Prognostic value of the Memorial Sloan Kettering prognostic score (MPS) in metastatic pancreatic adenocarcinoma

Abstract

BACKGROUND:

The Memorial Sloan Kettering (MSK) Prognostic Score (MPS), a composite of the neutrophil-lymphocyte ratio (NLR) and albumin and is an objective prognostic tool created as a more readily available alternative to the Glasgow Prognostic Score. Our prior analysis of patients with metastatic pancreatic adenocarcinoma (mPDAC) suggested that the MPS may predict survival, although did not control for clinically relevant factors.

METHODS:

MPS scores were calculated for mPDAC patients treated at MSK from January 1, 2011 to December 31, 2014. An MPS scale of 0–2 was utilized: MPS = 0 for albumin ≥ 4 g/dl and NLR ≤ 4 g/dl, MPS = 1 for either albumin < 4 g/dl or NLR > 4 g/dl, and MPS = 2 for albumin < 4 g/dl and NLR > 4 g/dl. Performance status, anti-neoplastic therapy, presence of thromboembolism (TE), radiation therapy, and metastatic sites were also analyzed. The associations with overall survival were examined using time-dependent Cox proportional hazard regression analyses.

RESULTS:

A multivariate model revealed that higher MPS scores at diagnosis (MPS=2 vs 0, HR = 1.41, 95%CI: 1.13–1.76), liver metastases, radiation therapy, hospital admissions, TE, and performance status were associated with worse overall survival. The median overall survival in patients with MPS 0, 1 and, 2 were 12.9, 9.0 and 5.4 months, respectively.

CONCLUSION:

The MPS, an easily calculated composite of albumin and neutrophil-lymphocyte ratio, is an objective tool that may predict survival in mPDAC independent of performance status, disease characteristics, and cancer therapy.

Lay summary:

The Memorial Sloan Kettering Prognostic Score (MPS) is a new scoring system that incorporates markers of inflammation found in individuals’ blood at diagnosis of metastatic pancreatic cancer. Data suggests the MPS may help determine prognosis.

Precis for use in Table of Contents:

The Memorial Sloan Kettering Prognostic Score (MPS), a composite of neutrophil-lymphocyte ratio and serum albumin, is a novel scoring system that may have prognostic utility in patients with metastatic pancreatic cancer, independent of performance status, disease characteristics, and cancer therapy.

INTRODUCTION

Pancreatic cancer is estimated to be the fourth leading cause of cancer-related deaths in 2020.¹ The incidence of the disease is projected to increase by as much as 55% by 2030.² Over 50% of pancreatic cancer patients have distant metastases at diagnosis, which correlates with a dismal 5-year relative survival of 2.9%.³ Amidst the growing number of individuals at risk for this disease arises a demand for accurate, non-invasive, and objective prognostic tools to aid clinicians in choosing appropriate treatment regimens, stratifying patients in trials, and communicating evidence-based life expectancy to patients and families.

As observed in many cancer types, chronic inflammatory states promote development of pancreatic neoplasms.⁴ Given this association, numerous inflammation-based scores have emerged as prognostic tools in pancreatic cancer and multiple other malignancies over the past few decades, including Glasgow Prognostic Score (GPS), modified Glasgow Prognostic Score (mGPS), neutrophil-lymphocyte ratio (NLR), and platelet-lymphocyte ratio (PLR).^5,6 Among the various scores, GPS/mGPS and NLR have the most robust data in metastatic pancreatic ductal adenocarcinoma.^7–12 The GPS, which measures albumin as well as c-reactive protein (CRP), has been found to have independent predictive prognostic value in resectable pancreatic cancer and is perhaps superior to other inflammatory-based scores such as NLR, PLR, prognostic index (PI), and prognostic nutritional index (PNI).^7,13,14 There are data to also suggest that GPS/mGPS have prognostic value in patients with unresectable, recurrent, and mPDAC.^8–10,15 Additionally, the mGPS was used to stratify patients in a randomized phase III study investigating ruxolitinib and capecitabine in mPDAC.¹⁶ Despite the existing data indicating mGPS can aid in prognosis of patients with advanced pancreatic cancer, the score has not been routinely incorporated as a tool for patient stratification in therapy trials.

Given the independent prognostic value of both NLR^11,12,17,18 and albumin^19–22 in pancreatic cancer, and the fact that they are routinely ordered (whereas CRP, in the GPS, is not), we combined these markers to form the MPS, which to the best of our knowledge is the first tool combining NLR and albumin to be studied in patients with mPDAC. We first evaluated the composite of NLR and albumin retrospectively in mPDAC patients treated at Memorial Sloan Kettering Cancer Center (MSK). Our initial analysis found that this combination of albumin and NLR, the Memorial Sloan Kettering (MSK) Prognostic Score (MPS), allowed these patients to be stratified into three discrete categories based on MPS.²³ However, this primary evaluation did not include clinical characteristics relevant to prognosis in pancreatic cancer. We therefore retrospectively calculated the MPS and abstracted clinical characteristics including performance status, metastatic site, anticancer therapy, and diagnosis of thromboembolism in mPDAC patients. Significant variables in a univariate analysis were then included in a multivariable analysis. The overall goal was to validate the prognostic value of a reproducible and easily calculatable score in pancreatic cancer.

METHODS

One thousand four hundred and fifty-five adult (>18 years) patients were retrospectively identified by ICD-O code for pathology-confirmed stage IV pancreatic ductal adenocarcinoma at time of presentation to MSK from January 1, 2011 to December 31, 2014 (Figure 1). From this cohort, n = 622 patients were excluded due to operable disease or unavailable pre-treatment NLR or albumin data. Patients missing pre-treatment NLR or albumin data had presented for a second opinion only or had been previously treated at outside institutions. The charts of all individuals indicated as “alive” in DataLine, an MSK institutional database, as of June 1, 2019 were manually investigated to ensure accurate vital status. Sixty patients were excluded due to incomplete follow-up data. The final analysis cohort included n= 773 patients.

Figure 1:

Study design.

PDAC, pancreatic ductal adenocarcinoma; MSK, Memorial Sloan Kettering Cancer Center; NLR, neutrophil-lymphocyte ratio.

An MPS scale of 0–2 was utilized: MPS = 0 for albumin ≥ 4 g/dl and NLR ≤ 4 g/dl, MPS = 1 for either albumin < 4 g/dl or NLR > 4 g/dl, and MPS = 2 for albumin < 4 g/dl and NLR > 4 g/dl (Figure 2). The NLR and albumin cutoff values were chosen based on findings from our prior work as well as that of others.^23–25

Figure 2:

Memorial Sloan Kettering Prognostic Score Calculation

MPS, Memorial Sloan Kettering Prognostic Score; NLR, neutrophil-lymphocyte ratio.

Various clinical and diagnostic characteristics were extracted utilizing DataLine to examine their association with overall survival (Table 1). Clinical characteristics in the DataLine search included age, sex, ethnicity, insurance status, marital status, number of inpatient admissions, weight, CA 19–9 level at presentation, presence of deep venous thrombosis/pulmonary embolism (hereafter referred to as “thromboembolism” or “TE”), date of death, and date of first chemotherapy administration.

Table 1:

Demographic Data of Patients with Advanced Pancreatic Adenocarcinoma at MSK Between January 1, 2011 to December 31, 2014

Characteristic	Value
Total, No.	773
Median Age at Cancer Diagnosis, (range)	68 (31, 95)
Female, No. (%)	339 (44%)
Male, No. (%)	434 (56%)
Ethnicity, No. (%)
Hispanic or Latino	36 (4.7%)
Not Hispanic or Latino	734 (95%)
Not specified	3
Baseline CA19–9, U/mL (range)	1195 (0, 4825530)
Unknown	84
ECOG Performance Status, No. (%)1
0	38 (5.5%)
1	372 (54%)
2	201 (29%)
3	68 (9.8%)
4	13 (1.9%)
Not specified	81
MPS, No. (%)
0	297 (25%)
1	308 (40%)
2	268 (35%)
Chemotherapy, No. (%)	609 (79%)
Summary of Chemotherapy2
Fluoropyrimidine-based	436 (70%)
Gemcitabine-based	459 (73%)
Fluoropyrimidine and gemcitabine combinations	73 (12%)
Erlotinib/Mitomycin/nab-paclitaxel monotherapy	9 (1.4%)
Clinical trial	95 (12%)
First-line Chemotherapy, No. (%)
FOLFIRINOX	302 (49.5%)
5FU monotherapy	4 (0.6%)
Gemcitabine and albumin-bound paclitaxel	102 (16.7%)
Other gemcitabine-based doublet3	55 (9%)
Gemcitabine monotherapy	106 (17.4%)
Gemcitabine and capecitabine	12 (2%)
GTX	6 (1%)
Clinical trial	22 (3.6%)
Total Lines of Therapy, No. (%)
0	149 (19%)
1	293 (38%)
2	182 (24%)
3+	149 (19%)
Radiation, No. (%)	43 (5.6%)
ERCP/stent placement4	58 (7.5%)
Site of Metastasis, No. (%)
Liver	544 (70%)
Lung	145 (19%)
Peritoneum	127 (16%)
Bone	19 (2.5%)
Thromboembolism, No (%)5	318 (41%)
Pre-treatment Thromboembolism	32 (4.1%)

MSK, Memorial Sloan Kettering Cancer Center; CA 19–9, Cancer Antigen 19–9; ECOG, Eastern Cooperative Group; MPS, Memorial Sloan Kettering Prognostic Score; FOLFIRINOX, 5-fluorouracil, leucovorin, irinotecan, oxaliplatin; 5FU, 5-fluorouracil; GTX, gemcitabine, capecitabine, docetaxel; ERCP, endoscopic retrograde cholangiopancreatography.

¹Performance status groupings documented as KPS were converted to ECOG based on the following: 0 = KPS 100 or ECOG 0; 1 = KPS 80–90 or ECOG 1; 2 = KPS 60–70 or ECOG 2; 3 = KPS 40–50 or ECOG 3; 4 = KPS 10–30 or ECOG 4.

²Generalized categories of all lines of chemotherapy administered.

³Includes gemcitabine plus oxaliplatin, cisplatin, irinotecan, or erlotinib.

⁴Includes biliary stent placement, pancreatic duct stent placement, and biliary drainage.

⁵Includes patients diagnosed with thromboembolism at any point after diagnosis.

In an attempt to account for biliary infections and/or inflammation that might falsely elevate inflammatory markers used in our model, we collected information on biliary procedures. The DataLine search included all ERCP (endoscopic retrograde cholangiopancreatography) CPT codes as well as codes related to biliary stent placement, pancreatic duct dilatation, or sphincterotomy. These procedures were included if they occurred within two weeks of diagnosis.

Metastatic involvement was abstracted from chart review as liver, lung, bone, or peritoneum based on cross-sectional imaging available at initial presentation to MSK (Table 1). Karnofsky Performance Status (KPS) or Eastern Cooperative Group (ECOG) Performance Status were classified as documented in the earliest clinical evaluation. ECOG or KPS scores documented as ranges were transcribed as only the lower or higher bounds of the range given, respectively. KPS scores were then translated into ECOG scores based on the following scale: 100 = 0, 80–90 = 1, 60–70 = 2, 40–50 = 3, and 10–30 = 4.²⁶ Chemotherapy regimens, total lines of chemotherapy, and the receipt of external beam or stereotactic body radiation to either primary or metastatic lesions were also abstracted from chart review (Supporting Methods).

The Kaplan-Meier method and the Cox proportional hazard regressions were used to estimate overall survival measured from date of diagnosis and examine its associations with clinical variables (Table 2). In variables present after diagnosis, the starting date of first chemotherapy, first stent procedure, hospital admission, TE, and the completion date of radiation were the transition time in the time-dependent regressions. Univariably important variables (p < 0.10) were included in multivariate analysis with the exception of baseline CA 19–9 level, considering a relatively large number of patients were missing this value (Table 1). A sensitivity analysis confirmed that inclusion of CA19–9 levels did not meaningfully impact the estimated effect of MPS on overall survival. To assess the discriminative ability of the multivariate model, the bootstrapping bias-adjusted concordance probability was estimated with 200 bootstrap replicates. All statistical analyses were performed in software packages R version 3.6.2 (The R Foundation for Statistical Computing).

Table 2:

Univariate Analysis of Clinical Characteristics in Advanced Pancreatic Adenocarcinoma Patients at MSK

Characteristic	n	HR	95% CI	p-value
Age at Cancer Diagnosis	773	1.07	0.99, 1.15	0.075
Gender	773			0.5
Female		—	—
Male		1.05	0.91, 1.21
Ethnicity	770			0.3
Hispanic or Latino		—	—
Not Hispanic or Latino		0.85	0.61, 1.21
CA19–9, logarithmic transformed	689	1.07	1.05, 1.10	<0.001
ECOG Performance Status	692			<0.001
0–1		—	—
2		1.47	1.24, 1.75
3–4		1.56	1.22, 1.98
MPS Group	773			<0.001
0		—	—
1		1.29	1.07, 1.54
2		1.96	1.63, 2.36
Site of Metastasis	773
Liver		1.32	1.13, 1.55	<0.001
Lung		0.99	0.82, 1.18	0.9
Peritoneum		1.15	0.95, 1.39	0.2
Bone		1.36	0.86, 2.15	0.2
Chemotherapy	609	0.79	0.66, 0.95	0.014
Biliary Procedure/Stent	58	1.46	1.05, 2.01	0.022
Hospital Admission	448	2.24	1.94, 2.6	<0.001
Radiation	44	2.1	1.52, 2.91	<0.001
Thromboembolism	318	2.21	1.91, 2.56	<0.001

MSK, Memorial Sloan Kettering Cancer Center; n, sample size; HR, Hazard Ratio; CI, Confidence Interval; CA 19–9, Cancer Antigen 19–9; ECOG, Eastern Cooperative Group; MPS, Memorial Sloan Kettering Prognostic Score.

RESULTS

Patient Characteristics

Of the n=773 patients analyzed, the median age was 68 years (mean 67.3) and 434 (56%) were male (Table 1). Almost all (n=734, 95%) identified as “not latinx”. Liver was the most prevalent site of metastatic involvement (n=544, 70%), followed by lung (n=145, 19%), peritoneum (n=127, 16%), and bone (n=19, 2.5%). Median CA 19–9 at time of diagnosis was 1,195 units/mL (range 0 to 4,825,530 units/mL). Thromboembolism was diagnosed in 32 patients (4.1%) prior to initiating chemotherapy and 318 (41%) at any point after diagnosis. The relative frequencies of ECOG scores 0–4 were as follows: 0 = 38 patients (5.5%), 1 = 372 patients (54%), 2 = 201 patients (29%), 3 = 68 patients (9.8%), and 4 = 13 patients (1.9%). Fifty-eight patients (7.5%) underwent ERCP with biliary stent placement, pancreatic duct stent placement, or biliary drainage within two weeks of diagnosis. After analyzing all lines of systemic treatment that n=609 patients received, the most prevalent chemotherapy classifications were gemcitabine-based (n=459,73%) and fluoropyrimidine-based (n=436,70%). Approximately half of patients were treated with first-line FOLFIRINOX (n=302;49%) (including modified FOLFIRINOX) and 102 (16%) were treated with combination gemcitabine/nab-paclitaxel (gem/nab-P).

Univariable and Multivariable Analysis/Overall Survival Analysis

On June 1, 2019, the final date of analysis, 10 (1.6%) patients remained alive. For the entire cohort, the median overall survival was 8.8 months (95% CI 8.0–9.6 months) and median chart follow-up time was 82.7 months (95% CI 54.3–107.1 months) on patients alive (Supporting Table 1). Several clinical characteristics were univariably associated with worse OS: ECOG performance status of 3–4 or 2 versus ECOG 1, hospital admission, presence of liver metastases, receipt of radiation therapy, TE diagnosed prior to initiation of treatment or during, requirement for biliary procedure/stents, higher CA 19–9 level, and higher MPS (Table 2). With respect to CA 19–9 levels, every natural number e (2.7) fold increase in baseline CA19–9 was associated with 7% increasing in the hazard of death (HR 95%CI: 1.07 (1.05, 1.10), p = <0.001). All characteristics univariably associated with overall survival remained associated with overall survival in multivariate analyses (p < 0.05) except biliary procedures/stents. The multivariable model had a bias-adjusted concordance probability of 0.657. In the multivariable analysis, higher MPS was associated with progressively worse survival: compared to MPS 0, MPS 1 hazard ratios (HR) = 1.18 (95% CI 0.97–1.43), and MPS 2 HR = 1.62 (95% CI 1.32–1.99) (Table 3). The median overall survival of patients with an MPS 0 was 12.9 months (95% CI 10.9–14.3 months), MPS 1 was 9 months (95% CI 7.4–10.3 months), and MPS 2 was 5.4 months (95% CI 4.3–6.6 months) (Figure 3 and Supporting Table 1).

Table 3:

Multivariate Analysis of Clinical Characteristics in Advanced Pancreatic Adenocarcinoma Patients at MSK ¹

Characteristic	HR (95%CI)	p-value
Age at Diagnosis	1.09 (1, 1.17)	0.040
ECOG Performance Status		<0.001
0–1	1
2	1.44 (1.21, 1.72)
3–4	1.32 (1.03, 1.68)
MPS		<0.001
0	1
1	1.18 (0.97, 1.43)
2	1.62 (1.32, 1.99)
Liver Metastasis at Diagnosis	1.60 (1.34, 1.91)	<0.001
Chemotherapy	0.56 (0.45, 0.69)	<0.001
Biliary Procedure/Stent	0.98 (0.70, 1.39)	0.928
Hospital Admission	2.12 (1.8, 2.5)	<0.001
Radiation	1.97 (1.37, 2.83)	<0.001
Thromboembolism	1.84 (1.57, 2.16)	<0.001

HR, Hazard Ratio; CI, Confidence Interval; CA 19–9, Cancer Antigen 19–9; ECOG, Eastern Cooperative Group; MPS, Memorial Sloan Kettering Prognostic Score.

¹n = 692

Figure 3:

Kaplan-Meier Curve of Patients Presenting with Advanced Pancreatic Adenocarcinoma at MSK Between January 1, 2011 to December 31, 2014 stratified by MSK Prognostic Score

DISCUSSION

Several inflammation-based prognostic scores have been developed and applied to a variety of malignancy types in order to aid clinicians and patients in objective, evidence-based, decision-making regarding clinical trial stratification, anti-neoplastic treatments, and advanced care planning. The neutrophil-lymphocyte ratio (NLR)^11,12,18,21 and serum albumin^19–22 have been independently investigated and found to have prognostic utility in pancreatic cancer. The Memorial Sloan Kettering (MSK) Prognostic Score (MPS) is, to our knowledge, the first score to combine serum albumin and neutrophil-lymphocyte ratio in mPDAC. Our study is also the first to demonstrate the prognostic value of the MPS on overall survival beyond clinically relevant factors. Additionally, our data confirms the prognostic significance of several patient and treatment characteristics including performance status, thromboembolic events, hepatic metastases, radiation therapy, and hospital admission. Although our data are from 2011–2014, the outcomes with mature follow up were consistent with present day overall survival values observed in patients with mPDAC at diagnosis.^27–30

Our study elucidated several clinical characteristics associated with overall survival; some of which corroborate or expand previously identified variables, and others which are novel. Performance status was among the first clinical indicators of prognosis established in all cancer types including pancreatic cancer.^31–33 Our results support the existing data that poorer ECOG performance status or KPS are associated poorer overall survival.^31,34 However, their relative subjectivity is an important limitation and underscores the need for precise objective clinical scores, such as the ones we utilize in our model. Early thromboembolic events have been found to confer poorer short term outcomes, including in 1,915 patients treated with chemotherapy at our center for exocrine pancreatic cancer.³⁵ Additional studies have had similar findings to ours in regards to venous thromboembolic disease portending poorer survival outcomes in pancreatic cancer patients, specifically in East Asian and hospitalized patients.^36,37 Metastatic involvement of the liver is well-reported in the literature as a poor prognostic indicator in pancreatic cancer, which is supported by our multivariable analysis.^38–40 The organization of our subgroups by metastatic site each includes patients with both isolated single-organ metastases and mixed metastatic disease, and thus does not allow for organ-specific correlations. However, further investigation of individuals with isolated lung metastases is important given the potential that this distribution of disease represents a biologically unique subset with differential outcomes and therefore may be used in treatment stratification.⁴¹

Radiation therapy in metastatic disease, while sometimes an effective method of providing pain relief, has an unclear relationship with mortality. In our multivariable analysis, radiation therapy was associated with an increased risk of death. However, only a small number of patients (n=52, 6.8 %) received radiation therapy. One single-institution study from the Netherlands in locally advanced or mPDAC patients (n=61; 62% metastatic) did not find any association between radiation therapy and overall survival.⁴² In contrast, a retrospective study of 72,763 stage IV pancreatic cancer patients from the National Cancer Data Base found that use of radiation therapy was associated with a 9% increased risk of death (HR 1.09; 95% CI [1.01, 1.17]).⁴³ These discrepant findings are intriguing and may be explained by large differences in sample size, fractionation of radiation schedule, the timing of radiation, and patient selection. Furthermore, our time-dependent model may augment the risk of death when compared with other models and therefore radiation administered closer to death can appear to be a greater risk factor for death. Radiation therapy is likely instead a proxy for severe pain associated with end-stage disease.

Strengths of our study include the large sample size, and manual abstraction of clinical data (which allowed for quality control and appropriate exclusion of some patient records), detailed treatment information, radiation therapy data, radiographic sites of metastases, performance status, CA 19–9 level, and thromboembolic diagnoses. In addition, our median chart follow-up was 82 months on patients alive, so the cohort was very mature at the time of analysis, which provides confidence around our conclusions. Our multivariable model achieved a concordance probability of 0.657 with internal biased correction, which corresponds to good discriminatory potential as a prediction tool. A larger study with independent external validation is warranted to further validate the predictability of MPS on overall survival, and to provide easy to interpret nomograms taking into account important time-dependent characteristics.

The MPS effectively divided our cohort of patients with stage IV pancreatic cancer into discrete groupings, independently associated with overall survival after controlling for several clinically relevant characteristics. Our study also identified that liver metastases, performance status, hospital admissions, radiation therapy, and TE were associated with poorer outcomes. The use of NLR and albumin as a composite to estimate prognosis supports the use of inflammatory markers in this setting as well as the association between high inflammatory states and more aggressive malignancies. Although the Glasgow Prognostic Score (GPS) and its updated version, the modified GPS (mGPS), are the best studied and widely utilized inflammation-based prognostic scores, they are less readily available for use in the United States, which does not routinely collect CRP levels. Serum albumin, neutrophil count, and lymphocyte count are routinely measured in pancreatic cancer patients internationally, making the MPS an accessible prognostic score worldwide. Routine application of the MPS and screening for the above clinical variables in patients with pancreatic cancer may aid in stratification when enrolling patients for clinical trials. These clinical tools will also aid oncologists in providing patients with more accurate survival predictions and may serve as triggers for palliative care or other specialty referrals to help patients and families navigate these serious illnesses. Evaluation of the MPS, prospectively in an independent population, is the vital next step in validation as an objective prognostication tool. Parallel investigations of the utility of the MPS in resectable pancreatic cancer and other malignancies are also warranted. Additionally, a dedicated comparison of the MPS to the mGPS in pancreatic cancer populations would determine the relative accuracy of each inflammatory marker-based score.

There are several key limitations to the MPS as well as our analysis overall. Firstly, our analysis was a single-institution retrospective analysis with a neither ethnically nor racially diverse population at a quaternary referral center. Only 55% (n=404) of our cohort received first-line gem/nab-P or FOLFIRINOX (or mFOLFIRINOX). This relatively low frequency of first-line administration of gem/nab-P or FOLFIRINOX is in part explained by our inclusion period 2011–2014; notably, the United States’ Federal Drug Administration approval of gem/nab-P for pancreas cancer occurred in late 2013. The MPS, as a composite of inflammatory markers, may be falsely elevated in patients harboring infections at time of serum collection. While we attempted to mitigate this potential confounder with our inclusion of biliary procedures, it is possible some patients had occult infections, rheumatologic diseases, or other unknown confounders. Performance status was not recorded in outpatient notes for approximately 80 (10 %) patients and therefore only 692 (90%) patients were included in the multivariate analysis. Many performance statuses were reported as ranges and we recorded these intervals only as the poorer value; this fact likely accounts for the relatively low frequency of ECOG 0 patients. Although the translation of KPS into ECOG performance status is another potential source of error, we utilized a conversion scale based on studies indicating the scales can be interchanged with high fidelity.^26,44

To the best of our knowledge, this is the first study that combines the NLR and albumin into a composite score with the potential to predict survival in patients with stage IV PDAC, independent of several clinically relevant treatment and disease characteristics.