'Consent-GPT S1 UK Sample'
(AsPredicted #187,429)
Download PDF

---

---

1) Have any data been collected for this study already?
No, no data have been collected for this study yet.

2) What's the main question being asked or hypothesis being tested in this study?
The main prediction is whether, or to what extent, people perceive consent delegation in medicine (either to a human junior doctor or a large language model (LLM)) to be morally valid*.

This study is also interested in evaluating whether, or to what extent, people feel it is justifiable to sue the hospital in the event of a complication occurring that the patient was either informed about or not, depending on the type of consent delegate.

3) Describe the key dependent variable(s) specifying how they will be measured.
The consent validity DV will measure agreement on a sliding scale (labelled: 0 (strongly disagree), 25 (somewhat disagree), 50 (neutral), 75 (somewhat agree), 100 (strongly agree)) with the following three statements:
Robin has given meaningful consent for the procedure to go ahead.
The consent process was sufficient to allow the treating surgeon to go ahead with the surgery.
If the treating surgeon proceeds with the surgery now, they'll be acting without Robin's proper permission. [reversed scored]

Assuming an acceptable Cronbach's alpha, the final consent validity DV will be the average measure of responses to these three statements, with responses to statement (3) being reversed scored.

The justified to sue DVs will measure agreement on a sliding scale (labelled: 0 (strongly disagree), 25 (somewhat disagree), 50 (neutral), 75 (somewhat agree), 100 (strongly agree)) with the following two statements:

Suppose a complication accidentally occurs during the procedure.
This complication is a known risk of the procedure and occurred even though the procedure was carried out carefully.
Before the procedure took place, ${e://Field/condition} did mention this possible complication.
Please indicate how much you agree or disagree with the following statement:
Under these circumstances, Robin is justified in suing the hospital.

Suppose a complication accidentally occurs during the procedure.
This complication is a known risk of the procedure and occurred even though the procedure was carried out carefully.
Before the procedure took place, ${e://Field/condition} did NOT mention this possible complication.
Please indicate how much you agree or disagree with the following statement:
Under these circumstances, Robin is justified in suing the hospital.

4) How many and which conditions will participants be assigned to?
Participants will be randomly assigned to one of 3 clinical vignettes in which Robin, the patient, provides consent to one of the following consent-seeker conditions: (1) the treating surgeon, (2) the junior doctor OR (3) Consent-GPT (an AI-generated app).

5) Specify exactly which analyses you will conduct to examine the main question/hypothesis.
We predict that H1) consent validity in the Junior Doctor condition will be higher than in the Consent-GPT condition, and H2) participants will consider it more justifiable to sue the hospital when the task was delegated to Consent-GPT.

To test consent validity, we will conduct a one-way ANOVA across the three conditions.

To test the justified to sue, we will conduct a 3x2 mixed ANOVA with between and within subject analyse.


To test these hypothesis, we will conduct two regressions models, one for consent validity and one for suing probability, with fixed effects for the experimental condition and socio-demographics:
lmer(consent ~ condition + sociodemographics)
lmer(sue ~ condition + sociodemographics)

Additionally, we predict that H3) negative attitudes towards AI moderate the effect of diminished consent validity in the Consent-GPT condition. To test this, we will repeat previous models adding interactions between conditions and AI attitudes:
lmer(consent ~ condition*AIattitudes + sociodemographics)
lmer(sue ~ condition*AIattitudes + sociodemographics)

6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations.
Participants will be excluded from the final sample for analysis if they (i) fail to answer the two main DVs, OR (ii) fail the comprehension check (a multiple choice question asking participants which of the following is NOT true about the clinical vignette, correct response = B) The consent conversation between [the treating surgeon / junior doctor / Consent-GPT] and Robin does NOT adhere to typical medical privacy and confidentiality standards.), OR (iii) fail the attention check (correct answer = 10), OR (iv) complete the survey in less than 3 minutes.

7) How many observations will be collected or what will determine sample size?
No need to justify decision, but be precise about exactly how the number will be determined.
We want 90% power to detect an effect size of Cohen's f= .20 with alpha .05. This requires 125 participants per condition (375 participants total). Anticipating exclusions (e.g. due to failing attention or comprehension checks, or incomplete surveys), we intend to recruit an additional 10%. Therefore, we aim to recruit a total sample size of 413 participants.

8) Anything else you would like to pre-register?
(e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?)
Additional exploratory DV's will measure agreement on a sliding scale (labelled: 0 (strongly disagree), 25 (somewhat disagree), 50 (neutral), 75 (somewhat agree), 100 (strongly agree)) with the following three statements:

If I were Robin, I would feel satisfied with having gone through the consent process with ${e://Field/condition}, based on the conversation described.
If I were Robin, I would feel satisfied with having gone through the consent process with ${e://Field/condition}, based on the conversation described.
If I were Robin, I would trust ${e://Field/condition} to give me all the relevant information I need to consent to the procedure.
If I were Robin, the fact that consent was delegated to ${e://Field/condition} makes me doubt whether the treating surgeon is acting in my best interest.
If I were Robin, I would feel comfortable sharing potentially embarrassing personal health information with ${e://Field/condition}.

For each of the above measures, we will run exploratory regression models analogous to the ones described previously:
lmer(DV ~ condition + sociodemographics)
lmer(DV ~ condition*AIattitudes + sociodemographics)

We will also be measuring participants' attitudes towards AI & the consent process.

We will also perform a Cronbach's alpha to assess the internal consistency of the perceived moral validity measures before combining them into an overall consent score.