search

medic / config-covid

3 stars 6 forks source link

VMMC / RapidPro: Isaac #16

Open helizabetholsen opened 4 years ago

helizabetholsen commented 4 years ago

Hey @isaacholeman and team, could we configure this as a demo for health worker daily check-ins, modeled off the Audere HW self-assessment SMS workflow? Awesome bonus points if we can show that a health worker has reported a fever and isn’t cleared to work in that HW’s profile in the web app (e.g. context card). There are a few inquiries right now about HW health checks, and it would be super useful to show a demo of what RPro + CHT can do. cc: @helenelizabeth @michael

isaacholeman commented 4 years ago

I drafted a flow in RapidPro to show what it might look like to implement a VMMC style intervention for covid follow up. It’s titled [IH] Interactive semi-automated follow up - March 2020 covid. A few key points:

  1. This could be used for a general population, but to begin with at least it might make the most sense for identified at risk persons. e.g. health workers, people directed to quarantine at home after port-of-entry screening or contact tracing.
  2. It sends an automated daily message asking if they have any symptoms. Most people are expected to say that they don’t have symptoms and these people receive an automated thanks and BCC/educational message. This reduces the number of people who might visit a clinic or want to text with a health worker.
  3. For people who say that they have symptoms, the flow asks them which symptoms. It might be appropriate for some symptoms to remain in the “quarantine at home” category. These people might get additional educational messaging beyond what people with no symptoms would get. Appropriateness of this needs to be checked by clinicians. If it’s safe and appropriate, this would be a second “off ramp” in which people can interact with the bot and don’t need to visit a clinic or text a health worker.
  4. Anyone who reports fever, cough, difficulty breathing, or multiple of those symptoms gets connected to a nurse. If they type something the bot doesn’t understand, it asks them to once to clarify. If they type something else again, they get connected to a nurse for texting. This hand-off is important, we may want to iterate on letting them know that they can choose to text with a nurse at any time.
  5. If they don’t respond, the bot nudges them again some hours later. If they don’t respond after that, we should create a task for a nurse to follow up via text or voice call.

This isn’t a working flow yet, it’s just enough of a skeleton to convey what it would look like to support human-to-human messaging, with some automated “off ramps” for people who feel well and don’t need to burden the health system. Issues that would need to be worked out include:

  1. Clinical vetting of description of symptoms and appropriateness of off ramps
  2. We should examine how the language etc aligns with the client-initiated workflow @Katanu @michael @Bara
  3. Additional behavior change or educational messaging enhancements
  4. Polishing what happens when people don’t respond (I haven’t really completed this part for all branches of this flow)
  5. Drafting a registration workflow within the CHT. This might involve assigning people to catchment areas so that if they get sick, they can be directed to a nurse who knows what’s going on with facilities in their area.
  6. Figuring out how routing would work for people who need to text with a live nurse. I would imagine that we could create a RapidPro webhook that generates a Task for a health worker in the same catchment area as the person just registered. Would the nurses then register them from a separate number? Or is it possible to add those people to a group or a new flow in RapidPro where each time they text in, the content of their message gets automatically forwarded to the appropriate CHT user?

In case it’s helpful, here’s the protocol for the VMMC study where we supported such a workflow for 14 days of post-surgical follow up in Zimbabwe (led by Caryl Feldacker at I-TECH/University of Washington). That study showed that this kind of interactive messaging was 1) safe for VMMC; 2) more cost effective than standard care; 3) usable and preferred by patients. Figure 1 is a visual of what the messaging flow looked like for that study. https://trialsjournal.biomedcentral.com/articles/10.1186/s13063-019-3470-9