Domains/Dimensions of data quality

One thing that should be noted regarding Domain 3 – External Consistency. Comparisons with population based surveys can be tricky – you have to ensure that the data are truly comparable. Often, surveys take a couple of years to compile and publish so the values you are using are already quite dated relative to the routine data. Also, population-based surveys cover the whole population, while routine values from HMIS only cover the people who avail themselves of facilities that report to the HMIS (which often exclude many private health facilities) – not the whole population. In addition, survey aggregation units may not be the same as administrative units used to aggregate the routine data. All these factors should be considered when comparing the routine values to population-based survey values.

Anyone have experience conducting such comparisons and best practices they can share to facilitate them?

Dave

These are important points to keep in mind when attempting to compare routine coverage estimates with survey coverage estimates.

WHO and UNICEF meet once each year to compare, for over 150 countries, the routine estimates of immunization coverage with the survey estimates and information from other sources. Their annual comparison graphs for each country can be downloaded from the following website: http://apps.who.int/immunization_monitoring/globalsummary/wucoveragecountrylist.html

Looking at the two first domains a) Completeness/timeliness, b) Internal consistency (plausibility) are actually normal and routine steps of any analysis of a data set. So if, within the district, monthly meeting are organized to analyse data and actually use them for concrete decisions, or if supervision are conducted regularly, outliers such those that you have shown, bob, should be corrected right away and not appear in the national data base. The point that I would like to make is that before data quality verification at te level of a national survey, the actual use of the information is the first condition for data quality.

Absolutely. Improved data quality will follow directly from analysis and use. However, we have a chicken and egg situation. Review of many HMIS datasets shows that there are significant issues with incompleteness of data and/or implausible data. This discourages users from analyzing and using the data.

I would suggest that data quality review be conducted simultaneously with analysis and use — including at sub-national levels.

Both processes (DQR and analysis) can be largely automated with a data management system like DHIS2. The impact of both processes depends upon active interpretation of the findings by multiple stakeholders — something that can never be automated.