Introduction
Multiple myeloma (MM), with an estimated annual incidence of 7.1 per 100,000 population, is the second most common hematologic neoplasm worldwide [ 1 ]. As per the Global Cancer Observatory report, there were 6988 new cases of MM and 4827 deaths in 2022 in Japan [ 2 ]. With a rapidly aging population, the incidence of MM is projected to increase [ 3 - 5 ].
The treatment choice for patients with newly diagnosed multiple myeloma (NDMM) is based on patient-based factors (age, frailty status, comorbidities, compliance, and lifestyle preference), disease-based factors (cytogenetics abnormalities, mutations, biochemical abnormalities, extramedullary disease, and International Staging System score), treatment-related factors (access to therapy, cost, and safety), and the eligibility to undergo autologous stem cell transplantation postinduction therapy [ 6 ]. In Japan, several key therapies have been approved for MM treatment over the past 2 decades: bortezomib (approved in 2006) [ 7 ], lenalidomide (approved in 2010) [ 8 ], pomalidomide (approved in 2014) [ 9 ], and thalidomide (approved in 2008) [ 10 ]. These agents have become standard treatment options, demonstrating clinically meaningful improvements in both progression-free and overall survival for patients with MM. For transplant-ineligible patients with NDMM and relapsed or refractory multiple myeloma (RRMM), novel therapies (such as panobinostat, elotuzumab, carfilzomib, ixazomib, daratumumab, and isatuximab) are commonly used treatment modalities [ 11 , 12 ].
Although patient survival is increasing with the advancement in treatment options and supportive care [ 4 , 5 ], MM is still considered an incurable disease as patient survival solely depends on the prescribed treatment regimen and the patient’s demographic and clinical characteristics [ 1 ].
Treatment continuation for an appropriate duration plays an important role in maximizing treatment efficacy in patients with MM, as it is directly correlated with patient survival [ 12 ]. Our previous real-world study using patients’ data from the Medical Data Vision (MDV) claims database in Japan found that the attrition rate was increased with each line of treatment (first line [transplant-conducted]-third line). Subsequently, the treatment duration became shorter in Japanese patients with MM from 2016 to 2020 (6.7-8.6 months) compared with 2003-2015 (8.8-10.0 months); the reduction was mostly seen in older adult patients (aged >80 y) [ 12 ]. Although the accurate reason for the shorter duration of therapy (DoT) was not established, it was hypothesized that older age, comorbid disease, availability of newer treatment regimens, safety and tolerability of treatments, and activities of daily living (ADLs) could have impacted the DoT in patients with MM [ 13 - 15 ]. Hence, accurate evaluation of the predictive factors that shorten the DoT could help in the treatment decision-making process.
Artificial intelligence (AI) using machine learning (ML) has been continuously used for the diagnosis and treatment of diseases. Using different tools and algorithms, AI eases the human experts’ tasks by computerization of the same. Informed evaluations are generated by detecting relationships in information [ 16 ]. Therefore, several studies have used different AI and ML algorithms for the diagnosis and risk stratification of MM using patient-level data as well as to create decision-making models [ 16 - 18 ].
This analysis is intended to identify predictive factors that shorten DoT at a threshold of 3, 6, and 12 months in patients with MM who had received first, second, and third-line treatments by applying an ML model using the patients’ data (type of first treatment and patient characteristics) from the MDV claims database.
Methods
Data Sources
This was a nationwide, retrospective observational cohort study conducted with secondary use of the Japan claims database over a 19-year period between 2003 and 2022. The study used longitudinal patient data from the MDV (MDV Co, Ltd) claims database, which is built over 450 Japanese hospital administrative data (covers approximately 26% of “Diagnosis Procedure Combination hospitals,” ie, the hospitals that perform higher than the lowest in the university hospital group), containing over 38 million patient data since the year 2003. The MDV database is a national, patient-level, anonymized, and longitudinal claims database including hospital claims, discharge summaries, hospitalization data, and outpatient and prescription data. The ICD-10 ( International Statistical Classification of Diseases, Tenth Revision ) was used for disease diagnosis, and the medical procedures were coded using Japanese Procedure Codes. Hospitalized death and other clinical evaluations at discharge were available from the “discharge summary” data.
The study period was divided into a look-back period, treatment (time-at-risk) period, and follow-up period. The look-back period for covariates was the period from the first record of MM in the MDV database till the index date. Time-at-risk was the period from the start of the treatment to the end of the treatment for MM, while follow-up time was the period from the end of the treatment for MM to the last record registered in the MDV database. The index date was the record at the start of the treatment line in patients with transplant-ineligible NDMM or RRMM between 2003 and June 2021.
Study Population
Male and female patients aged 18 years and older at index date with a confirmed diagnosis of MM (transplant-ineligible NDMM or RRMM), between 2003 and 2022, were included. Patients were required to have confirmed administration of any of the treatment regimens for first line for the transplant-ineligible NDMM or second line and third line for the RRMM. The unit of analysis (sample) in this study was an individual treatment of first line used for the patients who were transplant-ineligible NDMM or RRMM, second line, and third line. Patients whose DoT cannot be calculated were excluded from the ML analysis.
Variables and Assessment
This study was conducted on 4848 samples for MM using 647 variables extracted from the MDV database. Variables were selected based on the collected opinions of local clinical experts and publications, and the results of a feasibility analysis. Final variables were determined after consulting local clinical experts on the results of the feasibility analysis and were broadly classified as patient, MM treatment-related, and MM treatment-related comorbidity management variables ( Table 1 ). The primary end point was to identify the factors that correlate with the short DoT at the threshold of 3, 6, and 12 months. The DoT was defined as the time from the initiation of a line of treatment to the last day of the line of treatment.
Table 1
Feature variables extracted from the Medical Data Vision database to identify the important predictors associated with the duration of therapy.
a MM: multiple myeloma. b Variables used to develop machine learning models and understand the characteristics of clusters. c eGFR: estimated glomerular filtration rate. d ADL: activity of daily living. e Variables used only to understand the characteristics of clusters (ie, variables not used to develop machine learning model). f IV: intravenous. g SC: subcutaneous. h IMiDs: immunomodulatory drugs. i PI: proteasome inhibitors. j CD: cluster of differentiation. k SLAM7: Signaling Lymphocytic Activation Molecule Family Member 7.
| Patient variables | MM a treatment-related variables | MM treatment-related comorbidity management variables |
| Older age at the MM diagnosis (≥75 y, ≥80 y) b Gender b Height, weight b Comorbidity (Charlson Index) b eGFR c , blood pressure b Leukocytes, neutrophils b Lactate dehydrogenase b Prothrombin time b Independence (ADL d ), Performance status b Under nutrition b Cardiac function b Presence or absence of eating or swallowing dysfunction b Frequency and history of depressive symptoms b,e Frequency and history of use of antidepressants b,e Frequency and history of use of hypnotics b,e Year at diagnosis of MM b | Use of Bortezomib, Lenalidomide in previous treatment regimen b Number of previous lines of regimens b Use of triplet therapy in second line e Use of oral drugs in the next line of regimens e Stem cell transplantation conducted after the treatment regimen e Steroid dose, timing of dose reduction e Type of drug formulations in each regimen (eg, oral, with IV f , or with SC g ) e Type of hospital (ie, oncology-focused site), hospital size b Type of treatment at molecule level (eg, IMiDs h -based, PI i -based, PI or IMiDs-based, anti-CD j 28 mono, anti-CD38 combination, anti-SLAMF7 k , anti-SLAMF7 combination, and other) b | Frequency and history of neutropenia management treatment (history of use of granulocyte-colony stimulating factor, frequency of administration, etc) b,e Frequency and history of antibiotic use (levofloxacin) b,e Frequency and history of prophylactic antibiotics b,e History and frequency of antibiotic use for moderate to severe infections b,e Frequency and history of pneumonia b,e Frequency and history of thrombotic events b,e Frequency and history of aspirin or direct oral anticoagulant use e |
Data Analysis
Data analysis included 5 steps as discussed in Figure 1 .
Variable Extraction
ML models were constructed using 647 variables extracted from the MDV database to identify important predictors associated with DoT. The deep learning used in this study was applied to the exploratory research of biomarkers that are associated with the response of the immune checkpoint inhibitors [ 19 , 20 ]. The binary variables that represented DoT within 3, 6, or 12 months were categorized as target variables, while variables used to develop ML models and understand the characteristics of clusters listed in Table 1 were categorized as explanatory variables.
Feature Engineering
All the feature variables were classified into continuous variables, binary variables, and nominal categorical variables. Continuous variables are normalized to have a mean of zero and an SD of 1 to eliminate bias due to differences in scale between variables. Binary variables were encoded to −1 or 1, and one-hot encoding (1 and −1) was used for nominal categorical variables [ 21 ]. The variables with missing values were imputed with 0 because the 0-value input would not change the output in the neural network models [ 22 ]. The target variables were set as a binary class and presented in 3 threshold patterns (0 for DoT <3 months, <6 months, or <12 months; and 1 for DoT ≥3 months, ≥6 months, or ≥12 months).
Model Generation and Evaluation
Unlike the randomized controlled trials used in clinical trials, the patients enrolled in the MDV database used in this study are a heterogeneous patient population ( Figure 2 A). In other words, the impacts of each variable on DoT differ across patients. We used point-wise linear (PWL) models to predict DoT. The PWL model was used as the explainable deep learning model, which generated a custom-made logistic regression (LR) model for each sample by a meta-learning approach ( Figure 2 B) [ 23 ]. Especially, although the LR model has a weight vector w for the original variable vector x ( n ) ([n] is the sample index), the PWL model has a point-wise weight vector w ( n ) for the original variable vector x ( n ) ( Figure 2 B). On the other hand, extreme gradient boosting (XGBoost) [ 24 , 25 ] is a decision-trees–based machine learning model capable of addressing patients’ heterogeneity. Although XGBoost achieves as high a performance as deep learning models do, in general, XGBoost cannot calculate the weight, that is, the impact on the DoT of each variable. In the finalization of the DoT prediction model, the area under the receiver operating characteristic curve (AUROC), recall, precision, and F 1 -score were calculated. The prediction performance of the PWL model was compared with the XGBoost (gradient-boosted decision trees) [ 24 , 25 ] and the LR model with a combination of L1 (Lasso) and L2 (Ridge) regularization terms (commonly known as elastic-net). The LR model is similar to PWL in that it explains feature importance by referencing the magnitude of weights (model parameters). The key difference between LR and PWL is that LR assigns a single weight coefficient to each feature across all samples, whereas PWL can assign different weights to the same feature for different samples. Consequently, PWL retains the capability to perform individualized factor analysis for each sample. The prediction performances of these models were calculated by the area under the curve (AUC) evaluated by 10-fold double cross-validation [ 26 ]. A “10-fold cross-validation” procedure was used, including 10 repeated attempts after classifying the data into 10 quadrants, with 1 quadrant for evaluation and the remaining 9 for learning. To avoid overlearning in hyperparameter tuning, the learning data were further divided into 10 pieces, 1 of which was used for tuning evaluation. The generalizability of the DoT prediction model was evaluated using the test set of the outer loop of 10-fold double cross-validation, and the datasets were reconstructed or hyperparameters were fine-tuned in case any significant difference (>0.1) was confirmed between the mean AUC in the test set and the training set.
Clustering
A clustering analysis of 4848 individual samples was performed based on the similarity of the coefficients of custom-made LR models calculated from the PWL model to understand the number of patterns of the relationship between each sample and DoT (3, 6, and 12 months). All the samples were classified into subgroups using the regression coefficient vector tailored to each sample in the PWL model. Cluster analysis was conducted using the k-means method [ 27 , 28 ], and the number of clusters (k) was determined by the Elbow method [ 29 ]. The characteristics of each cluster were analyzed by using all the variables listed in Table 1 . The P values for continuous variables were assessed by t test and 1-way ANOVA, while the P values for categorical variables were assessed by chi-square test. To understand the characteristics within a particular cluster, we categorized all 4848 samples into 2 groups—those predicted not to continue treatment after the respective DoT threshold and those predicted to continue treatment after the respective DoT threshold. Continuous variables and categorical variables significantly different between the groups were assessed by 1-way ANOVA and chi-square test, respectively. The P values were adjusted by the Bonferroni correction method.
Ethical Considerations
All data used in this study were deidentified and anonymized patient records obtained from the MDV database for secondary research purposes. In accordance with the Ethical Guidelines for Epidemiological Research issued by the Japanese Ministry of Health, Labor, and Welfare, the study did not require informed consent or approval from an Institutional Review Board [ 30 ].
Results
Baseline Demographics and Clinical Characteristics
This analysis was conducted in 2762 patients (4848 individual samples) with MM. Of these, 1450 (52.5%) were male; 2079 (75.3%) patients were included during the index year 2016-2021. At the first MM diagnosis, the mean age, Charlson Comorbidity Index (CCI) score, and ADL score were 69.6 (SD 11.8) years, 0.5 (SD 1.1), and 14.6 (SD 7.2), respectively ( Table 2 ).
Table 2
Baseline demographics and clinical characteristics of patients with multiple myeloma.
a MM: multiple myeloma. b CCI: Charlson Comorbidity Index. c ADL: activity of daily living. d Missing data were present and these entries were not included in the table. e The number of patients with missing values was 1404. f The number of patients with missing values was 2110. g The number of patients with missing values was 2530. h The number of patients with missing values was 2759. i The number of patients with missing values was 2720. j The number of patients with missing values was 2040.
| Characteristics | Overall (N=2762) | |
| Age at first MM a diagnosis, mean (SD) | 69.6 (11.8) | |
| Age groups at first MM diagnosis (y), n (%) | ||
| 18-59 | 540 (19.6) | |
| 60-64 | 344 (12.5) | |
| ≥65 | 1878 (68) | |
| Sex, n (%) | ||
| Male | 1450 (52.5) | |
| Female | 1312 (47.5) | |
| Index year, n (%) | ||
| 2003-2015 | 683 (24.7) | |
| 2016-2021 | 2079 (75.3) | |
| CCI b score at first MM diagnosis, mean (SD) | 0.5 (1.1) | |
| CCI score at first MM diagnosis, n (%) | ||
| 0 | 2121 (76.8) | |
| 1 | 281 (10.2) | |
| 2 | 177 (6.4) | |
| 3 | 99 (3.6) | |
| ≥4 | 84 (3) | |
| ADL c score at first MM diagnosis, mean (SD) d,e | 14.6 (7.2) | |
| ADL score at first MM diagnosis, n (%) d,e | ||
| 0 | 111 (8.2) | |
| 1 | 38 (2.8) | |
| 2 | 31 (2.3) | |
| 3 | 18 (1.3) | |
| 4-19 | 466 (33.3) | |
| 20 | 694 (51.1) | |
| Degree of care requirement at diagnosis of MM, mean (SD) d,f | 0.4 (1.2) | |
| Degree of care requirement at diagnosis of MM, n (%) d,f | ||
| 0 | 564 (86.5) | |
| 1 | 24 (3.7) | |
| 2 | 14 (2.1) | |
| 3 | 16 (2.5) | |
| ≥4 | 34 (5.2) | |
| Tube and parenteral nutrition at first MM diagnosis, n (%) d,g | ||
| 0 | 214 (92.2) | |
| 1 | 17 (7.3) | |
| 2 | 1 (0.4) | |
| Stage of MM at first MM diagnosis d,h | ||
| Mean (SD) | 5.0 (0.0) | |
| 5, n (%) | 3 (100) | |
| NYHA at first MM diagnosis, mean (SD) d,i | 2.1 (1.1) | |
| NYHA at first MM diagnosis, n (%) d,i | ||
| 0 | 2 (4.8) | |
| 1 | 15 (35.7) | |
| 2 | 7 (16.7) | |
| 3 | 14 (33.3) | |
| 4 | 4 (9.5) | |
| Cancer at first MM diagnosis, n (%) d,j | ||
| Primary | 706 (97.8) | |
| Recurrence | 16 (2.2) |
Prediction Performance
The AUROC score of the PWL model to predict DoT at 3, 6, and 12 months was 0.61, 0.64, and 0.66, respectively. The corresponding values for the XGBoost model were 0.62, 0.64, and 0.66, respectively. The PWL model with the target variable DoT at 12 months performed the best; however, no model reached an AUC>0.7 ( Figure 3 ). Moreover, the slightly smaller SD in PWL suggested more consistent performance across different folds of cross-validation, making it the model of choice for this analysis. The recall, precision, and F 1 -score of all models are presented in the supplementary material ( Multimedia Appendix 1 ). The precision and F 1 -score were decreased when the DoT length increased, and the thresholds shortened.
Cluster Analysis
Overview
The PWL model generated regression models for each patient, allowing the clustering of patient populations based on the similarity of the coefficients of the regression models. When classified into different clusters, it was found that there were multiple populations with diverse groups of factors affecting the prediction. The samples were categorized into 2 clusters (cluster A and cluster B) at DoT of 3 months and 3 clusters (cluster A, cluster B, and cluster C) at each DoT of 6 months and DoT of 12 months. The baseline demographics and clinical characteristics of 4848 individual samples categorized as per different clusters at DoT of 3, 6, and 12 months are presented in Multimedia Appendix 2 . In addition, we presented the distribution of each cluster in the 2 major components of the variables transformed by principal component analysis in the supplementary materials ( Multimedia Appendix 3 ).
At DoT of 3 Months
Of the 4848 samples, 3287 (67.8%) samples were included in cluster A and 1561 (32.2%) were included in cluster B. Compared with cluster A, cluster B had more patients with age <75 years, male sex, index year 2003-2015, and CCI ≥4 ( Multimedia Appendix 2 ). In the probability analysis, cluster B reported a lower median probability than cluster A, with pointed and extended left tails ( Figure 4 A). From Figure 4 A, it can be estimated that many patients in cluster B might be in poor condition, but some were predicted to continue treatment for ≥3 months. Compared with cluster A, patients in cluster B had significantly more lines of previous therapy (0.5 vs 1.6; P <.001), higher pretreatment CCI (2.2 vs 2.5; P <.001), and a higher level of care requirement at the time of diagnosis of MM (0.4 vs 0.7; P= .001). Additionally, cluster B had significantly ( P <.001) higher proportions of patients with pneumonia, depression, cancer, gastroesophageal reflux disease (GERD), other bowel dysfunctions, primary hypertension, pain, purine and pyrimidine metabolism disorders, and type 2 diabetes mellitus (T2DM) than cluster A ( Multimedia Appendix 2 ).
At DoT of 6 Months
Of the 4848 samples, 2011 (41.5%) were included in cluster A, 1736 (35.8%) were included in cluster B, and 1101 (22.7%) were included in cluster C. Compared with cluster A and B, cluster C had more patients with age <75 years, female sex, the index year 2016-2021, and CCI ≥4 ( Multimedia Appendix 2 ). In the probability analysis, cluster C reported a lower median probability than clusters A and B, with flat and extended right tails ( Figure 4 B). From Figure 4 B, it can be estimated that many patients in cluster C had many comorbidities, but some were predicted to continue treatment for ≥6 months. Patients in cluster B had significantly lower estimated glomerular filtration rate ( P <.001) and higher neutrophils ( P <.001), while cluster C had significantly higher pretreatment CCI ( P <.001) compared to other clusters. Additionally, cluster C had significantly ( P <.001) higher proportions of patients with pneumonia, depression, cancer, GERD, other bowel dysfunction, primary hypertension, pain, purine and pyrimidine metabolism disorders, and T2DM than clusters A and B ( Multimedia Appendix 2 ).
At DoT of 12 Months
Of the 4848 samples, 3425 (70.6%) were included in cluster A, 1123 (23.2%) were included in cluster B, and 300 (6.2%) were included in cluster C. Compared with cluster A and B, cluster C had more patients with male sex, index year 2016-2021, and CCI ≥4 ( Multimedia Appendix 2 ). In the probability analysis, cluster C reported a lower median probability than cluster A and B, with peaked and extended right tails ( Figure 4 C). From Figure 4 C, it can be estimated that many patients in cluster C were in poor condition, but some were predicted to continue treatment for ≥12 months. Cluster C had significantly higher neutrophils ( P <.001), leukocytes ( P <.001), lactate dehydrogenase ( P <.001), previous treatment lines ( P <.001), pretreatment CCI ( P <.001), degree of care requirement at diagnosis of MM ( P <.001), and pretreatment tube and parenteral nutrition ( P <.001) among clusters ( Multimedia Appendix 2 ). Additionally, cluster C had a statistically ( P <.001) higher proportion of patients with comorbidities, including depression, pneumonia, cancer, GERD, other bowel dysfunction, primary hypertension, pain, purine and pyrimidine metabolism disorders, and T2DM, than cluster A and B.
Characteristics of Each Cluster Before and After the DoT Threshold
All 4848 samples were categorized before and after each DoT threshold (3, 6, and 12 months) to determine the factors responsible for shortened DoT. As cluster B at DoT of 3 months and cluster C at DoT of 6 and 12 months were predicted to continue the treatment, patients from these clusters were further divided into those predicted to continue treatment after the DoT threshold, and those predicted to discontinue treatment within the DoT threshold, and the incidence of events during and after treatment was compared. There were a lesser number of treatments included in DoT <3 months than DoT ≥3 months (n=1859, 38.3% vs n=2989, 61.7%), while DoT <6 (n=2753, 56.8%) and <12 months (n=3672, 75.7%) had a higher number of treatments compared with DoT ≥6 (n=2095, 43.2%) and ≥12 months (n=1176, 24.3%), respectively. In cluster B, a significantly ( P <.01) higher percentage of patients used immunomodulatory drugs (IMiDs) and used or maintained aspirin during MM therapy in DoT ≥3 than in DoT <3 months group. However, the rate of subsequent transplantation in the next treatment was significantly ( P <.01) higher in DoT <3 than DoT ≥3 months group. Similarly, in cluster C, a significantly ( P <.01) higher percentage of patients used IMiDs or aspirin during MM therapy in the DoT ≥6 months than in the DoT <6 months group. In cluster C, a significantly ( P <.01) higher percentage of patients used IMiDs during MM therapy in DoT ≥12 months than in the DoT <12 months group ( Table 3 ). A summary of the key clinical traits for each cluster after the DoT threshold analysis is presented in Table 4 .
Table 3
The characteristics between before and after treatment thresholds, as per the duration of therapy.
a Not applicable. b IMiD: immunomodulatory drug.
| Medications | Cluster B | Cluster C | Cluster C | ||||||
| <3 months, n/N (%) | ≥3 months, n/N (%) | Adjusted P value | <6 months, n/N (%) | ≥6 months, n/N (%) | Adjusted P value | <12 months, n/N (%) | ≥12 months, n/N (%) | Adjusted P value | |
| Samples | 1859/4848 (38.3) | 2989/4848 (61.7) | — a | 2753/4848 (56.8) | 2095/4848 (43.2) | — | 3672/4848 (75.7) | 1176/4848 (24.3) | — |
| Use of IMiDs b | 261/714 (36.6) | 624/847 (73.7) | <.01 | 311/763 (40.8) | 238/338 (70.4) | <.01 | 91/240 (37.9) | 43/60 (71.7) | <.01 |
| Use of aspirin | 206/714 (28.9) | 431/847 (50.9) | <.01 | 244/763 (32.0) | 170/338 (50.3) | <.01 | 69/240 (28.8) | 32/60 (53.3) | .77 |
| Continue aspirin | 125/714 (17.5) | 236/847 (27.9) | <.01 | — | — | — | — | — | — |
| Subsequent transplant | 90/714 (12.6) | 19/847 (2.2) | <.01 | — | — | — | — | — | — |
Table 4
A summary of the key clinical characteristics distinguishing clusters.
a DoT: duration of therapy. b CCI: Charlson Comorbidity Index. c IMiD: immunomodulatory drug. d LD: lactate dehydrogenase.
| Cluster | DoT a threshold | Key characteristics | CCI b ≥4, n/N (%) | IMiD c use (≥DoT), n/N (%) | Aspirin use (≥DoT), n/N (%) |
| Cluster B | 3 months | Higher comorbidity, more prior therapy | 393/1561 (25.2) | 624/847 (73.7) | 431/847 (50.9) |
| Cluster C | 6 months | High comorbidity, more comorbid events | 383/1101 (34.8) | 238/338 (70.4) | 170/338 (50.3) |
| Cluster C | 12 months | High neutrophils, LD d , leukocytes | 127/300 (42.3) | 43/60 (71.7) | 32/60 (53.3) |
Discussion
Principal Findings
This retrospective real-world observational study using longitudinal patients’ data from the nationwide MDV claims database analyzed the predictive factors that shorten DoT in Japanese patients with MM over a 19-year period (2003-2022). Additionally, this analysis used ML models to determine the factors responsible for shorter DoT using information, such as type of treatment and patient characteristics from the MDV claims database, categorizing all the data in different clusters at 3, 6, and 12 months. Finally, a clustering analysis was conducted to investigate the characteristics of the sample in each cluster and the relationship between predicted DoT and the comorbidity management before and after the treatment thresholds (3, 6, and 12 months).
Patients in cluster B than cluster A (at 3 months), and patients in cluster C than cluster A and B (at 6 and 12 months) were supposed to be in poor condition as the number of treatment lines and pretreatment CCI were high; patients had more comorbidities, and elevated laboratory parameters. Cluster B (3-month DoT) and cluster C (6- and 12-month DoT) consistently captured subgroups with high comorbidity burden and elevated pretreatment CCI, yet demonstrated longer treatment durations. Additionally, the use of IMiDs for MM treatment was significantly higher in patients who met predicted DoT at each threshold versus those who did not. In contrast to the clinical findings, the data from this study indicate that the choice of drug, irrespective of whether it is used for induction or maintenance therapy, affects the expected DoT.
Comparison With Previous Work
From our recent real-world study in Japan, it was found that the DoT decreased continuously with increasing the line of treatment in patients with MM, especially after 2016 when several newer therapy options were in force [ 12 ]. In clinical practice, treatment may be discontinued for various reasons; however, for the desired prognosis, it is important to continue the medications for an appropriate duration. The DoT has a strong association with the clinical outcomes and survival in patients with MM [ 31 , 32 ]; therefore, it was inevitable to determine the factors affecting the shortened DoT. Although it was evident that the availability of new drugs might increase the treatment options, making it easier to change treatment regimens, the huge pattern of treatment makes it difficult to identify the cause of shortened DoT. Therefore, this analysis was conducted using the national MDV claims database to determine the predictive factors that could help clinicians in the decision-making process while considering the management of MM. Additionally, this study found the characteristics of patients (age, CCI, comorbidity, treatment lines, laboratory parameters, and care requirement) who had shorter DoT as well as the characteristics of patients with frequent treatment change, which reflects the actual picture of clinical practice. This information is critical to continually improving MM management, achieving optimal patient care, and understanding the burden of disease and health services available to Japanese patients with MM in the real world.
In line with the published evidence [ 33 , 34 ], our study compared the prediction performance of the PWL model with the XGBoost and the LR models. The AUROC values of constructed models were all between 0.6 and 0.7 ( Figure 2 ). Values between 0.6 and 0.7 indicate satisfactory performance [ 35 - 37 ]. The ranges observed in this study are consistent with those reported in previous real-world data–based modeling studies [ 33 , 38 , 39 ]. Administrative claims data often face inherent data limitations, such as the absence of genomic or staging variables, which can constrain predictive performance. In real-world settings, such models can still provide clinically relevant insights for subgroup identification and hypothesis generation. Nonetheless, models built on such data can still yield clinically relevant insights—particularly for subgroup identification and hypothesis generation in real-world settings. In our study, we generated the models using the variables required in the claims database. For improvements in prediction performance, the use of MM-specific laboratory parameters might be required. Furthermore, in our study, the PWL model showed greater AUROC than the linear regression model at the DoT thresholds of 3, 6, and 12 months. However, these scores were the differences within the error bars between PWL and XGBoost models at each DoT threshold.
DoT Assessment Using PWL Models
DoT predictive models in our study were specifically designed to forecast the DoT in patients with MM, rather than to evaluate clinical effectiveness between specific treatment groups. Due to this focus, these models were not intended for direct application in clinical decision-making or disease diagnosis. Consequently, we did not perform propensity score matching or sensitivity analysis, which are typically used in studies evaluating treatment efficacy. Instead, our efforts were concentrated on carefully developing and comparing the PWL model with the XGBoost and LR models to thoroughly assess the factors influencing DoT. The PWL model differs from LR by generating a separate set of regression coefficients for each individual sample, rather than using a single global set. This approach allows identification of factors that are most relevant for each patient, which may enhance interpretability and personalization in real-world datasets where treatment decisions and outcomes are highly individualized. Unlike randomized controlled trials, the MDV claim database includes a diverse and unselected patient population, without adjustments for specific demographics or conditions. This makes the data more representative of real-world variability. Multiple patterns can thus be anticipated based on the combination of variables that affect DoT. The LR model cannot recognize multiple patterns because the model has only a single regression coefficient to a variable for the whole sample. On the other hand, since the PWL model has a custom-made regression coefficient, it is possible to perform individual analysis for each sample. Therefore, we performed the cluster analysis using the regression coefficient vector tailored to each sample in the PWL model to understand the number of patterns. The PWL model enabled case-level insights, highlighting which patient features most influenced predicted treatment duration. Such interpretability is especially valuable in clinical settings, where understanding the rationale behind risk stratification supports shared decision-making. In our study, the k-means method was used for cluster analysis to classify all the samples into subgroups, and the Elbow method was used to determine the number of clusters. The k-means method is an unsupervised ML approach widely used in research for clustering samples based on similarities [ 28 , 40 ]. Additionally, the k-means method could substantiate heterogeneous longitudinal data into distinct, homogeneous clusters [ 27 , 28 ]. The Elbow method is an established method to determine the appropriate number of clusters that rides the clustering algorithm many times with different values of k (number of clusters) [ 41 ], and defines the optimal number of clusters to efficiently categorize the data [ 42 , 43 ].
From the baseline demographics and predictors of cluster assignment, it was found that patients in cluster B at DoT of 3 months and cluster C at DoT of 6 and 12 months, compared with other clusters, had advanced disease and continued the treatment beyond the threshold duration. Interestingly, we observed that in these clusters, more patients had CCI ≥4, ≥1 line of treatment, and comorbidities (pneumonia, depression, cancer, GERD, and other bowel dysfunction) than in other clusters at each DoT threshold. Additionally, the laboratory parameters such as estimated glomerular filtration rate (MDV specific), leukocytes, neutrophils, and lactate dehydrogenase were significantly higher in the cluster C than the cluster A and B at DoT 12 months. An all-inclusive assessment of comorbidities and overall status (with the help of laboratory parameters) may help to determine a patient’s suitability for appropriate therapy as well as the adverse effects that may occur. Altogether, the results from this study support the understanding that the patient’s condition and management-related factors may be associated with shortened DoT, consistent with clinical findings [ 12 , 44 , 45 ]. Evidence suggests that severe disease, older age, and comorbidities are the prognostic factors for poor survival in patients with MM [ 46 , 47 ]. Although it was not in the scope of this analysis to comment on the causal relationship between predictive factors and patient survival, and their effect on longitudinal outcomes and DoT, our findings validate a depiction of shortened DoT in those with reduced disease burden. This is consistent with the findings of the retrospective study conducted by Ailawadhi S et al [ 32 ], which reported that each additional month of therapy was associated with a reduced probability of disease progression and death at 2 years. Similar results were also observed in a pooled analysis of 3 phase III trials that evaluated the advantage of novel agent-based continuous therapy versus fixed DoT in patients with newly diagnosed myeloma [ 48 ].
To understand the characteristics of the clusters, we analyzed the relationship between the explanatory variables and the flag that indicated the samples belonged to each cluster. Additionally, we calculated the association between the variables during and the flag to investigate the tendency to the event during and after treatment. The significance was detected by the test for noncorrelation at a significance level false discovery rate of <0.05 using the Benjamini-Hochberg method. The Benjamini-Hochberg method is a validated tool used in multiple correlation analyses that could decrease the false discovery rate [ 19 ].
Significantly, more patients used IMiDs and aspirin beyond the treatment thresholds (both ≥3 [cluster B] and ≥6 months [cluster C]) compared with before treatment thresholds (<3 and <6 months), and in cluster C at DoT ≥12 months, a higher proportion of patients used IMiDs during MM therapy compared with DoT <12 months group. The use of IMiDs was significantly higher when comparing sample populations predicted as DoT of <3, <6, and <12 months with those who did not. However, this observational association may reflect confounding factors, such as patient eligibility, tolerability, or physician preference, and does not imply causation. In patients with MM, lenalidomide, one of the commonly used IMiDs, is sometimes used in maintenance therapy, and the results from this study suggest that IMiD therapies are associated with treatment continuation for MM in patients with high CCI, severe disease, and comorbidities. Furthermore, these results imply that the use of appropriate IMiD may impact DoT in patients with MM. Since the treatment choices for patients with MM depend on several factors, including the type of previous therapy and comorbidity, these data may assist the clinical decision-making on whether IMiDs should be used in these patients. Similar to IMiD use, the management of thromboprophylaxis was also associated with treatment continuation for MM in patients with high CCI.
Factors Identified for Shortened DoT
Our analysis identified several key factors that might be associated with shortened DoT for patients with MM. The first factor is the patient’s disease status—patients with poorer health status (CCI ≥4, ≥1 previous line of treatment, and comorbidities including pneumonia, depression, cancer, GERD, or bowel dysfunction), along with elevated laboratory parameters, showed longer DoT. Second was the management-related factors, including use of IMiDs and management of thromboprophylaxis: patients who were treated with IMiDs for MM showed extended DoT.
Strengths and Limitations
The strengths of the study include (1) this study is a precedent of data-driven research to understand the characteristics of patients with MM whose DoT was shortened; (2) by using AI, based on ML techniques, we were able to build a prediction model using many factors that are difficult to build with a conventional statistical model; (3) in this study, subgrouping was automatically done by the ML-based DoT predictive models as subgrouping analysis is required when analyzing real-world data due to heterogeneity of the patient population; (4) this study used the latest data from nationwide MDV claims database which was based on Japanese general population, and represented real-world clinical practice.
This study also has certain limitations. First, the database is restricted to the Japanese population; hence, caution should be taken while generalizing the results. This database only included the data of the insured population, and it was impossible to follow patients outside of the hospital or those who switched hospitals. Not all laboratory tests were conducted in the entire patient population, which again restricts the generalizability of the findings. Furthermore, since the cause of death was unavailable, it could not be determined whether the shorter DoT was due to treatment failure, toxicity, patient preference, or unrelated mortality. Additionally, death was captured only for hospitalized patients, and important prognostic factors for MM, such as cytogenetic profile, R-International Staging System stage, and presence of all plasmacytomas, were not collected. This may affect the clinical granularity and predictive robustness of the models. Their inclusion may further enhance both robustness and interpretability of future predictive tools. We partially addressed this by incorporating surrogate measures, such as CCI and ADL scores. Furthermore, it is difficult to evaluate the causal relationship between predictive factors and DoT, even if the relationship between the factors and shortening of DoT is assumed, and the presence of unknown confounding that affects the continuation of treatment cannot be ruled out.
Conclusion
This study is the first to apply ML to assess predictive factors for shortened DoT in Japanese patients with MM. Using the PWL model, an explainable deep learning method, we identified key patient and management factors influencing DoT. These insights can contribute to improving patient management in hospitals, optimizing care, and enhancing understanding of the burden of disease and health services for Japanese patients with MM in real-world settings. Further statistical analysis is needed to refine patient segmentation and ensure the effective application of these insights in clinical environments.
The authors thank Rabi Panigrahy (PhD), Aafrin Khan, and Jyotsana Dixit (PhD) for writing assistance and Himabindu Gutha, CMPP, for additional editorial support (all from SIRO Clinpharm Pvt Ltd) for the development of the manuscript.
Authors' Contributions: Conceptualization: HH, TI, SO, SI, KW, CS, KK, DBCW, DY
Data curation: SN, YY, TS
Formal analysis: KW, SN, YY, TS
Funding acquisition: KW
Methodology: HH, TI, SO, SI, KW, CS, KK, DBCW, DY, SN, YY, TS
Software: SN, YY, TS
Validation: HH, TI, SO, SI, KW, CS, KK, DBCW, DY, SN, YY, TS
Writing – original draft: KW
Writing – review and editing: HH, TI, SO, SI, KW, CS, KK, DBCW, DY, SN, YY, TS
Conflicts of Interest: HH has received consulting fees and honoraria from Johnson & Johnson, Ono, and Takeda; grants from Takeda, Kyowa Kirin, and BMS. TI has received consulting fees and honoraria from Ono, Takeda, BMS, Johnson & Johnson, and Sanofi; research fundings support from Takeda, Janssen, Pfizer, and BMS. SO has nothing to declare. SI has received consulting fees and honoraria from BMS, Johnson & Johnson, Takeda, Ono, Sanofi, AstraZeneca, and Pfizer; research funding support from BMS, Johnson & Johnson, Daiichi Sankyo, Amgen, Ono, AbbVie GK, GSK, Eli Lilly, Caelum Biosciences, Inc., Pfizer, Takeda, and Sanofi; grants from Chugai, Takeda, Ono, and Sanofi. KW and CS are employees of Johnson & Johnson. KK was an employee of Johnson & Johnson when the study was planned and is currently employed by Gilead Sciences K.K. since September 2022 and have share in Gilead Science. DBCW and DY are employees of Johnson & Johnson. SN is an employee of Hitachi Ltd. YY and TS were employed by Hitachi Ltd. during the time this research was conducted.
Table S1: The precision, recall and F1 score of all models.
Table S2: Details of characteristics of patients with MM treatment stratified by clusters at DoT 3, 6 and 12 months. Missing values from total samples: a4428, b4217, c4221, d4420, e4254, f3624. CCI, Charlson Comorbidity Index; DoC, degree of care; DoT, duration of treatment; eGFR, estimated glomerular filtration rate; GERD, gastro esophageal reflux disease; HTN, hypertension; LoT, line of treatment; MDV, Medical Data Vision; MM, multiple myeloma; T2DM, type 2 diabetes mellitus.
Two-dimensional cluster analysis using PCA at DoT of 3, 6 and 12 months Dimensional compression was performed by PCA to confirm the relationship between clusters in two-dimensional plots. Each result was drawn to correspond to a previously defined cluster. DoT, duration of treatment; PCA, principal component analysis.
Abbreviations
activity of daily living
artificial intelligence
area under the curve
area under the receiver operating characteristic curve
Charlson Comorbidity Index
duration of therapy
International Statistical Classification of Diseases, Tenth Revision
gastroesophageal reflux disease
immunomodulatory drugs
logistic regression
Medical Data Vision
machine learning
multiple myeloma
newly diagnosed multiple myeloma
point-wise linear
relapsed or refractory multiple myeloma
type 2 diabetes mellitus
extreme gradient boosting
Data Availability
The data that support the findings of this study were derived from the MDV database. Restrictions apply to the availability of these data, which were used under license for this study.
Funding
This study was funded by Janssen Pharmaceutical K.K., Japan, and Johnson & Johnson.
1 Fazio M Del Fabro V Parrinello N Allegra A Markovic U Botta C Accardi F Vincelli I D Leotta S Elia F Esposito B Garibaldi B Sapuppo G Orofino A Romano A Palumbo G A Di Raimondo F Conticello C Multiple myeloma in 2023 ways: from trials to real life Curr Oncol 2023 30 11 9710 9733 10.3390/curroncol30110705 37999125 curroncol30110705 37999125 PMC10670159
2 World Health Organization (2022) Global Cancer Observatory: Cancer Today (Japan) Global Cancer Observatory: Cancer Today (Japan) World Health Organization 2024-03-05 https://gco.iarc.who.int/media/globocan/factsheets/populations/392-japan-fact-sheet.pdf
3 Suzuki N Okuyama T Akechi T Kusumoto S Ri M Inagaki A Kayukawa S Yano H Yoshida T Shiraga K Hashimoto H Aiki S Iida S Symptoms and health-related quality of life in patients with newly diagnosed multiple myeloma: a multicenter prospective cohort study Jpn J Clin Oncol 2022 52 2 163 169 10.1093/jjco/hyab178 34865077 6446274 34865077
4 Cancer RS2 Cancer Statistics in Japan—2023 Foundation for Promotion of Cancer Research 2024-03-05 https://ganjoho.jp/public/qa_links/report/statistics/pdf/cancer_statistics_2023.pdf
5 Cancer RS2 Cancer Statistics in Japan—2018 Foundation for Promotion of Cancer Research 2024-03-05 https://ganjoho.jp/public/qa_links/report/statistics/pdf/cancer_statistics_2018.pdf
6 Costello C Newly diagnosed multiple myeloma: making sense of the menu Hematology Am Soc Hematol Educ Program 2022 2022 1 539 550 10.1182/hematology.2022000404 36485145 493541 36485145 PMC9820388
7 Ogawa Y Tobinai K Ogura M Ando K Tsuchiya T Kobayashi Y Watanabe T Maruyama D Morishima Y Kagami Y Taji H Minami H Itoh K Nakata M Hotta T Phase I and II pharmacokinetic and pharmacodynamic study of the proteasome inhibitor bortezomib in Japanese patients with relapsed or refractory multiple myeloma Cancer Sci 2008 99 1 140 4 10.1111/j.1349-7006.2007.00638.x 17970782 CAS638 17970782 PMC11159153
8 Uno S Motegi Y Minehata K Aoki Y Lenalidomide treatment of Japanese patients with myelodysplastic syndromes with 5q deletion: a post-marketing surveillance study Int J Hematol 2023 118 4 432 442 10.1007/s12185-023-03634-7 37493867 10.1007/s12185-023-03634-7 37493867 PMC10522743
9 Matsue K Iwasaki H Chou T Tobinai K Sunami K Ogawa Y Kurihara M Midorikawa S Zaki M Doerr T Iida S Pomalidomide alone or in combination with dexamethasone in Japanese patients with refractory or relapsed and refractory multiple myeloma Cancer Sci 2015 106 11 1561 7 10.1111/cas.12772 26292221 26292221 PMC4714698
10 Murakami H Shimizu K Sawamura M Suzuki K Sugiura I Kosugi H Shimazaki C Taniwaki M Abe M Takagi T Phase II and pharmacokinetic study of thalidomide in Japanese patients with relapsed/refractory multiple myeloma Int J Hematol 2009 89 5 636 41 10.1007/s12185-009-0314-5 19399582 19399582
11 Uno S Midorikawa S Inoue K Ichikawa D Ito T Kuroda J Suzuki K Survival outcomes among patients with multiple myeloma in the era of novel agents: exploratory assessment using an electronic medical record database in Japan PLoS One 2023 18 5 e0285947 10.1371/journal.pone.0285947 37256880 PONE-D-22-34745 37256880 PMC10231788
12 Handa H Ishida T Ozaki S Mori A Kato K Iida S Treatment pattern and clinical outcomes in multiple myeloma patients in Japan using the Medical Data Vision claims database PLoS One 2023 18 4 e0283931 10.1371/journal.pone.0283931 37023056 PONE-D-22-33974 37023056 PMC10079007
13 Lee HC Ailawadhi S Gasparetto CJ Jagannath S Rifkin RM Durie BGM Narang M Terebelo HR Toomey K Hardin JW Wagner L Omel JL Dhalla M Liu L Joshi P Abonour R Connect® MM Registry Treatment patterns and outcomes in elderly patients with newly diagnosed multiple myeloma: results from the Connect MM Registry Blood Cancer J 2021 11 7 134 10.1038/s41408-021-00524-1 34301916 10.1038/s41408-021-00524-1 34301916 PMC8302734
14 Shin J Lee J Kim K Min C Lee J Suh C Kim JS Lee YJ Yoon S Jo J Lee HS Bang S Korean Multiple Myeloma 151 Investigators Venous thromboembolism in relapsed or refractory multiple myeloma patients treated with lenalidomide plus dexamethasone Int J Hematol 2019 109 1 79 90 10.1007/s12185-018-2540-1 30291558 10.1007/s12185-018-2540-1 30291558
15 Palumbo A Bringhen S Mateos M Larocca A Facon T Kumar S Erratum: Palumbo A, Bringhen S, Mateos M-V, et al. Geriatric assessment predicts survival and toxicities in elderly myeloma patients: an International Myeloma Working Group report. . 2015;125(13):2068-2074 Blood 2016 128 7 1020 10.1182/blood-2016-07-727081 31265505 S0006-4971(20)34244-0 PMC4375104 25628469
16 Allegra A Tonacci A Sciaccotta R Genovese S Musolino C Pioggia G Gangemi S Machine learning and deep learning applications in multiple myeloma diagnosis, prognosis, and treatment selection Cancers (Basel) 2022 14 3 606 10.3390/cancers14030606 35158874 cancers14030606 35158874 PMC8833500
17 Mosquera Orgueira A González Pérez M S Diaz Arias J Rosiñol Laura Oriol A Teruel AI Martinez Lopez J Palomera L Granell M Blanchard MJ de la Rubia J López de la Guia Ana Rios R Sureda A Hernandez MT Bengoechea E Calasanz MJ Gutierrez N Martin ML Blade J Lahuerta J San Miguel J Mateos MV PETHEMA/GEM Cooperative Group Unsupervised machine learning improves risk stratification in newly diagnosed multiple myeloma: an analysis of the Spanish Myeloma Group Blood Cancer J 2022 12 4 76 10.1038/s41408-022-00647-z 35468898 10.1038/s41408-022-00647-z 35468898 PMC9038663
18 Williams L Khosravi B Velimirovic M Khouri J Raza S Mazzoni S An ensemble machine learning model using gradient boosting identifies patients with disease progression in newly diagnosed multiple myeloma Blood. 2023 2023 142 1 1990 10.1182/blood-2023-188762
19 Benjamini Y Drai D Elmer G Kafkafi N Golani I Controlling the false discovery rate in behavior genetics research Behav Brain Res 2001 125 1-2 279 84 10.1016/s0166-4328(01)00297-2 11682119 S0166-4328(01)00297-2 11682119
20 Kumagai S Togashi Y Kamada T Sugiyama E Nishinakamura H Takeuchi Y Vitaly K Itahashi K Maeda Y Matsui S Shibahara T Yamashita Y Irie T Tsuge A Fukuoka S Kawazoe A Udagawa H Kirita K Aokage K Ishii G Kuwata T Nakama K Kawazu M Ueno T Yamazaki N Goto K Tsuboi M Mano H Doi T Shitara K Nishikawa H The PD-1 expression balance between effector and regulatory T cells predicts the clinical efficacy of PD-1 blockade therapies Nat Immunol 2020 21 11 1346 1358 10.1038/s41590-020-0769-3 32868929 10.1038/s41590-020-0769-3 32868929
21 Samuels J Samuels J. One-Hot Encoding and Two-Hot Encoding 2024 Lake Rumare Imperial 2024
22 Hirakawa Y Yoshioka K Kojima K Yamashita Y Shibahara T Wada T Nangaku M Inagi R Potential progression biomarkers of diabetic kidney disease determined using comprehensive machine learning analysis of non-targeted metabolomics Sci Rep 2022 09 29 12 1 10.1038/s41598-022-20638-1 36175470 PMC9523033 36175470
23 Shibahara T Wada C Yamashita Y Fujita K Sato M Kuwata J Okamoto A Ono Y Deep learning generates custom-made logistic regression models for explaining how breast cancer subtypes are classified PLoS One 2023 18 5 e0286072 10.1371/journal.pone.0286072 37216350 PONE-D-22-20429 37216350 PMC10202302
24 Boldini D Grisoni F Kuhn D Friedrich L Sieber SA Practical guidelines for the use of gradient boosting for molecular property prediction J Cheminform 2023 15 1 73 10.1186/s13321-023-00743-7 37641120 10.1186/s13321-023-00743-7 37641120 PMC10464382
25 Chen T Guestrin C XGBoost: a scalable tree boosting system 2016 22nd ACM SIGKDD International Conference 2026 Aug 16 San Francisco, California, USA Association for Computing Machinery 10.1145/2939672.2939785
26 Filzmoser P Liebmann B Varmuza K Repeated double cross validation Journal of Chemometrics 2009 23 4 160 171 10.1002/cem.1225
27 Genolini C Pingault J Driss T Côté S Tremblay R Vitaro F Arnaud C Falissard B KmL3D: a non-parametric algorithm for clustering joint trajectories Comput Methods Programs Biomed 2013 109 1 104 11 10.1016/j.cmpb.2012.08.016 23127283 S0169-2607(12)00213-1 23127283
28 Ko J Furby H Ma X Long JD Lu X Slowiejko D Gandhy R Clustering and prediction of disease progression trajectories in Huntington's disease: An analysis of Enroll-HD data using a machine learning approach Front Neurol 2022 13 1034269 10.3389/fneur.2022.1034269 36793800 36793800 PMC9923354
29 Madhulatha T An overview on clustering methods arXiv 2012 10.48550/arXiv.1205.1117
30 Ministry of Education, Culture, Sports, Science and Technology Ministry of Health, Labour and Welfare Ministry of Economy, Trade and Industry Ethical Guidelines for Medical and Biological Research Involving Human Subjects. Tokyo (JP) Ministry of Education, Culture, Sports, Science and Technology 2026-02-12 https://www.mext.go.jp/content/20250325-mxt_life-000035486-01.pdf
31 Hari P Romanus D Palumbo A Luptakova K Rifkin RM Tran LM Raju A Farrelly E Noga SJ Blazer M Chari A Prolonged duration of therapy is associated with improved survival in patients treated for relapsed/refractory multiple myeloma in routine clinical care in the United States Clin Lymphoma Myeloma Leuk 2018 18 2 152 160 10.1016/j.clml.2017.12.012 29395837 S2152-2650(17)31516-1 29395837
32 Ailawadhi S Ogbonnaya A Murty S Cherepanov D Schroader BK Romanus D Farrelly E Chari A Duration of frontline therapy and impact on clinical outcomes in newly diagnosed multiple myeloma patients not receiving frontline stem cell transplant Cancer Med 2023 12 3 3145 3159 10.1002/cam4.5239 36151787 36151787 PMC9939178
33 Boxley C Fujimoto M Ratwani RM Fong A A text mining approach to categorize patient safety event reports by medication error type Sci Rep 2023 10 26 13 1 18354 10.1038/s41598-023-45152-w 37884577 10.1038/s41598-023-45152-w 37884577 PMC10603175
34 Han J Kim SY Lee J Lee WH Brain age prediction: a comparison between machine learning models using brain morphometric data Sensors (Basel) 2022 10 21 22 20 8077 10.3390/s22208077 36298428 s22208077 36298428 PMC9608785
35 Divyanshu D Aggarwal D Agarwal G Agarwal G Shahid M Heart attack predication using machince learning Journal of Propulsion Technology 2024 45 2 908 918
36 Yamana H Okada A Ono S Michihata N Jo T Yasunaga H Machine learning-based models for outpatient prescription of Kampo formulations: an analysis of a health insurance claims database J Epidemiol 2024 34 1 8 15 10.2188/jea.JE20220089 36642513 36642513 PMC10701253
37 Golas SB Shibahara T Agboola S Otaki H Sato J Nakae T Hisamitsu T Kojima G Felsted J Kakarmath S Kvedar J Jethwani K A machine learning model to predict the risk of 30-day readmissions in patients with heart failure: a retrospective analysis of electronic medical records data BMC Med Inform Decis Mak 2018 18 1 44 10.1186/s12911-018-0620-z 29929496 10.1186/s12911-018-0620-z 29929496 PMC6013959
38 Fitri A Sjahrir H Bachtiar A Ichwan M Fitri FI Rambe AS Predictive model of diabetic polyneuropathy severity based on vitamin D level Open Access Maced J Med Sci 2019 7 16 2626 2629 10.3889/oamjms.2019.454 31777620 OAMJMS-7-2626 31777620 PMC6876802
39 Hurria A Mohile S Gajra A Klepin H Muss H Chapman A Feng T Smith D Sun C De Glas Nienke Cohen HJ Katheria V Doan C Zavala L Levi A Akiba C Tew WP Validation of a prediction tool for chemotherapy toxicity in older adults with cancer J Clin Oncol 2016 34 20 2366 71 10.1200/JCO.2015.65.4327 27185838 JCO.2015.65.4327 27185838 PMC5321104
40 Nicholson C Beattie L Beattie M Razzaghi T Chen S A machine learning and clustering-based approach for county-level COVID-19 analysis PLoS One 2022 17 4 e0267558 10.1371/journal.pone.0267558 35476849 PONE-D-21-20512 35476849 PMC9045668
41 Marutho D Handaka SH Wijaya E Muljono The determination of cluster number at k-mean using elbow method and purity evaluation on headline news 2018 2018 International Seminar on Application for Technology of Information and Communication 2018 Sep 21-22 Semarang, Indonesia 10.1109/isemantic.2018.8549751
42 Humaira H Rasyidah R Determining the appropiate cluster number using elbow method for k-means algorithm 2024 01 24 Determining the appropiate cluster number using elbow method for k-means algorithm. Proceedings of the 2nd Workshop on Multidisciplinary and Applications (WMA) , January , Padang, Indonesia; 2020 2018 Padang, Indonesia 10.4108/eai.24-1-2018.2292388
43 Hassan BA Rashid TA Hamarashid HK A novel cluster detection of COVID-19 patients and medical disease conditions using improved evolutionary clustering algorithm star Comput Biol Med 2021 138 104866 10.1016/j.compbiomed.2021.104866 34598065 S0010-4825(21)00660-0 34598065 PMC8445768
44 Terpos E Mikhael J Hajek R Chari A Zweegman S Lee HC Mateos M Larocca A Ramasamy K Kaiser M Cook G Weisel KC Costello CL Elliott J Palumbo A Usmani SZ Management of patients with multiple myeloma beyond the clinical-trial setting: understanding the balance between efficacy, safety and tolerability, and quality of life Blood Cancer J 2021 11 2 40 10.1038/s41408-021-00432-4 33602913 10.1038/s41408-021-00432-4 33602913 PMC7891472
45 Handa H Ishida T Ozaki S Iida S Mori A Sakai C Applying artificial intelligence machine learning to discover factors that shorten duration of treatment in patients with multiple myeloma using real-world patient-level longitudinal data from Medical Data Vision database Blood. 2023 2023 142 10.1182/blood-2023-172849
46 Kleber M Ihorst G Terhorst M Koch B Deschler B Wäsch R Engelhardt M Comorbidity as a prognostic variable in multiple myeloma: comparative evaluation of common comorbidity scores and use of a novel MM-comorbidity score Blood Cancer J 2011 1 9 e35 10.1038/bcj.2011.34 22829196 bcj201134 22829196 PMC3255252
47 Cordas Dos Santos David M Saliba RM Patel R Bashir Q Saini N Hosing C Kebriaei P Khouri IF Nieto Y Popat U Ahmed H Lee HC Manasanch EE Patel KK Thomas SK Weber DM Orlowski RZ Champlin RE Qazilbash MH Age is a prognostic factor for the overall survival of patients with multiple myeloma undergoing upfront autologous hematopoietic stem cell transplantation Biol Blood Marrow Transplant 2020 26 6 1077 1083 10.1016/j.bbmt.2019.11.028 31786242 S1083-8791(19)30834-1 31786242
48 Palumbo A Gay F Cavallo F Di Raimondo F Larocca A Hardan I Nagler A Petrucci MT Hajek R Pezzatti S Delforge M Patriarca F Donato F Cerrato C Nozzoli C Yu Z Boccadifuoco L Caravita T Benevolo G Guglielmelli T Vincelli D Jacques C Dimopoulos MA Ciccone G Musto P Corradini P Cavo M Boccadoro M Continuous therapy versus fixed duration of therapy in patients with newly diagnosed multiple myeloma J Clin Oncol 2015 33 30 3459 66 10.1200/JCO.2014.60.2466 26282661 JCO.2014.60.2466 26282661