Definition of a HPV coinfection

[carajbro]

How to define an HPV coinfection to accurately match our model outputs to observed data

[carajbro]

Cara: Original approach (spoiler, still use this one): Group HPV-positive people in the model into stages based on their clinical diagnosis, or most progressed type. For example, if a person has 9v-CIN2 and non-9v-CIN3, they would be considered to have CIN3. Then, I defined persons as 9v if their 9v type was most progressed, non-9v if their non-9v type was most progressed, and 9v if they had a HPV coinfection, or their types were equally progressed (ie. HPV-HPV, CIN1-CIN1, etc.). The example person above was grouped as non-9v because their non-9v type was most progressed.

Alternate approach (decided against this): Still group people into stages based on their clinical diagnosis, or most progressed type. However, then re-define a HPV coinfection as the general presence of both HPV types. Therefore, I would define persons as 9v if they had 9v infection only, non-9v if they had non-9v infection only, and 9v if they had both types. Our example person would therefore be considered coinfected, and put in the 9v group for comparison to observed data.

I think the underlying question regards the way HPV genotyping was done in the observed data. My understanding is that first that the clinician makes their clinical diagnosis as CIN1, CIN2, etc. Then they test for all types present in the cervix. The clinician then assumes that the lesion is due to the more high risk type (9v in our case), regardless of whether that type actually caused the lesion?

I think this is relevant because we are calibrating to type distribution in the 2000s when there is a substantial amount of HIV which increases HPV coinfections (using both definitions).

[carajbro]

Darcy: It’s tricky because even the empirical data are uncertain. As you may recall, when we were reviewing the literature to define the target type distribution for each stage, it was quite challenging because studies typically reported the prevalence of each HPV type, such that that sum of the prevalences would sum to more than 100% due to multiple infection. I didn’t see any articles that classified women as being prevalent only for the higher risk type if they had multiple infections. My approach to derive estimates from these data weighted the 9v types higher for some articles, but others seemed to do some adjustment of their own which was unreported or we used only data from women with single-type infections. Note this latter approach would probably weight HIV-negative women higher, but it was the best option I could think of.

So it is not clear to me that your alternate approach is necessarily more aligned with the data. I actually think your old approach might be more accurate - I imagine that most often if someone has coinfection, the 9v type is the causal type. But there will be some cases when that is not the case, and I think the way we defined the estimates reflects that to some degree.

[carajbro]

Cara: I agree that neither of these approaches will be a perfect match to the calibration data. With your context above, I worry that the second/new approach will overweight 9v-type infections so I think I'll stick with the original approach and we can just be aware of the limitations.