🔬 DAC2026 Summary & Wiki/AI: Research

Resources - Ressources
1

Research at DAC2026: Technical Methodologies, Case Studies, and System Implementations

:warning: Note: This summary was generated and reviewed by the dhis2 docs Ask AI tool and may contain errors. As this is a Wiki post, we encourage you to edit and improve this content with your own expertise, or reply with your questions for discussion!

This document compiles the research methodologies, system evaluations, and technical frameworks presented in the academic sessions at the DHIS2 2026 Annual Conference. [Research DAC2026]

1. Technical Framework: Four Streams of DHIS2 Research

The academic literature and studies surrounding DHIS2 are classified into four distinct research streams: [Research DAC2026]

  1. DHIS2 as a Primary Data Source: Utilizing routine, aggregated, or longitudinal HMIS database exports to conduct clinical, epidemiological, and demographic analyses [Research DAC2026].
  2. Data Quality Methodologies: Implementing and testing specialized validation models, audit applications, and data cleaning routines to improve dataset accuracy and completeness [Research DAC2026].
  3. Implementation Research: Assessing how organizational workflows, local capacity, training interventions, and user behavior affect system adoption and sustainability [Research DAC2026].
  4. Platform Technology Research: Studying the design patterns, software core developments, and standardizations required to balance international product configurations with localized country needs [Research DAC2026].

2. Action Design Research: Frugal Digital Innovation in Ethiopia

An ongoing Action Design Research (ADR) study conducted by HISP Ethiopia and the University of Oslo analyzed the design and deployment of a localized digital register to replace fragmented paper systems in public facilities [Research DAC2026].

The Iterative Design Cycles

The 2.5-year ADR process evolved through five distinct design cycles [Research DAC2026]:

  1. AI Modeling Proposal: The research team proposed automated predictive models to address data quality, which facility administrators rejected as mismatched to immediate operational requirements [Research DAC2026].
  2. Optical Character Recognition (OCR): The team piloted automated OCR scanning of paper registries. This cycle failed because the handwriting on the source paper forms was systematically incomplete or illegible [Research DAC2026].
  3. Tally Application: The team designed a standalone digital tally interface. This was rejected because it added an extra step without reducing the administrative time spent maintaining the physical paper registers [Research DAC2026].
  4. Digital Register & Raspberry Pi Hosting: The team co-designed a localized digital register mimicking the physical paper layout. The application ran on a low-cost, on-site Raspberry Pi micro-server, automatically aggregating entries to generate standard HMIS report payloads with a single operator-approval click [Research DAC2026].
  5. Paper Elimination Pilot: Within one month of deploying the local digital register, the facility officially terminated the use of physical paper registers [Research DAC2026].

Documented Outcomes

  • Cost Reductions: Seven months post-implementation, the local health department documented saving $20,000 USD by canceling a scheduled printing contract for paper registries. These funds were redirected to procure and install large-screen public dashboards showing facility metrics [Research DAC2026].
  • Design Framework: The study defines this approach as frugal digital innovation, prioritizing minimal hardware dependencies, high local configurability, and low maintenance overhead to achieve system sustainability [Research DAC2026].

3. Implementation Research: IMPULSE Phase 2 in Ethiopia

The IMPULSE study (Phase 2) evaluated the implementation of a targeted data auditing intervention across six hospitals in Addis Ababa to address gaps in newborn and stillbirth reporting accuracy [Research DAC2026].

The Technical Intervention

The intervention did not deploy new software, but instead configured the existing national ALQAS (monthly data accuracy check) application natively within the DHIS2 instance of the pilot sites [Research DAC2026].

  • System Configuration: The ALQAS metadata was restricted to compare reported DHIS2 newborn and stillbirth indicators directly against the physical paper register baselines monthly [Research DAC2026].
  • Supportive Components: The platform was linked to a visual data-accuracy dashboard and supported by structured clinical supervision rounds [Research DAC2026].

Documented Results

Initial results from the six pilot facilities showed:

  • A decrease in newborn indicators falling below the national accuracy threshold [Research DAC2026].
  • Acceptability: Rated at 94% to 100% by end-users across both evaluation phases [Research DAC2026].
  • Feasibility: Rated at 96% in both implementation rounds [Research DAC2026].

4. Retrospective Evaluation: Survive to Thrive (Zambia)

Researchers from the University of North Carolina Chapel Hill evaluated World Vision’s Survive to Thrive maternal and child health program in Luwingo District, Zambia, using the national DHIS2 HMIS database as the primary source of retrospective data [Research Lightning Talks].

Study Design and Modeling

The study mapped aggregate data across 18 intervention facilities and compared outcomes against non-intervention control sites [Research Lightning Talks].

  • Data Imputation: Missing DHIS2 data fields were modeled using a seasonally decomposed imputation method [Research Lightning Talks].
  • Model Controls: The regression models incorporated seasonal and autocorrelation terms to control for environmental and reporting trends over time [Research Lightning Talks].

Documented Findings and Null Results

  • Antenatal Care (ANC): Follow-up ANC visits showed a significant increase in the intervention group relative to comparator sites, showing an incidence rate ratio of approximately 3 [Research Lightning Talks].
  • Null Detections: There was no statistically significant effect detected on institutional deliveries, initial ANC visits, child diarrhea diagnoses, acute respiratory infections, or BCG immunization rates [Research Lightning Talks].
  • Methodological Limitations:
    1. Small Sample Size: Due to high rates of missing data, several facilities had to be excluded, reducing the final evaluation dataset to only 11 active intervention facilities, which limited the statistical power of the model [Research Lightning Talks].
    2. Paper-to-Digital Quality Gap: Because the researchers were external, they could not verify the accuracy of the DHIS2 records against the physical paper facility logs [Research Lightning Talks].
    3. Community Blind Spot: The national facility-level HMIS was unable to capture community-level preventative interventions delivered by CHWs outside of clinic environments [Research Lightning Talks].

5. Mobile Tracker Evaluation: Boma Health Initiative (South Sudan)

The Malaria Consortium conducted a hybrid evaluation of a mobile-based DHIS2 Tracker tool deployed on tablets for Community Health Workers (CHWs) under South Sudan’s Boma Health Initiative [Research Lightning Talks].

Study Methodology

The evaluation used a difference-in-differences design, comparing CHWs using tablets against a paper-based control group, and gathered survey feedback from 1,900 caregivers at baseline and endline [Research Lightning Talks].

Documented Findings

  • Clinical Knowledge: Both groups showed improvement in knowledge test scores from baseline to endline. However, difference-in-differences analysis showed no statistically significant difference in knowledge scores between the tablet-using and paper-using groups [Research Lightning Talks].
  • Quality of Care Confounders: The composite quality of care metric (correct diagnosis + correct treatment + correct referral) declined during the study period. Analysis showed that this decline was caused by recurrent drug stockouts. While CHWs routinely made correct diagnoses, the lack of available drugs on site categorized the treatment step as incorrect under the clinical protocol, showing how logistical failures can confound digital tool evaluations [Research Lightning Talks].
  • Caregiver Perception: Caregiver satisfaction and trust metrics remained high across both the paper-based and tablet-based groups [Research Lightning Talks].

6. Individual Patient Journey Reconstruction (MSF)

To track patient outcomes across specialized hospital departments (Emergency, ICU, Surgery, and Inpatient Wards), MSF developed a data cleaning and matching pipeline to reconstruct longitudinal patient journeys across discrete, unlinked DHIS2 Tracker databases [Research Lightning Talks].

Reconstructing the Patient Key

Because the databases lacked a unified patient identifier, researchers generated a composite patient key using three attributes: patient initials, sex, and approximate age [Research Lightning Talks].

  • Handling Entry Errors: To resolve common data entry errors where operators reversed the order of initials (e.g., entering “AB” in one program and “BA” in another), the matching script alphabetized the initials within the key (so both variations resolved to “AB”) [Research Lightning Talks].
  • Age Tolerances: To account for age calculation variations across multiple hospital admissions over the 2-year study window, the algorithm permitted a maximum deviation of two years [Research Lightning Talks].
  • Success Rate: This matching logic successfully matched and reconstructed patient journeys for 1,694 patients (representing 71% of the target emergency admission cohort) [Research Lightning Talks].

Cross-Program Length of Stay (LoS) Analysis

The linked multi-program datasets enabled hospital planners to run a Kaplan-Meier survival analysis, where “survival” represented a patient’s continuous stay within the facility:

  • Oncology vs. Trauma: Burn patients showed a median hospital length of stay of 16 days, compared to 7 days for trauma patients, a statistically significant difference providing planners with concrete metrics for bed capacity and resource forecasting [Research Lightning Talks].

7. What to Watch

The research presentations are available for review on the DHIS2 YouTube channel:

Community Call to Action

The research track at DAC2026 is a community-building effort as much as a knowledge-sharing one. If you are working on research that uses DHIS2 data, studies DHIS2 implementations, or evaluates programmes using HMIS data, this community wants to hear from you:

  • Share your own research — whether published, in progress, or at the “interesting field observation” stage. The thread below is a starting point.
  • Connect with the presenters if you are working on related topics — cross-programme patient journey reconstruction (MSF/Lilit), frugal digital innovation and ADR methods (Malaka/HISP Ethiopia), HMIS-based programme evaluation (Lucy Tantum/UNC), or CHW digital tool evaluation (Awen Norman/Malaria Consortium).
  • The SSM Health Systems special issue is active — contact Yan Sabu or Karen Grein if you have a paper in progress that fits the scope.
  • What research infrastructure do you need? Yan Sabu raised the question of what a fair and transparent DHIS2 research data infrastructure could look like — making aggregate HMIS data more accessible to researchers while protecting data governance. If you have views on this, share them below.
🗓️DAC2026 Thematic Summaries & Wikis/AI