We used the newly available linkage of the SEER registry (reporting incident cancers from 18 geographic areas covering 34.6% of the United States population) with Medicare claims and MDS; MDS includes direct clinical assessments of beneficiaries receiving care in NHs.²⁴ We selected patients diagnosed with DLBCL 2011–2015, aged ≥65, with continuous fee-for-service Medicare coverage from 1 year prior to lymphoma diagnosis, until ≥6 months after, 1 month before death, or December 2016, whichever occurred first (Figure S1 and Table S1). We chose 6 months as the minimum observed coverage because all claims-based outcomes of interest in this study (except all-cause mortality, recorded separately) would be observed within this timeframe. We excluded beneficiaries with central nervous system lymphoma, and those started on chemoimmunotherapy >120 days after diagnosis. Patients with DLBCL typically receive treatment shortly after the diagnosis; so the threshold of 120 days minimized potential confounding from initiation of lymphoma-directed therapy for another lymphoma diagnosis.¹⁹ We then identified beneficiaries who had an MDS assessment²⁶ between −120 and +30 days of DLBCL diagnosis (and before treatment), a definition of NH patients used in previous studies.^19,27 We also variedthe timeframe of MDS assessment by using −90 to +30 and −60 to +30 days, respectively, to examine the robustness of our results in the sensitivity analysis. For beneficiaries with several MDS assessments, we selected the one closest to the time of diagnosis. Beneficiaries without matching MDS assessments were classified as community-dwelling. The study was reviewed by the Brown University Institutional Review Board and was exempt from the regulations regarding the inclusion of human participants in research.

An MDS assessment is routinely performed at the time of NH admission and then every 3 months, both for patients receiving post-acute care in the NH and for long-stay residents. Reporting of the assessment results to Medicare is mandatory.²⁴ The assessment covers multiple geriatric domains, including physical and cognitive function. We used Morris activities of daily living (ADL) scale to quantify functional disability in each of the seven activities (bed mobility, transfer, locomotion on unit, dressing, eating, toilet use, and personal hygiene, Table S2).²⁷ We defined “dependency” for each ADL if the item was scored 3 or 4 on the self-performance scale, or if the ADL activity occurred ≤ twice over the seven-day period of assessment. Based on the count of ADLs with dependency, we classified the overall functional disability as none (0 ADL), moderate (1–4 ADLs), or severe (5–7ADLs).²⁷

To characterize cognition, we used Cognitive Function Scale (CFS),²⁸ constructed based on Brief Interview for Mental Status and Cognitive Performance Scale.²⁹ We classified cognitive impairment as intact, mild, or moderate to severe, based on a well-validated criterion²⁸ (Table S3).

Among all patients with DLBCL, we examined the receipt of chemoimmunotherapy (Table S1). Chemoimmunotherapy regimens were identified only among those receiving treatment in the outpatient setting, because specific antineoplastic agents are not available from inpatient Medicare claims.^30,31 Regimens that included anthracycline and comprised ≥4 antineoplastic agents (excluding corticosteroids) were classified as “multi-agent, anthracycline-based” or “intensive”. Those without anthracycline or containing <4 agents were considered as “attenuated” (Table S4). To generate a complete list of codes for chemoimmunotherapy drugs for DLBCL, we referenced multiple sources of drug codes (e.g., Centers for Medicare & Medicaid Services, American Academy of Professional Coders) and compared our codes with published studies using Medicare data to examine treatment patterns of B-cell lymphoma^19,30,32; three lymphoma clinicians (MD, AJO, SFH) experienced in using Medicare claims to study lymphoma treatment also reviewed these codes. To explore potential clinical scenarios where patients did not receive lymphoma treatment, we examined transitioning to hospice within 60 days of diagnosis or inpatient death without recorded chemoimmunotherapy.

Among patients receiving chemoimmunotherapy, we examined outcomes including all-cause mortality and hospitalization within 30 days from the start of treatment.^30,33 These early adverse outcomes reflect immediate poor patient outcomes related to treatment selection, decompensated lymphoma, or other baseline health conditions. We also examined overall survival (OS) among patients receiving treatment in the outpatient setting; we estimated OS from treatment initiation. We used OS over cause-specific survival because it is unequivocal, avoiding potentially inaccurate cause-of-death specification.³⁴

We used descriptive statistics to evaluate baseline characteristics, reporting medians with interquartile ranges (IQR) for continuous variables and counts with percentages for categorical variables.

We compared each outcome between older adults with DLBCL who had a NH stay and those who did not. We also compared outcomes of patients with: (1) moderate versus no disability, (2) severe versus no disability, (3) mild cognitive impairment versus intact cognition, and (4) moderate to severe cognitive impairment versus intact cognition.

We used multivariable Cox regression model to analyze OS, reporting hazard ratio (HR) with 95% confidence interval (CI). For all categorical outcomes (receipt of any chemoimmunotherapy, receipt of intensive regimens, transition to hospice, inpatient death before recorded therapy, 30-day mortality, and 30-day hospitalization), we used multivariable logistic regression models, reporting odds ratios (OR) and 95% CI. All models were adjusted for age, sex, race, DLBCL stage, comorbidity burden, Medicaid dual coverage, and type of NH stay (for models within the NH population; long- vs. short-term stay; long-term stay defined as >90 consecutive days prior to the diagnosis²⁷). In the Cox regression models, we additionally adjusted for type of treatment regimens (intensive vs. attenuated as defined above). We estimated comorbidity burden based on the National Cancer Institute-Charlson comorbidity index, using diagnoses derived from the patient’s Medicare claims within 1 year prior to the lymphoma diagnosis. We categorized comorbidity burden as low (index score ≤2), moderate,^3,4 high (≥5), or unknown.^19,35 All covariates were categorical in the regression models to avoid the assumption of linear association with outcomes.

In the sensitivity analyses, we performed the propensity score matching of community-dwelling and NH patients (ratio: 1:1, greedy nearest neighbor with caliper equal to 0.2 standard deviation of the propensity score), based on pre-defined age subgroup (≥65 to <75; ≥75 to <80; ≥80 to <85; or ≥ 85), sex, race, DLBCL stage (I/II, III/IV, unknown), comorbidity index, and Medicaid dual coverage. In the matched population, we compared the receipt of chemoimmunotherapy, early transition to hospice, and inpatient death without recorded lymphoma therapy. We replaced any missing data with “unknown” category (only stage and comorbidities had missingness in approximately 5% patients) and did not exclude any patients due to missingness. Outcome models used conditional logistic regression and reported OR and 95% CI. Given the anticipated small sample sizes in the NH sub-populations defined respectively for the other outcomes (e.g., overall survival and receipt of intensive regimens for patients receiving treatment in the outpatient setting), the performance of regression would most likely be superior to that of matching,^36,37 therefore, we only conducted multivariable regression for all the other outcomes.

To better characterize the experience of older adults with DLBCL in NHs, we used multivariable logistic regression models to evaluate all categorical outcomes according to degree of functional and cognitive impairments at the time of diagnosis. Among patients who received lymphoma-directed therapy in the outpatient setting, we compared OS according to the degree of impairment using multivariable Cox regression models, including the type of chemotherapy regimen as an additional covariate. We also compared the outcomes of interest between patients with short- and long-stay using similar methods.

All analyses were conducted using R v.3.5.1 (https://www.R-project.org/). p-values were two-tailed with a type I error rate α = 0.05.

Compared with community-dwelling adults with DLBCL, NH patients were less likely to receive chemoimmunotherapy (45% vs. 74%; adjusted OR, 0.34; 95%CI, 0.29–0.41); when treated, they were less likely to receive multi-agent, anthracycline-based regimens (47% vs. 71%; adjusted OR, 0.51; 95%CI, 0.37–0.72). NH patients were more likely to transition to hospice within 60 days of lymphoma diagnosis (adjusted OR, 2.69; 95%CI, 2.18–3.32) or die in the hospital without recorded treatment (adjusted OR, 2.97; 95% CI, 2.36–3.73). Among chemoimmunotherapy recipients, NH patients had higher rates of 30-day mortality (adjusted OR, 2.00; 95%CI, 1.43–2.78) and hospitalization (adjusted OR, 1.51; 95%CI, 1.18–1.93) than community-dwelling individuals (Figure 1, Figure S2). Among those receiving chemoimmunotherapy in the outpatient setting, NH patients had a shorter median OS than community-dwelling individuals (24.1 vs. 54.7 months, adjusted HR: 1.36, 95% CI: 1.11–1.65) (Figure 2). Results of these comparisons in the matched cohort (total N = 1290) were very similar to those from the main analyses shown in Figure 1 (Table S6).

Twenty percent of NH patients with DLBCL had moderate disability, while 61% had severe disability. Mild and moderate to severe cognitive impairment was present in 18% and 26% patients respectively (Table S5). Compared with patients with intact function, those with severe disability were less likely to receive chemoimmunotherapy (adjusted OR: 0.57, 95% CI: 0.38–0.87). The distributions of disability for each ADL item among chemoimmunotherapy recipients and non-recipients are shown in Figure S3. Those with severe disability (vs. no disability) were also more likely to transition to hospice shortly after lymphoma diagnosis (adjusted OR: 2.00, 95% CI: 1.15–3.47). Among chemoimmunotherapy recipients, patients with severe disability (vs. no disability) had a higher rate of 30-day hospitalization (adjusted OR: 2.02, 95% CI: 1.10–3.68). There were no statistically significant associations between disability and other categorical outcomes (including receipt of anthracycline-based, multi-agent regimens, inpatient death without recorded treatment, and 30-day mortality) (Table 2).

Compared with those with intact cognition, patients with any cognitive impairment were less likely to receive chemoimmunotherapy (adjusted OR for mild cognitive impairment vs. intact cognition: 0.64, 95% CI: 0.44–0.95; adjusted OR for moderate to severe impairment: 0.31, 95% CI: 0.19–0.50), and more likely to transition to hospice within 60 days of lymphoma diagnosis (adjusted OR for mild impairment: 1.65, 95% CI: 1.06–2.56; adjusted OR for moderate to severe impairment: 1.80, 95% CI: 1.09–2.97). Patients with moderate to severe cognitive impairment (vs. intact cognition) were more likely to die in the hospital without recorded chemoimmunotherapy (adjusted OR: 3.29, 95% CI: 1.95–5.56). Those with mild cognitive impairment (vs. intact cognition) had a higher rate of 30-day hospitalization after therapy (adjusted OR: 2.41, 95% CI: 1.28–4.52). There was no statistically significant association between cognitive impairment and other categorical outcomes (including receipt of anthracycline-based, multi-agent regimen, and 30-day mortality) (Table 2). The associations between receipt of chemoimmunotherapy and functional or cognitive impairment were largely unchanged using different time windows to define pre-treatment MDS assessment (e.g.,−60 to +30 days, or − 90 to +30 days of lymphoma diagnosis; Table S7).

In multivariable Cox models, the association between levels of functional or cognitive impairment and OS was not statistically significant among patients who received lymphoma-directed therapy (Table 3).

Patients with short- and long-stay had similar rates of functional and cognitive impairments, and probabilities of receiving treatment (Table S5). Median OS was lower among long-term residents (19.7 vs. 29.2 months). There was no statistically significant difference in any outcomes of interest between short- and long-stay residents (Table S8).