Read on Twitter
1/n Saw three fab talks today at #RSSConf2020 on target trial emulation. A summary thread 
Target trial emulation involves specifying a target trial (i.e. ideal RCT) and emulating this as best as possible using observational data
Slide by @RuthHKeogh
Target trial emulation involves specifying a target trial (i.e. ideal RCT) and emulating this as best as possible using observational data
Slide by @RuthHKeogh
2/n @RuthHKeogh gave an illustrative example of target trial emulation to estimate the effect of lung transplant on survival in people with cystic fibrosis 
First step is to specify elements of the target trial protocol (shown in 4th pic)
First step is to specify elements of the target trial protocol (shown in 4th pic)
3/n emulation then involves:
clone each person and assign one clone to each treatment
censor clones when their data stops being consistent with their group (e.g. censor controls when they receive transplant)
use appropriate methods to account for informative censoring
clone each person and assign one clone to each treatment censor clones when their data stops being consistent with their group (e.g. censor controls when they receive transplant) use appropriate methods to account for informative censoring
4/n preliminary results indicate that lung transplant helps improve survival 
5/n @camaringe presented a second example using target trial emulation to estimate the causal effect of surgical treatment on survival in patients with lung cancer disease
In this example, the issue of immortal-time bias (time between diagnosis and treatment) was highlighted
In this example, the issue of immortal-time bias (time between diagnosis and treatment) was highlighted
6/n We can use time-updated cox models to deal with immortal time bias, or use the cloning approach in target trial estimation
In the cloning approach the "immortal time" contributes to both the control arm and the treatment arm
In the cloning approach the "immortal time" contributes to both the control arm and the treatment arm
7/n There were differences in the results obtained using different approaches, but overall some evidence of an effect of surgery for older lung cancer patients.
In the pics:
-KM (1) ignores immortal time bias
-KM (3) ignores informative censoring
In the pics:
-KM (1) ignores immortal time bias
-KM (3) ignores informative censoring
8/n @bldestavola presented on the challenges in emulating target trials.
Two main challenges (previously discussed) are immortal time bias and inappropriate selection of comparative groups.
Bianca's talk focused on two different challenges:
Eligibility
Time frame
Two main challenges (previously discussed) are immortal time bias and inappropriate selection of comparative groups.
Bianca's talk focused on two different challenges:
Eligibility Time frame
9/n Missingness in eligibility
E.g: interested in effectiveness of palivizumab in preventing hospitalisation in high risk population (premature infants 
high risk)
Problem: we're missing ~25% data on gestational age (i.e. missing information on eligibility)
Missingness in eligibilityE.g: interested in effectiveness of palivizumab in preventing hospitalisation in high risk population (premature infants high risk)Problem: we're missing ~25% data on gestational age (i.e. missing information on eligibility)
10/n Potential solutions include confounding control and multiple imputation. Each have their own strengths and limitations
11/n Challenge : Selecting a time frame
E.g. do HbAlc levels at end of follow-up differ by second line treatments?
Problem: the entry to study is staggered and so the choice of time-frame (T) influences recruitment period and affects results. How best to choose T?
: Selecting a time frameE.g. do HbAlc levels at end of follow-up differ by second line treatments?Problem: the entry to study is staggered and so the choice of time-frame (T) influences recruitment period and affects results. How best to choose T?
12/n Bianca's recommendation: acknowledge these issues & include details of emulation steps in the description of the emulated trial!
Really enjoyed these talks, thanks to the presenters & organisers @NicolaFitz & @WJHarrisonPhD
Recommended references for learning more
Really enjoyed these talks, thanks to the presenters & organisers @NicolaFitz & @WJHarrisonPhD
Recommended references for learning more
