In Gordis Epidemiology, it is stated that the interview question Have you ever had asthma? can be used to measure cumulative incidence. Cumulative incidence was defined in the same book as:
$$\frac{\text{number of new cases in the population}}{\text{number of people at risk in the population}}\text{over a specified time period}$$
My question is, how can the answer to Have you ever had asthma? measure this? Would the proportion of the 'yes's to the total number of people somehow be the cumulative incidence proportion, and if so, what would be the "time period" implicit in the calculation? How exactly is the 'new cases' in the numerator explained since this, in my opinion, looks more like prevalence than incidence?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ "New cases over a time period" helps measure cumulative incidence over that time period. "Have you ever?" helps measure cumulative incidence since birth. $\endgroup$– HenryCommented May 11, 2021 at 14:47
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
3
Cumulative incidence is not the same thing as the incidence proportion (i.e. one kind of incidence rate), unless you understand "period of time" to be a variable length period which always begins at birth. This distinction is made very clear if you recognize that the "over a specific period of time" in the definition of incidence proportion is something like a calendar year, decade, or month, and if you also recognize that the word "ever" in cumulative incidence makes a very different kind of period measure (one of variable length). Cumulative incidence becomes more interesting when restricted to specific groups such as 'all people born in a given year,' or 'all people in this treatment group'. As Gordis' used the term, cumulative incidence is the complement of the survivor function: ${1 - S(t)}^{\dagger}$.
The reason why cumulative incidence is not prevalence, is because (for some conditions) having ever experienced the condition does not mean one currently experiences the condition. For example, have you ever had influenza? If so, you would appear in the denominator of cumulative incidence for influenza. However, if you do not currently have influenza, you would not appear in the numerator of prevalence of influenza. Cumulative incidence and prevalence may share a numerator if once one has the condition, one can never return to a state of not having the condition.
${}^{\dagger}$ That said, there are other uses of the term 'cumulative incidence' not appearing in Gordis which have different definitions.
-
1$\begingroup$ +1. To understand Gordis' motivation for the example: the question which most survey designers might have picked initially, "Do you have asthma?" would lead to an estimate of prevalence, but would not count people who previously had asthma. And so you start to understand the relationship between the survey question and the actual parameter here, specifically how subtle wording can lead to seemingly inconsistent findings. $\endgroup$– AdamOCommented May 11, 2021 at 15:15
-
$\begingroup$ @AdamO Thank you. Say: I am doing a new intro biostats course prep (the one that the Epi students take), and wonder if you have preferences for textbooks? I am evaluating Glantz's, Pagano & Gauvreau's, and Rosner's. I was thinking that a text like Rosner's might fit the bill... but then it does not have repeated measures in it. Do you have one you are partial to? $\endgroup$– AlexisCommented May 11, 2021 at 19:25
-
1$\begingroup$ Have you tried Vittinghoff et al, regression methods in biostatistics? $\endgroup$– AdamOCommented May 11, 2021 at 20:03