Ideally, the objective of a clinical trial is to determine with a high degree of certainty whether the intervention of interest is truly effective. In high-quality randomized trials, if the results indicate that the intervention is different compared with the comparator, the results are likely to be due to real treatment differences. With analytic observational studies, it is less straightforward to determine with certainty what the results really mean.
Arora et al in a multicenter cohort study were planning to investigate the association between early childhood feeding practices and dental caries development in preschool children. The investigators were interested particularly in examining potential preventive risk factors against tooth decay.
Participants were to be recruited from a birth cohort from disadvantaged communities in southwestern Sydney in Australia. Mothers would be invited to join the study soon after the birth of their child at the time of the first home visit by Child and Family Health Nurses. Data on feeding practices and dental health behaviors were planned to be collected via telephone interviews at 4, 8, and 12 months, and thereafter at 6-month intervals until the child was 5 years old. The authors planned to collect data on (1) initiation and duration of breastfeeding, (2) introduction of solid food, (3) intake of cariogenic and noncariogenic foods, (4) fluoride exposure, and (5) oral hygiene practices. The children were scheduled for dental and anthropometric examinations at 2 and 5 years of age. Apart from caries prevalence and incidence, oral health quality of life was the main study outcome.
This study is ongoing, but let us assume that it is finished, and that breastfeeding is associated with a low incidence of caries. Before we take this conclusion for granted, we must ask ourselves the following questions: (1) Is the observed association the result of systematic bias? (2) Is the observed association the result of confounding bias? (3) Is the observed association a chance finding? (4) Is the observed association real? and (5) If the observed results reflect a real association, is the relationship causal? We will consider these questions in the context of this study.
Is the observed association the result of systematic bias?
Bias can occur in all types of observational studies. As we have discussed in previous articles, the main types of systematic bias are selection and information bias.
Selection bias is distortion that occurs from processes used to select study subjects and from factors that influence study participation. It is caused when the association of exposure-outcome differs among those who participate and those who do not participate in the study. The authors of the study at hand selected exposed and unexposed participants from disadvantaged communities in southwestern Sydney by invitation. The investigators invited “mothers to join the study soon after the birth of their child at the time of the first home visit by Child and Family Health Nurses.” This is a cohort study collecting information for a relatively long period (birth to 5 years of age) in a disadvantaged community. If the response rate was low, the possibility of self-selection bias is high because mothers who refuse to participate often differ in important characteristics from participating mothers. This self-referral is considered a validity threat because the reasons for self-referral might be associated with the study outcome.
Information bias is the result of the method in which information is collected that concerns the exposure as well as the outcome of the study participants. It occurs either when participants’ recall of information on exposure and outcome is inaccurate or similarly when ascertainment for the exposure or the outcome by the investigator is inaccurate. In the study at hand, the investigators collected data on feeding practices and dental health behaviors via telephone interviews at 4, 8, and 12 months, and thereafter at 6-month intervals until the child was 5 years old. Collected information included initiation and duration of breastfeeding, introduction of solid food, intake of cariogenic and noncariogenic foods, fluoride exposure, and oral hygiene practices. Inaccurate exposure recording (misclassification: ie, placement of certain participants in the wrong category) by the interviewer might result in interviewer bias, a type of information bias. In this cohort study that uses telephone interviews, systematic bias could occur if the interviewers are aware of the exposure status of the subjects and query the breastfeeding and the not-breastfeeding participants differently about tooth decay development.
Another type of information bias in this study could be follow-up bias. Loss to follow-up arises when subjects no longer want to participate; systematic bias results when participants who are not monitored for the duration of the study differ from those who remain until the end of the follow-up. As a reminder, nondifferential misclassification of exposure (nonsystematic difference in reporting inaccuracy between exposed and unexposed subjects, independent from the appearance of disease) results in only dilution of the association, whereas differential misclassification can result in overestimation or underestimation of the association.