Gunnarson, Caden

Monoclonal gammopathy of undetermined significance (MGUS), smoldering multiple myeloma (SMM), and active multiple myeloma (MM) represent a continuum across the spectrum of plasma cell disorders. While patients with MGUS can be observed, it has a risk of progression to MM, while patients with SMM have a higher risk of progression to MM. Determining the risk of progression is important for proper patient care and elimination of anxiety, especially for patients identified at very low risk of progression. In contrast, identifying patients at high risk may reduce morbidity by more active surveillance or the introduction of therapy. We hypothesized that Fourier Transform InfraRed Spectroscopic analysis of serum combined with machine learning (FTIR/ML) could be used to facilitate this decision-making and correctly stratify patients' risk. Methods: After Institutional Review Board approval, we identified 99 patients with MGUS who did not progress to SMM/MM or AL amyloidosis, 96 patients with SMM who did not progress to active MM over a 2-year interval from diagnosis, and another 96 SMM patients who progressed to MM within two years of diagnosis. The patients had been diagnosed and followed up at Mayo Clinic in Rochester. Serological molecular fingerprint analysis was then performed using surface- and chemical-enhanced infrared absorption FTIR spectroscopy combined with artificial neural network ML. A randomized 10-fold cross-validation analysis was performed using 262 samples for training and the remaining 29 samples for validation to assess the model across ten independent iterations. Receiver operator characteristic (ROC) analysis was used to assess the FTIR/ML algorithm output compared to the clinical outcome data. Results: The patients with both MGUS and SMM had a similar median age (64.4 years), but patients with SMM tended to have lower hemoglobin compared to MGUS (median 12.4 versus 13.8g/dl respectively). As expected, patients with SMM also had a higher serum M-spike (2.0g/dl versus 1.2) and a higher clonal plasma cell burden in the bone marrow (20% versus 9%). FTIR/ML was able to correctly classify sera into MGUS in 95.8% of samples and SMM in 92.9% of samples. This gave an average accuracy of 94.8%. We performed ROC analysis and determined an area under the curve of 98.7%. The F1 score/precision was 94.9%, with a recall of 94.8%. After establishing a suitable threshold, the ML algorithm was able to correctly classify MGUS in 96.8% and SMM in 97.2% of cases. Conclusion: A single serum test at the time of diagnosis can correctly distinguish between MGUS that will not progress to another plasma cell disorder from smoldering multiple myeloma with a high degree of accuracy. Given that the patients with MGUS that were chosen had not progressed to a plasma cell disorder and were correctly identified, this methodology offers an opportunity to provide reassurance and minimize the need for repeat testing and follow-up of such patients.

The prognosis of patients with immunoglobulin light or heavy chain amyloidosis (AL/AH) has improved with the advent of more effective therapies. However, diagnosis is often delayed, with a negative impact on overall survival due to irreversible end-organ damage. Therefore, technologies that could facilitate the diagnosis or raise early suspicion of AL/AH amyloidosis in patients would be beneficial to improve outcomes. We hypothesized that Fourier transform infrared spectroscopic analysis of serum combined with machine learning (FTIR/ML) may be able to correctly diagnose AL/AH amyloidosis compared to patients with monoclonal gammopathy of undetermined significance or smoldering multiple myeloma. Methods: After institutional Review Board approval, we identified 90 patients with AL/AH amyloidosis, 99 patients with MGUS, and 192 patients with smoldering multiple myeloma (96 each who did/did not progress to active myeloma within two years of diagnosis) that had been diagnosed and followed up at Mayo Clinic in Rochester. We then performed surface- and chemical-enhanced infrared absorption FTIR/ML-based molecular fingerprint analysis on serum samples taken from the time of diagnosis to train an AI algorithm to distinguish between the two cohorts. A randomized 10-fold cross-validation analysis was performed using 253 samples (90% from each group) for training and the remaining 29 for testing to assess the model across ten independent iterations. We performed a Receiver Operator Characteristic analysis of the outcomes to measure the model's performance. Results: The patients with immunoglobulin-related amyloidosis had involvement of one or more of the following organs: heart, kidneys, liver, nerve, and gastrointestinal tract. Patients diagnosed with amyloidosis had lower serum albumin (3.0 versus 3.6 g/dl) and higher lambda immunoglobulin-free light chain values compared to those with MGUS (28.4 versus 1.92mg/dl). The predominance of lambda immunoglobulin-free light chains is not unexpected but not diagnostic by itself of AL/AH amyloidosis. Patients with SMM had higher serum M protein, higher immunoglobulin kappa free light chains, and higher bone marrow plasma cell burden (20%) compared to either MGUS (9%) or AL amyloidosis patients (10%). However, none of these are diagnostic features. FTIR/ML was able to correctly classify AL/AH amyloidosis in 80% of samples and MGUS in 83.8%, for an accuracy of 82% with an area under the curve of 89.2%. The precision was 82%, with a recall of 82%. After setting an appropriate threshold, the algorithm was able to correctly identify AL/AH amyloidosis in 86.1% of patients and MGUS in 86.6% of patients. When we compared AL/AH amyloidosis with SMM, FTIR/ML correctly classified AL/AH amyloidosis in 94.4% and SMM in 96.4% for an accuracy of 95.7% and an area under curve of 97.5%. The precision was 95.8%, with a recall of 95.7%. After setting a threshold, the algorithm was able to correctly diagnose SMM in 98.3% and AL/AH amyloidosis in 96.2%. Discussion: FTIR/ML analysis of serum can distinguish between AL/AH amyloidosis and both MGUS and SMM with a high degree of accuracy. Given that the diagnosis of immunoglobulin related amyloidosis is often delayed, our approach, based on a single analysis of a serum sample could potentially accelerate the diagnosis of AL/AH amyloidosis and implementation of life saving therapy.

Abstract Background: Smoldering multiple myeloma (SMM) is a state where patients have more than 10% clonal plasma cells or a serum monoclonal protein > 3g/dl but no evidence of end organ damage. Patients with SMM are at risk of developing active multiple myeloma (defined by the CRAB criteria). The best stratification system with respect to the risk of progression to active myeloma is the 20/2/20 (20% plasma cells, M protein >2.0 g/dl, and involved to uninvolved immunoglobulin free light chain ratio >20). The presence of ≥2 criteria predicts a 50 - 60% risk of progression within 2 years. Thus, many patients considered ‘high-risk’ may not progress in this time frame, and patients with ≤1 risk factor can progress to active myeloma. Therefore, improved risk stratification methods are needed to correctly identify patients at high risk of progression. We hypothesized that Fourier Transform InfraRed (FTIR) spectroscopic analysis of serum combined with artificial intelligence (AI) can correctly risk stratify patients with SMM. Methods: We identified 192 SMM patients: 96 who progressed and 96 who did not progress within 2 years of diagnosis. We used stored sera from the time of diagnosis to perform FTIR to determine differences in the spectra of patients who did/did not progress and train an AI algorithm. A randomized 10-fold cross-validation analysis was performed using 172 samples (86 from each group) for training and the remaining 20 for testing to assess the model across 10 independent iterations. Subsequently, we blindly tested the trained model against an independent set of serum samples (N=24) from patients who progressed ≤3 years of diagnosis. Receiver operator characteristic (ROC) analysis was used to assess the FTIR AI algorithm output compared to the clinical outcome data. Results: From the 96 patients who progressed, 48 had complete data for the 20/2/20 risk stratification. 11 patients (23%) had ≤1 of the criteria and so would have been considered at low likelihood to progress. Of the 96 who did not progress within 2 years, 49 had complete data for risk stratification. Of these, 27 (55.1%) met the high-risk criteria of the 20/2/20 classification and would have been predicted to progress within 2 years. FTIR/AI model correctly identified the patients at risk of progression (95% accuracy) with AUC of 0.96. Similarly, it was able to classify patients who did not progress within 2 years of diagnosis (94% accuracy) with an AUC of 0.96. FTIR/AI correctly predicted the patients at risk of progression within 2 years (90%) and 3 years (87.5%) in the blinded validation group with an AUC of 0.89 and 0.82, respectively. Conclusion: Analysis of serum in patients with newly diagnosed SMM with FTIR is able to correctly identify those at risk of progression within 2 or 3 years of diagnosis with an accuracy of ~90% in a blinded external validation study. FTIR may, therefore, be used to assess SMM patient risk before committing to potentially toxic therapy. Citation Format: David Dingli, Darlyna Khonkhammy, Caden Gunnarson, Grant Schlauderaff, Dan Q. Pham, Ali Khammanivong. Risk stratification of smoldering multiple myeloma using fourier transform infrared spectroscopy of serum [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6170.