Early warning and monitoring of COVID-19 using the Baidu Search Index in China.

The recent study published by Ben, et al. showed that the gastrointestinal symptoms can be used in predicting COVID-19 outbreak. We report another finding that using the diseases with similar symptoms in the early warning of COVID-19 outbreak and monitoring the epidemic trends in China.

The data used in analysis was from December 1, 2019 to March 15, 2020. The daily number of onset cases before January 20 was obtained from the publications2, 3, 4 by using GetData Graph Digitizer software. The daily number of laboratory confirmed cases after January 20 (including clinical diagnosis between 12 and 14 February) was obtained from China National Health Commission (http://www.nhc.gov.cn/xcs/yqtb/list_gzbd.shtml). The daily Baidu Search Index volume of “influenza (流感) or  (流行性感冒)”, “pneumonia (肺炎)”, “SARS (非典) or  (非典型肺炎)”, “coronavirus(冠状病毒) or  (新型冠状病毒)”, “COVID-19  (新冠肺炎) or(新型冠状病毒肺炎) or  (新冠)”, “MERS (中东呼吸综合征)”, “Avian influenza (禽流感) or  (人感染高致病性禽流感)” and “masks(口罩)  (including N95)” was achieved from the Baidu Index Platform (http://index.baidu.com/v2/index.html#/) and weighted by populations. All data analysis were separated in Wuhan, Hubei (excluded Wuhan) and China (excluded Hubei), respectively for comparison and separated into 3 phases according to the key events that may affect the search behaviors. Phase 1 was from December 1 (7 days before the first case onset) to 31, 2019, when the rapid response team from China CDC was sent for investigation, which caused the public's great concerned. Phase 2 was from January 1 (Huanan seafood market in Wuhan was closed) to 20 (COVID-19 was notifiable in China). Phase 3 was from January 21 to March 15 when strong intervention measures were taken to well control in China. The time series and Spearman correlation analysis were used with the lag time was set as 0–20 days for every one day interval. Data analysis was using R 4.0.2.

In phase 1, the search volume of “Influenza” in Wuhan was obviously increase on December 9, 2019 and keep in the highest level with the median (inter-quantile) was 66.7 (42.8–67.8) per million populations on December while it was only 2.9 (2.3–4.9) in Hubei (excluded Wuhan) and 6.7 (3.6–8.0) in China (excluded Hubei). An anomalous peak of “SARS” (as high as 8485.02/1000,000 populations) and “pneumonia” in Wuhan occurred on December 30–31, 2019 and keep in very high level with the number of cases increase and then substantially increased on January 16 with another anomalous peak was observed on January 20 (Fig. 1 A). The correlation coefficient of “influenza”, “pneumonia” and the number of cases (lag 0) in Wuhan was 0.69 and 0.59, respectively. While it was both only 0.31 in Hubei (excluded Wuhan).

Fig. 1

The changing trend of the daily number of cases by symptom onset date and the Baidu Search Index per 100,000 population of the different keywords from December 1, 2019 to January 20, 2020 in Wuhan (1A, 1B), Hubei excluded Wuhan (1C) and China excluded Hubei (1D). Note: The Baidu Search Index of “SARS” in Wuhan on December 31, 2019 was as high as 8485.02 per million population, as the relative small value of other keywords, for visualization, this peak was not fully display in Fig. 1A.

In phase 2, with the casual pathogen was public on January 7, the search term of “pneumonia”, “coronavirus”, “COVID-19” and “masks” in Wuhan increasing substantially (Fig. 1B). The correlation coefficient of “coronavirus” and “COVID-19” and the number of cases (lag 0) in Wuhan, Hubei (excluded Wuhan) and China (excluded Hubei) were 0.69∼0.89, while “influenza” change to −0.56 ∼ −0.64.

In phase 3, the search volume of all keywords in Wuhan (except “influenza”), Hubei (excluded Wuhan) and China (excluded Hubei) was significance increase. The daily trends of the number of cases were consistent with search volume of “COVID-19”, “pneumonia” and “coronavirus” while the peak displayed 10 days lag time in Wuhan before February 18 and 5 days lag time in Wuhan (after February 19), Hubei (excluded Wuhan) and China (excluded Hubei). The search volume of these three keywords were decreasing with the number of cases and keep in relative low level after February 28 (Fig. 2 ). A high association (rs>0.7) was observed of the search volume of “influenza”, “pneumonia”, “SARS”, “coronavirus”, “COVID-19” with 6–12 days earlier in Wuhan, 0–7 days earlier in Hubei (excluded Wuhan, except “influenza”) and 0–5 days China (excluded Hubei).

Fig. 2

The changing trend of the Baidu Search Index per 100,000 population (by different lead time) and the daily number of reported cases by diagnosis date for the different keywords from January 21 to March 15, 2021 in Wuhan (2A, 2B), Hubei excluded Wuhan (2C) and China excluded Hubei (2D).

Note: a. 2A plot with 10 days lead time while 2B, 2C and 2D plot with 5 days lead time of the Baidu Search Index compare to the number of cases. b. The number of reported cases in Wuhan and Hubei from February 12 to 14 including clinical diagnosis cases due to the Sixth Edition of COVID-19 Diagnosis and Treatment Scheme in China changed the reporting criteria. For visualization, the number of clinical diagnosis cases was not included in Wuhan as it was as high as 12,364, 1667, 922, respectively. But it was included in Hubei (excluded Wuhan) (2C).

COVID-19, as an emerging infectious disease, it is important to detect the outbreak as early as possible for taking actions. Baidu serve as the most popular search engine in China, we found that the abnormal and continuous high search volume of “influenza” and “pneumonia” on December 2019 can be used in the early warning in Wuhan. Although the causative pathogen was unknown, however, the continuous’ attention of “SARS” and “pneumonia” accompany with the increasing search volume of “coronavirus” and “mask” in Wuhan in the next 8 days may implicated the severity of this disease and the epidemic in Wuhan. After the causative pathogen was announced until COVID-19 was notifiable, the search term of “COVID-19”, “SARS”, “pneumonia”, “coronavirus” and “mask” in Wuhan were 6.3–24.5 folds higher than Hubei (excluded Wuhan) and China (excluded Hubei) may implicated the further development in Wuhan.

Even after the surveillance system was built no January 20, we can obviously observe that the Baidu Search Index data was ahead of the reported case data. The peak of search volume of “COVID-19”, “pneumonia” and “coronavirus” were 10–12 days earlier than the reported cases before February 18. This was consistent with the situation that 12 days interval from onset to diagnosis due to the medical resources shortage. , When the medical resource was relative sufficient, the peak of search volume was still 5 days earlier than the reported cases in the whole China. The high association between the number of cases and ‘pneumonia’, ‘coronavirus’ and ‘masks’ after January 8 in the whole China implicated the potential epidemic in China. Given COVID-19 as an emerging and high contagious disease with R0 2.2–3.84, , make full use of the internet search data should be recommended in the future for timely alert and the prediction of further development.

Declaration of Competing Interest

We declare that we have no conflicts of interest.