Cracking the code of cancer and clots: artificial intelligence and cancer-associated thrombosis

Artificial intelligence (AI)^* refers to the field of computer science that uses powerful computers and large amounts of real-world data to learn useful patterns to mimic the problem-solving capabilities of humans. Although this field has indeed captured the imagination of authors and the public for over a century with characters appearing in pop culture ranging from dystopian novels to popular ‘futuristic’ space television shows (think C3PO, the neurotic android crew member in Star Wars), the more recent public release of Generative AI models that allow users to generate research papers, poems and even images with minimal human input has sparked a renewed interest in its current and potential applications in our contemporary lives, not the least of which is clinical medicine. Indeed, AI has already found its way into medicine, ranging from programs that could serve as clinical scribes to complex analysis of ‘big’ data to improving surveillance and tracking of disease outbreaks.¹ As AI continues to make inroads in clinical studies, both scope and ability, we explore its potential role in thrombosis, including cancer-associated thrombosis (CAT).

Advantages of AI that can be leveraged for thrombosis

Artificial intelligence allows sophisticated algorithms to process large volumes of data to generate actionable insights that can be effectively employed to improve outcomes. These capabilities allow rapid processing of a large volume of data that would otherwise be unfeasible for humans to perform unassisted. Moreover, this advanced computer processing can allow efficiency in healthcare deliveries with automated checks to minimize and allow for early recognition of human errors that can improve patient safety.² The application of AI in clinical medicine lags comparatively to other sectors, such as technology or finance, where adoption and development were early. The stakes and barriers are arguably much higher in a clinical context. However, AI is now rapidly being deployed to improve healthcare delivery models, including operations, reimbursement, patient safety, and quality improvements.

Venous thromboembolism is a frequent global condition with a heavy price tag for morbidity and mortality.³ Yet, given the complexity of factors that contribute to the condition’s pathophysiology and the variety of settings in which it can occur, the nonspecific clinical findings and the treatment factors that can impact the clinical outcomes make it a challenging beast to tame. Artificial intelligence tools such as Image Recognition and Natural Language Processing can thus be valuable to assist clinicians individually and health care systems to prevent, diagnose and manage this condition. Cancer thrombosis presents its own complexities given the specific risk factors, differences in treatment and prevention strategies, and often higher stakes, such as bleeding, recurrent thrombosis and death, which have created opportunities for researchers to try and apply AI in this space.

*At present, artificial intelligence (AI) and machine learning (ML) are used interchangeably in popular and scientific literature to loosely denote the same concept.

Applications of AI in thrombosis

Natural Language Processing:

Natural Language Processing (NLP) applies principles of computational linguistics with machine learning to allow the processing of human language into data that can be analyzed by a computer/machine. Its application in thrombosis includes allowing for automatic surveillance of data in electronic health records, which can benefit epidemiological research and improve healthcare delivery through real-time detection of thrombotic events. Researchers have applied various NLP models to thrombotic outcomes with good performance, which is promising. For example, researchers have compared NLP interrogation of radiological reports, including ultrasounds, CT angiograms and V/Q scans, yielding a sensitivity, specificity, and positive and negative predictive values north of 90%. ⁴ Specifically in cancer patients, a recent study showed that in a longitudinal cohort of oncology patients a NLP-based model to detect thrombotic events had a c statistic of 0.93 (95% CI, 0.92–0.94), which when combined with an algorithm that was based on ICD codes and medications improved to a c statistic of 0.98 (95% CI, 0.97–0.98). ⁵

Image Recognition:

Artificial intelligence is being applied to allow for computer-aided radiological review of imaging studies that can enhance diagnostic accuracy with increased throughput and efficiency. These techniques have been applied to ultrasound,⁶ CT angiographies⁷ and VQ scans.⁸ Novel applications have included methods developed for automatically identifying appropriate diagnostic cross-sectional ultrasonographic images of the popliteal vein by disaster victims using portable ultrasound equipment.⁹ The potential for AI-enabled image recognition in the radiology approach in CAT is demonstrated in a retrospective single-center study on patients with cancer. Over 1800 elective chest CT scans were reviewed using an AI algorithm that found the algorithm had a very high sensitivity and specificity with greater detection of incidental pulmonary embolism; the majority were originally unreported, including a significant proportion of emboli proximal to subsegmental arteries.¹⁰

Prediction modeling and Biomarker discovery:

Applying AI offers opportunities for improvements in patient outcomes by complex information synthesis and automated analyses, with advantages over traditional modalities, such as regression models, including flexibility and scalability.¹¹ We performed a systematic review and meta-analysis that included 14 studies for VTE prediction and found that the use of machine learning appears to improve the accuracy of diagnostic and prognostic prediction of VTE over conventional risk models.¹² Notably, only one of these included CAT,¹³ there was a high risk of bias in most studies, and all lacked external validation. Groups have applied AI/ML techniques to developing clinical prediction models for CAT.¹⁴ Similarly, machine learning is being leveraged to develop novel biomarkers predictive of CAT in a well-curated prospective cancer cohort using a novel proteomic screen of baseline plasma.¹⁵

Concerns for the use of AI in clinical medicine and how to overcome them

Like any novel technology, AI is not immune to fallacies and pitfalls. Although machine learning is touted as ‘self-learning’, the reliability of AI algorithms depends on the availability of large, well-characterized feature-rich datasets. Regulatory and policy frameworks will need to evolve with the application of these technologies to avoid the emergence of ethical, non-discriminatory products that enhance accessibility rather than hinder it.¹⁶ Since AI models are often seen as ‘black boxes’, clinicians may hesitate to rely on them for clinical decision-making. Explainability and transparency are emerging as important features in AI techniques developed for adoption in the clinical world.¹⁷ Additionally, robust collaborative research to allow testing for clinical effectiveness is essential to facilitate the adoption of AI models in healthcare.¹⁸ A contemporary multidisciplinary national survey that evaluated current attitudes and practices for thrombosis prevention in hospitalized patients, several of these themes emerged as concerns for applying novel AI-based technologies in the prevention and management of VTE for today’s clinicians.¹⁹ Identifying these barriers will be instrumental in designing and implementing AI-based tools for VTE prevention and management.

Conclusion

Artificial intelligence / Machine Learning are being explored as novel technologies with the potential to enhance healthcare delivery models and clinical care for patients and providers. These techniques are being explored with promising results in several domains of thrombosis prevention, prediction and management, including surveillance, research, enhanced diagnosis through image recognition and accurate prediction modeling. Future directions will involve thorough auditing and prospective testing of AI tools to allow their application into routine clinical care.

References

1. Rajpurkar P, Chen E, Banerjee O, et al. AI in health and medicine. Nat Med 2022;28:31-38.
2. Choudhury A, Asan O. Role of Artificial Intelligence in Patient Safety Outcomes: Systematic Literature Review. JMIR Med Inform 2020;8:e18599.
3. Wendelboe AM, Raskob GE. Global burden of thrombosis: epidemiologic aspects. Circ Res 2016;118:1340-7.
4. Woller B, Daw A, Aston V, et al. Natural language processing performance for the identification of venous thromboembolism in an integrated healthcare system. Clin Appl Thromb Hemost 2021;27:10760296211013108.
5. Li A, da Costa WL Jr, Guffey D, et al. Developing and optimizing a computable phenotype for incident venous thromboembolism in a longitudinal cohort of patients with cancer. Res Pract Thromb Haemost 2022;6:e12733.
6. Kainz B, Heinrich MP, Makropoulos A, et al. Non-invasive diagnosis of deep vein thrombosis from ultrasound imaging with machine learning. NPJ Digit Med 2021;4:137.
7. Seo JW, Park S, Kim YJ, et al. Artificial intelligence-based iliofemoral deep venous thrombosis detection using a clinical approach. Sci Rep 2023;13:967.
8. Jabbarpour A, Ghassel S, Lang J, et al. The Past, Present, and Future Role of Artificial Intelligence in Ventilation/Perfusion Scintigraphy: A Systematic Review [published online ahead of print, 2023 Apr 18]. Semin Nucl Med. 2023;S0001-2998(23)00026-0. doi:10.1053/j.semnuclmed.2023.03.002
9. Nakayama Y, Sato M, Okamoto M, et al. Deep learning-based classification of adequate sonographic images for self-diagnosing deep vein thrombosis. PLoS One 2023;18:e0282747.
10. Wiklund P, Medson K, Elf J. Incidental pulmonary embolism in patients with cancer: prevalence, underdiagnosis and evaluation of an AI algorithm for automatic detection of pulmonary embolism. Eur Radiol 2023;33:1185-1193.
11. Steyerberg EW, van der Ploeg T, Van Calster B. Risk prediction with machine learning and regression methods. Biom J 2014;56:601-6.
12. Chiasakul T, Lam BD, McNichol M, et al. Artificial intelligence in the prediction of venous thromboembolism: A systematic review and pooled analysis [published online ahead of print, 2023 Oct 4]. Eur J Haematol. 2023;10.1111/ejh.14110. doi:10.1111/ejh.14110
13. Ferroni P, Zanzotto FM, Scarpato N, et al. Validation of a machine learning approach for venous thromboembolism risk prediction in oncology. Dis Markers 2017;2017:8781379.
14. Mantha S, Chatterjee S, Singh R, et al. Application of Machine Learning to the Prediction of Cancer-Associated Venous Thromboembolism. Preprint. Res Sq. 2023;rs.3.rs-2870367. Published 2023 May 8. doi:10.21203/rs.3.rs-2870367/v1
15. Zwicker JEE, Patell R, Marchetti M, et al. Personalized cancer-associated thrombosis risk assessment: integration of plasma proteomics, clinical characteristics, and machine learning [abstract]. Res Pract Thromb Haemost. 2021;5(Suppl 2):OC 23.1.
16. Jassar S, Adams SJ, Zarzeczny A, Burbridge BE. The future of artificial intelligence in medicine: Medical-legal considerations for health leaders. Healthc Manage Forum. 2022;35(3):185-189. doi:10.1177/08404704221082069
17. Chaddad A, Peng J, Xu J, Bouridane A. Survey of explainable AI techniques in healthcare. Sensors (Basel). 2023;23(2):634. doi:10.3390/s23020634
18. Stogiannos N, Malik R, Kumar A, et al. Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK. Br J Radiol. 2023;20221157. doi:10.1259/bjr.20221157
19. Lam BD, Dodge LE, Datta S, Rosovsky RP, et al. Venous thromboembolism prophylaxis for hospitalized adult patients: a survey of US health care providers on attitudes and practices. Res Pract Thromb Haemost. 2023;7(6):102168. doi: 10.1016/j.rpth.2023.102168.