Patterns of treatment failure in Primary Central Nervous System Lymphoma

Abstract

Objectives:

Progression of PCNSL remains a challenge with salvage therapies including risk for substantial morbidity and mortality. We report patterns of first tumor progression to inform opportunities for improvement.

Methods:

This is an institutional retrospective review from 2002–2021 of 95 consecutive patients with pathologically confirmed PCNSL, of whom 29 experienced progressive disease. Kaplan-Meier method, Log-rank test, and Cox proportional hazard models used to characterize associations of patient, tumor, and treatment variables with LC, PFS, and patterns of first failure.

Results:

Most patients were <65yo (62%) with KPS >70 (64%) and negative CSF cytology (70%). In 70 patients with MRIs, median tumor volume was 12.6cc (range 0.5–67.8cc). After median follow-up of 11 months, 1-yr PFS was 48% and 1-yr LC was 80%. Of 29 patients with progression, 24% were distant only, 17% were distant and local, and 59% were local only. On MVA, LC was associated with age (HR 1.08 per yr, p=0.02), KPS (HR 0.10, p=0.02), completion of >6 cycles HD-MTX (HR 0.10, p<0.01), and use of intrathecal chemotherapy (HR 0.03, p<0.01). On UVA, local only first failure trended to be increased with >14cc tumors (OR 5.06, p=0.08) with 1-yr LC 83% (<14cc) vs 64% (>14cc). There were no significant associations with LC and WBRT (p=0.37), Rituximab (p=0.12), or attempted gross total resection (p=0.72).

Conclusions:

Our findings re-affirm the importance of systemic and intrathecal therapies for local control in PCNSL. However, bulky tumors trended to fail locally warranting further investigation about role of local therapies or systemic therapy intensification.

Introduction:

Primary Central Nervous System Lymphoma (PCNSL) is an uncommon intracranial malignancy with up to 4.32 cases per 100,000 for people 70 years old or older.¹ PCNSL represents approximately 4% of intracranial neoplasms and 4–6% of extranodal lymphomas.² Patients with PCNSL have a median overall survival (OS) ranging from months^1,3 to over 8 years.^4–6 Typical prognostic factors for PCNSL include: age, performance status, location of tumors, serum LDH, and CSF protein.^4–6 However, size of initial disease has not previously been incorporated despite prognostic and predictive importance in systemic lymphoma⁷ even in the modern era with cell therapies.⁸ While local therapies have shown benefit in management of bulky systemic lymphoma,^9,10 PCNSL has almost universally been treated with non-focal treatments such as systemic therapy, intrathecal (IT) therapy, or whole brain radiotherapy (WBRT).

The backbone of treatment is high-dose—often 3.5–8 g/m²—methotrexate (HD-MTX).¹¹ Trials have incorporated radiation¹², IT chemotherapy^13,14, temozolomide (TMZ)¹⁵, stem cell transplants¹⁶, or the MATRix regimen¹⁷ of HD-MTX, Cytarabine, Rituximab, and Thiotepa, with mixed results. PFS can vary dramatically with older regimens reporting median PFS of 12 months, while IELSG-32 reported a 2year PFS of up to 80% with WBRT.^12,13,16 Given the paucity of cases, prospective randomized data is lacking, and there remains debate over optimal management.

When utilized, standard consolidative radiation treatment includes whole brain radiotherapy (WBRT). However, this technique is associated with significant toxicity, particularly in older patients and following HD-MTX.^18,19 Focal radiation treatment for consolidation, as opposed to WBRT, may help prevent tumor progression of initially involved disease without significant morbidity. In order to assess the utility of this approach, we sought to characterize the patterns of first tumor progression following systemic therapy alone and to evaluate if there are associations between patient, tumor, and treatment factors with intra-cranial patterns of failure.

Materials and Methods:

This was a single institution retrospective cohort study of consecutive patients with pathologically confirmed PCNSL from 2002 to 2021. Study design RAD4901–19 approved by Clinical and Translational Research Committee and Institutional Review Board (IRB Study 00000331) with a waiver of informed consent. We evaluated patient (age, Karnofsky performance status [KPS], HIV, and solid organ transplant), tumor (CSF cytology, size/volume (cc) of contrast enhancing lesion[s], and number of lesions), and treatment (number of cycles of HD-MTX [0–5 vs ≥6], first line chemotherapy agents, intrathecal (IT) chemotherapy, use of whole brain radiation therapy (WBRT), and biopsy vs. gross total [GTR] or sub-total resection [STR]) factors as covariates. Outcomes included progression free survival (PFS), local control (LC) and pattern of first disease failure (local only compared to distant only or distant and local failure). Local failure was defined based on RANO criteria,²⁰ with first failure location determined to be local only if all sites of contrast enhancing disease are contiguous with at least partial overlap at initial area(s) of disease on pre-treatment MRI. Patients censored for LC if first failure non-overlapping with initial disease. A priori exploratory analyses were also performed evaluating volume of disease as a dichotomous variable using the standard size cutoff for eligibility for single fraction radiosurgery (14 cc tumor volume). We stratified non-bulky disease (<14 cc tumor volume) with bulky disease (≥14 cc tumor volume) in analysis of LC as well as Overall Survival (OS), and Progression Free Survival (PFS). PFS defined as time from biopsy to any progression or death, censored at last follow-up.

Descriptive statistics were generated for all covariates. Time to event outcomes were estimated using Kaplan-Meier method and compared using log-rank tests. Univariate (UVA) cox regression analysis was used to determine associations with LF, patterns of first failure, OS, and PFS. Variables that were significant at an alpha of 0.2 were used for multivariable analysis (MVA). A multivariable cox regression analysis by backward selection method was used to select the covariates. Statistical analysis was performed using SAS 9.4 (SAS Institute Inc., Cary, NC) macros,²¹ and statistical significance was defined a priori at the 0.05 level.

Results:

Patient, tumor, and treatment characteristics

We identified 95 patients with pathologically confirmed PCNSL treated at our institution from 2002–2021 and were included in analyses of OS and PFS. Of these patients, 70 had evaluable pre- and post-treatment MRI Brain scans and were included in the LC and patterns of failure analyses. Median age for entire cohort was 58 years (range 18–85) and median KPS was 70 (range 20–100). In total 10 had HIV (10.5%) and 4 had a history of solid organ transplant (4.2%). Demographic details are summarized in Table 1. Histology was Diffuse Large B-Cell (n=87, 91.6%), Marginal Zone (n=5, 5.3%), or T-Cell lymphomas (n=3, 3.2%). Of 60 patients with known CSF cytology assessments at diagnosis, 18 had positive cytology (30.0%). The median number of intracranial lesions was 2 (range 1–22). Median size of contrast enhancing disease was 12.6 cc (range 0.5–67.8 cc). All patients who received systemic therapy (n=62) were treated with at least one cycle of HD-MTX. Of the 62 patients who received HD-MTX, median dose of first cycle was 7.4 g/m2 (range 2.7–8.3 g/m2) and 35 patients completed at least 6 cycles (56.5%). WBRT was used in consolidation after chemotherapy or as monotherapy for patients with poor performance status or HD-MTX ineligible for 16 patients (19.5%). Maximal resection was attempted prior to pathologic confirmation of diagnosis in 25 patients (26.3%). Ultimately 13 had a gross total resection (13.7%) based on operative report and post-operative MRI, while 12 had residual disease after sub-total resection (12.6%). The remaining 70 patients had biopsy alone (73.7%). Summary of demographics in Table 1.

Table 1.

Variable	Level	No	%
Age	< 65 years old	59	62.1%
	≥ 65 years old	36	37.9%
Sex	Male	52	54.7%
	Female	43	45.3%
KPS	≥ 70	61	64.2%
	< 70	34	35.8%
CSF Cytology	Positive	18	30.0%
	Negative	42	70.0%
	Missing	35	-
Number of lesions	1	34	48.6%
	2+	36	51.4%
	Missing	25	-
Size of initial lesion(s)	< 14 cc	38	54.3%
	≥ 14 cc	32	45.7%
	Missing	25	-
Chemotherapy	HD-MTX	62
	HD-MTX + Ritux	38	61.3%
	HD-MTX + Ritux + TMZ	21	33.9%
	HD-MTX + Intrathecal chemo	12	19.4%
WBRT	None	54	65.9%
	1st Line	16	19.5%
	Salvage	12	14.6%
	Missing	13	-
Surgery	Biopsy	70	73.7%
	Gross Total Resection	13	13.7%
	Sub-total Resection	12	12.6%

KPS stands for Karnofsky Performance Status. CSF stands for cerebrospinal fluid. HD-MTX designates high dose methotrexate used in initial treatment of patient with or without other agents. HD-MTX + Ritux designates methotrexate and rituximab with or without other agents were used in initial treatment. HD-MTX + Ritux + TMZ designates methotrexate, rituximab, and temozolomide with or without other agents were used in initial treatment. HD-MTX + Intrathecal chemo designates that methotrexate and intrathecal chemotherapy with or without other agents were used in initial treatment. WBRT stands for whole brain radiotherapy

Overall outcomes

After a median follow-up of 11.0 months (range 0.2–193.7) the 1-year/2-year OS, PFS and LC rates were 55.5% (95% CI 44.0% - 65.5%) / 50.1% (95% CI 38.6% - 60.5%), 47.7% (95% CI 36.6% - 58.0%) / 38.5% (95% CI 27.9% - 49.0%), and 79.8% (95% CI 67.4% - 87.9%) / 72.8% (95% CI 58.8% - 82.7%), respectively (Figure 1). On MVA, LC was significantly associated with completion of 6 or more cycles of HD-MTX (p=0.001), use of intrathecal chemotherapy (p=0.033), and use of Rituximab (p=0.036). UVA and MVA associations with LC are summarized in Table 2.

Figure 1.

Kaplan-Meier curve of Progression Free Survival (a) and Local Control (b) over time for entire cohort.

Table 2.

Associations with Local Control

		UVA		MVA
Variable	Level	HR (95% CI)	p-value	HR (95% CI)	p-value
Age	Continuous	1.06 (1.02–1.10)	0.004	1.03 (0.99–1.07)	0.130
KPS	≥ 70	-	-	-	-
	< 70	2.07 (0.88–4.87)	0.097
Immunodeficiency	Yes	0.68 (0.09–5.12)	0.711
	No	-	-
CSF Cytology	Negative	0.91 (0.33–2.51)	0.849
	Positive	-	-
Number of Lesions	2+	1.40 (0.59–3.34)	0.441
	1	-	-
Size of Lesions	≥ 14 cc	1.41 (0.60–3.34)	0.430
	< 14 cc	-	-
Cycles of HD-MTX	≥ 6	0.28 (0.11–0.71)	0.008	0.18 (0.06–0.51)	0.001
	< 6	-	-	-	-
Use of Rituximab	Yes	2.74 (1.04–7.23)	0.041	3.20 (1.08–9.50)	0.036
	No	-	-	-	-
Use of Temozolomide	Yes	2.48 (1.00–6.12)	0.049
	No	-	-
Intrathecal Chemotherapy	Yes	0.15 (0.02–1.12)	0.065	0.11 (0.01–0.83)	0.033
	No	-	-	-	-
1st Line WBRT	Yes	0.50 (0.11–2.19)	0.357
	No	-	-
Surgery	Biopsy	0.90 (0.29–2.77)	0.848
	GTR	0.73 (0.13–4.13)	0.720
	STR	-	-

UVA and MVA stand for univariate and multivariable analyses. HR stands for hazard ratio. CSF stands for Cerebrospinal Fluid. HD-MTX stands for high dose methotrexate. WBRT stands for whole brain radiotherapy. GTR stands for gross total resection. STR stands for sub-total resection. Backward selection with an alpha level of removal of 0.2 was used. The variables KPS and Use of Temozolomide were removed by selection criteria.

Patterns of Failure

There were 35 patients with progression, of whom 29 patients had evaluable pre- and post-treatment scans. Of the 29 evaluable patients: 17 patients had local only progression (59%), 7 patients had distant only (24%), and five patients had local and distant progression (17%). The median time to progression was 13.6 months (IQR 3.8 – 56.1). Local only progression trended to be higher with ≥14 cc tumors (OR 5.06, 95% CI: 0.83–30.75, p=0.078) and increased age (OR 1.06, 95% CI: 1.00–1.14, p=0.060). There was no association with negative CSF (OR 0.35, 95% CI: 0.05–2.41, p=0.286), multiple lesions (OR 1.71, 95% CI: 0.37–7.92, p=0.490), or completion of 6 or more cycles of HD-MTX (OR 0.50, 95% CI: 0.11–2.24, p=0.365). UVA odds ratios for pattern of first failure summarized in Table 3. Odds Ratios for progression with patients who had HIV, solid organ transplant, upfront WBRT, GTR, STR, or IT chemotherapy use are not reported due to small sample size. MVA not performed due to limited sample size.

Table 3.

Odds ratios for pattern of first failure (local only vs distant with or without local failure).

			UVA
Variable	Level	n (%)	OR (95% CI)	p-value
Age	Continuous	29 (100%)	1.06 (1.00–1.14)	0.06
KPS	≥ 70	18 (62%)	-	-
	< 70	11 (38%)	2.67 (0.43–13.43)	0.234
CSF Cytology	Negative	15 (68%)	0.35 (0.05–2.41)	0.286
	Positive	7 (32%)	-	-
Number of Lesions	2+	15 (54%)	1.71 (0.37–7.92)	0.490
	1	13 (46%)	-	-
Size of Lesions	≥ 14 cc	11 (39%)	5.06 (0.83–30.75)	0.078
	< 14 cc	17 (61%)	-	-
Cycles of HD-MTX	≥ 6	14 (48%)	0.50 (0.11–2.24)	0.365
	< 6	15 (52%)	-	-
Use of Rituximab	Yes	19 (66%)	1.71 (0.36–8.08)	0.496
	No	10 (34%)	-	-
Use of Temozolomide	Yes	12 (41%)	1.78 (0.38–8.23)	0.462
	No	17 (59%)	-	-

UVA and MVA stand for univariate and multivariable analyses. HR stands for hazard ratio. CSF stands for Cerebrospinal Fluid. HD-MTX stands for high dose methotrexate. WBRT stands for whole brain radiotherapy. GTR stands for gross total resection. STR stands for sub-total resection.

Bulky Disease Outcomes

Of the 29 patients with progression and available imaging, 11 had bulky disease (37.9%). Of the 11 patients who progressed with bulky disease, 9 had local only first failure (81.8%), 1 had distant only failure (9.1%), and 1 had distant and local failure (9.1%). Of the 70 patients with available imaging, there was no statistically significant differences in PFS or LC for patients with bulky disease (≥14 cc) compared to patients with non-bulky disease (Figure 2). However, numerically there was worse PFS for patients with bulky initial disease at 1-year (39.1% vs 56.9%), 2-years (32.0% vs 47.9%), and 5-years (17.8% vs 29.9%). Similarly, LC was worse for patients with bulky initial disease at 1-year (64.3% vs 83.4%), 2-years (58.4% vs 74.1%), and 5-years (48.7% vs 69.5%).

Figure 2.

Kaplan-Meier curves comparing bulky (≥14 cc) and non-bulky (<14 cc) disease among the 70 patients with available pre-treatment MRI Brain scans for Progression Free Survival (a) and Local Control (b).

Discussion:

We report on patterns of failure in a large single-institution cohort of patients with PCNSL. Overall, 2-year LC was 72.8% with better LC significantly associated with completion of induction HD-MTX and use of intrathecal chemotherapy. While comparisons between bulky and non-bulky disease are likely underpowered, patients with bulky disease (>14 cc) had numerically but not significantly worse outcomes. Furthermore, pattern of first failure was predominantly at initial site of disease (59% of patients). There was a trend towards higher risk of local only failure in patients with initially bulky disease, and 81% of patients who failed with bulky disease had local only site of first failure. This highlights a potential opportunity to improve PFS through local control strategies and identifies a subgroup who may have the most to benefit from this approach.

Historically and in countries with limited ability to offer treatments such as autologous stem cell transplants, WBRT can provide durable responses and intracranial control.^11,12,22 However, the most effective therapy to date is HD-MTX and when combined with WBRT can lead to severe neurotoxicity in as many as 49% of patients, almost double the rate of neurotoxicity as HD-MTX alone.¹² Focal radiation such as stereotactic radiosurgery, however, is known to be substantially less toxic when compared to WBRT in the management of brain metastases from solid tumors.^23–26 Furthermore, focal radiotherapy is standard of care for select patients with systemic lymphoma. Even with Rituximab, involved site radiation remains important for patients with systemic DLBCL with initial bulky disease¹⁰ or with poor response to induction therapy.²⁷ Consistent with a prior report of patients with progressive PCNSL, 66% of first progressions are at initial site of disease only,²⁸ our study showed that 59% of patients had progression at site of initial disease only. This highlights potential benefit for local therapies in selected patients. Alvarez-Pinzon et al showed in a retrospective study that Gamma Knife radiosurgery to 11–16 Gy—a safe radiotherapy dose even in sensitive areas of the brain—combined with HD-MTX led to median OS of 47.6 months with 100% complete response rate at 3–8 weeks after radiosurgery.^29,30 However, the tumors treated ranged from 1.5–6.4cc, which is substantially smaller than the average patient treated in our cohort. Furthermore, in a series of 38 patients treated from 1985–1994 with fractionated radiation to 30–70 Gy to partial brain, margins of 1–3.5 cm were associated with worse outcomes compared to patients treated with 4–5.5 cm margins (out-of-field recurrences in 83% vs 22%, p=0.0079).³¹ Given the reduced neurocognitive deficits of focal RT compared to WBRT,^25,32, the addition of consolidative focal RT to select patients may improve tumor control without adding significant morbidity.

The role of neurosurgery in care of PCNSL has typically been reserved to confirming pathology with a biopsy. Older studies showed no benefit of aggressive surgical resection, particularly STR.^3,33 However, an unplanned secondary analysis of German PCNSL Study Group-1 trial showed that resection improved OS and PFS, albeit p-value increased to 0.085 when controlling for number of lesions.³⁴ Using the National Cancer Database and Surveillance, Epidemiology, and End Results Program, Rae et al showed statistically significant increases in OS with use of craniotomy, particularly for patients without high risk features.³⁵ While our cohort showed no statistically significant benefit of GTR or STR, there were only 13 and 12 patients respectively who received each. As such further work is needed to identify patients who may benefit from local therapies before changing practice patterns.

The importance of HD-MTX cannot be overstated in the management of PCNSL. In this cohort we showed that completion of induction HD-MTX is critically important for local control, consistent with NCCN recommendations.¹¹ While systemic malignancies may benefit from combination of radiotherapy and systemic therapies,³⁶ Multiple systemic therapies have been added to a backbone of HD-MTX including rituximab,¹⁷ temozolomide,¹⁵, stem cell transplants,¹⁶ and other combinations.^13,14 In our study, we demonstrated improved LC with intrathecal therapy, but no benefit with use of temozolomide and detriment with addition of rituximab. Ongoing trials are investigating new agents for induction therapy such as lenalidomide, ibrutinib, or voraxaze^37,38 as well as multiple studies employing immune checkpoint inhibitors for relapsed or refractory disease and other techniques to optimize immune response.^39–41

This study has multiple limitations. As a retrospective, single institution study, practice patterns were not controlled and many patients had limited follow-up. Given the aggressive nature of PCNSL, our LC rates may be an over-estimate due to censoring. Some patients were initiated treatment at outside facilities and treated at relapse with incomplete details about initial staging and treatment, which could confound our findings. Additionally, limited sample size could result in Type II errors regarding impact of initial tumor burden or therapies on LC and patterns of first failure.

Our study re-affirms that systemic and intrathecal therapies are critical for local control in PCNSL. The majority of treatment failures included failure at sites of initially involved disease alone, suggesting local consolidative therapy such as focal radiation may improve tumor control for a subset of patients. Patients with larger volume tumors demonstrated the highest risk of local failure, though this study is limited by small sample size. While additional research may further characterize optimal patient selection, prospective clinical testing will be needed to guide the role of focal therapies such as radiosurgery in the management of selected PCNSL patients.