The Last Mile of Care: What Kenya's Community Health Workers Are Missing

Kenya has one of the most ambitious community health programmes in sub-Saharan Africa. Under the Community Health Policy 2020-2030 and the broader Universal Health Coverage agenda, the country has built a network of Community Health Promoters (CHPs) — until the Primary Health Care Act, 2023 reclassified them, these were Community Health Volunteers (CHVs) — each responsible for a defined household roster, each conducting regular household visits, referring sick members to higher-level facilities, and collecting health data on the populations they serve.¹ The current official norm, set out in the Ministry of Health's own 2024 clinical guidelines, is one CHP per 20 households, or about 100 people.²

In principle, this is a remarkable infrastructure. A trained health worker with a defined caseload, visiting households regularly, collecting consistent data, and linking the community to the formal health system. In practice, it is a remarkable infrastructure that, as of the government's own most recent published evaluation, is still well short of fully built — though, as §5 takes up directly, it is also already more digitised than a purely paper-based account would suggest.

The CHPs record household visits using the nationally approved MOH Household Register (Form MOH 513). A Community Health Assistant (CHA) — officially responsible for up to 20 CHPs each, per the same 2024 guidelines² — collects and reviews this material, and Community Health Units (CHUs) are expected to forward summarised data for upload into Kenya's national health information platform, DHIS2 (rebranded KHIS2), which Kenya was the first country in sub-Saharan Africa to roll out, in 2010-2011.³ By the time data reaches that level of aggregation, in the paper-register parts of the system that still exist, it has typically been manually re-entered, is often weeks old, and has been stripped of the contextual detail that made it useful at the household level.

This is not a criticism of the people in the system. A 2025 peer-reviewed photovoice study of Nairobi's CHPs found them occupying what its authors call a "paradoxical position" — essential to the health system, materially under-supported by it.⁴ The problem this paper argues is structural: CHPs have been given a mandate to generate operational health intelligence for Kenya's communities, and large parts of the system built to receive that intelligence were, until very recently, not built for the job.

The visit happens regularly, for every household. Much of what it produces still doesn't survive the trip back — though eCHIS has narrowed that gap.

1. A claim grounded in the government's own evidence, not ours

This paper argues that Kenya's community health data pipeline is still, on the government's own published and press-reported evidence, losing a meaningful share of what its frontline workers collect — and that the gap is tractable with better field software. Here is precisely what grounds that claim.

What it argues from. The Ministry of Health's own Community Health Policy 2020-2030 and its cited 2018 national evaluation of community-unit and workforce coverage, fetched and read directly.¹ The Ministry's 2024 clinical guidelines for Level 1 community services, fetched and read directly, for the current official household-ratio and supervision-ratio norms.² The Primary Health Care Act, 2023, fetched and read directly, for the legal basis of the CHP role and county governments' statutory duty to supply tools and stipends.⁵ A peer-reviewed mixed-methods study of CHW data quality in two Kenyan counties, read in full.⁶ A peer-reviewed account of Kenya's national eCHIS digitisation rollout, read in full.⁷ Press reporting on the frontline reality of that same rollout.⁸ A 2025 peer-reviewed photovoice study of Nairobi CHPs.⁴ Flux's own fieldwork in one Kiambu sub-county, labelled throughout as internal and unaudited.

What it does not argue from. No nationally representative, independently audited measurement of community-health data completeness exists; the national-level figures this paper cites (coverage gaps, workforce gaps) are themselves drawn from the Ministry's own evaluations, which have their own limitations as self-reported government data. Flux's Kiambu table in §3 is a single sub-county snapshot from Flux's own early research, not a peer-reviewed or independently audited study, and is not generalised beyond itself.

What it does not claim. It does not claim Kenya's community health system is still purely paper-based; §5 states directly, as a genuine complication to this paper's own prior framing, that a national digital system (eCHIS) has been rolled out to all 47 counties and reports high utilisation. It does not claim the digitisation already solves the problem this paper describes; §5 and §9 both engage press-reported frontline failures of that same system directly. It does not claim Flux's own Kiambu data generalises nationally; it is one sub-county's internal snapshot, presented as illustrative, not as proof.

What would refute it. Given in full in §11. The sharpest version: if Kenya's national eCHIS rollout is shown, by independent measurement rather than government or vendor self-report, to have already closed the specific gaps this paper describes — referral-outcome tracking, real-time supervisor visibility, longitudinal household records — the central practical argument of this paper fails, and the right response is to say so, not to relocate the argument to a narrower claim.

2. What community health data is supposed to do

The purpose of the CHP programme is not just service delivery — it is early detection. A CHP who visits households regularly is in a position to notice things that no clinic-based system can see: the child who has lost weight visit over visit, the elderly person whose mobility has declined, several household members developing coughs within the same week, the pregnant woman who has missed scheduled antenatal-care visits.

These are signals. In aggregate, across a sub-county, they are early warning systems for malnutrition trends, disease surveillance, maternal health gaps, and chronic condition deterioration. In a well-instrumented system, they would flow up the health hierarchy quickly, trigger responses, and generate a continuously updated picture of community health status.

Kenya's community health architecture was designed with this vision in mind. The question this paper asks is not whether the design is right — it is whether the data collection and transmission infrastructure, as it actually exists today across 47 counties, matches the ambition yet.

Where the infrastructure is still paper-based, or where a digital system exists but is unreliable at the point of care, these signals mostly still disappear in the way they always did: recorded in a register or an app that cannot sync, collected or transmitted at some later point, and by the time anything aggregated reaches a supervisor or facility manager, the specific household with the specific pattern has been lost in aggregation. The early warning system exists on paper, or on a phone with no connectivity, more often than it exists in practice.

3. The register problem, and what Flux's own fieldwork found

The paper register remains, in large parts of the system, the defining operational constraint. Household registration itself is still done using the nationally approved MOH Household Register, Form 513.¹ Everything a CHP does where digital tools have not yet reached, or are not working, is mediated through that register: household visits, referrals, follow-ups, births and deaths, new household members, nutrition screenings.

This creates several compounding problems where it is the operative system.

Longitudinal tracking is difficult. A paper register is organised by visit date. Understanding a household's health trajectory over six months — the unit of analysis that matters for chronic conditions, nutrition monitoring, and pregnancy tracking — means physically leafing through multiple register pages to find and reconcile entries for the same household across different months. In practice, this often does not happen; CHPs know their households from memory more than from structured records.

Referral follow-up is frequently broken. When a CHP refers a household member to a facility, the register records the referral. Whether the person actually went, whether they were seen, what happened — none of this reliably comes back to the CHP in a paper-only system. The link from community to facility functions, in that case, as a recommendation rather than a closed loop.

Data quality degrades in ways that are now independently documented, not just observed by us. A 2020 peer-reviewed mixed-methods study — conducted in Kitui and Nairobi counties in Kenya, and in Malawi — found that only 15% of community-health-worker data was reported consistently between CHWs and their supervisors over a three-month period across 339 registers and 72 summary reports.⁶ The same study's qualitative interviews documented CHWs and supervisors describing data "cooking" under reporting pressure — one supervisor's own words: "if they are not there and you are told to submit a report, you have to cook" — alongside register-completion gaps, tool shortages, and a general lack of trust in the data once it reached the next level.⁶ This is an independent, peer-reviewed finding, not an estimate this paper is making on its own authority.

One sub-county, Kiambu County, 2024 — Flux's own early-research fieldwork. Internal and unaudited; not a peer-reviewed or nationally representative measurement, and not claimed to be one.

This table is Flux's own data, from a single sub-county engagement conducted as part of early research for a possible community health product, and this paper states that plainly rather than letting it read as an independent study. What it shows — reasonable visit completion, collapsing data completeness further along the pipeline, real lags, and near-total failure of referral-outcome tracking — is directionally consistent with the independently published Regeru et al. 2020 finding above, though the two are not the same measurement and should not be cited as if they were.

4. What community health data quality looks like outside Kenya's own self-reports

It is worth asking directly whether the picture in §3 is Flux's own framing of a problem that happens to justify a Flux product, or whether it holds up against evidence Flux did not produce. The honest answer is: partly.

The Regeru et al. 2020 study cited above is the strongest independent evidence this paper has for the general pattern — community-health data degrading in both directions (both under- and over-reporting, with no clean pattern by site, worker, or month) as it moves from register to supervisor to national system.⁶ It was conducted in exactly two of the counties relevant to this paper's claims (Kitui, a largely rural county, and Nairobi, Kenya's largest urban county), which is a narrower base than "Kenya's community health system" as a whole, and it predates both the Primary Health Care Act, 2023 and the national eCHIS rollout discussed in §5 — meaning it describes the paper-based system this paper is partly arguing has since started to change.

The 2025 photovoice study of Nairobi CHPs adds a different kind of evidence: not about data quality directly, but about the working conditions — under-resourcing, inconsistent pay, being treated as "just like" the community members they serve rather than as health-system staff — that plausibly explain why register completion and follow-through are inconsistent in the first place.⁴ That paper does not measure data-quality outcomes itself, and this paper does not claim it does; it is cited for the structural account of CHP working conditions, not for a data-completeness figure.

5. The complication: Kenya has already started digitising this

Kenya's community health system is not still paper-only nationally. A national digital system has already been rolled out, and treating this as a purely paper-based problem would misdescribe where things currently stand.

Kenya's electronic Community Health Information System (eCHIS) piloted in Kisumu County in mid-2021, began nationwide rollout in mid-2022, and reached full national implementation across all 47 counties by mid-2024, with a reported 95% utilisation rate among the roughly 107,000 CHPs it targets — read directly from a 2025 peer-reviewed account of the scale-up.⁷ eCHIS supports household enrolment and service tracking, supply-chain management, client messaging, and a health information exchange intended to connect community-level records with facility electronic medical records, enabling what the same source describes as supported referrals from CHPs to health facilities, with standard forms and centralised dashboards.⁷ If that account is taken at face value, several of the problems this paper describes in §§2-3 — longitudinal tracking, standardised reporting, supervisor visibility — are substantially addressed already, for the roughly nine in ten CHPs the same source reports as active users.

The same peer-reviewed account, read in full rather than selectively, also documents real limits to that rollout: insufficient server capacity causing training postponements during the pilot; delayed smartphone deliveries in several counties, with some CHPs resorting to personal phones because "not all of us received them"; heavy donor dependence (reported at roughly 95% of funding); high CHP attrition after training; and one county, Tana River, that remained unimplemented for lack of partner support as of the study period.⁷

Press reporting from 2025 complicates the picture further, and in the direction this paper's original argument anticipated. A Nation Media Group investigation found CHPs in Nairobi and Nyamira counties reporting frozen phones, unreliable servers, and inadequate support, "forcing them back to notebooks and manual reconciliations" despite the national rollout — including a documented case of a CHP in Nairobi's Mukuru kwa Njenga settlement whose digital referral for a sick man was not acted on with any priority at the receiving facility, where he queued like any walk-in patient.⁸ This is press reporting, not a measured failure rate, and this paper treats it as order-of-magnitude evidence of a real gap between official utilisation figures and reliable field performance — not as a quantified estimate of how often the system fails.

Read together: the specific software gap this paper originally identified has been substantially, if unevenly, addressed by the government's own digitisation programme — a genuine complication for a paper partly written to justify building that software — but the scale-up's own documented limitations (connectivity, device reliability, donor dependence, uneven county coverage) mean reliable field-to-system data flow is not yet solved everywhere eCHIS has nominally reached.

6. What a well-instrumented system would look like

Independent of whether that system is eCHIS, a future iteration of it, or something else, a well-instrumented community health system would have several properties, and it is worth stating them plainly rather than assuming the current rollout already delivers all of them everywhere.

The CHP's device is the reliable source of truth, including offline. Not merely a digital form, but one that is offline-first by design — the visit is recorded to local storage regardless of connectivity and synced when available, so a server outage or a dead data connection at the facility end does not force a CHP back to a notebook, which is precisely the failure mode the 2025 press reporting documents.⁸

Longitudinal household records are built automatically, not reconstructed from registers. Every visit adds to a persistent record for each household and household member. Patterns — declining nutritional status, repeated respiratory symptoms, missed ANC visits — can be surfaced automatically rather than relying on a CHP or CHA to manually detect them across paper or disconnected app records.

I supervise my CHPs and I have to go through their reports and try to make sense of what happened. By the time I have finished, the month is almost over and I am already behind on the next one. I am always looking backwards, never forwards.

— Community Health Assistant, Kiambu County (Flux fieldwork)

Referral loops are closed, and verified, not merely logged. When a CHP makes a referral, the system flags it for follow-up. If the referred person does not show up at the facility within a defined window, the CHP is notified. Where facility data is available — which the eCHIS health-information-exchange component is designed to provide — the outcome should flow back to the CHP's own record, closing the loop rather than, as the Mukuru kwa Njenga case in §5 illustrates, carrying a referral forward without the receiving facility treating it as anything other than an ordinary walk-in.⁸

Supervisors have a real-time view that degrades gracefully, not catastrophically, when connectivity fails. A Community Health Assistant — officially responsible for up to 20 CHPs under current guidelines² — should not have to wait for either paper registers or a frozen app to understand what is happening across their caseload. The 2024 guidelines' own 1:20 ratio is itself optimistic relative to the workforce gap the 2020-2030 policy documented: as of its cited 2018 evaluation, Kenya had only 1,569 Community Health Assistants against an expected 10,379 — a gap of 85%¹ — meaning the real supervisory load in many places is likely well above the official norm, which makes a resilient, low-maintenance system more important, not less.

7. The harder problem: integration

Building a reliable field tool, on the evidence in §5, is now substantially underway in Kenya rather than purely a greenfield problem. The harder, still-unresolved problem is deep integration with the broader health system.

Kenya's facility-level health information system is DHIS2 — rebranded KHIS2 — which Kenya was the first country in sub-Saharan Africa to deploy as a fully online national system, completed in 2011, and which remains the backbone platform for facility-reported, rather than community-reported, data.³ The Community Health Policy 2020-2030 already specified that CHU-level summary data should be "analysed, summarised, and forwarded for uploading" into DHIS2 by CHAs and Community Health Officers¹ — a manual aggregation step, by the policy's own description, rather than a direct system-to-system feed. eCHIS's health-information-exchange component is a more recent attempt to close that gap directly, connecting community records to facility electronic medical records rather than routing through a manual summarise-and-upload step.⁷

For a community health tool to fulfil its potential, CHPs need to know which of their households have active facility records, which have upcoming appointments, and what happened when they referred someone previously. Facilities need to know which households in their catchment have been flagged by CHPs for follow-up. Neither serves the other well in informational isolation — which is exactly the failure the Mukuru kwa Njenga referral illustrates in §5, where the referral existed in a CHP's system but conferred no priority at the facility that received the patient.⁸

The Primary Health Care Act, 2023 gives this integration a statutory basis it did not previously have: it defines the CHP role in law, requires each county government to supply CHPs with "working tools" and "reporting tools" alongside training, stipend and supervision, and deems existing CHVs to be CHPs under the new framework.⁵ What the Act does not specify is a technical integration standard between the community-level and facility-level systems; that remains an operational, not a legal, gap.

This is the pattern we see across all of Flux's research areas: the technology to connect these systems exists, the APIs exist, the standards increasingly exist. What has been missing, where it is still missing, is operational software worth integrating — software that is actually used reliably at the point of care, that produces clean data, that generates enough institutional value to justify the integration investment.

8. Why this is on our roadmap

We have not yet built a community health product. We are in research. But we are in research with conviction, and with a clear-eyed sense of what already exists and what still does not.

Kenya's CHPs are among the most distributed operational health infrastructure in East Africa — roughly 107,000 trained workers nationally as of the government's most recent stipend rollout, each with a defined household roster, each conducting regular household-level contact with populations that no facility-based system can reach at this resolution.⁹ This is an extraordinary asset, and the national eCHIS programme has already moved it substantially closer to its potential than it was. It is not, on the evidence in §5, operating at that potential everywhere yet — connectivity, device reliability, and donor-dependent funding remain real, documented constraints.

The software problem that remains is narrower than it was: not "build a digital field tool from nothing," but "build, or contribute to, a field tool and integration layer that is reliable under the connectivity and device conditions Kenyan CHPs actually work in, and that closes referral loops all the way to the facility, not just to the next level of community supervision." We know how to build offline-first field tools; it is the core of our own AccessWASH product, deployed across Nairobi-area water utilities' field operations today. Whether that experience transfers cleanly to community health — a different regulatory environment, a different workforce, a system with its own national digitisation programme already underway — is precisely the question our current research is trying to answer, not something this paper assumes.

What we still need to understand is the specific operational workflow of the CHP under eCHIS as it actually runs today, the specific gaps Community Health Assistants experience in practice, and the specific points where Flux's own offline-first and closed-loop experience would add something the existing national system does not already provide. We are spending time in sub-counties, with CHPs, with CHAs, with facility managers. Building from observation, not assumption — and from what the government's own digitisation programme has already achieved, not around it.

When we build anything here, we will build offline-first and closed-loop, in whatever gap remains once eCHIS's own evolution is accounted for. And we will publish what we learn regardless of what the product becomes.

— Ken Ruto

9. Objections I take seriously

"Flux has a commercial interest in this conclusion." Flux is explicitly in research toward a community health product, and this paper's central argument — that a data-flow gap remains worth solving — is, if borne out, good for exactly that research programme. The honest mitigation is narrow: the Regeru et al. 2020 study, the eCHIS scale-up paper, and the 2025 press reporting are all independent of Flux and would show the same patterns whether or not Flux existed. What is not independent is the Kiambu fieldwork table in §3, which is Flux's own, unaudited, single-sub-county data — labelled as such throughout rather than presented as confirming evidence.

"The national eCHIS rollout has already solved this, and this paper is arguing against a problem that no longer exists." This is the objection §5 takes most seriously, and it is not fully answerable either way on current evidence. The 95% utilisation figure is from a peer-reviewed source, not Flux's own claim, and taken at face value it is a serious objection to this paper's framing.⁷ But utilisation is not reliability: the same source documents real device, connectivity and funding limitations, while 2025 press reporting documents CHPs reverting to notebooks in named locations.⁷⁸ This paper holds both things true at once — substantial progress, and a real remaining gap — rather than picking whichever figure is more convenient for Flux's research interest.

"A single sub-county's internal, unaudited data (§3) should not be used to illustrate a national argument." This is correct as stated, and §1 and the table caption in §3 say so directly rather than letting the table imply more than it supports. The table is retained because it is Flux's own direct field observation and is directionally consistent with the independently published Regeru et al. 2020 finding, not because it is independently probative on its own.

"Crediting a government programme's limitations as the reason a private company should build something is self-serving." It is a fair challenge, and this paper does not have a clean answer beyond naming it: Flux's own research interest and the public-interest case for better community health data flow point in the same direction here, and a reader should weigh the evidence in §§4-5 rather than the fact that Flux also benefits if the conclusion holds.

10. What this account doesn't establish

Kenya's community health system is not still primarily paper-based nationally. §5 states that directly, based on a peer-reviewed account of the eCHIS rollout.

It does not claim the eCHIS rollout has failed; it claims the rollout's own documented limitations (connectivity, device reliability, funding dependence, uneven county coverage) leave a real, if narrower, gap — a claim grounded in the same peer-reviewed source plus independent press reporting, not in Flux's own assertion.

It does not claim the Kiambu fieldwork table in §3 is nationally representative, peer-reviewed, or independently audited; it is explicitly labelled as Flux's own single-sub-county, internal research.

It does not claim a measured, population-level referral-outcome completion rate for Kenya's community health system as a whole; no such national measurement was found, and the 14% figure in §3's table is this paper's own sub-county snapshot, not a national statistic.

It does not claim Flux has independently verified the 95% eCHIS utilisation figure or the 107,000 CHP national workforce figure; both are taken from a peer-reviewed source and a Ministry/NGO press account respectively, read directly but not independently re-measured by this paper.

11. What would falsify this

Stated as commitments, in decreasing order of how much damage each would do to this paper's claims.

An independent, non-government, non-vendor measurement finds that eCHIS, as actually deployed, already closes the specific gaps this paper describes — referral-outcome tracking, longitudinal household records, real-time supervisor visibility — at or near the rates the Ministry and the 2025 peer-reviewed scale-up account report. That would directly contradict this paper's central claim that a meaningful gap remains, regardless of how well-documented the rollout's own stated limitations are, and the right response would be to withdraw the operational argument, not to relocate it to a narrower claim.

Kenya's own national data shows referral-outcome completion, longitudinal tracking, or supervisor-visibility metrics improving to a level comparable with the Regeru et al. 2020 finding's inverse (i.e., data consistency well above 15%) specifically because of eCHIS, independent of Flux's own involvement. That would be strong evidence the software gap this paper and Flux's own research programme are oriented around has already substantially closed, undercutting the commercial rationale in §9 directly.

Flux's own future community health research finds that the operational gap it expected to find, based on this paper's argument, does not exist at the scale assumed — i.e., that eCHIS as currently run already serves CHAs and facility managers adequately in the sub-counties Flux actually studies. That would not falsify the national picture, but it would falsify the specific premise motivating Flux's own roadmap, and should be reported here regardless of what it implies for the business case.

Press reporting of frontline eCHIS failures (§5, §9) turns out, on closer independent inquiry, to be unrepresentative or substantially overstated relative to the national 95% utilisation figure. That would narrow this paper's claim considerably, though it would not eliminate the documented structural limitations (donor dependence, uneven county coverage) the peer-reviewed scale-up account reports in its own right.

Appendix A — Evidence table

Appendix B — Citation register

CITATION-NEEDED — a national, independently audited measurement of community-health data completeness. Appendix A row 16. All national-level figures in this paper (CHU coverage, workforce gaps, eCHIS utilisation) are drawn from government or vendor/partner-affiliated evaluations; no fully independent national audit was found.

CITATION-NEEDED — a specified technical integration standard between eCHIS and DHIS2/KHIS2. Appendix A row 17. The Primary Health Care Act, 2023 creates a legal duty to supply "working tools" and "reporting tools" but does not specify a technical data-exchange standard between community- and facility-level systems; no such standard was found published.

Correction, not merely flagged. The previous version stated the ratio as "one CHP per 100 households" and "twelve to fifteen CHPs" per CHA. The Ministry's current (2024) clinical guidelines state 20 households (~100 people) per CHP, and 20 CHPs per CHA. Both corrected throughout; the 100-households figure appears to have conflated the per-CHP household count with the equivalent people-count the same guidelines state as an alternative measure, not an additional one.

Correction. The previous version's footnote cited "Ministry of Health, Kenya (2023); see also Amref Health Africa" without naming a document. This revision cites the Community Health Policy 2020-2030, the Primary Health Care Act 2023, and the 2024 clinical guidelines directly — each fetched and read in full.