Notes

Health Discovery and Innovation — Course Resources

Lecture notes, slides, readings and links for the UT Austin MSDS Health Discovery and Innovation course. Files are organized per module under docs/moduleN/.

Module 1 — Survival Analysis

Resources

• Lecture Notes Part 1
• R script — Lecture Video 6 — empirical survival & Kaplan–Meier on the 6-subject class dataset (survival package)
• R script — Lecture Video 9 — smoking-study (pharmacoSmoking) survival comparisons
• R script — Lecture Video 12 — Cox proportional hazards models on the smoking study
• R script — Lecture Video 14 — Cox model residuals (martingale, deviance), complementary log-log plots and cox.zph Schoenfeld tests on pharmacoSmoking and BrainCancer
• R script — Lecture Video 16 — model comparison: likelihood ratio test, AIC, BIC
• R script — Lecture Video 20 — ROC curves, AUC and incremental value with pROC (aSAH)
• R script — Lecture Video 24 — single split, LOOCV, k-fold and generalized cross-validation
• R script — Lecture Video 26 — time-dependent sensitivity/specificity/AUC by hand (IPCW) and with timeROC; Paquid activity
• R script — Lecture Video 29 — Stanford heart transplant (jasa): landmark analysis and tmerge time-varying covariates
• Homework 1 — Answers & Reasoning
• Homework 2 — Answers & Reasoning

Notes

1. Survival analysis: functions & estimation — Lectures 1.1–2.2 · core functions ($f$, $F$, $S$, $\lambda$, $\Lambda$) and their identities; parametric models (exponential, Weibull, gamma, log-normal) and MLE with censoring; non-parametric estimation and Kaplan–Meier, with worked examples.
2. Comparing survival curves — Lecture 2.3 · covariates, two-group comparison at a fixed time point, hypothesis testing on $S_1(t^) - S_0(t^)$; the log-rank test ($2 \times 2$ tables, hypergeometric nulls, $T_{LR} \sim \chi^2_1$).
3. Comparing survival curves, continued — Lecture 3.1 · continuation: multigroup and stratified log-rank comparisons.
4. Cox model diagnostics, model selection & prediction — Lecture 4.1 onward · martingale, deviance and Schoenfeld residuals; the proportional hazards assumption and its remedies; LRT / AIC / BIC; ROC, AUC, incremental value and cross-validation; time-dependent AUC with IPCW; time-varying covariates and landmark prediction.

Module 2 — Longitudinal Data Analysis

Resources

• Lecture Notes Part 2
• R script — Lecture Video 32 — tidyverse / ggplot2 visualization of the dental growth data: wide-to-long, boxplots, ggpairs correlations, spaghetti plots; bodyfat class activity
• R script — Lecture Video 35 — linear mixed models with lmer: time, group and interaction effects, linear time, BLUPs, random intercepts and slopes
• R script — Lecture Video 37 — GEE with geeglm, compared with the mixed model; aids class activity
• R script — Lecture Video 39 — joint longitudinal–survival model with JM::jointModel on the aids data
• Homework 3 — Answers & Reasoning
• Homework 4 — Answers & Reasoning

Notes

1. Longitudinal data: notation & visualization — Lectures 7.1–7.2 · $Y_{ij}$, $\Sigma_i$ and $\mathbf{r}_i$; the four facts about within-subject correlation; the three research questions (time, group, time × group); the dental growth data and how to look at it in R.
2. Linear mixed models & GEE — Lectures 8.1–8.4 · random intercepts, variance components and the ICC, REML, BLUPs and shrinkage, random slopes; marginal vs. subject-specific models, the three-part GEE specification, why coefficients differ for non-linear links; lmer and geeglm on the dental data.
3. Joint modeling & missing data — Lectures 10.1–10.3 · why separate Cox and mixed models fall short; the three-step joint model and JM; MCAR / MAR / MNAR, complete-case vs. available-data analysis, LOCF, multiple imputation and Rubin’s combining rules.