Dawa Chat

As public health practitioners and researchers we are keenly aware of the importance of adherence to prescribed medication by patients (or their caregivers) in fulfilling several health interventions. To mention a few, these include the treatment, at-home, of pneumonia, malaria, and iron deficiency anaemia, all in the under 5 population, plus management of non-communicable diseases (hypertension & type II diabetes) in the older population.

Specifically, we would like to make it possible for healthcare professionals to evaluate medication adherence in their patients. Unfortunately, monitoring this “last mile” is difficult. Conventional methods to evaluate adherence include surveys of households several days after treatment was administered. However this method is vulnerable to recall bias or social-desirability bias. The use of drug (moiety) concentration monitoring in urine samples, may give the accurate indications that drug adherence was done, but is complicated by the costs of equipment/reagents, drug specific characteristics (renal clearance, half life in urine etc.) - in short, this would not scale or be sustainable as a public health intervention.

How big is the problem?
At public health scale, this “last mile” is largely a black box. Little is known, because not much data is available on medication adherence itself. Proxy indicators of adherence include resolution of symptoms, and, not returning to the health facilities. However, it could largely be the case that suboptimal adherence still results in resolution of the initial symptoms. These scenarios breed grounds for antimicrobial resistance or failed treatment; the latter leading to re-emergence of the initial symptoms. 

Using the example of adherence to the 3 day schedule of malaria treatment/prophylaxis when using Artemisinin based combination therapy (ACT) proper adherence has been found to be as low as 42.1% locally in Western Kenya[1], 67.9% in the Gambia[2] and 34.1% in South America[3]. Not only are these results disconcerting, they (as others) are all based on research studies that have rather small sample sizes, and offer no public health intervention level evaluation metric or data.

Additionally, it is not enough to evaluate whether adherence is taking place, but also grasp the challenges (caregivers of) patients are facing in achieving optimal adherence. This can promptly address amenable issues (e.g. providing suitable alternatives to a drug that causes adverse reactions, correcting wrong beliefs etc.) hence improving the final outcome.

============
1. Onyango, E.O., Ayodo, G., Watsierah, C.A. et al. Factors associated with non-adherence to Artemisinin-based combination therapy (ACT) to malaria in a rural population from holoendemic region of western Kenya. BMC Infect Dis 12, 143 (2012). https://doi-org.ezproxyberklee.flo.org/10.1186/1471-2...
2.Amponsah AO, Vosper H, Marfo AF. Patient related factors affecting adherence to antimalarial medication in an urban estate in ghana. Malar Res Treat. 2015;2015:452539. doi: 10.1155/2015/452539. Epub 2015 Feb 12. PMID: 25767736; PMCID: PMC4342176.
3. Factors associated with non-adherence to the treatment of vivax malaria in a rural community from the Brazilian Amazon Basin. Eduardo Dias AlmeidaJosé Luiz Fernandes Vieira,https://doi-org.ezproxyberklee.flo.org/10.1590/0037-8...

Our intervention seeks to engage with patients in their at-home medication journey. We design Dawa Chat to address the multiple interventions where at-home adherence is the common Achilles tendon. Examples include pneumonia, iron deficiency anaemia, TB (where adherence is needed for several days/months) and HIV, Hypertension, type II Diabetes where the need for adherence is lifelong. 

To demonstrate Dawa Chat’s potential, we will, in this grant application, use the short 3 day course of oral medication required in the at-home treatment of Malaria in under 10 year old children. We intend the end-user to be the immediate caregiver of the patient, literate in English, Swahili or Teso (local language), with access to at least a feature phone.

Dawa Chat seeks, through two-way text conversation, to 1) remind mothers of the next time they are to dose their children on treatment and 2), to know, from each of the dosing instances, how this process actually turned out noting any challenges.  In this way, Dawa Chat is not only an unconventional, engaging method of monitoring adherence, but may also improve the process itself!

At its core, Dawa Chat will be a text conversation over the 3 days during which a child is being administered ACT Malaria treatment by the immediate caregiver. 

Below highlights what the chain of texts in a typical conversation may look like!

___________________

? : Hi Mama, just a reminder. Baby Tom’s next dose (1 tablet) is at around 4 pm today

? : OK

? : Hi Mama. Were you able to give baby Tom all her 2 doses today?

? : he vomited the evening dose

? : And the morning dose, did that go well?

 ? : Yes

??‍⚕️ :Does the medication make him vomit or was he vomiting before the medication started

? : He was vomiting even b4

??‍⚕️ : Okay. Continue with the medication today. Let me know how it goes. I will check on you tomorrow morning!

__________________

Key:

? : Caregiver

? : Automatically sent (as Dawa Chat) to caregiver,on pre-scheduled time

 ??‍⚕️: Human in the loop responds (as Dawa Chat) to the caregiver.

---

These conversations are NOT scheduled text alerts per se. Our intention is to go further and have humans-in-the-loop (HITL) who are trained HCPs, actually converse with the caregivers to better understand the context of an issue. Caregivers respond in either free text format, or using structured responses (numbers). In the former case, the HITL would interpret the information and reply with a personified response. This approach hence is a hybrid of automation and human text conversation. While the former serves to increase the scale of the implementation, the latter increases the engagement of the caregiver with the platform.

What processes and technology does Dawa Chat use? 

Dawa Chat is based on simple technologies and easily integratable processes. On the part of the end-users, no software is needed other than a feature/smartphone. No internet connectivity is required. We outline the process, and technologies used in the following steps.

1. The caregiver to a sick child is registered onto Dawa Chat by the attending HCP via USSD or a Web/Android app using the caregiver’s telephone number. Details of the treatment regimen to be followed at home are also collected.

2. The caregiver will receive timely reminders on his/her phone on when to next administer medication. They are also encouraged to text on how the administration went, highlighting any challenges.Caregivers are encouraged to respond to what they “have done” and not what they intend to do (procrastinate).

3. The caregivers responses are relayed to the cloud server, and if necessary presented to a HILT who would further engage the caregiver by continuing the conversation

4. Possible outcomes of step 3 above are 

   1. Encouraging the caregiver on, 

   2. Addressing any issues possible within the context of the home

   3. Advising the caregiver to return to the health facility

After the steps above, another natural language processing based (NLP) algorithm will look at the chain of conversation and determine the percentage adherence the caregiver has achieved. For the use case of Malaria treatment, this shall be calculated as the number of doses actually administered divided by the expected number, within the treatment duration. This would enable categorization of adherence, given thresholds, into poor, moderate, good and complete adherers.

How does it incorporate inclusive human-centered design? 

A lot of effort will be put into how to phrase the reminders/texts that are sent out to the caregivers. We will incorporate qualitative research methods  (FGD, IDI, Structure brainstorming) engaging caregivers in the local community and use what is learned to understand how best to have text conversations that 1) remind caregivers to adhere, 2) enable caregivers to communicate most accurately/honestly of how they are adhering to the drug regime and 3) are able to collect information on when challenges are faced. 

How is it better and/or complementary to existing methods in low- and middle-income countries? 

Compared to conventional methods that have been used in low and middle income countries to evaluate adherence, Dawa Chat is potentially more scalable, able to reach a wider population. It leverages the fact that mobile phone (feature phone) penetration is > 95% in Kenya (LMIC). It also addresses two key challenges of conventional techniques; recall bias and social desirability bias (caregivers may feel more comfortable telling the accurate outcomes via text than face-to-face surveys or voice conversations). 

Additionally Dawa Chat provides patient level data, personifying the attention a caregiver needs, while being able to do so at scale.

Who is collecting your primary health care data? How and Why?

Data will be collected as text responses from the caregivers. This is actually user level primary healthcare data, entered as free text, and hence enabling caregivers to be as expressive as possible - a feature why we would expect better understanding of the challenges of adherence through the platform. 

Dawa Chat, will serve the caregivers of children on ACT based Malaria treatment at home (outpatient).

It will positively impact the medical adherence to therapy in a direct and meaningful way. Close to 50% of children on medication don’t achieve full adherence, making them vulnerale to prolonged illness, further complications (e.g. anemia) and resistance to antimalarials in the long run. Dawa Chat will address the various challenges their caregivers face, improving treatment outcomes.

We are based in Busia county in Kenya and comprise a medical doctor/health informatician, a social scientist and 3 malaria entomologists, all with a wealth of experience in research in malaria vector control, clinical management. As a team, we lead a large cluster randomised malaria study, in Busia, which has the highest transmission rates of Malaria in Kenya.

Our work involves home treatment and hence drug adherence as a key component. We evidently appreciate its importance in the fight against malaria and see a missing gap. We have reported several cases of timely treatment, but poor adherence, leading to prolonged and/or more severe illness.

Additionally, as part of our work, we engage with the local community directly. We anticipate the positive effect Dawa Chat would have on the adherence, and at a larger scale than what we have achieved via one to one interactions with caregivers.

We continually witness first hand the gap in medication adherence. We are not able to develop effective strategies to address it due limited financing to do so. We are technically versed with how we would go about about designing and implementing the platform but would benefit from additional funding to bring our idea to life.

Drug adherence is a very important last mile determinant of health outcomes. Todate, many strategies to measure this are either vulnerable to the responder’s bias or biological markers (e.g. urine drug concentrations) that are proxy markers and cannot be applied at scale. To the best of our knowledge, no approach has been used before to evaluate adherence, and at such scale within the LMIC context 

Dawa Chat not only promotes adherence, but would also accurately document richer data about the adherence to medication while learning challenges that caregivers face in the process. 

Dawa Chat is an adaptable and transferable intervention to other health care interventions that rely on good adherence monitoring. This increases its use cases, and versatility.

With time, after accumulating a sizable amount of data, Dawa Chat can leverage more NLP to augment the responses to caregivers, while flagging only the conversations that need HITL, further increasing its scalability. 

It can also be applied to other interventions such as treatment of oneumonia, anemia, type II diabetes, etc.

Year 1 impact goals:

1. Demonstrate increased ability to evaluate medical adherence, and at a scale of 100+ follow ups per week, with limited humans in the loop. This will be achieved by constant improvement of the platform, and informed by input from end-users.

2. Demonstrate improved adherence or health outcomes as a result of Dawa Chat: conduct a randomised control study comparing careegiver on/off the platform against health outcomes

Year 5 impact goals:

1. Expansion of the demongraphic areas/populations served by Dawa Chat, to include other regions, health implementatiopn organisations. This will be achieved thrrough strategic partnerships, having proved the value proposition of Dawa Chat

2. More leverage of AI, in particular NLP to scale reach of Dawa Chat. To be achieved after accumulation of sizale datasets in various languages, enough to train/test models that provide responses to caregivers that are medically sound and indistiguishable from HITL.

3. Application of Dawa Chat to other disease's intervention. To be achieved via strategis partnerships 

The most important metrics, which are user centered, are

1. user acceptance/uptake of the platform, measured as those who register on to the platform as a proportion of those it is presented to.
2. proportion of those who use the platform through the whole treatment period (3 days) and the fallout rate at each of the days

Design and implementation shall be guided to improving these two core metrics.

Other metrics to demontrate the impact Dawa Chat has in improving health outcomes are

1. comparing adherence rates between those on the platform and those not on it
2. comparing repeat/prolonged cases on malaria between those on the platform and those not on it
3. documenting numbers of cases that benefitted from use of the platform to complete their regimens

*An interesting metric would be the number of caregivers who use the platform to the end, while admitting sub-optimal/poor adherence - demonstrating their engagement on it.

Dawa Chat relies on the following core technologies; 

SMS technology - through user shall be registered (USSD) and have conversations with the Dawa Chat platform

 Cloud base servers - from which caregivers' messages shall be sent, and their responses either replied or relayed to HITLs

NLP - which in the initial stages shalb be used to process caregivers'  responses and determine the percentage of adherence. In the later iteration, NLP shall also play a key role in directly responing to caregivers.

Mobile app (Android) - which shall be used by HITL to communicate with the caregivers, offering then personalised guidance.

Dawa Chat is offered to the caregivers/patients who directly interact with it, with no intermediary (data collector) between. 

The caregivers would be motivated to interact with the platform because 

1) it offers them a channel to communicate on the progression of the medication (and hence health) of their patients, and 

2) as Dawa Chat keeps track and reminds them of the next dosing time, sparing them the mental load, while acting in their interest 

To address diversity, Dawa Chat will be keen to address 3 local languages spoken by the population in the areas we will first prototype it, i.e. English, Swahili and Teso. In this region, GSM connectivity is available and mobile phone penetration is high. The service does not require internet connectivity and it shall be offered free, with no fees incurred on the part of the caregivers; this addresses access and equity to the use of the platform.

Inclusivity is key consideration in the design of the texts/chat as a spectrum of literacy/education, religious background shall be reached - all with whom it will be vital for Dawa Chat to make them feel engaged. Personalisation of the contexts the varied caregivers/patients come from, and the content of the chats especially if challenges/adverse events are communicated is thus a keen design and training (of HITLs) consideration for Dawa Chat.

The mature Dawa Chat platform would target, as customers, health programme funding organisations, implementing programmes, health management teams, etc carrying out vertical interventions at scale. These organisations ideally would be those that appreciate the importance of improving/monitoring/evaluating ‘final mile’ medical adherence in their interventions.

The beneficiaries of the ‘Dawa Chat’ platform are the patients/caregivers who have a chance to receive reminders, communicate their journey and receive any advice on any adverse reactions or challenges they may face.

The Dawa Chat platform shall be offered to customers as a platform as a service (PaaS), and would have a wide range of interventions it can be applied to.

 
Our product proposition would be that increased/assured ‘final mile’ adherence justifies all the upstream investments made in an intervention and hence would be a cost effective component in several health services.

We estimate that via this MIT solve grant, we would be able to demonstrate the potential and versatility of our non-conventional approach to understanding medication adherence. However, to create a scalable product, we will need sustained donations and grants to accumulate more data from the text conversations and add a trained NLP model to handle the more common patterns in conversations hence increasing scalability, while delegating more complex ones to HITLs.

Having done this, such a product will be sold as an adherence monitoring service to health organisations/ programmes that are keen on the same. 

The Malaria Branch and Entomology section at the KEMRI - Centre for Global Health Research are recipients of multiple research grants from a wide range of funders. A few examples include a US$9 million grant to support the epidemiological evaluation of spatial repellents ending in Dec 2023, a US$7 million grant to support the epidemiological evaluation of Attractive Targeted Sugar Baits ending in June 2024, among others, and there are multiple discussions with Wellcome Trust UK and the Bill and Melinda Gates Foundation about other clinical research grants. These research grants co-support the salaries of core staff on this MIT Solve application hence it will not be burdened by salary support.

Secondly and most importantly, Busia County is very receptive to research and change that is likely to improve service delivery. For example, we are currently conducting the evaluation of scannable registers for data entry in collaboration with QED (https://qed.ai/). These registers are much better than the paper registers traditionally used by health facilities and have seen the reduction in errors in the facilities by upto 70%. At the moment, the county is committing funds towards the expansion of these registers to cover the entire county and therefore, provided the county sees the value of the innovation, they would be receptive to embracing it and supporting its implementation.