Cause-specific mortality among Medicare beneficiaries with newly diagnosed non-Hodgkin lymphoma subtypes

Abstract

Background:

As the US population ages and non-Hodgkin lymphoma (NHL)-specific mortality declines, deaths from causes other than NHL will become increasingly important in treatment decision-making for older patients with NHL. This study sought to describe how the 5-year cumulative incidence of NHL-specific and other-cause mortality varies by subtype, age, comorbidity level, and time since diagnosis in older patients.

Methods:

Using the Surveillance, Epidemiology, and End Results cancer registry data linked to Medicare claims, we identified patients aged ≥66 at diagnosis with a first, primary NHL diagnosis from 2004–2013. Death certificate data and Fine-Gray competing risks models were used to estimate the 5-year cumulative incidence of NHL-specific and other-cause mortality by subtype, age, and comorbidity level. Estimates were displayed over time using stacked cumulative incidence curves.

Results:

Among 30,666 NHL patients, 32% died from NHL and 13% died from other causes within five years of diagnosis. The cumulative incidence of other-cause mortality increased with age and comorbidity level for all subtypes. Among aggressive subtypes, NHL-specific mortality exceeded other-cause mortality across all age groups, comorbidity levels, and years post-diagnosis. For indolent subtypes, other-cause mortality was similar to or exceeded NHL-specific mortality, especially among older patients with severe comorbidity or with the indolent marginal zone, lymphoplasmacytic, and mycosis fungoides subtypes.

Conclusions:

Findings suggest that mortality from causes other than NHL are important for patients with an older age, higher comorbidity level, and indolent disease. Evidence from this study can guide development of tools for estimating individual prognosis that inform NHL treatment discussions.

CONDENSED ABSTRACT:

This study sought to describe how the 5-year cumulative incidence of mortality from non-Hodgkin lymphoma (NHL) and other causes varies by subtype, age, comorbidity level and time since diagnosis in older patients. The study results suggest that mortality from causes other than NHL are important for patients with an older age, higher comorbidity level and indolent subtype, which can be used to inform development of tools for estimating individual prognosis and guiding treatment decisions.

INTRODUCTION:

Non-Hodgkin lymphoma (NHL) is the seventh most diagnosed cancer and eighth leading cause of cancer death among US men and women, with an estimated 74,680 new diagnoses and 19,910 deaths in 2017.¹ The demographic composition and survival of the NHL population has changed markedly over the past two decades. Notably, the proportion of new NHL diagnoses among older adults has risen since the late 1990s with the aging US population.² By 2030, two-thirds of new NHL diagnoses are expected to be aged ≥65.³ The aging population brings unique challenges to NHL treatment decision-making. Older patients are more susceptible to cancer treatment toxicities and have a greater number and severity of comorbidities than younger patients, which increases their likelihood of dying from causes other than cancer.^4–6

In addition to the aging NHL population, there has also been a decrease in NHL-specific mortality over the past two decades.^2,7 The decreasing NHL-specific mortality is largely attributable to the introduction of rituximab, a monoclonal antibody against CD20, as well as the development of other effective second- and third-line treatments.⁸ As patients live longer with NHL, their risk of dying from other causes increases. Going forward, treatment decision-making for older patients with NHL could benefit from information about the risk of mortality from causes other than NHL, known as competing causes.

The importance of competing causes of mortality in treatment decisions likely varies across NHL subtypes, which have heterogeneous demographic, clinical and tumor characteristics that differentially influence NHL-specific prognosis.^7,9,10 One key difference between subtypes is the average speed of tumor growth, with subtypes broadly grouped as having indolent or aggressive growth. Compared to indolent subtypes, aggressive subtypes progress more rapidly, potentially leading to earlier NHL-specific mortality ⁹ and more intensive treatment. However, aggressive NHL subtypes are also more likely to achieve complete cure. In contrast, indolent subtypes are characterized by patterns of disease remission and relapse requiring long-term management of the cancer, such as additional treatment that can lead to adverse events.¹¹ Although patients with aggressive disease face treatment that can impact comorbidity and subsequent non-cancer deaths, indolent patients are generally treated over a longer period with additional treatments due to relapses. Taken together, deaths from causes other than NHL may be a greater concern for treatment decisions among indolent rather than aggressive subtypes.

As mortality from competing causes becomes increasingly important among patients with NHL, cause-specific prognosis estimates are needed to inform NHL treatment decisions. In the cancer literature, the most commonly reported measures of cause-specific prognosis are net cancer-specific mortality risks, which remove or censor patients from an analytic cohort after they die of causes other than the cancer.¹² Net cancer-specific mortality risks are useful for isolating the effect of interventions on cancer mortality and for comparing cancer mortality across time or populations. However, by assuming that patients only die from NHL and ignoring patients who die of other causes, these measures can overestimate the probability of NHL-specific mortality.^13,14 In contrast, cumulative incidence functions (also called crude measures in the surveillance literature) are cause-specific mortality measures that account for competing risks.¹² These measures are calculated by retaining patients in the denominator population after death from a competing cause, preventing inflation of prognosis estimates so that the probabilities of NHL-specific mortality, non-cancer mortality, and survival add to 100%.^12,14–16 Though not commonly reported, cumulative incidence functions provide the best reflection of a patient’s prognosis in real-world settings and are the most useful measures for informing individual treatment decisions.^12,15

This study sought to describe patterns in the cumulative incidence of NHL-specific and other-cause mortality by prognostic factors for older patients with NHL, including subtype, age and comorbidity level.

METHODS:

Data source and study population

For this analysis, we used data from the 18 US Surveillance, Epidemiology, and End Results (SEER) cancer registries linked with Medicare insurance claims. SEER registries cover approximately 28% of the US and provide information on NHL diagnosis and mortality that are largely representative of those observed in the general US population.¹⁷ Medicare is a federally funded program providing health insurance to persons aged ≥65. We analyzed data from Medicare Part A (hospital, skilled-nursing facility, hospice, home health care) and Part B (physician and outpatient services) fee-for-service coverage¹⁸

Using the SEER-Medicare data, we identified patients aged ≥66 years at diagnosis with first, primary NHL between January 1, 2004 and December 31, 2013. Eligible patients were required to have continuous Medicare Parts A and B and no managed care coverage for the 12 months before the diagnosis date (set to the first day of the diagnosis month). Our study started in 2004 after the 1997 FDA approval and dissemination of rituximab to ensure that patients with B-cell subtypes responding to rituximab, which form the majority of NHL cases, had a similar opportunity to experience survival advantages from this drug.^19–21 B- and T-cell NHL cases were identified using the SEER site recode variable,²² which was created by SEER registries using ICD-0–3 site and histology codes and updated based on the World Health Organization’s (WHO) 2008 Classification of Tumours of Haematopoietic and Lymphoid Tissues (Table S1).²³ Cases with a SEER site recode variable for NHL were categorized by subtype using the SEER lymphoma subtype recode variable,²⁴ which was created using the International Lymphoma Epidemiology Consortium’s (InterLymph) subtype classification²⁵ based on the 2008 WHO classification system for hematological and lymphoid tissue malignancies. (Table S2).²³ Although chronic lymphocytic leukemia/small lymphocytic lymphoma is classified as an NHL subtype by the SEER lymphoma subtype recode variable, this condition mainly is classified as a leukemia in the SEER site recode variable and was not included in the current analysis. Waldenstrom macroglobulinemia is a subset of lymphoplasmacytic lymphoma, but the ICD-O-3 histology code for this subtype (9761) is not included in the SEER site recode variable for NHL; therefore, it was also excluded from the analysis. Using the SEER lymphoma subtype recode variable, we further excluded malignancies that have unspecified/unknown subtypes, that primarily affect non-lymphoid tissue, that occur in precursor or plasma cells, or that have a primary site in the blood rather than in lymph nodes. Excluded subtypes were lymphoblastic leukemia/lymphomas (ICD-O-3 histology codes 9811–9818, 9837), plasma cell/myelomas (9731–9732, 9734, 9762), and precursor lymphomas (9724–9729, 9735). We removed Sezary syndrome due to small numbers preventing stable stratification. Patients diagnosed at autopsy or death were also excluded.

Demographic and Clinical Variables

We grouped NHL into indolent and aggressive subtypes based on clinical expertise and prior knowledge about subtype-specific survival.^4,7,9 Aggressive subtypes included diffuse large B-cell lymphoma (DLBCL), peripheral T-cell lymphoma (PTCL), and Burkitt lymphoma. Although subpopulations of mantle cell lymphoma can exhibit indolent tumor growth,⁴ this subtype was categorized as aggressive since it displays a higher NHL-specific mortality than observed in typical indolent NHL.¹⁰ Indolent subtypes included follicular lymphoma, marginal zone lymphoma (MALT extranodal, nodal, and splenic), lymphoplasmacytic, and mycosis fungoides. Although mycosis fungoides is generally considered a cutaneous lymphoma that primarily affects the skin rather than the lymphatic system,²⁵ the subtype was included in the indolent NHL subgroup since it is characterized by a slow tumor growth.⁴

We assigned patients to an age group according to their age at diagnosis (66–74, 75–84, 85+ years). We also identified sex, race (white, black, Hispanic, other), and Ann Arbor cancer stage (I/II, III/IV) in the SEER data to further describe the population. Patients were assigned indicator variables if they received care for each of the 16 noncancer conditions from the Charlson Comorbidity Index²⁶, which were identified in Medicare claims using the International Classification of Disease, 9th edition, Clinical Modification (ICD-9-CM) codes for cancer populations.^28,29 These weights were log hazard ratios estimated using Cox proportional hazards models regressing each condition and interactions between the most prevalent conditions (diabetes, cardiovascular disease, COPD, congestive heart failure, peripheral vascular disease, renal disease, rheumatological disease, and ulcers) on non-cancer survival, controlling for age, sex, and race. Comorbid condition indicators were multiplied by weights and summed over the 16 conditions and nine interactions. Individuals were stratified into comorbidity groups (none=0, low or moderate=0–0.66, and high=>0.66) using cutpoints determined by grouping comorbidities according to their importance in cancer treatment decisions.³⁰ Table S3 in the Supplement provides the ICD-9-CM codes used to define each condition, the Mariotto et al. weights for each condition and interaction, and the prevalence of each condition in the cohort.

Cause of Death

Patients were followed from NHL diagnosis until death or the end of follow-up on December 31, 2013. We identified deaths using state death certificate data compiled by the National Center for Health Statistics and linked to SEER records.³¹ Deaths were linked to individuals with SEER data regardless of whether they died within or outside of a SEER registry.

Deaths were defined by major site groups on death certificates based on 3-digit International Classification of Disease version 10 codes. We defined death using the “SEER cause-specific death” variable, which adjusts for potential misattribution of NHL-specific deaths by considering tumor site, origin, and order, as well as secondary malignancies and comorbidities that commonly occur with NHL (e.g. HIV).³² For this variable, decedents diagnosed with a first, primary NHL who have a cause of mortality listed as a secondary treatment-related leukemia or another secondary cancer are relabeled as having an NHL-specific death. Patients with a first, primary NHL diagnosis who have a cause of death from HIV/AIDS or another NHL-related comorbidity are also relabeled as dying from NHL. Any death not classified as a cancer death was considered an “other-cause death.”

Statistical Analysis

For each NHL subtype, we calculated all-cause mortality as the complement of overall survival probabilities from Cox proportional hazards models. We estimated the cumulative incidence of NHL-specific and other-cause mortality by subtype, age, and comorbidity level using the Fine-Gray subdistribution hazards regression model,¹⁶ which accounts for competing causes of death preventing patients from experiencing the event of interest. When calculating the cumulative incidence of NHL-specific mortality, NHL death was the event of interest, and death due to other causes was the competing event. For the cumulative incidence of other-cause mortality, death from causes other than NHL was the event of interest, and NHL death was the competing event. We estimated 95% confidence intervals using the variance of the cumulative incidence functions from the cmprsk package in R.³³

We calculated the cumulative incidence functions of NHL-specific and other-cause mortality for each subtype, age group and comorbidity level. These values were graphed using stacked bar charts of 5-year cumulative incidence functions and stacked cumulative incidence curves over the five years post-diagnosis. The top of the stacked curves represented cumulative all-cause mortality. The area above the curves represented the overall survival probability at each time point after NHL diagnosis. Analyses were conducted using R open source software.³⁴

RESULTS:

From 2004–2013, 30,666 eligible adults aged 66+ were newly diagnosed with mature B- or T-cell NHL in the SEER-Medicare database (Figure 1). Of these individuals, 39.5% had indolent subtypes and 60.5% had aggressive subtypes. The most common subtype was DLBCL (47.7%), followed by follicular (21.9%) and marginal zone lymphoma (13.5%).

Figure 1.

Flowchart for the study population. CLL/SLL indicates chronic lymphocytic leukemia/small lymphocytic lymphoma; DLBCL, diffuse large B-cell lymphoma;NHL, non-Hodgkin lymphoma; PTCL, peripheral T-cell lymphoma; SEER, Surveillance, Epidemiology, and End Results.

Table 1 displays the characteristics of older adults newly diagnosed with NHL by subtype. In general, patients with indolent subtypes were more likely to be younger, female, and white, non-Hispanic. At the time of diagnosis, patients with indolent subtypes had less advanced disease than patients with aggressive subtypes. For all patients with NHL, the most prevalent baseline comorbidities were diabetes (23.8%), chronic obstructive pulmonary disease (14.9%), and congestive heart failure (11.2%). Patients with aggressive subtypes had a slightly higher burden of comorbidity at diagnosis than indolent subtypes, including a higher prevalence of diabetes (25.5% v 21.3%), cerebrovascular disease (8.3% v 6.4%), renal disease (8.8% v 6.5%), congestive heart failure (12.2% v. 9.6%), and HIV/AIDS (1.12% vs. 0.02%). Subtypes with more adults aged 85+ had a higher comorbidity level.

Table 1.

Demographic characteristics of 30,666 non-Hodgkin lymphoma patients age ≥66 at diagnosis from 2004–2013 by tumor growth group and subtype in the linked Surveillance, Epidemiology, and End Results cancer registry and Medicare claims database

			Aggressive										Indolent
Characteristics	Total (n=30,666)		Total Aggressive (n=18,565)		DLBCL (n=14,636)		PTCL (n=1908)		Mantle Cell (n=1742)		Burkitt (n=279)		Total Indolent (n=12,101)		Follicular (n=6703)		Marginal Zone (n=4128)		Lympho- plasmacyticb (n=677)		Mycosis Fungoides (n=593)
	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%
Age Group
66–74	12551	40.9	7139	38.5	5468	37.4	802	42.0	763	43.8	106	38.0	5412	44.7	3180	47.4	1680	40.7	257	38	295	49.7
75–84	13074	42.6	8108	43.7	6431	43.9	821	43.0	725	41.6	131	47.0	4966	41.0	2692	40.2	1742	42.2	311	45.9	221	37.3
85+	5041	16.4	3318	17.9	2737	18.7	285	14.9	254	14.6	42	15.1	1723	14.2	831	12.4	706	17.1	109	16.1	77	13
Sex
Female	16006	52.2	9280	50.0	7673	52.4	893	46.8	584	33.5	130	46.6	6726	55.6	3730	55.6	2378	57.6	339	50.1	279	47.0
Male	14660	47.8	9285	50.0	6963	47.6	1015	53.2	1158	66.5	149	53.4	5375	44.4	2973	44.4	1750	42.4	338	49.9	314	53.0
Race/Ethnicity
White, NH	25526	83.2	15241	82.1	12036	82.2	1475	77.3	1515	87.0	215	77.1	10285	85.0	5849	87.3	3403	82.4	574	84.8	459	77.4
Black, NH	1236	4.0	771	4.2	542	3.7	157	8.2	57	3.3	15	5.4	465	3.8	196	2.9	200	4.8	20	3.0	49	8.3
Hispanic	2000	6.5	1286	6.9	1034	7.1	122	6.4	106	6.1	24	8.6	714	5.9	393	5.9	260	6.3	36	5.3	25	4.2
Other	1593	5.2	1136	6.1	934	6.4	127	6.7	50	2.9	25	9.0	457	3.8	207	3.1	202	4.9	28	4.1	20	3.4
Missing	311	1.0	131	0.7	90	0.6	27	1.4	14	0.8	0	0.0	180	1.5	58	0.9	63	1.5	19	2.8	40	6.7
Ann Arbor Stage
I/II	13477	43.9	7895	42.5	6667	45.6	782	41.0	347	19.9	99	35.5	5582	46.1	3001	44.8	2136	51.7	84	12.4	361	60.9
III/IV	14984	48.9	9520	51.3	7159	48.9	922	48.3	1275	73.2	164	58.8	5464	45.2	3223	48.1	1624	39.3	555	82.0	62	10.5
Missing	2205	7.2	1150	6.2	810	5.5	204	10.7	120	6.9	16	5.7	1055	8.7	479	7.1	368	8.9	38	5.6	170	28.7
Comorbidity Levela
None	14881	48.5	8580	46.2	6665	45.5	913	47.9	882	50.6	120	43.0	6301	52.1	3595	53.6	2036	49.3	347	51.3	323	54.5
Low/moderate	5946	19.4	3672	19.8	2941	20.1	359	18.8	305	17.5	67	24.0	2274	18.8	1291	19.3	773	18.7	100	14.8	110	18.5
High	9839	32.1	6313	34.0	5030	34.4	636	33.3	555	31.9	92	33	3526	29.1	1817	27.1	1319	32.0	230	34.0	160	27.0

Abbreviations: DLBCL=diffuse large B-cell lymphoma; PTCL=peripheral T-cell lymphoma; NH=non-Hispanic

^aDefined based on comorbidity score and cut-points identified by Mariotto et al.²⁹ and Cho et al.³⁰

^bDoes not include Waldenstrom macroglobulinemia.

There was some variation in characteristics within tumor growth speed groups. Among aggressive subtypes, the percentage of patients diagnosed at age 85+ years ranged from 15% in Burkitt to 19% in DLBCL. Over 66% of mantle cell lymphoma patients were male compared to 48% of DLBCL patients. Patients diagnosed with mantle cell or Burkitt lymphoma were more likely to have advanced disease than those with other aggressive subtypes. For indolent subtypes, patients with marginal zone and lymphoplasmacytic lymphoma were generally older than patients with other subtypes. The majority of patients with lymphoplasmacytic lymphoma were diagnosed in advanced stages, while early stage diagnoses were more common for other indolent subtypes, including marginal zone lymphoma and the cutaneous NHL, mycosis fungoides.

Table 2 reports the number of deaths from cancer and other causes in the study period and the five-year NHL-specific and other-cause mortality by subtype. Forty-five percent of NHL patients died in the five years after a new diagnosis. Thirty-two percent of patients died from NHL and 13% of patients had deaths attributed to causes other than NHL.

Table 2.

5-year cumulative incidence functions of all-cause, non-Hodgkin lymphoma-specific, and other-cause mortality by tumor growth group and subtype for patients aged ≥66 diagnosed with non-Hodgkin lymphoma from 2004–2013 in the linked Surveillance, Epidemiology and End Results-Medicare data.

		Deaths in 5-years		5-year all-cause mortality		5-year cancer- specific mortality		5-year other-cause mortality
Subtype	Total (N=30,666)	Cancer (n=9679)	Non-cancer (n=4113)	CIF (%)	95% CI	CIF (%)	95% CI	CIF (%)	95% CI
Aggressive	18,565	7684	2453	61.8	61.0,62.6	45.6	44.8,46.3	16.2	15.6,16.8
Diffuse large B-cell lymphoma	14,636	5921	1965	60.5	59.6,61.4	44.1	43.2,45.0	16.4	15.7,17.1
Peripheral T-cell lymphoma	1908	869	250	66	63.5,68.3	50.3	47.8,52.7	15.7	14.0,17.7
Mantle cell lymphoma	1742	716	211	65.9	63.1,68.5	50.2	47.5,53.0	15.7	13.7,17.3
Burkitt lymphoma	279	178	27	77.2	71.1,82.1	65.5	59.5,71.0	11.8	8.0,16.2
Indolent	12,101	1995	1660	37.4	36.4,38.4	19.5	18.7,20.2	17.9	17.1,18.7
Follicular lymphoma	6703	1258	816	37.5	36.2,38.8	21.8	20.7,22.9	15.7	14.7,16.7
Marginal zone lymphoma	4128	534	649	36.2	34.5,37.9	15.7	14.5,17.0	20.5	19.0,222.0
Lymphoplasmacytic	677	127	113	46.9	42.0,51.4	23.1	19.6,26.9	23.8	20.0,27.8
Mycosis fungoides	593	76	82	35.5	30.6,40.1	16.2	12.9,19.8	19.3	15.6,23.4

CIF=cumulative incidence functions; CI=confidence interval

Patients diagnosed with indolent subtypes had a lower cumulative incidence of NHL-specific mortality (20% vs. 46%) and a slightly higher cumulative incidence of other-cause mortality (18% vs. 16%) at five years post-diagnosis than aggressive subtypes. Those with indolent marginal zone, lymphoplasmacytic, and mycosis fungoides subtypes had a higher cumulative incidence of other-cause mortality than NHL mortality at five years.

Figure 2 illustrates the cumulative incidence of NHL-specific mortality, other-cause mortality, and survival at five years by subtype, age group, and comorbidity level. Five-year NHL-specific mortality was higher for every age and comorbidity level in aggressive subtypes than in indolent subtypes. Among aggressive subtypes, 5-year NHL-specific mortality rose with increasing age and was more influenced by comorbidity in younger than older ages. Five-year other-cause mortality increased with age and comorbidity level for most subtypes and was highest among older patients with indolent subtypes with the greatest comorbidity level.

Figure 2.

Bar charts displaying non-Hodgkin lymphoma-specific mortality, other cause mortality, and survival probabilities at five years after patients aged ≥66 are diagnosed with non-Hodgkin lymphoma from 2004–2013 by subtype, comorbidity group, and age groups.

Figure 3 presents cumulative incidence curves for other-cause mortality stacked on cumulative incidence curves for NHL-specific mortality over the five years post-diagnosis by age group and subtype. Cumulative incidence curves for NHL-specific and other-cause mortality varied across NHL subtypes, though similar patterns were observed among subtypes with the same speed of tumor growth. Among aggressive subtypes, NHL-specific mortality increased rapidly within the first year of diagnosis and exceeded other-cause mortality throughout the five years post-diagnosis, regardless of age group. The increase in cumulative incidence for NHL-specific and other-cause mortality became steeper as patients aged. In general, the cumulative incidence of other-cause mortality increased more rapidly in older patients with indolent subtypes than for those with aggressive subtypes. Among older patients diagnosed with the indolent marginal zone, lymphoplasmacytic, and mycosis fungoides subtypes, other-cause mortality exceeded NHL-specific mortality for patients surviving three or more years post-diagnosis.

Figure 3.

Stacked cumulative incidence curves of non-Hodgkin lymphoma-specific (dark blue) and other-cause (light blue) mortality over the five years after patients aged ≥66 are diagnosed with non-Hodgkin lymphoma from 2004–2013 by subtype and age group.

Figure 4 displays the stacked cumulative incidence curves for NHL-specific and other-cause mortality in the five years post-diagnosis stratified by subtype and comorbidity level. Compared to patients diagnosed with aggressive subtypes with no or low/moderate comorbidity at diagnosis, patients with a high comorbidity level have a greater increase in other-cause mortality over the five years. This increase is most notable among indolent subtypes.

Figure 4.

Stacked cumulative incidence curves of non-Hodgkin lymphoma-specific (dark blue) and other-cause (light blue) mortality over the five years after patients aged ≥66 are diagnosed with non-Hodgkin lymphoma from 2004–2013 by subtype and comorbidity level.

The cumulative incidence curves for NHL-specific mortality generally increased at a faster rate in advanced stages than early stages among aggressive subtypes (Figure S1 in the supplement). In contrast, other-cause mortality increased at a slightly faster rate in early versus advanced stages. Similar patterns were observed among indolent subtypes.

DISCUSSION:

In this population-based study, we explored the risks of NHL-specific and other-cause mortality among older Medicare beneficiaries diagnosed with NHL during the rituximab era by subtype, age group, comorbidity level, and time since diagnosis. Our findings suggest that, for most subtypes, NHL-specific mortality increases with age and by comorbidity level among younger seniors. Other-cause mortality generally increases with age and comorbidity level, regardless of age. Similar patterns have been observed in other cancer sites.^35,36 At five years post-diagnosis, NHL-specific mortality is higher for aggressive subtypes compared to indolent subtypes. In contrast, the 5-year cumulative incidence of other-cause mortality is generally higher in each age and comorbidity group for indolent subtypes than aggressive subtypes, especially for patients diagnosed with the marginal zone, lymphoplasmacyctic, and mycosis fungoides subtypes. Patterns with indolent subtypes mirror those previously reported for early stage, solid tumor cancers, which are also slower growing.^35,36

Prior population-based studies of patients with NHL have also observed variation in overall survival^7,9,37 and net NHL-specific mortality estimates^7,10,38 across subtypes, age groups, and comorbidity levels. However, overall survival estimates do not provide specific information about the cumulative incidence of death from NHL or other causes, and net survival measures do not account for competing causes of death.^12,15 By exploring patterns of NHL-specific and other-cause mortality, our results contribute unique, population-level evidence about the impact of competing risks on survival in older NHL patients.

A strength of this study is use of the linked SEER-Medicare data, which is generally representative of the US population.³⁹ Patterns in the cumulative incidence of cause-specific mortality observed in the SEER-Medicare data are expected to reflect patterns among all older adults in the US. The SEER-Medicare data also provide an opportunity to measure prognostic importance of comorbid conditions present at the time of diagnosis. Other data sources, such as cancer registries, generally do not collect comorbidity data, while clinical trials generally exclude individuals with higher comorbidity levels, affecting the ability to translate prognostic trends observed by comorbidity levels in trial populations to those expected in the general population. Another strength is that we explore the cumulative incidence of cause-specific mortality by NHL subtype, which to our knowledge, has not been explored previously. Our findings demonstrate the variation in cause-specific mortality across subtypes, suggesting the subtypes that may benefit more from aggressive NHL interventions, such as Burkitt lymphoma, and the subtypes in which watch-and-wait strategies may be more beneficial, such as marginal zone lymphoma. Finally, this study provides information on mortality trends from a time period in which contemporary first-line treatment paradigms with rituximab were used for most patients with the two most common subtypes, follicular lymphoma⁴⁰ and DLBCL.⁴¹

There are also limitations of this analysis. SEER-Medicare only provides information on patients with NHL who are aged ≥65 years. Eighty-nine percent of patients are missing the International Prognostic Index (IPI)⁴² or other prognostic scores widely used by oncologists to inform treatment decisions. Although some components of the IPI score are available in SEER-Medicare data, including age and stage, other components cannot be identified directly in the data, including performance status, number of extranodal sites, and lactate dehydrogenase levels. Future studies should explore how cumulative incidence of NHL-specific and other-cause mortality vary by the IPI score. Despite use of the enhanced cause-specific death variable, cause of death may still be misclassified, leading the cumulative incidence of NHL-specific mortality to falsely appear higher or lower than other-cause mortality across subtypes and time periods.^32,43 Another limitation is that the 5-year crude mortality risks reflect death in the presence of treatments available for patients at the time of their diagnosis from 2004–2013. NHL treatment has changed since 2004, with increased use of rituximab and other targeted treatments and advancements in stem-cell transplants and cellular therapies (e.g. chimeric antigen receptor T-cell [CART] therapy). As less toxic, novel therapies are introduced, effective treatment options are now more accessible to patients with a higher comorbidity burden who otherwise may have not been treated with toxic chemotherapies. Due to treatment advances, 5-year crude mortality risks may look different for patients diagnosed in 2004 than those diagnosed in 2013. Although prior studies have shown mortality rates plateauing during this time period,⁷ relative measures utilizing expected survival data from life tables may be useful for exploring time trends in NHL prognosis.¹²

Our findings describe population-level patterns in the cumulative incidence of NHL-specific and other-cause mortality. These population-level results suggest that treatment decision-making for patients with indolent subtypes who are older or who have higher comorbidity levels may benefit from information comparing the cumulative incidence of non-cancer mortality to NHL-specific mortality. However, to improve outcomes among older NHL patients, individual-level estimates of the cumulative incidence of cancer-specific and other-cause mortality are needed, as well as tools that predict these outcomes according to a patient’s specific characteristics. Current NHL prognosis tools, such as the International Prognostic Index⁴² and Follicular Lymphoma Prognostic Index,⁴⁴ were developed to inform providers on a patient’s probability of overall mortality. However, these tools do not provide context regarding the patient’s cancer-specific mortality risks in the presence of competing risks, nor do they inform providers on the risk of death from causes other than NHL. Currently, the NCI is developing the SEER*CSC tool for prostate, breast, colorectal, and head and neck cancers, which will provide nomograms for predicting the cumulative incidence of surviving or dying from cancer or other causes based on a patient’s tumor, age, race, gender, and other measures of health status.^45,46 Our study informs the development of predictive tools like the SEER*CSC nomogram for NHL, which would generate highly personalized, actual prognosis measures for informing treatment discussions between providers and older patients with NHL.