GLM for HP (fear conditioning)

[camillerib33]

Hello,
I am running a GLM analysis for HP (fear conditioning) and have included a condition file. In this file, I specified that each condition, CS+ and CS-, has an 8-second duration, as this corresponds to the design of our study. However, when I run the GLM, I receive the following warnings:
Warning: Non-zero durations in condition CSplus detected. This is discouraged for modality hp.
Warning: Non-zero durations in condition CSminus detected. This is discouraged for modality hp.
Is this something I should be concerned about? Could it affect the validity of the analysis?
Thank you for your guidance.

[dominikbach]

The "duration" in the multiple condition file has a specific meaning, and non-zero durations will essentially "smooth" the canonical response.

In fear-conditioned bradycardia, however, the response is simply shifted according to the CS-US interval.

As such, you should set the duration to 0 in the multiple condition file, and instead encode the CS-US interval in the field "SOA" under "Basis functions".

[camillerib33]

Thank you for the clarification. I have a question regarding the extinction/retrieval phase: since the CS+ is not followed by a US in this phase, should I set the duration to 0? Also, during acquisition, since the reinforcement schedule is not 100%, what would be the best way to handle the duration and SOA in that context?

[dominikbach]

This is a good point. So, the important thing in all these models is is to model all trials (all CS+ and also all CS- and all extinction trials) in the same way (a) to avoid bias, (b) because the agent does not know for sure if a US will follow or not, and so if a CR is elicited on these trials, it will have the same shape as in reinforced trials (assuming that SOA is constant in reinforced trials).

So, SOA refers to the interval between CS onset, and the US on those trials where a US occurs.

PsPM takes care of this, you just need to specify CS onset, and the SOA.

[camillerib33]

Thank you, I would like to ask for your expertise regarding my settings and output.

I imported an Acknowledge file, selected my ECG channel, marker 1 (CS+) and marker 2 (CS-). I then trimmed the file because it combines the retrieval and reinstatement phases, and I wanted to focus only on the retrieval phase.

Next, I performed the data conversion from ECG to heartbeats (Pan & Tompkins), followed by conversion from heartbeats to heart period. When prompted for the sample rate, I wasn’t sure whether to enter the original sample rate (2000 Hz in my case) or to downsample it, so I entered 100 Hz.

I subsequently checked the QRS complexes with the ECG Editor. Finally, I created a condition file by copying the data for marker 1 and marker 2, specified the time units in seconds, and set the SOA to 8 seconds.

However, I am confused about the resulting graphs and the statistics for the reconstructed amplitude estimate. To illustrate, I ran the analysis on two different participants:

For the first participant, CS+ reconstructed amplitude = -41 and CS- reconstructed amplitude = -5

For the second participant, CS+ reconstructed amplitude = 76 and CS- reconstructed amplitude = -51

Additionally, why does the ‘plot data per condition’ graph show 10 seconds on the x-axis instead of 8, and why is the reconstructed response 18 seconds long? Is this normal?

Furthermore, is there a minimum number of trials required to obtain a reliable heart period response (HPR) per CS type? Conversely, is there a maximum number beyond which the data become less reliable? In my study, we used 4 trials per CS type during the retrieval phase, 16 trials per CS type in the extinction phase, and 8 trials per CS type in the acquisition phase.

Thank you very much for your time and expertise.

[ribbensca-ship-it]

Hi, I was wondering if it's still possible to help me with the above question?

[dominikbach]

Apologies for the late reply. Your work flow makes sense. To answer your questions:

1. The x-limits for the two visualisations are defaults used across different data modalities - they are not necessarily optimal for the HPR model. Unfortunately, PsPM does not offer better plots, but you could easily create nicer data plots by using the "extract segments" functionality.

2. In our studies, we have typically used many more trials, and the HPR is also somewhat persistent during extinction. As a rule-of-thumb I'd say more data does not harm, but I cannot give you a lower boundary. (All our data is publicly available in PsPM format though, so you could potentially find out).

3. Perhaps my previous posts were not fully clear: although the CS response should be modelled for all conditions (CS-, CS+ followed by US, CS+ not followed by US), these should be different conditions, because the US response strongly affects the estimated CS response. This is why we always report the CS+US- vs. CS- contrast.

Hope this helps!
Dominik