T02 — Causal Infer­ence in Biosta­tis­tics and Epidemi­ol­o­gy: DAGs, g‑Methods and Target Trial Emula­tion – A Tuto­r­i­al for Researchers and Educators

Day, Time, Duration

Sunday morn­ing, 9 am‑1 pm, 4h

Language offered

Englisch

Content Summa­ry

One of the most impor­tant tasks of deci­sion makers is to derive causal inter­pre­ta­tions from statis­ti­cal and epidemi­o­log­i­cal analy­ses of obser­va­tion­al and real-world datasets. Often an inter­ven­tion, action or risk factor is inter­pret­ed to have a “causal effect” on one or more outcomes (e.g., prob­a­bil­i­ty, rate, or mean of endpoint). There­fore, both the biosta­tis­ti­cian and the epidemi­ol­o­gist need tools to under­stand: (1) when effect esti­mates have a causal inter­pre­ta­tion and when they do not (using direct­ed acyclic graphs, DAGs); and (2) the causal frame­work for the hypo­thet­i­cal target exper­i­ment (e.g., causal research ques­tion, target trial emula­tion design), (3) the appro­pri­ate meth­ods to derive causal effects instead of mere­ly statis­ti­cal asso­ci­a­tions (e.g., tradi­tion­al multi­vari­ate regres­sion analy­sis or propen­si­ty score for time-independent confound­ing and causal g‑methods for time-dependent confound­ing). Over the last years, it has become evident, that bias in obser­va­tion­al stud­ies is often not main­ly caused by uncon­trolled confound­ing but by self-inflicted bias­es relat­ed to incor­rect analyt­ic designs lead­ing to incor­rect exposure/treatment assign­ment and time-related bias­es (e.g., immor­tal time bias). This tuto­r­i­al intends to provide basic knowl­edge on causal think­ing and visu­al, struc­tur­al, and statis­ti­cal tools to guide valid causal analy­sis and to know which esti­mates are suit­able for causal inter­pre­ta­tion. The tuto­r­i­al covers inno­v­a­tive causal designs as well as causal infer­ence concepts and meth­ods that are need­ed for the design and analy­sis of obser­va­tion­al data and prag­mat­ic trials with time-varying expo­sures or treat­ments. The tuto­r­i­al is struc­tured into the follow­ing sections: 

1. Intro­duc­tion to the prin­ci­ples of causa­tion in biosta­tis­tics and epidemi­ol­o­gy
2. Use of causal diagrams (direct­ed acyclic graphs, DAGs)
3. Demon­stra­tion of the target trial exper­i­ment and the target trial emula­tion approach
4. Brief intu­itive illus­tra­tion of the prin­ci­ples of g‑methods: a) g‑formula, b) margin­al struc­tur­al models with inverse prob­a­bil­i­ty of treat­ment weight­ing, and c) struc­tur­al nest­ed models with g‑estimation
5. Intro­duc­tion of the target exper­i­ment prin­ci­ple and appli­ca­tion of the target trial emula­tion approach combined with a coun­ter­fac­tu­al approach using the cloning-censoring-weighting (CCW) approach for dynam­ic treat­ment strate­gies
6. Appli­ca­tion of g‑methods in obser­va­tion­al stud­ies and prag­mat­ic trials with post-randomization confound­ing and selec­tion bias (treat­ment switching/non-adherence/2nd-line-treatment etc.)

Orga­niz­er

Felic­i­tas Kühne

Insti­tu­tion

UMIT TIROL — Univer­si­ty for Health Sciences and Technology

Contact

felicitas.kuehne [at] umit-tirol.at

Addi­tion­al Speaker

Uwe Siebert

Insti­tu­tion

UMIT TIROL — Univer­si­ty for Health Sciences and Technology