Interactive Voice Response (IVR): Technology and its role in health monitoring

Public health authorities in developed countries encourage excessive testing as a strategy to break the chain of COVID-19 infections and identify high risk areas. This strategy, while suitable for developed countries is unfeasible in several developing countries, such as Pakistan. Using high frequency data collection as a tool to monitor the behavior of individuals and thereby the changing health outcomes can be an additional asset as a rapid response mechanism.

Consequently, in this study, we follow respondents over a period of 3 months, during which we plan to collect survey data remotely from 24000 respondents. Data on the health status of household members, their adherence to preventive measures and the level of knowledge of the respondent on the Coronavirus is collected via bi-weekly calls.
Making use of this approach, we will experimentally test how to best collect information and promote preventative behavior remotely. The core outcomes of interest, including changes in response patterns on COVID-19 symptoms and awareness, will be captured over time and compared against a control group.

Findings of the high frequency data collected will be regularly presented on The Rapid Data for Action dashboard to offer an enhanced understanding of the situation and to identify suspected cases and sub-populations in utmost need of support.

The intervention is designed and implemented in close cooperation with Acted Pakistan, and National Rural Support Programme (NRSP). The target population are beneficiaries of the two NGOs across three provinces of Pakistan.
As of early September 2020, contacts of NGO beneficiaries were collected in 1139 villages. 12,324 individuals within these villages were randomized to receive enumerator led follow up interviews (4,312), IVR calls (1,848) or a mix of the two (6,164). In further rounds, an additional 12000 will be added in the same ratio, to get a final sample of 24000 individuals, with 7200 receiving IVR calls, and 16800 enumerator follow-ups.

External Link(s)

Registration Citation

Citation

Avdeenko, Alexandra et al. 2020. "Interactive Voice Response (IVR): Technology and its role in health monitoring." AEA RCT Registry. September 23. https://doi.org/10.1257/rct.6517-1.0

Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Experimental Details

Interventions

Intervention(s)

Over the biweekly data collection, two interventions are implemented on the population: 1) technology intervention and 2) IVR (Interactive Voice Response) context intervention.

In 1) one control and two treatment arms are formed. The base arm, or the control arm will consist of the sample that purely receives enumerator follow-up over the course of the data collection. The two treatment interventions within the technology design consist of a) a group that will purely receive IVR calls, and b) the mixed-technology group (IVR calls and enumerator follow-ups).

In 2) within the set of individuals that receive IVR calls, we introduce another treatment, by varying the delivery and context of the introductory messages within the IVR calls. Within the set of individuals allocated to IVR call treatment, the basic group will be those receiving the call that was recorded by a male enumerator. Against this control arm, three different treatment arms will be tested: female enumerator, religious leader and medical doctor

In 1) the enumerators follow-ups will also be randomized between men and women, as well as a mixed group, to understand the role of gender in achieving better responses.

Intervention Start Date

2020-09-23

Intervention End Date

2020-12-09

Primary Outcomes

Primary Outcomes (end points)

Take-up, response rate

Primary Outcomes (explanation)

Take-up - consent indicator if respondent agreed to participate in the interview.
Response rate-total number of sensitive questions answered.
Response rate-total number of sensitive health questions answered.
Response rate-total number of sensitive non-health questions answered.
Response rate -indicator if question on number of household members falling sick in the past 14 days, was answered by the respondent.
Response rate -indicator if question on number of household members falling sick with COVID-like symptoms in the past 14 days, was answered by the respondent.
Response rate-indicator if question on whether respondent has left the village in the past 7 days, was answered by the respondent.
Response rate-indicator if question on whether respondent has attended a social gathering in the past 7 days, was answered by the respondent.
Response rate-indicator if question on whether respondent has done paid work outside home in the past 7 days, was answered by the respondent.

Secondary Outcomes

Secondary Outcomes (end points)

Response quality

Secondary Outcomes (explanation)

Quality - under-reporting on sensitive health questions-number of household members reported to have been sick in the past 14 days.
Quality - under-reporting on sensitive health questions-number of household members reported to have been sick with COVID-like symptoms in the past 14 days.
Quality - under-reporting on sensitive non-health questions-number of days respondent traveled outside the village in the past 7 days.
Quality - under-reporting on sensitive non-health questions-number of days respondent worked out side home in the past 7 days.
Quality - under-reporting on sensitive non-health questions-indicator if respondent attended social gatherings in the past 7 days.

Experimental Design

We follow a two-stage randomization process. In the first stage, the technology innovation will be randomized across the entire population. Of the total population, 30% will be called via IVR calls in each round of follow-up, whereby 15% will receive follow-ups purely via IVR, while the remaining 15% will receive it in combination with enumerator follow-ups. Given the unequal distribution fo the sample between IVRs and enumerator follow-ups, the IVR calls will not take place on a biweekly basis for each individual, but bi-monthly instead. In contrast, the enumerator follow-ups in the mixed group will take place on a biweekly basis.
Thereafter, in the second stage, those who receive IVR calls will be randomized into the fours arms of equal size, i.e. 25% of the overall sample will each receive male enumerator, female enumerator, religious leader or male doctor calls. Once randomized into their groups, the individuals will always receive the same treatment. Those receiving enumerator follow-ups will be randomized into three arms: purely female enumerator follow-up, purely male enumerator follow-up and a group where male and female enumerators will call in alternative rounds.

Experimental Design Details

Randomization Method

We apply a computer based randomization for each round of data collection.

Randomization Unit

Randomization is at individual level.

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

0 - the study is not clusetered

Sample size: planned number of observations

24000

Sample size (or number of clusters) by treatment arms

Pure IVR calls- 3600 (Male enumerator IVR- 900, Female enumerator IVR- 900, Religious leader IVR- 900 and Male doctor IVR- 900)
Pure enumerator follow-ups: 8400 (TBD)
Mixed group (IVR+ enumerator follow-ups): 16800 (3600 IVR+ 8400 enumerator Follow-ups)

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Within the main treatment arms, a minimum detectable effect size of 75% is expected, at a base value of 32% response rate for enumerator calls. The IVR should encourage higher response rates to around 39% from the population. This is the most conservative estimate, where the MDES for less conservatives estimates were around 35%.

Supporting Documents and Materials

IRB