Survival outcomes with and without splenectomy in splenic marginal zone lymphoma


  • Adam J. Olszewski

    Corresponding author
    1. Division of Hematology/Oncology, Memorial Hospital of Rhode Island, The Warren Alpert Medical School of Brown University, Providence, Rhode Island
    • Memorial Hospital of Rhode Island, Assistant Professor, The Alpert Medical School of Brown University,111 Brewster St., Pawtucket, RI 02860
    Search for more papers by this author

  • Conflict of interest: Nothing to report.

  • Preliminary findings from this research were presented at the 2012 Annual Meeting of the American Society of Clinical Oncology, June 1–5, 2012, in Chicago, IL.

The optimal management of splenic marginal zone lymphoma (SMZL), a rare malignancy of the elderly, is not clearly defined [1, 2]. Splenectomy had been advocated in the past due to its palliative effect on abdominal discomfort, cytopenias, and prolonged remissions, even though most patients have disseminated disease in the bone marrow which subsequently relapses. No prospective studies compared outcomes after surgery or chemotherapy, so the choice of treatment is based on perception of individual benefits and risks. This study analyzed 1,251 cases of SMZL recorded in the Surveillance, Epidemiology, and End Results (SEER) database. Effects of splenectomy on survival outcomes were evaluated after balancing confounders with a propensity score (PS). While 52% of patients underwent spleen removal, no significant impact of the procedure on the risk of lymphoma-related death (LRD) (P = 0.66) or overall survival was detectable. The results strengthen the notion that, with the availability of safe and effective alternatives, splenectomy should no longer be considered the treatment of choice in SMZL.

SMZL is an indolent B-cell lymphoma, introduced as a provisional diagnostic category in the Revised European-American Lymphoma Classification in 1994, and then as a separate subtype in the World Health Organization classification in 2001. Due to its rarity (less than 1% of all non-Hodgkin lymphomas), prognostic factors and treatments are based on retrospective series subject to inherent bias [3, 4]. Several studies from the era when alkylating chemotherapy was the principal therapeutic alternative to splenectomy indicated better survival in patients undergoing surgery [5–7]. This paradigm has been increasingly questioned with excellent outcomes achieved using rituximab alone or in combination with purine analogues [8–11]. Since a randomized trial to answer the question is unlikely to be conducted, I carried out an observational study based on the SEER registry to evaluate previous assertions.

The SEER database contained 1,251 SMZL patients aged 25–96, diagnosed between 1995 and 2009 (Table I). The median age was 68 years, and the median follow-up 37 months, reflecting preponderance of cases diagnosed after 2005. Most patients were Caucasian, residing in metropolitan areas and with disseminated (stage IV) lymphoma. Overall, 52.1% underwent splenectomy and 1.2% radiotherapy as the first course of therapy. There was no available information on alternative management choices (watchful waiting, systemic therapy). Patients undergoing splenectomy were younger, more likely to have B-symptoms or Stage I/II disease. The frequency of splenectomy decreased over the period of study from about 70% in the 1990s to 50% in the late 2000s.

Table I. Patient Characteristics and Comparison of Treatment Groups
VariableTotal (% by column)Group P-valueAbsolute standardized difference
No splenectomy (% by row)Splenectomy (% by row)UnmatchedMatched
  1. Standardized differences were calculated for means of covariates between treated and untreated group in the overall population and after propensity score matching.

  2. IQR: interquartile range. SD: standard deviation.

Number of patients (%)1,251 (100)599 (47.9)652 (52.1)   
Age (years)      
 Mean (SD)67.3 (12.6)69.8 (12.7)65.0 (12.1) 0.3970.004
 Median (IQR)68 (58–77)72 (60–80)66 (56–77)<0.0001  
Sex (%)      
 Female670 (53.6)300 (44.8)370 (55.2)0.020.1350.015
 Male581 (46.4)299 (51.5)282 (48.5)   
Race   0.004  
 White1,123 (89.8)526 (46.8)597 (53.2) 0.1350.005
 Black60 (4.8)33 (55.0)27 (45.0) 0.0690.007
 Asian/Native American51 (4.1)25 (49.0)26 (51.0) 0.010.01
 Unrecorded17 (1.3)15 (88.2)2 (11.8) 0.3970.033
Year of diagnosis   0.003  
 1995–2000108 (8.6)37 (34.3)71 (65.7) 0.1510.025
 2001–2004460 (36.8)213 (46.3)247 (53.7) 0.0480.008
 2005–2009683 (54.6)349 (51.1)334 (48.9) 0.1410.022
ICD-O-3 histology designation   <0.0001  
 9689866 (69.2)452 (52.2)414 (47.8) 0.2480.023
 9699385 (30.8)147 (38.2)238 (61.8)   
Stage   <0.0001  
 Localized (I)289 (23.1)81 (28.0)208 (72.0) 0.3940.029
 Regional (II)81 (6.5)23 (28.4)58 (71.6) 0.1780.008
 Disseminated (III.IV)835 (66.8)467 (55.9)368 (44.1) 0.4340.033
 Unrecorded46 (3.6)28 (60.9)18 (39.1) 0.1170.033
B-symptoms   0.01  
 Absent523 (41.8)255 (48.8)268 (51.2) 0.030.001
 Present287 (22.9)116 (40.4)171 (59.6) 0.1560.033
 Unrecorded441 (35.3)228 (51.7)213 (48.3) 0.1150.03
No prior malignancy1,023 (81.8)505 (49.4)518 (50.6)0.0280.120.027
 Prior malignancy228 (18.2)94 (41.2)134 (58.8)   
Marital status   0.039  
 Married745 (59.6)336 (45.1)409 (54.9) 0.1370.007
 Widowed190 (15.2)94 (49.5)96 (50.5) 0.0270.005
 Other316 (25.2)169 (53.5)147 (46.5) 0.1360.004
Region   0.001  
 East Coast388 (31.0)159 (41.0)229 (59.0) 0.180.001
 Northern Plains178 (14.2)80 (44.9)98 (55.1) 0.0470.011
 Pacific Coast622 (49.7)333 (53.5)289 (46.5) 0.2270
 South West63 (5.0)27 (42.9)36 (57.1) 0.0440.015
Type of area   0.12  
 Metropolitan1,100 (87.9)536 (48.7)564 (51.3) 0.0870.042
 Urban/Rural151 (12.1)63 (41.7)88 (58.3)   
Percent persons below poverty level      
 Mean (SD)12.3 (5.2)12.6 (5.2)12.0 (5.2)0.030.1150.042
Lymphoma-related death203 (16.2)99 (48.8)104 (51.2)   
Any death376 (30.0)187 (49.7)189 (50.3)   
Median follow-up in months (IQR)37 (15–67)31 (12–60)40.5 (17–71)   

Altogether, 203 patients (16.2%) died from lymphoma (54% of all deaths). Other major causes of death included cardiovascular disease (17%), other malignancies (10%), and unspecified events (9%). The 5-year relative survival was 79.8% (95% confidence interval, 95% CI, 75.8–83.1) with no difference by gender (P = 0.62). This correlated with cumulative incidence of LRD, which was 17.4% (95% CI, 15.0–20.0) at 5 years. The Kaplan–Meier estimate of overall survival was 67.8% at 5 years (95% CI 64.7–70.9) with a median survival of 8.6 years (95% CI, 7.8–9.6). In a multivariate Cox model, only age over 65 (hazard ratio, HR, 3.09, 95% CI, 2.23–4.29, P < 0.0001) and the presence of B-symptoms (HR, 1.83, 95% CI, 1.27–2.66, P = 0.001) were significant factors associated with inferior cause-specific outcome. Gender, race, clinical stage, or year of diagnosis were not significant.

The PS analysis successfully balanced all covariates, as evidenced by standardized differences of means (a difference smaller than 0.1 implies adequate balance, Fig. 1). After the adjustment, 1,194 patients were included in the PS-matched cohort. A comparison of cumulative incidence curves indicated no difference between treated and untreated patients in the probability of LRD (P = 0.66) or competing event (P = 0.97, Fig. 2a). Likewise, there was no evident impact of splenectomy in the regression model on the risk of LRD (subhazard ratio, SHR, 1.09, 95% CI, 0.78–1.51, P = 0.63) or on overall survival (HR, 1.03, 95% CI, 0.81–1.30, P = 0.84, Fig. 2b). Consistently null results were also obtained in the subgroup of 568 patients diagnosed before 2005, with a longer median follow-up of 64 months (P = 0.77) as well as inthe unadjusted population using conventional multivariate regression (n = 1,250, SHR 1.00, 95% CI 0.74–1.34, P = 0.99).

Figure 1.

Absolute standardized differences of covariate means, comparing treated to untreated patients in the original and PS-matched samples. Differences smaller than 0.1 indicate adequate balance.

Figure 2.

Survival outcomes in the PS-matched population. (a) Cumulative incidence of LRDs and competing events. (b) Overall survival.

As a sensitivity analysis, in an attempt to minimize potentially unobserved treatment indication bias, the author evaluated 203 patients who actually died of SMZL, with a median survival of 17 months. There was similarly no significant effect of splenectomy in this highest-risk subgroup (log-rank test P = 0.16).

The current report, based on the largest cohort of SMZL studied to date, shows no evidence of either beneficial or detrimental impact of splenectomy on survival. The overall survival was lower in this population-derived series than in some prior reports originating from academic centers, but the cause-specific (relative) survival of about 80% at 5 years was similar to larger studies [3]. Splenectomy, while commonly performed in the 1990s, has become diagnostically obsolete in most cases thanks to improved clinical, immunopathologic, and molecular criteria (such as recurrent 7q deletion) [12–15]. Its palliative benefits need to be weighed against surgical morbidity in elderly patients and subsequent risk of sepsis with encapsulated organisms. The operation is often delayed for several weeks to allow appropriate vaccinations. In contrast, rituximab-based chemoimmunotherapy can be initiated immediately and leads to durable improvement in almost all cases (response rate 87–100%, including complete remission in 35–69%) with a reported median time to clinical response of 3 weeks [9, 10]. Some studies suggest that splenectomy may induce potentially unfavorable changes in the lymphomatous bone marrow infiltration, although clinical significance of these findings is unclear [8, 16].

Observational studies can generally only study associations and generate hypotheses, but in rare diseases such as SMZL, in which randomized trials are not practicable, they may represent the only attainable level of evidence. Careful methodology is paramount in order to reduce group imbalance and consider competing events, since many patients die from comorbidities other than lymphoma—a fact frequently unaccounted for in prior literature.

One weakness of the present study is database limitation with regards to some clinical factors that influence eligibility for therapy. Although treatment decisions are typically guided by symptoms, laboratory values (hemoglobin, LDH, albumin) provide prognostic information [3]. Nevertheless, in previous literature they were not significantly different between patients undergoing splenectomy or not [5, 8]. Age and presence of B-symptoms affect cause-specific survival in SMZL, as confirmed by our study, although in this large cohort gender was not of prognostic value, contrary to some smaller studies [15, 17, 18]. PS analysis, while balancing a retrospective cohort akin to a randomized trial, relies on the assumption of ignorable treatment assignment and cannot remove confounding due to unmeasured factors. Indolent lymphomas can be managed with watchful waiting, sometimes for prolonged periods. Comorbidities and extent of malignancy are only approximated by SEER data and patients undergoing splenectomy might fare worse by virtue of symptomatic disease warranting treatment. Conversely, a proportion of patients not needing aggressive therapy may have operations for diagnostic purposes. Additional treatment indication bias could be mitigated by analyzing datasets containing information on all treatment modalities and time to progression. However, previous reports quite uniformly indicate that at least 75% of SMZL patients do require either splenectomy or systemic therapy at diagnosis [3, 5, 8, 19]. Over 90% of the present cohort was diagnosed in the era of availability of efficacious chemoimmunotherapy, the principal therapeutic alternative beyond supportive care, palliative radiation, or interferon (for hepatitis C-associated cases). In an attempt to reduce the indication bias, patients at the extremes of PS were removed from the analysis, a method partly compensating for unmeasured confounding related to medical frailty [20]. The sensitivity analysis conducted in the highest-risk group of patients who died of SMZL demonstrated a consistently negative result. It should be noted that because of the exclusion of cases at high and low extremes of the PS, the results of this study are applicable to patients at average risk, in which splenectomy and systemic therapy might be contemplated as viable alternatives.

In conclusion, this analysis demonstrates lack of association between splenectomy and survival in SMZL. This may alleviate potential concerns when deciding between systemic therapy and surgery. Splenectomy may be considered for palliation or rapid correction of severe cytopenias, but with ongoing diagnostic and therapeutic advances, it should be regarded a historical rather than gold standard, as was the case in hairy cell and chronic lymphoid leukemia. The differential benefits of chemoimmunotherapy and splenectomy with regards to quality of life and remission duration remain to be investigated.


Data source, variables, and outcomes

The SEER program collects information about cancer cases from 18 geographic areas representing 26% of the United States population. It contains demographic and histopathological information on the type of lymphoma, site of origin, extent (stage), survival time from diagnosis, and cause of death. Information about the first course of treatment is also available, defined as a documented management plan formulated at the time of diagnosis and delivered within 12 months. Watchful waiting is treated as an individual treatment plan. Interventions in case of recurrence or progression during initial therapy are not included. Treatment modalities are limited to surgery and radiation, with no differentiation between diagnostic and therapeutic procedures and no record of chemotherapy.

Relative survival tables and case listings of SMZL patients diagnosed between 1995 and 2009 were extracted from the SEER database using the SEER*Stat software (Surveillance Research Program, National Cancer Institute, version 7.0.5.) [21]. The initial query included codes 9689 (SMZL) and 9699 (marginal zone lymphoma) with spleen as the primary site (C42.2) according to the World Health Organization International Classification of Diseases for Oncology, 3rd Edition. In order to avoid overlap with other types of marginal zone lymphoma, patients were excluded if the primary site was specified as lymph nodes or bone marrow. From the initial 1,261 cases, those with a dubious record of “non B-cell” grade (T-, NK-, or null-cell, three cases), identified through autopsy or death certificate only (two cases) or with unknown splenectomy status (five cases) were also excluded. The International Prognostic Index was recorded for less than 5% of cases and was not analyzed.

The primary studied endpoint was the incidence of LRD. For its definition, the author used cause-specific events from the database, as well as all deaths caused by any form of lymphoma or leukemia. This endpoint was validated against actuarial relative survival adjusted to age, sex, and race in the general population, which is relatively immune to cause misattribution on death certificate [22].

Statistical analysis.

Group characteristics were compared using Fisher's exact or Wilcoxon rank sum tests. PS analysis was employed in order to reduce the treatment selection bias [23]. This two-step procedure involves fitting a model predicting the probability of receiving treatment and subsequent adjustment in the outcome analysis. When successful in balancing confounders, the methodology reduces systematic bias and enables causal inference regarding treatment effects, although it can control only for observed variables [24]. In some aspects it mirrors the design of a controlled study, allowing for definition of a subpopulation eligible for potential randomization (i.e., with overlapping propensity to undergo the intervention), checking appropriate covariate balance and comparing survival curves. The relative merits of PS analysis and conventional multivariate regression depend on specific settings and desired form of outcome analysis [25].

In this study, PS was derived from a logistic regression model through an iterative process (see Supporting Information). Variables entered on the basis of association with the treatment or the need for covariate balance, including statistically significant interactions. For case–control matching, the radius method was chosen with a caliper equal to 0.2 times pooled standard deviation of the logit of PS. Within the radius, cases were matched to multiple controls, which were assigned appropriate weights. This matching method retained more cases and controls in the analysis than one-to-one matching procedures, thus preserving statistical power. Cases outside of the common support region were trimmed, removing 5% of treated patients with the lowest density of matching controls. Covariate balance diagnostics were based on standardized differences of means, cumulative distribution plots, and box plots. Matching on linear PS removed confounding more efficiently than stratification and allowed estimation of treatment effect for the treated, the outcome of interest.

In order to account for the potential impact of splenectomy on LRD as well as unrelated causes of death, competing risk analysis was utilized, since Kaplan–Meier estimates are known to produce biased estimates in such settings [26, 27]. Cumulative incidence function curves between treated and untreated patients were compared using Pepe–Mori test [28]. Fine–Gray regression model was employed to study effects of splenectomy on probability of LRD (reported as subhazard ratio, SHR) and Cox proportional hazard model for overall survival (reported as hazard ratio, HR) [29, 30]. The models were fitted including covariates significantly influencing the outcome. Proportional hazard assumptions were tested using graphical methods and violating variables were evaluated with time interaction. All statistical tests are reported with two-tailed P-values at an alpha level of 0.05, with 95% CI, using Stata software (version 11.2, StataCorp, College Station, TX) and psmatch2 module (Edwin L, Barbara S. version 405 18apr2012: Boston College Department of Economics; 2003).


The author acknowledges the efforts of the National Cancer Institute (NCI), the North American Association of Central Cancer Registries (NAACCR), and the SEER program registries in the maintenance of the SEER database as a research resource. The author also thanks Dr. Vincent Mor for his helpful suggestions.

Adam J. Olszewski*, * Division of Hematology/Oncology, Memorial Hospital of Rhode Island, The Warren Alpert Medical School of Brown University, Providence, Rhode Island.