Abstract
Lopez et al. (2002) prospectively examined the association between maternal periodontitis and adverse pregnancy outcomes among poor Chilean women. Out of 881 women screened, 263 had periodontitis and 618 had gingivitis or mild periodontitis. Those in the latter (comparison) group were offered treatment to make a sharper contrast; 159 refused treatment. There were two main weaknesses of this study that cast doubts on its validity.
Bias
In this study, the comparison group was treated and the exposed (periodontitis) group was not, violating the counterfactual principle (Rothman and Greenland, 1998). Moreover, 26% (159/618) of the women in the comparison group dropped out vs. none in the exposed group, causing selection bias. Selection bias could be in either direction, depending upon the characteristics of the women who dropped out. The biases could have been avoided by not offering any intervention and comparing women with periodontitis with those without. Alternatively, women with periodontitis consenting to treatment could have been randomized to treated or untreated groups. Either comparison would have been between groups of women who differed only by exposure, and therefore unbiased.
Residual Confounding
Smoking during pregnancy adversely affects birth outcomes (CDC, 2001). It is not clear here whether lifetime smoking or smoking during pregnancy was evaluated. Including non-smokers and smokers of ≤ 5 cigarettes/day in one group increased misclassification and consequent confounding. Since more women with periodontitis smoked than those in the comparison group (21.4% vs. 15.0%, p = 0.038), smoking should have been included in every model, whether or not it was statistically significant. Residual confounding from smoking would probably overestimate the results.