Segmented regression interrupted time series analysis - including lagged effect in the model

[SSK_011523]

Hello

I am using single ITS for one of my projects to assess the impact of a policy intervention on medication use. 

Outcome: Medication use (defined as aggregated means)

Exposure: Policy intervention

For the ITS model, I have defined the following parameters: 

Time (beta1): quarterly data (a total of 31 quarters). 

Intervention (beta2): The intervention was introduced in 12th quarter. I have defined intervention as three segments to take into account lagged effect. 0 indicates pre-intervention, 1 indicates transition/lagged period and 2 indicates post-intervention.

Time after intervention (beta 3): defined as 0 before the intervention 0 and intervention 1 and then as (1,2,3..19)

Since autocorrelation was detected in my data, I decided to use PROC AUTOREG. I used the following code:

********************************

PROC AUTOREG data = have;

model outcome = time intervention time_after_intervention/method = ml backstep nlag=5 dw=12 dwprob loglikl;

output out=want p =predicted r = residual;

run;

******************************

OUTPUT:

Variable	Estimate	Approx Pr > t
time	0.8	<0.0001
intervention	-2.9	<0.0002
time after intervention	-1.1	<0.0003

Interpretation:

I am not sure how to interpret 'intervention' as I have defined it as three segments (pre-intervention, lagged period, and post-intervention). Do I interpret it in the same way if two segments were used to define intervention? 

I tried using class statement thinking I would get separate estimates for each segment but it did nit work.

Any help on this would be appreciated. Thanks!

[sbxkoenk]

See here :

Interrupted time series with proc autoreg
https://communities.sas.com/t5/Statistical-Procedures/Interrupted-time-series-with-proc-autoreg/td-p...

(I provided the accepted solution there)

and here :

Did the Protocol Change Work?
Interrupted Time Series Evaluation for Health Care Organizations
Carol Conell. Division of Research. Kaiser Permanente Northern California, Oakland, CA
Alexander C. Flint. Department of Neuroscience. Kaiser Permanente, Redwood City, CA
https://www.lexjansen.com/wuss/2016/14_Final_Paper_PDF.pdf

BR,
Koen