Inference for Causal Interactions for Continuous Exposures under Dichotomization

Abstract

Summary Dichotomization of continuous exposure variables is a common practice in medical and epidemiologic research. The practice has been cautioned against on the grounds of efficiency and bias. Here we consider the consequences of dichotomization of a continuous covariate for the study of interactions. We show that when a continuous exposure has been dichotomized certain inferences concerning causal interactions can be drawn with regard to the original continuous exposure scale. Within the context of interaction analyses dichotomization and the use of the results in this paper can furthermore help prevent incorrect conclusions about the presence of interactions that result simply from erroneous modeling of the exposure variables. By considering different dichotomization points one can gain considerable insight concerning the presence of causal interaction between exposures at different levels. The results in this paper are applied to a study of the interactive effects between smoking and arsensic exposure from well water in producing skin lesions.

The joint effect of asbestos exposure, tobacco smoking and alcohol drinking on laryngeal cancer risk: evidence from the French population-based case–control study, ICARE

“…Although this group includes drinkers, no excess risk was found among men consuming 1–5 drinks per day when compared with never drinkers. We nevertheless conducted sensitivity analyses using different dichotomisation points, which led to similar conclusions; our findings are, therefore, not likely to be due to erroneous modelling of the exposure variables 32. In addition, although misclassification cannot be totally ruled out, as previously discussed, it has been shown that when investigating additive interaction, the probability of misclassification should be substantial (at least larger than ¼) to bias the results 33.…”

Dietary folate intake and modification of the association of urinary bisphenol A concentrations with in vitro fertilization outcomes among women from a fertility clinic

“…Second, measurement error and misclassification of both spot urine BPA concentrations and self-reported folate intake are possible, and this would likely attenuate associations [71]. However, tests for interaction, using the misclassified exposures, are valid provided the probability of misclassification satisfies certain bounds [72]. Measured exposures have correlation close to 0 and misclassification of them is likely independent, therefore, in similar scenarios [72], evidence for significant interaction remains.…”

Additive Interaction Between Continuous Risk Factors Using Logistic Regression

“…1,2 Certain publications, focusing on dichotomous risk factors, have explored additive interaction through rothman's indexes. [2][3][4][5][6][7] Continuous risk factors have been investigated for specific cases only 8 (eAppendix 1; http://links.lww.com/eDe/ A771). We consider here the setting of continuous risk factors X, Y for disease D (0 = absence, 1 = presence), such as the probability that D = 1 increases with higher values of X, Y.…”

“…In these data, the outcome was rare (<10%), and so the above indexes are applicable. 5 the association of hypertension with BMI and nonadherence to a Mediterranean diet, estimated through logistic regression, with and without their interaction term, is shown in the table and etable 2 (http://links.lww.com/eDe/ A771), respectively. In etable 2, both factors are shown to be positively associated with prevalence of hypertension (Or BMI = 1.14 and Or NMD = 1.18, respectively).…”