Complex methods for complex data: key considerations for interpretable and actionable results in exposome research

Complex multidimensional data are becoming more widely available and are drastically affecting the way epidemiological studies are designed and conducted. Novel frameworks such as the exposome-which encompasses the comprehensive and cumulative set of exposures affecting individuals throughout their lifetime and the complex mechanisms through which they act - provide a unique opportunity to transform how public health recommendations are identified at the population and individual level. This data revolution is accompanied by a growing interest in analytical approaches that can handle the complexity of these novel research questions. These include semi-parametric and non-parametric statistical and machine learning methodologies that provide compelling frameworks for analyzing large-scale databases while mitigating overfitting. Nevertheless, interpreting results from these complex methods is often challenging. While discussions on interpretability have largely focused on statistical inference, causal considerations and the practical applicability of the findings to inform the design of tangible interventions have received less attention-despite being essential components of epidemiological research. With this commentary we provide a general overview of these three levels of interpretability-statistical, causal, and actionable-and discuss available tools that can aid epidemiologists to improve results interpretability as they start utilizing more complex analytical approaches.