Assessing the Performance of the Iceland Screens, Treats, or Prevents Multiple Myeloma (iStopMM) Model in a Multicultural Bronx Cohort: Implications for Monoclonal Gammopathy of Undetermined Significance (MGUS) Risk Stratification

Introduction: Monoclonal gammopathy of undetermined significance (MGUS) is a pre-malignant condition with a 1% annual risk of progressing to multiple myeloma (MM) or related malignancies, making risk stratification and monitoring essential. The Iceland Screens, Treats, or Prevents Multiple Myeloma (iStopMM) risk stratification model is a multivariable model developed to predict the likelihood of ≥10% abnormal plasma cells in the bone marrow in individuals with IgG, IgA, or biclonal MGUS. However, the model was developed in a predominantly White, genetically homogeneous Icelandic population, lacking external validation. Our study aims to externally validate the performance of this model in a racially and ethnically diverse Bronx population.

Methods: The electronic medical records of patients at Montefiore Medical Center from 2002 to 2023 were searched to identify those diagnosed with MGUS using ICD-10 code D47.2 and ICD-9 code 273.1 and who had undergone a bone marrow biopsy. Patients were excluded if they did not meet revised International Myeloma Working Group (IMWG) criteria for MGUS, lacked available bone marrow biopsy results, had suboptimal biopsies, IgM MGUS, unclear immunoglobulin isotype, or unavailable free light chain ratio or SPEP. For the final cohort, variables such as MGUS isotype, M protein concentration, free light chain ratio, and total IgG, IgA, and IgM levels were entered into the iStopMM prediction model. Predicted and actual plasma cell percentages were recorded. Baseline characteristics were reported using frequencies for categorical values and medians for continuous variables. The area under the receiver operating characteristic (AUROC) curve assessed the iStopMM model's performance in predicting ≥10% plasma cells. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were evaluated for 10%, 20%, and 50% probabilities of having greater than or equal to 10% plasma cells based on the iStopMM prediction model. The optimal cut-point was determined using Youden's index. Statistical analysis was performed using IBM SPSS Statistics, Version 29.0.

Results: Of the initial 663 patients, 190 were included in the final cohort after exclusions. Of these 190 patients, 55.8% (n = 106) were females, and the rest were males. In terms of race, most individuals were Blacks or African-Americans (52.6%), followed by individuals categorized as Others (22.1%), Whites (14.7%), Unknown (8.4%), and Asians (2.1%). Regarding ethnicity, a majority identified as non-Hispanic/Latino (68.4%), 23.2% as Hispanic/Latino, and 8.4% were unknown ethnicity. Analysis of immunoglobulin isotypes among the MGUS cohort indicated a prevalent distribution of IgG (83.7%), followed by IgA (11.1%). A smaller proportion of individuals exhibit biclonal or light chain patterns, each representing 2.6% of the cohort. The iStopMM predictive model was able to predict greater than or equal to 10% plasma cells on bone marrow biopsy with an AUROC of 0.78 and confidence interval (0.71, 0.85). When set at a 10% threshold for predicting SMM or worse, the iStopMM model had a 93.3% sensitivity, 33.7% specificity, 55.3% PPV, and 85.0% NPV. Additionally, the optimal cut-point by Youden's index was calculated to be 33%, indicating that within our population, the sensitivity (70.8%) and specificity (77.2%) are optimal when the threshold is set at a 33% chance of having greater than or equal to 10% plasma cells on bone marrow biopsy by the IStopMM model.

Conclusion: Our study included 52.6% of patients identifying as Black or African American and 23.2% as Spanish/Hispanic/Latino, highlighting the unique patient composition, remarkably different from the homogenous population of the iStopMM study. This AUROC value of 0.778 suggests a reasonable discriminatory performance of the model in our racially and ethnically diverse study population. Moreover, the narrow confidence interval of 0.71-0.81 of the AUROC further underscores the consistency in the model's performance. Finally, given the limited sample size, we could not explore the iStopMM model's performance across different racial and ethnic subgroups or different MGUS isotypes. Further studies addressing such limitations and utilizing a different population subset could provide more insights into the external validity and clinical utility of the iStopMM model.