Search-based infodemiology

It provides a signal of a population's interest, concern or health awareness across time and place: how much relative search volume a condition attracts, when and where. This applies when you want to capture the dimension of public perception that healthcare records do not collect. Notified incidence measures cases seen by health services; search volume measures something else, and something complementary: how much attention, worry or lack of knowledge there is about a health topic among people who search online. In some settings that signal can run ahead of demand for care, because people search before they seek help, and it reflects concerns that never reach the clinic. It is worth putting it in its place from the start: it does not measure disease, it measures search interest, and it represents people who search online, not necessarily the whole population. Within that scope, it is a layer of information that conventional surveillance does not see.

No. It is a complementary layer, not a substitute for conventional epidemiological surveillance or for official records. This is a clarification of scope, because "search-based surveillance" can be read as an alternative to traditional systems, and it is not one. Conventional surveillance measures confirmed cases, with their clinical validity; search analysis adds the dimension of interest and attention, which records do not capture, and does so with an accessible, low-cost source. The value lies in the combination: the official record says how many cases there are; the search signal approximates which topics arouse interest or concern among people who search online, when that interest runs ahead of demand for care or accompanies it. Using it as a replacement for traditional surveillance would be a mistake, because it lacks that diagnostic validity; using it as an added layer covers an angle that traditional surveillance does not see.

The temporal pattern is assessed with formal statistical methods, not inferred from the curve "looking" cyclical. This applies to any search time series, where visual inspection suggests patterns that may not hold up. The stationarity of the series is assessed with tests such as KPSS and augmented Dickey-Fuller, and seasonality is analysed with dedicated methods for decomposition, time series modelling or testing recurrent patterns; the underlying trend is also estimated by regression. That way a temporal pattern rests on statistical evidence, not on visual inspection alone. This separates real seasonality, which recurs and withstands the analysis, from a fluctuation that only looked like it. The difference matters for the decision: a health education campaign scheduled around a seasonal pattern needs that pattern to be robust, not a coincidence of one particular year.

Not necessarily. A lower search volume may reflect less concern or, quite simply, less internet access, and confusing the two leads to wrong conclusions. This applies to any geographical comparison of search interest between countries or regions. That is why the comparison is not made on raw figures: it is stratified by development level, using UNDP indices, and the digital divide is built into the interpretation, so that what is really unequal access is not read as lack of interest. Comparisons between regions use non-parametric tests suited to the nature of the data. The result is a geographical comparison that helps interpret whether the observed difference may be due to interest, to access or to a combination of both, instead of penalising less connected populations as if less connectivity necessarily meant less health concern.

Aggregated relative search volume data are available through public tools such as Google Trends, which provide normalised RSV by topic, language and region, in aggregated, anonymous form. This applies to the whole service, which relies on an open source rather than on client data. That means you do not need to provide a database: what we need to define with you is the question: which condition or risk factor you are interested in, in which area and over what period. The definition of the search topics depends on it. That definition is a sensitive technical step: it draws on thesauri such as DeCS and MeSH to capture synonyms and cross-language equivalents, so that the search reflects how people actually name a condition, not just its technical term. The service extracts the RSV, checks the consistency of the downloads and of the term definitions, and builds the series; your role is to bring knowledge of the health issue and the target population, and to decide on and implement any resulting action.

Yes, it has known biases, and the analysis documents them and limits the conclusion to what the data represent. Search data reflect the aggregated behaviour of people using a particular platform, not the whole population: the digital divide leaves out people who do not search online or do not use that route for health topics. The volume is relative and normalised rather than a count of people, and platforms may suppress low-volume terms. News coverage, interface changes and algorithm updates can also move the signal without any underlying epidemiological change. We therefore document the platform, period, search terms and populations not represented, and compare spikes with other sources when those are available. Digital epidemiology works well as an early signal or as a complement to conventional surveillance, not as a representative prevalence estimate.

They differ in what their data capture. Search-based infodemiology analyses aggregated, anonymous search volume (what a population searches for about health, when and where) as a syndromic epidemiological signal, using time series methods. Social media and digital infodemiology analyses the conversation on platforms (what is said, how it spreads, with which narratives and what sentiment) using text and network analysis methods. They do not answer the same question or work with the same type of data: one measures search interest through trends, seasonality and geographical comparison; the other studies the content and circulation of messages. If the question is how a population's interest in or concern about a health topic is evolving, it is search-based infodemiology; if it is what is being said and how a message or narrative circulates, it is social media infodemiology. They are sister services covering different sides of the same information phenomenon.

You receive the analysis of how search interest in the defined health topic has evolved: the time series built from RSV, the estimated underlying trend, the seasonality analysis, the geographical comparison stratified by development level and explicit documentation of the sample's scope. This is for those who need to complement their surveillance with the interest or concern of people searching online, or to target education and communication campaigns at where and when the population searches for, or is unaware of, a condition. The deliverable records the definition of the search topics, the period and the geographical scope, and which population the data represent and which they leave out, so that the reading is traceable and stays within what the source allows you to claim. The search data are publicly available and the service extracts, analyses and interprets them; the health authority or the team in charge decides on and implements any resulting campaign or action.