Based on the identified problem and needs we determined that the primary outcome was to improve HPV-vaccination uptake among invited girls and the secondary outcome was to strengthen mothers' informed decision-making, reduce decisional conflict, and positively influence determinants of the HPV-vaccination decision. Below, we introduce POs, determinants of these POs, and accompanying COs for each outcome.

The expected intervention outcomes were subdivided into POs. The HPV-vaccination consists of two subsequent injections. In other words, the behavior has to be repeated only once (with an interval of 6 months). The POs are: (1) the mother makes a (informed) decision to have her daughter vaccinated against HPV; (2) the mother discusses her decision with her daughter and partner; (3) the mother guides her daughter toward receiving the first HPV-injection; (4) the mother guides her daughter toward receiving the second HPV-injection.

For each PO, we identified the reasons why mothers would take that action. These so called behavioral determinants were based on theory (e.g., the theory of reasoned action and socio-cognitive theory) (25, 26) and on empirical research (19, 20, 27). We selected determinants that met the criteria of importance and changeability (10). Importance of the determinants of PO 1 (i.e., deciding to get the HPV-vaccination) was based on the association (i.e., R2 effect size) (28) of the determinants with HPV-vaccination intention (21). For the POs 2–4 (i.e., discussing the HPV-vaccination and actually getting the first and second injection), importance was based on consensus among co-authors (RR, TP, MP, and HvK). Changeability (i.e., the strength of the evidence that the proposed change can be realized by the intervention), was also based on consensus among co-authors (RR, TP, MP, and HvK).

For the first PO, we returned to the needs assessment and selected the following determinants: attitude, beliefs, positive, and negative outcome expectancies, anticipated regret about both receiving and rejecting the HPV-vaccination, confidence in authorities, habit strength, risk perception having received (no) HPV-vaccination, subjective, and descriptive norms, relative effectiveness of the HPV-vaccination, ambivalence, and HPV-vaccination information processing (19, 21). Furthermore, according to Marteau et al. (27), an informed decision is based on sufficient and relevant knowledge, and a match between the person's values (i.e., their attitude toward the HPV-vaccination) and outcome behavior (i.e., whether mothers had their daughter vaccinated against HPV or not). Consequently, knowledge was selected as a determinant. Determinants that were selected for PO2 were attitude (29), and self-efficacy (26). Selected determinants for PO3 and PO4 were knowledge and beliefs (26). Determinants that were not selected were considered either unimportant (e.g., for PO1: self-efficacy) (19, 21) or unchangeable (e.g., for PO2: parenting style).

For each determinant, we identified COs. COs are the active ingredients of the intervention and function as a blueprint of the theoretical design rationale. Table 1 provides an overview of examples of the matrix of COs, the complete version can be found in Additional File 1.

The first task of Step 3 is to generate ideas for intervention theme, components, scope, and sequence. The product of this step is an initial plan that describes the program (10). We decided the intervention to be Web-based (see Step 1). The main theme of the intervention was “making an informed decision about the HPV-vaccination of your daughter.” The various components were logically clustered for improving usability. This resulted in four menu options: (1) two-sided information about the HPV-vaccination, (2) a decisional balance, (3) practical information, and (4) frequently asked questions (See Additional File 3 for screenshots of the four menus). The first menu enabled mothers to collect tailored information about the HPV-vaccination (e.g., such as information about the effectiveness of the HPV-vaccination). The various components were in line with mothers' preferences as indicated by earlier research (See “needs assessment”) (10). In the second menu, mothers could weigh their personal values regarding the HPV-vaccination in the form of a decisional balance and values clarification tool. In the third menu, mothers could gather practical information such as how and where to receive the HPV-vaccination. The fourth menu listed frequently asked questions about the HPV-vaccination. Here we also added a “problems with the website” component, providing mothers with help.

Furthermore, mothers were able to visit the intervention multiple times. The first time they visited the website, they were provided with an explanation of how the website worked. Then, they were introduced to the first menu. We used a combination of a freedom of choice and a tunneled design (i.e., a “hybrid design”) to guide mothers through the website (30). This means that mothers could choose themselves which component in which menu they wanted to visit (i.e., freedom of choice design). However, once they entered a component, they were guided through it in “tunnel fashion,” with navigation being limited to “next” and “prior” buttons. The reason for choosing such a hybrid design is that we wanted to profit from the strengths of both approaches (30). Specifically, the tunnel design was expected to increase intervention adherence and engagement and acquisition of knowledge (31). The freedom of choice design was expected to promote mothers' autonomy, which is important when motivating behavior change (32, 33). Furthermore, the hybrid design matched the differential needs concerning the amount and scope of information expressed by the mothers (19). Table 2 provides an overview of the scope and sequence of the intervention.

To identify theoretical change methods that help achieve the COs, we used an overview of methods provided by Bartholomew et al. (chapter 6) (10). The eHealth setting gave us the chance to apply effective strategies in an innovative way, namely by using computer-tailoring and interactions with virtual assistants (see sections below) (34). Computer-tailoring was selected as the main theoretical framework for development. Tailoring is a health communication strategy by which messages are individualized to personal preferences and needs (35). Meta-analyses have shown that tailored interventions are more effective than generic interventions in achieving behavioral outcomes [e.g., (36, 37)]. Beneficial effects of tailoring are attributed to improved exposure, information processing, appreciation, reading, and perceived personal relevance [e.g., (38, 39)]. Because computer-tailored interventions can reach large groups of people at relatively low costs, especially when delivered via the Internet, (40), they can have substantial impact at the population level (41). Also, tailoring matches the mothers' need for more interactive, personal information about the HPV-vaccination (Step 1) (19).

Computer-tailored feedback was used in three different ways throughout the intervention. First, it was used to tailor the feedback on participants' answers to statements and questions about specific aspects of the HPV-vaccination. For instance, mothers were first asked to estimate their daughters' chance to get an HPV-infection. Those who perceived this chance as low, received feedback which stated that this chance is rather high instead of low, whereas those who perceived the chance as high, received feedback that confirmed that the chance is indeed high. Second, computer tailoring was used to provide mothers the opportunity to weigh their personal values regarding the HPV-vaccination in a decisional balance. Another mean was the “value clarification” tool [a motivational interviewing strategy; cf. (32)]. Mothers were invited to list their central values for life, and were stimulated to relate these to the decision about vaccinating her daughter. Finally, computer-tailoring was used for guiding mothers through the website. The intervention kept track of the components that the mother had already visited by using logs. This enabled us, for instance, to highlight parts of the intervention which the mother had not seen yet. Also, if mothers were exposed to information that had already been discussed in another component, the intervention mentioned this in order to assure that the connection between the different types of information provided was clear.

We selected virtual assistants for delivering tailored feedback. A virtual assistant is an embodied conversational agent defined as a computer program with a human-like visual make-up and appearance on a computer screen (42). Virtual assistants were chosen to match the mothers' preferences for more interactive personalized feedback (19). They provide opportunities for two-way interactions, and can create a highly personal experience. Also, research has indicated that a social relationship between user and program is important (43, 44), as it supports the basic psychological need for relatedness (33, 45). This can be established by using virtual assistants (46–48). Also, several studies confirmed that the presence of a virtual assistant can further improve the effectiveness of the intervention (49–51). Specifically, the added value of using a virtual assistant over a text and picture-based website is that it improves recall of information (52), transfer of learning (53), amount of learning (54), self-efficacy expectations, literacy, and behavior change (49, 50, 55). In addition, the mere presence of such an animated interface agent has a positive effect on experiencing fun and engagement [e.g., (47, 50, 56)]. Two virtual assistants were visualized: a mother-like and a female doctor-like assistant as the combination of using an expert and a peer virtual assistant has been shown to be effective (57, 58). The main purpose of the virtual assistants was to provide mothers with social support, which is an important factor associated with positive health outcomes in general (59). The mother-like assistant was used to guide mothers throughout the website and helped weigh their personal values in the decisional balance. The doctor-like assistant was used to deliver feedback about the HPV-vaccination.

Table 3 provides examples of theoretical methods (column 2) for determinants identified in IM step 2 (column 1) for PO1 (i.e., mother makes the informed decision to have her daughter vaccinated against HPV). For each method, parameters for effectiveness were specified (column 3). We then translated theoretical methods into practical applications (column 4) that were appropriate for the population and the (Internet) setting. In Additional file 2, we also specify which POs and COs were targeted using which methods and applications in the various components. This can be seen as the most straightforward blueprint of the intervention. All COs were covered by the intervention.

The most important method aiming to reduce decisional conflict was the decisional balance (see Figure 2), which has proved a quick and efficient intervention by itself (63). Mothers were presented with a list of pros and cons of the HPV-vaccination by the mother-like assistant (left column). This list was based on pros (e.g., “the HPV-vaccination decreases the chance of my daughter getting cervical cancer”) and cons (e.g., “my daughter is too young to receive the HPV-vaccination”) that were considered most important to the mothers as indicated by the needs assessment (Step 1). For each pro or con, they indicated (1) whether they agreed (disagree/neutral/agree; middle column) and (2) how important the pro or con was to them (unimportant/neutral/important; third column). The latter was indicated by stars: the more stars, the more important the pro or con was to the mother. When mothers (dis)agreed, tailored feedback “popped up.” This was done to ensure mothers based their answer on correct information (see Figure 2). Furthermore, mothers were given the option to add pros and cons that were not in the list. Based on pros and cons mothers marked as most salient, a decisional balance (top right of the screen) revealed their current position on a scale ranging between not-wanting (left side) and wanting (right sight) to get their daughter vaccinated.

In order to anticipate an intervention that meets the requirements and preferences of the target group, we followed user-centered design procedures. This entails the iterative involvement of the end-users in the design process (64). We gathered feedback on different versions (static and interactive) or (parts of the) intervention, by online experimental pretests and focus groups, respectively. Experimental pretesting offers empirical support for the impact of strategies on determinants and serves as a guarantee for implementing adequate intervention materials (65, 66). We conducted three experimental online pretests. Since we did not find a clear consensus in the literature about the framing of risks, the first experiment (N = 375) was about testing the differential effect of providing statistical (i.e., probability rates) or/and narrative risk information (i.e., a personal story). Mothers were randomly assigned to one of four conditions in a 2 (statistical information: yes or no) × 2 (narrative information: yes or no) between-subjects factorial design. ANOVA revealed a significant main effect of statistical information on daughters' perceived susceptibility toward HPV [F_{(1, 371)} = 7.56, p < 0.01]. Mothers who received statistical risk information had a higher perceived daughters' susceptibility toward HPV (M = 4.11 on a 7-point scale, SD = 0.10) than mothers who did not receive statistical risk information (M = 3.73 on a 7-point scale, SD = 0.09). Thus, statistical risk information seemed most effective. We therefore decided to include statistical risk information to target risk perception with this intervention (Step 3).

In a second online experimental pretest (N = 561), we explored the best way to communicate about social norms; by providing negatively (i.e., discourage undesired behavior) vs. positively (i.e., encourage desired behavior) framed descriptive and/or subjective norms (67). It was suggested that the descriptive norm should be avoided in situations where the unhealthy behavior is prevalent (68). As for the HPV-vaccination, 39% of invited girls have not received the HPV-vaccination (15). Therefore, we also wanted to examine whether communicating a descriptive norm could have a potential adverse effect on HPV-vaccination acceptability. Mothers were randomized into one of four conditions in a 2 (norm: injunctive vs. descriptive) × 2 (frame: positive vs. negative) between subjects factorial design with an additional control condition. We found no indication for using one type of framing norms over the other (p's > 0.05; mean HPV-vaccination intention scores ranging from 5.51 to 5.77, on a 7-point scale). We also did not find any adverse effects of descriptive norms on HPV-vaccination acceptability (p > 0.05). As negatively framed norms were expected to be more difficult to process (69), we decided to just include positively framed descriptive norm by communicating about the national HPV-vaccination uptake rate. Within the component “talking about the HPV-vaccination,” mothers were taught how to deal with a potential contrasting subjective norm of important others (e.g., their daughter and partner).

Finally, being a relatively new vaccine, there remains uncertainty about potential long-term effects of the HPV-vaccination. This was also found to be a topic of interest among mothers for future communication (19). Therefore, in a third experimental pretest (N = 695), we investigated the effects of acknowledging vs. ignoring uncertainty about potential long-term effects of the HPV-vaccination. Mothers were randomly assigned to one of two conditions in which uncertainty about the HPV-vaccination was either (a) acknowledged or (b) ignored. Results showed that, compared to mothers who were exposed to information ignoring uncertainty, mothers who were exposed to information acknowledging uncertainty experienced more decisional conflict (acknowledged: M = 3.42, SD = 1.84, p < 0.01 vs. ignored: M = 3.05, SD = 1.74), were more ambivalent about their decision (acknowledged: M = 4.04, SD = 1.86 vs. ignored: M = 3.42, SD = 1.89, p < 0.001), and had a less positive attitude (acknowledged: M = 5.07, SD = 1.50 vs. ignored: M = 5.69, SD = 1.38, p < 0.01) and intention (acknowledged: M = 5.26, SD = 1.73 vs. ignored: M = 5.85, SD = 1.45, p < 0.01). These findings implicate not to communicate about long-term uncertainties. However, we chose to do so in the intervention, for the following reasons: first, the found effect sizes were small. Second, not communicating about long term uncertainties brings along the risk of mothers searching information about this elsewhere. This can be quite dangerous as many rumors about potential long-term effects, for which no prove exists, can be found (e.g., on the Internet) (70). Reading these (false) rumors without any refutation being offered aside (71), could have more detrimental effects on HPV-vaccination acceptability than when we ourselves provide the (correct) information. The latter enables us to inoculate mothers with arguments that become accessible in case they are confronted with (new) information that might challenge their initial positive intentions (i.e., psychological inoculation) (22, 23). Finally, mothers themselves expressed a need for full disclosure, especially when uncertainties were ignored, which was also found in a previous study (19).

At a later stage, we conducted several focus groups (N = 3) among mothers to test interactive prototypes of the intervention to ensure compatibility with the preferences of the target group. A first prototype of the intervention was tested in two focus groups. After we revised the prototype according to the feedback from these two focus groups, a second prototype of the intervention was tested in a third focus group. The protocol was similar for all focus groups: after a general introduction, mothers were given a laptop and headset to individually navigate through the website. They were given the opportunity to give feedback on every page of the website about features they (dis)liked (e.g., the “look and feel” of the page(s), and the tailored feedback of the virtual assistants). Then, they were asked to fill out a written questionnaire assessing their subjective evaluation of the virtual assistants (e.g., the extent to which feedback matched their responses) and the website (e.g., their evaluation of the different menus). Finally, in a group discussion mothers could elaborate on their opinion about the intervention, and offer suggestions for improvement.

Feedback was first gathered from the first two focus groups. For instance, in the first prototype, there was a component targeting anticipated regret by using imagery (72, 73). Mothers were asked how much regret they would have if they did not vaccinate their daughter against HPV and their daughter developed cervical cancer later in life. However, we discovered that asking this evoked much resistance. We therefore decided to remove this component from the intervention. As an alternative, we decided to target anticipated regret indirectly (e.g., by giving information about the high prevalence of HPV). Furthermore, in the first prototype, the written and spoken tailored feedback were provided at the same time. Mothers indicated that, therefore, they experienced difficulties listening to the virtual assistant. Hence, we created a new prototype, in which the written feedback appeared once the virtual assistant was done providing the tailored feedback.

In the third focus group, mothers indicated that they would like to see which components they had already visited. We therefore created an adapted version of the website in which logs were used to register the pages mothers had already visited and subsequently used these logs to visualize which components were completed. This was done by turning them into a different color (i.e., orange, see Figure 4) In addition, the mother-like virtual assistant was used to give advice about components to visit next, in order to maximize exposure to the intervention. If the virtual assistant advised on a component, the component was highlighted by an orange circle (see Figure 4).

After revising the intervention according to the feedback from the third focus group, a final prototype was pilot-tested online using various devices to ensure it worked adequately. This was done among a sample of mothers (N = 10) and among members of the project group.