2 of 18

Big Picture

There was an idea for a tool to help people decide on behalf of others.

People worried that this tool disrespected autonomy.

A new tool was proposed in response.

But: it still disrespects autonomy, and for very similar reasons.

3 of 18

Plan

Surrogate Decision-Making and the P3

Statistical Disrespect and the P4 Response

Propensity Disrespect and the Evidentiary Dilemma

Objections, Replies, and Discussion

4 of 18

Surrogate Decision-Making and the P3

Incapacity affects ~34% of adult inpatients, and up to ~83% in some ICUs.^1, 2

~2/3 of U.S. adults have no advance directive.³

Others must decide on their behalf, following the substituted judgment standard: decide as the patient would have.⁴

___________________

[1] Lepping et al. 2015 [2] Bertrand et al. 2019 [3] Yadav et al. 2017 [4] Buchanan and Brock 1989

5 of 18

Surrogate Decision-Making and the P3

But substituted judgment is often inaccurate: surrogates predict patient preferences only ~59–68% of the time.^1, 2

And it is often burdensome: about a third of surrogates show lasting symptoms consistent with PTSD.³

___________________

[1] Starke et al. 2025 [2] Shalowitz et al. 2006 [3] Wendler and Rid 2011

6 of 18

Surrogate Decision-Making and the P3

This has motivated an algorithmic solution: the patient preference predictor (PPP).^1, 2

The original PPP (a “P3”) infers a patient’s preferences from population-level data: e.g., age, diagnosis, religiosity.

P3s promise to improve accuracy and decrease burden.

(Bracketing two complications: consent and adjunctive use.)

___________________

[1] Shalowitz et al. 2007 [2] Rid and Wendler 2014a, 2014b

7 of 18

Statistical Disrespect and the P4 Response

(Naked) statistical evidence is population-level data used in support of an individual-level verdict.

Example: driver for the Blue Bus company.

A prominent objection¹ to the P3 concerns such evidence:

(1) Legal conviction by statistical evidence shows disrespect for autonomy.

(2) If so, then surrogate decision-making by statistical evidence does too.

(3) Use of the P3 is surrogate decision-making by statistical evidence.

(4) So, use of the P3 shows disrespect for autonomy.^{2, 3}

___________________

[1] Sharadin 2018 [2] John 2014 [3] Mainz 2023

8 of 18

Statistical Disrespect and the P4 Response

Some responses in defense of the P3:¹

Premise 1 is false: autonomy concerns outcomes.

Premise 2 is false: statistics are not disrespectful in medicine.

A popular response says that premise 3 is only true of the P3, but not the personalized P3 (the “P4”).

The P4 is an LLM fine-tuned on patient-level data; e.g., health records, social media, etc.²

___________________

[1] Jardas et al. 2022 [2] Earp et al. 2024

9 of 18

Propensity Disrespect and the Evidentiary Dilemma

Propensity evidence is individual-level data used in support of a character-based verdict.¹

Example: character witnesses in the Seinfeld finale.

An overlooked objection to the P4 concerns such evidence:

(1) Legal conviction by propensity evidence shows disrespect for autonomy.

(2) If so, then surrogate decision-making by propensity evidence does too.

(3) Use of the P4 is surrogate decision-making by propensity evidence.

(4) So, use of the P4 shows disrespect for autonomy.

___________________

[1] FRE 404(a)

10 of 18

Propensity Disrespect and the Evidentiary Dilemma

For 1: conviction by propensity evidence treats a person’s past behavior as determinative—“once a thief, always a thief.”^1-3

For 2: surrogate decision-making by propensity evidence treats a patient’s past behavior in the same way.

For 3: the P4 uses individual-level data to infer character patterns (preferences, etc.) in support of treatment decisions.

___________________

[1] Tillers 1998 [2] Matsumi and Solove 2025 [3] cf. Redmayne 2002

11 of 18

Propensity Disrespect and the Evidentiary Dilemma

There is parity between the objections from statistical and propensity disrespect.

Those who rely solely on the P4 in response to the former objection therefore face the latter, irrespective of its defense.

They also cannot re-appeal to the P3, on pain of statistical disrespect—the two arguments generate an evidentiary dilemma.

12 of 18

Objections, Replies, and Discussion

Objection 1: premise 1 is false or the conclusion is irrelevant, because our past behavior really is determinative.

Reply: distinguish between being determinative and treating as determinative.

Objection 2: premise 2 is false, because patients—unlike defendants—are incapacitated.

Reply: this is only pressing if there are no alternatives to propensity evidence.

13 of 18

Objections, Replies, and Discussion

Objection 3: there is a further parity argument:

(1) Legal conviction by propensity evidence shows disrespect for autonomy.

(2) If so, then surrogate decision-making by propensity evidence does too.

(3) The use of a human proxy is surrogate decision-making by propensity evidence.

(4) So, the use of a human proxy shows disrespect for autonomy.

Worry: if the propensity disrespect objection holds, then so does the parity argument—but that’s absurd.

14 of 18

Objections, Replies, and Discussion

Reply: human proxies don’t “really” use propensity evidence.

First, it’s unclear whether they use it at all (e.g., they might perceive character dynamically).

Second, it’s unclear how they use it (e.g., as determinative of character or of future decisions).¹

Third, it’s unclear whether they only use it (e.g., they also use their authority to decide).²

___________________

[1] Fetherstonhaugh et al. 2017 [2] Um 2022

15 of 18

Objections, Replies, and Discussion

Some implications:

First, a PPP could still be used consensually or adjunctively.

Second, a P4 could still be used in non-propensity fashion (e.g., to identify values and decisions consistent with them).^{1, 2}

Third, this entire discussion raises broader concerns about the autonomy of incapacitated patients (e.g., sovereignty).³

___________________

[1] Brudney 2019 [2] Schwan 2024 [3] Enoch 2022

16 of 18

Citations

Bertrand, P. M., et al. (2019). Disagreement between clinicians and score in decision-making capacity of critically ill patients. Critical Care Medicine, 47(3), 337–344.

Brudney, D. (2019). Changing the question. Hastings Center Report, 49(2), 9–16.

Buchanan, A. E., & Brock, D. W. (1989). Deciding for others: The ethics of surrogate decision making. Cambridge University Press.

Earp, B. D., et al. (2024). A personalized patient preference predictor for substituted judgments in healthcare: Technically feasible and ethically desirable. The American Journal of Bioethics, 24(7), 13–26.

Enoch, D. (2022). Autonomy as non-alienation, autonomy as sovereignty, and politics. Journal of Political Philosophy, 30, 143–165.

Fed. R. Evid. 404(a).

Fetherstonhaugh, D., McAuliffe, L., Bauer, M., & Shanley, C. (2017). Decision-making on behalf of people living with dementia: How do surrogate decision-makers decide? Journal of Medical Ethics, 43(1), 35–40.

Jardas, E. J., Wasserman, D., & Wendler, D. (2022). Autonomy-based criticisms of the patient preference predictor. Journal of Medical Ethics, 48(5), 304–310.

17 of 18

Citations

John, S. (2014). Patient preference predictors, apt categorization, and respect for autonomy. The Journal of Medicine and Philosophy, 39(2), 169–177.

Kapustin, D. A., et al. (2026). Beyond advance directives: A machine learning tool to aid end-of-life decisions in dementia care. Journal of Pain and Symptom Management, 71(6), e962–e963.

Lepping, P., Stanly, T., & Turner, J. (2015). Systematic review on the prevalence of lack of capacity in medical and psychiatric settings. Clinical Medicine, 15(4), 337–343.

Mainz, J. T. (2023). The patient preference predictor and the objection from higher-order preferences. Journal of Medical Ethics, 49(3), 221–222.

Matsumi, H., & Solove, D. J. (2025). The prediction society: AI and the problems of forecasting the future. University of Illinois Law Review, 2025(1), 1.

Redmayne, M. (2002). The relevance of bad character. The Cambridge Law Journal, 61(3), 684–714.

Rid, A., & Wendler, D. (2014a). Treatment decision making for incapacitated patients: Is development and use of a patient preference predictor feasible? The Journal of Medicine and Philosophy, 39(2), 130–152.

Rid, A., & Wendler, D. (2014b). Use of a patient preference predictor to help make medical decisions for incapacitated patients. The Journal of Medicine and Philosophy, 39(2), 104–129.

18 of 18

Citations

Schwan, B. (2024). Weighing patient preferences: Lessons for a patient preferences predictor. The American Journal of Bioethics, 24, 60–62.

Shalowitz, D. I., Garrett-Mayer, E., & Wendler, D. (2006). The accuracy of surrogate decision makers: A systematic review. Archives of Internal Medicine, 166(5), 493–497.

Shalowitz, D. I., Garrett-Mayer, E., & Wendler, D. (2007). How should treatment decisions be made for incapacitated patients, and why? PLoS Medicine, 4(3), e35.

Sharadin, N. P. (2018). Patient preference predictors and the problem of naked statistical evidence. Journal of Medical Ethics, 44(12), 857–862.

Starke, G., et al. (2025). Machine learning-based patient preference prediction: A proof of concept. NEJM AI, 2, AIoa2500265.

Tillers, P. (1998). What is wrong with character evidence? Hastings Law Journal, 49(3), 781–834.

Um, S. (2022). Autonomy, shared agency and prediction. Journal of Medical Ethics, 48(5), 313–314.

Wendler, D., & Rid, A. (2011). Systematic review: The effect on surrogates of making treatment decisions for others. Annals of Internal Medicine, 154(5), 336–346.

Yadav, K. N., et al. (2017). Approximately one in three US adults completes any type of advance directive for end-of-life care. Health Affairs, 36(7), 1244–1251.