Interrupted time series-counterfactual and lagged post intervention

[Radhika Singh]

Hi,

I’m currently analyzing an interrupted time series model to assess the impact of two key events on the frequency of cervical cancer screening tests (standardized to per 1000 women) among different age groups. The events in question are a guideline released in July 2018 (intervention date) and the COVID-19 lockdown in April 2020. My data covers quarterly intervals from January 2014 to December 2022.

I plan to use the ITSA package (Linden) for the analysis. The study includes all women as part of an observational study without separate control or intervention groups. However, I’m interested in comparing the effects on younger versus older women, which is similar to a multiple group, multiple intervention setup. Is this approach valid?

I have two main questions:

Question 1: I anticipate a one-year delay in the increase of diagnostic tests following the guideline release. How should I account for this lag in the syntax? Are there any resources or examples you can recommend?
Question 2: How can I generate counterfactual lines on the graph produced by ITSA?

Thank you for your time and assistance.

[Jared Greathouse]

Uhhhhhhhhhhh.... let's see. Do you think that the two groups are similar but for the intervention? Wouldn't there be baseline differences between the older group and younger cohort?

[Radhika Singh]

Thanks

[Radhika Singh]

Yes, there will be some inherent differences in baseline measurements. I've attached an image for reference. Two vertical lines indicate the two interventions. Older women tend to have higher baseline scores. Since the baseline needs to be consistent for ITSA to be valid, I can't directly compare the older and younger women. Will I need to analyze them separately? Do you have any other suggestions for making direct comparisons between the groups? My other two questions about delay post intervention and counterfactual remain.