Machine Learning-Based Clinical and Laboratory Values Associated with Serum Protein Electrophoresis Testing in US Veterans

Introduction: Currently, there is consideration of the role of SPEP testing in screening for monoclonal gammopathies, particularly for detecting plasma cell disorders such as monoclonal gammopathy of undetermined significance (MGUS). Furthermore, it is important to better understand the characteristics of the population undergoing SPEP tests for the assessment of monoclonal gammopathy, as well as the factors influencing the decision to conduct SPEP testing. Given the potentially complex and non-linear relationships inherent in SPEP testing, we employed a Random Forest model to analyze and elucidate the demographic and clinical features associated with SPEP testing.

Methods: We used data from the nationwide US Veterans Health Administration (VHA) database. We included Veterans aged 50-65 who had a total protein test. Veterans dually enrolled in VHA and Medicare were excluded due to the lack of laboratory data in Medicare. First SPEP orders were identified via CPT codes 84165 or 83883. A Random Forest model was used to select features to predict SPEP orders. This model handles high-dimensional data and captures complex interactions, providing robust feature importance metrics. These metrics were further refined using SHAP values. The SHAP value quantifies the contribution of each feature by evaluating all possible combinations of features and calculating the average marginal contribution of each, thereby aiding in understanding and controlling the impact of variables on the model's output. All potential clinical and laboratory variables were evaluated, such as chronic kidney disease, anemia, fractures, neuropathy, total protein levels, hypercalcemia, and body mass index (BMI), as well as sociodemographic factors like race, sex, socioeconomic status, and geographic location. We selected laboratory values prior to the time of the SPEP testing date. Disease/condition diagnoses were identified using ICD-9/10 codes. For Veterans with SPEP testing, data from one year prior to their initial test were analyzed, while for those without SPEP testing, data were collected within one year of their most recent total protein test.

Results: We identified 294,640 Veterans from 1999-2022. Among them, 17,035 (5.78%) underwent SPEP testing. The cohort was predominantly male (77%) and White (51%), with 11% being normal weight (BMI 18.5-24.4), 28% being overweight (BMI 24.5-29.9), and 42% being obese (BMI 30+). Additionally, 17% had neuropathy, 15% had anemia, 11% had chronic kidney disease, 4% had total protein levels > 8.5 gm/dL, and 2% had hypercalcemia, while pathologic and vertebral fractures were each present in 0.5% of cases. The top three features associated with SPEP testing were albumin, age, and thyroid-stimulating hormone values. Elevated total protein levels, chronic kidney disease, anemia, and calcium levels also emerged as significant clinical indicators based on their SHAP values. Increased body weight was associated with SPEP testing, as were elevated HbA1c values. Among the sociodemographic factors, race played a crucial role, with Black Veterans being more prone to receive SPEP testing.

Conclusions: We estimated the marginal contribution of clinical and laboratory values associated with SPEP testing. As expected, we found that SPEP tests were frequently ordered for conditions associated with multiple myeloma, such as chronic kidney disease, anemia, and abnormal calcium levels. Additionally, several laboratory values relevant to the differential diagnosis process for distinguishing MGUS from other conditions, such as albumin, HbA1c, and thyroid-stimulating hormone levels, were also associated with SPEP testing. Demographic factors such as age, weight, and Black race increased the likelihood of SPEP testing. Further research is needed to quantify the relative importance of these factors in SPEP testing, understand their combined influence on subsequent disease diagnoses, and identify individuals more likely to receive SPEP tests who ultimately receive an MGUS diagnosis.

No relevant conflicts of interest to declare.