In a study highlighting the utility of artificial intelligence-based algorithms for risk prediction, researchers developed a predictive model for multiple myeloma (MM) outcomes that incorporated the impact of ethnicity factors on patients’ risk status. The study results were published in the journal Frontiers in Oncology.
The researchers used machine learning algorithms to develop a risk prediction model with progression-free survival (PFS) and overall survival (OS) as the outcomes of interest for patients with newly diagnosed MM (NDMM). Datasets used for generating this risk stratification model came from the Multiple Myeloma Indian (MMIn) dataset, including patients based in India, and the Multiple Myeloma Research Foundation (MMRF) dataset, with patients based in the US. For each clinical or laboratory parameter used in modeling risk, threshold values were used to assign patients to high- or low-risk categories for the parameter.
In characterizing the model, the researchers employed a method that enables model interpretability, in order to establish which attributes included in the model appeared to be most relevant to risk prediction. The model was termed the Consensus-Based Risk-Stratification System (CRSS), and 3 risk categories were established with this model. CRSS-1 is the level with the lowest risk, CRSS-2 has intermediate risk, and CRSS-3 is associated with the greatest risk.
The MMIn datasest contained data for 1070 patients, and the MMRF dataset contained data for 900 patients. Data on high-risk cytogenetic aberrations (HRCAs) were available for 384 patients in the MMIn dataset and 800 of the patients in the MMRF dataset. The final model included beta-2 microglobulin (b2M) level, hemoglobin level, age, HRCAs, estimated glomerular filtration rate, albumin level, and calcium level as parameters used for risk stratification.
Between the 2 datasets, the researchers found significant differences in relevant thresholds for prognostic features in the best-performing models. For example, a cutoff age of 67 years appeared most useful when examining the Indian population, compared with 69 years for the American population. The threshold for b2M was also slightly lower with the MMIn dataset, while the hemoglobin threshold appeared higher for this dataset. Other thresholds also differed slightly.
The researchers identified b2M and hemoglobin levels as the features contributing the most to placement in CRSS-1 and CRSS-2 risk categories, with b2M exerting a strong impact in these categories. The presence of HRCAs was found to be the feature contributing the most to CRSS-3 status, although the magnitude of the effect of HRCA for CRSS-3 was not as strong as with b2M in other risk categories.
The researchers considered the model they developed in this study to be superior at allocating patients into risk groups, compared with the Revised-International Staging System (R-ISS). “The successful evaluation of our proposed staging system on both datasets establishes the utility of the proposed ethnicity-aware staging system for NDMM patients, treated largely with novel agents or a combination thereof, in a real-world scenario,” they also noted in their report.
Farswan A, Gupta A, Sriram K, Sharma A, Kumar L, Gupta R. Does ethnicity matter in multiple myeloma risk prediction in the era of genomics and novel agents? Evidence from real-world data. Front Oncol. 2021;11:720932. doi:10.3389/fonc.2021.720932