TLA provides broad, high-level views of user behavior in naturalistic settings over time. To our knowledge, TLA has not been widely applied at a user session level for exploring consumer health information seeking. In summary, the findings reported here suggest the following profile of user online navigation activities at ClinicalTrials.gov:
- Individual trial records (View Study; 40%) are the most frequently viewed page, followed by the query results (View Results; 25%).
- Users enter the site at a trial record (39%) more often than the homepage (24%).
- External Web sites refer a majority of users (69%); where Google (41%), other NIH-sponsored sites (18%), and other search engines (7%) are the top referring sites.
- The most common user session is viewing one or more trial record (40%), followed by viewing a results list and clicking on a study or simply viewing a results list (9% each).
Contrary to the initial expectations and intensions of the ClinicalTrials.gov Web designers [11
] and Web-design guidelines (e.g., [12
]), many users do not (1) enter from the homepage where a detailed menu of options is provided; (2) directly use the search and browse features provided by the site; and (3) spend time exploring the site and refining their search queries for their information needs. However, an original design goal of the ClinicalTrials.gov site, to require a minimal number of clicks from the home page to reach a trial record [11
], was supported.
One critical factor may be the increasing use of search engines, particularly Google, regardless of the availability of high-quality domain-specific resources [10
]. As a result, the most “relevant” sites/pages indexed by the search engine not only get the most exposure, but the most direct visits. For ClinicalTrials.gov, these are generally individual trial records (i.e., View Study
) and pre-specified URL-encoded search queries (i.e., View Results
Our preliminary results show that many users only view these pages and then return to the referring site using the Web browser’s Back
button to potentially click on another study link, indicative of the hub and spoke behavior pattern [7
]. While such users obtain study information, most do not take advantage of the other site features, such as Search within Results
(narrowing a query) or Resources
, (background information on clinical trials). It is likely that users do not realize that additional, highly relevant information might be accessible at the site.
New design strategies are required to aid users in finding features that may help satisfy information needs. Descriptive text and/or site instructions on a consumer health site homepage are not seen by users entering a site at low-level pages. Thus, providing links to background information and other search features on low-level pages (directly) might prompt users to continue searching within the site. Similar to recommendations to add site indexes throughout a document space [7
], this approach supports the hub-and-spoke behavior pattern and orients users, increasing the chance for exploring related information. In addition, there is evidence that users of Web sites containing visible local maps are (1) less likely to abandon information-seeking episodes, (2) delve deeper into the site hierarchy, and (3) use the Back
button less [13
]. Different research approaches and tools are needed to confirm these hypotheses.
Pilot User Study
We conducted a pilot user study to provide initial TLA validation by demonstrating convergence of the data from the two methods. An additional goal was to begin evaluating methods to compensate for TLA limitations and illuminate consumers’ information seeking goals on Web sites like ClinicalTrials.gov. As with the TLA referral data, participants used search engines to find clinical trials information. Participants also exhibited a hub-and-spoke navigation behavior pattern where search engine results pages and other information pages, such as MedlinePlus, served as hubs.
While four of the five users who arrived at ClinicalTrials.gov entered at low-level pages, three entered at View Results. TLA data indicated most users entered at the individual trial record level (View Study), so the two methods provided differing data. To investigate this trend, we reviewed queries for users referred to View Study pages associated with the scenario disorders to look for similarities. We found most queries used specific terms, (e.g. “provigil,”) or were complex queries, (e.g. “sleep apnea mirtazapine,”) linking to specific trials. Additionally, queries for users referred to View Results pages contained general terms “apnea” or “parkinson.” The study participants also searched with generic terms as they were instructed to find clinical trials on the disorders, indicating they may not be representative of the Web site users overall.
TLA and user studies may be complementary, but to integrate the approaches, strengths and weaknesses of each method must be assessed. TLA provides high-level trends over many user sessions, while user studies provide rich data about individual user actions. Thus, further research is needed to understand how to integrate methods for data at two levels of granularity (i.e., population and individual) and ways to correlate these data when the generality of the user group and the demographics of the population are unknown.
In this exploratory study, we used existing techniques of TLA and user studies to investigate the complex phenomenon of consumer information seeking on a health-related Web site. The pilot user study was intended to increase the understanding on consumer behavior online by providing possible correlations among data from different sources. Future research using both the TLA and user study methods, explored in this preliminary study, will be needed to validate this approach.
Sessions were estimated using cookies and time constraints as boundaries; system performance may impact this through load balancing and reboots. Data were analyzed for a three-month period; different time spans may change findings. Finally, the pilot study is based on one consumer health Web site and the data may not be representative of navigation behavior due to design, content or audience.