Comparison of Pancreatic Cancer Treatment Outcomes: A Real-World Data Analysis

Since the ground-breaking phase III MPACT trial showed clear benet of gemcitabine-nab-paclitaxel, this regimen has emerged as the standard of care for advanced pancreatic adenocarcinoma (PAC). Prior to this study, few studies have shown how results from randomized controlled trial translate to the real world. This study investigated how patients fared in the real world. This single-centre, retrospective study was conducted in the Vaasa Central Hospital, Finland. 148 patients with PAC (ICD-10 C25) between 1/2011-12/2016 were identied with resectable, locally advanced or metastatic disease. Information about the basic characteristics, treatment regimens and adverse effects (AEs) were extracted from patient les and analysed.


Conclusion
This study shows that active treatment is worth pursuing in most PAC patients and study results from this real-world data study differ from randomized controlled trials. Special caution should be applied when continuing chemotherapy in patients with ECOG 2 and who are older than 70 years.

Background
The diagnosis of pancreatic ductal adenocarcinoma (PAC) has been associated with poor outcomes in the vast majority of patients since it was rst identi ed. Although the mortality rates associated with the majority of cancers are falling, the incidence of PAC is increasing, and its associated mortality rate remains the same as it was a decade ago; the 5-year survival rate is 2-5% (1-3). Pancreatic cancer (PC) was the third most common cause of cancer-related death in Finland between 2013 and 2017(1), and globally, it was the seventh most common in 2015 (2). These poor outcomes might be partly explained by the fact that PAC is an aggressive cancer that silently progresses, and the majority of patients are diagnosed with stage IV metastatic disease (3). At this late stage, surgical intervention with curative intent is rarely possible, and few treatment options other than non-curative chemotherapy remain (4).
Gemcitabine (gem) monotherapy has been the standard rst-line treatment for PAC since Burris et al. in 1997 showed signi cantly better results with the newly approved gem than with the traditionally used 5uorouracil (5-FU)(5). Subsequent trials over the past two decades have unsuccessfully explored the use of gem-combination therapies, trying to increase the OS when compared with the results obtained with gem monotherapy (6-11).
Only recently in the phase III MPACT and PRODIGE4/ACCORD 11 studies have the use of combination regimens such as oxaliplatin, irinotecan and 5-FU/leucovorin (FOLFIRINOX) and gem plus nab-paclitaxel (gem + nab-P) been shown to have superior e cacy to gem monotherapy (12,13). These combination regimens resulted in improvements in the median OS and in disease-related symptoms (median OS 6.8 vs 11.1 months, p < 0.0001 for gem vs FOLFIRINOX and 6.7 vs 8.5 months p < 0.00015 for gem vs gem + nab-P) (14). Currently, both FOLFIRINOX and gem + nab-P are used as "gold standard" rst-line (1L) treatments rather than gem monotherapy for PAC and especially for metastatic PAC (15). However, both of these treatment regimens are associated with signi cantly higher levels of toxicity, and therefore, they are recommended only for patients with Eastern Cooperative Oncology Group (ECOG) performance status scores of 0-1(16). In cases in which neither FOLFIRINOX nor gem + nab-P can be used because the patient has a ECOG score of 2 or higher or because of the risk of grades 3-4 AEs, gem monotherapy is still the preferred treatment (16). To date, a few studies have developed treatment options following the progression of disease after 1L treatment (17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27), but a reliable standard second-line (2L) or later-line treatment has yet to be established; currently, many different regimens are being considered and offered (14)(15)(16)20).
Most of the studies that have been conducted in search of a new standard treatment for PAC have been set up as clinical trials, with very little real-world data (RWD) used for comparison or veri cation (28, 29).
Randomized controlled trials (RCTs) have long been considered the "gold standard" for obtaining clinical data on the e cacy and safety of treatments (30), but they are expensive and take a relatively long time to obtain results that could be utilized in regular clinical practice. RCTs are strictly controlled by a protocol designed to determine whether there is a bene cial treatment effect in a selected patient population.
Usually, treatments before the initiation of a protocol are highly restricted, while no restrictions are usually made after explorative treatment. In the real-world setting, it is challenging to recreate the same conditions, and therefore, RCT evidence is used to treat patients who are not actually represented in primary RCT study populations.
Recently, however, discussions about the potential role of using RWD or real-world evidence (RWE) to provide additional information that could improve health outcomes rather than relying on RCTs alone have started to take place (31). RWD from regular clinical practice better re ect the reality of treatment usage and the effects on patient outcomes, as the treatments administered to patients are not dictated by a protocol intended to address certain research questions. Treatment decisions are made by individual doctors and are in uenced by patients, relatives, hospital regulations, other social factors and the nancial situation of the patients. Thus, it is likely to become increasingly important to ensure that treatment regimens truly provide bene ts when used as a part of routine clinical practice in real populations, not just in RCTs (32). Comparisons of RWD and data from clinical trials serve as an important part of adapting new treatments for use in clinical practice and increase transparency between clinicians and those developing the treatment regimens; this process, in turn, is likely to help individual patients receive the right treatment at the right time (30,33).
In the present study, we conducted a retrospective analysis with RWD from Vaasa Central Hospital to evaluate the effects of various treatments on outcomes in patients with PAC, spanning the spectrum from no treatment to a maximum of seven lines.

Patients
Patients who were diagnosed with pancreatic cancer (PC), speci cally PAC, and admitted to the oncology clinic of Vasa Central Hospital between January 2011 and December 2016 were included in this retrospective study. First, a search was carried out in a patient register by a statistician using the ICD-10 code for PAC (C25). The list of eligible patients was extracted, and 160 candidates who met the criteria were identi ed (n = 160). (Fig. 1A). Then, two independent researchers in 2018-2019 carefully reviewed the patient les to con rm the diagnosis based on the histopathology reports of biopsies or postsurgical material. Patients who were diagnosed with a neuroendocrine tumour (NET) were excluded (n = 10).
Additionally, patients who were initially clinically diagnosed with PAC and started chemotherapy for this indication but were later found to have a different primary tumour (such as metastasized NSCLC or cholangiocarcinoma) were excluded (n = 2) (Fig. 1A).
Information such as patient age, sex, tumour characteristics (resectable, locally advanced or metastatic), operation status, comorbidities, baseline biochemistry test results (alanine aminotransferase (ALAT) and bilirubin levels and leukocyte and neutrophil counts), carbohydrate antigen 19 − 9 (CA19-9) levels, treatment regimens, performance status, hospitalization and number of chemotherapy treatment lines received was extracted from the patient les. A treatment line was de ned as a minimum of 3 cycles of the same chemotherapy regimen. The primary endpoints were overall survival (OS), the frequency and grade of adverse effects (AEs), type of treatment and number of treatment lines received.

Statistical Analyses
Patient and tumour characteristics, treatments and AEs are reported as frequencies and percentages for categorical variables and median with ranges or interquartile ranges or means with 95% con dence intervals for continuous variables. These characteristics are also reported in the subgroups.
Kaplan-Meier survival analysis was used to investigate OS and determine differences in survival between the subgroups. Signi cant differences between the subgroups were determined with the log-rank test, and the results are presented as the survival curves. Analyses were performed using R version 3.5.3.

Patient characteristics
A total of 148 patients who met the inclusion criteria were identi ed and included in the analysis (Fig. 1  a). Of these 148 patients, 109 had received chemotherapy as either curative neoadjuvant, adjuvant or rst-line non-curative treatment. The tumour and patient characteristics are listed in Table 1. The patients were divided into subgroups according to the stage of disease at diagnosis, and their baseline characteristics were analysed. The male to female ratio in our study was 77:71. Age was determined from the date of birth to the date of diagnosis, and the median age was 70.5 years. The distribution of ECOG scores at baseline was as follows: 22 patients had an ECOG score of 1 (of whom 10 had locally advanced disease, 9 had metastatic disease, and 3 had resectable disease), 106 patients had an ECOG score of 2 (of whom 39 had locally advanced disease, 49 had metastatic disease, and 18 had resectable disease) and 20 patients had an ECOG 3 (of whom 5 had locally advanced disease and 15 had metastatic disease) ( Table 1).
At baseline, the median biochemical values were 31 U/l for plasma ALAT (females/males: 27/31 in the locally advanced group, 32/30 in the metastatic group and 45/38.5 in the resectable group), 13 µmol/l for plasma bilirubin (13/16 in the locally advanced group, 13/12 in the metastatic group and 13/15.5 in the resectable group) and 284 kU/l for serum CA19-9 (49/176 in the locally advanced group, 2194/1211.5 in the metastatic group and 22/15 in the resectable group). Fifty-two patients (35%) had diabetes type I or type II (14 in the locally advanced group, 29 in the metastatic group and 9 in the resectable group). Ninety-four patients (63.5%) were using opioids for pain management during our study period. A total of 35 patients (23.6%) had major comorbid illnesses (26 in total) or secondary cancers (9 in total) before or during our study period, such as hypertension, ischaemic heart disease, kidney cancer, breast cancer, severe depression or brain infarction. (Table 1) Tumour characteristics The primary tumour was found in the head of the pancreas in 91 patients, in the body of the pancreas in 13 patients or in the tail of the pancreas or another unde ned location in 35 patients ( Table 1). The median tumour size was 3.7 cm (0.4-14 cm). For males and females respectively, the tumour sizes were 3.5 cm and 4 cm in the locally advanced group, 3.4 cm and 4.5 cm in the metastatic group and 3.2 cm and 3.5 cm in the resectable group. The tumour stage at diagnosis was metastatic in 49.3% of the patients, locally advanced in 36.5% and resectable in 14.2% (Supp Fig. 5).

Treatment results
For the patients who received treatment of any kind, the median OS was 10.4 months (8.6-13.3 95% CI), while the patients who did not receive any treatment had a median OS of 2.8 months (2.2-5.2 95% CI) (p<0.01) (Fig. 2 a). Figure 2 b shows the OS for all patients according to treatment characteristics. Patients who received neoadjuvant or adjuvant chemotherapy after surgical intervention had a median OS of 31.2 months (16.9-NA 95% CI), while patients who received chemotherapy as a 1L treatment had a median OS of 8.6 months (7.0-10.6 95% CI), and the patients who received no treatment had a median OS of 2.8 months (2.2-5.2 95% CI) (p<6e-10) (Fig. 2 b). A total of 77 patients received chemotherapy as a 1L treatment. Sixty-three patients received gem monotherapy, 7 received gem+nab-P, and 7 received some other regimen as the 1L. The group receiving gem monotherapy had a median OS of 7.6 months (6.6-9.8 95% CI), the group receiving gem+nab-P had an median OS of 10.6 months (6.3-NA 95% CI), and the others group had a median OS of 18.2 months (9.2-NA 95% CI) (p<0.107) (Fig. 3 a and Supp Fig. 6). To overcome the limitation of a small sample size and better assess the outcomes of the patients receiving gem+nab-P, Figure 4 a and b show the OS of patients who received gem+nab-P at any time during treatment versus the other patients. Figure 4 a shows all patients, excluding patients who underwent surgery, while Figure 4 b shows only those with metastatic disease. The patients who received gem+nab-P had a median OS of 18.3 months (11.2-31.9 95% CI), while those who received some other type of treatment had a median OS of 7.2 months (6.3-9.1 95% CI) (p<0.008) (Fig. 4 a). Among the patients with metastatic disease, those who received gem+nab-P had a median OS of 11.2 months (8.2-NA 95% CI), while those who received other types of treatment had a median OS of 6.3 months (4.4-9.5 95% CI) (p<0.112) (Fig. 4 b).

Safety
The frequency, types and grades of AEs are listed by subgroup in Table 2. AE classi cation was performed according to version 4 of the National Cancer Institute Common Terminology Criteria for Adverse Effects (CTCAE). The most commonly observed side effects in this study were grades 2-4 bone marrow toxicities, with 19 cases of anemia, 49 cases of leukopenia, 55 cases of neutropenia and 9 cases of thrombocytopenia. Febrile neutropenia was observed 7 times. Other commonly observed AEs were nausea, diarrhoea and infections related to the administered treatment. In our study, a total of 12 fatal AEs were recorded, with 1 fatal case of thrombocytopenia, 3 cases of fatal sepsis, 1 case of fatal leukopenia, 3 cases of fatal haemorrhage, 1 case of fatal cholecystitis, 1 case of fatal diarrhoea and 2 cases of fatal febrile neutropenia. An important observation is that while a total of 12 fatal AEs were recorded, they occurred in a total of 7 patients, with a few patients experiencing more than one fatal AE simultaneously. Five of the patients with fatal AEs had locally advanced disease (all of them were older than 70 years, and 4 of these had ECOG scores of 2, while 1 had an ECOG score of 1), 1 of the patients had metastatic disease (older than 70 years, with an ECOG score of 2) and 1 had resectable disease (aged 65, with an ECOG score of 2). The division by subgroup yielded the following results: out of all 281 recorded AEs of any grade, 116 occurred in patients with metastatic disease, 110 in patients with locally advanced disease and 55 in patients with resectable disease. Another 3 patients died within one month of the administration of treatment; however, because these fatalities did not occur in the hospital setting and there was no indication of a direct treatment-related AE, they were excluded from the toxicology analyses. None of these 3 patients were receiving their rst treatment cycle when they died.
AEs seen only in individual patients or rarely associated with chemotherapy were all grouped together in the Other group. (Table 2) Discussion To our knowledge, this study provides some of the most mature RWD about treatment outcomes from a single institute and serves as a report on how results from RCTs translate to patients in routine clinical practice.
Although the OS rate of PC patients is one of the worst in any disease, our data clearly show the bene t of active treatment. As seen in Fig. 2a, the treated group had a signi cantly longer OS than the untreated group. Active treatment results in several additional months of life. Extending the life expectancy from 2.8 months to 10.4 months (Fig. 2a) is signi cant, and it is important to discuss this possibility with every patient diagnosed with PAC. Furthermore, our analysis showed a clear bene t of surgery in patients with resectable disease and, to some extent, those with locally advanced disease. Patients who received neoadjuvant or adjuvant chemotherapy achieved a median OS of 31.2 months compared to 8.7 months for patients receiving 1L chemotherapy and 2.7 months for untreated patients (Fig. 2b). While locally advanced disease is usually inoperable, a few of our patients with locally advanced disease underwent surgery. Our study also clearly showed the generally more favourable median OS of patients diagnosed with locally advanced disease compared with the OS of those diagnosed with metastatic disease (9.2 vs. 4.9 months) (Fig. 2c). This indicates the need for better diagnostic tools, such as biomarkers, and better imaging technology, to increase the number of patients who are diagnosed at an earlier stage of disease, as other studies have also shown (34)(35)(36)(37)(38); novel therapies are also needed.
Increasing the treatment duration and number of treatment lines administered resulted in a signi cant increase in OS from 6.3 months with one line of treatment up to 19.1 months with 3 or more lines of treatment (p < 0.002) (Fig. 3b). This nding is subject to survival bias, with healthier patients living longer and therefore receiving more treatment. It is thus not necessarily the treatment that is prolonging their lives. Further research on the subject is needed before conclusions can be drawn. However, these results provide evidence of the bene t of treating younger patients with lower ECOG scores more aggressively for longer periods of time.
Even though our data largely show relatively short survival times after diagnosis, there were some individual cases that stood out. A few patients treated early during our study period were still alive after 5 years and showed no signs of disease, with one patient showing complete remission of disease (con rmed via radiological examination) after a single cycle of gem monotherapy (PAD was positive for PAC). While this outcome is rare and further research is needed, these cases may serve as a source of hope for researchers, clinicians and patients because they show that remission can occur.
Because the combination of gem + nab-P was approved for PAC rather late during our follow-up period, the number of our patients who received this combination was low. Seven patients received gem + nab-P as a 1L treatment, while 63 patients received gem monotherapy, and 8 patients received another treatment. This limited our analysis of the e cacy and safety of this combination compared to the e cacy and safety of gem monotherapy.
To better assess the e cacy of gem + nab-P combination treatment, we analysed the OS of patients who received gem + nab-P at any time during their treatment schedule (among all treated patients and in the subgroup of patients with metastatic disease). The results were fairly conclusive; in all patients using gem + nab-P, the median OS was 18.3 months, which was longer than the 7.3 months in patients receiving other types of treatment, including gem monotherapy (p < 0.008). This analysis included all patients (except patients who had undergone surgical resection of the tumour) (Fig. 4a). In the subgroup of patients with metastatic disease, the results were also promising, with those who received gem + nab-P obtaining a median OS of 11.2 months, while those who received other treatments obtaining a median OS of 6.3 months; however, this difference lacked statistical signi cance (p < 0.112) (Fig. 4b).
Due to its poor toxicity pro le and the older age and poor performance status of our population, no patients in our study received FOLFIRINOX. Our analysis shows that there were signi cant toxicities and safety concerns associated with the administered treatments. A total of 10 (7 with recorded AEs and 3 without known cause) patients died within 30 days of chemotherapy administration. The most commonly observed AEs were grades 2-4 bone marrow toxicities, nausea, diarrhoea and infections. The frequency of AEs was high, with almost every patient experiencing an AE and many patients experiencing more than one. Grades 2-5 AEs occurred most frequently in the metastatic subgroup (134 out of 281), closely followed by the locally advanced subgroup (86 patients).
Due to key similarities in patient characteristics and geographical location, we used the results from the 2019 study by Bloomberg et al. (29) as a reference and compared the results to our own. We also considered the results from the landmark MPACT study by Von Hoff et al. from 2013 (12) when analysing our own results. However, due to signi cant differences between the inclusion/exclusion criteria and the manner in which our study was completed, a direct comparison of results between the MPACT study and our retrospective report is not possible. The present study and the study by Blomstrand  The results of the study by De Vita et al (39) and a meta-analysis of 50 studies presented at an international conference suggest that with careful and appropriate patient selection, a median OS of longer than 12 months can be attained with 1L gem + nab-P (40).
The patients in the present study experienced AEs more frequently and with higher grades than those in the study by Blomstrand et al., the MPACT study and previous studies on the subject (41,42). The most common side effects in the aforementioned studies were bone marrow toxicities. Since no data on dosage have been published, we cannot with con dence determine the cause of this discrepancy. Both of the studies used for comparison collected data from 1L treatment only, while the present study included all treatment lines, with a large proportion of patients receiving 2 or more lines (36%) and 15 patients receiving 3 or more lines. This increased exposure to therapy regimens and longer follow-up time coupled with advanced age and high ECOG scores might contribute to the higher numbers of AEs observed.
While Blomstrand et al. published RWD, there are key differences in the design of our study. The primary difference is that they examined the e cacy of gem + nab-P combination therapy in a real-world population and compared their ndings with the established MPACT study. Our study is mainly a retrospective report of the outcomes in all treated PAC patients in a hospital in central Finland during an approximately six-year period. However, our study provides important and mature data with substantial value to clinicians. The present study included every patient diagnosed with PAC regardless of insurance status, comorbidities, age, sex, ethnic subgroup, or socioeconomic status and therefore provides a very close approximation to the outcomes of treatment with many of the accepted therapies today in patients in Finland diagnosed with PAC.
The present study involved a heterogeneous patient group, which made analysis and interpretation of the results di cult and, to some extent, impossible. The heterogeneity and lack of standardization in 1L treatment regimens was the most disruptive factor and is also the largest limitation of our study. Due to these limitations, direct evaluation of e cacy between individual treatment regimens was not feasible.
We believe that the treatment decisions made during our study period were in uenced by an array of factors, including age and ECOG score at baseline, which could partly explain why a relatively small number (only 77 out of 148 patients) received chemotherapy as a 1L treatment and only 7 patients initially received gem + nab-P combination therapy. Furthermore, the fact that change has been slow in the eld of PAC treatment, with no new regimens introduced between 1997 and 2011/2013 that were improvements compared with gem monotherapy, together with the many years of clinical experience with the use of gem monotherapy, can provide an explanation for the low number of patients receiving 1L combination therapy.
In our study, AEs frequently occurred during chemotherapy administration. The frequency was higher than that reported in the literature, mainly because our patients were older and had worse performance status scores. Seven patients died during treatment. Another three patients died without clear signs of disease progression; these deaths were registered as treatment-related deaths, but none of these patients died after the rst chemotherapy cycle. This means that the initial chemotherapy was tolerated, but as treatment was continued, the level of toxicity became too high in some patients. Notably, among the patients who died during treatment, one patient who died had resectable disease, was 65 years old, and had an ECOG score of 2; thus, this patient should have had a more favourable prognosis and should have been treated actively. Another patient who died from AEs had metastatic disease, was older than 70 years and had an ECOG score of 2. Similarly, 5 other deaths occurred in patients with locally advanced disease, who were over 70 years old and, with the exception of 1 patient with an ECOG score of 1, had ECOG scores of 2. Overall, in our study population, 106 patients (72%) had an ECOG score of 2 and a median age of 70.7 years, and 50% of the study patients were older than 70 years. Therefore, if active treatment had been denied to patients older than 70 years or who had ECOG scores of 2 or more, then most of our patients would not have had the chance to obtain the clear bene t they received from active treatment with chemotherapy. However, we think that performing more active follow-up during treatment and allowing temporary or permanent discontinuation of therapy might affect the occurrence in side effects and decrease the number of treatment-related deaths. In addition, based on our data, those patients who can tolerate it should continue therapy.
These ndings strongly suggest the need for more tolerable therapies and more research to further establish patient selection criteria for every treatment regimen; treatments are especially needed for patients with ECOG scores of 2 and those with metastatic disease who are in good condition. Further research is also needed on protocols and patient selection criteria for the continuation of treatment beyond the 1L. This coupled with de nitive protocols based on evidence regarding when and which patients should receive palliative care instead of continuing with active therapy could also save resources. Furthermore, this study clari es the need for RWD to supplement RCTs. Analysis of RWD could reduce severe side effects and deaths caused by overtreatment, providing a reference for individual physicians when determining the best treatment for patients with PC.
In the last few years, the e cacy of 1L FOLFIRINOX or gem + nab-P vs gem monotherapy for metastatic PAC has been widely investigated.
provided another prognostic model to predict which patients would bene t the most from 2L chemotherapy. In their study, poor Karnofsky performance status (less than 70%) and a neutrophil-tolymphocyte ratio > 5 (NLR) were independent negative prognostic factors for a bene t of 2L therapy. Tsang et al. (25) performed a multivariate analysis and found that a better ECOG score, locally advanced disease and 1L treatment with FOLFIRINOX were the best prognostic factors for receiving a bene t from 2L therapy. All of these proposed prognostic factors need to be validated in prospective studies. However, these studies provide a preliminary list of factors clinicians should pay attention to when deciding whether to administer 2L treatment to patients.
PAC is unfortunately often diagnosed in a late stage, thus substantially narrowing the treatment options and leading to a poor outcome. There is an urgent need for new techniques and biomarkers to facilitate the diagnosis of PAC in an earlier stage and to facilitate therapeutic decision making and the determination of follow-up strategies. Liquid biopsies (LBs) are a promising non-invasive approach to the generation of personalized cancer treatment strategies. Improvement in LB techniques is urgently needed as is the standardization of LB protocols. However, these techniques are becoming widely recognized as fundamental tools in clinical practice, as highlighted in the meta-analysis conducted by Luchini et al.

Declarations
Ethics approval and consent to participate Permission to conduct this retrospective study and access patient records was obtained as needed by the chief physician and the Vaasa Central Hospital medical research ethics committee, thereby waiving the need for individual patient informed consent to be obtained.

Consent for publication
Not applicable.
Availability of data and materials The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
The study was supported by Tampere EVO grant. The funding body had no role in the design, collection, analysis or interpretation of the data.
Authors' contributions AJ, the principal investigator, conceived the study and supervised the project. GS and JV collected data, TV performed statistical analyses, JV and NE interpreted the data and drafted the manuscript, AJ reviewed manuscript. All authors read and approved the nal version of the manuscript.    Survival according to treatment and stage A. Survival in the untreated and treated groups. This gure displays the overall survival (OS) between patients who received treatment of any kind versus those patients who received no treatment at all. B. Survival in groups receiving no treatment, chemotherapy secondary to surgical intervention and chemotherapy as the primary form of treatment. This gure shows the overall survival between patients who underwent surgery, received primary treatment chemotherapy and those who received no treatment at all. C. Survival in patients with metastatic vs. locally advanced pancreatic cancer. This gure shows the overall survival between patients diagnosed with metastatic disease and locally advanced disease regardless of treatment received.

Figure 3
Survival according to regimens and lines of chemotherapy A. Survival according to rst-line (1L) chemotherapy regimens in all treated patients. This gure displays the overall survival according to which regimen were given as rst-line (1L) treatment in patients who received chemotherapy as primary form of treatment. B. Survival according to the number of lines of chemotherapy. This gure displays the overall survival for patients according to how many treatment lines were given in total.

Figure 4
Comparison of overall survival with gem+nab-P vs other treatments according to stage of disease A. OS in those who received gem+nab-P in comparison with other treatments (all patients). This gure shows