Lessons from the field: implementing an electronic clinical decision support app for acute febrile illness in rural Cambodia.

To improve the management of acute febrile illness in rural low- and middle-income country (LMIC) primary care settings, we developed the Electronic clinical Decision support for Acute fever Management (EDAM) app and evaluated it through a cluster-randomised trial in rural Cambodian primary health centres (PHCs). In parallel, we conducted structured, mixed methods observations at a subset of participating PHCs to document key challenges in screening, enrolling and managing patients. A key lesson was the critical importance of user-centred design to align digital tools with health worker workflows and capabilities. The lessons from these observations may help guide researchers and policymakers developing novel digital health solutions for low- and semi-skilled health workers in rural LMIC settings. Our experiences could offer valuable insights, as there are few documented digital health interventions designed for these contexts. Furthermore, navigating the challenges of implementation and subsequent evaluation remains a significant hurdle for the field.

Introduction

Acute febrile illness (AFI) is a common presentation in primary health centres (PHCs) across South and Southeast Asia (1). Historically, malaria was the predominant cause of AFI, but its burden has significantly declined, particularly in Cambodia, where malaria cases decreased by 76% between 2015 and 2022 (2). Consequently, the relative burden of non-malarial AFIs has increased considerably (2). This trend exposes significant challenges in their management, at both the individual patient and health system levels.

This is amply illustrated in rural Cambodia, where PHCs are typically staffed by healthcare workers (HCWs) who face significant challenges in diagnosing and treating non-malarial AFI due to limited access to diagnostic tools, reliance on non-user-friendly paper-based guidelines, and limited clinical skill development (3). These factors contribute to the overprescription of antibiotics, a key driver of antimicrobial resistance (AMR), which in 2019, was estimated to directly cause over 3,000 deaths in Cambodia (4).

To help address these problems, we co-created theElectronic clinicalDecision support forAcute feverManagement (EDAM) app for use in rural PHCs in Battambang province with local medically trained expert clinicians and Provincial Health Department (PHD) officials. The EDAM app is a rule-based algorithm designed to optimise the diagnosis and management of AFI by integrating clinical symptoms, vital signs (including peripheral oxygen saturation), and malaria and C-reactive protein (CRP) rapid test results to formulate a syndromic diagnosis and recommend a management strategy. We aimed to identify the key challenges and lessons learned from implementing a pragmatic cluster-randomised trial of EDAM in 30 PHCs, with the primary outcome measure being reduction in antibiotic prescriptions (5).

Methods

Clinical trial procedures

The trial methodology is described elsewhere (5). Briefly, PHCs in three Operational Districts (ODs) were randomized 1:1 to one of two arms: 15 PHCs in the intervention arm (EDAM-guided management with associated training) and 15 PHCs in the control arm (routine care). Each PHC aimed to recruit 152 patients, totaling 4,560 participants aged ≥1 year. Prior to the trial, HCWs received two days’ training including clinical scenarios to familiarise them with its use. Enrollment support included access to standard operating procedures (SOPs) via an electronic drive and direct contact with the study team via a Telegram group chat.

Structured observation procedures

The mixed-methods study presented here was embedded as part of the trial implementation. In each Operational District, 4-5 PHCs were selected for observation. Selection was non-random, focusing in particular on PHCs where enrolment rates were low and/or where the ratio of screened to enrolled patients was lower than expected. All research staff underwent a two-day standardised training session prior to data collection. This included a detailed review of the EDAM algorithm and study SOPs, hands-on practice sessions (e.g., use of the clicker counter for respiratory rate measurement, blood pressure measurement, pulse oximetry, and CRP testing), as well as training on correct interpretation and error classification. At each PHC, these trained research staff observed 12-15 consultations of enrolled patients, aiming to represent a variety of different years of experience of the HCWs enrolling the patients and different age groups of the enrolled patients. A checklist (Supplementary Appendix 1) was designed to cover several key themes: observation details (including patient socio-demographic data), algorithm utilisation, technical issues, time to form completion, and study procedures. Each individual step of the algorithm was assessed by the research staff (Figure 1, B), with a description of any errors and perceived causes documented for incorrect use. In addition, the time taken to complete each step was recorded. Discrepancies in the clinical information documented in the PHC logbook and the EDAM app were documented and their possible cause evaluated.

Study outcomes and analysis

The outcome measures were the time taken to complete each section of the algorithm and the overall time to form completion; the frequency, description and possible causes of incorrect use of the app; and inconsistencies in collected clinical data. Outcomes were collated into two key themes according to their implications for future research, namely app-related and contextual challenges.

Results

Conclusion

This study yielded useful insights into the challenges and opportunities of implementing a digital health intervention in a rural LMIC setting. Key lessons include the importance of user-centred design, intensive training, stakeholder engagement, and contextual adaptation. While the app showed promise in optimising AFI management, its success depended on addressing time constraints, technological barriers, and local challenges, which may have contributed to the desired primary outcome of reduced antibiotic prescribing not being reached in the trial (5). The findings offer guidance for researchers and policymakers developing and evaluating digital health solutions for low- and semi-skilled HCWs in similar settings. The lessons learnt are valuable for these stakeholders, since digital health interventions intended for rural LMIC contexts are currently uncommon but will likely become more so, and there are unique challenges associated with their implementation and subsequent evaluation.

References

ShresthaPDahalPOgbonnaa-NjokuCNon-malarial febrile illness: a systematic review of published aetiological studies and case reports from Southern Asia and South-eastern Asia, 1980–2015BMC Medicine2020181299 doi.org/10.1186/s12916-020-01745-0
World Health OrganizationWorld malaria report 2023World Health OrganizationGeneva2023Licence: CC BY-NC-SA 3. 0 IGO
DittrichSTadesseBTMoussyFTarget Product Profile for a Diagnostic Assay to Differentiate between Bacterial and Non-Bacterial Infections and Reduce Antimicrobial Overuse in Resource-Limited Settings: An Expert ConsensusPLOS ONE2016118e0161721 doi.org/10.1371/journal.pone.0161721
Institute for Health Metrics and EvaluationGlobal Burden of Disease Study 2019 (GBD 2019) Resources: CambodiaIHMESeattle, WA2020cited 2025 May 12[Internet] Available from:
WynbergEMishraALiveraniMImpact of an electronic clinical decision support algorithm (eCDSA) on antibiotic prescribing in primary care in Cambodia: a cluster randomised controlled trialInt J Infect Dis2026154107876 doi.org/10.1016/j.ijid.2026.108382

Republished from the open web under CC-BY. Authors: Mishra A, Wynberg E, Liverani M, Vanna M, Chanpheakdey P, Nguon C, Callery JJ, Adhikari B, Tripura R, Chandna A, Fegan G, Waithira N, Peto TJ, Voeurng B, Davoeng C, Rekol H, Dysoley L, Day NPJ, Lubell Y, Chew R. Read the original.