Automated opioid risk scores: a case for machine learning-induced epistemic injustice in healthcare

In the paper

“That is to say, standard, non-ML-mediated epistemic practices between patients and physicians are already prone to put the patient in a position of epistemic weakness (Carel & Kidd, 2014; Carel et al., 2017).”

What the sources say

• Supported

  Carel & Kidd (2014) argue throughout that ordinary, non-exceptional healthcare epistemic practices structurally disadvantage patients: the Introduction states that health professionals are considered epistemically privileged and 'decide which patient testimonies and interpretations to act upon', and that what patients say 'will therefore be assigned a deflated epistemic status'; Section 3 ('Epistemic privilege') describes 'an asymmetry in the relationship owing to an implicit hierarchy assigning the health professional (normally a physician) a high epistemic status' and claims the structures and discourses of contemporary healthcare practice 'privilege certain styles of articulating testimonies... in a way that structurally disables certain testimonial and hermeneutical activities'. The 'non-ML-mediated' qualifier is the citing paper's own contrast frame, not something attributed to this source; the substance of the claim (standard patient-physician epistemic practices already tend to place the patient in an epistemically weak position) is what our copy states, including the vicious-circle passage in Section 2. (LLM, conf=0.88)

  cites Carel & Kidd, 2014 source
• Supported

  The paper states as its general thesis that ordinary clinical epistemic practices place patients at a disadvantage: 'A growing body of work has suggested that individuals suffering from ill health are more vulnerable to testimonial injustice, and this vulnerability exists across the different stages and epistemic practices of medical work' (Testimonial injustice, p. 3), and 'it is patients who have most to lose from the effects of such epistemic skewing' (Testimonial injustice, p. 4). It also notes that deferring to standard consultation roles — 'authoritative doctor' and 'submissive patient' — 'can on its own be epistemically unjust' (Evidence from patients, p. 7), and the Abstract (p. 1) holds patients to be 'particularly vulnerable to epistemic injustice'. The claim's hedged wording ('prone to', 'position of epistemic weakness') matches the source's level of confidence; the 'non-ML-mediated' contrast is the citing paper's own framing, not an attribution. (LLM, conf=0.85)

  cites Carel et al., 2017 source

In the paper

“As already mentioned, being NarxCare a proprietary algorithm, the possibilities to get meaningful insight into how it works and which criteria were decisive in estimating a patient’s high-risk score are very much constrained.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Oliva, 2022

In the paper

“However, patients are confronted with a risk score that is treated as grounded knowledge that reliably mirrors their drug consumption levels.”

In the paper

“Through careful consideration of the current flaws identified in this system's functioning, I show that it fulfills three fundamental conditions to identify forms of hermeneutical injustice suffered by patients affected by PDMPs' decision-making (“Defining epistemic injustice in ML” section).”

In the paper

“As a matter of fact, a study conducted by Picco et al. (2021, p. 8) reports that the number of patients being refused medical assistance and medication supply and of patients being discharged from practice has considerably increased due to the role taken up by PDMP platforms.”

What the sources say

• Not checked yetcites Picco et al., 2021 source