Abstract
Newly diagnosed patients with non-Hodgkin lymphoma (NHL) often have a history of other diseases, and these comorbidities can impact patient treatment and management options, as well as overall survival (OS). We developed the Lymphoma Epidemiology of Outcomes (LEO) comorbidity index (LCI) as a sum of 10 comorbidities adapted from the Self-Report-Generated Charlson Comorbidity Index (SRG-CCI) for use in the LEO cohort, a national prospective study of newly diagnosed NHL. Of the 5,502 participants with self-reported comorbidity data, 3,107 (56.4%) were male and the mean age at diagnosis was 60.9 years (range, 18–99 years). The LCI ranged from 0–6, with 48.6% having 0, 30.2% having 1, 21.2% having 2 or more comorbidities. With a median follow-up of 62.4 months among surviving participants, 2,099 patients had an event and 1,219 died. Continuous LCI similarly predicted both 1-year mortality (c-statistic=0.654) and OS (c-statistic=0.655), while it showed a weaker but still statistically significant predictive ability for lymphoma-specific (c-statistics=0.617) and event-free (c-statistic=0.574) survival. Participants with 1 (HR=1.21, 95% CI 1.05–1.39) and 2+ (HR=1.80, 95% CI 1.56–2.08) comorbidities had inferior OS compared to those with no comorbidities after adjustment for age and sex (c-statistic=0.654), and performance strengthened after adjustment for the International Prognostic Index (c-statistic=0.672). LCI predicted OS most strongly in marginal zone (c-statistics=0.748) and weakest in T-cell (c-statistic=0.579) lymphoma. The cumulative incidence of death due to lymphoma, lymphoma treatment, and other causes all increased with increasing comorbidities, with the greatest increase observed for death due to other causes. The LCI performs comparable to other published comorbidity indices, supporting its use in the LEO cohort to better model real-world outcomes and more generally providing an approach to implementing comorbidity indices in cancer survivorship cohorts.
Keywords: non-Hodgkin lymphoma, comorbidity, overall survival, event-free survival
1 |. Introduction
In the United States in 2025, an estimated 80,350 new cases of non-Hodgkin lymphoma (NHL) will be diagnosed and 19,390 will die of this malignancy (1). While the 5-year relative survival of NHL in the US has increased from 47% in 1975 to 74% in 2019 (2) there is still substantial mortality. Further, given that more than half of the patients are diagnosed with NHL at the age of 65 years or older (2), they often have a history of prior diseases. These comorbidities can impact patient treatment and management options, increase risk of hospitalization, reduce quality of life, increase health care utilization/costs, and decrease overall survival (OS) (3–5). Furthermore, patients with significant comorbidities are often excluded from randomized clinical trials, highlighting the importance of using real-world data to more fully understand the impact of comorbidities on NHL outcomes (6).
Multiple indices have been developed to measure the impact of comorbidities using a variety of approaches, including counts of individual conditions, organ or system-based approaches, weighted indices, case-mix approaches, and overall measures of physical health status (3, 7, 8). The data required for these indices have been obtained from administrative data, medical records, physician report and/or patient self-report, and they have been utilized in hospitalized patients, outpatients, and the general population. There is, however, no universally accepted gold standard for measuring comorbidities in cancer patients, and a preferred index for use in a specific study is likely to be the one that has been developed and validated in similar populations, taking into account the types, availability and robustness of data needed to construct the index (3).
The Lymphoma Epidemiology of Outcomes (LEO) cohort study prospectively enrolls and follows newly diagnosed patients with NHL. For this report, our aim was to adapt a validated comorbidity index for general use in the LEO cohort, including describing the burden of morbidity in our study population, assessing impact on outcomes, using it as a potential confounding or mediating factor, and conducting risk adjustment in downstream analyses. After reviewing available indices, we selected the Self-Report-Generated Charlson Comorbidity Index (SRG-CCI) (9) with the primary objective of adapting it for use in participants in the LEO cohort as the LEO comorbidity index (LCI) and assessing its performance in predicting 1-year mortality. As the SRG-CCI was developed in hospitalized patients, and most of the LEO participants were recruited as outpatients, in secondary analyses we also assessed the predictive ability of the LCI on continuous OS, lymphoma-specific survival (LSS) and event-free survival (EFS), which are more commonly modeled outcomes in lymphoma prognosis studies.
1 |. Material and Methods
2.1 |. Study Population
LEO is a prospective cohort study of newly diagnosed patients with NHL recruited from 8 academic centers in the U.S. (see Supplemental Methods). Eligibility criteria were age 18 years or older, newly diagnosed (within 6 months of diagnosis at consent), and able to provide written informed consent in English or Spanish. From 7/1/2015 through 5/31/2020, 7,735 participants were enrolled. All participants were contacted every six months for the first three years after initial diagnosis and then annually thereafter until death or disenrollment in order to update their disease status, new medical conditions, and patient reported outcomes. All deaths were reviewed by a LEO clinician and cause of death was assigned according to the LEO protocol. Full details on the LEO cohort have been previously published (10).
All participants were asked to complete a self-administered Enrollment Questionnaire (EQ) in English or Spanish, which was provided in either printed or digital form. In addition to demographic and anthropometric information in the EQ, participants were asked about selected medical conditions. While medical records were available to abstract comorbidity data, many participants did not have extensive local records preceding their enrollment into the LEO cohort and abstracting the comorbidities locally and from outside records presented a prohibitive burden. We also did not attempt to access administrative (claims) data for LEO participants given the potential variability in administrative coding across the eight centers.
2.2 |. Selection of the SRG-CCI
Due to having collected self-reported comorbidity data on the LEO EQ, we prioritized comorbidity measures based on self-report. After considering multiple available self-reported indices (8) (including the Comorbidity Symptom Scale (11), the Seattle Index of Comorbidity (12), the Self-Administered Comorbidity Questionnaire (13), the Disease Burden Morbidity Assessment (14), and a comorbidity index from the Health and Retirement Survey (15), we selected the Self-Report-Generated Charlson Comorbidity Index (SRG-CCI) (9) as this had the highest concordance with the available data in LEO and was practical to adapt to LEO. It was built from the original CCI (16) and an interviewer/self-administered version was developed and validated in over 7,700 hospitalized patients, where it performed well against administrative data in predicting 1-year mortality and health care utilization (17).
2.3 |. Adaptation of the SRG-CCI to the LEO Cohort
The Supplemental Methods and Supplemental Table 1 provides the details of how the SRG-CCI was adapted to the LEO cohort by providing the mapping of the SRG-CCI items to available LEO data. For 8 of 10 items on the SRG-CCI, we used self-reported data from the LEO EQ to assign the presence of a comorbidity. For HIV/AIDs, we did not have self-reported data, but it was abstracted from medical records as part of LEO clinical data collection. We also did not have a self-reported item for Kidney Disease, but we were able to ascertain it via the serum creatinine level abstracted from the medical record (also collected as part of LEO clinical data). We then used a published sex and age-specific creatinine cutpoint (18) to define kidney disease. The prevalence and missing data for each comorbidity is provided in Supplemental Table 2.
Of 7,735 individuals enrolled into the LEO cohort, 11 participants were subsequently found to be ineligible after enrollment (see Supplemental Figure 1 for CONSORT diagram). We excluded 2,131 (27.6% of 7724) participants who did not complete the EQ and thus had no self-reported comorbidity data, as well as a further 91 participants with missing data on two or more comorbidity items on the EQ (details in Supplemental Methods). When we compared LEO participants included in the analysis (n=5,502) to those excluded (n=2,131), we found no large or clinically meaningful differences between the groups in age at diagnosis, gender, race/ethnicity, NHL subtype, stage, ECOG performance status, B-symptoms, LDH status, or treatment group (Supplemental Table 3).
2.4 |. Outcomes
As the SRG-CCI (and the original CCI) was developed to predict 1-year mortality in hospitalized patients, we used this as our primary outcome, defined as vital status at one year after date of diagnosis. Secondary endpoints were OS, defined as time from date of diagnosis to date of death due to any cause or date of last follow-up; LSS, defined as time from date of diagnosis to death due to lymphoma or treatment-related toxicity; and EFS, defined as time from diagnosis to disease progression/relapse, re-treatment, or death due to any cause.
2.5 |. Data Analysis
We modeled the association for each of the 10 comorbidities with vital status at one year in age and sex adjusted logistic regression models (excluding those with incomplete follow-up at 1 year), and then we included all items in a single multivariable model adjusted for age and sex. We next modeled the sum of comorbidities as both a continuous and categorical (grouped as 0, 1, 2+) variable, and we used splines to evaluate the functional form of the association with vital status at 1 year. For secondary endpoints of OS, LSS and EFS, we used a similar modeling approach using Cox regression, adjusting for age and sex. We also assessed the models by NHL subtype and for several patient subgroups defined on age, International Prognostic Index (IPI) (19), and treatment. We assessed model performance using Harrell’s c-statistic. Cumulative incidence for competing risks of cause of death (COD), defined as due to lymphoma, lymphoma treatment-related, other causes, and unknown/missing, and tests of equality between LCI groups were calculated using the cuminc function from the cmprsk package in R (20).
3 |. Results
3.1 |. Analytic Cohort
Of the 5,502 participants in the analysis, 3,103 (56.4%) were male, the mean age at diagnosis was 60.9 years (SD 14.0), and 2,419 (44.0%) were aged 65+ years. At a median follow-up of 62.4 months (interquartile range, 49.1–80.9) among surviving participants, 2,099 had an event and 1,219 died.
After summing the comorbidities as the LEO Comorbidity Index (LCI), there was a range of 0–6 comorbidities, with 2,674 (48.6%) having 0, 1662 (30.2%) having 1, 786 (14.3%) having 2, 285 (5.2%) having 3, 72 (1.3%) having 4, 23 (0.4%) having 5+ comorbidities. For analysis, we grouped the LCI into 0 (n=2674, 48.6%), 1 (n=1662, 30.2%) and 2+ (n=1166, 21.2%) comorbidities. As shown in Table 1, increasing number of comorbidities was strongly associated with older age, poorer ECOG performance status, and higher LDH, with a modest inverse association with use of frontline systemic chemotherapy (73.2% for zero vs. 65.2% for 2+ comorbidities). Relative to no comorbidities, participants with 2+ comorbidities were slightly enriched for MZL, TCL and other NHL subtypes. There were no associations by gender, race/ethnicity, stage, or B-symptoms.
TABLE 1.
Association of number of comorbidities with demographic and clinical characteristics, LEO Cohort, 2015–2020.
| Number of comorbidities |
|||||||||
|---|---|---|---|---|---|---|---|---|---|
| Zero |
One |
Two+ |
|||||||
| N=5502 | % | N=2674 | % | N=1662 | % | N=1166 | % | p-value | |
|
| |||||||||
| Age at diagnosis | <0.001 | ||||||||
| Mean (SD) | 60.9 (14.0) | 57.1 (14.1) | 62.8 (13.0) | 66.9 (12.5) | |||||
| Range | 18 – 99 | 18 – 99 | 18 – 98 | 20 – 95 | |||||
| Age group | <0.001 | ||||||||
| <65 | 3083 | 56.0 | 1815 | 67.9 | 820 | 49.3 | 448 | 38.4 | |
| 65+ | 2419 | 44.0 | 859 | 32.1 | 842 | 50.7 | 718 | 61.6 | |
| Gender | 0.70 | ||||||||
| Male | 3103 | 56.4 | 1493 | 55.8 | 949 | 57.1 | 661 | 56.7 | |
| Female | 2399 | 43.6 | 1181 | 44.2 | 713 | 42.9 | 505 | 43.3 | |
| Race/ethnicity* | 0.22 | ||||||||
| Am Indian/AK Native | 20 | 0.4 | 12 | 0.5 | 4 | 0.2 | 4 | 0.4 | |
| Asian | 102 | 1.9 | 60 | 2.3 | 26 | 1.6 | 16 | 1.4 | |
| Black/AA | 323 | 6.1 | 135 | 5.3 | 104 | 6.4 | 84 | 7.4 | |
| Hispanic (all races) | 490 | 9.2 | 230 | 8.9 | 153 | 9.4 | 107 | 9.4 | |
| Native Hawaiian/PI | 7 | 0.1 | 4 | 0.2 | 2 | 0.1 | 1 | 0.1 | |
| White | 4348 | 81.6 | 2107 | 82.0 | 1322 | 81.5 | 919 | 80.9 | |
| Unknown | 40 | 0.8 | 23 | 0.9 | 12 | 0.7 | 5 | 0.4 | |
| Missing | 172 | 103 | 39 | 30 | |||||
| NHL subtype | <0.001 | ||||||||
| DLBCL | 1881 | 34.2 | 898 | 33.6 | 587 | 35.3 | 396 | 34.0 | |
| FL | 1190 | 21.6 | 643 | 24.0 | 334 | 20.1 | 213 | 18.3 | |
| MCL | 530 | 9.6 | 272 | 10.2 | 167 | 10.0 | 91 | 7.8 | |
| MZL | 522 | 9.5 | 213 | 8.0 | 175 | 10.5 | 134 | 11.5 | |
| TCL | 461 | 8.4 | 224 | 8.4 | 122 | 7.3 | 115 | 9.9 | |
| Other NHL | 797 | 14.5 | 369 | 13.8 | 240 | 14.4 | 188 | 16.1 | |
| Composites | 121 | 2.2 | 55 | 2.1 | 37 | 2.2 | 29 | 2.5 | |
| Stage group | 0.76 | ||||||||
| I-II | 1869 | 36.0 | 910 | 35.8 | 575 | 36.7 | 384 | 35.5 | |
| III-IV | 3323 | 64.0 | 1633 | 64.2 | 991 | 63.3 | 699 | 64.5 | |
| Missing | 310 | 131 | 96 | 83 | |||||
| ECOG PS Group | <0.001 | ||||||||
| <2 | 4630 | 88.9 | 2342 | 92.4 | 1390 | 88.4 | 898 | 81.7 | |
| ≥2 | 577 | 11.1 | 193 | 7.6 | 183 | 11.6 | 201 | 18.3 | |
| Missing | 295 | 139 | 89 | 67 | |||||
| B-symptoms | 0.063 | ||||||||
| No | 3986 | 72.4 | 1964 | 73.4 | 1217 | 73.2 | 805 | 69.0 | |
| Yes | 1380 | 25.1 | 644 | 24.1 | 405 | 24.4 | 331 | 28.4 | |
| Unconfirmed | 136 | 2.5 | 40 | 2.5 | 40 | 2.4 | 30 | 2.6 | |
| LDH group | 0.005 | ||||||||
| ≤Normal | 2851 | 61.7 | 1419 | 63.6 | 857 | 61.6 | 575 | 57.6 | |
| >Normal | 1770 | 38.3 | 811 | 36.4 | 535 | 38.4 | 424 | 42.4 | |
| Missing | 881 | 444 | 270 | 167 | |||||
| Frontline treatment group | <0.001 | ||||||||
| Observed | 803 | 14.6 | 380 | 14.2 | 255 | 15.3 | 168 | 14.4 | |
| R-monotherapy | 385 | 7.0 | 147 | 5.5 | 118 | 7.1 | 120 | 10.3 | |
| Systemic chemo | 3858 | 70.1 | 1943 | 72.7 | 1150 | 69.2 | 765 | 65.6 | |
| Other | 456 | 8.3 | 204 | 7.6 | 139 | 8.4 | 113 | 9.7 | |
Race categories exclude Hispanic ethnicity
3.2 |. Prediction of 1-year Mortality
Of the 10 comorbidities in the LCI, kidney disease, HIV/AIDS, heart disease, diabetes, and respiratory disease were statistically significantly associated with 1-year mortality in age and sex adjusted logistic regression models (Table 2). In a multivariable model including all 10 comorbidities, only diabetes, heart disease, HIV/AIDS, and kidney disease remained statistically significant. While not statistically significant, ORs were also elevated for respiratory disease (OR=1.33), and stroke (OR=1.36). The LCI modeled as a continuous variable was associated with an 35% greater risk of 1-year mortality for each increase of one comorbidity (OR=1.35, 95% CI 1.22–1.48) adjusted for age and sex (c-statistic=0.654). For the categorical model, compared to participants with no comorbidities at diagnosis, those with 1 (OR=1.70, 95% CI 1.30–2.21) and 2+ (OR=2.36, 95% CI 1.80–3.10) comorbidities had higher risk of 1-year mortality (c-statistic=0.654).
TABLE 2.
Association of comorbidities from the SSG-CCI with 1-year mortality in the Lymphoma Epidemiology of Outcomes (LEO) cohort.
| Single Comorbidity Models |
Multivariable Comorbidity Model |
||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Comorbidity* | N | %Dead | OR† | 95% CI | P-value | OR¶ | 95% CI | P-value | |||
|
| |||||||||||
| Respiratory disease | 0.032 | 0.081 | |||||||||
| No | 4945 | 6.6 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 557 | 9.3 | 1.40 | 1.02–1.90 | 1.34 | 0.96–1.89 | |||||
| Autoimmune disease | 0.21 | 0.17 | |||||||||
| No | 5026 | 7.0 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 475 | 5.1 | 0.76 | 0.48–1.14 | 0.72 | 0.45–1.16 | |||||
| Other cancer | 0.42 | 0.87 | |||||||||
| No | 5272 | 6.8 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 230 | 6.5 | 0.80 | 0.45–1.33 | 0.95 | 0.55–1.66 | |||||
| Diabetes | <0.001 | 0.0079 | |||||||||
| No | 4765 | 6.2 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 725 | 11.0 | 1.59 | 1.22–2.07 | 1.49 | 1.11–1.99 | |||||
| Digestive diseases | 0.94 | 0.96 | |||||||||
| No | 4989 | 6.7 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 489 | 7.2 | 1.01 | 0.69–1.44 | 0.99 | 0.66–1.47 | |||||
| Heart disease | <0.001 | 0.0039 | |||||||||
| No | 4509 | 5.7 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 993 | 11.8 | 1.70 | 1.33–2.17 | 1.49 | 1.14–1.96 | |||||
| HIV illness or AIDS | <0.001 | <0.001 | |||||||||
| No | 5426 | 6.6 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 76 | 21.1 | 6.22 | 3.34–11.1 | 6.75 | 3.67–12.40 | |||||
| Kidney disease | <0.001 | <0.001 | |||||||||
| No | 4100 | 5.6 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 696 | 14.1 | 2.16 | 1.64–2.81 | 2.02 | 1.54–2.67 | |||||
| Liver disease | 0.19 | 0.57 | |||||||||
| No | 5405 | 6.8 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 97 | 10.3 | 1.56 | 0.75–2.90 | 0.76 | 0.30–1.94 | |||||
| Stroke | 0.16 | 0.27 | |||||||||
| No | 5328 | 6.7 | 1.00 | reference | 1.00 | reference | |||||
| Yes | 155 | 11.6 | 1.44 | 0.84–2.33 | 1.36 | 0.79–2.35 | |||||
See methods for definitions of these comorbidities as implemented in the LEO cohort.
Each comorbidity was adjusted for age and sex.
Simultaneous adjustment for each comorbidity and age and sex.
3.3 |. Prediction of OS, LSS and EFS
Of the 10 comorbidities, respiratory disease, diabetes, heart disease, HIV/AIDS, kidney disease, liver disease, and stroke were each associated with OS after adjustment for age and sex in Cox models, while in a multivariable model that simultaneously adjusted for all variables, only liver disease was no longer statistically significant (Supplemental Table 4).
The LCI showed a linear association with OS (Supplemental Figure 2), and when modeled as a continuous variable it was associated with an 26% greater risk of death for each increase of one comorbidity (HR=1.26, 95% CI 1.19–1.32) adjusted for age and sex (c-statistic=0.655) (Table 3). Increasing categorical LCI score was associated with inferior OS as shown in a KM analysis (Figure 1; Log rank P<0.0001). Compared to participants with no comorbidities at diagnosis, those with 1 (HR=1.21, 95% CI 1.05–1.39) and 2+ (OR=1.80, 95% CI 1.56–2.08) comorbidities had inferior OS after adjusting for age and sex (c-statistic=0.654).
TABLE 3.
Association of the LEO Comorbidity Index (continuous and categorical) with overall, lymphoma-specific and event-free survival for NHL, selected subgroups and NHL subtypes.
| Overall Survival |
Lymphoma-Specific Survival |
Event-Free Survival |
||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HR* | 95% CI | c-stat | HR* | 95% CI | c-stat | HR* | 95% CI | c-stat | ||||||||||
|
| ||||||||||||||||||
| All NHL | Continuous | 1.26 | 1.19 | - | 1.32 | 0.655 | 1.18 | 1.10 | - | 1.26 | 0.617 | 1.13 | 1.08 | - | 1.18 | 0.574 | ||
| One vs No Comorbidities | 1.21 | 1.05 | - | 1.39 | 0.654 | 1.11 | 0.93 | - | 1.32 | 0.619 | 1.08 | 0.97 | - | 1.19 | 0.573 | |||
| Two vs No Comorbidities | 1.80 | 1.56 | - | 2.08 | 1.58 | 1.32 | - | 1.88 | 1.34 | 1.20 | - | 1.50 | ||||||
| IC-treated | Continuous | 1.24 | 1.15 | - | 1.34 | 0.655 | 1.17 | 1.06 | - | 1.29 | 0.623 | 1.11 | 1.03 | - | 1.18 | 0.579 | ||
| One vs No Comorbidities | 1.38 | 1.12 | - | 1.71 | 0.654 | 1.37 | 1.07 | - | 1.76 | 0.625 | 1.19 | 1.01 | - | 1.41 | 0.579 | |||
| Two vs No Comorbidities | 1.80 | 1.44 | - | 2.25 | 1.60 | 1.22 | - | 2.10 | 1.31 | 1.09 | - | 1.57 | ||||||
| Not IC-treated | Continuous | 1.27 | 1.19 | - | 1.36 | 0.657 | 1.18 | 1.08 | - | 1.30 | 0.619 | 1.15 | 1.09 | - | 1.21 | 0.571 | ||
| One vs No Comorbidities | 1.07 | 0.89 | - | 1.29 | 0.658 | 0.90 | 0.71 | - | 1.15 | 0.625 | 1.01 | 0.88 | - | 1.15 | 0.571 | |||
| Two vs No Comorbidities | 1.80 | 1.50 | - | 2.17 | 1.56 | 1.23 | - | 1.97 | 1.36 | 1.19 | - | 1.57 | ||||||
| On a clinical trial | Continuous | 1.27 | 1.01 | - | 1.58 | 0.649 | 1.39 | 1.08 | - | 1.79 | 0.663 | 1.12 | 0.95 | - | 1.32 | 0.571 | ||
| One vs No Comorbidities | 1.04 | 0.63 | - | 1.71 | 0.649 | 1.06 | 0.58 | - | 1.96 | 0.663 | 1.15 | 0.82 | - | 1.59 | 0.571 | |||
| Two vs No Comorbidities | 1.86 | 1.06 | - | 3.27 | 2.29 | 1.20 | - | 4.38 | 1.32 | 0.85 | - | 2.06 | ||||||
| Not on a clinical trial | Continuous | 1.25 | 1.18 | - | 1.31 | 0.655 | 1.15 | 1.08 | - | 1.24 | 0.614 | 1.13 | 1.08 | - | 1.18 | 0.576 | ||
| One vs No Comorbidities | 1.21 | 1.05 | - | 1.40 | 0.654 | 1.10 | 0.92 | - | 1.32 | 0.616 | 1.07 | 0.96 | - | 1.19 | 0.575 | |||
| Two vs No Comorbidities | 1.77 | 1.52 | - | 2.05 | 1.51 | 1.26 | - | 1.82 | 1.33 | 1.19 | - | 1.50 | ||||||
| No HIV | Continuous | 1.24 | 1.18 | - | 1.31 | 0.657 | 1.16 | 1.08 | - | 1.24 | 0.618 | 1.12 | 1.08 | - | 1.17 | 0.575 | ||
| One vs No Comorbidities | 1.18 | 1.03 | - | 1.36 | 0.656 | 1.08 | 0.91 | - | 1.29 | 0.619 | 1.07 | 0.96 | - | 1.19 | 0.574 | |||
| Two vs No Comorbidities | 1.73 | 1.50 | - | 2.00 | 1.50 | 1.26 | - | 1.80 | 1.32 | 1.18 | - | 1.48 | ||||||
| DLBCL | Continuous | 1.19 | 1.09 | - | 1.29 | 0.669 | 1.11 | 1.00 | - | 1.23 | 0.647 | 1.11 | 1.03 | - | 1.19 | 0.597 | ||
| One vs No Comorbidities | 1.34 | 1.07 | - | 1.67 | 0.667 | 1.25 | 0.96 | - | 1.64 | 0.647 | 1.26 | 1.05 | - | 1.52 | 0.597 | |||
| Two vs No Comorbidities | 1.61 | 1.27 | - | 2.04 | 1.39 | 1.04 | - | 1.86 | 1.33 | 1.09 | - | 1.64 | ||||||
| FL | Continuous | 1.33 | 1.14 | - | 1.55 | 0.723 | 1.30 | 1.05 | - | 1.62 | 0.696 | 1.10 | 0.99 | - | 1.22 | 0.556 | ||
| One vs No Comorbidities | 1.13 | 0.72 | - | 1.76 | 0.722 | 0.98 | 0.53 | - | 1.82 | 0.703 | 1.02 | 0.81 | - | 1.28 | 0.555 | |||
| Two vs No Comorbidities | 1.95 | 1.27 | - | 3.01 | 2.03 | 1.14 | - | 3.61 | 1.20 | 0.93 | - | 1.56 | ||||||
| MCL | Continuous | 1.30 | 1.09 | - | 1.55 | 0.624 | 1.28 | 1.03 | - | 1.59 | 0.616 | 1.17 | 1.02 | - | 1.35 | 0.559 | ||
| One vs No Comorbidities | 0.84 | 0.56 | - | 1.26 | 0.638 | 0.78 | 0.47 | - | 1.28 | 0.633 | 0.91 | 0.68 | - | 1.22 | 0.566 | |||
| Two vs No Comorbidities | 1.85 | 1.23 | - | 2.77 | 1.91 | 1.16 | - | 3.14 | 1.44 | 1.04 | - | 2.00 | ||||||
| MZL | Continuous | 1.35 | 1.09 | - | 1.67 | 0.757 | 1.43 | 0.97 | - | 2.09 | 0.684 | 1.15 | 1.01 | - | 1.32 | 0.581 | ||
| One vs No Comorbidities | 2.02 | 0.99 | - | 4.13 | 0.757 | 1.03 | 0.33 | - | 3.25 | 0.686 | 1.19 | 0.83 | - | 1.69 | 0.582 | |||
| Two vs No Comorbidities | 2.88 | 1.41 | - | 5.87 | 2.44 | 0.83 | - | 7.12 | 1.50 | 1.04 | - | 2.18 | ||||||
| TCL | Continuous | 1.13 | 0.99 | - | 1.29 | 0.583 | 1.04 | 0.89 | - | 1.22 | 0.554 | 1.01 | 0.90 | - | 1.14 | 0.539 | ||
| One vs No Comorbidities | 0.95 | 0.67 | - | 1.34 | 0.582 | 1.01 | 0.69 | - | 1.49 | 0.555 | 0.87 | 0.65 | - | 1.16 | 0.539 | |||
| Two vs No Comorbidities | 1.33 | 0.95 | - | 1.86 | 1.11 | 0.75 | - | 1.66 | 1.05 | 0.78 | - | 1.40 | ||||||
| Other NHL | Continuous | 1.40 | 1.24 | - | 1.59 | 0.655 | 1.25 | 1.05 | - | 1.48 | 0.589 | 1.26 | 1.13 | - | 1.40 | 0.599 | ||
| One vs No Comorbidities | 1.54 | 1.05 | - | 2.26 | 0.651 | 1.26 | 0.78 | - | 20.4 | 0.584 | 1.15 | 0.86 | - | 1.54 | 0.595 | |||
| Two vs No Comorbidities | 2.45 | 1.69 | - | 3.55 | 1.80 | 1.11 | - | 2.92 | 1.67 | 1.25 | - | 2.23 | ||||||
Adjusted for age and sex
Figure 1.
Kaplan-Meier curves for overall survival by categorical LCI, Lymphoma Epidemiology of Outcomes (LEO) cohort study.
In contrast to OS, the association of LCI with LSS and EFS (Table 3) were weaker when LCI was modeled as a continuous variable for LSS (HR=1.18, 95% CI 1.10–1.26; c-statistic=0.617) or EFS (HR=1.13, 95% CI 1.08–1.18; c-statistic=0.574) in age and sex adjusted models, with similar results for LCI modeled as a categorical variable. The KM curves for LSS and EFS are shown in Supplemental Figure 3.
3.4 |. Stratified and NHL Subtype Analyses
In analyses stratified on age, continuous LCI was more prognostic for OS in participants <65 years (HR=1.42, 95% CI 1.31–1.54; c-statistic=0.585) compared to those 65 years and older (HR=1.27, 95% CI 1.20–1.35; c-statistic=0.572). In analyses stratified on IPI, results were more prognostic in participants with IPI 0–2 (HR=1.35, 95% CI 1.25–1.46; c-statistic=0.570) compared to IPI 3+ (HR=1.20, 95% CI 1.13–1.28; c-statistic=0.553). In a model adjusted for age, sex and IPI, continuous (HR=1.22, 95% CI 1.16–1.29; c-statistic=0.670) and categorical (HR1vs0=1.17, 95% CI 1.02–1.35; HR2vs0=1.67, 95% CI 1.45–1.93; c-statistic=0.672) LCI remained prognostic.
As detailed in Table 3, LCI was also predictive of OS in participants treated in the first line with immunochemotherapy (HR=1.24, 95% CI 1.15–1.34; c-statistic=0.655) and those who were treated other management approaches (HR=1.27, 95% CI 1.19–1.36; c-statistic=0.657); for participants treated in the first line on a clinical trial (HR=1.27, 95% CI 1.01–1.58; c-statistic=0.649) versus not treated on a trial (HR=1.25, 95% CI 1.18–1.31; c-statistic=0.655); and after excluding participants with HIV/AIDS at diagnosis (HR=1.24, 95% CI 1.18–1.31; c-statistic=0.657). Consistent with the overall NHL results, LCI associations with LSS and EFS were weaker in all of these subgroups.
Continuous LCI was statistically significantly associated with OS for all major NHL subtypes (Figure 2) except TCL (HR=1.13, 95% CI 0.99–1.29; c-statistic=0.583), with the strongest effect observed for MZL (HR=1.35, 95% CI 1.09–1.67; c-statistic=0.757). LCI modeled as a categorical variable (0, 1, 2+) paralleled these results, and NHL subtype specific results for associations with LSS and EFS were generally weaker (Table 3).
Figure 2.
Forest plot of the association of LCI (continuous) with overall survival by lymphoma subtype, Lymphoma Epidemiology of Outcomes (LEO) cohort.
3.5 |. Cause of death
For all participants, the incidence of each type of cause of death increased with increasing number of comorbidities (all P<0.05), with lymphoma the leading cause of death at each comorbidity level but deaths due to lymphoma, lymphoma treatment, other causes, and unknown/missing becoming more prominent with more comorbidities (Figure 3). However, when we subset to IC-treated patients and DLBCL specifically (Supplemental Figure 4), both groups presumably treated for curative intent, there were no statistically significant differences for death due to lymphoma across LCI categories.
Figure 3.
Cumulative incidence for the competing risks of cause of death by LEO comorbidity index score.
4 |. Discussion
The LCI, which was adapted from the SRG-CCI for the LEO cohort study, predicted both 1-year mortality and OS after adjustment for age and sex, with weaker associations for LSS and EFS. The LCI predicted OS after further adjustment for IPI; in all common NHL subtypes (albeit more weakly and of only marginal statistical significance for TCL); and in subgroups defined on age, receipt of frontline immunochemotherapy, and participation in a frontline clinical trial. The cumulative incidence of death due to lymphoma, lymphoma treatment, and other causes all increased with increasing comorbidities, with the greatest increase observed for death due to other causes.
Strengths of this report include the large, prospective cohort study; geographic and racial/ethnic diversity that is largely representative of the US NHL population (10); expert pathology review; availability of high-quality clinical data; and virtually complete follow-up for key outcomes. There are also several limitations. While we had data on most of the SRG-CCI items, there were some differences in specific diseases within the broader comorbidity categories, and two of the items were derived from abstracted medical records instead of self-report. We did not have data on whether the comorbidities were active, controlled or historic. Reliability for self-report of individual comorbidities versus medical records in commonly used instruments (such as the SRG-CCI) varies by comorbidity (kappa range 0.66–0.86), but these instruments tend to have reasonable predictive ability (8). A significant proportion of the cohort was missing an LCI score (27.6%), mainly due to not completing an EQ. Nevertheless, there were few clinical differences between those with and without an LCI score, suggesting no strong selection bias in the subset with an LCI. Finally, for 14.4% of participants with missing data on either 1 (n=762) or 2 (n=28) of the 10 comorbidities, we assigned them as not affected for that specific comorbidity (or comorbidities) in the calculation of the LCI, which could lead to misclassification.
The performance of the LCI was only modestly weaker than that reported for the original SRG-CCI study (c-statistic 0.64 vs. 0.70) (9), but this was not unexpected, given its use in a new study population, imperfect adaptation of the SRG-CCI based on available data in LEO, and key differences in the original study populations. The latter issue is particularly important here as the SRG-CCI was developed on hospitalized patients while the LEO study population largely consists of ambulatory outpatients. Nevertheless, the performance of the LCI in LEO is similar to other comorbidity indices used in NHL for OS, LSS, across common subtypes (21–25) and supports its use in the LEO cohort. Finally, while the LCI is designed to be used internally for LEO analyses and not necessarily generalized to other studies or populations, our approach illustrates the feasibility of adapting the SRG-CCI to other study populations. Future approaches may be able to take advantage of extracting these data from electronic health records.
Beyond their usefulness in describing the burden of comorbidity in a population, comorbidity indices, including the LCI, are useful for conducting risk adjustment and to assess the potential confounding or mediating roles of comorbidities in an exposure (treatment) and outcome which is critical for ensuring both internal validity and generalizability of real-world results. The LCI was associated with both LSS and EFS, albeit not as strongly as for OS, and in particular non-lymphoma causes of death became more prominent in those with more comorbidities, while this was less pronounced of lymphoma-specific causes, particularly in patients treated with IC for presumably curative intent, as previously reported for DLBCL (23, 26). This highlights the importance of potential competing risks differentially impacting outcomes and the need to address comorbidities to understand outcomes of lymphoma treatments in both real-world studies and clinical trials, as well as its longer-term impacts on survivorship. Better assessment of comorbidities can also lead to advances in the management of lymphoma patients with comorbidities, including how to address the gaps in knowledge caused by many of these patients being excluded from the trials that lead to treatment recommendations (4). Optimal management of comorbidities may improve overall outcomes, as well as improving lymphoma-specific outcomes by expanding use of treatments with curative intent (23).
To conclude, the LCI predicted the impact of comorbidities on mortality outcomes in newly diagnosed NHL patients in the LEO cohort, providing a robust method to address comorbidities in the context of modeling prognosis and survivorship as well as an approach to implementing comorbidity indices in cancer survivorship cohorts more generally.
Supplementary Material
Acknowledgments
We thank Ms. Sondra Buehler for editorial assistance.
Funding:
This work was supported in part by the National Cancer Institute (U01 CA195568).
Footnotes
Conflict of interests
The authors declare no conflicts of interest.
Ethics Statements
The LEO cohort study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review boards of the eight centers, with approval for this analysis from the Mayo Clinic IRB.
Consent
Written informed consent was obtained from all participants.
Data Availability Statement
Data supporting the findings of this study are available on reasonable request to the corresponding author or through the LEO website (www.LEOcohort.org).
References
- 1.Siegel RL, Kratzer TB, Giaquinto AN, Sung H, Jemal A. Cancer statistics, 2025. CA Cancer J Clin. 2025;75(1):10–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cancer Stat Facts: Non-Hodgkin Lymphoma (https://seer.cancer.gov/statfacts/html/nhl.html); accessed 2/24/2024. [Google Scholar]
- 3.Sarfati D Review of methods used to measure comorbidity in cancer populations: no gold standard exists. J Clin Epidemiol. 2012;65(9):924–33. [DOI] [PubMed] [Google Scholar]
- 4.Chen W, Altshuler RD, Daschner P, Salvador Morales C, St Germain DC, Guida J, et al. Older adults with cancer and common comorbidities-challenges and opportunities in improving their cancer treatment outcomes. J Natl Cancer Inst. 2024;116(11):1730–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Yancik R Cancer burden in the aged: an epidemiologic and demographic overview. Cancer. 1997;80(7):1273–83. [PubMed] [Google Scholar]
- 6.Unger JM, Hershman DL, Fleury ME, Vaidya R. Association of patient comorbid conditions with cancer clinical trial participation. JAMA Oncol. 2019;5(3):326–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Yurkovich M, Avina-Zubieta JA, Thomas J, Gorenchtein M, Lacaille D. A systematic review identifies valid comorbidity indices derived from administrative health data. J Clin Epidemiol. 2015;68(1):3–14. [DOI] [PubMed] [Google Scholar]
- 8.Oemrawsingh A, Swami N, Valderas JM, Hazelzet JA, Pusic AL, Gliklich RE, et al. Patient-reported morbidity instruments: a systematic review. Value Health. 2020;23(6):791–811. [DOI] [PubMed] [Google Scholar]
- 9.Chaudhry S, Jin L, Meltzer D. Use of a self-report-generated Charlson Comorbidity Index for predicting mortality. Med Care. 2005;43(6):607–15. [DOI] [PubMed] [Google Scholar]
- 10.Cerhan JR, Maurer MJ, Link BK, Feldman AL, Habermann TM, Jaye DL, et al. The Lymphoma Epidemiology of Outcomes cohort study: Design, baseline characteristics, and early outcomes. Am J Hematol. 2024;99(3):408–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Crabtree HL, Gray CS, Hildreth AJ, O'Connell JE, Brown J. The Comorbidity Symptom Scale: a combined disease inventory and assessment of symptom severity. J Am Geriatr Soc. 2000;48(12):1674–8. [DOI] [PubMed] [Google Scholar]
- 12.Fan VS, Au D, Heagerty P, Deyo RA, McDonell MB, Fihn SD. Validation of case-mix measures derived from self-reports of diagnoses and health. J Clin Epidemiol. 2002;55(4):371–80. [DOI] [PubMed] [Google Scholar]
- 13.Sangha O, Stucki G, Liang MH, Fossel AH, Katz JN. The Self-Administered Comorbidity Questionnaire: a new method to assess comorbidity for clinical and health services research. Arthritis Rheum. 2003;49(2):156–63. [DOI] [PubMed] [Google Scholar]
- 14.Bayliss EA, Ellis JL, Steiner JF. Subjective assessments of comorbidity correlate with quality of life health outcomes: initial validation of a comorbidity assessment instrument. Health Qual Life Outcomes. 2005;3:51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lee SJ, Lindquist K, Segal MR, Covinsky KE. Development and validation of a prognostic index for 4-year mortality in older adults. JAMA. 2006;295(7):801–8. [DOI] [PubMed] [Google Scholar]
- 16.Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–83. [DOI] [PubMed] [Google Scholar]
- 17.Katz JN, Chang LC, Sangha O, Fossel AH, Bates DW. Can comorbidity be measured by questionnaire rather than medical record review? Med Care. 1996;34(1):73–84. [DOI] [PubMed] [Google Scholar]
- 18.Duru OK, Vargas RB, Kermah D, Nissenson AR, Norris KC. High prevalence of stage 3 chronic kidney disease in older adults despite normal serum creatinine. J Gen Intern Med. 2009;24(1):86–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.International Non-Hodgkin's Lymphoma Prognostic Factors P . A predictive model for aggressive non-Hodgkin's lymphoma. N Engl J Med. 1993;329(14):987–94. [DOI] [PubMed] [Google Scholar]
- 20.Gray RJ. A class of K-sample tests for comparing the cumulative incidence of a competing risk. Ann Stat. 1988;15:1141–54. [Google Scholar]
- 21.Gordon MJ, Duan Z, Zhao H, Nastoupil L, Iyer S, Ferrajoli A, et al. Comparison of comorbidity models within a population-based cohort of older adults with non-Hodgkin lymphoma. JCO Clin Cancer Inform. 2024;8:e2300223. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Gordon MJ, Duan Z, Zhao H, Nastoupil L, Ferrajoli A, Danilov AV, et al. A novel comorbidity score for older adults with non-Hodgkin lymphoma: the 3-factor risk estimate scale. Blood Adv. 2023;7(11):2632–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Wasterlid T, Mohammadi M, Smedby KE, Glimelius I, Jerkeman M, Bottai M, et al. Impact of comorbidity on disease characteristics, treatment intent and outcome in diffuse large B-cell lymphoma: a Swedish lymphoma register study. J Intern Med. 2019;285(4):455–68. [DOI] [PubMed] [Google Scholar]
- 24.Ellin F, Jerkeman M, Tornqvist J, Brudin L, Relander T. Impact of comorbidity on survival in peripheral T-cell lymphomas: A Swedish Lymphoma Registry study. Hematol Oncol. 2018;36(1):159–65. [DOI] [PubMed] [Google Scholar]
- 25.Antic D, Jelicic J, Trajkovic G, Balint MT, Bila J, Markovic O, et al. Is it possible to improve prognostic value of NCCN-IPI in patients with diffuse large B cell lymphoma? The prognostic significance of comorbidities. Ann Hematol. 2018;97(2):267–76. [DOI] [PubMed] [Google Scholar]
- 26.Juul MB, Jensen PH, Engberg H, Wehberg S, Dessau-Arp A, Haziri D, et al. Treatment strategies and outcomes in diffuse large B-cell lymphoma among 1011 patients aged 75 years or older: A Danish population-based cohort study. Eur J Cancer. 2018;99:86–96. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data supporting the findings of this study are available on reasonable request to the corresponding author or through the LEO website (www.LEOcohort.org).