Development of a Multivariable Model to Predict the Need for Bone Marrow Sampling in Persons With Monoclonal Gammopathy of Undetermined Significance : A Cohort Study Nested in a Clinical Trial

Abstract

Background: Monoclonal gammopathy of undetermined significance (MGUS) and smoldering multiple myeloma (SMM) are asymptomatic precursor conditions to multiple myeloma and related disorders. Smoldering multiple myeloma is distinguished from MGUS by 10% or greater bone marrow plasma cells (BMPC) on sampling, has a higher risk for progression, and requires specialist management.

Objective: To develop a multivariable prediction model that predicts the probability that a person with presumed MGUS has 10% or greater BMPC (SMM or worse by bone marrow criteria) to inform the decision to obtain a bone marrow sample and compare its performance to the Mayo Clinic risk stratification model.

Design: iStopMM (Iceland Screens, Treats or Prevents Multiple Myeloma), a prospective population-based screening study of MGUS. (ClinicalTrials.gov: NCT03327597).

Setting: Icelandic population of adults aged 40 years or older.

Patients: 1043 persons with IgG, IgA, light-chain, and biclonal MGUS detected by screening and an interpretable bone marrow sample.

Measurements: Monoclonal gammopathy of undetermined significance isotype; monoclonal protein concentration; free light-chain ratio; and total IgG, IgM, and IgA concentrations were used as predictors. Bone marrow plasma cells were categorized as 0% to 4%, 5% to 9%, 10% to 14%, or 15% or greater.

Results: The c-statistic for SMM or worse was 0.85 (95% CI, 0.82 to 0.88), and calibration was excellent (intercept, -0.07; slope, 0.95). At a threshold of 10% predicted risk for SMM or worse, sensitivity was 86%, specificity was 67%, positive predictive value was 32%, and negative predictive value was 96%. Compared with the Mayo Clinic model, the net benefit for the decision to refer for sampling was between 0.13 and 0.30 higher over a range of plausible low-risk thresholds.

Limitation: The prediction model will require external validation.

Conclusion: This accurate prediction model for SMM or worse was developed in a population-based cohort of persons with presumed MGUS and may be used to defer bone marrow sampling and referral to hematology.

Primary funding source: International Myeloma Foundation and the European Research Council.

Figure 1.

Flow diagram of the study population. FLC = free light-chain; MGUS = monoclonal gammopathy of undetermined significance.

Figure 2.

Optimism-corrected calibration plot of the prediction model (blue line, black dots with point range) and the Mayo Clinic risk stratification model (yellow dots with point range) shows the relationship between the observed and predicted probabilities of ≥10% BMPC. In our cohort of presumed MGUS, ≥10% BMPC is consistent with SMM or MM. For the evaluation of the prediction model, the sample is divided into 10 equally large groups of predicted probability and the mean observed probability of each group depicted as a black dot and point range, centered at the mean predicted probability of the group. The weighted scatterplot smoothing (LOWESS) relationship between the observed and predicted probabilities of 2000 bootstrap resamples with replacement and optimism correction are shown as individual thin gray lines, with the mean relationship shown as a blue line. For the Mayo Clinic risk stratification model, the mean observed probability of SMM or worse is shown for low-risk, low-intermediate risk, and high-intermediate risk MGUS, centered at the mean predicted probability. For both models, calibration is compared with the dashed black line, which represents a hypothetical perfect relationship between the observed and predicted probabilities. The sample distribution of predicted probabilities for the prediction model is presented as a blue marginal histogram and as a yellow marginal histogram for the Mayo Clinic risk stratification model. BMPC = bone marrow plasma cells; MGUS = monoclonal gammopathy of undetermined significance; MM = multiple myeloma; SMM = smoldering multiple myeloma

Figure 3.

Two-by-two tables for the Mayo Clinic risk stratification model and the prediction model showing the sensitivity, specificity, PPV, NPV, and number of persons with presumed MGUS for whom bone marrow sampling would be deferred using the models at 4 different low-risk thresholds. Panel A shows a low-risk threshold of 5%, which implies a willingness to subject 19 persons with presumed MGUS to unnecessary bone marrow sampling to identify 1 person with SMM. Panel B shows a low-risk threshold of 10%, which implies an acceptance of 9 unnecessary bone marrow sampling procedures for each case of SMM. Panel C shows a low-risk threshold of 15%, which implies an acceptance of 17 unnecessary sampling procedures for every 3 cases of SMM. Finally, Panel D shows a low-risk threshold of 20%, which implies an acceptance of 4 unnecessary sampling procedures for every 1 case of SMM. Note that using a 10% and a 15% low-risk threshold would lead to identical decision making using the Mayo Clinic risk stratification model, as in both cases bone marrow sampling would only be deferred among persons classified as having low-risk MGUS. MGUS= monoclonal gammopathy of undetermined significance; NA = not applicable; NPV = negative predictive value; PPV = positive predictive value; SMM= smoldering multiple myeloma.

Figure 4.

The net benefit of the prediction model (blue solid line) for the decision to defer bone marrow sampling compared with using the Mayo Clinic risk stratification model (yellow dashed line) to defer sampling, obtaining a bone marrow sample in all persons regardless of risk (black dashed line; default strategy), or deferring bone marrow sampling in all persons regardless of risk (red dashed line), over a range of plausible low-risk thresholds. Net benefit is illustrated in the unit of true negatives (persons with MGUS in whom bone marrow sampling is deferred) that have been appropriately weighted for false negatives (persons with SMM in whom bone marrow sampling is deferred) by the implied equivalence between the number of unnecessary bone marrow sampling procedures and identified cases of SMM associated with the chosen low-risk threshold. Applying a low-risk threshold of 10% to defer bone marrow sampling implies that one is willing to subject 9 persons with MGUS to unnecessary bone marrow sampling to identify a single person with presumed MGUS who has SMM by bone marrow criteria. MGUS = monoclonal gammopathy of undetermined significance; SMM = smoldering multiple myeloma.