Читать книгу Clinical Obesity in Adults and Children - Группа авторов - Страница 50

Questions about developmental origins of obesity by necessity require longitudinal data on early life exposures and outcomes into childhood or beyond. The majority of data on humans come from prospective and retrospective cohort studies, which are reasonably efficient in that they allow for simultaneous collection of information on many characteristics that can be considered as exposures, outcomes, or covariates in multiple analyses.

One example of a prospective cohort study is the Project Viva study, which one of us (EO) leads [21]. Project Viva recruited women in early pregnancy from eastern Massachusetts, United States, 1999–2002. The cohort includes 2128 singleton births who have been followed via in‐person research visits during pregnancy, after birth, and during infancy, early childhood, mid‐childhood, early adolescence, and mid/late adolescence. Additional data collection has occurred from mailed questionnaires, medical records, insurance claims, analysis of biospecimens, and output from personal monitors, such as spirometry for lung function, accelerometry for physical activity, and actigraphy for sleep. This cohort has served as the substrate for a number of analyses examining questions related to this topic, and we will use examples from several of these papers throughout this chapter.

The major concern related to analyses of observational data from cohort studies is that residual confounding from poorly measured or unmeasured variables may underlie observed associations [22]. Additionally, loss to follow‐up can result in bias. Well‐designed randomized controlled trials are the gold standard study design for addressing causal questions and eliminating confounding effects. Trials have been conducted to address some exposures, for example, treatment of gestational diabetes [23], lifestyle interventions to reduce gestational weight gain [24], and provision of nutrient supplements [25]. However, the utility of evidence from trials is limited because they are relatively expensive and thus rarely performed; only a single or small number of exposures can be examined per study; it is often infeasible or at least highly impractical to begin interventions before prenatal care begins, so typically they start at the end of the first trimester; and potentially harmful exposures (such as smoking) cannot be experimentally assigned.

Researchers have increasingly applied creative analytic approaches to observational data to bring us closer to causal interpretations of these analyses. One approach is to compare outcomes following discordant exposures among siblings [16], who presumably share similar genetic and sociodemographic characteristics; thus, it is more likely that any differences in obesity relate to the difference in exposure [16,26]. Examples include studies of siblings with discordant exposures to intrauterine diabetes or breastfeeding [16,26]. Alternatively, investigators have examined the same question in two or more cohorts with different social patterning of exposure [27]. This approach has been applied in studies of maternal diet and breastfeeding, among others [27,28]. If the exposure‐outcome relationship is similar in both settings, it is not likely explained by the social factors that predict exposure. Further, modern sophisticated statistical approaches have been described that minimize bias from loss to follow‐up and can support more confident causal inferences from observational data [29].

Quasi‐experimental study designs take advantage of “natural experiments” and thus minimize confounding. Mendelian randomization analysis is one type of quasi‐experimental design. This approach uses genetic variants to determine whether an observational association between a risk factor and an outcome is consistent with a causal effect [30]. Mendelian randomization relies on the natural, random assortment of genetic variants during meiosis yielding a random distribution of genetic variants in a population [30]. Examples of genetic variants studied in developmental origins of obesity research include genetic risk scores for gestational diabetes or birth size [31,32]. In the case of Mendelian randomization studies, it is Mother Nature herself who performed the “experiment.” Other quasi‐experimental studies leverage anthropogenic variation in exposure, for example state‐specific policies to limit smoking or second‐hand smoke exposure [33]. In both cases, exposure status should not be linked to other important predictors of outcome, and thus confounding is minimized. Finally, while there are many developmental and physiologic differences between animal models and humans, animal trials provide important confirmation that observed associations in human cohorts are likely to be causal and can more deeply investigate potential mechanisms.

In the sections that follow, we will discuss evidence for risk factors in the prenatal period and during infancy that are associated with later obesity within the context of these analytic considerations.