Community Health Toolkit

Adapting an existing CHT SMS-based innovation for COVID response: bot+human hybrid workflows

I recently worked with Medic Mobile @isaacholeman @korir to conduct a randomized control trial (RCT) that adapted the CHT model of nurse to community healthcare worker for direct two-way texting (2wT) between male circumcision (MC) providers and adult patients. This adaption has a lot of potential uses for COVID19 and beyond. For the RCT, the aim was to reduce the need for in-person follow-up visits. For MC, most countries and programs still require in-person follow-up to confirm healing. However, almost all men heal without issue, making these visits largely unnecessary. In our trial, instead of asking the men to return to the clinic to observe how they were healing, we counseled them on warning signs of complication, empowering them to be partners in their own healing. Then, we asked them to monitor their wounds at home. Over the next 13 days after MC, the Medic Mobile hybrid automated/interactive system sent a daily text to check in on the men, requesting a simple response of “0” if well or “1” if not text. If the men responded that they were well, that ended the interaction for the day. If the men reported a concern, a nurse would interact one-on-one with the client to determine if the man should return to the clinic for care. Many men were triaged via SMS, allowing them to remain home and heal after reassurance or additional guidance for wound care. There were no serious complications – the intervention was safe, and we reduced provider workload by 85%. It was cheaper than in-person visits. 2wT was also highly acceptable to patients: over 93% of men responded to at least one daily text. We have a lot of local buy-in for replication and scaling up, including a new model 2wT for MC project in South Africa. You can see more on the Zimbabwe RCT here.

This 2wT-based system has far-reaching potential for expansion to COVID19 or other acute conditions that require brief periods of follow-up (childhood diseases, respiratory infections, post-operative care that could similarly benefit from this type of intensive, direct provider-to-client communication. It has several advantages over other follow-up options. First, 2wT mixed automated and human-to-human messaging that optimized for individualized content, staff efficiency, and rapid follow up for the fraction of cases with that needed in-person care. Second, the short 13-day intervention maximizes client likelihood of response as phone theft, damage, and change of phone numbers are minimal over that brief period. Third, due to the high proportion of unnecessary visits, 2wT realizes immediate efficiency gains.

For COVID 19, the benefits of 2wT could also be quite advantageous. All those with phone numbers in a specific context could easily be enrolled if they’ve signed up for other forms of healthcare worker follow-up. We could use 2wT for contact tracing – sending a daily check in text on symptoms and having them respond (for themselves or their households) with the same “0” or “1”. Those with a 0 end their interaction while a nurse or lay healthcare worker could follow-up with those who have symptoms or concerns. The system could be used to triage patients, allowing those with mild symptoms to remain at home while giving support and directions for sicker patients on where to get care. The simple outgoing message about symptoms is not sensitive health information, reducing the risks to patients if the messages are seen by others. Scale-up and expansion of this system is timely and highly adaptable.


Thanks for sharing this idea Caryl!

I also really like the idea of adapting this kind of human-bot hybrid follow up for covid response. I’ve been working with @marc to create a demo for this and it looks like we’re pretty close to being able to support a draft workflow on our covid demo instance. We’re doing this through an integration with RapidPro, which enables some better interactivity than was possible in our initial trial, and makes it easier to connect new messaging channels. It currently works with SMS and Telegram, and we could look into Facebook Messenger and WhatsApp.

An important next step is to think about how we would approach clinical review and validation of this use case. To save health workers more time, it’s really helpful if most patients can have their needs met in an automated way by the bot, and only more difficult/clinical urgent cases or people with more complex questions get forwarded to a human. To that end, we need to clinically define one or more “off ramps” for people who messaged in to have their needs met, without getting forwarded to a human. The obvious first off ramp is for any person who says they have no symptoms.

Whether there’s potential for a second off ramp is where we’ll need good clinical/epidemiological review. Guidance for people with mild covid-like symptoms is typically to stay at home, so I’m wondering if we could have the bot provide that guidance, and only people who classify as having more severe symptoms get forwarded to a nurse. For example, we could say that anyone who has the symptom “difficulty breathing or hurts to breath” gets forwarded to a nurse, whereas anyone who has “fever, headache, aches & pains, or other symptoms not including difficult breathing” would be told that they have symptoms that may be associated with covid and should stay home, hand wash etc. What do you think? Who are the right people to review this?

A related question is, how important is it from an epidemiological perspective that the bot be able to record each individual symptom each person has? The challenge is that when we do this via messaging, the back-and-forth to note every single system could be tedious for the user and more costly for the system, and if it were too long some people might drop off before getting the guidance we want to give them. So, are there other specific symptoms we really want to track individually, or is some degree of catchall category acceptable?

To answer these questions, it might be helpful to focus in on a particular user persona. Given that health workers are 20-30x more likely to be infected, I’d be interested in targeting health workers with a daily prompt and symptom check. This group has the added advantage that many CHT implementations already have health workers’ phone numbers. Another possibility would be to create workflows for known vulnerable groups, e.g. persons over the age of 60.

Thanks again for sharing this idea Caryl, I’m stoked to build out this use case and demo. Part of why I think it’s so relevant here is that we’ve seen a remarkable amount of misinformation related to covid, with some even referring to covid as our first global infodemic. While bots save a lot of time, I don’t think simple automated systems are well set up to listen to complex questions, learn from the unexpected things people end up asking about, and respond in a personal way to the individual. But a health worker can do that remarkably well, even via text. If we make this concrete enough for the rest of the CHT community, and I think there’s a good chance people will want to implement it.


Hi @cfeld,

For the Zimbabwe RCT did you use a shortcode and reverse billing? If yes, who provided the shortcode and reverse billing service.

cc @Jonathan_Fernandes


1 Like

We didn’t use a short code or reverse billing for this study. We just used the Android SMS gateway with a normal econet SIM card and SMS bundles. Patients were given airtime at the end for participation in the study/to cover their SMS costs. @sue is this also the case for the scale up, or have we sorted out a short code? I know that we’ve done reverse billing for other projects with Africa’s Talking, but I don’t think they offer their service in Zimbabwe.


Thanks, @munjoma, for the question. @isaacholeman or @sue would have the answer for the Zimbabwe RCT context, but I will have other updates for the region coming soon. We are just about to start another project to scale up 2wT for male circumcision in RSA where we hope that the men would pay for the texts out of pocket with understanding that it would be cheaper than visits. In this same study, we will explore options to make it free for clients, including shortcode and reverse billing, if the initial results suggest that this user-pay model does not work. We will have findings in the coming year. We are also starting another project in Malawi on retention in ART care where we will also explore options that are free for clients. We will know more on what works best there in about 6 months. I’ll try to post those findings as we have them as well.


That is Correct @isaacholeman. We didn’t use the shortcode in Zimbabwe since Africa’s Talking does not provide the service in the country. We used the Android SMS gateway with a normal SIM Card and SMS bundles.