Table 4.
Causal contrast of interest and methods used to address different biases
| Index | Study | The estimand of interest | The measurement scale of the outcome(s) | The effect size measure used to quantify the causal contrast of interest | The statistical method used for analysing the primary outcome(s) | The statistical method used to adjust for baseline confounders | The statistical method used to account for time-varying confounders | The approach used to address immortal-time bias | The statistical method used to account for potential selection bias due to loss to follow-up |
|---|---|---|---|---|---|---|---|---|---|
| 1a (cohort analysis) | Dickerman et al. [24] |
ITT PP (treatment regimen) |
Time-to-event |
HR RD |
Pooled logistic regression | Outcome regression on the confounders | IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time | IPCW |
| 1b (case–control analysis) | Dickerman et al. [24] |
ITT PP (treatment regimen) |
Time-to-event | OR | Pooled logistic regression | Outcome regression on the confounders | IPTW | Cases and controls were sampled from the assembled cohort | IPCW |
| 2 | García-Albéniz et al. [25] |
ITT (point treatment) |
Time-to-event | RD | Pooled logistic regression | Outcome regression on the confounders | N.A. | Sequential trial emulations approach | Could not be determined |
| 3a (the addition of fluorouracil in stage II colorectal cancer) | Petito et al. [26] |
PP (point treatment) |
Time-to-event |
HR RD |
Pooled logistic regression |
1. Cloning approach + IPCW 2. Outcome regression on the confounders |
N.A. | Cloning approach + IPCW | Could not be determined |
| 3b (the use of erlotinib in advanced pancreatic adenocarcinoma) | Petito et al. [26] |
PP (point treatment) |
Time-to-event |
HR RD |
Pooled logistic regression |
1. Cloning approach + IPCW 2. Outcome regression on the confounders |
N.A. | Cloning approach + IPCW | Could not be determined |
| 4 | Dickerman et al. [4] |
ITT PP (treatment regimen) |
Time-to-event |
HR SD |
Pooled logistic regression | Outcome regression on the confounders | IPTW | Sequential trial emulations approach | IPCW |
| 5 | Dickerman et al. [27] |
PP (treatment regimen) |
Time-to-event |
RR RD |
PGF | PGF | PGF | Participants assigned to treatment groups at start of follow-up based on their data available at that time | PGF |
| 6a (single treatment versus no treatment) | Danaei et al. [28] |
ITT PP (treatment regimen) |
Time-to-event |
HR SD |
Pooled logistic regression | Outcome regression on the confounders | IPTW | Sequential trial emulations approach | Could not be determined |
| 6b (joint treatment versus no treatment) | Danaei et al. [28] |
ITT PP (treatment regimen) |
Time-to-event |
HR SD |
Pooled logistic regression | Outcome regression on the confounders | IPTW | Sequential trial emulations approach | Could not be determined |
| 6c (head-to-head comparison of two treatments) | Danaei et al. [28] |
ITT PP (treatment regimen) |
Time-to-event |
HR SD |
Pooled logistic regression | Outcome regression on the confounders | IPTW | Sequential trial emulations approach | Could not be determined |
| 7 | Zhang et al. [29] |
PP (treatment regimen) |
Time-to-event |
RR RD |
PGF | PGF | PGF | Participants assigned to treatment groups at start of follow-up based on their data available at that time | PGF |
| 8 | Atkinson et al. [30] |
PP (point treatment) |
Time-to-event | HR | Pooled logistic regression |
1. Cloning approach + IPCW 2. Outcome regression on the confounders |
N.A. | Cloning approach + IPCW | Could not be determined |
| 9 | Rojas‑Saunero et al. [31] |
PP (treatment regimen) |
Time-to-event |
RR RD |
PGF | PGF | PGF | Participants assigned to treatment groups at start of follow-up based on their data available at that time | PGF |
| 10 | Maringe et al. [14] |
PP (point treatment) |
Time-to-event | SD | Kaplan–Meier estimator | Cloning approach + IPCW | N.A. | Cloning approach + IPCW | CCA |
| 11 | Gilbert et al. [32] |
PP (treatment regimen) |
Time-to-event | HR | Pooled logistic regression | Outcome regression on the confounders | IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 12 | Caniglia et al. [33] |
PPa (point treatment) |
Binary | OR | Logistic regression | IPTW | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 13 | Althunian et al. [34] |
ITT PP (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Outcome regression on the confounders | Could not be determined | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 14 | Shaefi et al. [35] |
ITTa (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Outcome regression on the confounders | N.A. | Sequential trial emulations approach | Could not be determined |
| 15a (index trial emulation) | Bacic et al. [36] |
ITTa (point treatment) |
Time-to-event | HR | Cox proportional hazards model | IPTW | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 15b (high risk trial emulation) | Bacic et al. [36] |
ITTa (point treatment) |
Time-to-event | HR | Cox proportional hazards model | IPTW | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 16 | Rossides et al. [37] |
ITT (treatment regimen) |
Binary |
RR RD |
TMLE | TMLE | N.A. | Sequential trial emulations approach | TMLE |
| 17 | Xie et al. [38] |
ITT PP (treatment regimen) |
Time-to-event | HR |
1. Cox proportional hazards model (ITT) 2. Pooled logistic regression (PP) |
1. GPS (ITT) 2. IPTW (PP) |
IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 18 | Caniglia et al. [39] |
ITT PP (treatment regimen) |
Time-to-event | RD | Pooled logistic regression | Outcome regression on the confounders | IPTW | Sequential trial emulations approach | IPCW |
| 19 | Caniglia et al. [40] |
PP (treatment regimen) |
Time-to-event | SD | Pooled logistic regression |
1. Cloning approach + IPCW 2. Outcome regression on the confounders |
Cloning approach + IPCW | Cloning approach + IPCW | Could not be determined |
| 20a (historical comparison) | Caniglia et al. [41] |
Modified ITT (treatment regimen) |
Binary | RR |
1. Log-binomial regression 2. Poisson regression |
1. Adjusted for confounders at the design stage 2. Outcome regression on the confounders |
N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | IPCW |
| 20b (contemporaneous comparison) | Caniglia et al. [41] |
Modified ITT (treatment regimen) |
Binary | RR |
1. Log-binomial regression 2. Poisson regression |
1. Adjusted for confounders at the design stage 2. Outcome regression on the confounders |
N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | IPCW |
| 21 | Matthews et al. [42] |
ITTa (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | IPTW | N.A. | Sequential trial emulations approach | Could not be determined |
| 22 | Schmidt et al. [43] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model |
1. Propensity score matching 2. Outcome regression on the confounders |
N.A. | Sequential trial emulations approach | CCA |
| 23 | Al-Samkari et al. [44] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | IPTW | N.A. | Sequential trial emulations approach | Could not be determined |
| 24a (test the effect of hypoglycemia among individuals with dementia and diabetes, with respect to subsequent serious adverse events) | Mattishent et al. [45] |
PPa (point treatment) |
Time-to-event | HR | Cox proportional hazards model | Outcome regression on the confounders | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time |
1. CCA 2. MI |
| 24b (evaluate whether the effect of hypoglycemia was modified by the presence or absence of dementia) | Mattishent et al. [45] |
PPa (point treatment) |
Time-to-event | HR | Cox proportional hazards model | Outcome regression on the confounders | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time |
1. CCA 2. MI |
| 25 | Lenain et al. [46] |
ITT (point treatment) |
Time-to-event | SD | Kaplan–Meier estimator | Matching on time-dependent propensity score | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | CCA |
| 26 | Yiu et al. [47] |
ITT PP (treatment regimen) |
Binary |
RD RR |
Generalized linear model |
1. Propensity score matching 2. IPTW |
IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time |
1. CCA 2. Nonresponder imputation 3. Last observation carried forward 4. IPCW 5. MI |
| 27 | Wanis et al. [48] |
ITT (point treatment) |
Time-to-evet | SD |
1. Kaplan–Meier estimator 2. Pooled logistic regression |
Outcome regression on the confounders (pooled logistic regression) | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 28 | Lu et al. [49] |
ITT PP (treatment regimen) |
Time-to-event |
HR RD |
1. Cox proportional hazards model 2. Weighted Kaplan–Meier estimator |
IPTW | IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time | IPCW |
| 29 | Lyu et al. [50] |
PP (point treatment) |
Time-to-event |
HR RD |
Pooled logistic regression |
1. Cloning approach + IPCW 2. Outcome regression on the confounders |
N.A. | Cloning approach + IPCW | IPCW |
| 30 | Russell et al. [51] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model |
1. Propensity score matching 2. Outcome regression on the confounders |
N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 31 | Takeuchi et al. [52] |
ITT PP (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | IPTW | IPTW | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 32 | Abrahami et al. [53] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score methods (adjustment, stratification, fine stratification and matching) | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 33 | Secora et al. [54] |
ITT (treatment regimen) |
Time-to-event | HR | Time-to-event Fine and Gray regression model |
1. Outcome regression on the confounders 2. IPTW 3. Propensity score matching |
N.A. | Sequential trial emulations approach | Could not be determined |
| 34a (comparison of partly NRTI-sparing regimens) | Young et al. [55] |
ITTa (treatment regimen) |
Time-to-event | HR | Bayesian Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 34b (comparison of fully NRTI-sparing regimens) | Young et al. [55] |
ITTa (treatment regimen) |
Time-to-event | HR | Bayesian Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 35 | Czaja et al. [56] |
ITTa (treatment regimen) |
Time-to-event | OR | Pooled logistic regression | IPTW | N.A. | Sequential trial emulations approach | Could not be determined |
| 36 | Keyhani et al. [57] |
PPa (point treatment) |
Time-to-event | RD | Kaplan–Meier estimator | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37a (LEADER) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37b (DECLARE) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37c (EMPA-REG) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37d (CANVAS) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37e (CARMELINA) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37f (TECOS) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37 g (SAVOR- TIMI) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37 h (CAROLINA) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37i (TRITON- TIMI) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 37j (PLATO) | Franklin et al. [58] |
ITT (treatment regimen) |
Time-to-event | HR | Cox proportional hazards model | Propensity score matching | N.A. | Participants assigned to treatment groups at start of follow-up based on their data available at that time | Could not be determined |
| 38 | Fu et al. [59] |
PP (treatment regimen) |
Time-to-event | RD | Pooled logistic regression | Cloning approach + IPCW | Cloning approach + IPCW | Cloning approach + IPCW | Could not be determined |
Abbreviations: ITT Intention-to-treat effect, PP Per-protocol effect, HR Hazard ratio, RD Risk difference, IPTW Inverse probability of treatment weighting, IPWC Inverse probability of censoring weighting, OR Odds ratio, SD Survival difference, RR Risk ratio, PGF Parametric g-formula, CCA Complete case analysis, TMLE Targeted maximum likelihood estimation, GPS Generalised propensity scores, MI Multiple imputation
The symbol ‘a’ indicates that the information is not explicitly stated and was assumed given the methodological details provided