To our knowledge, this is the first study to empirically evaluate the impact of behavioural screening on intervention participation or intervention effects within the context of a multiple behaviour intervention trial. Behavioural screening exerted an influence on intervention participation in terms of the characteristics of participants included in the sample, but had no impact on the intervention dose received or trial retention rate. Behavioural screening also impacted on intervention effects, in ways that were anticipated and positive, but also unexpected and detrimental.
The detrimental impact of behavioural screening on outcomes that were not screened for suggests that some intervention effects may have been unintentionally affected to a small extent in the minority of multiple behaviour intervention trials that have used behavioural screening. While, in the context of this study, the absolute magnitude of the detrimental impacts were often small, and of potentially limited clinical significance, the findings from this study suggest that the impact of screening in multiple behaviour intervention trials warrants further consideration. Our results intimate that screening for physical activity is likely to have the anticipated consequence of improving intervention effects for that behaviour, while diet screening based on fruit and vegetable intake, may not. Hence, the greater propensity to screen for physical activity in single rather than multiple behaviour intervention trials may lead to an overall underestimation of the capacity of the latter to demonstrate successful changes in physical activity.
Our findings also have implications for sample sizes required in future trials. Hypothesised intervention effects for physical activity should take into account whether the sample will be screened or not, with unscreened samples being likely to achieve smaller effect sizes and thus require markedly larger sample sizes. An alternative to using pre-trial screening to exclude certain participants is to use sub-group analyses to derive similar benefits in terms of effect sizes. However, particularly for lower prevalence behaviours (e.g., smoking), the required sample sizes may become prohibitively large.
Our study also highlights some potential mechanisms through which screening may exert both positive and detrimental influences. The benefits of physical activity and combined screening on intervention effects for physical activity were due to relatively greater improvements in TC group participants when these screens were employed than when screening was not used. Presumably, the findings suggest reduced ceiling effects (i.e., participants with lower baseline levels of physical activity had a greater capacity to improve in response to the intervention). For the other detrimental impacts, the mechanisms were less clear cut: sometimes changes in the TC group were affected, sometimes UC, sometimes both. Since retention and intervention dose were unaffected by screening, these can be ruled out as possible explanations for the impact on intervention effects. The impact could relate to the effect of screening on baseline levels, and possibly the impact on participant characteristics (as certain types of participants achieve different levels of success in behavioural trials) .
Behavioural screening reduced the baseline levels of the behaviours being screened for, consistent with its purpose of targeting participants who have the greatest need for a behavioural intervention . However, physical activity and combined screening (unlike screening for diet only) also significantly and substantially affected baseline levels of behaviours not targeted by behavioural screening, specifically increasing baseline levels of total and saturated fat intake (as a percentage of daily calories). This is not entirely surprising as health behaviours (including physical activity and diet) tend to co-occur [25, 26]; hence excluding participants with higher levels of physical activity is also likely to exclude those with healthier diets. Although our comparisons were limited by the small numbers of participants excluded, behavioural screening also tended to exclude participants with certain demographic and health-related characteristics, thus resulting in a less representative sample than was achieved without screening. Similarly, other studies have reported that adherence with recommended levels of multiple health behaviours is greater in samples with certain demographic characteristics . The changes in the demographic profile of the sample occurring with screening resulted in the selection of a sample with characteristics associated with multiple unhealthy behaviours, who are therefore potentially in greater need of an intervention .
The results of this study must be interpreted in light of the fact that this is a secondary analysis, with 'behavioural screening' conducted retrospectively using baseline measures. In order to imitate feasible behavioural screening practices, we used behaviour-based questions and employed the national physical activity guidelines and fruit and vegetable intake recommendations as cut-off criteria. These criteria are not perceptibly different from those used by other studies, although our review of the literature did find that behavioural screening practices vary greatly in terms of both the specific aspects of the behaviours targeted (particularly for diet screening), and the criteria used to determine eligibility (particularly for physical activity screening). Despite our realistic simulation of behavioural screening, it is possible that true behavioural screening, occurring some time prior to baseline, and based on different, behavioural measures, may have had slightly different effects on intervention outcomes than what was found in our analyses. Both in our study and in general, the use of self-reported physical activity when screening is problematic, as over-reporting may lead to the exclusion of participants who would potentially benefit from the intervention.
A number of other issues must also be considered. As anticipated when targeting a group with chronic disease, behavioural screening excluded only a small number of participants due to the high prevalence of physical inactivity and unhealthy dietary practices in our study sample. If we had screened to exclude those with a health behaviour with a higher prevalence (e.g., non-smoking) or targeted a healthier population, we may have excluded a larger proportion of the sample and thus seen greater effects of behavioural screening [5, 29]. For multiple behaviour trials that target a more diverse array of behaviours (including smoking, hazardous drinking and sun exposure) the complexity of screening is increased, and further research on the impact of screening for less prevalent behaviours is warranted. Also, our study was powered on the full sample (n = 434), so the loss of significance of intervention effects observed in the screened groups should not be over interpreted; changes to the size of effect should also be considered. Finally, the impact of screening may relate to the mechanisms previously discussed, but alternatively could be due in part to phenomena related to measurement error and biases of self report, as we used self-report measures of physical activity and diet that may be subject to over- or under-reporting [30, 31]. For example, screening could have an impact on outcomes by tending to exclude participants who are comparatively more or less prone to socially desirable reporting. Without objective measures, this is not something we could assess.