Discussion

We analyzed search behavior on the Google search engine for Coronavirus related information with the Google Trends (Google, Mountain View, CA, United States) tool for India, as well as 10 states with highest and lowest caseloads. However, we could not identify any published studies using similar comprehensive search strategy or including GT data from states/UTs of India. ^{Reference Sharma and Sharma33,Reference Venkatesh and Gandhi34} For other countries, researchers have used only a few terms for Google Trend analysis of Covid-19 online search behaviour. ^{Reference Mavragani24–Reference Li, Chen, Chen, Zhang, Pang and Chen27,Reference Strzelecki29–Reference Husnayain, Fuad and Su32} We found 2 studies which reported the use of various methods to arrive at a detailed search strategy, but they have not reported the details. ^{Reference Higgins, Wu and Sharma28,Reference Yuan, Xu, Hussain, Wang, Gao and Zhang42} Categorization of the search terms is expected to answer 2 assumptions. First, the search behavior may simply reflect information-seeking out of curiosity, and apprehension among citizens generated by pandemic coverage by the media. Our search categories of disease terms, NPI terms, and government instruction terms is expected to reflect curiosity-induced search behavior. A study conducted to evaluate reliability of Google Trends™ (Google, Mountain View, CA, United States) for digital surveillance reports the influence of media coverage. ^{Reference Cervellin, Comelli and Lippi43} COVID-19 infodemiology studies have also reported similar GT data correlation of COVID-19 disease related terms with media coverage. ^{Reference Sousa-Pinto, Anto, Czarlewski, Anto, Fonseca and Bousquet44} A stark difference is seen in the United States of America (USA), where peak interest was reported on March 9 and 12, coinciding with the issuance of interim guidelines and the declaration of Novel Coronavirus pandemic by World Health Organization (WHO). ^{Reference Husain, Briggs and Lefebvre23} In India, search trends did not peak on these 2 dates however, as Indian mainstream mass media did not report these developments. Secondly, people seek health information to guide themselves following actual illness episodes in self or family members. The second assumption in search behavior would augment disease surveillance activities by providing a prediction of outbreaks earlier than traditional methods. Google Trends (Google, Mountain View, CA, United States) had been argued to have earlier predictive utility for H1N1 epidemic, dengue fever, Chikungunya, and Malaria, etc. ^{Reference Husnayain, Fuad and Lazuardi12,Reference Verma, Kishore, Kumar, Sondh, Aggarwal and Kathirvel22,Reference Zhang, Bambrick, Mengersen, Tong and Hu45–Reference Hossain, Kam, Kong, Wigand and Bossomaier47} Moreover, our study included GT data for a longer duration, unlike most of the other studies reporting Google Trend analysis for COVID-19 infodemiology.

The earliest search for the terms “coronavirus” and “corona” in India likely reflects media reports of the first confirmed case. The earliest spikes for disease-related terms likely coincides with media reports of the first cluster of cases reported among Italian tourists from Rajasthan. ^{Reference Awasthi48} The largest peak in the search for disease-related terms coincides with the declaration of ‘junta (citizen) curfew’ and lockdown by the Indian Prime Minister in a televised national address which was widely reported by various sections of the media. ^49,50 The same analogy should apply individually to the 10 states except Meghalaya. Cities with maximum searches related to “Covid-19” have the highest population densities in India and the highest caseloads in the initial phase of the pandemic. This may indicate true health-seeking behavior. COVID-19-related searches in India slowly began to decline after the last week of March, 2020 most likely due to flooding of information from various print, electronic and social media leading to information fatigue and disinterest.

NPI terms that we have used in our study have been used by few other studies as well for GT analysis. ^{Reference Bento, Nguyen, Wing, Lozano-Rojas, Ahn and Simon26,Reference Rovetta and Bhagavathula30,Reference Husnayain, Fuad and Su32,Reference Lin, Liu and Chiu41} The first search spike for these terms is likely related to the first case reported from Kerala, India. Searches for the term “Mask” in India reached 2 peaks much earlier than that for disease related search terms. Like disease related terms, the first peak coincides with the report of the first cluster of COVID-19 cases among Italian tourists. The second peak coincides with observation of citizen curfew likely reflecting curiosity. We argue that such health-information seeking may not correlate with healthcare seeking in India. This is in contrast with similar searches in internet-savvy countries during influenza epidemics. ^{Reference Cook, Conrad, Fowlkes and Mohebbi51} Our assumption is buttressed by the finding of poor trends for search behavior pertaining to 3 other NPI terms: “social distancing,” “hand wash,” and “soap.” Previous studies have shown that people in India are mostly aware of the effectiveness of handwashing with soap as healthy behavior, while the use of masks and social distancing are considered to be useful during epidemics like H1N1. ^{Reference Verma, Chayal and Kumar52,Reference Kamate, Agrawal, Chaudhary, Singh, Mishra and Asawa53} While handwashing has been shown to be consistently increasing, “Masks” and “sanitizers” have been infrequently used community-level preventive measures in Indian health system till the current COVID-19 pandemic. The increased search interest for “Mask” and “sanitizer” is also linked to the listing of these items under the essential commodity list by the Indian government. ⁵⁴ In states with low caseloads, users were more interested in searching about sanitizers than masks.

Finally, the lesser interest in search for the symptoms (specifically “fever” and “cough”) both in English and Hindi languages strongly supports our argument that the search behavior in India was mostly a result of curiosity rather than health-seeking. Only a few similar studies from India have reported the use of these search terms in their analysis. ^{Reference Sharma and Sharma33} Globally, we also identified only few published studies which used “fever” to analyze Google Trends™ results during the ongoing COVID-19 pandemic. ^{Reference Higgins, Wu and Sharma28} Google Trends™ analysis using the term “fever” has previously been reported for other diseases such as dengue fever and Influenza. ^{Reference Husnayain, Fuad and Lazuardi12,Reference Niforatos and Pescatore55,Reference Kang, Zhong, He, Rutherford and Yang56} Coexistence of Influenza and COVID-19, or Dengue fever and COVID-19 cannot be ruled out only by search behavior using this term during winter and rainy seasons respectively. For the term “cough,” we also could not identify any published study on Covid-19 from India using Google Trends™ analysis. Few other studies have used “cough” for Google Trends™ analysis in diseases like Sinusitis and COVID-19. ^{Reference Sharma, Sandelski, Ting and Higgins19,Reference Higgins, Wu and Sharma28,Reference Sousa-Pinto, Anto, Czarlewski, Anto, Fonseca and Bousquet44} The peak trends for “fever” and “cough” in India coincided with the declaration of the lockdown and with peak search trends for coronavirus-disease related terms and NPI terms. Similar peak trends in high-burden states reflect the curiosity following the lockdown declaration.

Regarding trends for terms related to government instructions (“lockdown,” “quarantine,” and “curfew”), peaks for “lockdown” both at the national level and for the high-burden states coincides with each of the 5 phases of the lockdown enforced by the Indian government. We argue that the only peak for “quarantine” and “curfew” is due to the curiosity generated through the observation of “junta (citizen) curfew.” The highest trend for the term “curfew” seen in Meghalaya and Mizoram states were due to local non-health issues.

We could not find any study from either India or other countries using Hindi terms related to COVID-19 for Google Trends™ analysis. Nearly 43.6% of Indian population uses Hindi as their first language. ⁵⁷ The term “sardi” which is the Hindi translation for both “common cold,” and “winter” had its peak on January 1 which coincides more aptly with the cold conditions of winter in India. This trend reflects media coverage for the extremely cold conditions in many parts of the country. The peaks for the terms “khansi” and bukhar” coincided with the Junta curfew, hence reflecting curiosity again. Using these terms may not help in the prediction of an outbreak through info-veillance.

This study showed high time-lag correlations by 10-14 days between the search behavior for Coronavirus related keywords using Google Trends™ and the number of new daily confirmed COVID-19 cases as well as tests. Shin et al., in their study on Middle East Respiratory Syndrome (MERS) using Google and Twitter keywords have showed similar results with daily cases and quarantined cases. ^{Reference Shin, Seo and An13} Various other studies have reported high time-lag correlation between disease related search terms and the increase in cases during an outbreak. ^{Reference Husnayain, Fuad and Lazuardi12,Reference Shin, Seo and An13,Reference Mavragani24,Reference Effenberger, Kronbichler, Shin, Mayer, Tilg and Perco25,Reference Li, Chen, Chen, Zhang, Pang and Chen27,Reference Higgins, Wu and Sharma28,Reference Husnayain, Fuad and Su32,Reference Sharma and Sharma33} The correlation was observed mainly for the terms “COVID 19,” “COVID,” “social distancing,” “soap,” and “lockdown” at the national level. Based on the available data, we are not sure whether these correlations are really due to information-seeking for healthcare or simply reflect curiosity. We have just explained that most of the peaks in search behavior for the 18 terms used for Google Trends™ analysis can be explained by curiosity emanating from media coverage of events like the declaration of lockdown, clustering of cases, etc. Community-based studies are required to ascertain the exact cause of health-information seeking.

This shows increased interest among the general population not only towards COVID disease related search terms such as “COVID 19” and “COVID” but also towards their own safety as observed by the high time-lag correlation throughout the period of 14 days for the terms “social distancing” and “soap.”

A high positive correlation between the number of infected cases and the Google trend values for the term “COVID 19” has been reported for India as part of the analysis for 8 major countries. ^{Reference Sharma and Sharma33} Similarly, high positive time-lag correlation between the number of infected cases and the Google trend values was seen for the term “COVID” in USA. ^{Reference Yuan, Xu, Hussain, Wang, Gao and Zhang42} While the term “Coronavirus” showed high time-lag correlation in several studies with cases, our study did not have such correlation. This might be due to the fact that these studies were done between January and March, 2020 when the term “COVID 19” was not used widely. ^{Reference Mavragani24,Reference Li, Chen, Chen, Zhang, Pang and Chen27,Reference Husnayain, Fuad and Su32} Increased apprehension of the population on whether the “lockdown” will end or whether it will be extended can be observed by the high time-lag correlation observed with daily cases at 14 days and with the daily tests at 13 days at the national level in India.

“Soap” was the term with the highest Pearson correlation coefficient at a time-lag of 10 days when analyzed against the actual daily COVID-19 cases and a time-lag of 1 day when analyzed against the daily COVID-19 tests at the national level. This might be due to either majority of population being well informed about the need for, and method of handwashing, or not giving due importance to handwashing over the use of other NPIs such as “sanitizer.” We hypothesize that media coverage of stock out of sanitizer might have prompted people to give more importance to sanitizer as compared to soap as a preventive method in COVID-19 epidemic. There were government instructions on including sanitizers in the list of essential commodities, which might further have deflected interest from soap to sanitizer. Therefore, we propose that “sanitizer” may be a more appropriate term for digital surveillance of COVID-19 outbreaks in India as compared to “soap” or “handwashing.” The high time-lag correlation observed in the high-burden states for these terms likely contributes to the observation at national level. These findings should form part of digital surveillance at state levels. It would be interesting to see whether such trends are seen in hitherto low-burden states which may develop outbreaks in future. The high correlation with the term “Fever” seen in the states of Maharashtra and Delhi is most likely predictive for an outbreak during the ongoing “COVID 19” epidemic in these states unlike the national level. This is in contrast to the results obtained by Higgins et al., who reported positive correlation between Google search for symptoms of COVID-19 and the COVID-19 cases at level of countries. ^{Reference Higgins, Wu and Sharma28}

These findings reflect increase in internet search activities almost 10 to 14 days before the increase in daily COVID-19 cases and deaths. These findings indicate that Google Trends™ can serve 2 purposes. First, it can be used as a tool to monitor public restlessness toward COVID-19 infections in India to decide the timing and location for risk communication by governments. Such timely and focused risk communication can help avoid “infodemics” of panicky health-seeking. Second, it should be further explored whether prediction of emergence and propagation of COVID-19 outbreaks by 2 weeks by GT can supplement the traditional field-based surveillance mechanism. Previous epidemics have supported the use of internet searches for digital outbreak surveillance, pointing towards the fact that the digital surveillance deserves greater investment by the public health agencies. ^{Reference Majumder, Santillana, Mekaru, McGinnis, Khan and Brownstein14,Reference Ginsberg, Mohebbi, Patel, Brammer, Smolinski and Brilliant58} The advantages of using a tool such as Google Trends™ (Google, Mountain View, CA, United States) is that data can be obtained earlier, more easily, and at a much lower cost than routine surveillance techniques adopted by the governments. A study has compared the surveillance platform of India known as Integrated Disease Surveillance Project (IDSP) and reported strong correlation with a time-lag of 2 to 3 weeks for Chikungunya, Dengue Fever, Malaria and Enteric Fever. ^{Reference Verma, Kishore, Kumar, Sondh, Aggarwal and Kathirvel22} Similar studies should be attempted for COVID-19 data from IDSP and Google Trend™. Longitudinal studies need to be conducted among internet users to identify and follow-up their online search behavior related to COVID-19 epidemic to verify the reason(s) for such searches. Such studies can provide a correction factor which can be used while using GT for complementing the traditional surveillance system. We hypothesize that the lack of such confirmatory studies led to demolishing of Google Flu search network. ^{Reference Lazer, Kennedy, King and Vespignani59}

Our study does have certain limitations. The selection of the keywords as well as the associated spelling might affect the overall results of the study. There is no globally accepted guideline for Google Trends™ (Google, Mountain View, CA, United States) analysis, though a procedure has been recommended by some authors. ^{Reference Mavragani and Ochoa7} A guideline is required from Google™ in this regard due to their custody of the search data, but none- sharing of the algorithm for search. ^{Reference Nuti, Wayda and Ranasinghe9} We did not have access to media coverage in India during this period to explore the correlation between COVID-19 related media coverage and GT data.