Summary
Multiple myeloma (MM) is a malignancy with varying survival outcomes and drivers of disease progression. Existing MM staging tools were developed using data from newly diagnosed patients. As patient characteristics and disease‐related factors change between diagnosis and the initiation of second‐line (2L) treatment, an unmet need exists for a tool that can evaluate risk of death at first relapse. We have developed a risk stratification algorithm (RSA) using data from patients with MM who were at 2L. Hazard ratios for independent predictors of overall survival (OS) were derived from a Cox models, and individual patient scores were calculated for total risk. K‐adaptive partitioning for survival was used to stratify patients into groups based on their scores. Relative risk doubled with ascending risk group; median OSs for patients in group 1 (lowest risk)–4 (highest risk) were 61·6, 29·6, 14·2 and 5·9 months, respectively. Differences in OS between risk groups were significant. Similar stratification was observed when the RSA was applied to an external validation data set. In conclusion, we have developed a validated RSA that can quantify total risk, frailty risk and disease aggressiveness risk, and stratify patients with MM at 2L into groups with profoundly different survival expectations.
Keywords: algorithm, multiple myeloma, overall survival, relapsed, risk stratification
Multiple myeloma (MM) is a malignancy of plasma cells that accounts for approximately 10% of all haematological cancers (Russell & Rajkumar, 2011; Vu et al, 2015; Moreau et al, 2017). MM is associated with considerable heterogeneity in terms of patient characteristics, drivers of disease progression, prognosis and treatment response (Moreau et al, 2017). This is exemplified by the variations observed in life expectancy. The median overall survival (OS) for MM patients is in the range of 5–7 years; however, data suggest that 13% of patients die within 24 months of diagnosis (Fonseca et al, 2017) and findings from a European chart review suggested that 36% of patients with MM do not reach the second line of treatment (Raab et al, 2016). A tool that could accurately evaluate overall risk of death and identify the drivers of disease severity would be of considerable value, especially at first relapse when patient and disease‐related information is available.
The International Staging System (ISS; Greipp et al, 2005) and the Revised International Staging System (R‐ISS; Palumbo et al, 2015a) are used to define disease progression in patients with MM. Clinical trial data from over 10 000 patients newly diagnosed with MM were used to develop the ISS, in which patients were stratified into one of three risk groups based on serum albumin and serum β‐2 microglobulin (Sβ2M) levels (Greipp et al, 2005). In the following years, studies highlighted the link between the presence of specific cytogenetic abnormalities (CA; del[17p], t[4;14] and t[14;16]), increased lactate dehydrogenase (LDH) levels and reduced OS (Fonseca et al, 2009; Terpos et al, 2010; Ross et al, 2012). To reflect this, the International Myeloma Working Group (IMWG) published a position paper in which three risk groups were defined based on the ISS criteria and CA at diagnosis (Chng et al, 2014); a revised ISS (R‐ISS) was published in 2015, in which CA and LDH levels were taken into account in addition to Sβ2M and serum albumin levels (Palumbo et al, 2015a). The ISS and R‐ISS have been widely adopted in clinic practice and clinical trials (Lonial et al, 2015; Stewart et al, 2015; Moreau et al, 2016; Dimopoulos et al, 2016a). It should be noted, however, that both tools were developed to stage patients using only information available at diagnosis, and neither tool has been fully validated in the relapse setting or takes patient frailty into consideration.
There is a need for physicians to assess patient prognosis systematically using all available evidence when making treatment decisions at the initiation of second‐line (2L) treatment. This has become particularly important in the past decade owing to the dramatic rise in the number of agents that have been approved for patients with relapsed or refractory MM (RRMM) (carfilzomib, Amgen Europe, Dublin, Ireland; elotuzumab, Bristol‐Myers Squibb S.r.l., Anagni, Italy; pomalidomide, Celgene Distribution B.V., Utrecht, Netherlands; daratumumab and bortezomib, Janssen Biologics B.V., Leiden, The Netherlands; ixazomib, Takeda GmbH, Singen, Germany). In the absence of a more suitable algorithm, the R‐ISS and ISS are frequently used to stratify patients with MM at relapse in clinical trials (Lonial et al, 2015; Moreau et al, 2016; Palumbo et al, 2016; Dimopoulos et al, 2016a; Dimopoulos et al, 2016b). In practice, however, the value of considering R‐ISS or ISS stage when making treatment decisions following first relapse is unclear. It is widely accepted that when making treatment decisions at 2L, physicians need to consider a range of patient and disease characteristics and experiences during first‐line (1L) treatment; therefore, any tool developed to stratify patients at this disease stage should reflect this thought process (Bird et al, 2011; Sonneveld & Broijl, 2016; Moreau et al, 2017). A recent analysis of real‐world data from patients with MM revealed that patient characteristics and predictors of OS differed between the initiation of 1L and 2L treatments (Hájek et al, 2016). Moreover, the predictors of OS at 2L included parameters that were based on information that was revealed during 1L therapy (Hájek et al, 2016). Herein, we describe the development and validation of a novel risk stratification algorithm (RSA) that can systematically assess the risk of death as well as the factors driving risk, such as frailty, whilst also reflecting the physician’s thought process when defining patient prognosis at first relapse. In addition, we present an interpretation of the results obtained from implementation of the RSA.
Materials and methods
Data source
Model development cohort
The Czech Registry of Monoclonal Gammopathies (RMG) is one of the largest registries of patients with MM and monoclonal gammopathies of unknown significance. It contains detailed data on a large number of patient characteristics and disease‐related parameters recorded at diagnosis and at first relapse. In addition, the RMG has mature OS data and is representative of the national and international patient populations. For these reasons, the RMG was selected for development of the RSA (Radocha et al, 2015). Data were collected from all 20 of the Czech centres that actively treat patients with MM; these centres cover approximately 80% of all patients with MM in the Czech Republic (Radocha et al, 2015). Data collection began in May 2007, and patients were observed from diagnosis until either death, loss to follow‐up or 26 April 2016.
Eligibility criteria
Because there is an unmet need to re‐define patients’ survival expectations at first relapse, the current analysis only included individuals aged 18 years or older who received at least one dose of anti‐myeloma treatment following first relapse. Patients who died or were lost to follow‐up before initiating 2L treatment were excluded from the analysis.
Development strategy
Step 1: identification of candidate predictors of overall survival at initiation of second line
Candidate predictors of OS were identified based on literature analyses, and the findings from a conceptual model of MM progression that was defined by a Delphi process involving leading experts in MM (Fig 1) (Gonzalez‐McQuire et al, 2019).
Figure 1.
Selection of predictors for the risk stratification algorithm. 1L, first line; 2L, second line; ECOG PS, Eastern Cooperative Oncology Group performance status; LDH, lactate dehydrogenase; RSA, risk stratification algorithm; Sβ2M, serum β‐2 microglobulin; SCT, stem cell transplantation.
Step 2: defining parameters to reflect clinical relevance
Predictors were classified into one of four types based on how the parameters were associated with risk of death: categorical in nature (e.g. presence of extramedullary disease); continuous with established clinical cut‐offs (e.g. LDH level), which were treated as categorical predictors; continuous up to a threshold point beyond which the risk of death remained constant (e.g. thrombocyte count); and fully continuous (e.g. age).
Time to initiation of 2L treatment was treated as a categorical, dichotomized variable (>24 months vs. ≤24 months). A boundary of 2 years may be regarded as too high and applicable only to patients with a good prognosis. However, time to initiation of 2L treatment is strongly influenced by treatment decisions, and a short time to initiation of 2L treatment may not correspond to a poor prognosis. Therefore, a high cut‐off level (24 months) was selected to allow the positive impact on prognosis associated with a longer time to initiation of 2L treatment to be captured. Other predictors, such as refractory status, occurrence of grade 3 or 4 adverse events (AE) and presence of hypercalcaemia or extramedullary disease, are likely to capture the increased risk of death associated with short time to initiation of 2L treatment.
Multiple imputation was used to estimate missing values for all variables except presence of CA (Eisemann et al, 2011). A category of ‘not available’ was included for CA to reflect clinical practice, in which CA data are not available for large numbers of patients, some of whom will have high‐risk CA, and risk CA (Groenwold et al, 2012).
Step 3: selecting the significant predictors of overall survival at the initiation of second line
To select the final predictors, Pearson’s correlations test identified and excluded the factors that correlated strongly with other parameters. The remaining predictors entered multivariable Cox regression models, in which OS from initiation of 2L treatment was the dependent variable. A backward selection was performed using Akaike’s information criterion (AIC).
One of the known disadvantages of predictor selection based on statistical analyses, such as backward selection, is the exclusion of known risk factors (Austin & Tu, 2004). Therefore, to ensure that the RSA included key parameters that are known to define disease severity and to reflect a physician’s thought process when selecting a 2L treatment, it was decided that the following predictors would be included in the model regardless of statistical significance: age, Sβ2M level, LDH level, Eastern Cooperative Oncology Group (ECOG) performance status (PS) (at initiation of 2L in all cases), time to initiation of 2L, refractory status and cytogenetic risk at diagnosis. High cytogenetic risk was defined as presence of del (17p) and/or translocation t (4;14) and/or t (14;16).
Step 4: calculating patient‐specific risk scores
The overall score for each patient was obtained by multiplying the hazard ratios (HRs) corresponding to the patient‐specific value for each parameter. Multiplying the risk scores, means that each score represents a relative increase in risk of death compared with a patient who has the lowest/lower risk for all predictors.
Step 5: defining risk stratification using patient‐specific risk score
The methodology required us to assume that four risk groups would be defined, and no group would contain less than 10% of the analysis population (unpublished observations). A K‐adaptive partitioning for survival (KAPS) algorithm was used to analyse the survival data and risk scores to determine whether patients could be partitioned into groups with statistically significant differences in survival expectations. KAPS was used in the development of the R‐ISS and has been applied to the Surveillance, Epidemiology and End Results Program run by the National Institutes of Health (Eo et al, 2014; Palumbo et al, 2015a).
Steps 6: defining the drivers of risk
To develop an understanding of the factors that drive risk in patients with MM at the initiation of 2L, the predictors were separated into those associated with patient frailty and those related to disease aggressiveness. While several definitions of frailty exist (Ethun et al, 2017), for this analysis frailty was defined as a measure of ECOG PS and age. Through multiplying the HR associated with the predictors in each subset, patient‐specific risk scores were obtained for frailty and disease aggressiveness.
Step 7: understanding the influence of second‐line treatment on prognosis
Treatment received during 2L was not a candidate predictor because the RSA aims to provide physicians with an insight into the appropriate management of patients at first relapse. In addition, adjusting for future variables can result in biased results (Cole et al, 2010). It was assumed that all patients with MM in the Czech RMG would receive the most effective, suitable treatment available. However, second‐line treatment patterns by risk group and the impact of treatment used in 2L on survival expectations were analysed descriptively.
Validation
The RSA was tested in an external data set containing information on patients with symptomatic MM from France, Germany and the UK who initiated 2L therapy in 2013. In total, 180 physicians were involved in a specially designed retrospective chart review study. The data were derived from patient medical chart audits based on a questionnaire sent to physicians. Regional and hospital type quotas were applied to ensure a representative sample. Further information on hospital type and region can be found in Table SI. Relevant data were abstracted onto a study‐specific case report form during the second and third quarters of 2017. The RSA was applied to the validation data set in the same manner as it was applied to the Czech RMG. To evaluate the performance of the RSA in the external data set, Harrell’s overall concordance index (C‐index) was calculated (which indicates the proportion of pairs of subjects whose observed and predicted outcomes agree).
Results
Patient baseline characteristics
Development cohort
Data from 1418 patients in the Czech RMG were included in the analysis. The median duration of follow‐up was 27·56 months (95% confidence interval [CI]: 25·05–30·09) (taking into account censoring owing to mortality), and the mean follow‐up was 38·34 months (95% CI: 36·06–40·61). The baseline characteristics of this cohort reflect the heterogeneous nature of the global real‐world MM population (Table 1). At the initiation of 2L treatment, over one‐fifth (21%) of patients were over 75 years of age, approximately three‐quarters (74%) had an ECOG PS of 0 or 1, and 70% had progressed within 24 months of initiating 1L treatment. Approximately one‐third (35%) of patients received proteasome inhibitors at 1L, 44% were treated with immunomodulators, and 6% received a combination containing both classes of agent.
Table 1.
Characteristics of patients in development and validation cohorts.
| Characteristic* |
Czech RMG (N = 1418) |
Validation dataset (N = 998) |
|---|---|---|
| Age (years) | ||
| <65 | 494 (34·8%) | 422 (42·3%) |
| 65–75 | 624 (44·0%) | 403 (40·4%) |
| >75 | 300 (21·2%) | 173 (17·3%) |
| Mean (SD) | 67·4 (9·8) | 66·4 (10·2) |
| Median (IQR) | 68·0 (61·0–74·0) | 67·0 (59–74) |
| β2 microglobulin (mg/l) | ||
| <3·5 | 648 (45·7%) | 339 (34·0%) |
| 3·5–5·5 | 370 (26·1%) | 405 (40·6%) |
| >5·5 | 400 (28·2%) | 254 (25·5%) |
| Mean (SD) | 5·7 (6·8) | 4·5 (2·5) |
| Median (IQR) | 3·7 (2·6–5·8) | 3·9 (3·0–5·6) |
| β2 microglobulin at diagnosis (mg/l) | ||
| <3·5 | 548 (38·6%) | 369 (37·0%) |
| 3·5–5·5 | 392 (27·6%) | 389 (39·0%) |
| >5·5 | 478 (33·7%) | 240 (24·0%) |
| Mean (SD) | 6·3 (6·8%) | 4·4 (2·3) |
| Median (IQR) | 4·2 (2·8–6·6) | 3·9 (3·0–5·5) |
| Albumin (g/dl) | ||
| <3·5 | 275 (19·4%) | 509 (51·0%) |
| ≥3·5 | 1143 (80·6%) | 489 (49·0%) |
| Mean (SD) | 4·0 (0·6) | 3·5 (0·9) |
| Median (IQR) | 4·1 (3·7–4·4) | 3·4 (3·0–3·8) |
| CA at diagnosis | ||
| Standard risk | 95 (6·7%) | 332 (33·3%) |
| High risk | 175 (12·3%) | 209 (20·9%) |
| NA† | 1148 (81·0%) | 457 (45·8%) |
| LDH (U/l) | ||
| ≤360 | 1280 (90·3%) | 744 (74·5%) |
| >360 | 138 (9·7%) | 254 (25·5%) |
| Mean (SD) | 262·3 (311·6) | 302·7 (147·3) |
| Median (IQR) | 205·4 (167·7–267·1) | 246 (201–361) |
| LDH at diagnosis (U/l) | ||
| ≤360 | 1338 (94·4%) | 784 (78·6%) |
| >360 | 80 (5·6%) | 214 (21·4%) |
| Mean (SD) | 210·7 (142·5) | 296 (157·8) |
| Median (IQR) | 183·8 (150·3–232·3) | 246 (200–350) |
| ECOG performance status | ||
| 0 | 213 (15·0%) | 144 (14·4%) |
| 1 | 838 (59·1%) | 566 (56·7%) |
| 2 | 279 (19·7%) | 258 (25·9%) |
| 3–4 | 88 (6·2%) | 30 (3·0%) |
| Thrombocyte count (× 109/l) | ||
| >100 | 1271 (89·6%) | 867 (86·9%) |
| ≤100 | 147 (10·4%) | 131 (13·1%) |
| Mean (SD) | 193·7 (85·2) | 180·3 (79·4) |
| Median (IQR) | 185·0 (139·0–236·0) | 176 (121–212) |
| Hypercalcaemia | ||
| No (calcium ≤2·75 mmol/l) | 1360 (95·9%) | 705 (70·6%) |
| Yes (calcium >2·75 mmol/l) | 58 (4·1%) | 293 (29·4%) |
| Mean (SD) | 2·3 (0·3) | 3·4 (2·6) |
| Median (IQR) | 2·3 (2·2–2·4) | 2·4 (2·2–2·9) |
| Bone marrow plasma cell count (%) | ||
| <20 | 760 (53·6%) | 240 (24·0%) |
| 20–70 | 552 (38·9%) | 668 (66·9%) |
| >70 | 106 (7·5%) | 90 (9·0%) |
| Mean (SD) | 23·7 (25·4) | 36·6 (22·7) |
| Median (IQR) | 17·6 (0·0–37·5) | 32 (20·0–50·0) |
| Extramedullary disease | ||
| No | 1257 (88·6%) | 879 (88·1%) |
| Yes | 161 (11·4%) | 119 (11·9%) |
| Time to initiation of 2L treatment (months) | ||
| >24 | 415 (29·3%) | 467 (46·8%) |
| ≤24 | 1003 (70·7%) | 531 (53·2%) |
| Refractory to previous treatment | ||
| Non‐refractory/refractory to other regimens without new drugs | 1083 (76·4%) | 787 (78·9%) |
| Refractory to bortezomib | 149 (10·5%) | 123 (12·3%) |
| Refractory to thalidomide | 142 (10·0%) | 1 (0·1%) |
| Refractory to other regimens with new drugs§ | 44 (3·1%)‡ | 87 (8·7%) |
| New bone lesions | ||
| No new lesions | 410 (28·9%) | 396 (39·7%) |
| >2 lesions at diagnosis and at 2L, or new lesions | 1008 (71·1%) | 602 (60·3%) |
| NA | – | – |
| Severe toxicities during/before 1L treatment (highest grade experienced) | ||
| 0–2 | 635 (44·8%) | 879 (88·1%) |
| 3 or 4 | 783 (55·2%) | 119 (11·9%) |
| Prior 1L therapy | ||
| Bortezomib only | 499 (35·2%) | 479 (48·0%) |
| Thalidomide only | 549 (38·7%) | 259 (26·0%) |
| Bortezomib and thalidomide | 66 (4·7%) | 110 (11·0%) |
| Bortezomib and lenalidomide | 14 (1·0%) | 10 (1·0%) |
| Lenalidomide | 75 (5·3%) | 30 (3·0%) |
| Other with new drugs | 93 (6·6%) | 262 (26·2%) |
| Other without new drugs | 211 (14·9%) | 103 (10·3%) |
Data are n (%) unless stated otherwise.
1L, first‐line; 2L, second‐line; CA, cytogenetic abnormalities; ECOG, Eastern Cooperative Oncology Group; IQR, inter‐quartile range; LDH, lactate dehydrogenase; NA, not available; RMG, Registry of Monoclonal Gammopathies; SD, standard deviation.
As measured at the initiation of 2L treatment unless otherwise stated.
Missing values were not imputed for cytogenetic abnormalities.
‘New’ drugs include carfilzomib, daratumumab, elotuzumab, ixazomib, panobinostat, pomalidomide and thalidomide.
Refractory to other regimens with new drugs – includes bortezomib plus thalidomide, lenalidomide only, bortezomib plus lenalidomide, and lenalidomide plus thalidomide.
Validation cohort
Chart data from 998 patients were collected; the baseline characteristics are presented in Table 1. When initiating 2L treatment, 17·3% of patients in the validation set were aged over 75 years, 71% had an ECOG PS of 0 or 1, and over half (53%) had initiated 2L treatment within two years of diagnosis. Of note, fewer patients in the validation set had experienced a grade 3 or 4 toxicity during or before 1L treatment than in the development set (11·9% vs. 52·2%).
Predictors of overall survival included in the risk stratification algorithm
Of the 29 pre‐selected candidate parameters, 16 were identified as independent predictors of OS following Pearson’s correlation test and a Cox regression analysis (Fig 1 and Table 2). Although they are considered to be important variables when predicting risk in patients with MM, creatinine level, time since diagnosis, time to progression (TTP), treatment regimen in 1L, duration and depth of response, M protein level and stem cell transplantation status were excluded because they correlated with other stronger predictors of OS. Infection and neuropathy (during 1L treatment), serum albumin at diagnosis and nature of relapse were eliminated during the backward selection process. Haemoglobin level was removed as a predictor following guidance from physicians because low levels are not considered to be indicative of disease severity.
Table 2.
Predictors of overall survival at 2L (Cox regression analysis).
| Predictor of overall survival* | Classification of predictor | Categories/thresholds |
Backward selection HR (95% CI) AIC = 9172·275 |
P value |
|---|---|---|---|---|
| Age (years) | Fully continuous | NA | 1·015 (1·007–1·023)† | 0·0002 |
| Albumin (g/dl) | Fully continuous | NA | 0·846 (0·745–0·960)† | 0·0095 |
| Bone marrow plasma cell count (%) | Fully continuous | NA | 1·008 (1·005–1·011)† | <0·0001 |
| Thrombocyte count (×109/l) | Continuous with threshold | (150 × 109/cells) | 0·995 (0·992–0·997)† | <0·0001 |
| Sβ2M (mg/l) | Continuous with threshold | (5·5 mg/l) | 1·063 (0·993–1·138)† | 0·0787 |
| Sβ2M at diagnosis (mg/l) | Continuous with threshold | (5·5 mg/l) | 1·090 (1·022–1·162)† | 0·0084 |
| LDH (U/l) | Continuous with clinically established cut‐offs | ≤ULN | Reference | |
| >ULN | 2·080 (1·651–2·622) | <0·0001 | ||
| LDH at diagnosis (U/l) | Continuous with clinically established cut‐off | ≤360‡ | Reference | |
| >360 | 1·297 (0·960–1·752) | 0·0904 | ||
| Calcium (mmol/l) | Continuous with clinically established cut‐off | ≤2·75 | Reference | |
| >2·75 | 1·406 (1·012–1·954) | 0·0422 | ||
| Time to next treatment (months) | Continuous with clinically established cut‐off | >24 | Reference | |
| ≤24 | 1·112 (0·915–1·353) | 0·2858 | ||
| ECOG performance status | Categorical | 0 | Reference | |
| 1 | 1·667 (1·227–2·265) | 0·0011 | ||
| 2 | 2·123 (1·520–2·964) | <0·0001 | ||
| 3 or 4 | 3·708 (2·496–5·506) | <0·0001 | ||
| CA at diagnosis | Categorical | Standard risk | Reference | |
| High risk | 1·643 (1·147–2·353) | 0·0067 | ||
| NA | 1·081 (0·789–1·481) | 0·6299 | ||
| Extramedullary disease | Categorical | No | Reference | |
| Yes | 2·331 (1·872–2·904) | <0·0001 | ||
| New bone lesions (X‐ray) | Categorical | No new lesions | Reference | |
| >2 at diagnosis and initiation of 2L§ or new lesions | 1·271 (1·075–1·502) | 0·0049 | ||
| Refractory status | Categorical | Non‐refractory | Reference | |
| Refractory to bortezomib | 1·533 (1·202–1·955) | 0·0006 | ||
| Refractory to thalidomide | 1·186 (0·942–1·493) | 0·1446 | ||
| Refractory regimens with new agents¶ | 1·427 (0·961–2·120) | 0·0776 | ||
| Severe toxicities during 1L treatment (any grade 3 or 4 toxicity) | Categorical | No | Reference | |
| Yes | 1·145 (0·984–1·332) | 0·0797 |
1L, first line; 2L, second line; AIC, Akaike’s information criterion; CA, cytogenetic abnormalities; CI, confidence interval; ECOG, Eastern Cooperative Oncology Group; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not available; Sβ2M, serum β‐2 microglobulin; ULN, upper limit of normal.
At initiation of 2L treatment unless otherwise stated.
HR per unit change.
ULN was 360 U/l in this data set.
Category comprises patients with accelerated osteoporosis/>2 lesions at diagnosis and 2L.
Comprising bortezomib plus thalidomide (n = 21), lenalidomide only (n = 20), bortezomib plus lenalidomide (n = 2) and lenalidomide plus thalidomide (n = 1); two patients were recategorized based on known previous treatment.
The 16 predictors were split into those associated with patient frailty (age and ECOG PS) and those that describe disease aggressiveness (all other predictors, including patient’s experience during 1L treatment). These predictors reflect factors that should be considered when defining patient‐specific disease severity: characteristics at diagnosis (3 predictors), characteristics at the initiation of 2L treatment (10 predictors) and patient’s experience during 1L treatment (3 predictors) (Table 2). CA are not routinely reassessed at relapse, so CA status at diagnosis was used as a predictor of OS, rather CA status at the initiation of 2L. Of the prespecified parameters that would be included in the RSA regardless of significance, only time to initiation of 2L treatment was forced into the model. The risk score calculation for a theoretical patient entering 2L treatment is shown in Table 3.
Table 3.
Calculation of risk scores for a theoretical patient entering 2L treatment.
| Parameter (at initiation of 2L unless stated otherwise) | Value | Calculation | HR for calculating risk scores | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Age | 72 years | exp (0·015 × 72) | 2·945 | ||||||||
| ECOG PS | 1 | NA | 1·667 | ||||||||
| Frailty risk score | 4·9 | ||||||||||
| Sβ2M | 3·5 mg/l | exp (0·061 × 3·5) | 1·238 | ||||||||
| Sβ2M at diagnosis | 5·5 mg/l | exp (0·086 × 5·5) | 1·605 | ||||||||
| CA at diagnosis | Standard risk | NA | 1 | ||||||||
| Calcium | >2·75 mmol/l | 1·406 | |||||||||
| LDH | Below ULN | NA | 1 | ||||||||
| LDH at diagnosis | Below ULN | NA | 1 | ||||||||
| Extramedullary disease | No | 1 | |||||||||
| New bone lesions | No | 1 | |||||||||
| Serum albumin | 3·8 g/l | exp (−0·168 × 3·8) | 0·528 | ||||||||
| Thrombocyte count | 220 × 109/l | exp (−0·005 × 150) | 0·472 | ||||||||
| Bone marrow plasma count | 65% | exp (0·008 × 65) | 1·682 | ||||||||
| Time to next treatment | >24 months | 1 | |||||||||
| Refractory status | Refractory to bortezomib | 1·533 | |||||||||
| Severe toxicities in 1L | No | 1 | |||||||||
| Disease aggressiveness risk score | 2·0 | ||||||||||
| Total risk score | 9·8 | ||||||||||
| Theoretical patient risk group classification | 3 | ||||||||||
| Total risk score cut‐off values for classification | Group 1 | Group 2 | Group 3 | Group 4 | |||||||
| Min | Max | Min | Max | Min | Max | Min | Max | ||||
| – | ≤3 | >3 | ≤7 | >7 | ≤15·4 | >15·4 | – | ||||
1L, first line; 2L, second line; CA, cytogenetic abnormalities; ECOG PS, Eastern Cooperative Oncology Group performance status; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not available; Sβ2M, serum 2‐microglobulin; ULN, upper limit of normal.
Patient‐specific risk scores
Development cohort
The KAPS defined four distinct patient groups based on survival expectations and total risk scores (Fig 2). Risk Group 2 contained the largest proportion of patients (n = 596; 42%), followed by Group 1 (n = 351; 25%) and then Group 3 (n = 318; 22%); Group 4 contained the smallest number of patients (n = 153; 11%). As specified by the methodology, no groups contained less than 10% of the analysis population (Fig 2). The calculation formula to measure risk score is included in Appendix S1.
Figure 2.
Overall survival stratified by risk group (A) development cohort; (B) validation cohort. CI, confidence interval; HR, hazard ratio; NR, not reached; OS, overall survival; Ref, reference.
Validation cohort
When the RSA was applied to the validation cohort, 17·8% of patients were placed in Risk Group 1, 34·6% were in Risk Group 2, 24·9% were in Group 3, and 22·6% were in Group 4 (Fig 2B).
Overall survival by risk group
Development cohort
The differences in median OS from the initiation of 2L treatment between groups were statistically significant. The median OS halved, and the risk of death doubled with ascending risk group. The OSs for patients in Risk Groups 1–4 were 61·6, 29·6, 14·2 and 5·9 months, respectively (Figure S1), and the HRs for differences in OS between Group 1 and Groups 2, 3 and 4 were 2·24, 4·30 and 10·88, respectively (P > 0·001 in all cases), with no overlap in the associated 95% CI (Fig 3).
Figure 3.
Patient frailty risk scores versus disease aggressiveness risk scores. Scores over 15 for disease aggressiveness have been excluded (n = 3).
Validation cohort
The median OS was not reached for patients assigned to Groups 1 and 2, 39·9 months in Risk Group 3 and 16·2 months for those in Risk Group 4. The HRs for the differences in OS between patients in Group 1 and those in Groups 2, 3 and 4 were 1·87, 4·61 and 8·51, respectively (Fig 2B). The C‐index when the RSA was applied to the external data set was 0·715 (95% CI: 0·69–0·74) (a score of 0·5 represents total random predictions; a perfectly discriminating model would have a score of 1) (Harrell, 2001).
Drivers of risk of death
To provide a deeper understanding of patients’ needs when initiating 2L treatment, risk scores for patients in the development cohort were assigned based on the two groups of predictors: those associated with patient frailty and those that describe disease aggressiveness. Mean frailty risk scores increased from 3·7 in Group 1 to 6·9 in Group 4; the differences between risk groups were greater still for mean disease aggressiveness scores (0·6 in Group 1 vs. 4·9 in Group 4; Figure S1).
Plotting frailty risk score against disease aggressiveness risk scores provides a graphic illustration of the factors driving risk for individual patients in addition to the total risk score. As can be seen in Fig 3, both disease aggressiveness and patient frailty contribute to risk of death for the majority of patients. However, a closer inspection of the data showed that a subset of patients in Risk Group 4 (shaded area A) did not have aggressive disease, and the high risk was driven by frailty alone. In contrast, shaded area B shows a population of patients in Risk Group 4 with low frailty risk scores whose poor survival expectations were driven by the particularly aggressive nature of their disease.
Patient characteristics by risk group
Exploring the characteristics of patients in each risk group of the development cohort showed that no single predictor governed underlying risk of death at the initiation of 2L treatment. For example, of the patients in Risk Group 4, 32 were aged 65 years or below, and 52% had LDH levels below the upper limit of normal. Of the patients in Risk Group 1, however, 45% of evaluable patients had high‐risk CA, and 46% had 3 or 4 AE at grade 3 or 4 during 1L treatment (Table SII).
Other treatment outcomes by risk group
Although the RSA was developed to predict OS, the stratification also holds for progression‐free survival (PFS) and TTP in the development cohort. Patients in Risk Groups 1–4 experienced a median PFS of 18·1, 13·2, 8·3 and 3·4 months, respectively, and a median TTP of 18·9, 14·1, 9·5 and 3·7 months, respectively (Table SIII). The proportions of patients in Groups 1–4 who achieved a very good partial response (VGPR), or better, during treatment were 29%, 24%, 19% and 13%, respectively (Table SIII).
Development cohort
Treatment patterns were similar across risk groups, suggesting that physicians in the Czech Republic were not making treatment choices based on risk (Table SIII). OS outcomes were broadly similar for each group regardless of treatment used. Among patients with the same regimen, OS expectation increased with increasing risk group.
Validation cohort
Lenalidomide‐based regimens were the most common treatment choice for patients in each of the risk groups (43·8 % in Group 3 to 57·1% in Group 4) and bortezomib‐based regimens were the second most frequently used treatment (24·3% in group 4–40·6% in Group 3). There was a trend for new drugs to be used more commonly in the lower‐risk groups (9·6%, 9·3%, 5·6% and 4·0% in Groups 1–4, respectively), whilst old drugs were used more in higher‐risk groups (3·9%, 4·3%, 6·4% and 10·2% in Groups 1–4, respectively) (Table SIII).
Discussion
The RSA defines four groups with significantly different OS from the initiation of 2L treatment. In addition, the algorithm also allows the physician to identify the drivers of disease in terms of patient frailty or disease aggressiveness on an individual basis for the first time. Existing risk stratification tools for MM were primarily developed using clinical trial data from newly diagnosed patients and used a small number of disease‐related predictors. The validated analysis presented here demonstrated that to assess systematically and quantify appropriately the total risk and the drivers of risk, more information is needed regarding patients’ experiences during 1L and during clinical presentation at the initiation of 2L therapy.
Existing staging tools for patients with MM (ISS and R‐ISS) have not been validated in the relapse setting using real‐world data and may therefore be of limited value to medical decision‐making at first relapse. Evidence from a European retrospective chart review of deceased patients with MM suggests that using the R‐ISS stage alone is not a reliable way to predict prognosis in real‐world patient populations. The study reported that 8% of patients with ISS stage 1 disease died before receiving treatment or during 1L treatment, whereas 38% of patients with R‐ISS stage 3 disease received five lines of treatment (Yong et al, 2016). In a recent study (Tandon et al, 2017), in which the R‐ISS and ISS were applied to an RRMM population, almost two thirds of patients in the analysis (65%) were placed into the R‐ISS Group 2. It is likely that the patients in that large group had heterogeneous disease characteristics and survival expectations; however, the R‐ISS was unable to discriminate between them (Tandon et al, 2017). RRMM is a complex disease; therefore, several factors need to be taken into consideration to assess risk appropriately in patients at the initiation of 2L treatment. When analysed in the RMG data set, the RSA Cox model outperformed the R‐ISS and ISS Cox models (data not shown; unpublished observations). The RSA covers multiple disease dimensions and, as a result, includes 16 parameters when assessing risk of death, with the exception of analysis of CA, all of which are routinely available. Tools with a narrow focus on a limited number of disease‐related factors may be inaccurate when applied to complex diseases and may not provide the right patient‐specific risk assessment to support therapy and management decisions. Instead, a balance of complexity and clinical practicality is required to provide a tool that will improve disease understanding and patient management.
While the R‐ISS and ISS have focused on disease‐related parameters, other studies have reported links between survival and frailty parameters in newly diagnosed cohorts. Real‐world data from patients with MM in the Netherlands have shown that relative survival decreased significantly with age (Schaapveld et al, 2010). The IMWG went further and assessed the impact of frailty on survival in a pooled analysis of 869 patients with MM (Palumbo et al, 2015b). Each patient was assigned a frailty score based on age, comorbidities, cognitive and physical conditions, and was stratified into one of three groups (fit, intermediate fitness and frail). Three‐year OS rates were significantly higher in fit patients than in intermediate fitness or frail patients (Palumbo et al, 2015b). Neither study considered the impact of disease aggressiveness. The RSA is the first integrated tool that allows physicians to quantify the relative impact of patient frailty and disease aggressiveness on survival outcomes in patients with MM at relapse.
Our algorithm also highlights the need to consider experiences during 1L to accurately determine the prognosis at the initiation of 2L. The influence of 1L outcomes on subsequent treatment lines was illustrated in a real‐world chart audit of patients from seven European countries; this showed that experiencing an AE during a line of treatment and not achieving a VGPR or better, each significantly reduced the chances of a patient receiving the next line of treatment (Yong et al, 2016). It can therefore be expected that, by preventing patients receiving the next treatment line, such experiences would also adversely affect their prognosis.
Although the higher OS values in the validation cohort may indicate a healthier population than in the development cohort, the patient characteristics do not support this conclusion. The patients in the validation cohort were treated more recently than the development cohort and therefore had the opportunity of being treated with novel agents at later treatment lines. With this exception, treatment patterns were broadly consistent across the risk groups in both the development and validation cohorts, and did not appear to be influenced by risk of patient death. This could suggest that physicians may not be able to predict patients’ prognoses accurately when making treatment decisions. Alternatively, it may highlight the complexity of determining patient prognoses and optimal treatment options at 2L. Traditionally, the class of agent would only be changed when the previous line of treatment has failed. Our tool suggests that 16 factors need to be considered to determine prognosis accurately. This could be perceived as challenging in the clinic, but most parameters are either measured routinely or readily available. While CA at relapse is not routinely measured in clinical practice, and therefore not suitable for inclusion in the RSA, data suggest that there is an association between the frequency of high‐risk cytogenetic lesions in patients at relapse and survival outcomes (Dimopoulos et al, 2015). Thus, it would be of interest in future work to measure CA at relapse and to investigate how this variable may contribute to predicting overall survival.
The RSA was developed and validated in a cohort that was treated with either proteasome inhibitors or immunomodulators initiating 1L treatment. While this is reflective of some European countries, where many patients do not receive combination therapies at 1L (Raab et al, 2016), it is becoming standard practice in other countries (e.g. the United States) to combine these agents at 1L (Rifkin et al, 2016) While current practice has started to combine these agents at 1L, robust OS data are not readily available because these patients have been treated very recently. Future work could test the RSA in a cohort of patients treated with combination therapy at 1L.
There is potential for the algorithm defined in our study to be converted into a tool to aid physicians with their patient management. In the future, this may also help with the complex treatment decisions that need to be made at first relapse. Cost constraints and the AE profiles of some double‐ and triplet regimens are likely to limit the extent to which the most effective regimens are used in the clinic. More research is needed to identify the optimum treatment strategy for patients in each risk group. However, through systematic assessment of risk, and the extent to which patient frailty and disease aggressiveness contribute to risk, the RSA can provide valuable information to help physicians select the treatment regimen and dosing schedule on an individual basis.
In conclusion, we have developed and validated an algorithm that can stratify patients in routine clinical practice according to their clinical expectations. The value of this RSA lies in its ability to define patient‐specific risk and combine both frailty and disease aggressiveness into a single tool that can help guide management decisions in response to the relapse of patients with MM, ultimately to improve outcomes.
Disclosures
RH has received research funding from Amgen and Celgene, consultancy fees from Amgen, Celgene and Takeda, and honoraria from Amgen, Bristol‐Myers Squibb and Janssen. MD has received research funding from Janssen and Celgene, and honoraria from Amgen, BMS, Celgene, Janssen and Takeda. MSR has received research funding from Amgen and Novartis, has received consultancy fees from Amgen, Bristol‐Myers Squibb and Novartis, and has participated in advisory boards for Celgene and Janssen, and in speaker bureau for Janssen. PS is an employee of Amgen Europe and a stockholder in Amgen Inc. LD is an employee of Amgen Ltd and a stockholder in Amgen Inc. IS has received research funding from Celgene, and honoraria from Amgen, Bristol‐Myers Squibb, Celgene, Janssen, Novartis and Takeda. JR has no conflicts to disclose. LP has no conflicts to disclose. SGM is an employee of Amgen Europe and a stockholder in Amgen Inc. WB is an employee of Pharmerit International who received funding from Amgen to conduct this research.
Supporting information
Fig S1. Total risk scores, frailty risk score and disease aggressiveness risk scores by group.
Table SI. Patient chart review demographics of the validation cohort.
Table SII. Characteristics of patients in each risk group of the development cohort.
Table SIII. Treatment outcomes stratified by risk group.
Appendix S1. Calculation formula to measure risk score.
Acknowledgments
Medical writing support was provided by Matthew Long PhD of Oxford PharmaGenesis, Oxford, UK and funded by Amgen Europe GmbH. This study was funded by Amgen Europe GmbH. SMG, LDC and WB performed the research. SMG, LDC and WB designed the research study. All authors analysed the data. All authors contributed to writing the paper by providing guidance and comments on its content.
References
- Austin, P.C. & Tu, J.V. (2004) Automated variable selection methods for logistic regression produced unstable models for predicting acute myocardial infarction mortality. Journal of Clinical Epidemiology, 57, 1138–1146. [DOI] [PubMed] [Google Scholar]
- Bird, J.M. , Owen, R.G. , D'Sa, S. , Snowden, J.A. , Pratt, G. , Ashcroft, J. , Yong, K. , Cook, G. , Feyler, S. , Davies, F. , Morgan, G. , Cavenagh, J. , Low, E. & Behrens, J. ; on behalf of the Haemato‐oncology Task Force of the British Committee for Standards in Haematology (BCSH) and UK Myeloma Forum . (2011) Guidelines for the diagnosis and management of multiple myeloma 2011. British Journal of Haematology, 154, 32–75. [DOI] [PubMed] [Google Scholar]
- Chng, W.J. , Dispenzieri, A. , Chim, C.S. , Fonseca, R. , Goldschmidt, H. , Lentzsch, S. , Munshi, N. , Palumbo, A. , Miguel, J.S. , Sonneveld, P. , Cavo, M. , Usmani, S. , Durie, B.G. & Avet‐Loiseau, H. ; International Myeloma Working Group . (2014) IMWG consensus on risk stratification in multiple myeloma. Leukemia, 28, 269–277. [DOI] [PubMed] [Google Scholar]
- Cole, S.R. , Platt, R.W. , Schisterman, E.F. , Chu, H. , Westreich, D. , Richardson, D. & Poole, C. (2010) Illustrating bias due to conditioning on a collider. International Journal of Epidemiology, 39, 417–420. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dimopoulos, M.A. , Terpos, E. , Niesvizky, R. & Palumbo, A. (2015) Clinical characteristics of patients with relapsed multiple myeloma. Cancer Treatment Reviews, 41, 827–835. [DOI] [PubMed] [Google Scholar]
- Dimopoulos, M.A. , Moreau, P. , Palumbo, A. , Joshua, D. , Pour, L. , Hajek, R. , Facon, T. , Ludwig, H. , Oriol, A. , Goldschmidt, H. , Rosinol, L. , Straub, J. , Suvorov, A. , Araujo, C. , Rimashevskaya, E. , Pika, T. , Gaidano, G. , Weisel, K. , Goranova‐Marinova, V. , Schwarer, A. , Minuk, L. , Masszi, T. , Karamanesht, I. , Offidani, M. , Hungria, V. , Spencer, A. , Orlowski, R.Z. , Gillenwater, H.H. , Mohamed, N. , Feng, S. & Chng, W.J. ; ENDEAVOR Investigators . (2016a) Carfilzomib and dexamethasone versus bortezomib and dexamethasone for patients with relapsed or refractory multiple myeloma (ENDEAVOR): a randomised, phase 3, open‐label, multicentre study. The Lancet Oncology, 17, 27–38. [DOI] [PubMed] [Google Scholar]
- Dimopoulos, M.A. , Oriol, A. , Nahi, H. , San‐Miguel, J. , Bahlis, N.J. , Usmani, S.Z. , Rabin, N. , Orlowski, R.Z. , Komarnicki, M. , Suzuki, K. , Plesner, T. , Yoon, S.S. , Ben Yehuda, D. , Richardson, P.G. , Goldschmidt, H. , Reece, D. , Lisby, S. , Khokhar, N.Z. , O'Rourke, L. , Chiu, C. , Qin, X. , Guckert, M. , Ahmadi, T. , Moreau, P. & Investigators, P. (2016b) Daratumumab, lenalidomide, and dexamethasone for multiple myeloma. New England Journal of Medicine, 375, 1319–1331. [DOI] [PubMed] [Google Scholar]
- Eisemann, N. , Waldmann, A. & Katalinic, A. (2011) Imputation of missing values of tumour stage in population‐based cancer registration. BMC Medical Research Methodology, 11, 129. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Eo, S.‐H. , Kang, H.J. , Hong, S.‐M. & Cho, H.J . (2014) K‐adaptive partitioning for survival data, with an application to cancer staging. arXiv:1306.4615v3 [stat.AP], https://arxiv.org/pdf/1306.4615.pdf.
- Ethun, C.G. , Bilen, M.A. , Jani, A.B. , Maithel, S.K. , Ogan, K. & Master, V.A. (2017) Frailty and cancer: implications for oncology surgery, medical oncology, and radiation oncology. CA: A Cancer Journal for Clinicians, 67, 362–377. [DOI] [PubMed] [Google Scholar]
- Fonseca, R. , Bergsagel, P.L. , Drach, J. , Shaughnessy, J. , Gutierrez, N. , Stewart, A.K. , Morgan, G. , VanNess, B. , Chesi, M. , Minvielle, S. , Neri, A. , Barlogie, B. , Kuehl, W.M. , Liebisch, P. , Davies, F. , Chen‐Kiang, S. , Durie, B.G. , Carrasco, R. , Sezer, O. , Reiman, T. , Pilarski, L. & Avet‐Loiseau, H. ; International Myeloma Working Group . (2009) International Myeloma Working Group molecular classification of multiple myeloma: spotlight review. Leukemia, 23, 2210–2221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fonseca, R. , Abouzaid, S. , Bonafede, M. , Cai, Q. , Parikh, K. , Cosler, L. & Richardson, P. (2017) Trends in overall survival and costs of multiple myeloma, 2000–2014. Leukemia, 31, 1915–1921. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gonzalez‐McQuire, S. , Dimopoulos, M.A. , Weisel, K. , Bouwmeester, W. , Hajek, R. , Campioni, M. , Bennison, C. , Xu, W. , Pantiri, K. , Hensen, M. , Terpos, E. & Knop, S. (2019) Development of an initial conceptual model of multiple myeloma to support clinical and health economics decision making. Medical Decision Making Policy and Practice, 4 10.1177/2381468318814253 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Greipp, P.R. , San Miguel, J. , Durie, B.G. , Crowley, J.J. , Barlogie, B. , Blade, J. , Boccadoro, M. , Child, J.A. , Avet‐Loiseau, H. , Kyle, R.A. , Lahuerta, J.J. , Ludwig, H. , Morgan, G. , Powles, R. , Shimizu, K. , Shustik, C. , Sonneveld, P. , Tosi, P. , Turesson, I. & Westin, J. (2005) International staging system for multiple myeloma. Journal of Clinical Oncology, 23, 3412–3420. [DOI] [PubMed] [Google Scholar]
- Groenwold, R.H. , White, I.R. , Donders, A.R. , Carpenter, J.R. , Altman, D.G. & Moons, K.G. (2012) Missing covariate data in clinical research: when and when not to use the missing‐indicator method for analysis. CMAJ, 184, 1265–1269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hájek, R. , Jarkovsky, J. , Bouwmeester, W. , Treur, M. , DeCosta, L. , Campioni, M. , Brozova, L. & Gonzalez‐McQuire, S. (2016) Predictors of overall survival in patients with multiple myeloma initiating first‐ and second‐line treatment in the Czech Republic. Blood, 128, Abstr 3607. [Google Scholar]
- Harrell, F.E. (2001) Regression Modeling Strategies. Springer, New York. [Google Scholar]
- Lonial, S. , Dimopoulos, M.A. , Palumbo, A. , White, D. , Grosicki, S. , Spicka, I. , Walter‐Croneck, A. , Moreau, P. , Mateos, M.‐V. , Magen‐Nativ, H. , Belch, A. , Reece, D.E. , Beksac, M. , Taniwaki, M. , Röllig, C. , Singhal, A.K. , Katz, J. , Bleickardt, E.W. , Poulart, V. & Richardson, P.G. (2015) ELOQUENT‐ 2: A phase III, randomized, open‐label study of lenalidomide (Len)/dexamethasone (dex) with/without elotuzumab (Elo) in patients (pts) with relapsed/refractory multiple myeloma (RRMM). Journal of Clinical Oncology, 33, abstr 8508. [Google Scholar]
- Moreau, P. , Masszi, T. , Grzasko, N. , Bahlis, N.J. , Hansson, M. , Pour, L. , Sandhu, I. , Ganly, P. , Baker, B.W. , Jackson, S.R. , Stoppa, A.M. , Simpson, D.R. , Gimsing, P. , Palumbo, A. , Garderet, L. , Cavo, M. , Kumar, S. , Touzeau, C. , Buadi, F.K. , Laubach, J.P. , Berg, D.T. , Lin, J. , DiBacco, A. , Hui, A.M. , van deVelde, H. & Richardson, P.G. ; TOURMALINE‐MM1 Study Group . (2016) Oral ixazomib, lenalidomide, and dexamethasone for multiple myeloma. New England Journal of Medicine, 374, 1621–1634. [DOI] [PubMed] [Google Scholar]
- Moreau, P. , San Miguel, J. , Sonneveld, P. , Mateos, M.V. , Zamagni, E. , Avet‐Loiseau, H. , Hajek, R. , Dimopoulos, M.A. , Ludwig, H. , Einsele, H. , Zweegman, S. , Facon, T. , Cavo, M. , Terpos, E. , Goldschmidt, H. , Attal, M. & Buske, C. (2017) Multiple myeloma: ESMO clinical practice guidelines for diagnosis, treatment and follow‐up. Annals of Oncology, 28, iv52–iv61. [DOI] [PubMed] [Google Scholar]
- Palumbo, A. , Avet‐Loiseau, H. , Oliva, S. , Lokhorst, H.M. , Goldschmidt, H. , Rosinol, L. , Richardson, P. , Caltagirone, S. , Lahuerta, J.J. , Facon, T. , Bringhen, S. , Gay, F. , Attal, M. , Passera, R. , Spencer, A. , Offidani, M. , Kumar, S. , Musto, P. , Lonial, S. , Petrucci, M.T. , Orlowski, R.Z. , Zamagni, E. , Morgan, G. , Dimopoulos, M.A. , Durie, B.G. , Anderson, K.C. , Sonneveld, P. , San Miguel, J. , Cavo, M. , Rajkumar, S.V. & Moreau, P. (2015a) Revised international staging system for multiple myeloma: a report from International Myeloma Working Group. Journal of Clinical Oncology, 33, 2863–2869. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Palumbo, A. , Bringhen, S. , Mateos, M.V. , Larocca, A. , Facon, T. , Kumar, S.K. , Offidani, M. , McCarthy, P. , Evangelista, A. , Lonial, S. , Zweegman, S. , Musto, P. , Terpos, E. , Belch, A. , Hajek, R. , Ludwig, H. , Stewart, A.K. , Moreau, P. , Anderson, K. , Einsele, H. , Durie, B.G. , Dimopoulos, M.A. , Landgren, O. , San Miguel, J.F. , Richardson, P. , Sonneveld, P. & Rajkumar, S.V. (2015b) Geriatric assessment predicts survival and toxicities in elderly myeloma patients: an International Myeloma Working Group report. Blood, 125, 2068–2074. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Palumbo, A. , Chanan‐Khan, A. , Weisel, K. , Nooka, A.K. , Masszi, T. , Beksac, M. , Spicka, I. , Hungria, V. , Munder, M. , Mateos, M.V. , Mark, T.M. , Qi, M. , Schecter, J. , Amin, H. , Qin, X. , Deraedt, W. , Ahmadi, T. , Spencer, A. & Sonneveld, P. (2016) Daratumumab, bortezomib, and dexamethasone for multiple myeloma. New England Journal of Medicine, 375, 754–766. [DOI] [PubMed] [Google Scholar]
- Raab, M.S. , Cavo, M. , Delforge, M. , Driessen, C. , Fink, L. , Flinois, A. , Gonzalez‐McQuire, S. , Safaei, R. , Karlin, L. , Mateos, M.V. , Schoen, P. & Yong, K. (2016) Multiple myeloma: practice patterns across Europe. British Journal of Haematology, 175, 66–76. [DOI] [PubMed] [Google Scholar]
- Radocha, J. , Pour, L. , Spicka, I. , Scudla, V. , Gregora, E. , Frankova, H. , Hajek, R. , Schutzova, M. , Sykora, M. , Kessler, P. , Adamova, D. , Wrobel, M. , Sedlarikova, L. , Sevcikova, S. , Horinek, D. , Pelcova, J. , Brozova, L. , Jarkovsky, J. & Maisnar, V. (2015) Registry of Monoclonal Gammopathies (RMG) in the Czech Republic. Blood, 126, Abstr 4514. [Google Scholar]
- Rifkin, R.M. , Abonour, R. , Durie, B.G.M. , Gasparetto, C.J. , Jagannath, S. , Narang, M. , Shah, J.J. , Terebelo, H.R. , Toomey, K. , Kitali, A. , Zafar, F. , Srinivasan, S. & Hardin, J.W. (2016) Treatment patterns from 2009 to 2015 in patients with newly diagnosed multiple myeloma in the United States: a report from the Connect® MM Registry. Blood, 128, 4489–4489. [Google Scholar]
- Ross, F.M. , Avet‐Loiseau, H. , Ameye, G. , Gutierrez, N.C. , Liebisch, P. , O'Connor, S. , Dalva, K. , Fabris, S. , Testi, A.M. , Jarosova, M. , Hodkinson, C. , Collin, A. , Kerndrup, G. , Kuglik, P. , Ladon, D. , Bernasconi, P. , Maes, B. , Zemanova, Z. , Michalova, K. , Michau, L. , Neben, K. , Hermansen, N.E. , Rack, K. , Rocci, A. , Protheroe, R. , Chiecchio, L. , Poirel, H.A. , Sonneveld, P. , Nyegaard, M. , Johnsen, H.E. on behalf of the European Myeloma Network . (2012) Report from the European Myeloma Network on interphase FISH in multiple myeloma and related disorders. Haematologica, 97, 1272–1277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Russell, S.J. & Rajkumar, S.V. (2011) Multiple myeloma and the road to personalised medicine. The Lancet Oncology, 12, 617–619. [DOI] [PubMed] [Google Scholar]
- Schaapveld, M. , Visser, O. , Siesling, S. , Schaar, C.G. , Zweegman, S. & Vellenga, E. (2010) Improved survival among younger but not among older patients with multiple myeloma in the Netherlands, a population‐based study since 1989. European Journal of Cancer, 46, 160–169. [DOI] [PubMed] [Google Scholar]
- Sonneveld, P. & Broijl, A. (2016) Treatment of relapsed and refractory multiple myeloma. Haematologica, 101, 995. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Stewart, A.K. , Rajkumar, S.V. , Dimopoulos, M.A. , Masszi, T. , Spicka, I. , Oriol, A. , Hajek, R. , Rosinol, L. , Siegel, D.S. , Mihaylov, G.G. , Goranova‐Marinova, V. , Rajnics, P. , Suvorov, A. , Niesvizky, R. , Jakubowiak, A.J. , San‐Miguel, J.F. , Ludwig, H. , Wang, M. , Maisnar, V. , Minarik, J. , Bensinger, W.I. , Mateos, M.V. , Ben‐Yehuda, D. , Kukreti, V. , Zojwalla, N. , Tonda, M.E. , Yang, X. , Xing, B. , Moreau, P. , Palumbo, A. & Investigators, A. (2015) Carfilzomib, lenalidomide, and dexamethasone for relapsed multiple myeloma. New England Journal of Medicine, 372, 142–152.25482145 [Google Scholar]
- Tandon, N. , Rajkumar, S.V. , LaPlant, B. , Pettinger, A. , Lacy, M.Q. , Dispenzieri, A. , Buadi, F.K. , Gertz, M.A. , Hayman, S.R. , Leung, N. , Go, R.S. , Dingli, D. , Kapoor, P. , Lin, Y. , Hwa, Y.L. , Fonder, A.L. , Hobbs, M.A. , Zeldenrust, S.R. , Lust, J.A. , Gonsalves, W.I. , Russell, S.J. & Kumar, S.K. (2017) Clinical utility of the Revised International Staging System in unselected patients with newly diagnosed and relapsed multiple myeloma. Blood Cancer Journal, 7, e528. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Terpos, E. , Katodritou, E. , Roussou, M. , Pouli, A. , Michalis, E. , Delimpasi, S. , Parcharidou, A. , Kartasis, Z. , Zomas, A. , Symeonidis, A. , Viniou, N.A. , Anagnostopoulos, N. , Economopoulos, T. , Zervas, K. & Dimopoulos, M.A. ; Greek Myeloma Study Group . (2010) High serum lactate dehydrogenase adds prognostic value to the international myeloma staging system even in the era of novel agents. European Journal of Haematology, 85, 114–119. [DOI] [PubMed] [Google Scholar]
- Vu, T. , Gonsalves, W. , Kumar, S. , Dispenzieri, A. , Lacy, M.Q. , Buadi, F. , Gertz, M.A. & Rajkumar, S.V. (2015) Characteristics of exceptional responders to lenalidomide‐based therapy in multiple myeloma. Blood Cancer Journal, 5, e363. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Yong, K. , Delforge, M. , Driessen, C. , Fink, L. , Flinois, A. , Gonzalez‐McQuire, S. , Safaei, R. , Karlin, L. , Mateos, M.V. , Raab, M.S. , Schoen, P. & Cavo, M. (2016) Multiple myeloma: patient outcomes in real‐world practice. British Journal of Haematology, 175, 252–264. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Fig S1. Total risk scores, frailty risk score and disease aggressiveness risk scores by group.
Table SI. Patient chart review demographics of the validation cohort.
Table SII. Characteristics of patients in each risk group of the development cohort.
Table SIII. Treatment outcomes stratified by risk group.
Appendix S1. Calculation formula to measure risk score.