Web searches

Considering differences in the use of search engines, languages and organization of health systems between countries, we only analyzed searches done in English from the US, using aggregated data for 2015. By focusing in only in the US we were able to obtain more homogenous results at the prize of losing generalization to other Western countries and assuming the bias to higher-educated population. All data used in this study was obtained from public advertising interfaces (APIs) available in the two major search engines, Google and Bing/Yahoo. No user identification procedure was performed at any time. Therefore, informed consent was not requested from users as we used aggregated results. For the same reason, we could not do session analysis, in which we would track the sequence of requests of each user (e.g. first searching a symptom such as “weakness”, then for an illness such as “multiple sclerosis”, then a drug such as “interferon beta” or an indication of disability such as “wheelchair”). For this reason, we focused in more precise terminology such as disease diagnosis and did not performed more general, common or vague terms such as “pain”, which prevented the analysis by symptoms. The ethical committee of Hospital Clinic of Barcelona, Spain, approved this study.

Databases

We used the list of diseases from the National Institute of Neurological Disorders and Stroke (http://www.ninds.nih.gov/disorders), which provided a list of 448 neurological diseases and enriched the list with its common disease abbreviations.

In addition, we created the list of FDA approved drugs for neurological diseases (including pain) from Centerwatch (https://www.centerwatch.com) and Medilexicom (http://www.medilexicon.com) including the generic and brand name. Because some commonly used drugs in Neurology are old (e.g. phenobarbital) or may have not been approved for these indications (off label use), we enriched the list of drugs with the list of drugs reimbursed by Medicare for neurological diseases as described in[15], leading a list of 212 drugs (see S1 File for the full list of term searches used).

Analysis

We evaluated people’s interest in brain diseases by using the volume of search terms related to neurology, either disease diagnosis or CNS drugs. That is, we considered the frequency or volume of a query (or group of queries that have the same meaning) as a good proxy for the interest of people. Similarly, we used the real-time bids for web queries by advertisers to evaluate the interests of the healthcare industry. In search advertising, a second price auction is performed for each query, where for each click the advertiser pays what the next advertiser is willing to pay. Therefore, the bid value is a good proxy for the interest of the industry (advertiser) for a given query. Hence, the total value of the market for that query, spend, is proportional to the query volume multiplied by the bid. This market value can be estimated from different search engines and will give different results due to their different number of users, the number of advertisers, and their technology (e.g. how the bid is suggested to a new advertiser; or how polysemy of terms is handled). For our analysis, we use the standard advertising APIs of Google and Bing/Yahoo restricted to the US market, as they are the major search engines nowadays. For the case of Google, we also include a competition factor that estimates the competitiveness for the market of each query, or in other words the number of companies interested in a given term (https://support.google.com/adwords/answer/3022575?hl=en).

Frequency of neurological diseases in web searches

We evaluated the interest of people in neurological diseases by using the volume of search terms related to neurological diseases (full name or common abbreviations) in the two most important search engines Google and Bing/Yahoo (S1 File). We found that the most searched neurological diseases (Fig 1A) were highly prevalent diseases such as Dementia or Attention Deficit/Hyperactivity Disorder (ADHD), as well as medium frequency diseases with high social impact such as Parkinson disease (PD), Multiple Sclerosis (MS) or Amyotrophic Lateral Sclerosis (ALS) (S2 File). Surprisingly, the two most prevalent neurological diseases, headache and stroke, were not ranked in the top 5 or 10, but they were in the top 30.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Average monthly searches and bid price of neurological diseases and CNS drugs.

A) Figure shows the top 15 diseases (blue bars) and its suggested ad bid (green bars, in USD) searched in the Bing/Yahoo and Google search engines. Full names and well accepted acronyms were used for the analysis. Abbreviations: ADHD: Attention-Deficit/Hyperactivity Disorder; MS: Multiple Sclerosis; PSP: Progressive Supranuclear Palsy; AD: Alzheimer Disease; IBM: Inclusion-Body Myositis; ALS: Amyotrophic Lateral Sclerosis. B) Figure shows the top 15 drugs with their indication (blue bars) and suggested ad bid (green bars, in USD) searched in the Bing/Yahoo and Google search engines.

https://doi.org/10.1371/journal.pone.0178019.g001

Moreover, highly ranked searches included polysemic terms such as “whiplash” or “shingles”. Both engines handle polysemic terms in a different way, which can explain some differences in the frequency of searches. In our queries, we included the name of the disease as well as common disease abbreviations. However, abbreviations can refer to different terms. Nevertheless, the effects of such abbreviations were not significant because they did not modify the ranking of a given condition, with three exceptions: IBM (Inflammatory Body Myositis vs IBM corporation), PSP (Progressive Supranuclear Palsy vs PlayStation Portable) and TM (Transverse Myelitis vs Trademark) (S2 File).

Second, we analyzed the bids for queries by advertisers to evaluate the interest of the healthcare industry in neurological diseases (S2 File). Although the cost of each ad varies among search engines, the bid price is not related to frequency of the disease nor based solely on the frequency of the searches, but also reflects market value in the health sector[16]. Neurological diseases ranking higher in the bids included ADHD, PD, Alzheimer’s disease, MS and meningitis (Fig 1A).

Frequency of CNS drugs searches

We evaluated the searches for drugs for neurological indications (S2 File). The most commonly searched drugs were generic drugs used for pain, followed by sleep disorders, dementia, ADHD, stroke, or PD (Fig 1B). Moreover, drugs that were approved more recently also ranked high considering their prescription rates compared to those of the massively used drugs described above, such as the case of immunomodulatory drugs for MS, and several anti-epileptics and drugs for PD (S3 File). Regarding the bids for queries by advertisers for CNS drugs, Over The Counter (OTC) drugs including painkillers, drugs for headache, epilepsy and insomnia ranked the highest (Fig 1B).

Neurology’s market analysis based in web searches

We analyzed the relation between query volume and query bid (value per click) for drugs and diseases for both search engines (Fig 2A–2D). ADHD, Alzheimer disease, MS, ALS, meningitis, and hypersomnia ranked high in both dimensions. Regarding CNS drugs, painkillers, triptans and drugs for neuropathic pain, dementia or insomnia scored the highest in both dimensions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relationship of query volume and advertising expenditure (ad bid) for CNS drugs and diseases searches.

A-D) show the average monthly searches (Y axis) for CNS drugs or diseases and the suggested ad bid (X axis, in log scale) for either drugs or diseases for searches in Google or Bing/Yahoo. In each graph, queries that are in the top represent the people’s interest (searches) while the queries that are in the right side represent the health industry’s interest (suggested ad bid). E-H) show data for Google (g) or Bing/Yahoo (b) in log scale for searches of neurological diseases. Comparisons are done using the total search spend (E), the Google competition against bid (see Methods for definition) (F), the average monthly searches for Google and Bing/Yahoo (G), and the average monthly searches against the advertising spend for Google (H).

https://doi.org/10.1371/journal.pone.0178019.g002

In addition, we analyzed the correlation of potential advertisement spend among search engines, the correlation of volume among search engines, the correlation of Google’s competition factor and suggested bid, and the correlation of Google’s volume and spend (Fig 2E–2H). The correlation was high for all of them, except for competition against bid, as the competition factor depends on the number of advertisers for each query. The spend (volume * bid) correlated in both search engines (Fig 2E), the competition between advertisers and bid showed a lower correlation for Google (Fig 2F), reflecting that the competition is for the most expensive drugs. Average searches were highly correlated between search engines, with some exceptions for Google due to its dominant position (Fig 2G). As expected, correlation between spend and volume of searches was high for Google, with some exceptions where the volume was significantly higher than the spend (dots above the correlation line, Fig 2H), because not all diseases or drugs have the same interest for industry.