Feasibility of risk classification with newly proposed international myeloma working group (IMWG) high-risk definition: A single-center analysis using natural language processing

Abstract Introduction The International Myeloma Working Group (IMWG) recently published an updated high-risk (HR) definition for multiple myeloma (Avet-Loiseau et al, JCO 2025). This system attempts to refine risk classification by capturing the nuance of cooperating cytogenetic or laboratory abnormalities. However, the complexity of this system raises concerns about missing data, especially in the case of inadequate bone marrow specimens for full cytogenetic analysis. Consequently, we performed an analysis of patients with multiple myeloma who were diagnosed and staged at our center between 2018 and 2024 to compare degree and impact of missing data on the IMWG-HR classification versus the traditional International Staging System (ISS) and its two revisions (R-ISS, R2-ISS). Methods Eligible patients were enrolled in our multiple myeloma tissue bank (IRB: 201102270). Data for classification was obtained via an associated protocol that downloads electronic medical record data on patients with multiple myeloma at regular intervals and converts unstructured text (i.e., pathology and cytogenetic reports) to discrete data fields using natural language processing (IRB: 202410213). Patients were staged by the ISS, R-ISS, and R2-ISS and defined as high or standard risk by the IMWG-HR criteria per published criteria. Notably, no patients at our center undergo single-gene testing for TP53 abnormalities and t(14;20) has not historically been included in our cytogenetics algorithm. These criteria were consequently excluded from the IMWG-HR calculation. Risk-determining features were reported as present, absent, or unknown. For the initial analysis, patients with missing data were reported as unclassified. For the secondary analysis, missing data was treated as the absence of a high-risk feature (i.e., presumed to be normal). Results 305 patients were included in the overall analysis. The median age was 66 (range: 34 – 92). 60.3% of patients were male and 82.0% were White. Missing laboratory values were rare (B2M: 2.6%, albumin: 0.0%, LDH: 0.0%, creatinine: 0.0%). Missing cytogenetic data were common (del(17p) : 10.8%, t(4;14) : 29.2%, t(14;16) : 40.7%, del(1p) : 47.5%, gain(1q) : 47.5%). Unclassified risk was common in more recent risk systems (ISS: 2.6%, R-ISS: 42%, R2-ISS: 48.5%, IMWG-HR: 50.5%). Inadequate cytogenetic data was the primary cause of unclassified patients (total patients: ISS: N/A, R-ISS: 41.0%, R2-ISS: 48.5%, IMWG-HR: 49.8%). Missing cytogenetic data for IMWG-HR was not associated with bone marrow aspirate assessed as “adequate” (n = 214) by the reading pathologist (51% vs. 50%, p = 0.91). In patients with adequate aspirates, patients with completed cytogenetic assessment had significantly higher median plasma cell aspirate percentage (median: 29% vs. 15%, p = 0.01). In patients with no missing data, 34.7%, 36.7% and 28.6% of patients were stage I, II, and III by ISS, 24.3%, 48.6% and 27.1% were stage I, II, and III by R-ISS, and 12.7%, 21.7%, 47.8% and 17.8% of patients were stage I, II, III, and IV by R2-ISS. 61.6% and 38.4% were standard and high by IMWG-HR. In patient patients with missing data, 35.2%, 47.7%, and 17.2% were stage I, II, and III by R-ISS, and 21.6%, 45.9%, 29.7%, and 2.7% patients were stage I, II, III, and IV by R2-ISS. 87.2% and 12.8% were standard and high risk by IMWG-HR. Distribution of risk classification was significantly different among unclassified patients versus patients with no missing data with all recent risk systems (R-ISS: p = 0.04, R2-ISS: p <0.001, IMWG-HR: p <0.001). The difference in high-risk classification in patient with or without missing data was especially striking for R2-ISS (stage III-IV: 32.4% vs. 65.6%) and IMWG-HR (12.8% vs. 38.4%). Conclusion While improved risk classification systems may improve bedside prognostication in theory, significant barriers remain to their practical implementation. In particular, data from standard-of-care FISH-based cytogenetic analysis may be inadequate to support “multi-hit” risk classification, including the new IMWG-HR definition and R2-ISS. Alternative approaches, including the use of whole genome sequencing, may help circumvent some of these challenges.