📄 medicine

Exploring Clinical Phenotypes of Pancreatic Cancer in 2-year Survival and Treatment Patterns: A Retrospective Registry-Based Cluster Analysis

This retrospective registry-based study utilized unsupervised hierarchical clustering on 1,158 pancreatic cancer patients to identify six distinct clinical phenotypes with varying two-year event-free survival rates and treatment patterns, offering a framework to better address disease heterogeneity and guide therapeutic strategies.

Original authors: Louisa Schwarz, Christina Justenhoven

Published 2026-08-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Louisa Schwarz, Christina Justenhoven

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the human body as a bustling, complex city. Sometimes, a rogue construction crew shows up and starts building chaotic, dangerous structures that shouldn't be there. This is cancer. In most cities, the police can spot these bad builds early and stop them. But in the city of the pancreas—a small, vital organ tucked behind the stomach that helps digest food and control blood sugar—the trouble often starts silently. By the time the "construction" is noticed, it has usually spread far and wide, making it one of the most dangerous types of cancer out there.

For a long time, doctors have treated all pancreatic cancer cases somewhat the same way, mostly based on how big the tumor is or if it has spread. But just like every city block has its own unique vibe, every patient's cancer is different. Some tumors are slow and quiet; others are aggressive and fast. Some patients respond well to surgery, while others do better with medicine. The big question scientists are asking is: Can we stop treating everyone as if they are the same? Can we sort these patients into different "neighborhoods" or groups based on their specific traits, so we can give them the exact right help? This is where the idea of "phenotypes" comes in. Think of a phenotype not as a medical jargon word, but as a "profile" or a "character sheet" that describes a patient's age, the type of tumor they have, and the treatments they receive. If we can find these distinct profiles, we might finally crack the code on how to treat this tricky disease better.

This is exactly what Louisa Schwarz and Christina Justenhoven set out to do in their new study. They acted like digital detectives, using a massive computer program to sift through the medical records of 1,158 patients with pancreatic cancer. Instead of guessing which patients were similar, they let the data speak for itself using a method called "clustering." Imagine you have a giant box of mixed-up LEGO bricks of different colors, shapes, and sizes. If you shake the box and let them sort themselves out based on how they fit together, you might end up with piles of red bricks, blue bricks, and weirdly shaped ones. That's what the computer did here. It looked at everything about the patients—their age, whether they were male or female, where the tumor was, how aggressive it looked under a microscope, and what treatments they got—and grouped them into six distinct "neighborhoods" or phenotypes.

The results were fascinating and revealed a clear map of risk. The study found six very different groups, each with its own story. At the "safest" end of the map was a small group of younger patients with low-risk tumors and no spread to nearby lymph nodes. These patients had the best chance of surviving two years without the cancer coming back. On the other end of the map was a group of younger patients who, unfortunately, already had the cancer spread to distant parts of their body. This group faced the highest risk, with a chance of an event (like death or the cancer getting worse) five times higher than the safest group.

In the middle, the researchers found the most common "neighborhoods." These included older women and men with low-grade tumors that hadn't spread far. While these patients had a better outlook than the most advanced cases, they still faced double the risk of the safest group. Interestingly, the study noticed that many of these "middle-ground" patients received surgery but didn't get follow-up medicine (chemotherapy) afterward. The data suggested that for these specific groups, adding that medicine might actually lower their risk, hinting that they might be missing out on a helpful treatment.

Then there were the groups with high-grade, aggressive tumors. One group had tumors that were very advanced but still caught early enough for partial surgery. Another group, mostly older patients, underwent a massive, total removal of the pancreas. Both of these groups faced significantly higher risks, with the total-removal group facing the highest risk among those who had surgery. The study suggests that for these patients, the danger comes from the aggressive nature of the tumor itself, rather than just the type of surgery they received.

What makes this study so exciting is that it didn't just guess; it used real-world data from over a thousand people to prove that these six groups are real and distinct. The researchers showed that these groups aren't just random collections of people; they follow the rules of how the disease actually behaves. The "safe" group stayed safe, the "dangerous" group stayed dangerous, and the "middle" groups had their own specific challenges.

However, the authors are careful to point out that this is a starting point, not the final answer. They found that while the groups are clear, the data didn't always show a perfect link between a specific treatment and a better outcome for every single group, partly because the numbers in some groups were small. They also noted that their data was limited to basic information and didn't include things like family history or lifestyle factors.

So, what's the takeaway? This study suggests that pancreatic cancer isn't just one big, scary monster; it's actually a collection of six different "monsters," each with its own strengths and weaknesses. By recognizing these differences, doctors might be able to stop using a "one-size-fits-all" approach. Instead, they could look at a patient's "profile" and say, "Ah, you are in Group 3, which means you might need this specific treatment," or "You are in Group 5, and you might benefit from adding medicine after your surgery." It's a step toward a future where cancer treatment is as unique and personalized as the people it's meant to help.

Technical Summary: Exploring Clinical Phenotypes of Pancreatic Cancer in 2-year Survival and Treatment Patterns

Problem Statement
Pancreatic cancer (PC) is characterized by high lethality and significant heterogeneity. Current treatment paradigms are primarily driven by tumor severity (resectability), yet effective therapies for non-resectable tumors remain limited, and treatment responses exhibit high inter-patient variability. The authors posit that this heterogeneity may reflect distinct biological and clinical phenotypes based on observable patient, tumor, and treatment characteristics. There is a recognized need to explore these subgroups to better understand the disease spectrum and support improved personalized clinical management.

Methodology
The study employed a retrospective, registry-based design using a dataset of 1,158 adult patients diagnosed with histologically confirmed pancreatic cancer (ICD-10-GM C25) between January 1, 2016, and December 31, 2023, from the Cancer Registry Rhineland-Palatinate. Follow-up data extended to December 31, 2025.

Data Processing: The analysis focused on unsupervised phenotypic stratification using patient demographics, tumor characteristics (location, histology, TNM staging, grading), and treatment modalities (surgical resection type, systemic therapy). Patients who died within three months of diagnosis or from unrelated causes within two years were excluded.
Clustering Algorithm: The authors utilized hierarchical clustering with Ward's linkage, an agglomerative method that minimizes total within-cluster variance. To accommodate the mixed data types (numeric and categorical) and varying scales inherent in real-world registry data, the Gower distance metric was applied.
Cluster Determination: The optimal number of clusters ( $K$ ) was determined using the Dunn index.
Survival Analysis: The relationship between identified phenotypes and 2-year event-free survival (defined as death, tumor progression, recurrence, or new metastasis) was evaluated using Kaplan-Meier curves and univariate Cox proportional hazards regression. Phenotype 2 was designated as the reference group.
Treatment Association: Separate multivariate Cox regression models were applied within each phenotype to assess the association between specific treatments (systemic therapy, partial resection, pancreatectomy) and 2-year event-free survival.

Key Results
The analysis identified six distinct, homogeneous phenotypes:

Phenotype 1 (Elderly Females, Low Grade): The largest cluster ( $n=307$ ), characterized by elderly women with low-grade, non-metastatic tumors in the pancreatic head. Most underwent partial resection.
Phenotype 2 (Young, Low Grade, Node-Negative): The smallest cluster ( $n=92$ ), featuring younger patients with uniformly low-grade, node-negative tumors. Notably, this group had a high proportion of uncommon histological subtypes and received minimal systemic therapy (2.2%).
Phenotype 3 (Elderly, Pancreatectomy): ( $n=191$ ) Elderly patients with high-grade tumors, predominantly in the pancreatic head, who underwent total pancreatectomy.
Phenotype 4 (High Grade, Node-Positive): ( $n=162$ ) Patients with high-grade, node-positive tumors (mostly invasive ductal adenocarcinoma) treated almost exclusively with partial resection.
Phenotype 5 (Males, Low Grade): ( $n=234$ ) Similar to Phenotype 1 but predominantly male, with low-grade, non-metastatic tumors in the pancreatic head.
Phenotype 6 (Young, Metastatic): ( $n=172$ ) The youngest cohort, defined by a high proportion of distant metastases (78.5%) and unknown tumor/nodal status, treated primarily with systemic therapy.

Survival and Risk Associations:

Reference: Phenotype 2 demonstrated the highest 2-year event-free survival and served as the baseline.
Moderate Risk: Phenotypes 1 and 5 showed approximately double the risk of an event compared to the reference (HR 2.32 and 2.23, respectively).
High Risk: Phenotypes 3 and 4 exhibited even higher risks (HR 3.19 and 2.81, respectively), associated with high-grade tumors and lymph node involvement.
Highest Risk: Phenotype 6 had the highest hazard ratio (HR 5.06), consistent with its metastatic profile.

Treatment Patterns:
Within the multivariate analysis, systemic therapy was significantly associated with a reduced hazard of events in Phenotype 1 (HR 0.67) and showed a trend toward significance in Phenotype 5 (HR 0.72, $p=0.077$ ). No significant association was found between surgical methods (partial resection vs. pancreatectomy) and survival within the specific phenotypes, suggesting that the observed risk differences are driven more by disease stage and biology than the specific surgical approach.

Significance and Claims
The paper claims that the six identified phenotypes successfully capture existing clinical profiles and align with established treatment guidelines, ranging from early resectable disease to metastatic stages. The study asserts that these phenotypes vary significantly in 2-year event-free survival, validating them as clinically meaningful subgroups.

The authors suggest that these findings could facilitate the development of novel therapeutic strategies and promote more informed treatment decisions, particularly regarding the potential benefit of systemic therapy in resectable, low-grade cases (Phenotypes 1 and 5) where it appears underutilized. Furthermore, the distinct profile of Phenotype 2 highlights the potential need for a histopathological taxonomy that addresses uncommon subtypes separately. The authors conclude that their data-driven approach offers a robust framework for understanding PC heterogeneity, though they note limitations regarding the need for external validation and the inclusion of more granular treatment sequencing data in future work.

Technical Summary: Exploring Clinical Phenotypes of Pancreatic Cancer in 2-year Survival and Treatment Patterns

More like this