# Can I use indirect standardization with the WHO Standard World Population?

I have to compare mortality rates among countries for a variable where the age-specific mortality rate is not available, but the overall rate is.

I want to use indirect standardization with the WHO World Standard Population and from my understanding, this means that I need to have the age-specific mortality rates from the reference population. I was planning to use country census data to get the age structure for each country, and compute a standardized mortality ratio.

However I looked everywhere I could think of and while age structure for the WHO World Standard Population is readily available, the age-specific mortality rates are not. How do people usually proceed in that case? Are there other standard populations providing standard mortality rates? Or am I supposed to choose for myself, for example taking the age-specific mortality rates from last year and apply those to the standard population?

• 1/2 A good motivation to do indirect age standardization is to make narratives of the form "How fast would people have died in population X if they had died at the same rates by age as people in the reference population, relative to how fast they actually did die" For example, “Steven Woolf and coauthors calculated that 25 percent of all deaths in Virginia between 1996 and 2002 would have been averted if the mortality rates of the five most affluent counties and cities had applied statewide.” Sep 26, 2020 at 1:31
• 2/2 In other words, the narrative power of indirect age adjustment results from projecting age-specific death rates from a special population (the reference population), onto a group of comparator populations. How fast would people in nations around the world die of SARS-cov-2, if they died at the same rates by age as people in the US died of SARS-cov-2 relative to how fast they actually did die? Sep 26, 2020 at 1:33
• @Alexis Yes, I am aware of the reference population idea. But I thought that the WHO standard population was designed to act as such. Of course, I could take some other reference, such as a specific country. Is that what I should do in your opinion, although I am not targeting the population of a country in particular in the report? Would it make more sense to use an extreme case then, such as the country with the highest/lowest crude rate? Sep 26, 2020 at 1:52
• Reference population ≠ standard population. The latter is for direct age standardization, and serves a very different narrative purpose. It is not clear in what way the WHO Standard World Population is special in a way that presents narrative insight. Sep 26, 2020 at 3:45

Here, we are apparently assuming a normalization exercise based on available and accurate (expressly for the reference population, which could be a pooled aggregate of the more 'advanced' countries in achieving longevity) age-specific mortality statistics. The latter are appropriately weighted by a country's percentage population by age-grouping.

To assist in interpretation, I postulate a perhaps implicit model, namely:

$${Country_c\text{ } Mortality\text{ } Rate = AdjFactor_c * (Weighted\text{ } Average\text{ }Age\text{-Specific}\text{ }Mortality\text{ }Rate)}$$

For the reference population, the cited adjustment factor is unity. The model inherently assumes a uniform factor severity adjustment across age groups.

For some countries, this could be appropriate in measuring the impact of say natural disasters (earthquakes, floods, fire, hurricanes, ..) or perhaps wars, which may decimate a population without respect for age differentials.

However, for heat waves or some pandemics or wide spread starvation, there could be selectivity for the old or very young. Noteworthy, for poor countries, there is a significantly higher infant/child mortality rate.

As such, care should be exercised in judging the meaning of adjustment factor by country. For countries with more elderly that are particularly subject to a pandemic, or pre-existing conditions, which may be more present in countries without socialized medicine, the implicit model's Adjustment Factor, per the equation above, would likely be harder to assess for parameter accuracy/bias.

On the question: "Or am I supposed to choose for myself, for example taking the age-specific mortality rates from last year and apply those to the standard population?", Yes. Using the last years mortality rates, with ample disclosure and perhaps part of my modeling assumptions, may assist in assessing results.