In behavioral intervention trials, members of the control group may have access to the intervention, which can potentially affect the outcomes of interest. If the members of the comparison (control) group are actually affected by the intervention, then contamination or “spillover” is said to occur. Referred to as contagion in the case of experimental evaluation, contamination can pose a major threat to the investigator’s ability to draw unambiguous conclusions about the effectiveness of the intervention under study. For example, suppose an investigator is trying to conduct a behavioral trial of an intervention to improve mood and reduce loneliness and isolation in a congregate living facility for older adults. He or she randomly assigns half of the residents in this community to the intervention group and half to the control condition. However, in this case, the potential of contamination effects from the intervention group to the control is very high because the residents of this community may talk to one another about the study and may even share intervention materials if they are available, given their close proximity and shared living space.
So, how should the investigator try to control for, or minimize, contamination? One possibility is to explicitly ask participants not to talk with one another about the study until the intervention trial is over. Although this may reduce the extent or severity of spillover, it will not entirely eliminate it, given that some residents may still share details of their experience with the interventions with those who are not in the treatment group. Also, this request may be perceived as burdensome to some study participants. These exchanges between participants can be difficult or impossible to document. Another related option is to try to limit access to the intervention materials, and ask intervention participants not to share materials with anyone. For example, in a memory-training study we conducted in a retirement community, all handouts were collected at the end of each training class, and participants were not allowed to take the intervention manuals home with them. However, there were still instances where participants asked to keep the handouts, or even tried to smuggle them out of the classroom; so the potential for sharing materials with others still may exist despite the investigator’s best efforts.
A stronger approach to dealing with contamination involves the deliberate design of an intervention trial to minimize the possibility of spillover. For example, in the Baltimore Prevention trial with first and second graders, we used both
internal and external control groups to derive a statistical estimate of how much spillover occurred from the intervention conditions to the control groups within 19 public elementary schools in Baltimore City (Dolan et al., 1993; Kellam, Rebok, Ialongo, & Mayer, 1994). Six of the 19 schools were randomly assigned to a classroom behavior intervention, 7 to a reading intervention, and the remaining 6 were assigned as no-treatment or external control schools. Because the latter schools had neither intervention, it was considered unlikely that contamination would have occurred through exposure to the intervention. However, we also employed internal controls within the 13 schools that had either the behavioral or reading interventions. By comparing the treatment effect size of the interventions under the two different types of control conditions, one can get an estimate of just how large the spillover effect might be. If no spillover effects had occurred, then one would expect there to be no difference in the treatment effect size under either comparison condition. However, if spillover occurred, then the treatment effects should be relatively larger in the external control schools, because the treatment effects may have been attenuated in the internal control schools owing to contamination of the control classrooms. However, in this digital age of electronic information sharing, even geographic distance of intervention sites does not guarantee control conditions, and contamination may still occur (Sanson-Fisher, Bonevski, Green, & D’Este, 2007). And these effects are exacerbated when the intervention is implemented at the community level such as in a school system or in community health care facilities.