Reasoning and Design

Summary: Outlines the steps of the scientific method and the FINER criteria for generating strong research questions.

• Generate testable question
• Gather data and resources
• Form a hypothesis
• Collect new data
• Analyze the data
• Interpret data and compare with hypothesis
• Publish
• Verify results

FINER method

• Feasible
• Interesting
• Novel
• Ethical
• Relevant beyond scientific community

Controls

• Verify results
• Can separate experimental conditions
• Positive controls: ensure change in dependent variable is expected
  • Example: HIV tests using known HIV⁺ samples
• Negative controls: ensure no change in dependent variable when none expected
  • Samples known HIV⁻
  • In E. coli antibiotic test, use saline instead of treatment
• Often used to assess placebo effect

Causality

• If–then relationship
• Manipulate independent variable, measure dependent

Error

• Bias usually minimal but exists if opinions incorporated
• Bias = systematic error → caused by inaccurate tool → lack of validity
• Random error always present → overcome with large sample size

Accuracy vs Precision

• Accuracy (validity): ability to measure true value
• Precision (reliability): ability to get consistent readings

Experimental approach

• Randomization: algorithm determines subject placement
• Blinding: subjects/researchers unaware of group
  • Single-blind / double-blind
• Data analysis
  • Variables: binary, continuous, categorical
  • Regression: linear, parabolic, exponential, logarithmic patterns

Observational approaches

• Cohort studies
  • Sorted by exposure, assessed for outcome
  • Longitudinal, prospective
• Cross-sectional
  • Different groups at one point in time
  • Example: prevalence of cancer in smokers vs nonsmokers
• Case–control
  • Identify subjects with/without outcome
  • Look backward for exposure history

Hill’s Criteria (more = stronger causal argument)

• Temporality
• Strength
• Dose–response
• Consistency
• Plausibility
• Consideration of alternatives
• Experiment
• Specificity
• Coherence

Error sources

• Bias: flaw in data collection
  • Selection bias: nonrepresentative subjects
  • Detection bias: differences in how professionals screen
    • Example: physicians screening obese patients more → inflated stats
  • Observational bias (Hawthorne effect): behavior changes because being studied
• Confounding: incorrect relationship characterized
  • Example: red hair, pain tolerance, opiate tolerance
  • Strong statistical link but not causal

Four core ethical tenets

• Beneficence: act in patient’s best interest
• Nonmaleficence: avoid treatments causing harm
• Respect for autonomy: honor patient’s decisions
• Justice: treat similar patients similarly

Respect for persons

• Honesty between subject and researcher
• Generally prohibits deception
• Informed consent required
• No coercion
• Vulnerable persons (children, pregnant individuals, prisoners) get special protection
• Confidentiality important

Justice in research

• Applies to study topic and execution
• Morally relevant differences exist (e.g., age)
  • Example: transplant priority for child
• Population size matters—large population impact increases benefit
• Race, ethnicity, sexual orientation, SES are not morally relevant differences
• Religion may or may not be relevant, depending on autonomy
• Risks must be equally distributed

Beneficence in research

• Net positive change for study + general population
• Minimize harm
• Use least invasive, least painful, least traumatic methods
• In two-treatment studies:
  • Neither treatment should be known to be superior (equipoise)
  • If one becomes clearly superior, trial must stop

Populations vs samples

• Population: complete group meeting criteria
  • Using entire population gives a parameter
• Sample: subset meant to represent population
  • Data from sample is a statistic

Generalizability

• Internal + external validity
• Low generalizability: narrow selection criteria, not reflective of population
• High generalizability: broadly representative of target population

Supporting interventions

• Evaluate whether data is strong enough to support/exclude a therapy or treatment plan

Statistical vs clinical significance

• Statistical significance: result unlikely due to chance
  • Magnitude of change doesn’t matter
• Clinical significance: effect meaningfully impacts health
• Lack of clinical significance is valid criticism of a study