ones.

Appendix 1 shows different AI platform development strategies used in the studies reviewed. Machine learning models were the most commonly used, and only four cases did not specify any algorithm.

The majority of researchers include data pertaining to the presence or absence of commonly recognized risk factors for IHD progression in the list of data to be interpreted by the AI platform. Most of these risk factors are non-modifiable, with anemia being a modifiable one. However, only 10% of researchers included this parameter in their analysis of IHD risk. D'Ascenzo et al. [25] developed the only algorithm to determine the risk of IHD in the presence of anemia. Only Ohara et al. [24] evaluated red blood cell counts as a predictor of anemia. At the same time, anemia is a significant IHD risk factor and correlates with worse outcomes [3].

According to Appendix 2, all AI systems developed for recording laboratory parameters of anemia can be classified into two groups which consider (54 papers) and disregard (6 papers) CBC parameters.

Based on this fact, anemia seems to be underestimated as a prognostically adverse factor that significantly increases the risk of adverse IHD outcomes [1–3]. Bleeding during IHD surgery is also not fully considered, although it is an independent adverse predictor: only one of the developed AI systems included this factor (D'Ascenzo et al.) [25].

Table 1 shows the frequency of using endpoints for assessment of outcomes predicted by the platform developers.

 

Table 1. Frequency of using endpoints to evaluate outcomes predicted by authors of papers
Predicted outcome	Number of papers using the outcome	Percentage of papers using the outcome out of the total number (n=38) of papers meeting inclusion criteria and having no exclusion criteria
Anemia	1	2.60
Non-ST-elevation acute coronary syndrome	3	7.80
Unstable angina	9	23.60
Acute myocardial infarction recurrence	7	18.40
Pericarditis	3	7.84
Atrial fibrillation	4	10.50
Cardiomyopathy	4	10.50
Angioedema	1	2.60
Chronic heart failure	2	5.20
Acute heart failure	2	5.20
Myocardial hypertrophy	1	2.60
Hypertrophic cardiomyopathy	2	5.20
Cardiac death	3	7.84
Stent thrombosis	2	5.20
Brugada syndrome	2	5.20
White coat hypertension	1	2.60
Pulmonary hypertension	1	2.60
Metabolic syndrome	2	5.20
Arrhythmia	4	10.50
Bleeding	6	15.47
Death (all causes)	8	21.0

 

An evaluation of outcome prediction parameters suggests that the focus is predominantly on clinical endpoints such as death, arrhythmia, bleeding, while morphological adverse endpoints such as myocardial hypertrophy and stent thrombosis are evaluated in only 7.8% of cases. We believe that the focus on clinical endpoints is justified because it is consistent with the principles of recommendations to be followed in routine clinical practice.

DISCUSSION

Artificial intelligence can improve the quality of risk prediction for adverse outcomes in IHD patients. However, our study shows no current consensus on the minimum sufficient set of parameters to be analyzed using AI for prediction of cardiac surgery outcomes. In the list of data to be interpreted, most researchers include mainly medical history of commonly accepted risk factors, most of which are non-modifiable. It should be noted that anemia is a modifiable risk factor. However, only 10% of the developers included this parameter in their analysis of IHD risk. The Ohara and D'Ascenzo algorithms are the first published IHD prediction platforms that consider anemia, and both algorithms were developed for patients with bleeding and/or baseline anemia [24, 25]. In patients without anemia or no history of bleeding, none of the developers included anemia as a potential outcome to be predicted.

We believe that anemia is not considered to predict IHD progression due to the relatively short history of AI use in healthcare. Researchers seem to improve their algorithms using a relatively narrow set of the same interpretable data and outcomes (see Appendix 2), and in the next phase a wider range of parameters is expected to be included.

It should be noted that there are no AI platforms that support clinical decision making to recommend a personalized treatment or diagnostic strategy. The only exception is the system developed by Noh et al. [6] to assess the “need for revascularization.”

The relatively small number of developments is remarkable. At the same time, AI platforms are being developed in Russia to improve the quality of life of IHD patients and optimize use of healthcare resources. There is a high demand for such services among healthcare professionals and patients, but no available services have been offered.

CONCLUSION

In some countries, AI platforms have been created to predict IHD progression. They address prognostic issues and do not focus enough on treatment strategies. At present, the methods for AI platform development for IHD patients are still in their infancy and are based on a limited set of input data. Anemia is not considered by most teams developing AI models for patients after cardiac surgery. Therefore, it is important to develop AI platforms for predicting IHD progression, considering anemia as a risk factor. This will increase their medical, social, and economic effectiveness.

ADDITIONAL INFORMATION

Appendix 1. Analysis of interpreted and predicted parameters in publications on developed artificial intelligence systems. doi: 10.17816/rmj635256-4221253

Appendix 2. The frequency of use of indicators included in the data interpreted by the authors of the articles. doi: 10.17816/rmj635256-4221255

Funding source. The work was performed as part of a scientific study of the Saratov State Medical University named after V.I. Razumovsky of the Ministry of Health of the Russian Federation on the subject of the state assignment of the Ministry of Health of the Russian Federation "Development of an expert artificial intelligence system for personalized diagnosis, correction of cognitive impairment and prediction of outcomes in chronic coronary heart disease depending on the presence of anemic syndrome" (EGISU R&D No. 124020600005-4, PTNI 1023022700025-8 dated 06.02.2024)

Competing interests. The authors claim that there is no conflict of interest in the article.

Authors’ contribution. All authors confirm compliance of their authorship with the international ICMJE criteria. The largest contribution is distributed as follows: T.Yu. Kalyuta — research concept and design, writing the text, editing; I.P. Emelyanova — collection and processing of material, writing the text; V.V. Suvorov — writing the text, compilation of the list of literature; A.S. Fedonnikov — writing the text, editing.

About the authors

Tatyana Y. Kalyuta

Saratov State Medical University named after V.I. Razumovsky

Author for correspondence.
Email: tatianakaluta@yandex.ru
ORCID iD: 0000-0003-3172-0804
SPIN-code: 4982-7861

MD, Cand. Sci. (Medicine)

Russian Federation, Saratov

Irina P. Emelyanova

Saratov State Medical University named after V.I. Razumovsky

Email: irisha-9966@mail.ru
ORCID iD: 0000-0002-4178-9437
SPIN-code: 1766-8528
Russian Federation, Saratov

Valeriy V. Suvorov

Saratov State Medical University named after V.I. Razumovsky

Email: valeriy_s@inbox.ru
ORCID iD: 0000-0002-4181-9034
SPIN-code: 4757-5250

Cand. Sci. (History)

Russian Federation, Saratov

Alexander S. Fedonnikov

Saratov State Medical University named after V.I. Razumovsky

Email: fedonnikov@mail.ru
ORCID iD: 0000-0003-0344-4419
SPIN-code: 2248-5246

MD, Dr. Sci. (Medicine)

Russian Federation, Saratov

References

Supplementary files