Week 1 HW: Principles and Practices

Question 1

I would like to develop a biological engineering tool that helps slow the onset or progression of Alzheimer’s disease, ideally by detecting risk early and enabling preventive intervention before major neurodegeneration occurs. This matters to me personally because my grandpa has Alzheimer’s despite maintaining a very healthy lifestyle (daily exercise, a diet focused on vegetables and whole grains, consistent sleep). His situation makes me think that for many people, Alzheimer’s risk is not simply the result of lifestyle choices, but also involves ageing-related biology and other non-behavioral factors. If we had better tools for early detection and prevention, we might reduce suffering and preserve independence for patients and families even when lifestyle isn’t the primary driver. Alzheimer’s is usually caused by build of protein plaques and neurofibrillary tangles (https://www.mayoclinic.org/diseases-conditions/alzheimers-disease/symptoms-causes/syc-20350447), therefore I was thinking of developing a tool that could combine using biomarkers detection, such as certain types and levels of proteins with an intervention

Question 2

Goal 1: Prevent harm (non-malfeasance) a) Sub-goal 1.1: Make sure the tool is accurate before it’s used i)Require clinical testing to prove it works. ii)Set minimum accuracy standards so it doesn’t wrongly label people “high risk.” b)Sub-goal 1.2: Prevent harm from unnecessary treatment i)Only allow interventions when benefits clearly outweigh risks. ii)Require monitoring for side effects and long-term impacts. Goal 2: Protect Autonomy and Consent a)Sub-goal 2.1: People choose whether they want to know their risk b)Sub-goal 2.2: Results must be communicated responsibly i)Require counselling/support when giving high-risk results. ii)Force clear language about uncertainty (“higher risk” ≠ “you will get Alzheimer’s”).

Question 3

Government: Expand anti-discrimination protections to cover Alzheimer’s risk data (not just genetic tests) Purpose Now: Canada’s Genetic Non-Discrimination Act (GNDA) protects people from being forced to take or disclose genetic test results for services/contracts/employment. But Alzheimer’s tools may also generate non-genetic risk scores Propose: A policy that prevents insurers/employers from using any Alzheimer’s predictive risk score (genetic or biomarker/AI-based) to deny coverage/jobs or raise premiums. Design Actors: Federal lawmakers + provincial regulators (insurance is often provincially regulated), privacy commissioners, employers/insurers. What’s needed: Define “predictive neurodegenerative risk data” (includes biomarker panels, algorithmic risk scores, genetic results). Enforcement + penalties (complaint process, audits). Clear allowed exceptions (e.g., medical use with consent). Assumptions People will avoid testing if they fear discrimination (likely true, but varies by population). Extending protections won’t destabilize insurance markets (insurers may argue it affects risk modeling). Risks of failure & “success” Failure risk: Loopholes—companies might still discriminate using proxies like family history or “medical record patterns.” Success risk: If protection is strong, uptake increases—good, but could strain healthcare capacity (more follow-ups, clinics, demand for counseling).

Companies + academic labs: Technical “privacy-by-design” + limited data sharing (federated analysis, minimal retention) Purpose Now: Many health tools centralize sensitive data (biomarkers, cognitive scores, even voice samples), increasing breach/misuse risk. Propose: Make privacy a technical requirement: collect the minimum data, process it as locally as possible, and restrict sharing. Design Actors: Companies building the tool, university research teams, hospitals/clinics, ethics boards (REB/IRB), privacy commissioners. What’s needed: Data minimization (only store what’s necessary). De-identification + strong access controls. Prefer federated learning/analysis (models learn across sites without pooling raw patient data in one place) where feasible. Clear deletion timelines and audit logs. Assumptions Federated or privacy-preserving approaches will still produce accurate models. Clinics and hospitals have the infrastructure to support this. Risks of failure & “success” Failure risk: Too little data sharing reduces model performance, especially for underrepresented groups → biased predictions. Success risk: If it works really well, people may over-trust it (“the algorithm says I’m fine”) and delay real medical evaluation.