Summary and impact
We have identified clinically significant discrepancies between the built-in DHIS2 calculation of weight-for-length/height z-scores (WLZ/WHZ) and calculations using the WHO Child Growth Standards methodology. We discussed this with University of Oslo/DHIS2 colleagues on 23 September 2026 and were encouraged to submit a formal ticket to document the issue and support further review.
This is more than a minor difference in numerical precision. In testing a large multi-country dataset, approximately 20% of children who would meet anthropometric criteria for treatment using WHO-consistent calculations did not meet those criteria using the current DHIS2 calculation.
Importantly, the discrepancies we observed were unidirectional rather than random: the current DHIS2 calculation consistently made children’s nutritional status appear better than it would under the WHO-consistent calculation. At the individual level, this creates a risk of systematically under-identifying children with acute malnutrition or classifying them into a less severe nutritional category, potentially affecting treatment eligibility and program assignment.
This direction of bias also has important population-level implications. WLZ/WHZ is used to estimate wasting and global acute malnutrition (GAM) prevalence. If children’s nutritional status is systematically classified as better than it is under WHO-consistent calculations, DHIS2-derived data could underestimate the prevalence and severity of acute malnutrition. This could affect monitoring of malnutrition trends and, in some settings, whether a population falls above or below thresholds used to characterize the severity of a nutrition emergency and inform humanitarian planning, resource allocation, and response.
Given the potential significance for both individual children and population-level nutrition monitoring, we hope to work collaboratively with UiO to identify an appropriate solution within DHIS2 core, so that WHO-consistent calculations are available by default to the broader DHIS2 and nutrition community, without requiring implementation-specific configuration or workarounds.
Issues identified and proposed approach
Our review identified several key differences between the current DHIS2 calculation and the WHO methodology replicated using WHO igrowup reference data. In collaboration with Simprints, we have explored potential approaches to address these differences. Based on our assessment and validation to date, the main issues and proposed solutions are:
1. WfL versus WfH reference selection
WHO growth standards use separate weight-for-length (WfL) and weight-for-height (WfH) reference distributions because recumbent length and standing height are different measurements and are not directly interchangeable. In standard anthropometric practice, children 6–23 months are generally measured recumbently and children 24–59 months standing. Because age may not always be reliably known, 87 cm provides a practical proxy for this transition, with measurements <87 cm evaluated using WfL and measurements ≥87 cm using WfH.
- Current DHIS2 approach: DHIS2 appears to use the WfL reference through 110 cm, switching to WfH only above that point. This means that children between 87 and 110cm who should be evaluated against the WfH reference are instead evaluated against the WfL reference.
- WHO-consistent approach: Use 87 cm as the operational threshold for selecting the appropriate reference: <87 cm → WfL and ≥87 cm → WfH. Where actual measurement position is captured, a more precise implementation allows this information to be used directly to select the appropriate reference (recumbent length → WfL; standing height → WfH), as is supported by established anthropometric packages.
- Why this matters: Standing height is systematically shorter (0.7cm on average) than recumbent length for the same child, so the WfL and WfH reference distributions are not interchangeable. Applying the WfL reference to children who should be evaluated using WfH systematically shifts the resulting z-scores toward better apparent nutritional status and can therefore affect classification of acute malnutrition. This was the most consequential difference identified in our testing: correcting the WfL/WfH reference selection accounted for approximately 85% of the clinical misclassifications observed in our comparison.
- Proposed approach: Update the DHIS2 core calculation to use 87 cm as the default threshold for selecting between WfL and WfH (<87 cm → WfL; ≥87 cm → WfH), while allowing implementations to configure the reference-selection logic where measurement protocols differ (for example, where children are measured recumbently across a broader age range, or where selection is based on age rather than height). Where measurement position is recorded directly, this could also be used to select the corresponding reference. Because other anthropometric indicators, such as height-for-age z-score, also depend on whether a measurement represents recumbent length or standing height, the same underlying logic should be applied consistently across relevant DHIS2 calculations and reference populations.
2. Reference data and granularity
- Current DHIS2 approach: DHIS2 appears to rely on reference tables at 0.5-cm intervals, containing weights corresponding to integer z-scores (−3, −2, −1, 0, +1, +2, +3 SD), with intermediate z-scores derived from these values.
- WHO-consistent approach: The WHO igrowup reference files provide the underlying L, M, and S parameters at 0.1-cm intervals. These parameters describe the distribution of weight at each length/height and allow the z-score to be calculated directly using the WHO LMS method rather than approximated from rounded integer-SD weights.
- Why this matters: Z-scores near clinical thresholds can determine whether a child is classified as severely malnourished (SAM), moderately malnourished (MAM), or neither. Using a coarser reference grid and approximating z-scores from rounded integer-SD weights can introduce differences that are small numerically but clinically meaningful for children close to these thresholds.
- Proposed approach: Base the DHIS2 calculation on the WHO LMS parameters and 0.1-cm reference grid, substantially improving agreement with WHO-compatible anthropometric calculations.
3. Flooring versus interpolation
- Current DHIS2 approach: Measurements appear to be effectively floored downward to the lower 0.5-cm reference point. For example, a recorded measurement of 76.8 cm may be evaluated as 76.5 cm. This systematically discards some of the precision of the observed measurement.
- WHO-consistent approach: The recorded measurement is retained. If it falls exactly on a WHO 0.1-cm reference point, the z-score is calculated at that point. If it falls between reference points, for example, 87.23 cm, the z-score can be calculated at the adjacent 0.1-cm values and interpolated according to where the observed measurement falls between them using the LMS approach.
- Why this matters: Flooring always moves a measurement in the same direction rather than introducing random rounding error. A child measured at 76.8 cm, for example, is evaluated as though they measured 76.5 cm. This systematic loss of measurement precision can shift the resulting z-score and, for children near clinical cutoffs, change nutritional classification. After correcting the WfL/WfH reference selection, this flooring behavior accounted for much of the remaining discrepancy in clinical classification. The effect of the flooring alone resulted in a systematic difference of 0.05 SD between the WHO and DHIS2 approach.
- Proposed approach: Retain the measured length/height rather than flooring downward and use interpolation between adjacent 0.1-cm reference points where needed.
4. Additional discrepancy at the upper end of the WfH reference
We also identified a substantial discrepancy for female children at approximately 112 cm, which falls above the range affected by the WfL/WfH reference-selection issue described in #1 above. For example, for a female child with a recorded height of 112.37 cm and weight of 15.3 kg, the WHO-consistent calculation produces a WHZ of −2.70, while the current DHIS2 calculation produces −1.64 – a difference of more than 1 SD and sufficient to change the child’s classification from malnourished to non-malnourished by WHZ.
Because this observation is above the approximately 110 cm point at which the current DHIS2 calculation already appears to use the WfH reference, the discrepancy cannot be explained by selection of WfL instead of WfH. This suggests an additional problem with the underlying reference values or their implementation at the upper end of the current WfH table. We observed the same pattern in multiple female observations with a height of 112cm.
· Proposed approach: Updating the calculation to use the WHO 0.1-cm LMS reference data described above should address this issue, but we highlight it separately because it demonstrates that clinically significant discrepancies are not confined to the 87–110 cm range affected by WfL/WfH reference selection.
5. Implementation and computational performance
A direct implementation of the full WHO reference data may result in very large DHIS2 program-rule expressions and create performance concerns, particularly for Tracker implementations operating offline or on lower-powered devices.
In collaboration with Simprints, we have developed both reference and optimized implementations. Initial testing suggests that fitted cubic splines can substantially reduce the size and computational burden of the calculation while maintaining z-score accuracy to approximately 0.01 SD.
· Proposed approach: Consider an optimized implementation that reproduces the WHO-consistent calculation to approximately 0.01 SD while remaining computationally practical within DHIS2 Tracker.
The WHO calculation also uses a modified linear-tail calculation outside ±3 SD, rather than indefinitely extrapolating the standard LMS equation. This differs from the current DHIS2 approach of constraining results to ±3.5 and should also be considered as part of aligning the core calculation with WHO methodology.
Validation
To investigate the discrepancies and test potential solutions, we developed independent Stata and Python implementations using the WHO reference files from igrowup and compared the resulting calculations with established anthropometric calculations in a dataset of 7,321 observations with complete height, weight, sex, and age data from a humanitarian context.
The WHO-consistent implementations showed extremely close agreement:
- Mean difference between zanthro and our combined WfL/WfH calculation: ~0.0015 SD
- Mean difference between zanthro and our WfL calculation: ~0.0014 SD
- Agreement in clinical classification as SAM, MAM, or neither was 100% in this validation.
Supporting materials and next steps
We hope this assessment and the work already undertaken by Action Against Hunger and Simprints can provide a useful starting point for UiO’s review. Our goal is to work collaboratively toward a solution within DHIS2 core, rather than only addressing the issue through a project-specific workaround, so that improvements can benefit the broader DHIS2 and nutrition community.
We would be happy to share:
-
WHO 0.1-cm WfL and WfH reference files derived from igrowup
-
Independent Python implementation of the WHO-consistent calculation
-
A selection of 200 de-identified example observations with substantial differences between the current DHIS2 calculation and the WHO-consistent calculation resulting in direct misclassification.
-
The optimized cubic-spline implementation developed to maintain approximately 0.01-SD precision while reducing computational burden, which can serve as a technical reference for DHIS2’s consideration.