Development, design, and experience related to Android application-based data collection for the MAASTHI cohort study in India

INTRODUCTION

In low and middle-income countries, traditional paper-based forms are used mostly for collecting data in most research studies.The entire process of data collection and management requires more extended time and constant monitoring.Before proceeding with the data analysis, further data quality checks by experienced researchers have to done to ensure the high quality of the data with minimal errors.^1,2 However, organizing the paperbased questionnaire is a challenge in conducting large epidemiological studies in developing countries. This requires organizing multiple printed questionnaires for repeated assessments involving multiple individuals over a long period in cohort studies. As a result, the data management processes tend to be tedious and have multiple disadvantages. First, the use of paper-based forms offers fewer opportunities for data validation at the point of entry. Second, these are prone to physical damage, require printing (often several times) that ultimately leads to paper wastage, and necessitates secure storage facilities until publication or legislative requirements. Third, initial entry or edits entirely rely on manual supervision. Fourth, it is often difficult to supervise and to ensure real-time or uniform quality standards in data collection. Finally, transcription costs from paper to computer system are unavoidable and is error-prone^3-5. There is an imperative need for capturing the data electronically, given it overcomes the challenges in using paper-based forms. A web or mobile/tabletbased application (app) is the primary alternative for data collection. It is currently possible to design the applications (apps) to provide clean data sets with minimal data entry errors. The complexity of data management reduces significantly.^2,6-9 Also, it is possible to automate the data collection, avoid duplication, and ensure real-time syncing through. The advantage offered by a thorough tracking system and audit logging system ensures the accountability of the system without costing much compared to paper-based forms. The use of an electronic platform for data collection ensures coordinating multiple staff on a real-time basis. The upfront cost for the development of the tool and software ecosystem can be well justified by several means. These include modularity, maintainability, extensibility, data-sanity, checksand-control, security, and integrative capabilities offered by mobile apps. Despite the increasing use of app-based technology for capturing field data in the developing world, there remains a dearth of publications that share field experiences and challenges. Through our paper, we highlight the development process of the app, field-level experiences, and challenges in a cohort setting in India (Table 1).

The purpose of our paper was to present the development and usage of a specially designed app for data collection of the "Maternal Antecedents of Adiposity and Studying the Trans generational role of Hyperglycaemia and Insulin (MAASTHI)" cohort.10 The cohort conducted at public health facilities of Bangalore, South India, aims to prospectively assess the effects of glucose levels during pregnancy and adverse infant outcomes, especially in predicting the possible risk markers of chronic diseases in adolescence and adulthood. 

Materials and Methods

The cohort initiated in April 2016 is currently on going in public sector prenatal clinics in Bangalore, Karnataka, South India. The study design and methodology of the cohort have been comprehensively described elsewhere10. The cohort was planned to recruit and capture information regarding 3000 pregnant women (respondents) and their offspring born during the four-year study period. The cohort was divided into baseline and follow-up periods, and interaction with respondents was done accordingly. The method of data collection was face-to-face interviews with respondents along with anthropometric measurements of mother-child pairs at different study time points. Due to elaborate details required for the cohort study, the interface was developed for secure use by authorized personnel onlybased on their roles. The principal investigator developed the concept of the app, with the intended data points and requirements. The app was then developed by a trained professional with over 12 years of experience in developing software systems for various verticals such as public health, finance, etc.

The data collection is performed using a secured Android app specially developed for the MAASTHI cohort. The app screens were constructed using user-experience (UX) best practices and design guidelines to ensure a consistent and uniform look throughout the interface. The app is specially designed for compatibility with devices of multiple screen-sizes. Additionally, an extensive range (96.5%) of Android versions (from 4.4 to 9.0) are covered for maximum backward compatibility.

Development process and costs

After training of the field researchers and Research Supervisors on the use of the mobile app by the app developer, the first beta version was tested for baseline recruitment. Post three months of extensive use, the first release candidate was introduced. Subsequently, version 1.4 was developed for follow-up visits. Thus the entire system was redesigned and released under version 2.0, with the current version of the app 2.1.0.

Initially, a Backend System version 1.0 was developed on Microsoft .NET 4.5 framework using ASP.NET MVC 4. Later, with the release of mobile app version 1.4, the backend system was upgraded to ASP.NET MVC 5. We chose to run the backend on Microsoft Azure. With version 2.0 of the mobile app, the complete backend was migrated to Microsoft.NET Core 2.1, and the front-end for a web app is developed in Angular 6. Presently the system runs on a private cloud hosted on Linux with Microsoft SQL Server Express edition as the database.

The entire process of developing the app and the platform amounted to USD 5886.6 (1 USD = INR 67.95 as in 2017). While the costs of running the present cloud server and the app has been reduced from INR 30,000 to INR 15,000per month (USD 420.4 to USD 210) due to the migration to Microsoft .NET Core 2.1 and private cloud hosted on Linux with Microsoft SQL Server Express edition. Cost of printing one questionnaire which is a 150 page booklet is INR 112.5 and the total cost for printing this booklet for 5000 respondents is INR 562,500 and the cost of data entry of the entire questionnaire is INR 20, 00,000.

Data entry process

Since the data collection is done during antenatal clinic visits during pregnancy and immunization visits and household visits, the app was designed to allow quick and easy. The field staff logs in to the system with her credentials. For fresh recruitment, the New Respondent workflow for baseline is followed. At any point, after the mandatory sections, the flow can be stopped, and data can be synced to the server. Once synced, the data is removed from the device. If the RA wants to resume a partial interview, she can do so by seeking the respondents via different filters. At this time, the appropriate response is sent back from the server. Follow-ups are such resumes, albeit after baseline. Validations and checks and controls to prevent parallel seek and sync are implemented. In the case of ad ministerial intervention, a separate module is provided on the web app.

Data storage, security/privacy, and output

The automation of the workflow tasks is welldefined and intuitive. The data is stored on the SQL Server Express database on the backend. The server is protected with an SSL certificate to ensure maximum protection by encrypting. The mobile apps collect and store the data in JSON format locally. As the system follows the disconnected data model, the mobile app and the backend use JSON for transactions. JSON is the current industry-standard for data interchange between disparate systems. Modifications after entry are by edit modules provided in the web app, which are role-based. Some key data elements, such as inclusion and exclusion criteria, are protected.

The system produces output in four ways:

i. HTML web-based dashboard indicators via graphical and tabular elements

ii. Regular reporting in Microsoft Excel for dayto-day operations by the staff

iii. Entire questionnaire as a PDF document for the archival purpose

iv. Different datasets in Microsoft Excel for data analysis according to the pre-defined codebook conventions

The data lives inside a secure environment of the private cloud with OS-level and databaselevel securities. The APIs are secured with an SSL certificate and use the JWT authentication system.

Basic aggregation based on pre-determined criteria is provided in the web app as graphical and tabular dashboards. Excel file reports are designed to assist the detailed analysis in the host software (Excel or statistical analysis software). User-roles control access to dashboards, reports, and datasets. These can be used directly on the host software (Excel or statistical analysis software) without any additional plugins.

Ethics consideration

Ethics approval for data privacy and storage was obtained through the Institutional Ethics Committee of Indian Institute of Public Health - Bangalore campus, After data entry and quality check, data can be accessed by the senior team members and Principal investigator only.

Results

Details on data collection can be seen in Table 2 (below). 

Thus far, the app has been used by the field staff to collect data from 5542 respondents, and 4270 follow up visits. Each field staff/ researcher has been provided with an individual tablet with the installed app. A unique identification number (UIN)is assigned to every researcher so that accountability and transparency are maintained for data entered. The researcher’s names are present in the output file to track data collection. Researchers are trained to use the app by senior team members with field experience. In case of end of contract or resignation of the researcher, the developer, and Principal investigator have rights to block the said researcher’s login credentials in order to bar access.

Researchers were requested for feedback on the app through an online anonymized survey. According to the researchers, the app was easy and straightforward to use and preferred over paperbased data collection. Most researchers were able to comprehend the use and function of the app within a day with hands-on usage and adequate training. The majority of the staff thought the app was suitable for even household and Anganwadi (outreach) visits, while few thought that ‘English language ‘was a barrier for use. Among features, the researchers appreciated were the calculation of gestational age, use of tick marks for each question, and skip and drop down options. Suggestions included GIS tagging for capturing the location, the inclusion of more options for twin pregnancy details, individual follow up reminders for each researcher, and improving login without the use of the internet.

The validation function in the app prevents several inaccurate and out of range entries into the system thus saving a lot of time in data validation. As and when the study progressed several data fields (anthropometry, lab values) were validated based on initial responses. Description notes were added in the app wherever technical jargons were used. Based on the response; the app would skip prospective questions that were not applicable; this would save a lot of time for the research team. The supervisors checked the collected data once in 15 days and flagged the errors for correction.

The research supervisors can download several reports and datasets in real-time in minimal time (less than 2 minutes). Some of the reports generated are a primary dataset, food recall dataset, physical activity dataset, a dataset of ineligible respondents. The website maintains an audit trail of all the entries made in the app so that the supervisor can track the entries made by the field staff (Figure 1). The website also has a dashboard that shows a graphical representation of monthly recruitment (Figure 2). The app locks a respondent identification number (RID) as soon as a field staff seeks it for data entry so that there are no multiple entries made for a single respondent, the supervisor has the option to revoke the seek when needed. Errors arising in data parsing while new versions of the app were released brought to the notice of the developer and were quickly resolved. As most of the fields in the questionnaire were validated, the time spent in data cleaning is minimal compared to paper forms.

Discussion

We have described how we have used app-based data collection in the on going MAASTHI cohort study. We have provided the process of collecting large amounts of data from an on going MAASTHI cohort comprising of 3000 women respondents and their offspring through multiple time points in the cohort study.¹¹ Globally there has been a boom in the use of mobile technology including use of tablets, and other handheld devices.¹² The app was designed to provide alternate options that follow the mandate of digitization of India by providing stable and secure infrastructure for high level data pertaining to pregnant women and their offspring.¹³ Compared to paper-based data collection, we have demonstrated that use of an android–based application worked especially for collection of large number of variables in the challenging environment of a developing country as India. The aim of development of the appspecific for the cohort was done to improve data collection and subsequent data management, rapid and secure access to real-time data for quick fixes rather than retrospective checking, reduction of long hours of data entry after paper-based collection, and uncomplicated data extraction. The immediate digitization and transmission of data from the point of survey promises more efficient, more cost-effective and more accurate surveys. As seen through our paper, the advantages of using an android app for the cohort outweighed the benefits of the traditional rather rudimentary paper-based method. The data collection by using the app helped the MAASTHI cohort by considerably reducing the time for data collection, improving the quality of the data, reducing transcription errors, and allowing for real-time supervision by senior team members.^4,14,15 The features in our app were specifically designed to provide seamless data collection and extraction procedures, among field investigators with minimal training on the handson use of the app and qualifications. The use of app-based technology that includes such features has been widely used in the developing world and has expanded to mHealth for use by health workers for surveillance, follow-up; and even selfmonitoring among chronically ill patients.^16-19

Our study demonstrated acceptability among the population in a South Indian public health facility setting through the use of apps that proved feasible for continuous data collection, with an optimal platform recommended for similar studies and settings. This mode of data collection explores the usability and acceptability of real-time data capturing by means of rapidly developing technology used for large cohort studies that has potential for following participants over the years. We highly recommend the use of technology as such for future cohort studies and other populationbased studies in similar settings. 

Funding

This work was supported by the Wellcome Trust/ DBT India Alliance Fellowship [Grant No. IA/ CPHI/14/1/501499] awarded to Giridhara R Babu.

Conflict of Interest The authors declare that they have no competing interests.