In time-to-event settings, g-computation and doubly robust estimators are based on discrete-time data. However, many biological processes are evolving continuously over time. In this paper, we extend the g-computation and the doubly robust standardisation procedures to a continuous-time context. We compare their performance to the well-known inverse-probability-weighting estimator for the estimation of the hazard ratio and restricted mean survival times difference, using a simulation study. Under a correct model specification, all methods are unbiased, but g-computation and the doubly robust standardisation are more efficient than inverse-probability-weighting. We also analyse two real-world datasets to illustrate the practical implementation of these approaches. We have updated the R package RISCA to facilitate the use of these methods and their dissemination.
ShermanREAndersonSADal PanGJet al. Real-world evidence – what is it and what can it tell us?. N Engl J Med2016; 375: 2293–2297. DOI: 10.1056/NEJMsb1609216.
2.
HernánMA. A definition of causal effect for epidemiological research. J Epidemiol Community Health2004; 58: 265–271. DOI: 10.1136/jech.2002.006361.
3.
RosenbaumPRRubinDB. The central role of the propensity score in observational studies for causal effects. Biometrika1983; 70: 41–55. DOI: 10.1093/biomet/70.1.41.
4.
RobinsJMHernánMABrumbackB. Marginal structural models and causal inference in epidemiology. Epidemiology2000; 11: 550–560. DOI: 10.1097/00001648-200009000-00011.
5.
LuncefordJKDavidianM. Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study. Stat Med2004; 23: 2937–2960. DOI: 10.1002/sim.1903.
6.
VansteelandtSDanielRM. On regression adjustment for the propensity score. Stat Med2014; 33: 4053–4072. DOI: 10.1002/sim.6207.
7.
AliMSGroenwoldRHHBelitserSVet al. Reporting of covariate selection and balance assessment in propensity score analysis is suboptimal: a systematic review. J Clin Epidemiol2015; 68: 112–121. DOI: 10.1016/j.jclinepi.2014.08.011.
8.
Le BorgneFGiraudeauBQuerardAHet al. Comparisons of the performance of different statistical tests for time-to-event analysis with confounding factors: practical illustrations in kidney transplantation. Stat Med2016; 35: 1103–1116. DOI: 10.1002/sim.6777.
9.
HajageDTubachFStegPGet al. On the use of propensity scores in case of rare exposure. BMC Med Res Methodol2016; 16: 38. DOI: 10.1186/s12874-016-0135-1.
10.
AustinPC. The performance of different propensity score methods for estimating marginal hazard ratios. Stat Med2013; 32: 2837–2849. DOI: 10.1002/sim.5705.
11.
KingGNielsenR. Why propensity scores should not be used for matching. Polit Anal2019; 27: 434–454. DOI: 10.1017/pan.2019.11.
12.
RobinsJ. A new approach to causal inference in mortality studies with a sustained exposure period-application to control of the healthy worker survivor effect. Math Model1986; 7: 1393–1512. DOI: 10.1016/0270-0255(86)90088-6.
13.
SnowdenJMRoseSMortimerKM. Implementation of g-computation on a simulated data set: Demonstration of a causal inference technique. Am J Epidemiol2011; 173: 731–738. DOI: 10.1093/aje/kwq472.
14.
KeilAPEdwardsJKRichardsonDRet al. The parametric g-formula for time-to-event data: towards intuition with a worked example. Epidemiology2014; 25: 889–897. DOI: 10.1097/EDE.0000000000000160.
BreskinAEdmondsAColeSRet al. G-computation for policy-relevant effects of interventions on time-to-event outcomes. Int J Epidemiol2021; 49: 2021–2029. DOI: 10.1093/ije/dyaa156.
17.
SjölanderADahlqwistEZetterqvistJ. A note on the noncollapsibility of rate differences and rate ratios. Epidemiology2016; 27: 356–359. DOI: 10.1097/EDE.0000000000000433.
18.
GollobHFReichardtCS. Taking account of time lags in causal models. Child Dev1987; 58: 80–92. DOI: 10.2307/1130293.
19.
KreifNDiazOrdazK. Machine learning in policy evaluation: New tools for causal inference. Oxford Research Encyclopedia of Economics and Finance2019. DOI: 10.1093/acrefore/9780190625979.013.256.
20.
TanZ. Comment: Understanding OR, PS and DR. Stat Sci2007; 22: 560–568. DOI: 10.1214/07-STS227A.
21.
LendleSDFiremanBvan der LaanMJ. Targeted maximum likelihood estimation in safety analysis. J Clin Epidemiol2013; 66: S91–S98. DOI: 10.1016/j.jclinepi.2013.02.017.
22.
VansteelandtSKeidingN. Invited commentary: G-computation-lost in translation?. Am J Epidemiol2011; 173: 739–742. DOI: 10.1093/aje/kwq474.
23.
ChattonALe BorgneFLeyratCet al. G-computation, propensity score-based methods, and targeted maximum likelihood estimator for causal inference with different covariates sets: a comparative simulation study. Sci Rep2020; 10: 9219. DOI: 10.1038/s41598-020-65917-x.
24.
HernánMABrumbackBRobinsJM. Marginal structural models to estimate the joint causal effect of nonrandomized treatments. J Am Stat Assoc2001; 96: 440–448. DOI: 10.1198/016214501753168154.
25.
AalenOOCookRJRøyslandK. Does cox analysis of a randomized survival study yield a causal treatment effect?. Lifetime Data Anal2015; 21: 579–593. DOI: 10.1007/s10985-015-9335-y.
26.
HernánMA. The hazards of hazard ratios. Epidemiology2010; 21: 13–15. DOI: 10.1097/EDE.0b013e3181c1ea43.
27.
SchemperMWakounigSHeinzeG. The estimation of average hazard ratios by weighted cox regression. Stat Med2009; 28: 2473–2489. DOI: 10.1002/sim.3623.
28.
RoystonPParmarMK. Restricted mean survival time: an alternative to the hazard ratio for the design and analysis of randomized trials with a time-to-event outcome. BMC Med Res Methodol2013; 13: 152. DOI: 10.1186/1471-2288-13-152.
29.
PooleC. On the origin of risk relativism. Epidemiology2010; 21: 3–9. DOI: 10.1097/EDE.0b013e3181c30eba.
30.
SjölanderA. Estimation of causal effect measures with the R-package stdreg. Eur J Epidemiol2018; 33: 847–858. DOI: 10.1007/s10654-018-0375-y.
31.
XuSRossCRaebelMAet al. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health2010; 13: 273–277. DOI: 10.1111/j.1524-4733.2009.00671.x.
32.
ColeSRHernánMA. Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed2004; 75: 45–49. DOI: 10.1016/j.cmpb.2003.10.004.
33.
BreslowN. Discussion of the paper by D. R. Cox. J R Stat Soc Series B1972; 34: 216–217. DOI: 10.1111/j.2517-6161.1972.tb00900.x.
34.
LuqueFernandezMASchomakerMRachetBet al. Targeted maximum likelihood estimation for a binary treatment: A tutorial. Stat Med2018; 37: 2530–2546. DOI: 10.1002/sim.7628.
35.
HernánMAHernández-DíazSRobinsJM. Randomized trials analyzed as observational studies. Annals of Internal Medicine2013, 159, 560–2. DOI: 10.7326/0003-4819-159-8-201310150-00709.
36.
HudgensMGHalloranME. Toward causal inference with interference. J Am Stat Assoc2008; 103: 832–842. DOI: 10.1198/016214508000000292.
37.
AustinPC. Variance estimation when using inverse probability of treatment weighting (iptw) with survival analysis. Stat Med2016; 35: 5642–5655. DOI: 10.1002/sim.7084.
38.
MorrisTPWhiteIRCrowtherMJ. Using simulation studies to evaluate statistical methods. Stat Med2019; 38: 2074–2102. DOI: 10.1002/sim.8086.
39.
R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing, 2014.
40.
LaplaudDACaseyRBarbinLet al. Comparative effectiveness of teriflunomide vs dimethyl fumarate in multiple sclerosis. Neurology2019; 93: 1–12. DOI: 10.1212/WNL.0000000000007938.
41.
MassetCBoucquemontJGarandeauCet al. Induction therapy in elderly kidney transplant recipients with low immunological risk. Transplantation2020; 104: 613–622. DOI: 10.1097/TP.0000000000002804.
42.
VanderWeeleTJShpitserI. A new criterion for confounder selection. Biometrics2011; 67: 1406–1413. DOI: 10.1111/j.1541-0420.2011.01619.x.
43.
MartinussenTVansteelandtSAndersenPK. Subtleties in the interpretation of hazard contrasts. Lifetime Data Anal2020; 26: 833–855. DOI: 10.1007/s10985-020-09501-5.
44.
PetersenMLPorterKEGruberSet al. Diagnosing and responding to violations in the positivity assumption. Stat Methods Med Res2012; 21: 31–54. DOI: 10.1177/0962280210386207.
45.
ConnerSCSullivanLMBenjaminEJet al. Adjusted restricted mean survival times in observational studies. Stat Med2019; 38: 3832–3860. DOI: 10.1002/sim.8206.
46.
HajageDChauvetGBelinLet al. Closed-form variance estimator for weighted propensity score estimators with survival outcome. Biom J2018; 60: 1151–1163. DOI: 10.1002/bimj.201700330.
47.
EfronB. Estimation and accuracy after model selection. J Am Stat Assoc2014; 109: 991–1007. DOI: 10.1080/01621459.2013.823775.
48.
BlakelyTLynchJSimonsKet al. Reflection on modern methods: when worlds collide—prediction, machine learning and causal inference. Int J Epidemiol2020; 49: 2058–2064. DOI: 10.1093/ije/dyz132.
49.
Le BorgneFChattonALégerMet al. G-computation and machine learning for estimating the causal effects of binary exposure statuses on binary outcomes. Sci Rep2021; 11: 1435. DOI: 10.1038/s41598-021-81110-0.
50.
KangJDYSchaferJL. Demystifying double robustness: A comparison of alternative strategies for estimating a population mean from incomplete data. Stat Sci2007; 22: 523–539. DOI: 10.1214/07-STS227.
51.
KreifNGruberSRadiceRet al. Evaluating treatment effectiveness under model misspecification: A comparison of targeted maximum likelihood estimation with bias-corrected matching. Stat Methods Med Res2016; 25: 2315–2336. DOI: 10.1177/0962280214521341.
52.
DíazIColantuoniEHanleyDet al. Improved precision in the analysis of randomized trials with survival outcomes, without assuming proportional hazards. Lifetime Data Anal2019; 25: 439–468. DOI: 10.1007/s10985-018-9428-5.
53.
BenkeserDCaroneMGilbertPB. Improved estimation of the cumulative incidence of rare outcomes. Stat Med2018; 37: 280–293. DOI: 10.1002/sim.7337.
54.
KeilAPEdwardsJK. A review of time scale fundamentals in the g-formula and insidious selection bias. Current epidemiology reports2018, 5, 205–213. DOI: 10.1007/s40471-018-0153-0.
55.
YoungJGStensrudMJTchetgen TchetgenEJet al. A causal framework for classical statistical estimands in failure time settings with competing events. Stat Med2020; 39: 1199–1236. DOI: 10.1002/sim.8471.
56.
DanielRCousensSDe StavolaBet al. Methods for dealing with time-dependent confounding. Stat Med2013; 32: 1584–1618. DOI: 10.1002/sim.5686.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
