2.1 Study design and data source

This was a retrospective analysis of a random sample of patients with NDMM initiating 1LT from an electronic healthcare record (EHR) database, the Integrated Oncology Network (ION) database, with data supplemented further with medical chart review data. Treatment data obtained from the chart review was verified independently by an expert MM oncologist. The ION EHR contains disease- and treatment-specific variables from 350 unique providers from over 25 geographically diverse, large, community-based oncology practices encompassing more than 650,000 patients. The data are certified as de-identified in line with the US Health Insurance Portability and Accountability Act statistical de-identification rules. Institutional review board (IRB) approval for this study was obtained from the Advarra IRB.

The study period was January 1, 2012 through March 31, 2018. Treated MM patients were initially identified during the enrollment period from July 1, 2012 through December 31, 2017. The index diagnosis date was defined as the first chronologically occurring MM diagnosis during the enrollment period. Subsequently, the first chronologically occurring date of an MM chemotherapy during the enrollment period was defined as the index treatment date. The baseline period was the 6-month period prior to index treatment date and was used to characterize the patients. The follow-up period (including the index treatment date) was variable for all patients; patients were followed longitudinally until death, loss to follow-up (defined as no additional data available prior to the end of study period), or end of study period (March 31, 2018). See Figure SS1.

2.2 Patient identification criteria

The population of interest in this study were patients with NDMM who did not undergo SCT as part of 1LT. Patients were included in the study if they had a MM diagnosis between July 1, 2007 through March 31, 2018; received MM-directed chemotherapy following the index diagnosis date during the enrollment period; and were aged ≥18 years on the index diagnosis date. Patients were excluded if they had evidence of any anti-cancer systemic therapy or SCT prior to index diagnosis date (except for dexamethasone given for <90 days), no evidence of MM-directed chemotherapy, missing gender or date of birth, or if they did not have activity in the EHR after index diagnosis date.

Of the patients who met the study selection criteria, a random sample of 300 patients were selected from the pool of patients who initiated 1LT after July 1, 2012, and manual chart review was performed for these patients to provide a clinically enriched dataset. Randomly selecting patients from this time period provided the most contemporary analysis of treatment patterns and outcomes available in the data set. Additional details provided in Figure 1.

2.3 Study variables

2.4 Outcome measures

First-line DOT was defined as the time from the start of the 1LT induction regimen to the end of all drug components of the 1LT regimen, including dose-attenuation (if observed). A regimen that ended because of discontinuation or death (event) was not censored. A regimen that ended because of the end of study/follow-up, was considered incomplete; and therefore, these observations were censored at the date of last follow-up. Gaps between drug administrations were not assessed, and end of DOT was defined based on treatment discontinuation date; if a discontinuation date was not noted, it was assumed that the patient was still on therapy unless they died.

TTNT, a proxy for PFS, was defined as the time from the start of 1LT to the start of 2LT or death (whichever event occurred earlier). Patients were censored at the date of last follow-up if they did not initiate 2LT, died within the follow-up period, or were lost to follow-up.

PFS was defined as the time from the start of 1LT to the earliest of the date of progression (event) or death (event) before the start of 2LT; dates of progression events were abstracted from patients' medical charts based on documentation by treating physicians, (described above). Patients who did not experience disease progression or death before the start of 2LT were censored at the beginning of 2LT (if they initiated 2LT) or at the end of follow-up (if they did not initiate 2LT), whichever occurred first.

OS was defined as the time from the start of 1LT to the date of death from any cause for patients who died. Observations for patients who were alive until study end date or were lost to follow-up were censored on the date of last follow-up/study end date (whichever occurred first).

2.5 Statistical analysis

Descriptive analyses were performed to examine demographics, baseline clinical characteristics, induction treatment patterns, and dose-attenuated treatment patterns in the overall population and stratified by DOT time intervals (≤3 months, >3 and ≤6 months, >6 and ≤ 12 months, and > 12 months [derived from observed treatment duration patterns]). Univariate comparisons between the DOT groups were conducted using Chi-square tests for categorical variables, Fisher's exact tests for binary variables when expected counts were <5 observations, Monte Carlo exact tests for categorical variables when expected counts were <5 observations and ANOVA for continuous variables.

Kaplan–Meier analysis was used to examine DOT among all patients and by receipt of dose-attenuated 1LT. Statistical comparison was done using the log-rank test. TTNT, PFS, and OS were described for the overall population. Median TTNT, PFS, and OS and rates at 1 and 2 years were estimated using Kaplan–Meier analyses.

Marginal structural models (MSM)²⁵ were used to evaluate the impact of DOT on TTNT, PFS, and OS to account for the time-dependent nature of DOT and time-varying confounders. First, the probability of continuing 1LT at each month from initiation of 1LT was estimated using a logistic regression model with being on treatment at each month as the dependent variable. Second, the probability of staying in the study at each month from the initiation of 1LT was estimated also using logistic regression, with being in the study at each month as the dependent variable. In both logistic regression models, covariates that remained constant were age, gender, race, region, year of index treatment start, ISS stage, immunoglobulin class, cytogenetic risk, frailty, CCI, and time from diagnosis to initiation of 1LT, while the time-varying covariates that were updated monthly included ECOG PS, diabetes, thromboembolic disease, peripheral neuropathy, non-MM cancer, and cardiovascular disease. The propensity scores derived from the logistic regression models were used to generate weights at each month, by multiplying the probability of being on 1LT at each month and the probability of being in the study at each month. Indicators (yes vs. no) for the outcomes (TTNT, PFS, and OS) at each month were defined based on whether or not the outcome has occurred in each month. These probabilities reflect the extent to which observations with certain characteristics are under-represented or over-represented in the sample with the respect to a target population in which these characteristics are balanced across DOT groups. The outcome indicators were modeled using logistic regression including the weights and the covariates to determine the average-odds of having the outcome of interest for one additional month since the time of initiation of 1LT at 2 years of 1LT. The outcomes (TTNT/PFS/OS) were truncated at 2 years, given extensive censoring and a small number of patients remaining following 2 years of follow-up (n = 33, 27, and 62 for TTNT, PFS, and OS, respectively). Results from the MSM were summarized as multivariable adjusted average ORs with associated 95% CIs, indicating the average adjusted odds of having a given outcome with each additional month of continuous 1LT among those who continued versus the average adjusted odds among those who discontinued 1LT.

All analyses were performed using SAS® version 9.4 (SAS Institute).

3.1 Baseline characteristics

A total of 207 patients were eligible for inclusion in the study following chart abstraction (Figure 1). Patients were categorized by DOT of 0–3 months (n = 64), 3–6 months (n = 45), 6–12 months (n = 45), and 12+ months (n = 53). Overall, patient characteristics were similar across the DOT categories with few exceptions, ECOG PS, CI score, modified frailty index score, and thromboembolic disease (Table 1). Over half of the patients were 75 years of age or older (55.1%), and across all DOT groups, the 0–3 DOT had the highest proportion of patients in this age group and the 12+ DOT group had the lowest proportion of patients in this age group. The majority of patients were male (59.4%) and Caucasian (55.6%). In terms of comorbidities, 65.7% of patients had a CCI score ≥2, with the 0–3 DOT group having the highest proportion of patients with CCI ≥2 and the 12+ DOT group having the lowest proportion of patients with CCI ≥2 across the DOT groups. The modified frailty index classified 47.3% of patients as being frail, and the proportion of frail patients was highest in the 0–3 DOT group and lowest in the 12+ DOT group. ECOG PS was ≥2 for 23.2% of patients (29.5%, unknown); patients with 12+ months of DOT tended to have lower ECOG PS with 11.3% of patients having an ECOG PS of ≥2 (35.8%, unknown) relative to those with DOT <12 months. ISS stage 3 was reported in 29.0% of patients (45.9%, unknown) and the most common immunoglobulin classes were IgG (44%), light chain (24.6%), and IgA (18.4%). High-risk cytogenetics were reported in 18.8% of patients (42%, unknown). All patients had ≥1 CRAB symptom, with 92.3%, 66.7%, 49.8%, and 22.4% having anemia, renal insufficiency, bone disease, and hypercalcemia, respectively. Diabetes was the most common other comorbid condition (22.2%).

The median (interquartile range [IQR]) time from MM diagnosis to initiation of 1LT was 0.6 (0.9) months, and median (IQR) follow-up time from treatment initiation was 14.2 (24.5) months.

3.2 Treatment patterns

Figures 2A and 2B present induction and dose-attenuated regimens in 1LT. Induction therapy consisted of 3-drug regimens or greater in 50.2% and 2-drug regimens or fewer in 49.8% of patients. The most common induction 1LT regimens were bortezomib, lenalidomide, dexamethasone (VRd, 30.4%), Vd (27.1%), Rd (18.4%), and bortezomib, cyclophosphamide, and dexamethasone (VCd, 16.4%). Correspondingly, the most common drug categories were PI/IMID-based (30.4%), PI-based (28.5%), IMID-based (19.8%), and PI/alkylator-based (18.8%) regimens. Reasons for discontinuing treatment with VRd, Vd, Rd, and VCd are provided in Table 2.

The majority of patients did not undergo dose-attenuation during 1LT (n = 146, 70.5%). However, as first-line DOT increased, the number of patients receiving dose-attenuation increased (p < 0.0001).

Among patients who did not receive dose-attenuated therapy during 1LT (n = 146), 87.7% (n = 128) discontinued induction therapy; the most common reasons for discontinuation were death (22.7%), progression of disease (21.1%), and adverse event (AE)/toxicity (16.4%). The majority of patients who discontinued induction therapy had ≥60 days of follow-up after discontinuation (59.4%). Of those who discontinued 1LT induction without dose-attenuation, 69.9% of patients did not initiate 2LT within 60 days of discontinuing 1LT; this may indicate patients who discontinued 1LT (without dose-attenuation) did not rapidly progress, suggesting that these patients had the opportunity to receive dose-attenuation.

Among patients who received dose-attenuated therapy (n = 61, 29.5%), the most commonly administered dose-attenuated 1LT regimens were lenalidomide (36.1%), Rd (26.2%), bortezomib (19.7%), and Vd (16.4%). Among these patients, 68.9% (n = 42) discontinued the dose-attenuated therapy during the study period. The most common reasons for discontinuation were progression of disease (n = 10, 23.8%), AE/toxicity (n = 9, 21.4%), and death (n = 7, 16.7%).

3.3 First-Line DOT

The median DOT for 1LT was 20.9 months (95% CI: 13.9, 26.4) versus 4.2 months (95% CI: 3.2, 4.9) for patients receiving dose-attenuation versus those who did not receive dose-attenuation, respectively (p < 0.0001) (Figure 3). One- and 2-year discontinuation rates in the dose-attenuation cohort were 27.9% and 59.3%, respectively, and 90.4% and 95.4% in the cohort without dose-attenuation, respectively. The unadjusted analyses of first-line DOT for the top four induction regimens are provided in the Appendix S1.

3.4 Unadjusted and adjusted analysis of 1LT TTNT, PFS, and OS

Unadjusted results, based on Kaplan–Meier analyses, indicated the 1LT median TTNT was 10.4 months (95% CI: 8.0, 17.1) and median PFS was 12.3 months (95% CI: 8.2, 16.6). Median OS was not reported due to extensive censoring (Figures 4A, 4B, 4C). The 1LT 1-year rate of TTNT was 49.6%, PFS was 50.0%, and OS was 76.1%. The 2-year 1LT TTNT, PFS, and OS rates were 31.4%, 28.0%, and 59.7%, respectively. Additional unadjusted analyses of outcomes (TTNT, PFS, and OS) stratified by the top four first-line induction regimens (VRd, Vd, Rd, and VCd) are provided in the Appendix S1.

In the adjusted analyses, conducted using MSM to account for survival bias due to the time-dependent nature of DOT, and patient and disease characteristics, TTNT, PFS, and OS at 2 years were improved with longer 1 L DOT (Table 3). Each additional month on 1LT was associated with reduced risk of initiating the next LOT or death at 2 years (OR: 0.76 [95% CI: 0.75, 0.78; p < 0.0001]). Similarly, each additional month on 1LT was associated with reduced risk of disease progression at 2 years (OR: 0.87 [95% CI: 0.86, 0.88; p < 0.0001]) and death at 2 years (OR: 0.72 [95% CI: 0.70, 0.74; p < 0.0001]).

4.1 Limitations

This study has limitations that are typical of retrospective analyses. First, conceptually, there is no standard definition of dose-attenuated therapy, limiting the ease of identifying these patients retrospectively; patients identified as receiving dose-attenuated therapy in this study were verified by an expert MM oncologist with review of patient charts to ascertain the intention of drug or dose changes. Next, the sample size of patients with non-SCT NDMM receiving dose-attenuated therapy in 1LT was relatively small. Clinical characteristics such as cytogenetic abnormalities, ISS stage, and ECOG PS were impacted by missing data, limited or inconsistent testing and documentations across clinical sites. The reasons for not pursuing SCT were not collected during the medical chart review, however, other factors which determine a patient's transplant eligibility were collected and reported. Patients with the shortest DOT more commonly had an ECOG PS of 2 or greater, had high-risk cytogenetics, and had higher CCI and modified frailty index scores of 2+, compared to patients with a longer DOT; however, these were adjusted for in the statistical analyses. Additionally, progression was determined based on physicians' documentation in the medical records; however, the criteria used to define progression by individual site physicians cannot be ascertained. Interestingly, our study observed a slightly shorter 1LT median TTNT compared to median PFS, which can be a reflection of variability in routine care practices or even chart documentation conventions; future studies should consider examining further both TTNT and PFS in other real-world healthcare settings. Dosing was not captured in this study which would allow for cumulative dose intensity to be determined and may have served as a better correlate of outcomes as compared to DOT. In order to clinically enrich the patient data and verify accurate outcomes of interest not captured in structured data, a retrospective chart review was conducted on a random sample of patients from among NDMM without front-line SCT patients that were identified in the ION database. Due to the resource-intensive nature of chart review, this study included 300 randomly selected patients; thus, future research should consider other larger samples from routine care setting to confirm these findings. Finally, this research focused on community oncology clinics, so future studies should consider other oncology care settings.