This study provides insight in the long-term efficacy of four intervention conditions (i.e. printed basic, printed environmental, Web-based basic and the Web-based environmental) of the Active Plus PA intervention in (subgroups of) adults aged over fifty. These insights provide indications for the feasibility of the large-scale implementation of printed or online tailored PA interventions (with or without an environmental approach) among older adults.
Overall, when considering only the complete cases, the Active Plus intervention as a whole was effective in increasing weekly days of sufficient PA (ES = .18), but only borderline effective in increasing weekly minutes of PA (ES = .20; p = .071). When performing similar analyses on a dataset in which the outcome measures for participants who dropped-out during follow-up were imputed, the Active Plus intervention was found to be significantly effective on both outcome measures. By imputing the missing data power was increased, resulting in a significant overall intervention effect on minutes of PA.
More in-depth analyses showed that 12 months after the intervention started, only the printed conditions resulted in significantly increased weekly minutes and days of PA. As short-term results (i.e. 6 months after the intervention started) showed that also the Web-based conditions were able to increase the weekly minutes of PA (Peels DA, Van Stralen MM, Bolman C, Golsteijn RHJ, De Vries H, Mudde AN, Lechner L: The differentiated effectiveness of a printed versus a Web-based tailored intervention to promote physical activity among the over-fifties, Submitted), this indicates that the effects of the Web-based conditions are less well maintained than the effects of the printed conditions. Effect sizes of the printed interventions regarding the effect on weekly minutes of PA decreased when follow-up time increased, but less sharply than the effect sizes of the Web-based interventions. Our results are in line with recent meta-analyses [20, 21] which both showed decreased intervention effects when follow-up time increased. In contrast to our study, in the meta-analyses no difference in intervention effects between different delivery modes was found (i.e. print, computer, telephone, etc.).
Maintenance (i.e. sustainability) of behavioural intervention effects is of major importance to achieve an impact on public health . More insight into factors that can explain or stimulate the maintenance of interventions is therefore of major relevance. The Active Plus programme evaluation  showed that the Web-based intervention materials were less often used, less often saved and less well appreciated than the printed intervention materials. Increasing the appreciation and usability of the Web-based intervention materials might increase the sustainability of the Web-based interventions. Evidence [23, 24] suggests that intervention features such as provisions for peer and counsellor support, email contact with supervisors and regular website updates, were related to increased exposure in internet-delivered health behaviour interventions. These features can be considered as intervention boosters. The current Active Plus website only provided participants with the tailored advice, none of the suggested intervention features were implemented in the website. Since research has identified a clear dose–response relationship between the intensity of the intervention and the resulting behaviour change, it can be expected that an increase in website engagement is an important factor for the long-term effectiveness of the Web-based intervention . Integrating the suggested intervention features  might help stimulate the sustainability of the intervention.
No statistical significant difference in the intervention effect was observed between the basic and the environmental conditions, in neither the printed nor the Web-based intervention. In practice, however, the difference in effect between the basic and the environmental condition (i.e. of about one hour per week when comparing the effect of the printed basic intervention condition to the printed environmental intervention condition) might still have implications for both health and for policymaking.
The effect of the printed environmental condition (which was the intervention condition in which the participants received the most information) was moderated by the participants’ PA intention; weekly days of PA only increased in participants with a high baseline intention. Possibly only people with a high intention are willing to use all additional environmental information that is provided to them. Participants with a low intention might perceive this intervention as an information overload, resulting in less effect. The intervention effect was not moderated by the participants’ age, gender, BMI, SES, or the presence of a chronic physical limitation. These equal effects in all subgroups might indicate that the intervention was successfully and sufficiently tailored to the different relevant characteristics of the participant, as tailoring is supposed to do.
Whereas our measurements relied on self-reported data through validated questionnaires, the responses can be biased by social desirability. Although self-reports may be less accurate than objective observations, self-administered questionnaires are the most commonly used, and most inexpensive method to use in large-scale studies. Validating intervention effects with objective measurements in the future would be recommendable.
Attrition analyses showed that participants in the Web-based and in the environmental conditions were more likely to dropout from the 12-month assessment. A possible explanation for the higher drop-out within the Web-based conditions might be that it requires more planning to fill in a Web-based questionnaire (i.e. print-delivered questionnaires can be filled in anywhere at any time, while filling in online questionnaires restricts one to a computer). Furthermore, in a printed questionnaire it is easier to resume filling in the questionnaire after pausing, rather than resuming a Web-based questionnaire, due to loading times and additional log-ins . One possible explanation for higher drop-out in the environmental conditions could be (as shown in focus group interviews among intervention participants (Peels DA, Golsteijn RHJ, Lechner L: Advice report on the feasibility of the Active Plus intervention: in depth analyses from focus group interviews with participants, Unpublished)) that the environmental intervention components motivated participants to search for additional information outside of the Active Plus intervention. This might make returning to the Active Plus website less necessary for these participants. Another explanation could be that receiving both tailored advice and additional environmental information could be experienced as an information overload, resulting in intervention dropout. This possible overload of information might also explain why the environmental information does not result in significantly increased intervention effects.
Furthermore, analyses showed that dropout was higher among younger participants and participants with a low PA intention. Due to this selective dropout, the overall effectiveness of this study might be biased. Longitudinal studies often result in missing data due to dropout during follow-up. Using multilevel regression analyses or using multiple imputation methods are often used methods to handle missing data. In the current study, both methods were applied with comparable results, although there were some differences. Which analyses are the most reliable is a major discussion point. Considering only the complete cases regarding the outcome measures provides the best prediction of parameter estimated, but it also results in a loss of power  and the effects of the intervention might be overestimated. Multiple imputations do not really add more information but the sample size is increased and the standard error reduced. Since sufficient power is essential when performing moderation analyses (i.e. subgroup analyses), performing the moderation analyses on an imputed dataset might result in the best estimates. However, analyses applied to the imputed dataset showed that the intervention effect was not moderated by any of the assessed participant characteristics. Several studies concluded that it was not necessary using multiple imputations before performing a multilevel analysis on longitudinal data [19, 25]. Based on these studies, we might assume that outcomes from the multilevel analyses without applying multiple imputations could be considered to provide the best estimates for our intervention effects.
In this paper the effect of four different intervention conditions are studied and compared to each other, resulting in multiple testing. By performing multiple tests we were able to show the results of this study in a broader perspective and to give a more nuanced picture of the study results. A disadvantage of multiple testing is an increase of the probability of making a Type I error. A Bonferroni correction, however, assumes that all of the hypothesis tests are statistically independent, which is not the case in the current study. The probability of making a Type I error would be less than Bonferroni assumes, and the Bonferroni would be an over-correction. Therefore, we did not apply a Bonferroni correction to the current study results. However, P-values found in the current study were so strong, that even when correcting for multiple testing, results would still have been significant.