MEL Flashcards
1
Q
Ethical principles for Medical Research involving Human subjects
A
- Research Ethics Committee
- review protocol before study starts
- monitor study compliance
- authorise protocol amendments
- receive final report
2.
- Informed consent
- Privacy and confidentiality
- Vulnerable groups and individuals
- Risks, Burdens, Benefits
3.
- Scientific requirements and research protocols
- Use of placebo
- Unproven interventions in clinical practice
4.
- Post trial provisions
- Research registration, publication and dissemination of results
2
Q
***Protecting participants who cannot give consent
A
- Risk of negative discrimination
- Surrogate consent / patients’ best interest followed by assent / informed consent at a later date
- Minimise risk / burden, societal necessity if no expected health benefit
DOH position on vulnerable groups:
- Special protection
- Responsive to health needs of this groups and cannot be carried out in non-vulnerable group
- Net benefit
3
Q
Duty to ensure informed consent
A
Each individual who initiates, direct, engage in experiment
- cannot be delegated
4
Q
Risks vs Benefits
A
- Risks must be balanced by benefits to individuals / society
- Assess likelihood + magnitude of harm
- Identify ways to minimise risk
- Enhance potential benefits of intervention - not incentives
5
Q
***Unproven interventions in clinical practice
A
- When proven interventions not exists / other known intervention ineffective:
- Physician, after expert advice, with patient / legal representative IC may use an unproven intervention if:
- Physician’s judgement that it offers hope of saving life, re-establishing health / alleviating suffering
- There should be subsequent research to evaluate its safety and efficacy
- new information must be recorded and made publicly available where appropriate
6
Q
Problems in managing data
A
- Fabrication
- Falsification
- Plagiarism
- Self-plagiarism
- Conflicts of interest
Honest errors / differences of opinion are not misconduct
7
Q
Ethical duty to complete research cycle
A
Prevent publication bias even though a significant amount of research is never published / presented
8
Q
***Research registration and publication + Dissemination of results
A
- Study registered with a publicly accessible database
- Duty to make results available
- Adhere to guideline for report
- Declare affiliations and Conflict of interests
- Post-trial provisions: sponsor, researchers, host governments should make provision for post-trial access for all participants who still need an intervention identified as beneficial in the trial
9
Q
Peer Review
A
- Quality control measure for research
- prevent flawed medical research papers from being published
10
Q
Big data: Precision Medicine and Precision Prevention
A
- Beneficence:
Patients:
- Earlier diagnosis
- ↓ Morbidity / mortality via tailored treatments
Provider:
- Better targeting of resources
- cost savings (↓ ADR, ↓ in-patient stay etc.) - Autonomy
- is seeking informed consent practicable?
- control over use of personal data
- can opt out?
- obligation to share own data?
- “free-riding”
- ***Confidentiality
—> anonymisation achievable? —> genetic exceptionalism (probably not achievable)
- implications for family
- risk of downstream harms and need for re-identification - Non-maleficence
- Continuous monitoring and privacy
—> promote better self management
—> remote monitoring for chronic conditions (worried well?)
—> real time monitoring of older people (have choice to opt out? Privacy implications?)
5. Justice Global: - health inequalities in / between counties - representative participation Local: - ***best use of funding / resources?
11
Q
Summary of ethical challenges of Big Data
A
- Informed consent, Confidentiality
- Privacy, Data security, Accountability
- Accidental data breaches
- Data hacking, Cyber attacks
- Misuse / Sale of data
—> Who is accountable? (Software developer, Manufacturer, Clinician, Hospital?) - Big brother, Discrimination
- Constant monitoring
- Imbalance of power: Organisational to individual
- Exploitation of behaviours
—> incentivisation e.g. smart devices partnering with insurance companies to lower premiums
—> risk of paternalism
—> equal access - Commercial exploitation
- Limitations of variable data entry
- Potential harm if data re-used / recombined with additional data by other organisations
—> Beyond control of organisation originally collected / used the data
5. Justice Global: - health inequalities in / between counties - representative participation Local: - ***best use of funding / resources?
12
Q
Direct to Consumer testing
A
- Consent:
- Ultimate expression of autonomy?
—> convenient, easy, extended accountability
- Informed consent?
- Wider implications for family - Interpretation of data:
- Testing for bad genes with strong penetrance
- Difficulty interpreting results - Confidentiality:
- General data management concerns
—> leaks may affect insurability / employability
- Shared nature of genetic material
—> 3rd party’s right to know / NOT to know - Governance
- on health databases and biobanks
- e.. Personal Data (Privacy) Ordinance of Hong Kong
13
Q
AI in health care
A
- Beneficence
- societal / individual health benefits - Autonomy
- informed consent
- physician authority
- data privacy + surveillance - Non-maleficence
- safe guarding free-will
- data security and commercial exploitation - Justice
- discrimination
- accountability - Governance
- human agency and oversight
- technical robustness and safety
- privacy and data governance
- transparency
- diversity, non-discrimination, fairness
- environmental + societal well being
- NO existing legislation in HK
- accountability
14
Q
Artificial intelligence vs Machine learning vs Deep learning
A
DL (範圍最小):
- computers improving performance without being programmed by humans
ML:
- method of data analysis that automates analytical model building
- systems learning from data/pattern recognition with minimal human intervention
AI (範圍最大):
- computers carrying out human tasks
15
Q
Benefits of AI
A
- Remote health care delivery
- help doctors to evaluate patients more efficiently
- less expensively
- no fatigue / time limitations
- allow screening of more rural populations
- predictive analytics - identifying high risk patients - Enhancing diagnostics
- detect cancer / other diseases based on imaging
- facial profiling aid diagnosis of syndromes
- prediction and prevention of sepsis
- classifying skin lesions based on images - Enhance drug development
- Virtual human simulation in medical education
—> limitations: reduced human interaction, dependent on clinicians for software development and evaluation
Future:
- Healthcare virtual assistant
- Personal life coach
- Advanced analytics and research including precision medicine
16
Q
Concerns of AI
A
Human related:
- Informed consent to use data - option to opt out (Autonomy)
- physicians need to understand how AI works and limitations for informed consent
- patient need to be aware of extent of AI’s role in treatment
- would AI undermine clinician authority?
- treatment recommendation by an algorithm without rationale?
- Automation bias - Privacy / Confidentiality (Autonomy)
- Overt / Covert monitoring of activities?
—> right to be left alone
—> who could use the data?
—> commercial exploitation
- Behaviour predictions to urge individuals to make smart decisions
- Loss of free choice to make inadvisable decisions
- Penalty if fail to act on prediction?
- Held accountable for future predicted actions? - Discrimination, Data mining (Justice)
- assumption of objectivity —> inaccurate / misrepresentative data leads to replication of human bias
- applicability to data set —> difficult to assess reliability, bias, detect malicious attacks
- assumption of infallibility - Manipulation - reinforce existing biases
- Protection from cyber-attack
- Data security and Commercial exploitation (Non-maleficence)
- manipulations can fool AI systems / change behaviour of AI systems
- potential for hacker exploitation
- potential for fraudulent commercial exploitation
Computer / Data related:
1. Blackbox
- Limitations of data sets (Non-maleficence)
- financial motivations of AI product vendors
- applicability to different populations (failure outside range of data which AI trained)
- constant requirement for updating —> failure to recognise emerging infections
- need for comprehensive data including social + economic factors - Accountability for errors (Justice)
- black box —> difficult for human to understand how ML conclusions reached
- who’s accountable for errors that lead to mismanagement?
- would healthcare behaviour change?
—> no longer seeks physician’s opinion
—> self-diagnosis / self-medication
- who would regulate the industry?
17
Q
Mitigation for risks of AI
A
- Independent verification of AI predictions
- ***critically appraise black box output - Clinical trials to compare AI recommendations with standard care
- Transparency when AI related errors occur
- process for disclosure
- publication of errors
- consider how to correct errors
- clinician engagement with every stage of development and evaluation
18
Q
AI cannot
A
- Assess patient preferences
- Take into account access to medication
- Take into account cost implications
- AI need means of “cooling down” and energy supply
