From the moment I learned about precision medicine, I was fascinated by its potential to benefit all parties involved in treating a patient’s health issues. I was truly privileged to have the opportunity to be a data science intern at Scipher Medicine, a precision immunology biotech company headquartered in Waltham, MA. Scipher Medicine has a precision medicine product on the market in rheumatology called PrismRA. PrismRA is a blood test administered on patients with Rheumatoid Arthritis (RA) to detect how their body will respond to the most often used FDA-Approved RA Treatment, TNFi therapies. Using this precision immunology blood test ensures that patients and insurance companies do not waste time or money on an ineffective treatment for RA.
My internship was in Scipher’s Data Science Department. Fortunately, I spent all of my time in Scipher’s Waltham Headquarters, except for the company-wide remote work on Fridays. Being in person was especially meaningful in that it taught me the feeling of working in a small company’s office.
Although I learned lots of Python in Mr. Schlenker’s Computer Science class during my junior year, I had to learn a new language and certain Python packages for my internship. I spent the week before my internship learning Structured Query Language (SQL), a language used to access data in a database; Matplotlib, a Python graphing package; Seaborn, a Python data visualization graphing package; Pandas, a Python data analysis package; Pingouin, a Python statistics package; and GitHub, a code development platform.
On my first day, I met with my internship coordinator, Dr. Mark Zielinski, a senior data scientist at Scipher, and we discussed what my overarching project would revolve around. We decided that I would do a project for the data science department by analyzing changes through the clinical trials for patient-reported outcomes (PROs) related to mental health. Eventually, my goal was to figure out how mental health PROs, specifically sleep, anxiety, and depression, improved or worsened for patients based on whether their physician prescribed them an RA treatment by following or not following the recommendation of the PrismRA blood test. My findings in this project could inspire the data science department to do a full-scale analysis of a larger data set to see if the same trends occur. The results could be used for business development and marketing if the data showed that the patients of physicians who prescribe RA medication based on PrismRA’s results have a decrease in mental health PROs compared to patients whose physicians prescribe the opposite RA medication of PrismRA’s results.
In my first few days at my internship, I learned how to access their clinical data set using Amazon Web Services (AWS). In AWS, I accessed one of their data centers. Next, I used Athena, AWS’s analytics service, to access data from PrismRA’s clinical trials by making an SQL query in the platform. Once my query results were returned, I downloaded the data table as a CSV file and then brought the data table into my Jupyter Notebook, my coding environment, by using the Pandas Python package.
Once I had brought Scipher’s clinical trial data into my Jupyter Notebook, it was time to begin the first phase of my project. I did a simple, surface-level analysis of the data using Pandas and NumPy. Next, I graphed this data using Matplotlib to see overall trends in mental health PROs over the clinical trial, regardless of what medication each patient took. I used various graphs to show these changes, such as bar plots, violin plots, regression plots, and kernel density estimate (KDE) plots. This data was promising in that it showed an overall improvement in patients’ sleep, anxiety, and depression, all three of the mental health PROs.
Next, I made improvements to my previous analysis by filtering the data to include only patients who had completed all of their visits in the clinical trial and who had also rated their mental health PROs in our questionnaire during each of their visits. My graphing results produced the same overall trends after cleaning up the data.
Although I had no prior knowledge of statistics, Dr. Zielinski, along with other members of the data science department, was able to quickly teach me some basic statistical tests, such as the t-test, ANOVA, and the chi-squared test, to figure out if the average mental health PRO rating was significantly different between a patient’s first and final visit. Using the Pingouins statistics package in Python and these three statistical tests in particular, I found that the p-value between the patient’s first and final visit in Scipher’s clinical trial was significantly below 0.05, which rejects the null hypothesis of no difference in mental health PROs between a patient’s first and final visit.
After working with these statistics, I then categorized patients in the clinical trials into particular groups based on whether the medication they received was recommended or not recommended by Scipher’s PrismRA test. I used SQL and Pandas to combine a few different data sets in AWS to get a patient’s clinical trial PROs, their PrismRA’s test result, and their medication history. Using NumPy, Pandas, and Pingouin, I did my most in-depth analysis of this data and plotted graphs that suggested that patients who were prescribed an RA medication based on PrismRA’s results had their mental health PROs improve between the start and the end of the clinical trial. My graphs also suggested that the symptoms of patients whose treatment prescriptions did not follow the recommendation of PrismRA’s test either worsened or stayed the same. My statistical analysis using multiple t-tests, ANOVAs, and chi-squared tests suggested the same conclusion. Although there were limitations to this data analysis, including limited sample size and selective data filtration, this data suggested promising trends for the impact of PrismRA in that mental health PROs improved only if a patient was prescribed the treatment that PrismRA recommended.
After completing my main project, I started a smaller project that extended off a previous market share project. This previous project, which was presented at the European Alliance of Associations for Rheumatology (EULAR) Conference in June, looked at the percentages of RA treatments prescribed by physicians based on the test results of PrismRA. To add to this project, I looked at how a patient’s assessment of their pain and a physician’s assessment of their patient’s pain changed based on how they were prescribed in relation to the PrismRA test results. I used similar data analysis from my main project in this smaller study, but I was not able to make any conclusive findings. Nonetheless, seeing the positive impact of Scipher’s work on patients made both of these projects particularly meaningful.
On my final day, I presented both of my projects to members of the Data Science and Medical Affairs Department at Scipher’s Waltham Headquarters.
Throughout both of these projects, it was fascinating for me to see how connected working in data science is with business development and marketing. The insights gained from data analysis of clinical trials, particularly regarding patient health in relation to their PrismRA result and prescribed treatment, directly inform many of PrismRA’s data-driven selling points in business development and marketing.
In addition to these projects, I am also extremely thankful to have been invited to attend numerous meetings, such as weekly data science department meetings and company-wide town halls, where the CEO even surprisingly mentioned me by name a few times!
I am grateful for my experience at Scipher Medicine as I learned so much about computer science, data science, and statistics, along with how to act, communicate, and work in the office. Thank you to Dr. Mark Zielinski and Dr. Slava Akmaev for having me at Schiper Medicine and for teaching me about your company’s work. Thank you to Mr. Schlenker and Mr. Chris Ehrlich for making this opportunity possible. I am truly grateful for the opportunity to have this unique internship, and I cannot thank all of you enough.
תגובות