The Internet has become a popular tool which offers unlimited possibilities for finding health information. Indeed, a new health consumer identity is emerging, sometimes referred to as the "on-line self-helpers” [1]. In their quest for health information, users are usually faced with several challenges, namely that of searching for relevant information, of accessing a good quality of information, and of understanding the information retrieved. Health professionals are also experiencing an identity shift, moving from the authoritative figure to that of facilitator [2]. Professionals’ information and knowledge are, among other ways of distribution, commonly mediated on the Internet via health related websites. The health information provider has often limited insight about the user's understanding of the information given by means of these health-related websites.

Information understanding can put strong cognitive demands on the users, especially the older users, and is closely related to that of the construct of contextualization of information. The notion of context appears in several disciplines including cognitive psychology and computer science. Context in our research is studied within the framework of communication, specifically man-machine health communication. Durnell-Cramton [3] has shown that the lack of contextualization is a common problem in communication under dispersed and technology mediated conditions. Contextualization is about providing supportive information to explain a core message (in this case a health message). The contextualization of the information can help improving the user's understanding and sustain an effective man-machine communication [4, 5]. In particular, contextualization of information can help to reduce the user’s cognitive distance (i.e. the difference in knowledge between the user and the website as the provider of information). In this paper, we report on a study investigating the impact of context on the user's understanding.

In order to investigate the impact of contextualized information on the user’s understanding of the retrieved information, we were guided by the model of communication proposed by Te’eni [4] whose central idea is that of communication within a well-defined context. In his model of communication, Te'eni introduces the notion of communication complexity as the communicator's sense of the cognitive effort needed to ensure effective communication. Effective communication results from the use of cognitive resources to overcome the difficulties in understanding and uncertainty about the message. Another construct of this model is that of mutual understanding (MU) which requires that the communication be comprehensible according to the sender's meaning and the user's cognitive capacity.

Central to the model is the idea of context. Contextualization requires cognitive resources and is seen as a communication activity aimed at increasing mutual understanding by reducing complexity. In other words, contextualization is about providing contextual information to reduce its complexity and thereby support understanding of a core message. Te'eni found that communication complexity is due to the intensity of information, the multiplicity of views, the incompatibility between representations and use of information, and affected by the heterogeneity of the communicators involved, and so by the terminology they used to communicate. Heterogeneity triggers communicators to incorporate more contextual information in order to improve mutual understanding. Contextualization, if effective, reduces communication complexity and thereby increases mutual understanding (see Fig. 1).

Fig. (1). Te’eni model of communication.

The Te’eni model comprises six hypotheses around the various aspects of communication complexity, heterogeneity, contextualization and mutual understanding [4]. Given the scope of this study, we selected the hypothesis no. 5 which states that:The impact of contextualization on mutual understanding is contingent on heterogeneity: for heterogeneity, contextualization leads to higher mutual understanding, but for homogeneity, it does not. This hypothesis was adapted for our study reported here and formulated as followed: The impact of contextualization on mutual understanding depends on the cognitive distance between the user’s knowledge of the relevant health topic and the knowledge available in the website. The hypothesis states that a user with a low cognitive distance (i.e. a person knowledgeable in the field) will have a high level of mutual understanding whether or not they use contextualization. Conversely, a user with a high cognitive distance (e.g. a non-knowledgeable person) is expected to have a low mutual understanding with no contextualization (e.g. a missing level of information, a missing link between the levels of information) and a higher mutual understanding with contextualization (see Table 1). The model of communication proposed by Te’eni has been tested in certain domains such as engineering (machine assembly) but has not yet been used in the healthcare area.

The aim of the study was twofold: 1) To test the hypothesis on the impact of contextualization when searching for health information in an informative website, and 2) to uncover types of contextualization which are relevant to the users searching for health related information.

The website SeniorGezond (http://www.seniorgezond.nl) was used as the test-bed for our investigation. This website has been developed in our institution and is dedicated to the older people and their care givers in the area of fall prevention [6,7]. Falling is a major health problem within the older population.

About one third of the older people fall at least once a year, and due to demographic ageing, the group of older people will expand [8].

The website SeniorGezond is built to inform and educate the older people and their caregivers about the risks of falling and present them with possible actions that they can take to prevent falls. The website has no commercial purpose and is purely informative. While the website provides information about products and services, it is up to the website visitor to make an informed decision/choice about what products and services suit him or her best. Whether the user takes action and buys such a product is beyond this study.

The information within the website is structured around problems of fall incidences, and contains four levels of information (see Fig. 2). Navigation within the information structured is done using pull-down menus. The top level Causes of falls includes frequently occurring problems in the domain of fall incidences e.g. dizziness. The second level Solutions focuses on possible interventions and advices associated with the causes of falling e.g. use of a walking aid. Solutions are supported by the third level Products & Services e.g. walking aids, fitness programmes. There is no overlap in information between level 2 and level 3. Level 3 shows pictures or descriptions of the products and services; the rationale for choosing a product or a service is given in level 2. From the Products & Services level, users can access the Supportive Facts which makes up the fourth and lowest level of the information trees. Supportive Facts contain addresses about where to purchase products and services, as well as insurance related information. The four levels take the user from general to specific information.

Fig. (2). Information structure in the SeniorGezond website.

Manipulation of the contextualization was achieved by removing the Solutions level from the navigation (2^nd level) as well as cross-links to the other levels. This level embeds relevant contextualization since its acts as a buffer between the background information level and the Products and Services level. In particular, the Products and Services level only gives descriptions about products such as walking-aids and services such as sport and physical exercises but no information about the reasons one should select such a product or service (given in level 2 Solutions).

Forty participants were recruited via advertisements in local newspapers and via personal contacts. Prerequisites for the selection included familiarity with Internet usage and not being a healthcare professional.

A randomized controlled experimental design was chosen. The study was composed of a pre-test for gathering participants' profile, the intervention with the website and a post-test for measuring participants' mutual understanding (see Fig. 3). The exposure to contextualization was randomly distributed amongst the participants. In this design, context within the website was the independent variable while participant’s MU was the dependent variable. Possible confounding variables which were considered include participants’ domain knowledge and participants’ cognitive style.

Fig. (3). Study design.

In the gathering phase, participants’ characteristics (baseline instrument) and their knowledge about the domain (domain knowledge instrument) were collected as well as their cognitive styles via the GEFT test [9].

During the intervention phase, participants with the context and no-context manipulated website were given the same task to carry out. Participants were given the task of finding an appropriate walking aid for a fictive older family member who has a problem of dizziness

To complete the task, all participants (independently whether they themselves had previously experienced a fall) were asked 1) to select an appropriate walking-aid for this family member and 2) to explain the reasons why (using structured questions, for example regarding the distance one can walk with the selected walking-aid). Participants were instructed to read the information from the website (rather than just guessing). In the last phase, participants were tested for their understanding of health information they had found in the website.

Four instruments were needed for data collection and hypothesis testing: Three instruments were self-developed (Baseline instrument, Domain Knowledge instrument and Mutual Understanding instrument); a cognitive style instrument (Group Embedded Figure Test; GEFT) from the field of psychology was used unchanged [9].

Forty participants took part, half of whom were randomly given exposure to contextualization and the other half had no contextualization exposure. The total group of participants (n = 40), before the random assignment, had a mean age of 56,7 (SD 14,4) years old (varied from 20 to 80 years old), and there were more female (n=24) than male (n=16) participants. The educational level² of the participants was 12,5 % (n=5) low education level, 37,5 % (n=15) medium education level and 50% (n=20) a high education level. Most of the participants used Internet on a daily basis (67.5%; n=27) or at least once per week (31.5%; n=13). Almost half of the number of participants (47,5%; n=19) looked for health-related information on the Internet for at least once per month. In 57,5% (n=23) of the participants someone in his environment (such as family members, relatives or friends) has had a fall accident and 25% (n=10) of the participants has had a fall incident themselves.

The frequency of Internet usage with respect to participants’ age group is shown in Fig. (4).

Fig. (4). Frequency of Internet usage.

After the randomization there were no statistical significant differences between the two groups regarding age, sex, educational level, Internet usage, frequency of searching health related information, and fall incidences (Mann Whitney U test, Chi square test or Fisher’s exact test, where appropriate). Data were collected between November 2006 and March 2007.

Fig (5) shows that the mean domain knowledge of all the participants (n=40) was 57% (SD 14%). Experts (n=12) from LUMC also completed the domain knowledge questionnaire and as expected scored higher than the participants (mean domain knowledge was 75% (SD 11%). We defined participants to be knowledgeable in the field of falling (threshold) when they have a score higher than 57% (21 participants >57%; 19 participants < 57%).

Fig. (5). Distribution of the participants’ domain knowledge.

Analysis of participants’ MU shows that participants mainly scored in the medium range (see Fig. 6). Contrary to our expectations, almost no low or high MU score was found. Results show that the impact of the manipulated website on MU only for non-knowledgeable participants is statistically significant (p=0.60 and p=0.04; knowledgeable and non-knowledgeable, respectively; Student’s t-test). In other words contextualization tends to increase understanding for non-knowledgeable users (see Figs. 7,8).

Fig. (6). Distribution of MU scores.

Fig. (7). Impact of website on MU for knowledgeable participants.

Fig. (8). Impact of website on MU for non-knowledgeable participants.

Two-way ANOVA revealed neither statistically significant main effects for type of website (F=3.76; df=1; p=0.06) and level of knowledge (F=0.09; df=1; p=0.76) nor their interaction effects (F=1.35; df=1; p=0.25).

Results from the GEFT indicated that a majority of the participants (65%, n=26) can be classified as Field Independent (FI) and the rest (35%, n=14) as Field Dependent (FD). Further results showed that the impact of the type of cognitive style (FD and FI) on MU is not statistically significant for both types of websites (p=0.65 and p=0.06; FD and FI respectively; Student’s t-test). In other words participants’ different cognitive styles do not have an impact on mutual understanding in case of using a normal website as well as using a manipulated website (see Figs. 9,10).

Fig. (9). Impact of website on MU for Field Dependent participants.

Fig. (10). Impact of website on MU for Field Independent participants.

Two-way ANOVA, however, revealed a statistically significant main effect for the type of cognitive style on mutual understanding (f=6.08; df=1; p=0.02) with a large effect size (Partial Eta Squared=0.15). The main effect for the type of website (F=2.30; df=1; p=0.14) and the interaction effect (F=0.45; df=1; p=0.51) did not reach statistical significance.

Transcripts from the MU instrument helped to identify participants’ contextualization themes. Results indicate that participants bring their own contexts when searching and deciding which information is relevant. We have identified two types of contexts, namely social context and psychological context.

Results from this empirical investigation corroborate the hypothesis from the Te’eni model, namely, that exposure to contextualization increases understanding for non-knowledgeable users. However, care should be taken when interpreting this result. Indeed there maybe two possible factors which have affected the outcomes:

1. Participants’ MU threshold: almost no low or high MU score was found. The participants’ outcome measure of MU fell mainly within the median score threshold.
2. Distribution of FI vs FD participants: More participants were identified as Field Independent (n=26) than Field Dependent (n=14). This may partly account for the participants’ outcomes of the MU scores.

As stated in the introduction, the communication model of Te’eni is complex and comprises a set of six hypotheses. Our study has been based on one of those hypotheses. Work on contextualization will need in the future to take into account additional interactions of the different components (see Fig. 1) as foreseen within the remaining hypotheses. In particular, of immediate interest are hypothesis #1 “Greater heterogeneity of communicators leads to more contextualization”, and hypothesis #3 “More contextualization reduces communication complexity”.

It should also be stressed that the website was not developed with functionalities of contextualization in mind. As stated in the Methods section, the website was used as a test-bed for our current investigation. The contents was structured using the Precaution Adoption Process Model, a health behavioral change stage theory [14]. Users need information that fit the stage they are in, and that helps them to move to the next stage of the model where eventually, they might take and maintain action to prevent from falling [15]. The stages of the model were translated into a users’ information need and thereupon converted into the information levels of SeniorGezond.

Kim et al. [12] have shown that the cognitive style (Field Independent / Field Dependent) plays a role when searching for information on the Internet. In particular, they have found that a Field Independent person tends to perform efficient searches, and use active and analytic searches more often. Conversely, a Field Dependent person will choose tools that are salient but not necessarily useful for completing the information searching task, use Home keys more often and tend to be distracted easily. Our finding confirms that the cognitive style Field Independent / Field Dependent is a significant factor in information seeking. However, in the scope of our study, we have not examined which web functions participants used (such as keyword search or “back” button). It is certainly an aspect which needs to be investigated further. An analysis of the web logs collected from the participants who worked with the non-context version of the website would be a starting point.

Other dimensions of cognitive styles have also been examined in the field of web information seeking, but not extensively. For example, work has been carried out [16] to measure the cognitive style “verbaliser” versus “imager” in relation to information seeking on the Internet. It is clear that other dimensions of cognitive styles should not be over looked in our future investigations.

Results show that participants bring other non-informational contexts (namely social and psychological) when searching for answers and justify their choice. Indeed, information seeking does not occur in a vacuum but rather arises from and is conditioned by the circumstances in which the information seeker is [17]. People tend to make sense of the information they have at their disposal in the light of their own needs and their own interpretations. Wilson’s model of information behavior shows how psychological and interpersonal environmental characteristics influence a person’s information seeking process and interpretation [18]. It is therefore possible that participants in our study have relied on psychological and social factors to 'reinforce' the information they found and selected.

Furthermore, within this sphere of social and psychological contexts, social perception [19] and social judgment [20] seem to play here an important role. Social perception is the study of how we form impressions of and make inferences about other people. The central idea of social judgment theory is that attitude change is mediated by judgmental processes and effects used to persuade people. In our study, fear of being stigmatized as invalid and handicap as reported by some of our participants is a good illustration of such social perception and social judgment. Not surprisingly, these remarks of stigmatization came from the older participants. Stigma and discrimination of being old remain for the older participants a real concern. In defense against the stigma, older individuals tend to deny their old age through for example physical disguises, the attainment of active mastery in areas traditionally closed to elders, avoidance of social interactions with other age groups, or self-other identification [21]. In our study, the choice of a walking aid such as a stick which is less visible (than a 4 wheel-walker) may be seen as an example of such attitude against stigmatization.

While contextualization of information such as the interventions level of SeniorGezond enhances some participants’ understanding of the domain area, our results indicate that understanding is coupled with other cognitive abilities such as:

Designing websites with functionalities of contextualization can strengthen the already existing informative functions and is a step towards getting closer to the user's own situation and preferences. However, while this top-down approach (from experts to users) is important, one should not ignore the opposite way – from users to experts – which takes into account the users' own non-informational contexts. There is a gap between the informational contexts proposed to the website visitors (such as the level of possible solutions) and the user’s non-informational contexts. This gap needs to be bridged to insure that both top-down and bottom-up approaches to contexts can be coupled in order to develop coherent contextualization guidelines. Within the setting of our study, we had chosen the whole level of interventions as context. However, it is not the unique possibility. More work needs to be carried out in order to investigate which website contexts as well as users’ contexts should be implemented in a given application. Such investigation cannot be restricted to only one website. We plan to use other existing Dutch health portals for this future investigation.