BMJ  2006;333:1131 (2 December), doi:10.1136/bmj.39044.369745.BE

Editorials

Diagnosis using search engines

Probably heralds a much more sophisticated web resource

Thousands of computer systems targeted at medical diagnosis (often described as expert systems) have been developed during the past 50 years. Most have had relatively little impact on day to day clinical practice; for example, because they are not easily accessible at the point of care; have a complex interface; can deal with only a narrow focus (one symptom or clinical problem); are not integrated with clinical information systems; depend on particular software or hardware platforms; or require labour intensive construction and are therefore expensive to maintain and extend.

In this week's BMJ a study by Tang and Ng assesses the effectiveness of a web search engine (Google) as a diagnostic aid.¹ Using general purpose web search engines as a diagnostic aid is new, although using computers to aid diagnosis is not. The article has provoked a widespread response, with one daily newspaper concluding that "Researchers have found that a simple Google search can solve diagnostic problems which mystify even the best general practitioners."²

This conclusion is misleading in at least three respects. Firstly, the Google searches reported in the study were not simple. Although the search required typing only a few words and clicking a button, doctors used extensive knowledge and experience to choose the search terms effectively in the first place. Secondly, Google did not solve the diagnostic problems. It returned a set of documents ranked according to their relation to the search terms, using metrics essentially based on counting the occurrence of word stems. Inferring a diagnosis from the returned documents depended on the skill of the doctors. Thirdly, the implication that diagnostic problems can be ordered on a linear scale of difficulty, against which human and machine performance can be compared, is unwarranted. The study shows that in a certain class of diagnostic challenges—perhaps not a large class given current web technology—experts may gain an advantage by including a web search in their deliberations.

Current search engines offer unprecedented accessibility at no direct cost to the enquirer, a natural all purpose user interface, and an immediate return of information on almost any topic. Most impressive of all is the volume of new material that is indexed automatically every day. However, the current web is based on documents in HTML (hypertext markup language) format, and consequently has two serious limitations. Firstly, search engines cannot automatically index resources in other formats; for example, information in relational databases. Secondly, HTML indicates only how a document should be rendered, not what it means. Search engines cannot identify which words in a document represent symptoms, diagnoses, drug names, or parts of the body, let alone reason about these concepts.

In fewer than 10 years of common use the web has already had a considerable impact on clinical medicine. Some would argue, however, that current web technologies are merely a stepping stone towards a much more powerful infrastructure—the semantic web—which aims to create a universal medium for information exchange by putting documents with computer processeable meaning on the worldwide web. The semantic web can overcome current limitations and offer much stronger support for tasks such as medical diagnosis.³

Work towards the development of the semantic web is already under way. Two key technologies are RDF (resource description framework) and OWL (web ontology language).⁴ RDF is designed to allow future web search engines to access data and resources in addition to HTML documents. OWL is an international standard language for building web enabled ontologies—that is, domain knowledge that can be downloaded freely and used by computer systems such as search engines.

Web enabled medical ontologies and the semantic web will be important resources for clinical medicine in the next decade. Special interest groups have been established (for example, W3C Semantic Web Health Care and Life Sciences Interest Group; www.w3.org/2001/sw/hcls/) and several targeted ontologies have been constructed (for example, the National Cancer Institute thesaurus; www.mindswap.org/2003/CancerOntology/), but a much wider coordinated effort is needed.

No clinician diagnosing patients should feel insecure about the findings reported by Tang and Ng. Their study should encourage us, however, to think about strategies for maximising the benefits in daily clinical practice of the evolving semantic web.

¹ Department of Computing Science, University of Glasgow, Glasgow G12 8QQ

---

Competing interests: None declared.

References

1. Tang H, Ng JHK. Googling for a diagnosis—use of Google as a diagnostic aid: internet based study. BMJ 2006 doi: 10.1136/bmj.39003.640567.AE
2. Researchers impressed by Google's medical diagnoses. Independent 10 November 2006.
3. Berners-Lee T, Hendler J, Lassila O. The semantic web. Sci Am 2001;May:34-43.
4. Shadbolt N, Hall W, Berners-Lee T. The semantic web revisited. IEEE Intelligent Syst 2006;21:96-101.

CiteULike    Complore    Connotea    Del.icio.us    Digg    Reddit    Technorati    What's this?

Related Article

Googling for a diagnosis—use of Google as a diagnostic aid: internet based study

Hangwi Tang and Jennifer Hwee Kwoon Ng
BMJ 2006 333: 1143-1145. [Abstract] [Full Text] [PDF]

Rapid Responses:

Read all Rapid Responses

It helps me!!

Sherin Jos Payyappilly
bmj.com, 4 Dec 2006 [Full text]

Accuracy of internet searches

Andrew J Larner
bmj.com, 5 Dec 2006 [Full text]