Sir, the study by C. M. Pine et al. (BDJ 2006; 200: 45–47) presents data from 5-year-old children from Scotland. The study population included 3,273 (46.8%) who had no caries experience and 3,691 who had caries of which 224 (6%) had all carious teeth restored. This dataset was used to present evidence that the proportion of children with sepsis increases with the number of untreated carious teeth. There is no problem using this dataset for this purpose because very few carious teeth were restored in these young children. However, the authors also conclude that this disadvantage can be mitigated if more of the caries experience is treated and write in their discussion that the reduced level of fillings and extractions in these children is a significant contributor to their oral sepsis. This conclusion is not and cannot be supported by their data.

The data presented are from a cross-sectional survey conducted in 1999/2000 in Scotland on children with a mean age of 5.3. In this population the proportion of dmft that is ft is reported to be 6% in children with sepsis and 10% in children without sepsis.

The authors use this dataset to try to persuade the reader that restoring primary teeth reduces the prevalence of sepsis. However, the small number of children with fillings in the dataset and the cross-sectional study design, which the authors correctly state precludes the drawing of causal inferences, make any possible link between sepsis and restored primary teeth impossible to demonstrate. To substantiate the authors' views on restoration a very strong relationship between restoring primary teeth and a reduced prevalence of sepsis needs to be demonstrated but this cannot be done with this dataset because so few children had their teeth restored.

If the authors want to investigate the relationship between filling primary teeth and the prevalence of sepsis it would have been more sensible to use a cohort approach. Why did the authors not follow these children from 1999/2000 to 2004/2005?

There are also methodological issues of which readers should be aware. The authors ignore two obvious sources of confounding. The first is that children with filled or extracted teeth must come from families that have visited a dentist whereas children with no restorations or extractions might never have been to a dentist. It is possible that attendance at a dentist and the preventive care administered might lead to a reduction in sepsis prevalence. The second is that children with a large number of carious teeth are less likely to have all their teeth restored than those with a small number of carious teeth. This leads to a large and important bias because children with multiple carious teeth are many times more likely to have sepsis. To allow the reader to assess these potential biases the authors should now at minimum describe in relation to the total number of carious teeth the proportion of children who had ever visited a dentist and the number of teeth each child had restored. This is important because the published finding that one in 10 children with untreated decay had sepsis whereas only one in 100 children with all their teeth restored had sepsis is, I believe, misleading and an artefact of the authors inappropriately ignoring obvious confounding from the fact that children with many carious teeth are more likely to have sepsis but much less likely to have all their carious teeth restored. Young children with many carious teeth are unlikely to have been taken regularly to a dentist and if they are taken to a dentist, the dentist is much more likely to advise diet control, oral hygiene instruction and fluoride rather than attempt to restore four or more carious teeth.

The second set of methodological issues relate to the undertaking and presentation of the stepwise logistic regression model. When undertaking such an analysis it is usual to include all independent variables that might be important in explaining the observed occurrences of the dependent variable. Those independent variables which do not improve the fit of the model to the data are rejected based upon predefined limits. In this paper I believe the authors have failed to present sufficient data to enable a detailed assessment of their method or the final logistic regression model. For example, it would be helpful if the authors described which statistical package they used, the models that were tested and if they presented the statistics about the goodness of fit of these models. The authors might also want to comment and present the statistical reasons why the presence or absence of plaque was not included in the final model.

The data in this paper suggest that five years ago in Scotland too many 5-year-old children had caries and that this was leading to sepsis and presumably pain. The obvious conclusion from this data is that efforts were needed to reduce caries. Water fluoridation might have been discussed. For reasons best known to the authors this paper instead focuses on restoration and ignores prevention. While it is possible that restoring primary teeth might reduce sepsis the authors should acknowledge that this dataset cannot and does not support this view.