Search

Background Avoiding “causal” language with observational study designs is common publication practice, often justified as being a more cautious approach to interpretation. Objectives We aimed to i) estimate the degree to which causality was implied by both the language linking exposures to outcomes and by action recommendations in the high-profile health literature, ii) examine disconnects between language and recommendations, iii) identify which linking phrases were most common, and iv) generate estimates by which these phrases imply causality. Methods We identified 18 of the most prominent general medical/public health/epidemiology journals, and searched and screened for articles published from 2010 to 2019 that investigated exposure/outcome pairs until we reached 65 non-RCT articles per journal (n=1,170). Two independent reviewers and an arbitrating reviewer rated the degree to which they believed causality had been implied by the language in abstracts based on written guidance. Reviewers then rated causal implications of linking words in isolation. For comparison, additional review was performed for full texts and for a secondary sample of RCTs. Results Reviewers rated the causal implication of the sentence and phrase linking the exposure and outcome as None (i.e., makes no causal implication) in 13.8%, Weak in 34.2%, Moderate in 33.2%, and Strong in 18.7% of abstracts. Reviewers identified an action recommendation in 34.2% of abstracts. Of these action recommendations, reviewers rated the causal implications as None in 5.3%, Weak in 19.0%, Moderate in 42.8% and Strong in 33.0% of cases. The implied causality of action recommendations was often higher than the implied causality of linking sentences (44.5%) or commensurate (40.3%), with 15.3% being weaker. The most common linking word root identified in abstracts was “associate” (n=535/1,170; 45.7%) (e.g. “association,” “associated,” etc). There were only 16 (1.4%) abstracts using “cause” in the linking or modifying phrases. Reviewer ratings for causal implications of word roots were highly heterogeneous, including those commonly considered non-causal. Discussion We found substantial disconnects between causal implications used to link an exposure to an outcome and the action implications made. This undercuts common assumptions about what words are often considered non-causal and that policing them eliminates causal implications. We recommend that instead of policing words, editors, researchers, and communicators should increase efforts at making research questions, as well as the potential of studies to answer them, more transparent. Summary box

Policymakers use results from randomized controlled trials to inform decisions about whether to implement treatments in target populations. Various methods - including inverse probability weighting, outcome modeling, and Targeted Maximum Likelihood Estimation - that use baseline data available in both the trial and target population have been proposed to generalize the trial treatment effect estimate to the target population. Often the target population is significantly larger than the trial sample, which can cause estimation challenges. We conduct simulations to compare the performance of these methods in this setting. We vary the size of the target population, the proportion of the target population selected into the trial, and the complexity of the true selection and outcome models. All methods performed poorly when the trial size was only 2% of the target population size or the target population included only 1,000 units. When the target population or the proportion of units selected into the trial was larger, some methods, such as outcome modeling using Bayesian Additive Regression Trees, performed well. We caution against generalizing using these existing approaches when the target population is much larger than the trial sample and advocate future research strives to improve methods for generalizing to large target populations.