Evaluation of a mobile health approach to improve the Early Warning System of influenza surveillance in Cameroon

Abstract Background Rapid reporting of surveillance data is essential to better inform national prevention and control strategies. Objectives We compare the newly implemented smartphone‐based system to the former paper‐based and short message service (SMS) for collecting influenza epidemiological data in Cameroon. Methods Of the 13 sites which collect data from persons with influenza‐like illness (ILI), six sites send data through the EWS, while seven sites make use of the paper‐based system and SMS. We used four criteria for the comparison of the data collection tools: completeness, timeliness, conformity and cost. Results Regarding the different collection tools, data sent by the EWS were significantly more complete (97.6% vs 81.6% vs 44.8%), prompt (74.4% vs n/a vs 60.7%) and of better quality (93.7% vs 76.1% vs 84.0%) than data sent by the paper‐based system and SMS, respectively. The average cost of sending a datum by a sentinel site per week was higher for the forms (5.0 USD) than for the EWS (0.9 USD) and SMS (0.1 USD). The number of outpatient visits and subsequently all surveillance data decreased across the years 2017‐2019 together with the influenza positivity rate from 30.7% to 28.3%. Contrarily, the proportion of influenza‐associated ILI to outpatient load was highest in the year 2019 (0.37 per 100 persons vs 0.28 and 0.26 in the other 2 years). Conclusion All sentinel sites and even other disease surveillance systems are expected to use this tool in the near term future due to its satisfactory performance and cost.


| INTRODUC TI ON
In recent years, influenza surveillance that was essentially virological expanded to include more epidemiological information to complement the virological data collected by the Global Influenza Surveillance and Response System (GISRS). 1 The 2009 influenza pandemics highlighted the need for rapid reporting of cases to assess the severity of the disease, define risk factors for severe outcome and to better inform national prevention and control strategies. This has urged many countries to establish surveillance systems for the early detection of public health emergencies and detection of potential pandemic influenza strains. 2 Reporting of surveillance data has mostly made use of paper-based systems, mobile phone-based systems and Web-based systems. Among these, mobile and Internet technologies have been successfully used for EWS in several countries and settings. [2][3][4][5] In Cameroon, there has been progress in the collection tools for influenza epidemiological data from forms to SMS (short message service) to smartphone using the Internet in order to improve on the timeliness of data collected. The implementation of the EWS, a Web-based system that makes use of smartphones, within the influenza surveillance in 2017 started with a few sentinel sites in Cameroon for more real-time analyses of data collected and in the preparedness of a future pandemic event.
We evaluate here the performance of the EWS as compared to prior tools for collecting influenza epidemiological data and estimate the annual proportion of influenza-associated illness among total outpatient visits in Cameroon.

| Description of the influenza surveillance system
For more than a decade, the Centre Pasteur of Cameroon has been designated the National Influenza Centre of Cameroon by the Ministry of Health and by the World Health Organization. In 2019, the influenza surveillance system comprised 16 sites distributed in 7 of the 10 administrative regions of the country. Among these, 13 sites collect data from outpatients, while 3 sites collect data from hospitalized patients with a severe acute respiratory infection (SARI). This surveillance system generates two main types of data: epidemiological data from sentinel sites and virological data from laboratory analysis of samples collected. Epidemiological data are collected weekly from sentinel sites and comprise information on the number of consultations, number of febrile illness, number of acute respiratory infections (ARI), number of influenza-like illness (ILI) and number of samples collected. Meanwhile, virological data obtained mostly comprise the influenza status of each individual sample collected. Nasopharyngeal and/or oropharyngeal swabs collected from the sites are analysed for the presence of influenza virus using the gold standard assay, rRT-PCR, as previously described. 6

| Evolution in the tools for collecting epidemiological data
Tools for the collection of epidemiological data from sentinel sites have gradually evolved over the years from forms (paper-based system) to SMS to the smartphones (EWS). Initially, all epidemiological data were sent through the paper-based system together with the respiratory samples. However, some major issues encountered with this system were the lack of complete data and timeliness. In September 2012, weekly reporting by SMS started at the sentinel sites in addition to the paper-based system. Data sent by SMS comprised reduced information as compared to the forms, with two parameters reported by age groups, that is number of consultations and number of ILI.
This reduced reporting via SMS was implemented to enable timely reporting of the minimum essential data in the WHO FluID platform Recently, reporting via the EWS with smartphones was initiated in a few sentinel sites in order to improve still on the timeliness of data received. The EWS makes use of Event Capture, an Android application which enables to capture and submit events (https://play. google.com/store /apps/detai ls?id=org.hisp.dhis.andro id.event captu re&hl=en). This system first started in January 2017 with sites located in the same town as the NIC (Yaounde) for a better coordination of this novel tool, and then was extended to sites located in the Northern region of Cameroon (Garoua) in August 2018. The EWS started with weekly reporting, but changed during the second phase of implementation to daily reporting for a better preparedness to a future pandemic event or in case of any unusual rise in influenza activity. Daily data sent through the EWS are aggregated into weekly data and extracted automatically in the server at the NIC.
Of the 13 sites which collect data from persons with ILI, 6 sites send data through the EWS, while the remaining 7 sites make use of forms and SMS. Of the 6 sites supposed to send data through the EWS, one had not sent any data and was discarded in the analysis.

| Method of comparison of collection tools
We used four criteria for the comparison of the epidemiological data collection tools: completeness, timeliness, conformity and cost.
Proportions of each criterion were compared among all three tools.
Completeness refers to data of the 52 epidemiological weeks that was successfully sent. For the EWS, completeness also involved sending all five or six daily data corresponding to the working days of the week.
Timeliness refers to data that were sent timely, that is within three days following the end of the reporting period.
Conformity refers to data that had no errors. We considered here as errors data with totals of each parameter wrongly calculated, incoherence of data (number of ILI > number of ARI OR number of febrile illness > number of consultations), errors in selecting the epidemiological week and presence of missing values in data sent.
Cost corresponds to the average cost in USD of sending one datum by a sentinel site per week. The cost of sending one datum through the EWS comprised the weekly cost of Internet provision necessary to send the data. The cost of sending one datum through the SMS comprised the cost of the SMS in accordance with the network provisioner. The cost of sending data through the paper-based system comprised the transport cost for sending the notification forms alongside the collected samples. We exclusively use 2019 data for comparisons among the different tools to ease analyses and to minimize bias.

| Comparison of the tools for collecting surveillance data
Concerning the tools used in collecting epidemiological data; of the 364 data that were expected to be sent by forms, 81.6% were even- SMS conformity on the other hand was 86%-93% for most sites and 62% for BASB. One of the sentinel sites sent no SMS data. Figure 2 shows the performance of each sentinel site based on the three data collection tools.

| D ISCUSS I ON
This study aimed to compare the performance of the EWS to the paper-based system and to the SMS in reporting influenza epidemiological data with respect to four selected criteria. Results showed that the EWS had significantly better performance in sending complete, prompt and conform data at a low cost. The workload has been reported by the focal points as the main reason for not sending timely data. Timeliness of the SMS was lower than reported by other influenza surveillance systems in Africa 2,9 but higher than that observed in 2014-2015 with the IDSR in Madagascar. 10 Regarding data quality, there were fewer errors in data sent through the EWS than data sent by forms or SMS. This is not surprising since the most commonly noted sources of error with the forms and SMS were corrected during the implementation and programming of the EWS. However, some data presented with missing information in the EWS, and this was corrected automatically in the system once the error was identified. A similar study in Kenya reported as well less errors in smartphones compared to the paper-based questionnaire. 11 Generating automated weekly bulletins for reporting performance, trends and summary of data collected by each site could help identify erroneous data rapidly, improve on site performance and help in driving public health actions as noted by other EWS. 3 The average cost of sending a datum by a sentinel site per week was lower for the SMS (0.1 USD) than for the forms (5 USD) and EWS (0.9 USD). However, SMS data still need to be entered manually in the database and this could be a potential source of error. The cost of sending data by the paper-based system was high because the forms are generally sent together with the samples. Meanwhile, the annual average cost for sending data through the EWS did not take into consideration the cost of setting up the electronic data collection system which is greater due to the high cost of electronic equipment and operating software. However, once these initial expenses have been handled, the EWS remains more cost-effective than using the paper-based system and SMS especially considering the possibility of analysing the data on real time. Similar findings were reported in Kenya where the EWS was found to be more cost-effective than the paper-based system. 11 The estimated incidence of influenza-associated ILI outpa-  Cameroon has indeed confirmed higher transmission rates of influenza virus in this age groups probably due to high contact rates in schools. 6 There was one peak of influenza activity in 2019 between week 39 and week 52 and this was slightly correlated with ILI levels. This could be used in setting up the alert thresholds in the EWS. This result corroborates with previous findings which showed that the major period for influenza activity in Cameroon is between the months of September to December. 6 Although ILI and ILI% are better indicators for use in EWS, as they are easily generated, these indicators may results in bias since illnesses other than influenza may present with ILI. 14,15

| CON CLUS ION
At the end of this study, which aimed to evaluate the performance of the EWS in collecting epidemiological data as compared to the paper-based system and the SMS, we found that the EWS had significantly satisfactory performance based on the four selected criteria for evaluation. Also, after implementation, considering the low cost of approximately 0.9 USD for sending one complete surveillance data per site, this tool could be proposed for national surveillance systems. All sentinel sites and even other disease surveillance systems are expected to use this tool in the near term future due to its satisfactory performance and cost. The next step in the EWS is to integrate alert threshold for influenza virus circulation in Cameroon based on previous surveillance data.

ACK N OWLED G EM ENTS
We are grateful to all the focal points of the influenza surveillance system in Cameroon for their involvement in data collection. This work received a grant from the US Department of Health and Human Services, DHHS (Grant Number 6 DESP060001-01-01), and from the WHO PIP Implementation project in Cameroon.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests.