I would just like to verify a statement made in the following paper: Peripheral blood HIV-1 DNA dynamics in antiretroviral-treated HIV/HCV co-infected patients receiving directly-acting antivirals

The authors state: "Factors associated with the increase of total HIV-1 DNA were analysed using univariable and multivariable logistic regression". They then go on to show a table, table 2, indicating the significant factors to this increase.

Now my question is, is this as simple as generating a logistic regression model with two dummy variables (i.e. whether or not there was an increase in DNA), and then identifying the significant factors ? Or is there more to it given that they are only interested in the factors affecting the increase of HIV DNA ?



1 Answer 1


The amount of HIV DNA is a continuous value (or at least approximately). And if one is interested in the effect of a treatment on the increase of HIV DNA, one should rather use a model with continuous outcome rather than logistic regression. E.g. you might deem a linear model to be appropriate. Then you identify the treatment, devise a model of the amount of HIV DNA that is linear in the treatment, make sure that the experiment is properly randomized or that you have controlled for all relevant confounders, and then the effect can be obtained from the fitted coefficient of the treatment.

  • $\begingroup$ Thanks for your response. In this instance though, the authors are ONLY interested in factors associated with an increase in HIV DNA above a particular threshold - 0.5log10. They claim to use logistic regression on samples falling within this range. In this case, how would one formulate the problem ? $\endgroup$
    – h3ab74
    Sep 22, 2022 at 20:03
  • $\begingroup$ My point is that this change of a continuous situation into a binary one is questionable. Consider two changes from below the threshold to above the threshold, the first with just a tiny step from shortly below the threshold to shortly above the threshold, the second being a giant leap. In the binary model, this would be considered the same, even though we have a much larger effect in the second case, which is unfortunate. But if that is not what is bugging you, could you clarify what is? $\endgroup$
    – frank
    Sep 23, 2022 at 2:32
  • 1
    $\begingroup$ Nothing is particularly bugging me, I just don't seem to understand how this problem is formulated (is it just a regular logistic model with 1 and 0 , 1 being an increase above the threshold?) .. I do see your point though tbh, and one cannot deny that. It does seem that for measuring factors associated w/ change in viral loads in HIV studies (I am new to this btw), it is the standard method. See the following paper: academic.oup.com/jac/article/69/3/753/786442. Thank you again btw for this discussion ! $\endgroup$
    – h3ab74
    Sep 23, 2022 at 14:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.