wsd extra

What are the 3 core components of epidemiology?

Its distributions (patterns of health events), its determinants (causes and risk factors), and the application of this knowledge to control health problems and improve population health.

Distinguish between the Hawthorn effect and "Social Desirability" bias

• Hawthorn effect: Participants alter their actual behavior because they know they are being observed (falls under objective measures).

• Social Desirability bias: Participants give untruthful, favorable answers to look good (falls under self-reported/survey measures).

What are the primary limitations of the WHO and Huber definitions of health?

• WHO ("complete" well-being): Arguably unattainable and risks medicalizing normal human variations.

• Huber (ability to adapt/self-manage): Less precise and harder to operationalize for empirical measurement or policy action.

What occurs when you alter the diagnostic threshold for a positive screening test?

• Lowering the threshold: Increases sensitivity (catches more true cases) but decreases specificity (creates more false positives/unnecessary anxiety).

• Raising the threshold: Increases specificity but decreases sensitivity (creates more false negatives/missed cases).

How does changing the definition of "disability" impact population statistics?

It can cause reported disability rates to shift dramatically from roughly 10% to over 25% within the exact same population, demonstrating how sensitive metrics are to structural definitions.

Why do we use standardisation when comparing different population groups?

To remove the distorting effects of confounding variables (like age profiles). By forcing both samples into an identical, hypothetical reference population structure, we can safely compare disease frequencies across countries or over time.

What is the hidden danger of over-relying on a massive sample size to achieve statistical power?

While a large sample size provides the power to detect tiny, subtle effects, it also means almost everything becomes statistically significant, even if the real-world clinical or public health effect size is completely negligible.

Why can fundamental "biological causal effects" be validly studied in non-representative samples, while descriptive work cannot?

Basic biological mechanisms (e.g., tobacco smoke causing cellular mutations leading to lung cancer) are universally shared across the human species. In contrast, non-biological effects (e.g., low income causing ill health) are deeply modified by social, economic, and environmental contexts, requiring strictly representative data to describe accurately

Compare the outputs and limitations of Cohort and Case-Control studies.

• Cohort: Time-consuming and costly, but directly produces incidence rates, risk ratios, and rate ratios.

• Case-Control: Quick, inexpensive, and ideal for rare diseases, but highly vulnerable to recall bias and difficulties in selecting a proper control group.

How do Narrative Reviews differ from Systematic Reviews?

• Narrative: Addresses broad questions using non-specified, potentially biased search approaches and subjective selection criteria.

• Systematic: Addresses a specific question using an explicit, replicable search strategy and strict, criterion-biased selection rules to minimize bias.

Distinguish between Risk Ratios (Ratio measures) and Risk Differences (Difference measures).

• Ratio measures: Answer "how much more likely?"—vital for understanding etiology and underlying biology.

• Difference measures: Answer "how many extra cases?"—essential for resource allocation and calculating Population Attributable Risk Percent (PARP).

What do negative values signify in Risk Differences and Odds Ratios?

• Negative Risk Difference: Indicates a lower absolute risk among the exposed group compared to the unexposed group.

• Negative Odds Ratio (or $OR < 1$): Indicates that the exposure is associated with a lower likelihood of the outcome occurring (protective effect).

Why has the public health sector shifted toward identifying "shared component causes" over isolated individual causes?

Because intervening on a shared component cause (e.g., an unhealthy food environment or community violence) targets an entire mechanism that affects a large number of people simultaneously, maximizing public health impact.

When evaluating data, why can you not adjust for a mediator if you want to understand the total effect of an exposure?

Adjusting for a confounder is necessary to reduce bias. However, adjusting for a mediator (a variable on the causal pathway) results in over-adjustment, which actively blocks and hides the real, total causal effect of the exposure.

Distinguish between Non-Differential and Differential measurement errors.

• Non-Differential (Random): Error is unrelated to other variables; it dilutes results towards the null, causing groups to look blurry and similar to each other.

• Differential (Non-Random): Error differs systematically between groups (e.g., exposed vs. unexposed), leading to an outright overestimation or underestimation of the true effect.

Why do Large Language Models (LLMs) hallucinate, and how does "temperature" affect this?

LLMs are purely probabilistic text predictors trained to guess the next token; they do not query a factual database. Lowering the temperature makes outputs more deterministic (choosing only the highest-probability words), reducing novel variations and erratic errors.

What is the core structural advantage of Open-Source LLMs over Closed-Source LLMs for scientific replication?

Open-source models allow researchers to download and freeze the model version locally, ensuring a replicable study environment and keeping sensitive patient data secure. Closed-source models (like Gemini or ChatGPT) can change their weights or go offline without warning, threatening reproducibility.

What is the core premise—and primary critique—of Fundamental Cause Theory?

It argues that health inequalities persist because socioeconomic privilege allows the advantaged to adopt life-saving knowledge and treatments first. The critique is that it is mainly descriptive and somewhat unfalsifiable, meaning it explains why inequalities last rather than predicting precisely where they will show up next.

Why is targeting specific behavioral mediators (e.g., diet apps) insufficient if Socioeconomic Position (SEP) is a fundamental cause of disease?

Because if SEP is the fundamental cause, eliminating one mediator won't stop the disparity; the underlying systemic forces will simply give rise to new risk factors that distribute themselves unequally across the social strata.

Why is it methodologically difficult to definitively prove biological mediation (like cortisol pathways) in social epidemiology?

Biological consequences accumulate over an entire life course. Measuring these shifts accurately at multiple time points is technically demanding, and measurement errors in biological samples make it tough to isolate exactly how much inequality is truly driven by that specific pathway.

Explain how reactive gene-environment correlation (rGE) complicates simple parenting-to-child health models.

An apparent environmental effect (like harsh parenting causing poor child mental health) can be genetically confounded: a child's genetically driven behaviors may actively provoke or influence how the parents and surrounding environment react to them.

Compare the "Health Literacy" vs. "System Integrity" pathways linking childhood IQ to adult health.

• Health Literacy: Higher cognitive function improves executive control and long-term planning, making individuals better at navigating healthcare systems and modifying behavior.

• System Integrity: Childhood IQ is simply an early marker of healthy, robust neurodevelopment, meaning the IQ-health link is a reflection of overall bodily integrity rather than direct cognitive mediation.

Why is measuring cognitive ability during childhood highly prized in cognitive epidemiology?

Because it captures cognitive function well before major chronic diseases or age-related declines manifest, drastically reducing the risk of reverse causality confusing the results.

Why are skyrocketing obesity rates in deprived neighborhoods viewed as structural rather than purely behavioral?

Because deprived areas feature obesogenic environments—unequal structural conditions characterized by a lower variety, poorer quality, and restricted availability of fresh produce, alongside a high density of cheap, energy-dense foods.

How has the socioeconomic gap in BMI evolved across modern generations?

The BMI gap between the highest and lowest social classes has expanded dramatically over time, growing from 2.0 kg/m² in the 1946 birth cohort to 3.9 kg/m² in the 1970 birth cohort, with individuals in newer cohorts shifting to an overweight status before age 10.

Why must late-life declines in BMI be interpreted with extreme caution in epidemiological tracking?

A falling BMI in older age groups is frequently a clinical marker of poor health (such as muscle wasting, hidden chronic illness, or impending neurodegeneration) rather than a sign of improved body composition or successful lifestyle modification.

what is negative confounding?

When the confounder hides/reverses the true relationship between an exposure and outcome, making it look weaker than it actually is.

• Could make beneficial factor look harmful, or vice versa

How does comparing results from different study design help with understanding importance of confounding bias?

If a completely different study design—with an entirely different set of potential confounders—comes up with the same result, your confidence grows. It becomes highly unlikely that a single hidden confounder is distorting the data, because that confounder would have to exist in both completely different contexts.

How does comparing results from different contexts with different confounding structures help with understanding importance of confounding bias?

If Study A shows a massive effect, but Study B (which successfully controls for a specific confounder) shows a much smaller effect, the difference between the two results explicitly measures how much that confounder was warping the reality.

what is rate ratio?

A comparison of how fast a disease happens in an exposed group versus an unexposed group.

What actually is attributable proportion?

The percentage of cases among the exposed group only that was directly caused by the exposure.

What actually is population attributable risk proportion?

The percentage of cases in the entire population (both exposed and unexposed people) that is caused by the exposure. It depends heavily on how common the risk factor is in the community.

Why is the odds ratio used for case-control studies?

Because you choose the number of cases and controls yourself (e.g., recruiting 100 cancer patients and 100 healthy patients), you do not know the actual number of people in the real world who were exposed or unexposed.

How does a risk ratio of more than 1.0 impact risk?

The risk of getting the disease is greater in those who were exposed to the risk factor

Why do those who drop out of cohort studies differ systematically?

Systematic → people who leave are fundamentally different from the people (if people who drop out are completely random = no biased results)

• Often, the sickest, poorest, or most marginalized participants are the ones who drop out.

• People experiencing the worst side effects/symptoms more likely to quit study

How does dropping out of cohort studies cause bias?

Distorts true relationship between exposure and disease

• can hide real danger by underestimating risk

• can create a fake relationship by overestimating risk

can you explain the prediction that IQ should only predict health outcomes where knowledge matters?

IQ can only protect your health if there is actual information available to use

• If no one knows what causes a disease or how to treat it, a higher IQ gives a person 0 biological advantage

• IQ is a resource used to navigate health literacy