Time from diagnosis to 2nd treatment is a promising surrogate for overall survival in patients with advanced stage follicular lymphoma

Abstract

It is difficult to demonstrate an overall survival (OS) benefit in trials of immediate therapy vs observation in follicular lymphoma (FL). Time to 2^nd treatment (TT2T) may be a preferred endpoint. We identified 584 consecutive patients at our institution with advanced stage FL grade 1–3A for whom intention was observation (n=248) or therapy (n=338). Median time to 1st treatment (TT1T), TT2T, and OS were estimated (subdistribution function). Modified Kendall's tau (mKτ) was used to assess correlation between survival endpoints. Among initially observed patients, median TT1T was 3.3 years, TT2T was 12.1 years, 10-year treatment-free survival was 23%, and 10-year OS was 82%. TT2T was strongly correlated with OS following initial observation (mKτ 0.46, p=0.004) or therapy (mKτ 0.53, p<0.0001), while duration of observation was not. TT2T is a potential surrogate for OS. Given the outstanding survival in this population, early intervention trials should focus on identifying high risk patients.

Introduction

Therapy may be safely deferred in asymptomatic patients with advanced stage follicular lymphoma (FL) until onset of symptoms or organ failure without compromising overall survival (OS).^1-6 Given the outstanding OS among patients with FL in the rituximab era, and in light of recent reports that patients who are initially observed might have an expected survival comparable to the general population,^7,8 it is difficult to adequately power a trial of early therapy vs initial observation for OS.

Predicting OS using surrogate endpoints has been a major challenge in randomized trials of patients with follicular lymphoma. Time to New Treatment (TTNT) and Progression-Free Survival (PFS) have been used in trials comparing early therapy vs initial observation, but the value of these endpoints is debatable given the lack of a difference in OS.^3,4 Similarly, PFS has not reliably correlated with OS in previously untreated patients with FL on randomized trials evaluating novel treatment approaches.^9-12 We therefore sought to define a novel surrogate endpoint that would correlate with OS and be a more appropriate primary endpoint in trials of initial observation vs immediate therapy.

We hypothesize that TT2T is an accurate surrogate for survival. The duration of response following initial therapy, reported as event-free survival at 12 months (EFS12) and progression-free survival at 24 months (PFS24), identifies a cohort of FL patients with poor survival.^7,13 TT2T captures this duration of clinical benefit following initial therapy, and thus might be more strongly associated with survival compared to endpoints such as TT1T and PFS that only capture a pre-treatment interval among initially observed patients. Furthermore, we posit that TT2T allows comparison of populations with balanced treatment exposure and is therefore a more appropriate primary endpoint in trials of initial observation vs immediate therapy.

We therefore performed a comprehensive analysis of survival endpoints in patients with follicular lymphoma to determine the utility of Time from Diagnosis to 2^nd Treatment (TT2T) as a relevant endpoint for trials testing immediate therapy vs initial observation. In addition, we evaluated TT2T as a surrogate for OS.

Patients and Methods

Patients

This retrospective research protocol was approved by the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Review Board, and a waiver of consent was granted due to its retrospective nature. We identified 883 consecutive patients diagnosed with FL at MSKCC from 1998 to 2007. The cutoff was chosen to ensure adequate follow-up to assess TT2T. We performed chart review to determine whether the physician recommended observation (n=340), therapy (n=542), or unknown (n=1). Of patients recommended observation (n=340), we excluded stage I/II disease (n=93), grade IIIB histology (n=1), or surgical resection (n=1). The intent to observe was based on clinician recommendations, which generally follow Groupe d’Etude des Lymphomes Folliculaires (GELF) criteria at our institution,^2,14 however GELF criteria were not used as the basis for selection in this retrospective analysis. Of patients recommended immediate therapy (n=542), we excluded stage I/II disease (n=184), grade IIIB histology (n=26), histologic transformation at diagnosis (n=2), or concurrent active cancer (n=1). The final Intent to Observe and Intent to Initiate Therapy Cohorts included 246 and 338 patients, respectively (Supplementary Fig 1).

Statistical methodology

Patient and disease characteristics were summarized using descriptive statistics, and frequencies and proportions were calculated. TT1T, TT2T, and OS were defined from time of follicular lymphoma diagnosis. OS was estimated using the Kaplan-Meier method, and TT1T and TT2T were estimated using the subdistribution function to adjust for semi-competing risk of death on treatment initiation.¹⁵ TT2T was defined a priori for this retrospective analysis. A modified proportional hazards model was used to compare the subdistribution hazard between cohorts.¹⁶ The standard Cox proportional hazards model was used to compare the hazard of death between cohorts.

Modified Kendall’s tau (modified Kτ) was used to assess the correlation between TT1T, TT2T, and OS, accounting for censoring and semi-competing risk in these quantities.¹⁷ Tests of modified Kendall’s tau against 0 (i.e. no correlation) were performed, with tau defined as the concordance probability minus the discordance probability.¹⁸ In the landmark analyses, OS was calculated using the landmark as the time origin.

Expected survival accounting for age- and gender-matched healthy population were calculated using the general US population data (https://wonder.cdc.gov/ucd-icd10.html) and the conditional method¹⁹ with R function survexp in package survival, and compared to the survival rates in our patient cohorts using standardized mortality ratios (SMR) of observed to expected deaths. Chi-squared test was used to compare the expected and observed survival rate.

All statistical tests were two-sided at a significance level of 0.05. Statistical analyses were performed using R 3.3.1 for Windows.²⁰

RESULTS:

Patient Characteristics

Patient characteristics are in Table 1. Of 246 patients who were initially observed, there was a slight female predominance (1.16:1) and the median age was 56 years (range 25-88). Of 338 patients who were recommended immediate therapy, there was a slight male predominance (1.18:1) and median age was 56 years (range 22-89). Patients who were recommended immediately therapy were more likely to have high risk FLIPI scores (42.9% vs 22%, p<0.001).

Table 1:

Patient characteristics at diagnosis

Baseline characteristics	Initial Observation	Immediate Therapy	p value
	n=246	n=338
Age at diagnosis
Median (range)	56 (25-88)	56 (22-89)	0.236
Gender
Female	132 (53.7%)	155 (45.9%)	0.075
Male	114 (46.3%)	183 (54.1%)
FLIPI Score
High	54 (22.0%)	145 (42.9%)	< 0.001
Intermediate	81 (32.9%)	85 (25.1%)
Low	43 (17.5%)	33 (9.8%)
Missing	68 (27.6%)	75 (22.2%)
Number nodal areas
≥ 5	117 (47.6%)	189 (55.9%)	0.051
< 4	124 (50.4%)	143 (42.3%)
Missing	5 (2.0%)	6 (1.8%)
Lactose dehydrogenase
Elevated	21 (8.5%)	80 (23.7%)	<0.001
Not elevated	144 (58.5%)	164 (48.5%)
Missing	81 (32.9%)	94 (27.8%)
Hemoglobin
< 12 g/dL	11 (4.5%)	65 (19.2%)	<0.001
≥ 12 g/dL	202 (82.1%)	227 (67.2%)
Missing	33 (13.4%)	46 (13.6%)
Date of diagnosis
Range	01/1998 - 12/2007	01/1998 - 12/2007
Treatment history
1st-line therapy administered	179 (72.8%)	338 (100%)	<0.001
Rituximab plus chemotherapy	90 (50.3%)	224 (66.3%)
Rituximab monotherapy	64 (35.8%)	48 (14.2%)
Chemotherapy	15 (8.4%)	61 (18%)
Radiotherapy alone	5 (2.8%)	1 (0.3%)
Other	5 (2.8%)1	4 (1.2%)2
2nd-line therapy administered	101 (41.1%)	191 (56.5%)	<0.001
Rituximab plus chemotherapy	41 (40.6%)	77 (40.3%)
Rituximab monotherapy	41 (40.6%)	61 (31.9%)
Chemotherapy	7 (6.9%)	18 (9.4%)
Radiotherapy alone	7 (6.9%)	14 (7.3%)
Radioimmunotherapy	4 (4%)	13 (6.8%)
Surgical resection	1 (1%)	2 (1.0%)
Other	0 (0%)	6 (3.1%)3

Baseline characteristics are shown by cohort.

⁽¹⁾Includes splenectomy (n=1) or enrollment on clinical trial with targeted (n=2), immune, (n=1) or antimicrobial therapy (n=1)

⁽²⁾Includes splenectomy (n=2) or enrollment on clinical trial with targeted therapy (n=2)

⁽³⁾Includes enrollment on clinical trial with targeted (n=5) or immune therapy (n=1).

Survival Outcomes

At a median follow-up of 9.6 years, patients recommended initial observation have a median TT1T of 3.3 years (95% CI 2.8-3.8) and 10-year treatment free survival of 23% (95% CI 0.18-0.29). The median TT2T is 12.1 years (95% CI 10.7-16.8). The 10-year OS is 82% (95% CI 0.77-0.88) and the median OS is not reached (Fig 1). The observed mortality rate of initially observed patients exceeds the matched US general population (SMR 1.56, 95% CI 1.15-2.07, p=0.002, Supplementary Fig 2).

Figure 1:

Time from diagnosis to 1^st treatment, 2^nd treatment, and death in initial observation cohort

TT1T, time to 1st therapy; TT2T, time from diagnosis to 2nd treatment; OS, overall survival. TT1T, TT2T, and OS, of patients who are initially observed are shown.

At a median follow-up of 9.1 years, patients recommended immediate therapy have a median TT2T of 5.2 years (95% CI 3.8-8.3). The 10-year OS is 74% (95% CI 0.69-0.79) and the median OS is not yet reached (Supplementary Fig 3). The observed mortality rate in immediately treated patients exceeds the matched US general population (SMR 1.70, 95% CI 1.36-2.10, p<0.001, Supplementary Fig 2).

TT2T but not TT1T is Correlated with Overall Survival

We directly assessed the association between TT1T, TT2T, and OS (Table 2, Fig 2). TT1T is not correlated with OS among patients who are recommended initial observation (modified Kτ 0.055, 95% CI −0.19-0.22, p=0.6), and is moderately correlated with OS among patients who are recommended immediate therapy (modified Kτ 0.20, 95% CI 0.045-0.31, p=0.016).

Table 2:

Association of Time to 1st and Time to 2nd Treatment with Overall Survival

	Association	modified Kτ	95% CI	P value
Initial Observation	TT1T and OS	0.055	−0.19 to 0.22	0.603
	TT2T and OS	0.46	0.22 to 0.59	0.004
Immediate Therapy	TT1T and OS	0.20	0.05 to 0.31	0.016
	TT2T and OS	0.53	0.40 to 0.61	<0.001

TT1T, time to 1^st therapy; TT2T, time from diagnosis to 2^nd treatment; OS, overall survival; modified Kτ, modified Kendall’s tau. TT2T is strongly correlated with OS among patients who are recommended initial observation or immediate therapy, while TT1T is not.

Figure 2:

Graphical representation of the modified Kendall’s tau analysis. Each dot represents one of eight hundred randomly chosen pair of subjects. The difference in the surrogate endpoint (TT1T or TT2T) between the two subjects is represented on the X axis and the difference in OS is represented on the Y axis. When Kendall’s tau is high (i.e. strong correlation), the two differences are more likely to have the same sign, and thus more dots are seen in the first or third quadrant.

TT2T is strongly correlated with OS among patients who are recommended initial observation (modified Kτ 0.46, 95% CI 0.22-0.59, p=0.004) as well as among patients who are recommended immediate therapy (modified Kτ 0.53, 95% CI 0.40-0.61, p=<0.0001). While both TT1T and TT2T are correlated with OS in patients who are recommended immediate therapy, the correlation between TT2T and OS is of a much greater magnitude and strength.

Failure to Achieve a Time to 2^nd Treatment of 4 years is associated with Inferior Survival

We undertook a landmark analysis in the Intent to Observe Cohort comparing OS between patients who achieved a TT2T of 4 years and those who did not. We selected the 4 year cutoff to target a failure rate of 15-20%, as previous landmark endpoints associated with high risk biology similarly capture 15-20% of patients.^7,13 Among 211 patients evaluable for this landmark analysis (i.e. alive and not censored at 4 years), 39 (18.5%) failed to achieve a TT2T of 4 years. Patients with TT2T <4 years had an increased risk of death (HR=2.1, 95% CI: 1.1 to 4.2, p=0.027, Supplementary Fig 5) compared to those with TT2T ≥4 years, and a 3.6-fold increased mortality compared with the general US population (SMR 3.6, 95% CI 1.9-6.3, p<0.001). The subsequent 5- year and 10-year OS among patients with TT2T <4 years was 78% (95% CI 66-93%) and 58% (95% CI 40-82%), respectively.

Treatment Free Survival of 4 years is associated with Reduced Likelihood of Requiring Therapy but Not Survival

We identified a clear change in the slope of the subdistribution function curve for TT1T, with a transition point occurring between 4-5 years (Fig 1). The event rate decreases dramatically suggesting the subset of patients who remain treatment-free beyond this landmark might have unique disease biology and a low risk of requiring therapy. We undertook a landmark analysis in the Intent to Observe Cohort comparing OS between patients who initiated therapy versus those remaining treatment-free at 4 years from diagnosis. This allowed us to estimate the likelihood of requiring therapy among patients who are treatment-free at 4 years. We selected the 4-year landmark because it provides a direct comparison with the landmark analysis with TT2T in the previous section, however we conducted a sensitivity analysis using the 5-year landmark which confirmed similar results. Among 211 evaluable patients (i.e. alive and not censored at 4 years) for the landmark analysis, 89 (42.2%) remained treatment-free and 122 (57.8%) required therapy within 4 years. There was no difference in OS between these groups (p=0.64, Supplementary Fig 4), which is consistent with the finding that no association exists between TT1T and OS among patients who are initially observed. Among patients who achieved a treatment-free survival of 4 years, the probability of not requiring any treatment in the next 5 years is 57.5% (95% CI 47.4-69.8%).

Time to New Treatment is a Poor Measure of Clinical Benefit

Because TTNT has been the primary endpoint for trials comparing early therapy vs initial observation in FL,^21,22 we next compared TTNT between cohorts as a measure of treatment effect (Supplementary Fig 6). Direct comparisons of the effect of immediate therapy vs initial observation using these data are biased because the therapy and observation cohorts were selected in this retrospective analysis based on physician recommendation. However, comparisons of relationships between the surrogate endpoints (TTNT and TT2T) with OS are informative. As expected, patients who require immediate therapy at diagnosis have an inferior OS (HR 1.4, 95% CI: 1.0 to 2.0, p=0.047) when compared with those recommended initial observation, but a comparison of TTNT reveals that this patient population has a superior TTNT (subdistribution HR 0.7, 95% CI: 0.5 to 0.8, p<0.0001). When we compare the two cohorts using TT2T, we find that this patient population has an inferior TT2T (subdistribution HR 1.8, 95% CI: 1.4 to 2.3, p<0.0001), which correlates with OS.

DISCUSSION

Early treatment has not been shown to improve overall survival among patients with follicular lymphoma who do not meet established criteria for treatment, thus the standard remains initial observation until onset of symptoms or organ failure.^1,2,4-6,22 We have described Time from Diagnosis to 2^nd Treatment (TT2T) as a novel primary endpoint for trials evaluating immediate therapy vs initial observation, which provides balance in treatment exposure. In addition, we have demonstrated that TT2T is strongly correlated with OS. Given these data, TT2T might be a more reliable surrogate for OS for both patients suitable for initial observation and those needing immediate treatment.

In our series, the median TT2T is 12.1 years among patients who are initially observed. We have shown that TT2T is strongly correlated with OS making it a promising surrogate for OS. This is probably due to the prognostic weight of the response duration following upfront therapy. Failure to achieve EFS12 or PFS24 following initial therapy predicts for poor survival,^7,13 and this critical period of clinical benefit following initial therapy is captured by TT2T but not by TT1T, TTNT, or PFS. These data are also expected based on an intuitive explanation: as n approaches the maximum number of therapies, Time from Diagnosis to Treatment n (TDnT) should approach OS. While our data demonstrate that TT2T is a potential candidate surrogate for OS, more is needed to rigorously establish surrogacy.²³

Patients who require immediate therapy have an inferior OS compared with those who are observed, which is an expected finding due to confounding by indication. However, the data presented above demonstrate that TTNT was superior among patients receiving immediate therapy compared with those who were initially observed. This finding demonstrates the limitation of unbalanced endpoints such as TTNT or PFS in a randomized trial when the comparator group is not receiving active therapy. Indeed, TTNT and PFS tend to favor the immediate therapy arm independent of clinical benefit or harm. When we compare the two cohorts using TT2T, we find that patients who received immediate therapy have an inferior outcome, which correlates with inferior OS. This analysis was not performed to directly compare the two cohorts. Rather, these data emphasize that comparison of TT2T better reflected long-term clinical benefit. Early therapy only benefits patients if it leads to a TT2T benefit (which requires an increase in post-treatment remission and observation that exceeds the period of initial observation forfeited with immediate therapy). Otherwise, early treatment might harm patients by unnecessarily exhausting available treatments and exposing 23% of patients to therapy they might otherwise never receive. These data support our assertion that TT2T allows comparison of populations with balanced treatment exposure and might serve as a more appropriate surrogate endpoint in trials of initial observation vs immediate therapy.

Though TT2T is a promising surrogate for OS which provides a balanced endpoint for trials of immediate therapy versus observation, the median TT2T among patients with FL who are initially observed is 12.1 years. This endpoint is clearly too long to be used in a prospective trial. Therefore, we performed a landmark analysis to identify an earlier endpoint for assessing outcomes in trials of early therapy vs initial observation in FL. Failure to achieve a TT2T of 4 years identifies a cohort of initially observed patients with FL who have inferior outcomes compared to those who achieve a TT2T of 4 years. Patients who fail to achieve a TT2T of 4 years have a subsequent 10-year OS of 58% and a 3.6-fold increased risk of death compared with the general US population. TT2T at 4 years may provide prognostic insight, but requires external validation before it can be used in trials of early therapy vs initial observation in FL, or for identifying patients who might benefit from a more aggressive treatment approach.

Duration of initial observation is not prognostic for OS, which is likely due its vulnerability to lead time bias. The median duration of observation in our series is 3.3 years, and the probability of remaining treatment-free at 10 years is 23%. These data are consistent with previous reports.^4,22 We identified a clear transition point between 4-5 years followed by a much lower event rate, suggesting the hypothesis that patients who remain treatment-free beyond this landmark might have unique disease biology. However, the landmark analysis of treatment-free survival at 4 years demonstrated no difference in OS (p=0.64), thus this inflection point is not a reflection of favorable underlying biology but rather a difference in the need for initial therapy. Among patients who achieved a treatment-free survival of 4 years, only approximately 40% went on to require therapy. These data are useful for counseling patients, and might support less frequent surveillance imaging for patients who remain treatment-free beyond 4 years.

In the current series, patients with advanced stage FL who are initially observed have an OS that is slightly inferior to the age- and gender-matched US general population. In a similar analysis from the Mayo Clinic, initially observed patients had an OS comparable to the general population.⁷ This discrepancy might also represent patient selection bias between institutions. We also excluded stage I/II FL (32% of patients in the Mayo Clinic analysis).⁷ Notably, the overall population in our series (both initially observed and immediately treated) had an inferior survival relative to the general US population (SMR 1.65, 95% CI 1.38-1.95, p<0.001), so this discrepancy cannot be fully attributed to a difference in the threshold to recommend immediate therapy and likely represents differences in background survival between the treated populations.

Overall survival is the definitive endpoint for the demonstration of benefit in therapeutic trials in indolent non-Hodgkin lymphoma, and we have shown that TT2T is a promising surrogate for OS. However, the outstanding OS and TT2T among patients with FL who are initially observed calls into question the viability of any early therapy approach in an unselected population of patients with FL who do not meet established criteria for treatment, which would require thousands of patients and decades of follow-up to achieve adequate statistical power. Future trials of early therapy would need to restrict eligibility to high risk patients expected to have inferior survival. Efforts are ongoing to identify high risk patients with FL at diagnosis (e.g. m7-FLIPI²⁴ and intrafollicular proliferation index²⁵).

Our analysis is limited by its single-institution design without a validation cohort and requires external validation. Additionally, to rigorously establish TT2T as a surrogate for OS, these findings should be validated in a post hoc analysis of a prospective trial with randomization of patients not meeting established criteria for treatment to immediate therapy vs initial observation.⁴ Another potential limitation is that our analysis included patients diagnosed in the pre-rituximab era, but the vast majority of patients received rituximab-based therapy as initial systemic therapy. Given additional limitations inherent in retrospective studies (e.g. selection bias and confounding by indication), we also did not explore the impact of first line therapy following observation on TT2T. However, these data can be extrapolated from existing literature.

In conclusion, Time from Diagnosis to 2^nd Treatment (TT2T) with a 4-year landmark analysis is proposed as a preferential primary endpoint for trials evaluating immediate therapy vs observation in FL, and warrants evaluation in other malignancies that may be initially observed (e.g. chronic lymphocytic leukemia, smoldering myeloma). TT2T is also a potential surrogate for OS in FL, although more is needed to rigorously establish surrogacy. Duration of initial observation is not prognostic for OS, and endpoints like TTNT or PFS cannot exclude the possibility of harm in the absence of an OS benefit. Existing randomized trials evaluating immediate therapy vs observation in FL should be reanalyzed with TT2T as the endpoint. Future trials might restrict eligibility to high risk patients expected to have inferior survival, and efforts are ongoing to develop biomarkers that identify a population from diagnosis enriched with high risk patients who might be more likely to benefit from early therapy.