A Review of Time Scale Fundamentals in the g-Formula and Insidious Selection Bias

Keil, Alexander P.; Edwards, Jessie K.

doi:10.1007/s40471-018-0153-0

Cited by 8 publications

(7 citation statements)

References 86 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, we studied only logistic and Cox PH regression: other working models could be applied. Keil and Edwards 54 proposed a review of possible models for Q false( A , L false) with a time-to-event outcome. Third, we considered only a reduced number of covariates, which could explain the abovementioned equivalence between the GC and the IPW with the extreme censoring rate.…”

Section: Discussionmentioning

confidence: 99%

G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting

Chatton

Borgne

Leyrat

et al. 2021

Stat Methods Med Res

View full text Add to dashboard Cite

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.

show abstract

Section: Discussionmentioning

confidence: 99%

G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting

Chatton

Borgne

Leyrat

et al. 2021

Stat Methods Med Res

View full text Add to dashboard Cite

show abstract

“…Besides, although not relevant to our cohort since we had the all-cause mortality as the outcome and no loss to follow-up, right censoring and informative loss to follow-up could be handled by parametric g-computation and the existing R package by simulating data on participants as though they had not been censored. 66 It is worth mentioning that other methods could also handle the methodological considerations that g-computation addresses—consideration of complex time structure and reporting of adjusted survival curves— and have been applied in air pollution epidemiological research, including Inverse Probability of Treatment Weighting (IPTW). 6 Furthermore, some recent approaches such as the targeted maximum likelihood estimation can also be used to directly evaluate individualized dynamic intervention strategies of continuous exposures and provide doubly robust estimates that are less vulnerable to model misspecification with valid statistical inference when data-adaptive/machine-learning methods are incorporated.…”

Section: Discussionmentioning

confidence: 99%

Using parametric g-computation to estimate the effect of long-term exposure to air pollution on mortality risk and simulate the benefits of hypothetical policies: the Canadian Community Health Survey cohort (2005 to 2015)

Chen

Donkelaar

et al. 2023

Preprint

View full text Add to dashboard Cite

Background Numerous epidemiological studies have documented the adverse health impact of long-term exposure to fine particulate matter (PM2.5) on mortality even at relatively low levels. However, methodological challenges remain to consider potential regulatory intervention's complexity and provide actionable evidence on the predicted benefits of interventions. We propose the parametric g-computation as an alternative analytical approach to such challenges. Method We applied the parametric g-computation to estimate the cumulative risks of non-accidental death under different hypothetical intervention strategies targeting long-term exposure to PM2.5 in the Canadian Community Health Survey cohort from 2005 to 2015. On both relative and absolute scales, we explored benefits of hypothetical intervention strategies compared to the natural course that 1) set the simulated exposure value at each follow-up year to a threshold value if exposure was above the threshold (8.8 μg/m3, 7.04 μg/m3, 5 μg/m3, and 4 μ g/m3); and 2) reduce the simulated exposure value by a percentage (5% and 10%) at each follow-up year. We used the three-year average PM2.5 concentration with one-year lag at the postal code of respondents' annual mailing addresses as their long-term exposure to PM2.5. We considered baseline and time-varying confounders including demographics, behavior characteristics, income level, and neighborhood socioeconomic status. We also included the R syntax for reproducibility and replication. Results All hypothetical intervention strategies explored led to lower 11-year cumulative mortality risks than the estimated value under natural course without intervention, with the smallest reduction of 0.20 per 1000 participants (95% CI: 0.06 to 0.34) under the threshold of 8.8 μg/m3, and the largest reduction of 3.40 per 1000 participants (95% CI: -0.23 to 7.03) under the relative reduction of 10% per interval. The reductions in cumulative risk, or numbers of deaths that would have been prevented if the intervention was employed instead of maintaining status quo, increased over time but flattened towards the end of follow-up. Estimates among those ≥65 years were greater with a similar pattern. Our estimates were robust to different model specifications. Discussion We found evidence that any intervention further reducing the long-term exposure to PM2.5 would reduce the cumulative mortality risk, with greater benefits in the older population, even in a population already exposed to low levels of ambient PM2.5. The parametric g-computation used in this study provides flexibilities in simulating real world interventions, accommodates time-varying exposure and confounders, and estimates adjusted survival curves with clearer interpretation and more information than a single hazard ratio, making it a valuable analytical alternative in air pollution epidemiological research.

show abstract

“…This is the first study to investigate mechanisms explaining the effect of CBT and UCT on ASB using counterfactual outcome mediation analysis [33]. We analyzed a unique, and methodologically robust dataset which applied a 2×2 factorial design conducted in a hard-to-reach population, incorporating a high number of variables and relevant information to control the analysis for confounders, as well as very few missing data (5.6%).…”

Section: Discussionmentioning

confidence: 99%

Combining cash transfers and cognitive behavioral therapy to reduce antisocial behavior in young men: A mediation analysis of a randomized controlled trial in Liberia

Carpena

Paula

Mola

et al. 2022

Preprint

View full text Add to dashboard Cite

Background Interventions that combine cognitive behavioral therapy (CBT) with unconditional cash transfers (UCT) reduce the risk of antisocial behavior (ASB), but the underlying mechanisms are unclear. In this paper, we test the role of psychological and cognitive mechanisms in explaining this effect. We assessed the mediating role of executive function, self-control, and time preferences. Methods We used data from the Sustainable Transformation of Youth in Liberia, a community-based randomized controlled trial of criminally engaged men. The men were randomized into: Group-1: control (n=237); and Group-2: CBT+UCT (n=207). ASB was measured 12-13 months after the interventions were completed, and the following mediators were assessed 2-5 weeks later: (i) self-control, (ii) time preferences and (iii) executive functions. We estimated the natural direct effect (NDE) and the natural indirect effect (NIE) of the intervention over ASB. Results Self-control, time preferences and a weighted index of all three mediators were associated with ASB scores, but the intervention had an effect on time preferences only [B= 0.09 95%CI (0.03; 0.15)]. There was no evidence that the effect of the intervention on ASB was mediated by self-control [B NIE = 0.007 95%CI (-0.01; 0.02)], time preferences [B NIE = -0.02 95%CI (-0.05; 0.01)], executive functions [B NIE = 0.002 95%CI (-0.002; 0.006)] or the weighted index of the mediators [B NIE = -0.0005 95%CI (-0.03; 0.02)]. Conclusions UCT and CBT lead to improvements in ASB, even in the absence of mediation via psychological and cognitive functions. Findings suggest that the causal mechanisms may involve non-psychological pathways.

show abstract

A Review of Time Scale Fundamentals in the g-Formula and Insidious Selection Bias

Cited by 8 publications

References 86 publications

G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting

G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting

Using parametric g-computation to estimate the effect of long-term exposure to air pollution on mortality risk and simulate the benefits of hypothetical policies: the Canadian Community Health Survey cohort (2005 to 2015)

Combining cash transfers and cognitive behavioral therapy to reduce antisocial behavior in young men: A mediation analysis of a randomized controlled trial in Liberia

Contact Info

Product

Resources

About