I am aiming to perform a direct standardisation to calculate the prevalence rate of a certain condition in an area. What I did was to merge different routinely collected health data to be able to detect all individuals affected by the condition in the area of interest. After doing so I came up with a final sample of ~ 6,000 cases (in a population of 4 million approx). I have then calculated the prevalence rate (x 100,000) and relative CI using the Byar's method.
As this data refer to a single region I would like to standardise the data using the direct standardisation method and the European population as reference.
However, I realised that for 92 individuals I do not information on the year of birth. Therefore, the calculation of the standardised rate would be biased. Although this is a very small proportion I still think that this matters, especially when calculating the standardised prevalence rates. Multiple imputation here is not an option as I have very few variables I could use as predictors. Also, all missing values come from the same data source (I used four different data sources and only one presented this issue), so missing would not be at randome. What options do I have? I was thinking to replace the missing values with the mean (or median) age value of the corresponding sex/district category. Practically speaking I would calculate the mean (or median) age for each group, defined by age and district of residence. For those having missing I would replace the missing value with the corresponding mean value. However, I am not sure whether this is technically sound in this case? What do you think? A reference could be a great help!