Historically, most of the evidence regarding the effect of interventions came from  randomized controlled trials. However, there is an increase in the availability of already collected data and large-scale data (sometimes called “Big data”), and these data are often used to evaluate the effects of medical, public health and social interventions on health.

Because data have been previously collected, the use of quasi-experimental methods is necessary. Importantly, the use of large sample sizes requires the use of rigorous research approaches ideally guided by pre-established, published, theories of changes or logic models and analytical frameworks to avoid spurious statistical correlations.

In addition, research using large-scale routinely collected data should consider challenges such as representativeness (e.g., poor representation of ethnic minorities in the UK Biobank), potential large proportions of missing data, unmeasured confounding, and others. When reporting research using routinely collected data, RECORD (REporting of studies Conducted using Observational Routinely-collected Data) is used.

In research that uses observational data, the triangulation of findings from different methodological approaches is being increasingly used to strengthen causal claims. The different approaches are often quasi-experimental designs or methods that can deal with different sources of bias or that have different assumptions (e.g., absence of unmeasured confounding).

The use of triangulation suggests that if we obtained similar or same directional effect estimates using different methods, we are more certain about our causal effects.  This is especially important when assessing the effects of interventions on health, where few studies can inform policy changes.