Using longitudinal targeted maximum likelihood estimation in complex settings with dynamic interventions

Abstract

Longitudinal targeted maximum likelihood estimation (LTMLE) has very rarely been used to estimate dynamic treatment effects in the context of time-dependent confounding affected by prior treatment when faced with long follow-up times, multiple time-varying confounders, and complex associational relationships simultaneously. Reasons for this include the potential computational burden, technical challenges, restricted modeling options for long follow-up times, and limited practical guidance in the literature. However, LTMLE has desirable asymptotic properties, ie, it is doubly robust, and can yield valid inference when used in conjunction with machine learning. It also has the advantage of easy-to-calculate analytic standard errors in contrast to the g-formula, which requires bootstrapping. We use a topical and sophisticated question from HIV treatment research to show that LTMLE can be used successfully in complex realistic settings, and we compare results to competing estimators. Our example illustrates the following practical challenges common to many epidemiological studies: (1) long follow-up time (30 months); (2) gradually declining sample size; (3) limited support for some intervention rules of interest; (4) a high-dimensional set of potential adjustment variables, increasing both the need and the challenge of integrating appropriate machine learning methods; and (5) consideration of collider bias. Our analyses, as well as simulations, shed new light on the application of LTMLE in complex and realistic settings: We show that (1) LTMLE can yield stable and good estimates, even when confronted with small samples and limited modeling options; (2) machine learning utilized with a small set of simple learners (if more complex ones cannot be fitted) can outperform a single, complex model, which is tailored to incorporate prior clinical knowledge; and (3) performance can vary considerably depending on interventions and their support in the data, and therefore critical quality checks should accompany every LTMLE analysis. We provide guidance for the practical application of LTMLE.

The Impact of Delayed Switch to Second-Line Antiretroviral Therapy on Mortality, Depending on Definition of Failure Time and CD4 Count at Failure

“…Our results comparing immediate switch to no switch yield similar conclusions to other studies which used other definitions of failure, which were done in different patient populations, for different follow-up times, and used different methodological approaches [5][6][7]17 . Like Rohr et al 7 we show the that the effectiveness of switching strategies depends on disease severity, though in a more refined way given that we modelled the relationship non-linearly for different patient groups.…”

“…As the number of HIV-positive patients with access to ART has increased, so has the number of patients that have experienced failure of first-line ART. Patients with virological failure on firstline ART should, in principle, switch to second-line therapy as soon as possible, as a delay in switching treatment regimens has been shown to lead to increased mortality [2][3][4][5][6][7] . South African guidelines recommend switching from two nucleoside reverse transcriptase inhibitors (NRTIs) and one nonnucleoside reverse transcriptase inhibitor (NNRTI) to two NRTIs and one protease inhibitor (PI) if two consecutive viral loads on first line therapy are greater than 1000 copies/mL.…”

“…The effect of delayed switch to second-line therapy on mortality has been investigated in several observational studies which adjusted for measured time-varying confounders using causal inference methods. Gsponer et al 5 showed the drastic reduction in mortality for patients switching to second-line compared to no switch based on an immunological criteria of failing, as well as the benefit of switching early. Petersen et al 6 estimated the effect of delayed switch after confirmed virological failure on survival and quantified the relative benefit of earlier switch based on the assumption of a linear relationship between timing of switch and probability of death.…”

The impact of same-day antiretroviral therapy initiation under the WHO Treat-All policy

“…We facilitated this step using extensive super learning to avoid model mis-specification (see Supplementary Table 2). 41,42…”

“…We facilitated this step using extensive super learning to avoid model mis-specification (see Supplementary Table 2). 41,42 Several supplementary analyses were performed. We compared same-day ART with rapid ART initiation defined as ART initiation 1-7 days after HIV care enrolment (rather than early ART) as per WHO recommendations.…”