Understanding Medical Research: course
Yale Courses Youtube-channel course as presented by F. Perry Wilson.
01 - Process of medical research.
The best salesman in the World can only show you the box. What is in the box?
The US spends more than $100 billion on medical research annually. Human genome project $3.8 billion in government funding, 1 trillion in economic growth.
How research is born? Question, hypothesis, testing hypothesis, funding, writing a grant / revise a grant. Grant funded? Conduct a study, collect data, analyse, write a manuscript. Submit a manuscript, Pierre Review, Revise manuscript, manuscript accepted.
Research in industry? Promising agent, conduct study, collect data, analyze data, write manuscript, submit manuscript. Pierre Review, revise manuscript, manuscript accepted.
Publications do not pay to researchers. Publications often implement paywalls. Outside paywalls, you maybe can access publication by asking a librarian, through Nih.gov Pubmed, Google Scholar - or by asking the authors directly.
Open Access. Good: authors can offer their work free for everyone, broader dissemination. Bad: the researchers pay for open access, some journals are predatory/scams, hard to see which are which.
Predatory journals. Get me off your F* Mailing List accepted after "peer review" in IJoACT.
Journal Metrics. Heard of the journal? Impact Factor: the number of citations.
Perverse Incentive. Running a lab is a small business, people depend on success. Negative studies considered a failure. Medical science enterprise is not pure science, encourages some bad behavior.
02 - Medical Manuscript
Structured documents with rules for reading and writing them.
Parts of a randomized water drinking study.
Title: exposure (to X), outcome (as Y). Exposure: drinking water. Outcome: less thirst.
Abstract: short description of the study.
Background/introduction: motivation, context, hypothesis, objective.
Methods: study setting in/exclusion criteria, intervention, PRIMARY outcome, and/or statistical analysis description.
Results: tables, description of the data outcomes. No editorialism. Primary outcome.
Discussion: speculation, conclusions, spin, conjecture. Readers responsibility.
Proposed objectivity in Methods and Results, subjectivity in background and discussion.
03 - Replication Crisis
Pick a number 1 to 20, psychic powers. Excited to get a positive result? (5% probability)
Four types of studies: positive study of a true hypothesis, negative study of a true hypothesis, positive study of a false hypothesis, negative study of a false hypothesis.
A positive study: data supports the hypothesis. Quantify how strongly data supports the hypothesis, set a treshold for success. Got lucky?
A negative study: data doesn't support the hypothesis. Unlucky?
Why replicate: false negative studies are bad, missing out on important outcomes, but false positive studies are really bad, lead to useless or harmful drugs, or dead-ends. Of 1000 example studies, 2,7% fully/partially replicated. Time, money and low career impact as constraints to replication. NIH Grant Review Criteria: 1 Significance, 2 Innovation, 3 Approach. FDA requires replication: 2 large controlled clinical trials.
Medical Reversal: new trial of better design contradicts current medical practice. Gerstein 2008: effects of intensive glucose lowering in T2 diabetes, Woodcock 2003: Control of exposure to mite allergen.
One study is never enough. A positive study is never enough: positive study is not necessarily true, a negative study is not necessarily false. Replication is not incentivized, innovation is. Skepticism @ "medical breakthrough".
04 - What is the Question?
Which question do you want to ask in a study? The exposure: outcome paradigm. Study on association between gun violence and Gun Shows. "Does gun ownership lead to death?" vs "Does the time-period surrounding a gun show in a state with blah blah lead to gun-retelated violence blah?"
What is an exposure? Something we do to you, you are, some measurement like cholesterol level. Varying exposures: medication, cholesterol. Static exposures: your DNA, your ethnic background.
What is an outcome? Some event or measurable about you: death, weight. Two flavors of outcomes; hard outcomes: birth, Quality of Life, Death, and everything else: money, fame, musclemass.
Conceptual models: Exposure > Outcome. Smokeing-Cancer. Coffee-Success. Illustrate the Causal Pathway. Smoking-Lung Damage-Lung Cancer. Confounder. Foie Gras associated with longevity - is the outcome result of eating foie gras, or other factors like economic status leading to better health care.
Exposure/Outcome game: in study-context, phenomena can be either-or. Some factors can be both exposure and outcome.
Formulate correct question. Hard outcomes are the best outcomes.
05 - Types of Data
Categorical data: model, sex, marital status. Here's the "race" concept again.
Continuous data: weight, price, cholesterol level.
Ordinal data: how much you like the car?
Normal distribution, presented with a poor example where the outcome /middleboard is pre-selected with the input /drop in the middle. Population age example of non-normal distribution.
Qualitative data: non-quantified, focus groups, interviews.