Comparison of epidemiologic surveillance and Google Trends data on asthma and allergic rhinitis in England.

Abstract

Epidemiologic surveillance is the ‘ongoing systematic collection, analysis, and implementation of health data essential to the planning, implementation and evaluation of public health practices, closely integrated with the timely dissemination of these data’.1 Recently, the use of data from internet users’ activity—measured with tools such as Google Trends (GTs)—has been proposed to complement ‘classical’ surveillance approaches.2 In this context, infectious diseases have been widely studied,2 but allergic diseases have not, as they are less often targeted by surveillance programmes. In England, however, the Oxford-Royal College of General Practitioners Research Surveillance Centre (RSC) monitors asthma, allergic rhinitis (AR) and respiratory infections, publishing weekly surveillance reports on their incidence. Such reports are based on ‘the number of patients [of all ages] consulting [a set of regionally representative primary care centres in England] with new episodes of illness’, differentiating incident from follow-up consultations3; methods for ensuring data quality are routinely applied.3 English general practice is a highly computerized registration-based system, with computerized medical records following the patient when they move practice, ensuring a reliable rate of disease to be reported. In this study, we assessed the correlation of English surveillance and GTs data on asthma, rhinitis and upper respiratory tract infections (URTI) to assess whether similar patterns can be observed with these two different approaches. We retrieved weekly surveillance data from the RSC reports on the incidence in England per 100,000 inhabitants of ‘AR’, ‘asthma’, ‘URTI’ and ‘common cold’ (a subset of URTI) for a period of 4 years (January 2016–January 2020).3 We then used GTs to retrieve relative search volume data for the topics ‘AR’, ‘pollen’, ‘asthma’ and ‘common cold’ (topics are groups of search terms concerning the same concept) for England during the same time period. Some of these topics had been previously used.4 Weekly surveillance and GTs data were correlated using Spearman correlation coefficients (ρ). Using English surveillance data, ‘asthma’ incidence was strongly correlated with ‘common cold’ incidence (ρ = 0.800) and ‘URTI’ (ρ = 0.765), but negatively correlated with ‘AR’ incidence (ρ = −0.517) (Table 1; Figure 1). 0.800 [0.733;0.845] 0.987 [0.980;0.990] −0.517 [−0.601;−0.420] −0.517 [−0.586;−0.433] −0.580 [−0.650;−0.505] 0.490 [0.380;0.593] 0.474 [0.352;0.573] 0.480 [0.349;0.586] −0.426 [−0.535;−0.303] 0.716 [0.644;0.773] 0.722 [0.659;0.774] 0.780 [0.729;0.819] −0.835 [−0.868;−0.795] 0.556 [0.458;0.647] −0.639 [−0.706;−0.563] −0.637 [−0.690;−0.573] −0.701 [−0.750;−0.642] 0.866 [0.813;0.902] −0.320 [−0.444;−0.197] −0.816 [−0.849;−0.774] −0.628 [−0.695;−0.549] −0.634 [−0.697;−0.557] −0.693 [−0.753;−0.628] 0.902 [0.859;0.930] −0.361 [−0.474;−0.250] −0.808 [−0.841;−0.772] 0.915 [0.875;0.943] GTs data on ‘asthma’ were moderately correlated with GTs data on ‘common cold’ (ρ = 0.556) and negatively correlated with GTs on ‘AR’ (ρ = −0.361). GTs data on ‘AR’ and ‘pollen’ were strongly correlated (ρ = 0.915). GTs data on ‘common cold’ displayed a strong correlation with English surveillance data on ‘asthma’ (ρ = 0.716), ‘common cold’ (ρ = 0.780) and ‘URTI’ (ρ = 0.722). ‘AR’ incidence was strongly correlated with GTs data on ‘AR’ (ρ = 0.902) and ‘pollen’ (ρ = 0.866). This study has some limitations, the most important being its ecological design and the assessment of single-country data. We analysed 2016–2019 data, as the COVID-19 pandemic could have influenced our results (in March 2020, there was a media-driven GTs peak for asthma,5 and increased inhalers prescriptions). However, we obtained similar results when analysing 2016–2021 data (Table S1). Correlation does not imply causation, so we must be careful when interpreting results. Nevertheless, applying the Granger causality test for positive correlations (which assesses whether one time series is useful in forecasting another), we mostly observed biologically plausible results, with time series on URTI/common cold being useful to forecast asthma, and asthma time series not being useful to forecast URTI/common cold (Table S2). The incidence of ‘asthma’ was strongly and positively correlated with that of ‘URTI’, but negatively correlated with that of ‘AR’ and with GTs data on ‘pollen’/’AR’. These results are mostly explained by the different seasonality of AR and asthma, and suggest that, even though AR and pollen exposure may be linked to worsening asthma, viral infections may be a more relevant trigger for asthma exacerbations (as previously shown using GTs for ‘common cold’ and asthma hospitalizations in five countries6). Strong correlations were also observed between surveillance and GTs data on ‘AR’ incidence, suggesting the potential of GTs as a complementary tool in rhinitis monitoring. On the other hand, GTs data on ‘pollen’/‘AR’ displayed negative correlations with GTs and surveillance data on ‘common cold’, suggesting that patients and primary care teams may know how to distinguish seasonal allergies from URTI (which are less common during pollen season) despite some common symptoms. Taken together, these results point to the complementarity of classical surveillance and of GTs data (even though GTs does not solely assess incident disease and may be influenced by media coverage5). For example, for each topic, GTs displays a list of the top-related queries searched by internet users, possibly helping to context epidemiologic fluctuations and to raise new hypotheses (eg, on exacerbations triggers). In addition, analysis of GTs data may possibly help planning health care needs for allergic respiratory diseases (as in most countries surveillance of asthma or AR does not occur). Open access funding enabled and organized by ProjektDEAL. None declared. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.