Identification of sub-groups of chronic obstructive pulmonary disease patients using clustering analysis of small airway histopathology
Project Overview
During my internship, I explored how statistical methods can uncover useful patterns in complex clinical data while still accounting for uncertainty and data quality. I used clustering to investigate patterns of small-airway morphology in COPD and how these related to clinical characteristics and within-patient variation. In addition to this, to gain more experience in the field of health data science, I also reconstructed and validated a cardiac-surgery mortality prediction model, before extending the analysis to explore whether its performance differed between women and men and changed over time. The project highlighted the importance of careful validation and cautious interpretation when translating statistical patterns into clinically meaningful findings.
What were the key results of your research project?
My project produced two main sets of findings linked by a common theme: identifying useful patterns in clinical data while being careful not to over-interpret them.
For the COPD analysis, I identified three recurring patterns of small-airway morphology: epithelial-high, lower-thickness and wall-high. The wall-high group was particularly interesting, with a high proportion of patients with COPD and current smokers. Adding lung-function measures changed the grouping of some patients, especially within the lower-thickness cluster, while pack-years had much less effect. Repeated airway samples also showed both consistency and heterogeneity within patients, suggesting that a single summary can capture the dominant pattern while still missing some airway-to-airway variation.
In the cardiac prediction work, I reconstructed a published mortality model and then assessed its performance separately in women and men. Discrimination was almost identical between the groups (AUC 0.732 vs 0.731), with no clear persistent overall difference in calibration. However, women were less represented in the development data, and year-by-year performance was more variable. A female-only model was also highly unstable because there were very few outcome events available for training.
Overall, the project highlighted the importance of validation, representation, uncertainty and careful interpretation when using statistical models to draw clinically meaningful conclusions.
How do you feel you have benefitted from completing this internship and has it made you consider future career paths?
This internship has been really valuable because it gave me experience of what research actually looks like beyond a taught project. I had to work with real clinical data that was messy, incomplete and sometimes difficult to interpret, and I learnt that a large part of good analysis is not just choosing the right statistical method, but understanding the data, checking assumptions, questioning unexpected results and being careful about what the evidence can genuinely support.
I also developed much more confidence in working independently. Over the course of the project I moved from exploratory data cleaning and visualisation into clustering, sensitivity analyses, regression modelling, prediction-model validation and an exploratory fairness analysis. Discussing the work with my supervisor and responding to feedback also helped me become more comfortable applying statistical concepts I had learnt in my course to the realm of health data, something I was previously unfamiliar with.
The internship has also made me much more certain that I would like to pursue a career involving health data science and medical research. I found both the COPD work and the clinical prediction modelling especially interesting because they sit at the intersection of statistics, computing and real healthcare questions. It has made me want to explore areas such as clinical prediction, patient stratification, model validation and the responsible use of data-driven methods in healthcare in more depth.
Before this internship, research was something I was interested in conceptually, but completing the project has made it feel much more tangible and has shown me that I really enjoy the combination of analytical problem-solving, scientific curiosity and the possibility of producing work that could eventually contribute to better understanding of patients and impactful issues in health.
Download presentation slides
Internships 2026 - Urbi Roychowdhury