Features Associated with Survival in Metastatic Melanoma Patients Treated with Patient-Specific Dendritic Cell Vaccines

Abstract

Previously, a 54% 5-year survival was reported for metastatic melanoma patients treated with patient-specific vaccines consisting of autologous dendritic cells loaded with antigens from autologous proliferating tumor cells. This study attempted to determine which clinical and laboratory factors best explained long-term survival in this group of patients. Univariate analyses were used to identify factors associated with continuous survival after initiating vaccine therapy. Multivariate logistic regression was used to identify independent factors to classify survival at 3.5 years. Survivors were followed a minimum of 3.7 years (median: 5.7). Univariate analyses identified eight features associated with improved survival: Eastern Cooperative Oncology Group (ECOG) performance status (PS) of 0, no measurable disease at study entry, receiving 8 vaccinations, age <50 years, normal baseline lactate dehydrogenase, no history of visceral metastases, anergy to standard skin tests, and failure of interferon-gamma (IFN-γ) to induce apoptosis in autologous tumor cells. After examining 54 variables for which complete information was available over all patients, the best multivariate regression for survival at 3.5 years utilized six features: prior radiation therapy, younger age, male gender, ECOG PS 0, higher numbers of cells administered during the first 3 injections, and lower numbers of viable cells administered during the first 3 injections. This model correctly classified survival for 28 of 32 patients (87%) and death for 20 of 22 (91%). When features with incomplete information were included in the analysis, addition of IFN-γ-induced apoptosis (n=49) improved predictive accuracy to 27 of 29 (93%) for survival and 19 of 20 (95%) for death. Dependencies between variables were common, but these multivariate linear models yielded high classification accuracy for survival at 3.5 years and identified two features of the vaccine itself as being of independent significance.

Introduction

Despite current standard therapies, metastatic melanoma is typically a lethal disease. Most therapeutic progress in the past two decades resulted from biological products with immunotherapy effects, including interferon (IFN)-alpha, interleukin-2 (IL-2), and the anti-cytotoxic T lymphocyte-4 monoclonal antibody ipilumumab.¹ Vaccine strategies have been especially attractive, because they are typically associated with minimal side effects compared with other approaches and could induce endogenous antitumor effects, but so far there are no vaccine products with proven benefit.¹

Theoretically, one of the best strategies for a vaccine approach is active specific immunotherapy utilizing antigens from autologous proliferating tumor cells. Previously, lengthy progression-free survival was reported in a subset of patients treated with patient-specific autologous dendritic cells loaded with antigen from such cells,² and an encouraging 54% 5-year survival rate was found for all patients.³ In this single-arm, open label, phase II trial, 54 metastatic melanoma patients were treated with repeated injections consisting of 500 μg of granulocyte-macrophage colony stimulating factor (GM-CSF) admixed with autologous dendritic cells that had been loaded with antigens by incubating the dendritic cells with proliferating autologous tumor cells derived from a short-term cell culture.³ The treatment plan consisted of 3 weekly subcutaneous vaccinations followed by 5 monthly vaccinations for a total of 8 vaccinations. Pursuant to the previous report, a series of univariate analyses were performed in this study to determine which clinical and laboratory features were associated with better survival, and a multivariate logistic regression analysis was carried out to identify a combination of independent factors that best classified survival or death by 3.5 years after vaccine therapy had been initiated. This analysis was performed at a time when all surviving patients had been followed for at least 3.7 years, with a median follow-up of 5.7 years.

Materials and Methods

Patients

The characteristics of the patient population were described in detail previously.³ There were 34 men and 20 women with metastatic melanoma, who had a median age of 50.5 years. All patients had undergone one or more surgical procedures; 20% had received radiation therapy (RT) and 83% had previously received systemic therapies such as single-agent biological therapies including 37% IFN-alpha, 35% GM-CSF, 24% IL-2 as a single modality, 13% other investigational vaccines, 30% chemotherapy as a single modality, and 35% IL-2-based biochemotherapy. Their most advanced stage of disease was 35% M1a (distant skin, subcutaneous, or lymph node metastases), 6% M1b (lung metastases as only site of visceral disease), and 59% M1c (other visceral metastases and/or elevated lactate dehydrogenase [LDH]).⁴ Brain metastases had been previously treated in 20%. Distant metastases (stage IV) had been previously documented in 78%, whereas 22% had experienced only recurrent regional metastases (stage III). Based on Response Evaluation Criteria in Solid Tumors (RECIST) criteria at the time vaccine therapy was initiated,⁵ 28% had objective measurable disease, 48% had detectable but nonmeasurable disease, and 24% had nonmeasurable disease. Eastern Cooperative Oncology Group (ECOG) performance status (PS) was 70% ECOG-0, 26% ECOG-1, and 4% ECOG-2. At the time of the present report, 32 patients had survived at least 3.7 years after starting the vaccine, and 24 were still alive 3.7–8.5 years after starting the vaccine.

Clinical and laboratory features

This analysis centered on 76 clinical and laboratory features measured relative to survival as the dependent variable. Clinical features included baseline clinical characteristics at study entry, prior treatments, and clinical laboratory measurements taken at baseline and at week 4 after the first three vaccinations. Laboratory features included characterization of the vaccine products, such as the characteristics of cells collected by leukapheresis, the changes in expression of markers on dendritic cells as they matured in cell culture, characteristics of the tumor cells used in the vaccine product, and quantity of cells administered, as a surrogate for quantity of tumor-associated antigens that were injected in the first three vaccinations.

The specific clinical variables used to group patients for comparisons were gender, age >50 years, residence in the state of California, ECOG PS of 0 (fully active), M1c metastatic disease, measurable disease by RECIST criteria, completing the planned eight vaccinations, prior therapies including RT, immunotherapy with IL-2 and/or IFN-alpha or GM-CSF, or chemotherapy, whether patients were anergic to baseline standard skin tests for Candida or Trichophyton, and whether patients exhibited a positive delayed-type hypersensitivity (DTH) reaction to an injection of 1 million irradiated autologous tumor cells at baseline or at any time during treatment. Clinical laboratory variables included baseline and week-4 LDH, S100b, and the changes in these measurements from week 0 to week 4.

The methods for characterizing the components of the vaccine products and assessing immune response are described in other publications.^3,6

–10 The specific tumor cell-related laboratory variables used to group patients for comparisons included the expression of S100 on the original pathology specimen, the number of passages of tumor cells to obtain sufficient numbers for the vaccine product, the length of time in culture, the level of expression of several melanoma-associated antigens (Mage, Mart-1, Mel-5, HMB45, S100, and tyrosinase), sensitivity to the apoptosis-inducing effects of IFN-γ as measured by Annexin V and 7-amino-actinomycin D (7-AAD), enhancement of Human Lymphocyte Antigen (HLA)-1 and HLA-2 histocompatibility lymphocyte antigen expression after coculture with IFN-γ, and autologous lymphocyte reaction to tumor cells by ELISPOT tests to measure IFN-γ secretion. Dendritic cell-related variables included the numbers of lymphocytes and monocytes collected during leukapheresis to obtain peripheral blood mononuclear cells, the level of expression and changes in various dendritic cell-associated antigens (CD11c, CD80, CD83, CD86), and the total number of cells and number of viable cells administered to patients during the first three injections. Research laboratory variables included baseline and week-4 serum levels of disialogangliosides and thymus activation-regulated chemokine (TARC), and the changes in these measurements from week 0 to week 4.

Statistical analysis

The dependent variable of interest was survival after initiating vaccine therapy. Multiple univariate analyses calculated by Dr. Dillman were used to identify factors associated with significant differences in continuous survival based on log-rank test comparisons of Kaplan–Meier survival curves (p<0.05).¹¹ These were generated for subsets of patients based on 22 clinical variables and 27 laboratory variables at a time when all patients had been followed for at least 3.7 years, including 24 survivors who had been followed for a median of 5.7 years. Five patients were censored alive between 3.7 and 5 years (none had been lost to follow-up); 22 patients were censored alive at 5 years.

Stepwise regression was used by Dr. Fogel to identify models relative to the dependent variable of survival at 3.5 years.¹² Patient survival at 3.5 years was recoded to 1=survival (n=32) and 0=death (n=22). Stepwise regression was used to determine which linear combination of features best separated these two data classes. The first approach was limited to the 39 features for which complete information was available for all 54 patients. The second approach made use of both complete and incomplete data for all 54 patients and included a range of features that were available for 24 to 53 patients.

Results

Clinical features

Table 1 shows the comparative groups and log-rank tests for the 22 clinical variables selected for analysis. Seven clinical features were associated with improved survival with p-values<0.05: ECOG of 0 (p<0.0001), not having measurable disease at the time of vaccine therapy (p=0.0005), receiving eight vaccine injections, (p=0.0008), age<50 years at the time of vaccine therapy (p=0.003), not having M1c metastatic disease (p=0.020), having a baseline LDH in the normal range (p=0.005), and not being anergic based on reactivity to standard skin tests (p=0.049). There was no association between survival and any prior therapy received, gender, or detection of a positive DTH reaction to an injection of 1 million irradiated autologous tumor cells at baseline or at any time during treatment.

Table 1.

Univariate Analyses of Clinical Variables and Actuarial Survival

Variable	Arms	Number of patients	Median survival in months	% 5-Year observed survival	p-Value log-rank test
ECOG PS	0	38	NR	66	<0.0001
	1 (n=14) 2 (2)	16	13	12
Measurable disease	No	39	NR	65	0.0005
	Yes	15	18	13
No. of vaccines received	8	42	67	57	0.0008
	3 to 7	12	9	25
Age in years	<50	27	70	66	0.003
	≥50	27	24	33
LDH at baseline (U/L)	167 to 446	27	71	66	0.005
	447 to 2370	27	39	37
Distant met stage	M1A and M1B	22	NR	68	0.020
	M1C	32	33	37
Anergic	No	42	NR	58	0.049
	Yes	11	34	27
Δ S100b week 0 to week 4 (μg/L)	−0.64 to 0.0	25	67	61	0.097
	+0.01 to 1.62	21	43	46
Prior irradiation	Yes	11	NR	73	0.150
	No	43	43	44
Prior chemo±IL-2	Yes	30	69	60	0.153
	No	24	39	36
Prior IL-2 only	Yes	13	26.7	38	0.242
	No	41	63.5	53
Prior biochemotherapy	Yes	19	66	63	0.332
	No	35	43	42
Best tumor DTH response week 0 or week 4	Positive	8	NR	63	0.465
	Negative	46	44	45
Prior IL-2±chemo	Yes	26	64	58	0.466
	No	28	44	42
Prior GM-CSF	Yes	19	63	53	0.660
	No	35	50	48
Tumor cell DTH ever positive	Yes	13	60	50	0.741
	No	41	52	49
State where the patient live	California	36	60	50	0.750
	Elsewhere	18	52	49
Prior interferon alpha	Yes	20	54	49	0.787
	No	34	63	50
Baseline S100b (μg/L)	0.01 to 0.04	25	50	48	0.807
	0.05 to 3.29	21	72	60
Gender	Female	20	60	50	0.960
	Male	34	54	49
Change LDH week 0 to week 4 (U/L)	−512 to −6	30	NR	53	0.968
	+6 to +962	24	42.7	46

ECOG PS, Eastern Cooperative Oncology Group Performance Status; IL-2, interleukin-2; LDH, lactate dehydrogenase; DTH, delayed-type hypersensitivity; GM-CSF, granulocyte-macrophage colony stimulating factor; NR, not reached.

Laboratory determinations

Table 2 shows the comparative groups and log-rank tests for the 27 laboratory measures available for analysis. The only research laboratory variable associated with improved survival was whether the patient's autologous proliferating tumor cells were resistant to the apoptosis-inducing effects of IFN-γ (n=49), based on flow cytometry measurements of annexin-V and 7-AAD as a means of detecting apoptotic cells (p=0.0006).¹⁰ None of the surface marker changes in dendritic cells, the expression of individual antigens on cultured melanoma cells, changes in HLA antigens in response to IFN-γ, or Elispot tests were associated with a difference in survival. Baseline TARC and ganglioside levels were not prognostic of survival, and changes in their levels after three vaccinations were not predictive of survival benefit.

Table 2.

Univariate Analyses of Laboratory Variables and Continuous Survival

Variable	Comparison groups	Number of patients	Median survival in months	% 5-Year observed survival	p-Value log-rank test
% Apoptosis by 7-AAD	>8%	26	27	25	0.0006
	<8%	23	NR	73
Apoptosis after IFN-γ	Yes	25	27	28	0.0006
	No	24	NR	75
% Apoptosis by Annexin V	>5%	25	27	28	0.0006
	<5%	24	NR	75
%CD86↑postincubation	Yes	27	68	63	0.126
	No	14	41	27
Elispots week 4/100k cells	0 to 7.3	18	63	54	0.142
	8.0 to 172	24	36	36
HMB-45 on cell line	No	10	70	70	0.174
	Yes	42	42	45
No. of cell line passages	3 to 7	31	43	42	0.222
	8 to 16	23	NR	60
MIF HLA2 pre-post IFN-γ	−252 to +142	27	43	44	0.222
	156 to 1114	27	64	55
Baseline serum gangliosides (mg/dL)	<20	11	54	50	0.234
	>20	13	64	62
No. of lymphs recovered (×10⁹)	1.5 to 3.4	28	42	46	0.289
	3.5 to 7.5	26	63	53
Elispots at week 0 per 100k cells	0 to 6.8	21	63	52	0.308
	8.5 to 172	21	40	37
Mel-5 on cell line	No	16	66	62	0.441
	Yes	36	43	44
% CD83 (6 up, 35 down)	0.20 to 1.46	20	42	45	0.472
	1.5 to 7.0	21	NR	57
Tyrosinase on cell line	No	9	43	44	0.449
	Yes	43	63	51
Time in months to cell line	1.6 to 4.2	27	69	56	0.562
	4.4 to 11.2	27	42	44
S100 on cell line	No	34	49	47	0.603
	Yes	18	63	56
Mage-1 on cell line	No	26	45	46	0.603
	Yes	26	66	54
Change TARC week 0 to week 4 (pg/mL)	−765 to 2819	26	54	50	0.615
	3062 to 5859	27	NR	52
No. of viable cells, first three injections	11.2–33.5	26	45	46	0.617
	34.0–73.2	28	68	59
ELISPOTs Δ week 0 to week 4	−152 to 0	18	53	49	0.646
	0.1 to 135	24	43	41
No. of monocytes recovered (×10⁸)	1.6 to 5.9	30	45	47	0.700
	6.1 to 10.1	24	62	54
Ganglioside levels declined week 0 to week 4 (mg/dL)	No	9	64	67	0.816
	Yes	15	43	47
% CD 80↑week 0 to week 4	Yes	23	63	52	0.847
	No	18	53	49
MIF HLA1 pre-post IFN-γ	7 to 415	27	49	48	0.901
	434 to 1533	27	NR	52
No. of cells, first three injections	14.5–45.1	27	69	50	0.912
	47.2–99.2	27	63	50
% CD11c (39 up, 2 same)	0.59 to 0.90	19	NR	53	0.922
	0.91 to 1.0	22	60	50
TARC at baseline (pg/mL)	82 to 499	28	63	50	0.947
	542 to 1892	25	63	52

MIF, median intensity of fluorescence; 7-AAD, 7-amino-actinomycin D; IFN-γ, interferon gamma; TARC, thymus activation-regulated chemokine.

Table 3 shows that the regression developed from those features with complete information over all 54 patients was composed of six features: prior RT, age, gender, ECOG PS, total number of cells administered during the first three injections, and total number of viable cells administered during the first three injections. All six features had p-values<0.015 when used in combination in this model, which provided quite good separation of the two patient groups for survival. For a decision threshold of 0.62 to the output of this model, survival at 3.5 years was correctly predicted for 28 of 32 patients (87%), with 4 false negatives (13%). Death prior to 3.5 years was correctly predicted for 20 of 22 patients (91%), with 2 false positives (9%).

Table 3.

Six-Term Linear Regression Using Threshold of 0.62 on the y-Axis to Classify Survival at 3.5 Years (n =54)

Count	54
Number missing	0
R	0.758
R squared	0.575
Adjusted R squared	0.521
RMS residual	0.343

ANOVA table	DF	Sum of squares	Mean square	f-Value	p-Value
Regression	6	7,494	1.249	10.591	<0.0001
Residual	47	5.543	0.118
Total	53	13.037

Regression coefficients	Coefficient	Standard error	Standard coefficient	t-Value	p-Value
Intercept	0.842	0.224	0.842	3.759	0.0005
Radiation therapy	0.310	0.121	0.254	2.555	0.0139
Age	−0.013	0.004	−0.364	−3.514	0.0010
Gender (0=F, 1=M)	0.263	0.100	0.259	2.646	0.0110
ECOG PS	−0.476	0.096	−0.527	−4.967	<0.0001
No. of cells, first three injections	0.038	0.011	1.459	3.423	0.0013
No. of viable cells, first three injections	−0.037	0.012	−1.276	−3.023	0.0040

Patients who survived were considered as 1 on the x-axis; those who did not survive were considered as 0. For survival at 3.5 years, y=0.842+(0.310×prior RT)−(0.013×age)+(0.263×gender)−(0.476×ECOG PS)+(0.038×total number of cells for the first three injections)−(0.037×total number of viable cells for the first three injections). Predicted survival from the linear regression indicated on the y-axis was 28/32 (88%) for survival and 20/22 (91%) for nonsurvival.

RMS, root mean square; ANOVA, analysis of variance; R, correlation coefficient; DF, degrees of freedom; F, female; M, male; RT, radiation therapy.

Although this equation resulted in a useful adjusted R ², three terms in this model (total number of cells, number of viable cells, and gender) had very low correlation coefficients, suggesting that they contributed very little to the overall performance of the model. After removing these three parameters, the p-value associated with RT was no longer significant; so it was also removed. Table 4 shows the resulting two-term regression for survival at 3.5 years. The same threshold of 0.62 provided reasonable classification accuracy. Survival at 3.5 years was correctly predicted for 27 of 32 patients (84%), with 5 false negatives (16% of surviving patients), and death by 3.5 years was correctly predicted for 16 of 22 patients (73%), with 6 false positives (27%). The p-value on both terms was <0.01.

Table 4.

Two-Term Linear Regression Using Threshold of 0.62 on the y-Axis to Classify Survival (y) at 3.5 Years (n =54)

Count	54
Number missing	0
R	0.611
R squared	0.373
Adjusted R squared	0.349
RMS residual	0.400

ANOVA table	DF	Sum of squares	Mean square	f-Value	p-Value
Regression	2	4.864	2.432	15.177	<0.0001
Residual	51	8.173	0.160
Total	53	13.037

Regression coefficients	Coefficient	Standard error	Standard coefficient	t-Value	p-Value
Intercept	1.311	0.215	1.311	6.105	<0.0001
Age	−0.12	0.004	−0.322	−2.726	0.0088
ECOG PS	−0.379	0.107	−0.420	−3.558	0.0008

The resulting two-term regression for survival at 3.5 years was y=1.311 − (0.012×Age) − (0.379×ECOG PS). Predicted survival from the linear regression indicated on the y-axis was 27/32 (88%) for survival and 16/22 (84%) for nonsurvival.

The best stepwise multiple regression for survival utilized all possible features, including those with incomplete data over the patients (n=49). Best multiple linear regression included the same six features as before, with the addition of one of the three measures of IFN-γ-induced apoptosis (7-AAD) as a seventh feature, as shown in Table 5. In this model, survival at 3.5 years was correctly predicted for 27 of 29 patients (93%), with 2 false negatives (7% of surviving patients), and death by 3.5 years was correctly predicted for 19 of 20 patients (95%), with 1 false positive (5%). The p-value for both features was <0.03.

Table 5.

Seven-Term Linear Regression Using Threshold of 0.58 on the y-Axis to Classify Survival (y) at 3.5 Years (n =49)

Count	49
Number missing	5
R	0.817
R squared	0.668
Adjusted R squared	0.611
RMS residual	0.310

ANOVA table	DF	Sum of squares	Mean square	f-Value	p-Value
Regression	7	7.904	1.129	11.770	<0.0001
Residual	41	3.933	0.096
Total	48	11.837

Regression coefficients	Coefficient	Standard error	Standard coefficient	t-Value	p-Value
Intercept	0.991	0.214	0.991	4.619	<0.0001
Radiation therapy	0.259	0.115	0.212	2.255	0.0295
Age	−0.014	0.004	−0.372	−3.818	0.0004
Gender (0=F, 1=M)	0.273	0.094	0.268	2.912	0.0058
ECOG PS	−0.406	0.092	−0.459	−4.399	<.0001
No. of cells, first three injections	0.039	0.010	1.507	3.795	0.0005
No. of viable cells, three injections	−0.037	0.011	−1.289	−3.293	0.0020
7-AAD/IFN-γ	−0.019	0.005	−0.336	−3.487	0.0012

Resulting formula was y=0.991+(0.259×RT)−(0.14×Age)+(0.273×Gender)−(0.406×ECOG PS)+(0.039×cells first three vaccines) −(0.037×viable cells in first three vaccines)−(0.019×% of cells positive by 7-AAD). Predicted survival from the linear regression indicated on the y-axis was 27/29 (93%) for survival and 19/20 (95%) for nonsurvival.

Discussion

These metastatic melanoma patients had a 5-year survival rate of 50%, which is much higher than that previously observed in published melanoma phase II chemotherapy trials in which median survival rates of only 4–6 months were typical.¹³ Based on a meta-analysis of phase II cooperative group trials in metastatic melanoma conducted between 1975 and 2005, the upper boundary (95% confidence level) for survival at 1 year in a phase II trial of 54 patients would be estimated to be 45%.¹³ However, in this patient-specific vaccine trial, the actual observed 1-year survival rate was 85%, and the 5-year survival rate was 50%.³ The encouraging survival rate may be due to immune enhancing effects of the vaccine, and it is also possible that the results are due to treating a population of patients with more favorable prognostic features. For instance, the presence of measurable disease at the time of treatment was a requirement for the studies that were included in the meta-analysis.¹³ Although all of the patients enrolled in this trial had sites of measurable metastatic melanoma during the course of their disease, most did not have measurable disease by RECIST criteria at the time vaccine treatment was initiated. The 15 patients who did have measurable metastatic disease at the time vaccine therapy was initiated had a 1-year survival rate of 73%, which is still beyond the upper boundary for phase II studies of similar sample size.¹³

Tumor burden and ECOG PS were closely correlated, and because of this, no measures of tumor burden were critical independent factors in the regression model. A good PS, whether measured by ECOG or Karnofsky scales, is a well-validated clinical prognostic feature that has been associated with improved survival in clinical trials of patients with metastatic melanoma.^13
–15 It is not surprising that ECOG PS would be highly prognostic in trials in which most patients would die within a few months, but patients with a poor ECOG PS were not included in this trial, and the median survival was nearly 5 years rather than 5 months. In the multivariate analysis, the combination of ECOG PS and age accurately classified 84% of patients who were alive at 3.5 years after starting vaccine treatment and 73% of patients who had died by that time. In the univariate analyses, ECOG PS was associated with the most robust p-value for the log-rank test comparisons of Kaplan–Meier survival curves (p<0.0001) and younger age was also associated with a significant p-value (p=0.003).

Older age is usually a negative prognostic factor for survival in clinical trials for a variety of reasons, including death from other causes, the presence of more highly mutated tumors, and perhaps because of senescence of the immune system. However, only one of the 22 deaths that occurred by 3.5 years was unrelated to melanoma. Despite this, age was still a strong independent predictor of survival.

Other clinical features associated with improved survival in this trial are also well-known prognostic features, including various measures of tumor burden such as the presence of measurable disease and serum levels of LDH and S100b.^13

–18 Although the number of vaccines received could be an important variable, this is misleading, because patients who experience progressive disease during treatment are more likely to discontinue treatment early, and in most cancer clinical trials, progressive disease is associated with increased likelihood of death.¹⁹ This is one of many variables likely to be strongly correlated with other features and therefore unlikely to be an independent variable. For instance, ECOG PS may actually be a more sensitive measure of tumor burden than radiographic evidence of disease or a putative serologic tumor marker. None of these other clinical features independently contributed to the models derived by regression analysis.

Many laboratory measurements that might correlate with enhanced immune response, such as the phenotype of tumor cells, the expression of HLA antigens, tumor-specific reactivity by autologous lymphocytes as measured in Elispot testing, or numbers of tumor cells injected (as a surrogate for antigen quantity), were not associated with improved survival in univariate analyses. Levels of serum gangliosides and TARC, which were previously shown to be associated with improved progression-free survival in smaller subsets of patients,^7,8 were not associated with improved overall survival in this analysis. Although it had been postulated that short-term cell lines associated with fewer passages might be preferable in such vaccines,⁹ the present analysis revealed no difference in survival by number of passages or duration of time needed to establish a cell culture. Faster growing cell lines might have been derived from tumors that were more biologically aggressive, whereas patients whose tumors grew slowly in vitro may have had less aggressive cancers. Further, patients whose tumors grew more slowly had to survive longer without rapidly progressive disease to become candidates for treatment in this clinical trial. In an earlier analysis, lack of increased S100b at 4 weeks was associated with improved survival,⁵ but in the present analysis the log-rank test for survival difference did not quite reach statistical significance (p=0.097).

The only immunologic laboratory assay feature associated with a difference in survival was IFN-γ-induced apoptosis based on the Annexin V and 7-AAD assays that were used to measure apoptosis.¹⁰ All three determinations were associated with the same robust p-value of 0.0006. Increased apoptosis in response to IFN-γ might have facilitated phagocytosis and antigen presentation by the dendritic cells used in the vaccine, which may have enhanced a response to melanoma-associated antigens in vivo.^20,21 However, if this were the case, a positive correlation with survival would have been expected rather than a negative correlation. On the other hand, such resistance may also have prevented cells from inducing or enhancing tolerance to tumor-associated antigens.^20,22 It is also possible that this in vitro observation correlates with lack of tolerance to tumor-associated antigens in vivo, such that patients whose proliferating melanoma cells exhibited increased resistance to apoptosis were more likely to have an ongoing antitumor immune response that could be enhanced by the vaccine.

Stepwise regression makes use of a process to analyze one feature at a time and successfully combine features to explain a dependent variable—in this study, patient survival at 3.5 years. The two features of greatest value in this approach were ECOG of 0 and younger age, both of which were statistically significant at the p<0.008 level. These two clinical parameters alone correctly classified survival at 3.5 years for about 75% of patients. However, the addition of other features to the model substantially increased classification accuracy, even though these other variables, including prior RT and numbers of cells included in the vaccinations, were not statistically significant in univariate analysis.

Meta-analyses of melanoma phase II chemotherapy trials have identified female gender as an independent factor associated with improved survival.^13,14 In this immunotherapy trial, gender was not associated with survival in the univariate analysis (p=0.96). In the regression analysis, male gender had a positive rather than negative independent association with survival, but this feature had a very low coefficient.

Previously, it was reported that a positive DTH reaction to intradermal injections of autologous irradiated tumor cells was associated with prolonged survival in metastatic cancer patients who were treated with irradiated, proliferating autologous tumor cells.²³ This included 66 patients with metastatic melanoma.²⁴ However, with longer follow-up, in the final analysis of all 74 melanoma patients who treated in this manner, there was merely a trend toward better event-free survival for those who had a positive DTH reaction to an i.d. injection of 1 million irradiated tumor cells at baseline, or converted to positive after three injections, compared with those whose DTH remained negative, and there was no difference in overall survival.²⁵ In the more recent trial,³ the data from which is the focus of this report, there was no association between the tumor cell DTH reaction and survival as shown in Table 1. Some have suggested that reactions to such tumor cell DTH tests are just epiphenomena, perhaps as another test of relative anergy to a less common antigen such as fetal bovine protein, rather than an independent predictor of survival. In this trial, of the 11 patients who were anergic to standard skin tests at baseline, all 11 were also anergic to the baseline tumor DTH test and at week 4, but 2 of 8 retested at week 24 were positive for the week-24 test only. These 2 had survivals of 13.9 and 36.3 months. Among the 43 patients who were not anergic to standard skin tests at baseline, 1 had a positive tumor DTH reaction at baseline (weakly positive), 7 converted by week 4, and 3 of 27 tested at week 24 were positive for the first time at week 24.

As far as characteristics of the vaccine itself, the two features related to numbers of cells injected had very low coefficients, but did enhance the model. Neither of these variables was identified as having significance in the univariate analysis. As all of the tumor-associated antigens that may be important for enhancing an antitumor immune response were not known, the total number of cells was considered as a potential surrogate for quantity of tumor-associated antigens injected. A previous study of patient-specific autologous tumor cell-based vaccines, which did not include dendritic cells, also failed to show an association between survival and the numbers of tumor cells or viable tumor cells injected during the first 3 weeks of a vaccination program.²³ In the more recent trial,³ the final vaccine product was mostly dendritic cells, but there were also variable numbers of nonviable irradiated tumor cells that had not been phagocytosed during coculture. It may seem paradoxical that the regression analysis showed a positive contribution from the total number of cells injected, but a negative contribution from the total number of viable cells injected. However, this raises the possibility that nonviable tumor cells in the final vaccine product (included in the total cell numbers, but not in the viable cell numbers) may be an important component of the vaccine. This is consistent with the suggestion that injections of apoptotic tumor cells themselves might be sufficient to induce an effective immune response in vivo, thus obviating the need to load dendritic cells ex vivo.

It is not surprising that clinical features dominated the regression analysis, because they were documented for the entire cohort as part of the clinical trial enrollment process, and measurements for many of the laboratory features were only available for smaller subsets of patients. A major limitation of the regression analysis is that complete information was available for only 9 of the 27 research laboratory features chosen for the analysis, and many of these were different measures of the same feature. The major reason for this was the exhaustion of supplies of patient cells and serum needed to replicate the laboratory experiments. Measurements of apoptosis induction by IFN-γ were available for 49 patients. Of the laboratory assays for which there was incomplete data, resistance to IFN-γ-induced apoptosis was the only one that increased the accuracy of the regression model.

This logistic regression model may be thought of as a training set for establishing a predictive model of outcome in populations of melanoma patients receiving these patient-specific vaccines. The authors will test this as a prognostic and predictive model in the randomized phase II trial that is in progress (Clinicaltrials.gov: NCT00436930), in which two patient-specific melanoma vaccines are compared: one consisting of only irradiated autologous tumor cells,²⁶ and the other consisting of autologous dendritic cells that have been loaded with continuously proliferating autologous tumor cells.³

Footnotes

Acknowledgments

The authors acknowledge the assistance provided by clinical research nurses Cristina De Leon and Cheryl Mayorga in the collection and documentation of clinical information and by Hoag Cell Biology Laboratory research associates Denysha Carbonell and Abner Fowler in the production of vaccine products and performance of assays. This work was supported by the Hoag Hospital Foundation from philanthropic gifts for Hoag Cell Biology Laboratory cancer research, and by Natural Selection, Inc.

Disclosure Statement

The authors declare that they have no proprietary, financial, professional, or other personal interest in any product, service, and/or company that could be construed as influencing the position presented in this manuscript.

References

Dillman

. Cancer immunotherapy. Cancer Biother Radiopharm, 2011; 26:1.

Dillman

, Selvan

, Schlitz

. Patient-specific dendritic cell vaccines for metastatic melanoma. N Engl J Med, 2006; 355:1179.

Dillman

, Selvan

, Schiltz

et al. Phase II trial of dendritic cells loaded with antigens from self-renewing, proliferating autologous tumor cells as patient-specific anti-tumor vaccines in patients with metastatic melanoma: Final Report. Cancer Biother Radiopharm, 2009; 24:311.

Balch

, Gershenwald

, Soong

S-J

et al. Final version of 2009 AJCC melanoma staging and classification. J Clin Oncol, 2009; 27:6199.

Therasse

, Arbuck

, Eisenhauer

et al. New guidelines to evaluate the response to treatment in solid tumors. J Natl Cancer Inst, 2000; 92:205.

Schiltz

, Dillman

, Korse

et al. Lack of elevation of serum S100B in patients with metastatic melanoma as a predictor of outcome after induction with an autologous vaccine of proliferating tumor cells and dendritic cells. Cancer Biother Radiopharm, 2008; 23:214.

Selvan

, Dillman

, Fowler

et al. Monitoring response to treatment in melanoma patients: Potential of a serum glycomic marker. Int J Cancer, 2008; 122:1374.

Cornforth

, Lee

, Fowler

et al. Increases in serum TARC/CCL17 levels are associated with progression-free survival in advanced melanoma patients in response to dendritic cell-based immunotherapy. Clin Immunol, 2009; 29:657.

Selvan

, Carbonell

, Fowler

et al. Establishment of stable cell lines for personalized melanoma cell vaccine. Melanoma Res, 2010; 20:280.

10.

Cornforth

, Fowler

, Carbonell

et al. Resistance to the proapoptotic effects of interferon-gamma on melanoma cells used in patient-specific dendritic cell immunotherapy is associated with improved overall survival. Cancer Immunol Immunother, 2011; 60:123.

11.

Lagakos

. Statistical analysis of survival data. Bailar

III , Mosteller

. Medical Uses of Statistics, 2nd. Boston: NEJM Books, 1992; 281.

12.

Fisher

, Van Belle

. Association and prediction: Multiple regression analysis, linear models with multiple predictor variables. Fisher

, Van Belle

. Biostatistics: A Methodology for the Health Sciences. New York: John Wiley and Sons, 1993; 496.

13.

Korn

, Liu

P-Y

, Lee

et al. Meta-analysis of phase II cooperative group trials in metastatic stage IV melanoma to determine progression-free and overall survival benchmarks for future phase II trials. J Clin Oncol, 2008; 26:527.

14.

Brand

, Ellwanger

, Stroebel

et al. Prolonged survival of 2 years or longer for patients with disseminated melanoma. An analysis of related prognostic factors. Cancer, 1997; 79:2345.

15.

Manola

, Atkins

, Ibrahim

et al. Prognostic factors in metastatic melanoma: A pooled analysis of Eastern Cooperative Oncology Group trials. J Clin Oncol, 2000; 18:3782.

16.

Deichmann

, Benner

, Bock

et al. S100-Beta, melanoma-inhibiting activity, and lactate dehydrogenase discriminate progressive from nonprogressive American Joint Committee on Cancer stage IV melanoma. J Clin Oncol, 1999; 17:1891.

17.

Torabian

, Kashani-Sabet

. Biomarkers for melanoma. Curr Opin Oncol, 2005; 17:167.

18.

Linette

, Carlson

, Slominski

et al. Biomarkers in melanoma: Stage III and IV disease. Expert Rev Mol Diagn, 2005; 5:65.

19.

Prentice

. Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med, 1989; 8:431.

20.

Fonseca

, Dranoff

. Capitalizing on the immunogenicity of dying tumor cells. Clin Cancer Res, 2008; 14:1603.

21.

Fry

, Shand

, Milliron

et al. Antigen loading of DCs with irradiated apoptotic tumor cells induces improved anti-tumor immunity compared to other approaches. Cancer Immunol Immunother, 2009; 58:1257.

22.

Wang

. Functional control of regulatory T cells and cancer immunotherapy. Semin Cancer Biol, 2006; 16:106.

23.

Dillman

, Beutel

, Barth

et al. Irradiated cells from autologous tumor cell lines as patient-specific vaccine therapy in 125 patients with metastatic cancer: Induction of delayed-type hypersensitivity to autologous tumor is associated with improved survival. Cancer Biother Radiopharm, 2002; 17:51.

24.

Dillman

, DeLeon

, Beutel

et al. Short-term autologous tumor cell lines for the active specific immunotherapy of patients with metastatic melanoma. Crit Rev Oncol Hematol, 2001; 39:115.

25.

Dillman

, DePriest

, DeLeon

et al. Patient-specific vaccines derived from autologous tumor cell lines as active specific immunotherapy: Results of exploratory phase I/II trials in patients with metastatic melanoma. Cancer Biother Radiopharm, 2007; 22:309.

26.

Dillman

, Beutel

, DePriest

et al.

Cancer vaccine potency: Is there a dose/response relationship for patient-specific vaccines and clinical outcomes?

Cancer Biother Radiopharm, 2005; 20:373.