Background It is generally presumed that a child who has lost weight or grown slower than their peers, is more likely to be malnourished than a child who was born small and has tracked the same low centile. A measure that identifies children with low weight velocity could be expected to be more discriminating than attained weight alone for predicting adverse outcomes of malnutrition such as stunting or mortality.

In low/middle-income countries nutritional assessment most commonly uses the latest weight, without reference to previous measures or the overall growth trajectory. This study explores whether using weight velocity in addition to the latest weight improves the prediction of future stunting or mortality in the first two years of life.

Methods Analysis of a combined data set of 3447 children from Pakistan, Malawi, South Africa with prospectively measurement data collected monthly in the first year and height and survival recorded till 24 months. The main exposures were weight-for-age z-score (WAZ) at the end of each 2-month period and weight velocity-for-age z-score (WVZ₂) across that period. The outcomes were wasting, stunting or all-cause mortality in the next 1-2 months. As a sensitivity analysis, we also used velocity over 6 months. Cox proportional hazard models with repeated growth measures were used to study the association between exposures and mortality. Mixed Poisson models were used for stunting and wasting.

Results Children who were already stunted or wasted were most likely to remain so, but when WAZ, WVZ₂ and HAZ were added to the model, WAZ and HAZ were both associated with increased risk of stunting, but WVZ₂ added minimal prediction performance (difference in AUC = 0.005) compared to a model including only WAZ. For wasting models, including WVZ₂ had a marginally stronger prediction performance than that of WAZ only (difference in AUC = 0.015). Low velocity over 6 (but not 2) months was associated with increased mortality, but only added marginal prediction performance to a model including only WAZ (difference in AUC = 0.016).

Conclusion Surprisingly, measures of weight change over time added minimal value to predictions of incident wasting, stunting, and mortality, once the latest weight was known. The key determinant appears to be how far below the normal range the child’s weight is, rather than how they reached that position. Whether the marginal value added by WVZ would be useful clinically requires further economic analysis.