| Data and study |
Public observational SUPPORT cohort: 5,735 critically ill patients. Original study · Patient summary and data sources |
| Time zero and eligibility |
The first time a patient met the SUPPORT eligibility criteria; the original study selected patients who remained alive and in hospital beyond 48 hours. |
| Treatment A |
Right heart catheterisation (RHC) within the first 24 hours, recorded as swang1: 2,184 received RHC and 3,551 did not. |
| Outcome Y |
All-cause death within 30 days of enrolment, recorded as dth30: 1,918 deaths and 3,817 survivors. Death on day 30 is counted. |
| Estimand |
The standardised difference in 30-day mortality risk between RHC and no RHC, averaged over the baseline distribution of all 5,735 patients. |
| Adjustment variables |
Chapters 1–7 mainly use the original sex variable; Chapter 8 extends the baseline to sex, age, race and 12 medical-history variables. |
| Initial fits |
g predicts receipt of RHC; Q predicts mortality from baseline and treatment. The multivariable version uses main-effects logistic regressions with intercepts. |
| Encoding |
The 15 baseline variables produce 17 design columns, including an intercept; Q adds treatment. Age is standardised using the full-cohort mean and standard deviation with divisor n; white is the reference race category. |
| Computation |
g-computation → AIPW one-step correction; TMLE updates Q along an EIF-guided direction and standardises again. Every chapter uses the same patient records. |
| Causal interpretation |
An ATE interpretation requires consistency, conditional exchangeability and positivity. Time zero, the treatment window and cohort selection jointly define the population to which the result applies. |
| Julia code |
Chapter 1 · Chapter 8 · Complete analysis |
| R code |
Chapter 1 · Chapter 8 · Patient-level TMLE |