How can we improve the quality of health searches ?

A friend was doing a search on the causes of high bilirubin levels in adults. He found a lot of garbage – mostly bulletin board postings and anecdotal case reports. He was also swamped by a lot of irrelevant results about neonatal jaundice.

I repeated the search using the same search engine, but using a better keyword – “ hyperbilirubinemia “ – and this improved the results dramatically.

Searching for “ adult hyperbilirubinemia differential diagnosis “ gave the best results . The problem was solved – and took just a few minutes of an expert’s time, and saved the user hours of frustration !

1. Health and medical information is the No 2 reason people use search engines
2. Many users are frustrated with the results of the searches they get

There are two groups of reasons for this.

1. The search engines. These are not clever enough to separate the wheat from the chaff, which means they deliver too many results, many of which are poor quality; or irrelevant.
2. The searchers. Many are naïve and don’t know how to do a search
properly . They make spelling errors; are unfamiliar with medical terms; and cannot evaluate the quality of the information they find.

This means there is a lot of searching – but very little finding !

What do users want when they do a search ?

They want results which are instant; reliable; relevant; understandable – and free ! While present search engines offer instant searches which are free, they often fail to provide reliable and relevant results, because Garbage In is Garbage Out.

So how can we improve the quality of health searches ? Most of the focus so far has been on making the search engines better. One approach would be to automatically provide the search engine with better inputs. For example, searches could be personalized using context sensitive information from the user’s a personal health record ( PHR) .If the user is searching for diabetes, is it a 12 year old ( in which case you should be looking for juvenile diabetes); or a 60 year old ? It is this ability to provide context that we pay doctors so much money. They specialize in being able to quickly apply a subset of their medical knowledgebase to individual patients, based on : the patient; and their clinical experience.

We need to start by trying to create a taxonomy of health searches . I don’t think this has been done. It would be easy to do – and would provide interesting insights !
For example, health searches can be classified as
Pre Diagnosis/ Post diagnosis
Simple/ complex
Well defined/ poorly defined
With / without spelling mistakes
With / without medical terms

It would also help to know why the search was being done. Is the user looking for specific information and just needs background knowledge of a particular illness ? Or is this a comprehensive search to look for elusive nuggets of information ? This first step will give us a better idea of what terms people use today to find information; and why.

Give the fact that there is so much information on the web, this means that it’s virtually a certainty that the information the user wants is there on the web – it’s just that the user cannot get to it. This can be very frustrating.

The holy grail would be to use a semantic search engine which had access to information from the searcher’s personal health record ( PHR ), so that the information is customized and filtered) ; and then to use artificial intelligence to display the latest , most uptodate information only from selected reliable websites.

Till we reach this stage, can we do anything to improve the user who is doing the searching ? One way of doing this would be to provide tutorials or help functions, but the fact remains that most users do not use these well at all.

I am interested in exploring whether using a human intermediary ( a doctor or nurse or librarian in India, for example) to answer questions would help improve searches.

The model would be to charge a micro-fee ( say $ 1 for a search) ; guarantee an answer
( to questions selected by the infomediary only) in 24 hours ; and allow the infomediary to select only the easy questions to answer ( those which were well-formed and well-defined). This is likely to help a significant chunk of users ( who only need some hand-holding) .

The expert human infomediary has the ability to use the search engine intelligently to provide much better results ! For example, he knows the importance of iteration to refine results ; and how to evaluate the quality of the websites found. This can be very cost effective, because these answers not only save time ; they also provide reliable information ! He also knows when to search the “deep web” ; what specialized databases can be tapped; and which searches are unlikely to provide useful results !

Searches done by the infomediary could also serve to provide feedback to the search engine algorithms, so they could be continually refined and polished.