Characteristics of hemoglobin distributions in preschool children and non-pregnant women of reproductive age and their implications for establishing quality control criteria for hemoglobin data in field surveys: evidence from 483 surveys conducted in refugee settings worldwide.

Abstract

Background: Currently, there is a lack of clear guidance on hemoglobin (Hb) data quality parameters and plausible flagging ranges for population-representative surveys. There is a need to determine which properties of Hb data indicate lower data quality and increased measurement error and which represent intrinsic statistical properties of Hb distributions rather than quality problems.

Methods: We explored statistical characteristics of Hb distributions and plausible exclusion ranges in population-representative surveys of non-pregnant women of reproductive age (WRA) (15-49 years, n = 401 surveys) and children (6-59 months, n = 461 surveys) conducted in refugee settings by the United Nations High Commissioner for Refugees (UNHCR). Hb distribution characteristics [standard deviation (SD), skewness and kurtosis] were compared to those from Demographic and Health Surveys (DHS).

Results: Overall, 0.08% of child and 0.14% of WRA Hb values were outside of the previously proposed 4.0-18.0 g/dL plausible range. Surveys conducted in Uganda tended to have unusually high SD compared with surveys from other settings, possibly an indication of problematic measurement quality. We therefore used summary results on SD, skewness and kurtosis excluding surveys from Uganda when comparing with DHS results or proposing plausible ranges. Both WRA and child Hb distributions tended to be left-skewed and had excess positive kurtosis. Mean survey-level SD was greater, mean skewness more negative, and mean kurtosis more positive in WRA surveys compared to child surveys. All these findings were broadly similar to those from DHS surveys. Mean SD in DHS surveys was higher than that in our data for both children (1.48 vs. 1.34) and WRA (1.58 vs. 1.43).

Conclusions: We observed several statistical characteristics of Hb distributions that may not necessarily be indicative of data quality problems and bear strong similarities with the characteristics found in DHS surveys. Hb distributions tended to be negatively skewed and positively kurtotic, and SD in many surveys exceeded 1.5 (previously proposed upper plausible range). Based on our empirical evidence, surveys with skewness above + 0.2 and kurtosis below -0.5 or Hb SD outside the range of 1.1-1.55 g/dL for children (6-59 mo) or 1.1-1.65 g/dL for non-pregnant WRA (15-49 y) may require further quality investigation.