Core concept: differential information bias in a case–control outbreak investigation
This scenario describes a classic
case–control study embedded in a foodborne outbreak investigation. When the list of all exposed persons is unknown, investigators compare ill residents (cases) with well residents (controls) and ask about food exposures. The goal is to identify which food item is associated with illness by calculating an
odds ratio for each exposure.
The key flaw here is that
the two groups were not questioned in the same way. Cases were interviewed face to face in the hospital by a nurse who knew they were ill, and each food item was asked about up to three times. Controls were questioned once by telephone by a different staff member. This unequal intensity of questioning creates
interviewer bias, which is a form of
differential information bias.
Differential information bias occurs when the accuracy or completeness of exposure reporting differs between cases and controls. Because cases were probed repeatedly and in person, they were more likely to recall and report specific foods than controls who received a single telephone question. This systematically inflates the reported exposure prevalence among cases relative to controls, which in turn
inflates the odds ratio away from the null value of 1.
The distinction from
recall bias is important. Recall bias is a specific type of differential information bias in which cases remember exposures better because they are motivated to find a cause for their illness. However, the pretest showed that when cases and controls were questioned the same way, both groups reported foods equally completely. That finding rules out a true difference in memory or willingness to report. The problem is not the participants’ memory; it is the
unequal data collection method applied by the interviewers.
Selection bias is also ruled out. Controls were chosen at random from barangay household lists and all agreed to participate. Random selection with full participation minimizes the chance that controls differ systematically from the source population that produced the cases. Therefore, the observed inflation of the odds ratio cannot be attributed to how controls were selected.
Nondifferential misclassification would occur if both cases and controls were equally likely to have their exposure status misclassified. In that situation, the odds ratio would be biased
toward the null, meaning it would be pulled closer to 1 rather than inflated. Since the problem states the odds ratio was inflated, nondifferential misclassification is not the correct explanation.
| Type of bias | Mechanism | Effect on odds ratio | Why ruled out or supported here |
|---|
| Interviewer bias | Unequal probing or questioning intensity between groups | Inflates odds ratio away from 1 | Supported — cases probed up to 3 times in person; controls asked once by phone |
| Recall bias | Cases remember exposures better due to illness motivation | Inflates odds ratio | Ruled out — pretest showed equal reporting when questioned identically |
| Selection bias | Controls not representative of source population | Distorts odds ratio in either direction | Ruled out — random selection from household lists with full participation |
| Nondifferential misclassification | Both groups equally misclassified on exposure | Biases odds ratio toward 1 | Ruled out — would not inflate the odds ratio |
The broader principle from the evidence base is that
information bias arises when the method of measuring exposure differs between comparison groups. In case–control studies, any systematic difference in how data are collected from cases versus controls can produce a spurious association. The educational primer on bias in clinical research emphasizes that information bias is a systematic error creating a difference between observed and true values, and that it must be distinguished from selection bias and confounding. The birth defects case–control study similarly highlights that differential participation or data collection can distort effect estimates, underscoring the need to evaluate whether observed associations reflect true relationships or artifacts of study methods.
Key point! When cases and controls are interviewed differently, the resulting bias is classified by the
source of the unequal reporting. If the difference comes from the participants’ memory or motivation, it is recall bias. If it comes from the interviewer’s technique or knowledge of case status, it is interviewer bias. Here, the pretest eliminated recall bias, leaving the unequal probing as the direct cause of the inflated odds ratio.