Introduction to Statistical Learning

TSHS Resource Review Post, by Ming Hu, PhD, Mayo Clinic.

In today’s healthcare environment, data-driven insights play a pivotal role, making statistical learning an essential part of the toolkit for medical science majors. For instructors of graduate-level courses in MPH or MD programs, Introduction to Statistical Learning (ISL) by Gareth James, Daniela Witten, Trevor Hastie, and Robert Tibshirani offers a valuable introduction to foundational methods. It focuses on prediction, classification, and pattern recognition, and has proved versatile in its application to different areas of public health.

Balancing Theory and Application

ISL stands out for its balance between theoretical concepts and practical applications. Rather than overwhelming students with dense derivations, ISL presents concepts through accessible language and visual aids. This is particularly valuable for students whose strengths lie in applying statistical methods to data analytic problems motivated by public health questions, rather than an interest in engaging with abstract mathematical proofs.

The book's structure supports course goals typically seen in machine learning, as it builds from linear regression and classification to more complex topics like neural networks and support vector machines. ISL’s practical orientation, combined with foundational chapters that introduce linear regression and logistic regression, ensures that students gain a strong base before advancing to more complex topics, and also helps them to appreciate that machine learning methods are a natural extension of traditional statistical modeling, and not a different field. This gradual build-up helps students develop competencies that are central to data science in public health, such as summarizing and analyzing health data for critical decision-making.

Hands-On Learning with R

One of ISL’s strengths is its extensive use of R programming. Every chapter includes R code snippets that demonstrate how to implement statistical learning techniques, offering students the immediate opportunity to apply what they’ve learned. This approach resonates well with project-based learning objectives and aligns with course requirements that focus on practical skill-building through homework and project assignments. It is also a great opportunity to practice the concept of the ‘flipped’ classroom using hands-on opportunities. R’s accessibility also ensures that students can focus on the methods rather than struggling with a steep software learning curve.

For instance, in a section on logistic regression, I adapted the default dataset from the book by introducing a dataset on cardiovascular health, asking students to form groups and to practice teaching each other how to predict patient outcomes based on social determinants of health. This small adaptation bridged the gap between the book’s examples and healthcare applications, and the students responded positively to seeing how logistic regression could identify high-risk patients as they learned proactively during this session.

Limitations and Supplementary Resources

While ISL provides a thorough introduction, it does not cover some advanced machine learning topics—such as graphical models or sequential data analysis—which are essential in a health-focused machine learning curriculum. Courses that cover these areas may require supplementary materials, such as The Elements of Statistical Learning (ESL) for a deeper dive. ESL’s advanced sections on techniques like boosting and support vector machines can provide interested students with a greater depth of understanding. And lately I have seen more students in health sciences who have a better understanding of the statistical and programming foundations who are using these for their research.

Another area where ISL may fall short for health-focused students is in data pre-processing, a critical step in handling healthcare datasets that often contain missing or inconsistent values. In my experience, it was beneficial to include lectures or lab sessions (if possible) on data wrangling and pre-processing, using additional R packages like tidyverse to ensure students were prepared to handle real-world data complexities.

Using ISL in a health science-oriented machine learning class would benefit from additional contextualization. Many students appreciate when the material is presented with real-life examples that resonate with their field of study. For example, in medical sciences, understanding model interpretability is crucial. Students often found ISL’s treatment of methods like Lasso regression valuable because it combines prediction with feature selection, allowing them to identify significant predictors in complex health datasets. This skill is directly applicable when interpreting outcomes from patient data analysis, where clear explanations of how models arrive at predictions can have significant implications for clinical decision-making.

Conclusion

For instructors aiming to introduce machine learning to medical science majors, Introduction to Statistical Learning is a highly accessible, well-organized resource. Its hands-on approach using R, focus on key statistical learning methods, and accessible language make it an ideal primary text for a graduate biostatistics course. While ISL is not exhaustive in covering every machine learning technique, its clear structure and practical orientation offer students a strong foundation. With supplementary examples tailored to healthcare applications, ISL can be a transformative resource, bridging the gap between theoretical statistics and practical data analysis in healthcare.