02.Causation08.Interrupted time series - sporedata/researchdesigneR GitHub Wiki

1. Use cases: in which situations should I use this method?

Interrupted time series (ITS) analysis is used to evaluate the impact of an intervention (policy, event, program, etc.) that was implemented at one specific point in time for all the populationd, dividing the time into the "pre-intervention" and "post-intervention" periods. The main objective of an ITS is to determine whether there is a difference between the data observed pre-intervention and post-intervention.

Some research questions that can be modeled with ITS:

  • Does the rate of post-surgical complications reduce after a certain treatment policy?
  • Does a safety policy implemented within a hospital reduce the number of hospital-acquired infections?
  • Does the implementation of an awareness campaign reduce the number of traffic accidents?

2. Input: what kind of data does the method require?

  1. Pre-requisites for causal-analysis
  2. Data set with multiple measurements of the outcome of interest over time, before and after the intervention.

3. Algorithm: how does the method work?

Model mechanics

The issue with using interrupted time series analysis is that the basis for the counterfactual (creating a theoretical group that represents the lack of intervention) could be improved. We can use a method called causal impact that improves upon this method by using three simultaneous methods to create its counterfactual:

  1. One based on auto-regression, or the idea that past observations predict future ones,
  2. another is based on control, contemporary group. This is kind of the traditional way, and
  3. the third one is based on predicting what a counterfactual group should look like based on variables that could predict it. For example, you take race, gender, etc., to predict the pre-even progression of the outcomes and then use that to predict the counterfactual, i.e., the progression without the intervention.

Auto-regressive component is almost like a requirement. Check this for the lists of three methods is is also stated below:

"In contrast to classical difference-in-differences schemes, state-space models make it possible to (i) infer the temporal evolution of attributable impact, (ii) incorporate empirical priors on the parameters in a fully Bayesian treatment, and (iii) flexibly accommodate multiple sources of variation, including local trends, seasonality and the time-varying influence of contemporaneous covariates".

Also in the paper:

"Broadly speaking, there are three sources of information available for constructing an adequate synthetic control. The first is the time-series behaviour of the response itself, prior to the intervention".

Here are the equations: Equation

The t+1 portion is auto-regressive, since it's predicted by t or the time that just preceded it. The previous portion of the paper mentions all ARIMA (autoregressive integrated moving average) models being a type of state-space model, which is what these equations are accomplishing.

Reporting guidelines

Methodology and reporting characteristics of studies using interrupted time series design in healthcare

Data science packages

Suggested companion methods

Learning materials

  1. Books *
  2. Articles

4. Output: how do I interpret this method's results?

Mock conclusions or most frequent format for conclusions reached at the end of a typical analysis.

Tables, plots, and their interpretation

5. SporeData-specific

Templates

Data science functions

References