Between Web search engines and artificial intelligence: what side is shown in laboratory tests?

Introduction

The use of Internet as a source of health-related information has increased greatly in recent years, largely due to the improvement achieved in the performance of search engine service algorithms and the widespread availability of portable devices (e.g. smartphones) that allow access to search engines and Web pages [1]. Moreover, it is now possible to create new Web contents more easily and quickly, so not only large (private or public) institutions and editorial groups, but also single users, can generate professional-like websites, providing engaging contents. Interestingly, the rapid growth of online Web pages and resources providing health-related information has not gone hand in hand with regulatory updates, and the quality of their content is often controlled only by their authors. In 2010, Tonsaker and colleagues reported that for many of the 70% of Canadians who searched health-related information up online, the Internet, rather than their physician, was the first source of medical information [2]. In the United Kingdom, in 2015, more than 20% of people chose to diagnose their own diseases by using Internet resources [3]. Individuals who experience increased arousal when viewing online medical information may be at risk for developing cyberchondria, a phenomenon in which repeated Internet searches regarding medical information result in excessive concerns about physical health [4]. Interestingly, cyberchondria was found correlated with health anxiety and with a decreased quality of life [4].

Patients searching online for medical answers may improve their engagement in the management of their diseases in synergy with their physician [5], but sometimes it is difficult to check the quality and accuracy of the information they obtain. Any quality check of online health information is a challenge for Internet users, considering that results often include non-documented and unreliable information or, in the worst case scenario, are based on unscientifically founded, or harmful, health practices [2], especially nowadays, when “fake news” can become viral thanks to the Internet and its social media platforms [6].

Therefore, public institutions, scientific societies and large editorial groups, which usually implement a scientific revision process before publishing online, face a growing number of personal blogs, small private laboratories’ websites and product sellers’ online shops, in which information is not always properly controlled and reviewed.

One of the most important organisations aiming to help users identify the quality of information provided, the Health On the Net Foundation (Chêne-Bourg, Switzerland), a non-profit organisation, promotes transparent and reliable health information over the Internet [7]; many major/important health-related sites undergo the Health on the Net Foundation Code of Conduct (HONcode) certification process to prove that their content is reliable and validated by health professionals. Yet, not all Internet users understand the difference between a certified and a clickbait-driven website.

Most of the laboratory medicine sources of information available online are from public or private institutions, laboratories and publishing groups, while others are from patients’ associations, foundations for research promotion or scientific societies. These resources provide a wide range of information, explaining, for example, the meaning of laboratory tests, how to get ready for them, how to interpret their results, etc.

It is now possible, by means of a Google Web Search (Alphabet Inc., Mountain View, CA, USA) flow, to identify about 75% of the total online searches [8], and users’ behaviour data can be retrieved, aggregated and anonymised, using Google Trends public service tools, already employed in many different ways for biomedical research [9]. Google Web Search, as its market competitors, is constantly increasing its use of artificial intelligence and deep learning algorithms to achieve constant improvement in results and to “rank” them (to define their order of appearance in the search results page) while considering an ever increasing number of variables and conditions [10].

The aim of the present study was to assess the quality of online retrievable information in the Italian language by investigating the Google Web Search engine for laboratory medicine tests, and analysing Web search results of three example tests: urinalysis, cholesterol and prostate-specific antigen (PSA). These three tests were chosen because they are widely known by the general public, and commonly performed in clinical laboratories [11], [12], [13].

Materials and methods

Results

Web search evaluation using Google Trends

The first parameter considered in Google Trends results was “interest over time” in the chosen laboratory tests. As reported by Google, the numbers obtained represent search interest relative to the highest point (always set to 100) on the chart for the given region and time (a value of 100 represents peak popularity for the term, and a value of 50 means that the term is half as popular) [14].

In Figure 1, which reports Google Trends data, an increase in the mean “interest over time” value is identifiable for all the considered laboratory tests. Comparing the first 3 months of 2014 and first 3 months of 2019, PSA-related searches had the highest increase in “interest over time”, with an increase in the number of searches of 25%; cholesterol had an increase of 19% and urinalysis, an increase of 15%. The number of searches showed a seasonal trend, with minimums in the middle of August and in the last week of December (Christmas period), for all the tests considered.

Figure 1:

Google Trends “interest over time”, from April 16, 2014 to April 15, 2019.

The first five “related queries” found for urinalysis were (in brackets the English translation): “analisi delle urine” (“urinalysis”), “analisi urine” (“urine analysis”), “esame delle urine” (“examination of urine”), “esami delle urine” (“examinations of urine”), “esami urine” (“urine tests”). For cholesterol, related queries were: “colesterolo” (“cholesterol”), “colesterolo alto” (“high/elevated cholesterol”), “colesterolo hdl” (“hdl cholesterol”), “hdl” (“hdl”), “il colesterolo” (“the cholesterol”). For PSA, related queries were: “prostata psa” (“prostate psa”), “psa” (“psa”), “psa libero” (“free psa”), “psa totale” (“total psa”), “valori psa” (“psa values”).

Discussion

The fact that nowadays patients look for health information online is well established [2], [3], and the trend has increased worldwide, thanks also to groups of patients’ forums, and enthusiasts’ blogs, which are easily accessible through Web search engines, addressing the user searches on the Web. Every Web search engine crawls all the pages it receives (or finds by surfing the Web on following links) and creates an index [15]. To provide results for each query (the text corresponding to the user’s question), search engine uses different algorithms in order to rank all possible results available in its index, and then displays a list of Web pages ordered by “relevance”, the first result being considered the one most relevant for the user, and the query made [16]. Then the user can click (or tap, depending on the type of device used) on the results to directly navigate to the Web page. In view of the variety of search engines and their back-end algorithms, we chose the Google Web Search, the engine most widely used by the population in Italy and in many other countries [8].

The first element we analysed was the “interest over time” trend. The overall increase of “interest” in these arguments can be considered a confirmation of the growing number of individuals searching for health-related information on the Web, probably linked to the increase in health-related searches by already connected users, but also the spread of Internet access to new users, who now find it easier than ever to surf the Web: there is greater availability of mobile devices with fast connections and more user-friendly interfaces, where Web search engines have a strong presence. The PSA test had the greatest growth in trend in the latest 5 years, perhaps because the test itself is younger than the other two tests, which are well established in laboratories. Regarding seasonality in the “interest over time”, the reduction in the number of searches in August and in the last week of December appears to correspond to a reduction in the number of requests for the three tests in laboratory routine.

In the second part of the investigation, we identified the owners or publishers of the Web contents in the search results (Table 1). In particular, from Table 1, the mean percentages of results (considering all three tests) from Web search of pages owned by editorial and publishing groups were 35.5% and 30.0% of the first three and the first 10 results, respectively; the mean percentages of results (considering all three tests) from Web search owned by private hospitals and laboratories were 28.9% and 34.7% of the first three and the first 10 results, respectively. Public health agencies and scientific societies were displayed only rarely when the first three results were considered, and accounting for only 2–10% on considering the first 10 results.

Furthermore, we looked at the Web pages in the search engine results in order to ascertain how many fulfil some basic quality features, such as source citation or scientific revision processes of the contents (Table 2). Most of the Web contents were written by health professionals and included cited sources, while HONcode certificate was not frequently used. Interestingly, HONcode represents a certification that focuses on the reliability and credibility of health and medical online information and sets rules for website to (a) hold basic ethical standard in the presentation of information and (b) facilitate readers to know the source of the information [7].

Finally, this study presents some specific limitations, such as the possibility that patients use keywords different from those initially queried and the utilization of a single Web search engine for obtaining information on online laboratory tests.

In conclusion, although we evaluated a small number of clinical laboratory tests, and did not analyse trends and results on social media or video sharing platforms, our findings confirm those made in other studies in the literature, demonstrating that online Web searches can lead patients to inadequately written or reviewed health information. Web search engines are providing results to an ever increasing number of users, but it is hoped that artificial intelligence and deep learning applications in Web search algorithms will give ever better and ever more accurate results [17], higher ranking being conferred to contents with cited sources, and those which undergo scientific revision by professionals.