Construction and Validation of a Progression‐Free Survival Prediction Model for Newly Diagnosed Multiple Myeloma Patients With 1q21 Gain/Amplification Authors: Yuexuan Rao, Shiyou Li, Aoyang Yu, Wenjie Yu, Zhiyuan Yao, Huanxin Zhang, Zhenyu Li Published: 2026 ABSTRACT ABSTRACTObjectiveTo investigate the clinical characteristics of newly diagnosed multiple myeloma (NDMM) patients with 1q21 gain/amplification (1q21+), construct and validate a progression‐free survival (PFS) prediction model, and explore the clinical implications of model‐based risk stratification.MethodsWe retrospectively analyzed 186 newly diagnosed multiple myeloma patients with 1q21+ treated between 2018 and 2024. Patients were randomly assigned at a 7:3 ratio to a training cohort (n = 131) and an internal split‐sample validation cohort (n = 55). Independent prognostic factors for PFS were identified using multivariate Cox regression to construct a nomogram (the HLBP model). Model performance was evaluated through bootstrap resampling, time‐dependent receiver operating characteristic (ROC) curves, calibration plots, and Kaplan–Meier survival analysis. Furthermore, the model's predictive accuracy was compared with conventional staging systems and established prognostic models. Finally, risk‐stratified exploratory analyses were conducted to assess treatment‐associated outcomes across different risk groups.ResultsHigher hemoglobin was protective, whereas elevated lactate dehydrogenase, increased β2‐microglobulin, and TP53 deletion were adverse prognostic factors. The HLBP model yielded AUCs of 0.805, 0.885, and 0.847 at 6, 12, and 24 months in the training cohort, and 0.747, 0.827, and 0.712 in the internal split‐sample validation cohort. Patients stratified into high‐ and low‐risk groups showed significantly different PFS outcomes. Risk‐stratified analyses suggested heterogeneous prognostic patterns associated with induction treatment regimens and autologous stem cell transplantation across risk groups. Overall, the HLBP model demonstrated superior predictive performance compared with existing staging systems and previously reported prognostic models.ConclusionThe HLBP model showed promising preliminary performance for predicting PFS in patients with 1q21+ NDMM and may provide an exploratory framework for individualized risk stratification.Keywords: 1q21 gain/amplification, multiple myeloma, predictive model, progression‐free survival Introduction Multiple myeloma (MM) is a hematologic malignancy characterized by clonal plasma cell infiltration in the bone marrow, monoclonal immunoglobulin production, and multi‐organ damage. MM ranks as the second most common hematologic malignancy, accounting for approximately 10% of all hematologic cancers [1, 2]. The widespread application of proteasome inhibitors (PIs), immunomodulatory drugs (IMiDs), monoclonal antibodies, and the maturation of autologous stem cell transplantation (ASCT) have significantly improved patient outcomes. However, MM remains incurable, exhibiting substantial clinical heterogeneity in therapeutic response and survival outcomes [3, 4, 5]. Therefore, early identification of high‐risk patients, accurate prognostic stratification, and guidance for individualized therapy remain critical clinical and research challenges. Several risk stratification systems have been developed to optimize treatment decisions. The Durie‐Salmon (DS) system initially assessed disease stage based on tumor burden and organ involvement but lacked molecular‐level prognostic insight and has gradually been supplanted [6]. The International Staging System (ISS), based on serum β2‐microglobulin and albumin levels, provided a simple and widely adopted prognostic tool [7]. With evolving treatment modalities, ISS alone cannot fully capture the impact of molecular cytogenetic abnormalities. In 2015, the Revised ISS (R‐ISS) incorporated lactate dehydrogenase levels and high‐risk cytogenetic abnormalities, including t(4;14), t(14;16), and del(17p), enhancing prognostic discrimination [8]. More recently, the European Myeloma Network (EMN) proposed the second revision (R2‐ISS), adding 1q21+ as a high‐risk factor and further stratifying risk into four levels [9]. Despite these advances, existing systems have limitations. R‐ISS and R2‐ISS only partially incorporate high‐risk cytogenetic features and fail to fully capture MM's complex biology. Among cytogenetic abnormalities, 1q21+ is the most frequent, occurring in approximately 30%–40% of MM cases [10], and is associated with increased tumor burden, drug resistance, and poor prognosis [11, 12, 13]. However, clinical characteristics and prognostic significance in 1q21+ patients vary across studies, and large‐sample systematic analyses are lacking [14, 15, 16]. This study retrospectively analyzed 186 NDMM patients with 1q21+ treated at our center from 2018 to 2024. Using PFS as the endpoint, univariate and multivariate Cox regression analyses identified independent prognostic factors. These factors were integrated into a Nomogram (HLBP model) to quantify individual PFS risk. Internal validation assessed model discrimination and calibration, and patients were stratified into high‐ and low‐risk groups to explore prognostic differences in induction therapy and ASCT. This study aims to provide a practical, reliable tool for individualized risk stratification in 1q21+ NDMM patients. Subjects and Methods Subjects This retrospective study included 186 NDMM patients with 1q21+ diagnosed at the Department of Hematology, Affiliated Hospital of Xuzhou Medical University from August 2018 to August 2024. Patients with incomplete baseline or follow‐up data were excluded. Among the included patients, no significant differences in baseline clinical characteristics were observed compared with excluded patients. PFS was calculated from diagnosis to disease progression, death, or last follow‐up (August 2024), with patients without events censored at last follow‐up. The median follow‐up duration was 31.5 months (range, 4.2–59.0 months). Diagnosis adhered to the 2014 International Myeloma Working Group (IMWG) criteria [17]. Inclusion criteria: (i) newly diagnosed at our center; (ii) FISH‐confirmed 1q21+; (iii) complete clinical and follow‐up data. Exclusion criteria: (i) presence of concurrent active malignancies or severe comorbid conditions, including advanced cardiac, hepatic, or renal disease; (ii) incomplete baseline or follow‐up data. Data Collection Candidate predictors were prespecified based on clinical relevance and prior literature, and included demographic characteristics (age, years; sex), laboratory parameters [hemoglobin (Hb, g/L), platelets (×109/L), white blood cell count (WBC, ×109/L), lactate dehydrogenase (LDH, U/L), serum creatinine (μmol/L), albumin (g/L), calcium (mmol/L), β2‐microglobulin (β2‐MG, mg/mL)], bone marrow plasma cell percentage (%), M‐protein type, cytogenetic abnormalities [del(13q), IgH rearrangement, TP53 deletion, Rb deletion], and clinical staging parameters (DS, ISS, R‐ISS, R2‐ISS). Clinically significant factors were defined as variables with established prognostic relevance in multiple myeloma based on prior literature and statistical significance in univariate Cox regression analysis. PFS was defined from diagnosis to disease progression or death. FISH Cytogenetic Analysis Prior to treatment, bone marrow aspirates (10 mL) were subjected to FISH using locus‐specific probes targeting 1q21, IgH, 13q14, and TP53. For each probe, 300 interphase nuclei were evaluated under fluorescence microscopy. 1q21 gain/amplification was defined as the presence of three or more 1q21 signals in ≥ 10% of analyzed nuclei. In accordance with contemporary definitions, patients with three 1q21 copies were classified as having 1q21 gain, whereas those with four or more copies were classified as having 1q21 amplification. Copy‐number burden of 1q21 was further evaluated as a candidate prognostic factor. Treatment Regimens Treatments followed current guidelines [18, 19, 20, 21]. Among the 186 patients, induction therapy regimens were categorized into three primary groups: PI or IMiD‐based doublets (n = 66), including bortezomib plus dexamethasone (VD, n = 58) and lenalidomide plus dexamethasone (RD, n = 8). PI or IMiD‐based doublets (n = 66), including bortezomib plus dexamethasone (VD, n = 58) and lenalidomide plus dexamethasone (RD, n = 8). PI plus IMiD‐based regimens (n = 83), including bortezomib, lenalidomide, and dexamethasone (VRD, n = 66), ixazomib, lenalidomide, and dexamethasone (IRD, n = 4), ixazomib, pomalidomide, and dexamethasone (IPD, n = 3), and daratumumab, bortezomib, lenalidomide, and dexamethasone (DVRD, n = 7). PI plus IMiD‐based regimens (n = 83), including bortezomib, lenalidomide, and dexamethasone (VRD, n = 66), ixazomib, lenalidomide, and dexamethasone (IRD, n = 4), ixazomib, pomalidomide, and dexamethasone (IPD, n = 3), and daratumumab, bortezomib, lenalidomide, and dexamethasone (DVRD, n = 7). Other combination regimens (n = 37) primarily involving cytotoxic agents or monoclonal antibodies, such as bortezomib, cyclophosphamide, and dexamethasone (PCD, n = 10), melphalan‐based regimens (MVD, n = 8; MRP, n = 2), daratumumab plus VD (DVD, n = 14), and bortezomib, epirubicin, dexamethasone, and thalidomide (PADT, n = 6). Other combination regimens (n = 37) primarily involving cytotoxic agents or monoclonal antibodies, such as bortezomib, cyclophosphamide, and dexamethasone (PCD, n = 10), melphalan‐based regimens (MVD, n = 8; MRP, n = 2), daratumumab plus VD (DVD, n = 14), and bortezomib, epirubicin, dexamethasone, and thalidomide (PADT, n = 6). Additionally, 43 patients (23.1%) underwent consolidation with ASCT following induction therapy. Statistical Analysis This study was a retrospective prognostic prediction model development study conducted in accordance with the TRIPOD statement [22]. The completed TRIPOD checklist is provided in Table S1. Analyses were performed using R 4.4.3. Continuous variables were summarized as mean ± standard deviation (SD) for normally distributed data or median with interquartile range (IQR) for non‐normally distributed data, while categorical variables were presented as frequencies and percentages. Normality and homogeneity of variance were assessed using the Kolmogorov–Smirnov test. Comparisons between groups were conducted using the Student's t‐test or Mann–Whitney U test for continuous variables and the χ2 test or Fisher's exact test for categorical variables, as appropriate. Patients were randomly assigned at a 7:3 ratio to the training cohort and internal split‐sample validation cohort using simple random sampling in R with a fixed random seed of 123. The distributions of PFS events and censored observations in the training cohort and internal split‐sample validation cohort are summarized in Table S3. PFS was the outcome. Candidate predictors were selected based on clinical relevance and data availability at baseline. Univariate Cox proportional hazards regression analysis was first performed to screen potential predictors associated with PFS, and variables with p < 0.05 were entered into a multivariable Cox regression model to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). Clinically and statistically significant predictors were retained in the final multivariable model. Based on the final Cox model, a prognostic nomogram (HLBP model) was developed to estimate 6‐, 12‐, and 24‐month PFS probabilities. Internal validation was performed using bootstrap resampling with 1000 repetitions, with predictor selection fixed during resampling. Model discrimination was assessed using time‐dependent ROC curves, and calibration was evaluated by calibration plots. To further assess the robustness of variable selection, a LASSO Cox regression with 10‐fold cross‐validation was conducted as a sensitivity analysis (Figure S1), showing consistency with the Cox regression results. Patients were stratified into high‐ and low‐risk groups according to the total nomogram score. Kaplan–Meier survival curves were generated, and differences between risk groups were compared using the log‐rank test. Risk‐stratified subgroup analyses were further performed to explore the prognostic impact of induction treatment regimens and ASCT within HLBP‐defined high‐ and low‐risk groups. To further evaluate model performance, the predictive accuracy of the HLBP model was compared with conventional staging systems (ISS, R‐ISS, and R2‐ISS) and previously published prognostic models using time‐dependent ROC curve analysis. All statistical tests were two‐sided, and a p value < 0.05 was considered statistically significant. Results Construction of a PFS Prediction Model for NDMM Patients With 1q21+ Clinical Characteristics of the Cohort Among 186 patients, 104 (55.9%) were male and 82 (44.1%) female; median age was 66 years (57–72). The training cohort included 131 patients, the internal split‐sample validation cohort 55; baseline characteristics were comparable (all p > 0.05). Median PFS for the overall cohort was 20.0 months (95% CI 18.1–23.4; Table 1). TABLE 1Baseline of clinical characteristics of NDMM patients with 1q21+.CharacteristicsTotal (n = 186)Train (n = 131)Test (n = 55) p Sex0.124Male104 (55.9%)68 (51.9%)36 (65.5%)Female82 (44.1%)63 (48.1%)19 (34.5%)Type of M protein0.609IgG86 (46.2%)60 (45.8%)26 (47.3%)IgA38 (20.4%)24 (18.3%)14 (25.5%)IgM11 (5.9%)8 (6.1%)3 (5.5%)Light chain type, not excluded IgD39 (21.0%)31 (23.7%)8 (14.5%)Non‐secretory12 (6.5%)8 (6.1%)4 (7.3%)IgH rearrangement0.535Yes131 (70.4%)90 (68.7%)41 (74.5%)No55 (29.6%)41 (31.3%)14 (25.5%)Del(13q)0.999Yes120 (64.5%)85 (64.9%)35 (63.6%)No66 (35.5%)46 (35.1%)20 (36.4%)TP53 deletion0.677Yes53 (28.5%)39 (29.8%)14 (25.5%)No133 (71.5%)92 (70.2%)41 (74.5%)RB deletion0.693Yes99 (53.2%)68 (51.9%)31 (56.4%)No87 (46.8%)63 (48.1%)24 (43.6%)Chromosome 1q copy number0.443390 (48.4%)61 (46.6%)29 (52.7%)≥ 496 (51.6%)70 (53.4%)26 (47.3%)Age/years66.00 (57.00, 72.00)67.00 (57.00, 72.00)65.00 (57.00, 70.00)0.364BMPC/%48.00 (36.00, 65.00)50.00 (37.00, 65.00)47.00 (33.00, 59.00)0.918HGB/(g/L)80.00 (64.50, 103.50)83.00 (63.00, 103.00)79.00 (65.00, 105.00)0.776WBC/(×109/L)4.65 (3.60, 6.23)4.60 (3.70, 6.30)4.80 (3.40, 6.20)0.466PLT/(×109/L)159.00 (109.00, 212.00)158.00 (105.00, 216.00)160.00 (116.00, 198.00)0.884Scr/(μmol/L)84.50 (58.00, 156.50)92.00 (58.50, 175.00)73.00 (55.50, 113.00)0.127LDH/(U/L)239.00 (163.50, 331.50)246.00 (175.50, 367.50)231.00 (171.00, 307.50)0.602β2‐MG/(mg/mL)4.92 (3.30, 8.05)5.22 (3.45, 8.25)4.01 (2.83, 6.51)0.0693Alb/(g/dL)34.60 (29.13, 40.43)34.70 (29.90, 40.00)34.50 (28.20, 41.50)0.923Calcium/(mg/dL)9.16 (8.52, 10.21)9.24 (8.52, 10.52)9.04 (8.64, 9.56)0.295Number of 1q21+ in FISH43.00 (22.50, 82.00)43.00 (21.00, 84.00)43.00 (31.00, 79.00)0.765 Note: Continuous variables are presented as median (Q1, Q3); categorical variables are presented as n (%). Baseline of clinical characteristics of NDMM patients with 1q21+. Note: Continuous variables are presented as median (Q1, Q3); categorical variables are presented as n (%). Prognostic Model Construction Univariate Cox regression in the training cohort identified significant PFS predictors, which were then analyzed with multivariate Cox regression. Higher Hb was an independent protective factor (HR = 0.978, 95% CI 0.969–0.988, p = 0.001), whereas elevated LDH, increased β2‐MG, and TP53 deletion were independent adverse factors (HR = 1.105, 95% CI 1.018–1.198, p = 0.018; HR = 1.051, 95% CI 1.002–1.102, p = 0.039; HR = 1.655, 95% CI 1.038–2.638, p = 0.034). The proportional hazards (PH) assumption was tested using Schoenfeld residuals and showed no evidence of violation for any covariates (Table 2). TABLE 2Univariate and multivariate analyses of factors influencing PFS in NDMM patients with 1q21+.FactorsUnivariateMultivariateHR95% CI p HR95% CI p Sex (male vs. female)0.1790.777–1.7890.438Type of M protein (IgG vs. others)0.9780.839–1.1400.776IgH rearrangement (yes vs. no)1.0930.676–1.7660.718Del(13q) (yes vs. no)1.2170.775–1.9100.394TP53 deletion (yes vs. no)1.9361.251–2.9950.0031.6551.038–2.6380.034RB deletion (yes vs. no)1.2190.799–1.8600.358Chromosome 1q copy number (≥ 4 vs. 3)1.1860.881–1.5950.260Age, years1.0170.993–1.0420.169BMPC, %1.0030.992–1.0130.600HGB (g/L)0.9750.967–0.984< 0.0010.9780.969 ~ 0.9880.001WBC (×109/L)1.0260.944–1.1150.546PLT (×109/L)1.0000.998–1.0030.775Scr (μmol/L)1.0511.026–1.089< 0.0011.0031.000–1.0100.344LDH (U/L)1.0371.013–1.0520.0021.1051.018–1.1980.018β2‐MG (mg/mL)1.1201.051–1.201< 0.0011.0511.002–1.1020.039Alb (g/dL)1.0030.976–1.0300.851Calcium (mg/dL)1.0240.999–1.0490.064Number of 1q21+ in FISH1.0010.997–1.0040.653 Univariate and multivariate analyses of factors influencing PFS in NDMM patients with 1q21+. These were integrated into the Nomogram (HLBP model) predicting 6‐, 12‐, and 24‐month PFS (Figure 1). Nomogram for PFS in NDMM patients with 1q21+. LASSO Cox regression with 10‐fold cross‐validation identified Hb, LDH, β2‐MG, and TP53 deletion as stable predictors. The results were highly consistent with those obtained from univariate and multivariate Cox regression, supporting the robustness of variable selection (Figure S1). The events‐per‐variable (EPV) in the training cohort was 22.5, exceeding the recommended threshold of 10. Model Validation In the training cohort, the model achieved AUCs of 0.805 (95% CI: 0.706–0.904), 0.885 (95% CI: 0.820–0.949), and 0.847 (95% CI: 0.769–0.925) for 6‐, 12‐, and 24‐month PFS, respectively. In the internal split‐sample validation cohort, the corresponding AUCs were 0.747 (95% CI: 0.536–0.958), 0.827 (95% CI: 0.700–0.954), and 0.712 (95% CI: 0.564–0.863), indicating good discriminative ability and stability across different time points (Figure 2). ROC curves for 6‐month (A), 12‐month (B), and 24‐month (C) PFS prediction in the training set and validation set of NDMM patients with 1q21+. Internal validation using bootstrap resampling demonstrated that the calibration curves for 6‐, 12‐, and 24‐month PFS in the training cohort closely matched the observed outcomes. Similar results were observed in the internal split‐sample validation cohort, confirming the Nomogram's predictive accuracy and robustness (Figure 3). Calibration curves for 6‐month PFS prediction in the training set (A) and validation set (B), for 12‐month PFS prediction in the training set (C) and validation set (D), and for 24‐month PFS prediction in the training set (E) and validation set (F) of NDMM patients with 1q21+. In addition, quantitative calibration assessment using calibration slope and intercept is provided in Table S2. Overall, these metrics further support satisfactory agreement between predicted and observed survival probabilities across all time points. Risk Stratification Based Based on risk scores calculated from the Cox model, patients were stratified into high‐ and low‐risk groups using the median risk score of the training cohort as the cutoff. Kaplan–Meier analysis revealed that high‐risk patients had significantly shorter PFS than low‐risk patients in the training cohort (median PFS: 8.9 vs. 29.8 months, p < 0.001), with consistent findings in the internal split‐sample validation cohort (median PFS: 7.9 vs. 23.1 months, p = 0.018; Figure 4). Prognostic stratification and risk‐adapted treatment outcomes in NDMM patients with 1q21+. (A) Training Set: Kaplan Meier PFS by HLBP Risk Group: Kaplan–Meier curve for PFS stratified by HLBP‐defined high‐ and low‐risk groups. Low‐risk patients exhibited significantly longer PFS compared with high‐risk patients (p < 0.0001). (B) Validation Set: Kaplan–Meier PFS by HLBP Risk Group: Kaplan–Meier curve for PFS stratified by HLBP risk groups, confirming the prognostic discrimination of the model: Low‐risk patients had superior PFS relative to high‐risk patients (P = 0.018). (C–F) Induction Regimen Outcomes by Risk Group and Cohort: (C) High‐risk (Training Set): PI+IMiD showed a trend toward longer PFS (p = 0.083). (D) Low‐risk (Training Set): PI+IMiD was associated with significantly prolonged PFS (p = 0.0025). (E) High‐risk (Validation Set): PI+IMiD correlated with improved PFS (p = 0.017). (F) Low‐risk (Validation Set): PI+IMiD consistently conferred a significant PFS benefit (p = 0.0025). (G–J) ASCT Outcomes by Risk Group and Cohort: (G) High‐risk (Training Set): ASCT was associated with markedly longer PFS (p = 0.002). (H) Low‐risk (Training Set): ASCT correlated with significantly prolonged PFS (p = 0.019). (I) High risk (Validation Set): ASCT remained significantly linked to improved PFS (p = 0.016). (J) Low‐risk (Validation Set): ASCT conferred a significant PFS advantage (p = 0.03). To further explore the clinical implications of the HLBP‐based risk stratification, we conducted exploratory subgroup analyses to evaluate the prognostic impact of induction treatment regimens and ASCT within each risk group. Regarding induction therapy, in the training cohort, no statistically significant differences in PFS were observed among the three induction regimens within the high‐risk group (median PFS: 6.5 vs. 15.0 vs. 8.8 months, p = 0.083). In contrast, within the low‐risk group, patients receiving PI + IMiD‐containing induction therapy demonstrated significantly prolonged PFS compared with those treated with PI‐ or IMiD‐based doublet regimens or other regimens (median PFS: 37.4 vs. 23.1 vs. 24.2 months, p = 0.028). In the internal split‐sample validation cohort, subgroup analyses according to induction therapy further supported a risk‐adapted treatment effect. Within the high‐risk group, patients treated with PI + IMiD‐containing induction regimens experienced significantly longer PFS than those receiving PI‐ or IMiD‐based doublet regimens or other regimens (median PFS: 19.3 vs. 5.4 vs. 10.7 months, p = 0.017). Similarly, among low‐risk patients, PI + IMiD–based induction therapy was associated with superior PFS compared with both PI‐ or IMiD‐based doublet regimens and other regimens (median PFS: 30.4 vs. 18.9 vs. 23.4 months, p = 0.025). With respect to ASCT, subgroup analyses in the training cohort demonstrated that patients who underwent ASCT had significantly prolonged PFS compared with those who did not, in both the high‐risk and low‐risk groups defined by the HLBP model (median PFS: 41.4 vs. 7.9 months, p = 0.002; and 30.1 vs. 27.1 months, p = 0.019). In the internal split‐sample validation cohort, ASCT remained significantly associated with improved PFS in both risk groups (median PFS: 32.4 vs. 10.4 months, p = 0.016 for the high‐risk group; and 37.5 vs. 24.3 months, p = 0.030 for the low‐risk group) (Figure 4). Comparison With Conventional Staging Systems and Published Prognostic Models We further compared the predictive performance of the HLBP model with conventional clinical staging systems (ISS, R‐ISS, and R2‐ISS) and previously published prognostic models [23, 24] in the 1q21+ NDMM subgroup using ROC curve analysis. For 6‐month PFS prediction, the HLBP model achieved an AUC of 0.776 (95% CI: 0.679–0.873), compared with 0.666 (95% CI: 0.560–0.773) for ISS, 0.368 (95% CI: 0.263–0.494) for R‐ISS, 0.706 (95% CI: 0.610–0.797) for R2‐ISS, 0.631 (95% CI: 0.526–0.735) for Liu et al., and 0.726 (95% CI: 0.642–0.811) for Xiong et al. For 12‐month PFS prediction, the HLBP model yielded an AUC of 0.857 (95% CI: 0.795–0.919), whereas ISS, R‐ISS, and R2‐ISS produced AUCs of 0.676 (95% CI: 0.604–0.752), 0.358 (95% CI: 0.250–0.485), and 0.734 (95% CI: 0.661–0.808), respectively. Liu et al. and Xiong et al. models had AUCs of 0.646 (95% CI: 0.557–0.735) and 0.776 (95% CI: 0.709–0.844). For 24‐month PFS prediction, the HLBP model AUC was 0.810 (95% CI: 0.728–0.871), compared with 0.685 (95% CI: 0.604–0.767) for ISS, 0.610 (95% CI: 0.527–0.694) for R‐ISS, 0.741 (95% CI: 0.662–0.820) for R2‐ISS, 0.682 (95% CI: 0.594–0.769) for Liu et al. and 0.741 (95% CI: 0.661–0.819) for Xiong et al. Across all evaluated time points (6, 12, and 24 months), the HLBP model consistently exhibited higher AUC values than conventional staging systems and previously published models, indicating superior discriminative ability and accuracy in predicting PFS for patients with 1q21+ NDMM (Figure 5). Comparison of ROC curves for the HLBP model, staging systems, and published prognostic models at 6 months (A), 12 months (B), and 24 months (C). Discussion MM is an incurable hematologic malignancy characterized by marked heterogeneity in clinical presentation, treatment response, and prognosis [3, 15, 25]. Therefore, the rapid and accurate identification of high‐risk patients and the implementation of individualized therapy are crucial for maximizing treatment efficacy and improving patient outcomes. With the widespread clinical application of FISH, significant progress has been made in the cytogenetic study of MM [26]. Numerous studies have confirmed that 1q21+ is an adverse prognostic factor in MM patients [27, 28, 29, 30]. However, most previous investigations have focused solely on its prognostic implications, lacking individualized predictive tools for this specific patient population. The present study aimed to analyze the clinical characteristics and prognostic heterogeneity of patients with 1q21+ NDMM, providing a robust framework for optimized risk stratification and generating hypotheses for future risk‐adapted therapeutic research. Based on 186 patients with 1q21+ NDMM at initial diagnosis, we developed and validated a PFS prediction model, termed the HLBP model. Using multivariate Cox regression and a nomogram, Hb, LDH, β2‐MG, and TP53 deletion were incorporated into the risk score calculation, enabling visualized individualized risk stratification and survival prediction. The results demonstrated that the model exhibited good discriminative ability and robustness in both the training and internal split‐sample validation cohort and significantly outperformed conventional clinical staging systems (ISS, R‐ISS, R2‐ISS), providing a potential decision‐making tool for the clinical management of 1q21+ NDMM patients. Regarding independent prognostic factors, decreased Hb levels reflect impaired bone marrow hematopoietic function and serve as a sensitive indicator of disease burden. Anemia is not only associated with abnormal plasma cell infiltration in the bone marrow microenvironment but may also affect cellular sensitivity to therapy through tissue hypoxia and metabolic dysregulation, thereby increasing the risk of disease progression [31]. This observation is consistent with prior reports [32, 33], highlighting the pivotal role of anemia in predicting survival outcomes in 1q21+ NDMM patients. Elevated LDH generally indicates active tumor cell proliferation and enhanced tissue‐destructive potential and may be closely linked to metabolic reprogramming and microenvironmental alterations that promote disease progression [34]. Studies in MM have demonstrated that increased LDH levels are associated with shorter survival and poor treatment response, emphasizing its significance in disease monitoring and prognostic assessment [23, 35, 36]. β2‐MG, as an integrated marker reflecting tumor burden and renal function, is elevated in high‐risk 1q21+ subtypes, indicating enhanced plasma cell proliferation and secretory activity, thereby confirming its value in PFS risk assessment [37]. Furthermore, TP53 deletion, a classic high‐risk genetic abnormality, impairs cell cycle regulation and DNA damage repair, facilitating tumor cell evasion from therapeutic intervention and markedly shortening PFS [38, 39]. Our findings are highly consistent with previous reports, underscoring the central role of these multidimensional factors in prognostic evaluation of high‐risk 1q21+ NDMM patients. Compared with conventional staging systems (ISS, R‐ISS, R2‐ISS) and previously published prognostic models (Liu et al. [23], Xiong et al. [24]), the HLBP model offers several advantages. By integrating clinical, laboratory, and genetic information, the model comprehensively reflects disease burden and biological characteristics. The visualized nomogram enables quantitative estimation of PFS risk, facilitating more refined risk stratification and follow‐up planning. In comparative analyses, the HLBP model consistently demonstrated superior discriminative ability and predictive performance in 1q21+ NDMM patients. These results support the value of integrating high‐risk cytogenetic abnormalities into prognostic assessment frameworks. Beyond prognostic model development, we further explored treatment‐associated outcomes within different risk strata defined by the HLBP model. By applying the proposed model to classify patients into high‐ and low‐risk categories, we observed distinct prognostic patterns with respect to induction therapy regimens and ASCT, suggesting potential heterogeneity in treatment outcomes across risk groups. With regard to induction therapy, PI plus IMiD–based doublet regimens were consistently associated with prolonged PFS in low‐risk patients across both the training and internal split‐sample validation cohort, whereas the benefit of intensified induction appeared more heterogeneous among high‐risk patients. These findings align with current evidence supporting the use of combination induction regimens incorporating novel agents as standard therapy in transplant‐eligible NDMM. For example, guidelines and clinical trials have established that triplet induction regimens such as VRd (bortezomib, lenalidomide, dexamethasone) generally achieve higher response rates and deeper remissions compared with doublet regimens, which may translate into improved outcomes after consolidation therapy including ASCT [40, 41]. Our observations suggest that patients classified as low risk by the HLBP model may derive greater prognostic benefit from combination induction strategies, while high‐risk patients remain vulnerable to early progression despite intensified induction therapy, reflecting the dominant impact of adverse disease biology and highlighting the need for novel or more aggressive approaches in this subgroup. Similarly, subgroup analyses demonstrated that ASCT was associated with improved PFS in both high‐ and low‐risk groups, although the magnitude of benefit varied across cohorts. These results are consistent with prior reports that early ASCT following induction therapy contributes to superior disease control compared with non‐transplant approaches in contemporary treatment contexts; for example, randomized trials have shown that ASCT combined with induction regimens can significantly prolong PFS even among high‐risk cytogenetic subgroups. Additionally, retrospective analyses in 1q21+ NDMM patients have indicated that ASCT confers significantly longer PFS and OS compared with non‐transplant therapy, supporting the notion that transplant remains an important component of therapy for patients with adverse cytogenetic features [42, 43]. While these exploratory findings reinforce the potential role of ASCT across risk strata, they also underscore the necessity for prospective, risk‐adapted clinical studies to clarify optimal integration of transplant and induction strategies. Importantly, these treatment‐related analyses were exploratory and conducted after model development, and treatment variables were not incorporated into the prognostic model itself. Therefore, the observed associations should not be interpreted as evidence of causal treatment effects or definitive therapeutic recommendations. Rather, they underscore the potential of the HLBP model to serve as a platform for generating hypotheses regarding risk‐adapted treatment strategies in high‐risk cytogenetic subgroups. Nonetheless, several limitations should be acknowledged. First, this was a single‐center retrospective study, and the cohort size was relatively limited. Although bootstrap resampling and an internal split‐sample validation cohort were used to evaluate model performance, independent external validation was not available. Therefore, the predictive performance of the HLBP model should be interpreted as preliminary and requires further confirmation in larger, multicenter cohorts. In addition, because the internal split‐sample validation cohort was small, the estimates of time‐dependent AUCs and subgroup analyses may be less stable. Second, the model‐building strategy also has inherent limitations. Although candidate predictors were selected based on clinical relevance and baseline availability, the final multivariable model was constructed after univariate Cox regression screening, which may introduce instability in variable selection and may miss clinically relevant factors that did not reach statistical significance in univariate analysis. LASSO Cox regression was therefore performed as a sensitivity analysis and showed generally consistent results, but it did not replace the original modeling strategy. Moreover, patients received heterogeneous treatments during the study period according to contemporaneous guidelines, clinical condition, transplant eligibility, and treatment availability. Because the HLBP model was intended as a pretreatment prognostic model based on variables available at initial diagnosis, treatment‐related variables were not incorporated into the final model. Nevertheless, treatment heterogeneity and potential treatment allocation bias may still have influenced PFS, and treatment‐related subgroup analyses should be regarded as exploratory. Future prospective studies with larger sample sizes, independent external validation, standardized treatment information, and additional prognostic factors such as MRD status, functional imaging biomarkers, and immune microenvironment‐related molecular imaging markers may further improve and validate the model [44, 45]. In summary, the HLBP model integrates baseline clinical, laboratory, and high‐risk genetic indicators to provide an individualized and visualized approach for predicting PFS in patients with 1q21+ NDMM. Compared with conventional staging systems, it showed promising prognostic performance and may serve as a useful exploratory framework for risk stratification. Further large‐scale, multicenter prospective studies are warranted to externally validate and optimize this model. Conclusion In this study, we developed and internally validated the HLBP nomogram, a visualized prognostic model integrating baseline clinical, laboratory, and high‐risk genetic features to predict PFS in NDMM patients with 1q21+. The model showed promising discriminative ability and calibration and outperformed conventional staging systems and previously published prognostic models in this cohort, suggesting its potential value for individualized risk stratification. Exploratory risk‐adapted analyses suggested that low‐risk patients tended to have more favorable PFS outcomes with combination induction regimens, whereas high‐risk patients remained prone to early progression despite intensified therapy, indicating the need for further therapeutic optimization in this subgroup. Overall, the HLBP model may provide a useful exploratory framework for prognostic stratification and future risk‐adapted research. Author Contributions Yuexuan Rao: conceptualization, methodology, data curation, formal analysis, validation, supervision, writing – original draft. Shiyou Li: conceptualization, formal analysis, methodology, validation. Aoyang Yu: data curation, investigation. Wenjie Yu: investigation, validation. Zhiyuan Yao: methodology, software. Huanxin Zhang: resources, visualization. Zhenyu Li: funding acquisition, project administration, supervision, writing – review and editing. Funding This work was supported by the Construction Project of High Level Hospital of Jiangsu Province (GSPJS202414, GSPJS202417, and GSPJS202420). Ethics Statement Approved by the Ethics Committee of the Affiliated Hospital of Xuzhou Medical University (XYFY2024‐KL002‐01), this study adhered to the Declaration of Helsinki. Patient data were de‐identified; informed consent was waived due to the retrospective design. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: LASSO Cox regression for variable selection. (A) Partial likelihood deviance plot with optimal λ selected by minimum and 1‐SE criteria. Numbers at the top indicate the number of non‐zero coefficients.(B) Coefficient profiles of all variables across log(λ) values. Table S1: TRIPOD checklist. Table S2: Calibration characteristics of the HLBP model in the training and validation cohorts at 6‐, 12‐, and 24‐month follow‐up. Table S3: Distribution of PFS events and censored observations in the training and internal split‐sample validation cohorts. Acknowledgments We sincerely thank all patients and their families for their participation in this study. We also acknowledge the contributions of the clinical and laboratory staff at the Department of Hematology, Affiliated Hospital of Xuzhou Medical University, for their support in data collection and patient management. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Figure S1: LASSO Cox regression for variable selection. (A) Partial likelihood deviance plot with optimal λ selected by minimum and 1‐SE criteria. Numbers at the top indicate the number of non‐zero coefficients.(B) Coefficient profiles of all variables across log(λ) values. Table S1: TRIPOD checklist. Table S2: Calibration characteristics of the HLBP model in the training and validation cohorts at 6‐, 12‐, and 24‐month follow‐up. Table S3: Distribution of PFS events and censored observations in the training and internal split‐sample validation cohorts. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.