Mortality

Morbidity is having the disease (prevalence of the disease). Mortality has to do with death. There are several measures. Note that high morbidity doesn't mean you have high mortality!

"Death is not subject to diagnostic variability."

You have 10,000 per 100,000. You have 15,000 per 100,000. Can you conclude that the latter has a higher mortality? Nope. The latter can have a larger group of older people! You have to age-adjust. It's not just a stupid number without context: it must be anchored to a population and in a time frame. Denominators shift over time! This is why you have many measures and cannot rely on a single summary statistic.

Crude Mortality Rate

This is what you'd typically think "mortality" means.

$$\frac{Deaths}{Population} \times 1000$$

That deaths part must be within a time interval. That population part must be mid-interval. You need to understand population structure: the composition of the population across key demographic factors (age, sex).

When you want isolate differences in risk versus differences in structure you will need a different measure! If you have populations P1 and P2 that have the same underlying risk, Crude Mortality Rate can emerge differences in structure.

Cause-Specific Mortality Rate

$$\frac{Deaths from Cause}{Population} \times 1000$$

Helpful to think of this as a component of the Crude Mortality Rate (assuming all causes are independent). Sum everything up and you get crude! This depends on the accuracy of the "causeo of death" label.

Age-Specific Mortality Rate

$$\frac{Deaths in Age group}{Population in Age Group} \times 1000$$

What it says on the tin. Stratified according to age. This is another decomposition of Crude Mortality (think about it). Problem here is that not everybody dies at the hospital and their death is not recorded.

Age Standardization Methods

You try to remove the confounding effects of age. Age is a big factor and a confounders. It distorts the causal link between exposure and outcome.

Take the ASMR from our population and see how this fares with the standard population. One way is Direct Standardization. This technique works with other confounders too! You apply the ASMRates to the standard population across various strata.

What if we don't have a reliable ASMRate for the standard population? Like in a small population. You cannot do Direct Standardization! You have to use Indirect Standardization.

So we know the popoulation. We know the mortality rate but not the age-specific rates (there's just one doctor in town). So you apply the standard age-specific rates to this population. So,

• Study Age Specific Rates → Standard Population (Direct - Changing Weighting Schema)
• Standard Population Age Specific Rates → Study Population (Indirect - Changing Benchmark)

Goes without saying that you will need a well-characterized Standard Population.

SMR: Standardized Mortality Ratio: $\frac{Observed}{Expected} \times 100$. Enrichment if over 100. Depletion if under 100.

Infant Mortality Rate

$$\frac{Infant Deaths (Age under 1 year)}{Live Births} \times 100$$

during a fixed time interval. Here we are anchoring to the size of a "birth cohort".

You can get a lot of information from this. It focuses on a vulnerable population: how well are we doing with taking care of them? This is reported widely and can be used for international comparisons.

Life Expectancy at Birth

This is a summary measure. Average number of years a newborn would be expected to live if current age specific mortality rates were to remain constant over their lifetime. Pretty useful too.

Random

Takes months to get a death certificate. This is problematic.

If you do "verbal autopsies" you can have a lot of recall bias.

'Windowing' refers to establishing the time span. Multiple visits can be the same event.

Some encounters are better than the others. You want in-person because of continuous monitoring! I trust it more and require less. It's more rigorous. For this you use visit_occurrence!