Carbohydrate antigen 19-9 change during chemotherapy for advanced pancreatic adenocarcinoma

Authors


Abstract

BACKGROUND:

Radiologic assessment of tumor response in pancreatic cancer is complicated by desmoplastic reactions within or around the tumor. The objective of this study was to evaluate the correlation between a decline in carbohydrate antigen 19-9 (CA 19-9) and survival in patients with advanced pancreatic cancer who received upfront chemotherapy.

METHODS:

CA 19-9 serum basal values were measured in 247 patients with advanced pancreatic cancer who were enrolled in 5 consecutive trials between 1997 and 2007. Survival curves were compared among patients who had a predefined CA 19-9 nadir variation (<50%. Group 1; 50% to 89%, Group 2; or >89%, Group 3). To eliminate guarantee-time bias, survival analysis was repeated using the landmark method.

RESULTS:

In both univariate and multivariate analysis, the basal CA 19-9 value significantly predicted survival. The median survival was 15.5 months for 34 patients who had normal basal CA 19-9 values, 11.9 months for 108 patients who had basal values between 38 U/mL and 1167 U/mL, and 8 months for 105 patients who had basal values >1167 U/mL. At least 1 CA 19-9 follow-up value was available for 204 patients who had baseline values greater than normal. A significant difference in overall survival was observed in univariate and multivariate analyses between Groups 1 and 2, between Groups 1 and 3, and between Groups 2 and 3. The results were confirmed using the landmark method.

CONCLUSIONS:

In this study, baseline CA 19-9 was confirmed as an independent prognostic factor for survival, and it may be considered as a stratification factor in trials in patients with advanced pancreatic cancer. Biochemical response may be used as a complementary measure to radiologic response to provide a better assessment of chemotherapy activity and to drive treatment decisions in clinical practice. Cancer 2009. © 2009 American Cancer Society.

Patients with stage III and metastatic pancreatic cancer1 have a dismal prognosis; and, despite the relevant number of prospective trials performed, the advantage in overall survival (OS) yielded by chemotherapy doublets over single-agent gemcitabine has been either absent2-6 or negligible and of marginal clinical significance.7, 8 Although combination chemotherapy with 4 agents9 more consistently improved the outcome in phase 3 comparisons with standard gemcitabine, and although the survival improvement was not achieved at the cost of impaired quality of life,10 the purpose of treatment remained palliative. In this context, a reliable assessment of tumor response to chemotherapy may assume a crucial role in driving treatment decisions aimed at obtaining the maximum possible benefit, such as whether to continue an active therapy, to stop an inactive therapy, or to change the regimen when effective second-line chemotherapy will be validated. In fact, it may allow clinicians to distinguish patients who likely will benefit from continuing therapy from those who may avoid a futile treatment that unnecessarily exposes them to the risk of quality-of-life impairment and who eventually may be considered for salvage therapy or best supportive care. Among the available tools for tumor response assessment, the evaluation of radiologic responses, even with newer imaging techniques, is considered rather unreliable because of the vigorous desmoplastic reaction, including inflammation and fibrosis, within and around the tumor.11

Among the surrogate markers of treatment efficacy, a decrease in carbohydrate antigen 19-9 (CA 19-9) has been proposed. CA 19-9 is a sialylated Lewis blood group, tumor-associated antigen that is not synthesized by individuals who lack the Lewis antigen glycosyl-transferase. There have been several retrospective studies that identified a correlation between a decline in serum CA 19-9 value and prolonged survival.12-18 However, because of the limited numbers of patients in those studies and the various definitions of CA 19-9 response, a consensus was not reached about the frequency with which tests should be performed or about the meaningful cutoff level of marker change. Furthermore, a recent retrospective analysis of prospectively collected data from a randomized phase 3 trial suggested that a decrease in CA 19-9 during chemotherapy (analyzed by correcting for guarantee-time bias) is not a valid surrogate endpoint for survival in clinical trials.19 Data from the latter trial are difficult to interpret because of the high percentage of missing values (72 of 247 patients; 29%) and because a decline in CA 19-9 concentration of ≥50% was related significantly to survival in the analysis that was not corrected for guarantee-time bias and in the Cox regression model.19 The objective of the current study was to assess the correlation between a decline in CA 19-9 value and survival in a large cohort of patients treated at a single institution and pooled from 5 consecutive trials that examined the role of first-line therapy in patients with advanced pancreatic cancer.9, 20-23

MATERIALS AND METHODS

Patients

All chemotherapy-naive patients with cytologically or histologically proven stage III or metastatic adenocarcinoma of the pancreas1 who were treated between April 1997 and January 2007 at our institution on 5 consecutive trials9, 20-23 were considered eligible for the current study. Patients aged ≤75 years who had a Karnofsky performance status (PS) >50; at least 1 measurable indicator lesion; adequate bone marrow function (absolute neutrophil count [ANC] ≥1500 cells/mm3, platelet count ≥100.000 cells/mm3, and hemoglobin ≥10 g/dL), kidney function (serum creatinine ≤1.5 mg/dL), and liver function (serum total bilirubin ≤1.5 mg/dL and serum transaminases ≤3 times the upper limit of laboratory normal [ULN]) received either a 4-drug regimen (cisplatin, epirubicin, 5-fluorouracil, and gemcitabine [PEFG]; cisplatin, epirubicin, capecitabine, and gemcitabine [PEXG]; or cisplatin, docetaxel, capecitabine, and gemcitabine [PDXG])9, 20-23 or gemcitabine alone.9 Cycles were repeated every 28 days for a maximum of 6 courses or until there was evidence of either unacceptable side effects or progressive disease. Patients with ampullary tumors or other histologic variants of pancreatic carcinoma were not considered for this analysis. CA 19-9 measurements were performed immediately before the start of chemotherapy and every 4 weeks thereafter. Jaundice, when present, was resolved surgically or endoscopically before study enrolment and before repeating basal CA 19-9 measurement. The ULN for CA 19-9 was 37 U/mL. Baseline and follow-up measurements for any given patient were performed at the same laboratory using the same testing method. The CA 19-9 nadir was defined as the lowest value measured at any time for each individual patient compared with the baseline value. A computed tomography scan was obtained at baseline, every 8 weeks during treatment, and every 12 weeks thereafter. Response to treatment was assessed according to either the World Health Organization24 for the earlier 3 studies9, 20, 21 or according to the Response Evaluation Criteria in Solid Tumors25 for the 2 most recent studies.22, 23 Progression-free survival (PFS) was calculated as the interval between the initiation of treatment and the occurrence of progressive disease or death from any cause, and survival was measured from the initiation of treatment to the date of either death from any cause or the last follow-up assessment.

Statistical Analysis

In univariate analyses, OS and PFS curves were estimated with the Kaplan-Meier method and were compared among patients with an arbitrarily predefined CA 19-9 change at nadir with respect to the baseline value (<50%, Group 1 [nonresponders]; 50%-89%, Group 2 [minor biochemical response]; and ≥90%, Group 3 [major biochemical response]) by using the log-rank test.

To eliminate guarantee-time bias, the survival analysis was repeated using the landmark method26, 27 on Day 57 (±7 days; ie, on Day 1 of the third cycle). In that analysis, patients who had disease progression or who died before the date of the landmark evaluation were excluded, and the response groups were defined on the basis of response status at the landmark regardless of any subsequent shifts in tumor response status only for patients who remained on the study. Then, OS was measured from the landmark time to the date of either death from any cause or the last follow-up assessment. Thus, probability estimates and statistical tests were conditional based on the response status of patients at the landmark time.

Multivariate analysis using a Cox proportional-hazards model was carried out to estimate independent risk factors that could affect OS. All probability (P) values were from 2-sided tests. Clinical characteristics and response rates were compared using the chi-square test or the Fisher exact test for categorical variables, as appropriate.

RESULTS

The final analysis was performed on May 19, 2008, when 235 of 252 patients (93%) had died and 17 patients who remained alive had completed ≥13 months of follow-up (median follow-up, 21 months; range, 13-50 months). No patients were lost to follow-up. Patients' characteristics are reported in Table 1. No significant difference among patient subgroups was observed apart a significantly greater number of patients with stage IV disease in the group with higher basal CA 19-9 values. Twenty patients underwent previous curative surgery and had metastatic failure at the time of enrollment. Thirty-five patients underwent previous palliative surgery.

Table 1. Patient Characteristics
CharacteristicNo. of Patients (%)
Basal CA 19-9, U/mLChemotherapy Regimen
Total>1167 U/mL38-1167 U/mL<38 U/mLGPEFG/PEXGPDXG
  • CA 19-9 indicates carbohydrate antigen 19-9; G, gemcitabine; PEFG, cisplatin, epirubicin, 5-fluorouracil, and gemcitabine; PDXG, cisplatin, docetaxel, capecitabine, and gemcitabine; ECOG, Eastern Cooperative Oncology Group; PS, performance status.

  • *

    The median PS for all groups was 1.

  • P < .0001 (chi-square test).

No. of patients247105108343318430
Age, y       
 Median61616162626161
 Range30-7433-7435-7330-7241-6930-7438-72
Sex       
 Men150 (61)64 (61)64 (59)22 (65)18 (55)117 (64)15 (50)
 Women97 (39)41 (39)44 (41)12 (35)15 (45)67 (36)15 (50)
ECOG PS*       
 089 (36)32 (30)44 (41)13 (38)10 (33)64 (35)15 (50)
 1144 (58)67 (64)59 (54)18 (53)21 (64)108 (59)15 (50)
 212 (5)6 (6)4 (4)2 (6)2 (6)10 (5)0 (0)
 Unknown2 (1)0 (0)1 (1)1 (3)0 (0)2 (1)0 (0)
Stage       
 III90 (36)22 (21)52 (48)16 (47)9 (27)71 (39)10 (33)
 IV157 (64)83 (79)56 (52)18 (53)24 (73)113 (61)20 (67)
Regimen       
 G33 (13)11 (11)17 (16)5 (15)   
 PEFG/PEXG184 (75)82 (78)75 (69)27 (79)   
 PDXG30 (12)12 (11)16 (15)2 (6)   
Basal CA19-9, U/mL       
 >1167    11 (33)82 (44)12 (40)
 38-1167    17 (52)75 (41)16 (53)
 <38    5 (15)27 (15)2 (7)

Baseline Carbohydrate Antigen 19-9 Value

Baseline CA 19-9 was detected in 247 of 252 patients in the current study (98%) and was >1 time the ULN (>37 U/mL) in 213 patients (86%). Among patients who had baseline values >1 time the ULN, the median CA 19-9 value was 35.4 times the ULN (1167 U/mL; range, 38-368,450 U/mL). Patients who had baseline CA 19-9 <1 time the ULN (N = 34) had longer OS compared with patients who had basal CA 19-9 values between 38 U/mL and 1167 U/mL (P = .05) and compared with patients who had basal CA 19-9 values >1167 U/mL (P = .00006) (Fig. 1A). Patients who had pretreatment CA 19-9 values between 38 U/mL and 1167 U/mL (N = 108) had significantly longer survival than patients who had CA 19-9 values greater than the median (N = 105; P = .0004) (Fig. 1A). In a Cox multivariate analysis that was stratified for age, PS, stage, sex, and treatment arm, the CA 19-9 value at baseline independently predicted OS (hazard ratio [HR], 1.47; 95% confidence interval [CI] 1.21-1.97; P = .0002).

Figure 1.

Overall survival (A) and progression-free survival (B) on the basis of baseline carbohydrate antigen 19-9 (CA 19-9) values. Thick line indicates CA 19-9 <38 U/mL (Group A); thin line, CA 19-9 38-1167 U/mL (Group B); dashed line, CA 19-9 >1167 U/mL(Group C); O, total number of observed events; N, number of patients registered.

Patients who had baseline CA 19-9 values <1 time the ULN (N = 34) had a PFS similar to that of patients who had basal CA 19-9 values between 38 U/mL and 1167 U/mL (P = .33) (Fig. 1B) and had longer PFS compared with patients who had basal CA 19-9 values >1167 U/mL (P = .002) (Fig. 1B). Patients who had basal CA 19-9 values between 38 U/mL and 1167 U/mL had a longer PFS compared with patients who had basal CA 19-9 values >1167 U/mL (P = .003) (Fig. 1B).

Carbohydrate Antigen 19-9 Nadir Response to Chemotherapy

At least 1 CA 19-9 follow-up value was available for 204 of 213 patients (96%) who had baseline values >1 time the ULN. Nine patients had no follow-up value because of a protocol violation (N = 6) or a treatment interruption caused by clinical deterioration (N = 2) or refusal (N = 1). Forty patients had only 1 CA 19-9 value after baseline. When considering the nadir value, an increase in the CA 19-9 value or a decrease <50% was observed in 67 patients (33%; Group 1 [nonresponders]), including 49 patients with metastatic disease (73%); a decrease between 50% and 89% was observed in 75 patients (37%; Group 2 [minor biochemical response]), including 47 patients with metastatic disease (63%); and a decrease >89% was observed in 62 patients (30%; Group 3 [major biochemical response]), including 36 patients with metastatic disease (58%). The differences in terms of stage III and metastatic patients across the 3 groups were not statistically significant. The median time to CA 19-9 nadir was 3.5 months (range, 1-6 months). Outcome data for the 3 groups are reported in Table 2. A significant difference in OS was observed between Groups 1 and 2 (Fig. 2A) (P = .0002), between Groups 1 and 3 (Fig. 2A) (P < .00001), and between Groups 2 and 3 (Fig. 2A) (P = .0001). These values did not change if the subsets of patients with stage III disease (Group 1 vs Group 2, P = .05; Group 1 vs Group 3, P = .003; Group 2 vs Group 3, P = .001) and patients with metastatic disease were analyzed separately (Group 1 vs Group 2, P = .001; Group 1 vs Group 3, P < .00001; Group 2 vs Group 3, P = .002). In a Cox multivariate analysis that was stratified for age, sex, PS, stage, CA 19-9 value at baseline, and treatment arm, CA 19-9 response to chemotherapy independently predicted OS (HR, 0.45; 95% CI, 0.36-0.55; P < .00001). A significant difference in PFS was observed between Groups 1 and 2 (Fig. 2B) (P = .00007), between Groups 1 and 3 (Fig. 2B) (P < .00001), and between Groups 2 and 3 (Fig. 2B) (P = .0002).

Figure 2.

Overall survival (A) and progression-free survival (B) on the basis of carbohydrate antigen 19-9 (CA 19-9) nadir response to chemotherapy. Thick line indicates decline in CA 19-9 level >89% (Group A); thin line, decline in CA 19-9 level between 50% and 89% (Group B); dashed line, decline in CA 19-9 level <50% (Group C); O, total number of observed events; N, number of patients registered.

Table 2. Outcome Based on Carbohydrate Antigen 19-9 Nadir Response to Chemotherapy
VariableGroup 1, n=67*Group 2, n=75Group 3, n=62
  • OS indicates overall survival; IQR, interquartile range; CI, confidence interval; PFS, progression-free survival.

  • *

    Group 1: patients who had an increase or a decrease in carbohydrate antigen 19-9 (CA 19-9) <50%.

  • Group 2: patients who had a decrease in CA 19-9 between 50% and 89%.

  • Group 3: patients who had a decrease in CA 19-9 >89%.

Median OS, mo6.5 (IQR, 3-9-9.6)10 (IQR, 7.6-14.1)16.7 (IQR, 12-21.7)
1-Y OS rate, %16 (95% CI, 7-25)31 (95% CI, 20-43)73 (95% CI, 62-84)
Median PFS, mo2.5 (IQR, 1.7-5.5)6.2 (IQR, 5.2-9.5)9.1 (IQR, 7.5-13-0)
6-Mo PFS rate, %18 (95% CI, 9-27)52 (95% CI, 41-63)68 (95% CI, 56-80)

There was no statistically significant difference among patients' characteristics based on chemotherapy regimen (Table 1). Biochemical response based on chemotherapy regimen is reported in Table 3. Patients who received PEFG or PEXG (P = .03) and patients who received PDXG (P = .002) had major biochemical responses more often than patients who received gemcitabine alone.

Table 3. Carbohydrate Antigen 19-9 Response, Overall Survival, and Progression-free Survival Based on Chemotherapy Regimen
VariableChemotherapy Regimen
G, n=33PEFG/PEXG, n=184PDXG, n=30
No. (%)95% CINo (%)95% CINo. (%)95% CI
  • G indicates gemcitabine; PEFG, cisplatin, epirubicin, 5-fluorouracil, and gemcitabine; PDXG, cisplatin, docetaxel, capecitabine, and gemcitabine; CI, confidence interval; OS, overall survival; PFS, progression-free survival.

  • *

    Biochemical response was available for 27 patients who received G, for149 patients who received PEFG/PEXG, and for 28 patients who received PDXG. Nonresponders were patients who had an increase or a decrease in carbohydrate antigen 19-9 (CA 19-9) <50%, minor responders were patients who had a decrease in CA 19-9 between 50% and 89%, and major responders were patients who had a decrease in CA 19-9 >89%.

  • Interquartile range.

Biochemical response*      
 Nonresponders14 (52)35-6947 (32)25-396 (21)6-36
 Minor responders11 (41)24-5857 (38)31-459 (32)15-49
 Major responders2 (7)0-1645 (30)23-3713 (47)28-64
OS      
 Median, mo105.4-12.510.57.3-17.9115.8-18.7
 1-Y, %249-394336-504022-58
PFS      
 Median, mo3.31.9-5.27.24.8-10.57.13.6-10
 6-Mo, %153-275851-655739-75

Similarly, the PEFG or PEXG regimen and the PDXG regimen yielded a significantly better OS than a regimen of gemcitabine alone (Fig. 3A<F 3>) (PEFG/PEXG vs gemcitabine, P = .04; PDXG vs gemcitabine, P = .05), whereas no difference in OS was observed between the PEFG or PEXG regimen and the PDXG regimen (P = .39). PFS was improved significantly in patients who received PEFG or PEXG and in patients who received PDXG with compared with patients who received gemcitabine alone (Fig. 3B) (PEFG/PEXG vs gemcitabine, P = .0003; PDXG vs gemcitabine, P = .005). No difference in PFS was observed between the PEFG or PEXG regimen and the PDXG regimen (P = .42).

Figure 3.

Overall survival (A) and progression-free survival (B) on the basis of chemotherapy regimen. Thick line indicates treatment with combined cisplatin, docetaxel, capecitabine, and gemcitabine (Group A); thin line, treatment with cisplatin, epirubicin, 5-fluorouracil, gemcitabine/cisplatin, epirubicin, capecitabine, and gemcitabine (Group B); dashed line, treatment with gemcitabine (Group C); O, total number of observed events; N, number of patients registered.

In a Cox multivariate analysis that was stratified for age, sex, PS, stage, and CA 19-9 value at baseline, treatment arm independently predicted OS (PEFG/PEXG vs gemcitabine: HR, 0.88; 95% CI, 0.80-0.98; P = .02; PDXG vs gemcitabine: HR, 0.87; 95% CI, 0.77-0.99; P = .03).

Carbohydrate Antigen 19-9 Response to Chemotherapy: Landmark Method

To eliminate guarantee-time bias, survival analysis was repeated by the landmark method at the time of the first radiologic assessment (Day 57 ± 7; ie, Day 1 of the third cycle), as proposed previously.19, 26, 27 Survival data for the 3 groups are reported in Table 4. Twenty-five patients had a missing value at this time point because of protocol violation (N = 6) or treatment interruption because of progression, toxicity, refusal, or death (N = 19). A significant difference in OS was observed between Groups 1 and 2 (P = .003), between Groups 1 and 3 (P = .003), and between Groups 2 and 3 (P = .02).

Table 4. Carbohydrate Antigen 19-9 Response: Landmark Method
VariableGroup 1, n=85*Group 2, n=78Group 3, n=16
  • OS indicates overall survival; IQR, interquartile range; CI, confidence interval; PFS, progression-free survival.

  • *

    Group 1: patients who had an increase or a decrease in carbohydrate antigen 19-9 (CA 19-9) <50%.

  • Group 2: patients who had a decrease in CA 19-9 between 50% and 89%.

  • Group 3: patients who had a decrease in CA 19-9 >89%.

Median OS, mo7.2 (IQR, 4.4-10.3)11.6 (IQR, 6.9-16.1)14.8 (IQR, 7.7-20.5)
1-Y OS rate, %20 (95% CI, 11-29)44 (95% CI, 33-55)63 (95% CI, 39-87)
Median PFS, mo3.6 (IQR, 0.9-6.6)6.2 (IQR, 3.6-9)8.1 (IQR, 4-11.1)
6-Mo PFS rate28 (95% CI, 18-38)53 (95% CI, 42-64)63 (95% CI, 39-87)

In a Cox multivariate analysis that was stratified for age, sex, PS, stage, CA 19-9 value at baseline, and treatment arm, CA 19-9 response to chemotherapy independently predicted OS (HR 0.59; 95% CI 0.45-0.77; P = .0001). A significant difference in PFS was observed between Groups 1 and 2 (P = .01), between Groups 1 and 3 (P = .005), and between Groups 2 and 3 (P = .04).

Correlation Between Radiologic Response and Carbohydrate Antigen 19-9 Response

One hundred eleven patients (44%) had a complete or partial response to chemotherapy, 76 patients (30%) had stable disease, 57 patients (23%) had progressive disease, and 8 patients (3%) were not assessed. Among the responders, 89 patients (80%) had basal CA 19-9 values >1 times the ULN and at least 1 follow-up CA 19-9 measurement. In these patients, an increase in the CA 19-9 value or a decrease <50% was observed in 4 patients (5%; median survival, 15 months), a decrease between 50% and 89% was observed in 33 patients (37%; median survival, 10.7 months), and a decrease >89% was observed in 52 patients (58%; median survival, 16.4 months). Among those with stable disease, 64 patients (84%) had basal a CA 19-9 value >1 time the ULN and at least 1 follow-up CA 19-9 measurement. In these patients, an increase in CA 19-9 value or a decrease <50% was observed in 23 patients (36%; median survival, 8.5 months), a decrease between 50% and 89% was observed in 32 patients (50%; median survival, 10 months), and a decrease >89% was observed in 9 patients (14%; median survival, 20.1 months). Among those with progressive disease, 49 patients (86%) had basal a CA 19-9 value >1 time the ULN and at least 1 follow-up CA 19-9 measurement. In these patients, an increase in the CA 19-9 value or a decrease <50% was observed in 40 patients (82%; median survival, 4.2 months), a decrease between 50% and 89% was observed in 8 patients (16%; median survival, 6.2 months), and a decrease >89% was observed in 1 patient (2%; survival, 2.6 months).

DISCUSSION

The current analysis confirms that the CA 19-9 value at baseline in patients with stage III and/or metastatic pancreatic adenocarcinoma is correlated independently with survival and that there is a significant relation between CA 19-9 decline during chemotherapy and survival. The prognostic value of baseline CA 19-9 value was reported previously by many authors.5, 18, 19, 28 The median value of baseline CA 19-9 (1167 U/mL) and the median survival of patients with CA 19-9 basal levels below or above the median (11.9 months and 8 months, respectively) observed in our series were consistent with those reported in previous series in patients with advanced pancreatic cancer (958-2000 U/mL, 10.3-14.9 months, and 5.5-7.4 months, respectively).5, 15, 19, 28 Accordingly, our findings endorse the application of the pretreatment CA 19-9 value as a stratification factor for future trials in advanced pancreatic cancer.

Several clinical trials analyzed the use of the CA 19-9 response to chemotherapy as a potential surrogate marker of treatment efficacy.12, 13, 15-17, 19, 28 Usually, a decline in the basal CA 19-9 level by ≥20%12, 15, 17 or by ≥50% is defined as biochemical response.13, 16, 28 In contrast to those studies, a recent series analyzing larger number of patients with more robust methodology and statistical tools questioned the value of CA 19-9 response at the nadir as a reliable surrogate endpoint for survival.19 In our current retrospective analysis of prospectively collected data, a significant correlation of CA 19-9 response both at the nadir and at a fixed time point (with adjustment for guarantee-time bias) with survival was observed. Important strengths of this study are the large number of patients; the small rate of missing follow-up values (11% vs 29%-43% in previous series15, 19); the completeness of the Cox model (only 1 missing value); and the involvement of a single institution, which reduced the risk of heterogeneity in laboratory and testing CA 19-9 methods, in clinical and radiologic criteria on which treatment decisions are based, and in patient management strategies at the time of treatment failure. Apart from these factors, the difference in results may rely on the more active chemotherapy regimens administered to most of our patients in terms of both tumor response and disease control duration. The overall radiologic response rate was 44%, the “biochemical response” rate (a decline ≥50% in the CA 19-9 value) was 64% (137 of 213 patients who had initial levels >1 time the ULN), and the median PFS was 6.2 months as opposed to radiologic and biochemical response rates of 9% and 45%, respectively (112 of 247 patients), and a median PFS 4.1 months reported in other series.19 The difference in activity also was suggested by the time to CA 19-9 nadir (3.5 months vs 2 months in previous series19), implying more enduring chemotherapy activity. More active treatment may be better for addressing the validity of a candidate tool for tumor response assessment. In the current analysis, a CA 19-9 decrease ≥90% was associated with a remarkable prolongation of survival compared with decreases between 50% and 89% (Tables 2, 4). The 90% cutoff level may be more appropriate than the 50% level for defining a “biochemical response.” This also was suggested by our analysis of the correlation between biochemical and radiologic responses. A correspondence between radiologic partial or complete response and a CA 19-9 decline ≥50% was observed in 80% to 100% of patients in previously reported series.3, 19, 29, 30 Consistently, in the current analysis, that correspondence was observed in 95% of patients. However, a “major biochemical response” was associated with longer survival compared with a “minor biochemical response” (median survival, 15 months vs 10.7 months). Among those with stable disease, 67% to 71% of patients in other series and 64% of patients in our current series had a ≥50% tumor marker decrease.3, 19, 29, 30 Again, a major biochemical response was associated with a doubling of the median survival compared with a minor biochemical response (20.1 months vs 10 months), which resulted in little improvement with respect to nonresponders (8.5 months). Finally, among patients who had radiologic progressive disease, a tumor marker response ≥50% was reported previously in 30% to 48% of patients.3, 19, 29, 30 In our series, 18% of patients who experienced progressive disease had a biochemical response, and only 1 of those patients had a major biochemical response. The OS for patients with radiologic progressive disease and a biochemical response, as previously reported by others,16 was significantly better than the OS for patients with a radiologic response but without a biochemical response.

The interpretation of the relation between biochemical response and survival is troublesome, because the hypothesis cannot be rebutted that response may represent a prognostic marker only because it may select the patients who would have survived longer even if the therapy had no effect at all.26, 27 In fact, patients who eventually become responders must survive long enough to be evaluated as responders, whereas patients who die before response assessment automatically are included in the group of nonresponders.26, 27 Thus, the risk of death is overestimated for nonresponders and underestimated for responders. In an attempt to limit the bias that response may act as a selection mechanism of a prognostically favorable subgroup of patients based on pretreatment characteristics and to clarify whether survival depends on response status, we used the landmark method by choosing a guarantee time that was as long as the time to the first radiologic response evaluation, as proposed previously.19, 26, 27 The hypothesis that survival from landmark depended on response status at landmark was confirmed by both univariate and multivariate analyses that stratified patients for known prognostic factors for survival. However, this observation does not exclude the possibility that response is merely a marker of unknown prognostic factors. The causality of the relation between response and survival can be proven only with randomized trials comparing treatments that differ both in terms of activity and efficacy.26, 27 Data from 5 prospective series, which used the same inclusion criteria and included 1 randomized series comparing single-agent gemcitabine with the PEFG regimen, were used in the current analysis.9, 20, 21-23 Treatment groups did not differ significantly in terms of baseline characteristics (Table 1). Single-agent gemcitabine had significantly inferior efficacy compared with 4-drug–based chemotherapy combinations both in univariate and multivariate analysis (Table 3, Fig. 3). Although the current analysis does not provide a level of evidence comparable to that obtained from a randomized trial, the difference in terms of biochemical response between single-agent gemcitabine and combination chemotherapy reinforces the possibility that survival is influenced not only by the chance that a patient is classified into 1 response group or the other but also by response to treatment.

On the basis of the current findings, minor and major biochemical response may be proposed as complementary measures to radiologic response, which is particularly troublesome in this disease because of desmoplastic tissue within and/or around the tumor, for more appropriate assessments of the activity of new chemotherapy agents and combinations. CA 19-9 represents an ideal endpoint for clinical studies because it is quick, safe, and easy to measure and allows the assessment of patients with “unmeasurable” disease. CA 19-9 is not expressed in 10% to 15% of patients with pancreatic cancer who are negative for the Lewis blood group antigen; therefore, the current considerations cannot be applied to this subset of patients. The information provided by this tool also may be useful in clinical practice to drive treatment decisions in critical situations, such as whether to continue with palliative chemotherapy in patients who have radiologically stable or progressive disease.

Conflict of Interest Disclosures

The authors made no disclosures.

Ancillary