Making DHIS2 data good enough to decide with: a data-quality workflow for municipal health teams

Nepal runs its national health management information system on DHIS2, managed through the Department of Health Services. Reporting completeness is generally high, because reporting is what the system rewards. Usability at the point where decisions are made — a municipal health department planning next quarter's outreach — is a different matter. The gap is rarely the software. It is the absence of a repeating, owned, boring routine for checking data before anyone relies on it.

1. The WHO metric set you should standardise on

WHO's Data Quality Assurance toolkit — published as 'Data quality review: a toolkit for facility data quality assessment, Module 1: framework and metrics', developed with the WHO Collaborating Centre at the University of Oslo — defines the dimensions that any routine review should cover. Adopting these definitions locally means your findings are comparable with provincial and national assessments instead of being an internal exercise nobody else can interpret.

• Completeness and timeliness of reporting: did every expected facility report, and did it report on time?
• Internal consistency: are values plausible over time, consistent between related indicators, and free of extreme outliers?
• External consistency: does routine data agree with an independent source such as a survey?
• Consistency of population data: are denominators plausible and consistently applied?
• Correctness of values against expected ranges for that facility type — the dimension DHIS2's own data-quality app operationalises directly.

2. The DHIS2 tooling that already exists

DHIS2's implementation documentation states that its data-quality approach follows the WHO DQA guidelines, and recommends reading both together. Three components matter in practice: the core Data Quality app (validation rule analysis, outlier and follow-up analysis, min-max ranges), the WHO Data Quality Tool app that implements the WHO methodology with configurable dashboards, and the newer DQ Workbench, which lets implementers define reusable data-quality stages — metadata integrity checks, outlier detection and validation-rule runs — that can be executed repeatedly rather than manually.

3. What the Nepal evidence says

A 2024 study in PLOS One reviewed DHIS2 data for maternal, newborn and child health indicators in Lumbini Province, assessing completeness and consistency — the kind of country-specific analysis that should be replicated province by province. A multi-country study in Population Health Metrics (2023) examined routine health data quality at the onset of the COVID-19 pandemic across five countries including Nepal, documenting how service disruption interacts with reporting behaviour. The Department of Health Services Annual Health Report series remains the official compiled record of HMIS-derived statistics by fiscal year, and is the correct place to source programme-level figures rather than quoting them second-hand.

4. The monthly routine that actually works

1. 01Day 1-5 of the month: reporting follow-up. Pull the reporting-rate summary, phone every non-reporting facility, and log the reason. Reasons repeat, and the log tells you which are structural.
2. 02Day 6-8: outlier and validation review for the priority indicator set only — not all indicators. Ten flagged values investigated beats four hundred flagged values ignored.
3. 03Day 9-10: register verification for a sample. Pick two facilities per month on rotation and compare three data elements against the source register. This is the only step that detects transcription and definition errors.
4. 04Day 11-12: internal consistency check across related indicators (for example ANC first visit versus ANC fourth visit, or screening versus treatment initiation).
5. 05Day 13-15: one-page data-quality note to the health chief: what was wrong, what was corrected, what remains and who owns it.
6. 06Quarterly: recalculate denominators, review min-max ranges by facility type, and retrain any facility that failed verification twice.

Data quality is not a report you produce. It is a habit with a calendar, an owner and a consequence.

5. The three failure modes to look for first

Definition drift

The same data element is counted differently in two facilities because staff turned over and nobody re-read the definition. It produces a stable, believable, wrong series — the most dangerous kind. Fix it with a one-page indicator definition sheet posted where the register is filled, not in a manual on a shelf.

Denominator improvisation

Coverage indicators quietly become meaningless when facilities use different population estimates. Standardise the denominator source across the municipality and record which source and year was used.

Reporting for compliance

If nobody local ever uses the data, staff optimise for submitting on time rather than submitting correctly. The cure is visible local use: put facility-level data on a wall chart and discuss it in the monthly meeting. Quality improves when the people producing data see it used.

6. From clean data to a decision dashboard

Once a priority indicator set is trustworthy, a small dashboard is worth building — five or six indicators, by facility, with trend and target. Whether it lives in DHIS2 dashboards or in a business-intelligence tool matters less than three design rules: every indicator must have an owner, every chart must answer a question someone actually asks, and any indicator whose data quality has not been verified should be visibly marked as provisional. A dashboard that mixes verified and unverified indicators teaches its users to distrust all of it.