Abstract
Recent reports have suggested that variations in the volume of Internet searches relating to suicide, and risk factors for suicide such as depression and divorce, are associated with population suicide rates [1,2]. Furthermore, search activity in relation to specific methods of suicide have been shown to mirror high profile media reporting of an unusual method of suicide in the UK and Japan [3]. Seasonal variations in the volume of Internet searches relating to depression, mirroring seasonal fluctuations in the incidence of depression, have also been noted [4]. There is perhaps the potential to use trends in Internet searching relating to suicide as a kind of barometer of actual suicidal ideation and behaviour in a community [3], in the same way as it has been suggested it may be useful in identifying the onset of epidemics of infectious disease [5]. Readily available Internet resources may be a way of augmenting routinely collected sources of mortality, hospital admissions and primary care sector data (the availability of which is often lagged by a number of years) as part of a strategy of syndromic surveillance of levels of suicidal behaviour and psychological distress in communities.
Surveillance of Australian Suicidal Behaviour Using the Internet?
We investigated whether internet searches using Google in Australia relating to ‘ways to commit suicide’ showed any seasonal trends (similar to those shown in previous studies of suicidal behaviour in Australia [6]) or were related to unemployment rates (a known correlate of suicidal behaviour in the Australian population [7,8]) using Unobserved Component Models (UCM) [9] (Figure 1). Monthly data on suicide search terms for the period February 2004 to March 2011 were obtained from Google Insight (http://www.google.com/insights/search) using previously identified search terms relating to suicidal behaviour [10], which included: ‘how to commit suicide’, ‘ways to kill yourself’, ‘suicide pact’, or ‘suicide hanging’. Trends in these search terms were restricted to Australia, and entered individually and in combination. Google does not provide absolute numbers of searches but a relative figure based on search activity for the study period, normalised to a scale of 100. The month in the selected period with the highest number of searches is assigned the value 100, and other months are scaled accordingly.
Trends in suicide-related search terms in Google and unemployment in Australia. Suicide search terms included: ‘how to commit suicide’, ‘ways to kill yourself’, ‘suicide pact’, or ‘suicide hanging’. Trends in these search terms were restricted to Australia, and entered individually and in combination. Combined searches were pooled to obtain a single index by summing across search terms and re-scaling to 100 by dividing each monthly scaled score by the maximum monthly scaled score for the study period. ‘Observed’ refers to observed scaled Google search frequencies (maximum is 100). ‘Expected’ refers to expected scaled Google search frequencies derived from the fitted unobserved component model adjusting for slope, seasonality, unemployment and suicide pacts (dotted lines show 95% confidence intervals). Monthly unemployment rates were obtained from the Australian Bureau of Statistics (ABS) [18].
There were two sharp spikes in searches using these terms over the period, relating to two high profile reported suicide pacts in Australia: (i) the case of two 16-year old girls who hanged themselves together in the Dandenong Ranges National Park (Victoria, Australia) in April 2007 [11], and (ii) the case of Australian identical twins who simultaneously shot themselves at a Colorado (United States) firing range in November 2010 [12]. An additional peak was also evident in October 2005, corresponding with the series of simultaneous suicide bombings in Bali (Indonesia), which was associated with widespread media coverage in Australia [13]. There was weak evidence for an association of monthly percentage increases in unemployment with increases in Google searches (β = 0.26, t = 1.8, P = 0.077 from model). However, there was no evidence for substantial upward or downward trends or any seasonal patterns in Google searches using suicide terms over the period. Additional models investigating non-seasonal periodicity identified two cycles in suicide search terms, one recurring at an approximate 9-month interval and the other at a 17-month interval; the significance of these is uncertain.
No Straightforward Indicator
Our findings indicate that trends in Internet searches of suicide terms using Google are not a sufficiently straightforward indicator of the levels of suicidal behaviour in Australia, and showed no seasonality, and limited evidence for an association with unemployment trends. However, analyses could not investigate corresponding monthly counts of suicide given the current lag in availability of mortality data in Australia. A contemporaneous rise in suicide search terms corresponding with the sharp rise in Australian unemployment (February-March, 2009) might have been expected during the period of the global financial crisis given previous studies showing increased suicide rates during periods of economic crisis in Australia [14] and internationally [15]. Explanations of the observed non-seasonal periodicity in searching would need to be linked to plausible population-level exposures associated with suicidal behaviour. The strong associations between high-profile media reports and suicide search terms may have more utility in understanding patterns of suicide clusters and the effects of media coverage on ‘copy-cat’ suicides [16,17], than in understanding population-levels of suicidal behaviour in Australia.