Letter to the Editor: About lost data in cohort studies

Clemmensen et al. (2016) address a very important and interesting topic on how hyper-theory-of-mind (HToM) correlates with psychotic experiences. Their findings correlate with previous research (Clemmensen et al. 2014). This research provides new information upon HToM, is a prospective design, has a large sample, and the instruments are validated.

The Copenhagen County Child Cohort 2000 includes 6090 children which is a representative sample of the children born in Denmark that year (Olsen et al. 2007). One important limitation of the present research is the high loss to follow-up (LFU). Only 1630 subjects were included on the analysis out of 2354 that completed the Development and Well-being Assessment. LFU is a type of selection bias; authors have found that when LFU is large, an impact on results is observed (Kristman et al. 2004). The latter is especially important when characteristics of subjects not included in the study differ from the final sample. Jeppesen et al. (2015) showed that there are significant differences among the two groups on the variables ‘education of mother’, ‘maternal age’ and other variables (p < 0.05).

Moreover, in the study by Clemmensen et al. (2016), the χ² test was used for the bivariate analysis of categorical variables and the association between ‘infancy contact problems’ and HToM does not meet the assumptions for the uncorrected χ² test; the Fisher exact test is a better option in this case (Rosner, 2015). On the other hand, showing unadjusted (as well as adjusted) odds ratios in Table 4 would help a better understanding of the association between the exposure and the outcome.