Embedding Life Course Interventions in Longitudinal Cohort Studies: Australia’s GenV Opportunity

What happens when the worlds of trials and cohorts collide? Some see a fundamental immiscibility: oil and water. Others see a coming of age, a new ability to “stop describing and start fixing”¹  that could transform the influence of life-course research.

Trials test interventions in an experimental paradigm, whereas cohorts by tradition are fundamentally observational. Trials can provide strong evidence of what we should or should not do; it has been said that “a day without randomization is a day without progress.”²  Trials address what causes outcomes, what interventions improve them, by how much in real people, and where to put (or not put) time and money. Seeking obvious effects rather than nuance, they have typically not reached into the domains of cohorts: prevalence, precedents, patterns, pathways, prediction; diversity, disparity, and divergence over time, whether secular (across successive individuals) or life course (within individuals); and the myriad social, economic and environmental determinants that have been poorly amenable to trials yet determine the bulk of our outcomes. Like Tennyson’s Lady of Shalott, cohorts have tended to describe, through a backward mirror, a world that they were powerless to change.

In our view, trials and cohorts must come together (Table 1) if we are to speed up progress in solving today’s complex issues. This requires designing the next generation of life-course cohorts to be capable of testing interventions. Some are already taking up this challenge. Trailblazing examples are 2 small cohorts each recruited antenatally through a single birthing service: the Born in Bradford Better Start (BiBBS)³  and ORIGINS⁴  cohorts in Northern England and Western Australia respectively. Born in Bradford Better Start is pragmatically testing over 20 community-mounted interventions for children’s social and emotional development, whereas ORIGINS focuses on randomized trials with participation determined by eligibility criteria.

The Australian Generation Victoria (GenV) initiative is now exploring how a much larger life-course cohort can create a ‘by design’ interventional laboratory for an entire state. It offers the opportunity to study the effects of simple and complex interventions in a “whole of life course” framework, including the multiple levels of biology, physical measures, demography, generations, and place.

GenV is a prospective, whole-of-state cross-generational study in Victoria (population 6.7 million, 18% of whom are 0–14 year-old children), Australia, targeting all the expected 150 000 newborns and their parents over 2 full years from late 2021. Because it targets an entire population, it covers all service sectors, geographies, morbidities, and socioeconomic and demographic groups. Although relatively affluent, the full range of advantage-disadvantage exists; Victoria’s multiple ethnicities speak more than 250 languages, around 1% identify as Aboriginal and/or Torres Strait Islander, and over 70% of parents report that their child has at least 1 ongoing health or developmental problem at every age from 2 to 15 years. By virtue of GenV’s size and its inclusion principles, it should thus include large numbers with social or health vulnerabilities, who have the most to gain from early intervention. GenV comprises (1) consent in the days after birth, (2) a statewide −80°C Biostore of repurposed antenatal and GenV-collected postnatal biosamples, (3) plans to link to antenatal and postnatal curated administrative and unstructured service data sets, (4) GenV-collected data, and (5) a data and IT platform oriented to open science (eg, designed according to FAIR [findable, accessible, interoperable, reusable] principles). During GenV’s planning process, we have also designed (6) a novel capacity to support wide-ranging integrated research including trials (Fig 1, a conceptualization GenV’s approach to solutions for the First 1000 Days).

We could find no blueprint to guide the systematic integration of interventional research into a cohort study like GenV. We started identifying approaches by brainstorming and prioritizing technical needs with expert trialists.⁵  Our trials working group then developed a statement of intent⁶  proposing models and processes for trials to integrate within or alongside GenV, and we designed GenV’s protocol and participant information and consent forms to enable trials participation. Next, we developed an outcomes framework drawing on a systematic review of core outcome sets⁷  to ensure that GenV would prefer intervention-relevant outcomes in the domains collectively most important to families, children, services, and policymakers across health, education, and social care. At time of writing, we are building necessary IT capabilities such as the ability to select eligible participants, to notify participants of new trials, to share contact details with trials teams, and to record data-sharing agreements. The first collaborative trials will drive the order in which we develop generalizable IT capabilities; these will grow according to the needs of subsequent trials.

Over the next 2 years, we will develop a framework of priorities, processes, and services to support multiple trials over many years. We believe, but are not certain, that GenV will prove most appealing to trials that are of low intensity, require large samples, span services and policies across multiple sites and administrations, involve whole communities, have long time horizons and/or have likely variation in response because of preintervention circumstances. For population or registry-type trials wholly within GenV, GenV will provide all outcomes and health service use cost data and consent via traditional, waiver, or trials within cohorts (TWiCs) models. Trials alongside GenV will obtain consent from their own participants who were born within the GenV window; GenV may help identify potential participants via consent or opt-out waivers to pass contact details to the trial for purposes of invitation. Data-sharing reduces collection and response burdens, enriches trials with outcomes, previous data, and/or access to linked data contingent on custodian agreements, and supports modeling of causal effects to the population and between-trials comparisons of costs, benefits, and utility. It will also facilitate the incorporation of cost-effectiveness analyses to inform decision-making on implementing trial findings into service or policy practice. The embedded nature of the trials in GenV, with cross-site collaboration, should assist translation readiness if new interventions are effective. Some methodological processes, such as mixed methods approaches, will be relevant to many kinds of studies integrated with GenV, whereas others will be trial-specific, such as interim data access for data and safety monitoring committees.

Trials do not necessarily mean testing new things. They could test translation of interventions shown to work at small scale into real-world opportunities, alone or stacked together to show the power of mutual benefit over the life course. GenV has the opportunity for interventional efficiency from 3 potential approaches: (1) analytic techniques that use observational data to approximate trials,⁸  (2) trials of single interventions embedded at low impost into our existing cohort, and (3) trials of multiple “stacked” interventions either longitudinally over the life course and/or ecologically (eg child, parent, community) that are purposefully planned and implemented, for example to demonstrate dynamic complementarity.⁹  This is in addition to the opportunities to maximize causal understanding through natural experiments relating to geography, timing of natural and man-made events, and variable exposures through policy rollouts.

We anticipate challenges in many areas. One challenge is measurement, in which philosophies may vary between cohorts and trials. Life course cohorts tend to preference deep measurement of exposures and confounders. In trials, the intervention is the exposure, with confounders considered balanced by randomization. Therefore, trials may prefer measurement of characteristics that underlie trial selection or intervention response, and of defined primary and secondary outcomes. Such outcomes can be brief (through GenV collected or universal services) or administrative (eg, attendance at high-quality child care), but some specialized trial outcomes may be incompatible with the low-burden approach needed by a whole-population cohort. Trials additionally impose the burden of the intervention itself on participants and the burden of measuring process fidelity, especially for more complex interventions. If participants drop out of a trial and/or cohort because of burden, this could be to the detriment of both; therefore, compromises may need to be negotiated at the outset. Finally, the need of many trials for real-time monitoring of data for adverse events may not fit the “batched” approach common to cohorts, releasing each wave of data only once all participants have completed it.

Further challenges relate to the limits of our collaborative, conceptual, and technical abilities. Mounting either a single cohort or trial is hard; achieving GenV’s goal of integrating multiple trials may yet prove impossible. This goes beyond manpower, understanding of mutual benefit, and willingness to compromise and to share data. It will also require brilliance in technical knowledge and skills, tenacity, leadership, collaboration, IT and documentation systems, and a durable ability to see the bigger picture while supporting and coordinating the moving parts.

If we can achieve this, then GenV should be able to help generate faster, larger-scale evidence than previously possible. It opens doors to master protocols, stacked interventions, dynamic complementarity and causal modeling and simulation to the whole population from trial results. In the context of multiple interventions over decades, the power to disentangle what caused what and to understand interactions will be vital design considerations especially if we are to translate findings into more precise policy action.¹⁰  The larger sample size to achieve this may be unwelcome to trialists and their funders. Therefore, we need to debate, understand, communicate, and gain support for any benefits of powering to this bigger agenda.

In conclusion, we paraphrase John F. Kennedy: “Let us not seek the cohort answer or the trials answer, but the right answer...Let us accept our own responsibility for the future.”