Targeting anaemia without measuring it: surrogate prediction, district decision uncertainty and the value of repeat measurement in India

Background & objectives: India's fifth National Family Health Survey measured anaemia in all 707 districts, whereas the sixth survey did not. Anaemia is now assessed through a venous blood survey covering 183 districts and reported only at the national level. Using the most recent district-level measurements, we examined whether the remaining survey indicators could predict district anaemia, whether omitting district anaemia altered programme prioritisation, and the value of repeating district-level measurement. Methods: We estimated district anaemia prevalence and uncertainty for children aged 6-59 months and non-pregnant women aged 15-49 years using small-area estimation with design-based variances. We evaluated prediction from the retained survey indicators using both district-level and leave-one-State-out validation, compared district prioritisation under three information scenarios using matched preference draws, and estimated the value of repeating measurement of the same underlying prevalence. Results: Median standard errors of district estimates were 3.57 percentage points for children and 2.22 percentage points for women. The best predictive surrogate had a root mean squared error of 10.14 percentage points for children, of which 9.44 percentage points reflected structural error, representing approximately 2.5 times the root mean squared measurement error. In leave-one-State-out validation, predictions performed worse than the training-set mean. Among the 71 districts prioritised using current estimates, 19.1% were not among the latent top 71 districts. Measuring 183 districts recovered 46.3% of this prioritisation gap when districts were selected according to decision value, compared with 9.6% under equal allocation across States. Interpretation & conclusions: Available survey indicators did not adequately substitute for direct measurement of district anaemia. When measurement resources are limited, the choice of districts to be measured has a greater influence on programme prioritisation than the total number of districts measured, provided differences between measurement platforms are addressed before comparison.