# STATISTICS 536B (2014-15, Second half of Term 2)

Statistical Theory for the Design and Analysis of Clinical Studies (1.5 Credits)

Lectures: Tuesday and Thursday, 9:30-11:00, ESB 4192, starting Feb. 24th.

Instructor: Paul Gustafson, ESB 3128, gustaf at stat dot ubc dot ca.

Prererequisites: Open to any interested graduate students in the Department of Statistics. Note that 536A (taught by Lang Wu in the first-half of the term) is not a prerequisite. (But Lang and I are coordinating to avoid excess repitition across the two courses.)

Graduate students from other departments are welcome, provided they have sufficient statistical and mathematical backgrounds. Such students should consult the instructor about suitability.

To be clear, this course is primarily aimed at training statisticians or biostatisticians, so understanding the math and computing behind the methods is a key part of the course. It isn't an appropriate course for someone who just wants to use statistical methods; it's for students seeking conceptual and theoretical understanding of methods and models. Against that, though, note that the Plan B evaluation scheme described below is offered to make the course more accessible to students from other units.

Description: While the official calendar title of this course is "Statistical theory for the design and analysis of clinical studies," a more apt title might simply be "Topics in Biostatistics." The planned topics cleave nicely into three groups. So the plan is to use our 13 lecture sessions to cover:

• Slicing and dicing exposure-disease data (4 lectures): issues of retrospective versus prospective study data, dichotomization, stratification, and matching.
• Meta-analysis (3 lectures): key concepts of statistically combining results from multiple studies.
• "Modern" causal inference (6 lectures): propensity scores, instrumental variables, models based on counterfactuals, time-dependent confounding.
In each of these three units we will try to survey a few key statistical ideas that seem most exciting and relevant.

Lecture Format: For a typical lecture:

• There will be some reading assigned beforehand.
• By 6pm the day before the lecture, each student will e-mail me a very brief comment or question indicating something they found surprising or confusing or interesting in the reading.
• The lecture will start with review and emphasis of selected ideas from the reading (perhaps 30 minutes). Note that for each lecture I will post skeletal slides in advance, and we will ``flesh these out'' in the lecture.
• Then we will break into small groups (freshly randomized each time), to work on a discussion or brainstorming or problem-solving activity (perhaps 20 minutes).
• Then I'll present some further material (perhaps some R-based empirical work) pertaining to one of the topics at hand (perhaps 25 minutes).

Textbooks: Nothing to buy! I will try to keep things relatively simple by assigning readings only from titles that the UBC library has in its e-book collection. The following list may be ammended as we progress:

Evaluation: Plan A (mandatory for STAT students)
• Participation - 20%: This includes the pre-lecture e-mails, participation in the in-class activities, and general participation in the lecture discussion. You will get credit here for reasonable participation - you don't need to try to out-participate your classmates, and it would be a mistake to replace quality of participation with quantity.
• Assignment Questions - 40%: Problems will be set as we go along, likely to be collected and graded in two batches.
• Final assignment in lieu of exam - 40%: This will be due near the end of the exam period. It will involve preparing a short report on a topic that we didn't have time to cover in the lectures.

Evaluation: Plan B Non-STAT students may elect to replace the Assignment Questions with a (second) short report on a topic we didn't have time to explore fully in the lectures. This is in recognition that students from other units may not have the same level of math/stat/computing backgrounds.

Evolving schedule of readings for lectures:

• Tuesday Feb. 24: Vitingoff, Sections 5.1 through 5.3. (The earlier parts of this may well be review - so feel free to just start reading at whatever point some of the ideas become new to you.)
• Thursday Feb. 26: Steyerberg, Ch. 9 (quite a short chapter).
• Tuesday Mar. 3: Holford, section on stratified analysis (bottom of p. 63 through to middle of p. 73) - but focus on what goals are trying to be achieved, not on the formulae.
• Thursday Mar. 5: Holford, first part of Ch. 10 (pp. 253-261).
• Tuesday Mar. 10: Borenstein et. al., Ch. 1-3 (don't worry - these are *very* short and non-technical chapters).
• Thursday Mar. 12: Algra and Rothwell (2012, Lancet Oncology) (access via UBC library, we are reading this to see meta-analysis ``in action'').
• Tuesday Mar. 17: NO-CLASS (PG out-of-town)
• Thursday Mar. 19: Dias et al (2013, Medical Decision Making) (access via UBC library, just skim, looking for a flavour of what `indirect comparisons and network meta-analysis' are about).
• Tuesday, Mar. 24. No advance reading, as I'd like to introduce propensity scores without any preconceptions.
• Thursday Mar. 26. To get some flavour for propensity scores in practice, take a quick look at Mongelluzzo et. al. (2008, JAMA) (access via UBC library).
• Tuesday, Mar. 31. No advance reading, as I'd like to introduce instrumental variables without any preconceptions.
• Thursday, Apr. 2. Hernan and Robins, Ch. 1.
• Tuesday, April 7. No advance reading - the plan is to introduce the notion of "mediation," whereby an exposure can have both "direct" and "indirect" influence on the outcome.
• Thursday, April 10. No *formal* advance reading, i.e., no need to send an email. But if you get a chance to skim this short discussion piece, or at least can bring a copy to the lecture, then you may get more out of the lecture.