Dementia is a clinical syndrome caused by neurodegeneration of brain tissue. It encompasses a variety of diseases, including Alzheimer’s disease (AD), vascular dementia, Lewy body dementia, and frontotemporal dementia [1]. It is well known that dementia becomes more prevalent with increasing age, and that this increasing prevalence of dementia is significantly affecting the lives of a large and growing number of older adults around the world. Dementia also generates a substantial effect on public health, the social care system, and societal costs [2].

Worldwide, the number of people over the age of 60 is projected to increase to 1.25 billion by 2050, and will account for 22% of the total world population [3]. According to the World Alzheimer Report released by Alzheimer’s Disease International (ADI), nearly 36 million people worldwide are believed to have been living with AD or other dementias in 2013; that number is expected to reach 66 million by 2030 and to increase to more than 115 million by 2050 [4]. ADI reported that overall payments for people aged 65 years and older with dementia—including health care, long-term care, and hospice services—are expected to reach US $203 billion in 2013, increasing to US $1.2 trillion by 2050 [1]. Additionally, the Aging, Demographics, and Memory Study (ADAMS), a nationally representative study taking place in the United States, recently reported that the yearly societal costs in the United States attributable to dementia amounted to US $41,689 per person, with 68% of the total costs stemming from direct costs for health and social care, and the remainder due to the cost of informal care [5].

Informal caregivers have recently become an important target group for long-term care policy making. The Organisation for Economic Co-operation and Development (OECD) and other researchers have suggested that informal caregivers need to learn and obtain more skills for care and more knowledge to avoid the depression and isolation which could potentially lead to more medical utilization [6-8]. However, a challenge that has been raised in much of the literature regarding caregivers has been identifying this group of informal caregivers and determining a way to reach out to them [9,10].

Although there is no known cure for dementia, some pharmaceutical drugs, treatments, and activity interventions may potentially help to improve or maintain the symptoms [11-13]. Thus, it has been suggested that these treatments and interventions be delivered at an earlier stage of dementia, rather than at a later stage, to achieve maximum benefits, including reduced direct medical costs [12,13], delayed institutionalization [3], and maintenance of lower symptom severity levels for longer periods [14,15]. However, current diagnoses tend to be made at a relatively later stage in the course of the disease, partly due to the lack of awareness of symptoms associated with dementia, as well as to the stigma and denial of its existence [16,17]. Under such serious financial burdens on society, all governments and the World Health Organization are set to establish dementia as a public health priority in order to reach out to individuals with early-stage dementia and their caregivers through creative strategies [18].

One of the main challenges of dementia is the lack of appropriate tools and channels for helping individuals identify themselves as having symptoms of dementia, as well as for their family members to quickly examine those they care for. The stigma associated with dementia increases this challenge [17]. A dementia diagnosis is generally perceived negatively by both family members and patients. Stigma and stereotypes are significant obstacles to well-being and quality of life [18-21]. The high rates of Internet searches for health information have suggested that the Internet may be an important resource for patients and family members, who may try to access information for self-diagnosis first, rather than visit their physicians for a more formal diagnosis [22]. Thus, the possibility of reaching out to people with dementia and their family members through the Internet merits further study.

In recent years, the Internet has become a popular medium for people searching for health-related knowledge and information for self-diagnosis [23]. A recent online health survey conducted in the United States reported that in 2012, 35% of US adults looked online to figure out medical conditions, and 72% of Internet users looked online for health information [22]. Of these online health seekers, 77% began at a search engine such as Google or Yahoo. About half of all online health searches are reportedly made on behalf of someone else [22]. These increasing numbers of online searches through search engines creates trend data which can be analyzed in real time [24].

The use of aggregated search queries has considerable potential for syndromic surveillance, as proven by comparing the evidence found in the numbers of clicks on a keyword-triggered link in Google with the epidemiological data from the 2004-2005 flu season in Canada [25]. Since 2004, Google has provided two services for trend analysis: Google Flu Trends (GFT) and Google Trends (GT). In a landmark study published in the journal Nature, GFT has been identified as a powerful tool used in influenza surveillance in the United States, identifying influenza epidemics up to 7-10 days before detection by the Centers for Disease Control and Prevention's (CDC’s) influenza surveillance network [26]. During periods of infectious disease prevalence, GFT has also been able to predict emergency department visit volumes [27,28].

However, GFT failed to correctly estimate the scale of the 2009 H1N1 pandemic in the United States. Several mistakes that led to GFT’s overestimation of H1N1 incidence were caused by the limited transparency of Google’s treatment of data and its dynamic algorithm, due to Google’s business considerations [29,30]. To overcome this, GFT recently used aggregated Google search data in a model created in collaboration with the CDC to estimate influenza activity in the United States, and the results are available in the CDC’s weekly US Influenza Surveillance Report [31].

Following the same logic, GT can be extended to any area for researchers to graph the frequency of searches for a specific term or phrase [24]. These graphs are normalized on a relative basis and can also be restricted to specific time intervals or geographic regions [32]. GT data is available in the United States for city, country, or subnational areas; but for other countries worldwide, including Taiwan, GT data is mostly available on a national basis [24,33]. By tracking health-seeking behavior, GT has predictive capability to monitor the epidemic curve of food-borne illnesses, such as peanut butter-associated outbreaks of Salmonella enterica serotype Typhimurium [34], as well as the incidence of human immunodeficiency virus [19,35].

This tool has more recently been extended to study the relationships among macroeconomic conditions and mental illness. A recent study reported that a 5% rise in the unemployment rate is followed in the next 12 months by an approximate 14% increase in searches around problem drinking [36]. Another study showed that searches for key terms such as “divorce,” “asthma,” and “social welfare” led the suicide death data for 2 months [23]. Additionally, a previous study demonstrated how restaurant table availability has the potential to monitor the incidence of influenza-like illness [37]. Nowadays, this type of research has become a new discipline collectively termed infodemiology [25,38,39]. Based on GT’s algorithm, mechanism, and research findings, it may be expected that what the general public searches for today will have predictive power for what will occur in the near future.

To the best of our knowledge, most previous studies of GT data have been in the field of infectious diseases, and quite a few have been in the field of noncommunicable and chronic disease. Additionally, most of the previous studies took place in the United States. By contrast, such studies in Asia, and in Taiwan in particular, are more rare.

The first objective of this study was to investigate and validate the search terms that could be the best predictors of new dementia cases and outpatient visits. The second objective was to further evaluate the lead pattern of GT search trends related to new dementia cases and outpatient visits. The long-term goal is to contribute to developing a surveillance system for the early detection of dementia in Taiwan.

Only three key terms—“dementia,” “Alzheimer’s disease,” and “neurology”—came above the GT threshold and were recorded in GT. The other 20 terms either did not generate sufficient search volume, and thus were excluded from GT, or were not relevant to dementia. GT reported that searches for “dementia” and “dementia + Alzheimer’s disease” steadily increased throughout the study period. Figures 1 and 2 illustrate the temporal relationship between search term trends and the number of new dementia cases (Figure 1) and outpatient visits (Figure 2). Scatterplots were constructed to compare new dementia cases and outpatient visits with data gathered from GT (see Figures 3 and 4).

From the comparisons of new dementia cases and outpatient visits with search data gathered from GT, Pearson correlation coefficient analysis was conducted.

This study contributes to an understanding of how GT search trends are related to dementia cases and outpatient visits by analyzing data gathered between the years of 2009 and 2011 in Taiwan. Furthermore, this study’s results demonstrate the possibility of monitoring the scale and the temporal and spatial patterns of online search behaviors that may plausibly be associated with future clinical diagnoses of dementia.

Regarding nowcasting capability, findings showed that GT data temporally coincided with the number of new dementia cases and outpatient visits. This finding is consistent with prior research showing that GT provides near-real-time surveillance data up to 7-10 days in advance of CDC’s influenza surveillance network [26]. As for long-term forecasting capability, three search terms—“dementia,” “dementia + Alzheimer’s disease,” and “neurology”—had leading effects on reported new dementia cases and outpatient visits. This finding also confirms previous studies’ results, which suggested that a set of suicide-related search terms, whose trends either temporally coincided with or preceded trends in suicide data, were associated with suicide death [23].

Another important finding in this study is that the search term “dementia” yielded more sensitivity in prediction power among women (with a 6-month leading effect) than among men. This finding may be reasonably explained by a recent study which suggested that women would be more likely than men to report subjective memory complaints (SMCs), which is one of the earliest symptoms observed in the prodromal phase of dementia, whereas men would be more likely to report difficulties or restrictions in instrumental activities of daily living (IADLs), which may become apparent at a later stage of the dementia diagnosis [45].

According to statistics from the Ministry of Health and Welfare Taiwan, in 2012, female patients generated more than 1.55 times the number of mental disorder clinical visits than male patients [46]. This indicates that female patients and their families may possibly take the step of going to see a doctor for complaints earlier than male patients do. In addition, men usually are not expected to fulfill IADLs such as housekeeping, cooking, and washing clothes, so once restrictions in instrumental activities are noted and reported, the dementia will likely be at a later stage [47]. Therefore, we interpreted the results of the search term “dementia” to indicate that families of women with dementia would search online for dementia-related information or knowledge at an earlier stage than would families of men with dementia. Conversely, owing to the stigma and stereotypes of dementia [18], men tended to search for “neurology” instead of “psychiatry” in Taiwan [46]. This may reasonably explain why the search term “neurology” yielded more sensitivity in prediction power among men (with a 3-month leading effect) than among women.

In Taiwan, dementia prevalence at ages 65-69 was 3.4% in 2013 [48]. In Taiwan in 2015, according to a survey by ClickForce [49], just 5.2% of adults aged 55 and over had at least one experience of searching on the Internet. Therefore, in this study we assumed that most searches were initiated by family members, whether before or after patient diagnosis. In addition, we anticipated that searches originating from dementia patients themselves would reflect concurrently in a no lag period after their outpatient visits, rather than precede a visit or a new diagnosis.

Most people in Taiwan have the basic knowledge to understand disease-related divisions in hospitals, such as neurology, and this is believed to be related to the increasing tendency for patients and family members to use Internet searches for health information prior to seeing a doctor. Such search trends can now be applied as leading indicators which are capable of predicting disease incidence and related increases in outpatient visits earlier than conventional surveillance [22,24,32].

In comparison with conventional surveillance methods, GT has several advantages. First, GT is a free and easily accessible search tool. Second, GT is now available for more than 70 countries worldwide, and thus allows for tracking of searches in different languages and regions. Last, GT is updated weekly, which permits more frequent monitoring of symptoms and could facilitate early detection of dementia. In this regard, GT can be used to notify health policy makers and hospital managers to get ready for upcoming outpatient visits, where screening work is expected to achieve early detection and treatment and thus save more medical and care costs [50,51].

Additionally, from a public health point of view, health policy makers and health care professionals may consider using key search terms to inversely feed dementia-related information, such as health care resources, care skills, and social welfare information, to those formal and informal caregivers who are difficult to identify in the dementia care system. It is possible to embed dementia-related keywords in online articles, where they can be easily found and recorded by Googlebot—Google’s Web-crawling bot that discovers new and updated pages to be added to the Google index—and thus delivered to users who search for these keywords on the Internet [52]. These users are likely to be the people conducting dementia-related searches on behalf of someone they care for. Using these inverse search strategies to deliver information may improve various aspects of caregivers’ well-being, such as increasing confidence and reducing depression, as well as lowering the potential utilization of medical and health care services by caregivers of persons with dementia [6-8].

Our research suggests the need for the following further research. First, further qualitative research is needed to validate the results of this study. In particular, an examination of the Internet search behaviors of caregivers for people diagnosed with dementia is necessary. Also, the gender difference reported in the argument on early report of SMC by female patients needs validation [45]. Second, additional research at different geographic scales would be worthwhile, in places where GT offers complete subnational data points. Third, aside from GT, Google AdWords and Baidu trends also have the potential to play similar roles in the surveillance of dementia, and research on these tools merit further study. Lastly, more dementia-related keywords and a longer period of data points merit further study to validate GT’s lead pattern. This study contributes to the literature by expanding the GT research field to the disease of dementia. To the best of our knowledge, no existing empirical research addresses the relationship between GT and dementia or ways of estimating potential medical needs for people with Alzheimer's disease.

Several limitations of this study should be noted, mostly related to the GT algorithm. First, owing to insufficient and incomplete data sources, only 3 years of GT data were included in this study. Future research with longer periods of data points is recommended to validate the lead patterns. Second, there was difficulty in identifying search trends that were generated by true cases. In particular, GT tends to be influenced by media exposure of specific diseases (eg, drug advertisements), which drives more nonpatients to search related terms and thus increases the search volumes [53]. Lastly, the calculation of GT depends on Google’s assumptions and normalization, which are not clearly reported.

In conclusion, GT serves as an easily accessible, real-time surveillance tool. Despite the limitations, this study highlights GT as a useful tool for establishing a plausible relationship between search terms and new dementia cases and dementia-related outpatient visits. This tool can lead to better management of medical resources and budgets in Taiwan’s health and social care system. It is possible that GT could be further developed as a surveillance platform to help lower the individual and social costs of this troublesome disease associated with aging populations.