Age-Adjusted Index, Machine Learning Are Significant Factors for Predicting Thrombosis in Polycythemia Vera

Two posters presented at the 65th American Society of Hematology Annual Meeting and Exposition found age-adjusted Charlson comorbidity index (ACCI) and machine learning (ML) are valuable tools for predicting thrombosis risk and survival in patients with polycythemia vera (PV).

Polycythemia vera, a rare slow-growing blood cancer with an increase in the number of red blood cells | Dr_Microbe - stock.adobe.com

The first poster aimed to identify the prediction factors for thrombosis and overall survival (OS) in patients with PV using demographic, clinical, and laboratory parameters with neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), thrombosis data, ACCI, and the presence of cardiovascular (CV) risk factors, including arterial hypertension, diabetes, smoking, and hyperlipidemia.¹

The study included a total of 816 patients diagnosed with PV who had been treated between 2005 to 2022. The mean (SD) age of these patients was 62 (18-87) years. Additionally, 26.5% of these patients were categorized as low-risk and 73.5% were categorized as high-risk. According to ACCI, 11.8% of these patients had no comorbidities, 34.4% had mild PV, 36% moderate PV, and 17.8% severe PV.

Previous thrombosis (P = .029), ACCI groups (P < .001), presence of 1 or more CV factor (P = .001), or secondary malignancy (P < .001) were predictors of venous thrombosis development, while previous thrombosis (P = .03), male gender (P = .006), standard risk score (P = .001) ACCI (P < .001), 1 or more CV factor (P < .001), NLR of 5 or greater (P = .008), and PLR of 500 or more (P < .001) were predictors of arterial thrombosis development.

ACCI was found to be the most reliable predictor of thrombosis development, which the researchers believe would allow for better survival and would improve identification of high risk patients with PV.

A second poster aimed to develop a ML approach to predict the risk of thrombosis, which remains the leading cause of morbidity and mortality for patients with PV.²

The study included a total of 526 patients with manually verified PV who were identified using electronic health records and a research database repository. Data completeness varied with more than 90% availability in over 200 features and found that features with less than 50% missingness could be imputed without corrupting the dataset.

A total of 470 patients had clinic visits through March 31, 2020, in which 75% were used in the model training cohort and 25% were used for model evaluation. Additionally, a risk calculator was applied to models corresponding to user-entered clinical data and computed 1-year thrombosis risk.

More than 1 million models were trained to predict near-term thrombosis risk in patients with PV, based on factors such as age, blood type, disease events, and short term changes such as body mass index (BMI). Moreover, the full model performed very well, with the full 21-feature model and high-performing models showing clear separation for 1-year thrombosis risk prediction. Furthermore, the features of importance identified were blood counts, BMI, and time since diagnosis.

Therefore, the researchers of this study believe their ML tool’s ability to identify patients with PV with a high risk for thrombosis can be applied as inclusion criteria for prospective clinical trials, which would reduce the time and number of patients needed in these trials.

References

1. Lekovic D, Ivanovic J, Arsenovic I, et al. Age-adjusted charlson comorbidity index is a significant factor for predicting thrombosis development and survival in polycythemia vera. Posted presented at: American Society of Hematology Annual Meeting and Exposition. December 9-12, 2023; San Diego, CA.

2. Krichevsky S, Abu-Zeinah G, Scandura JM. Using machine learning to predict near-term thrombosis risk in patients with polycythemia vera. Posted presented at: American Society of Hematology Annual Meeting and Exposition. December 9-12, 2023; San Diego, CA.