Exam practice tips Flashcards
(9 cards)
Common confounders and effect modifiers to look out for?
Examples of Common Confounders and Effect Modifiers:
Confounders (common ones to watch out for):
Age: Often a confounder, because older people may be exposed to different risks than younger people, and age is related to both the exposure and the outcome.
Gender: Differences in gender can affect both exposure and outcome, making it a common confounder.
Socioeconomic status: People from different socioeconomic backgrounds might have different exposures and different health outcomes.
Smoking: Smoking could be a confounder in studies on lung cancer because it is linked to both the exposure (e.g., diet, occupation) and the outcome (lung cancer).
Effect Modifiers (common ones to watch out for):
Genetics: Genetic factors might modify how exposure affects the outcome. For example, genetics can affect how smoking influences lung cancer risk.
Age: Sometimes age can be an effect modifier, especially in studies where the relationship between exposure and outcome differs across age groups.
Sex/Gender: In some studies, gender can modify the effect of an exposure on an outcome, such as how certain medications work differently in men and women.
Health status: Someone who is already ill might experience a different effect from a treatment compared to someone who is healthy, making health status an effect modifier.
In studies to do with exercise, diet is a common confounder
What do you do when commenting on external validity?
always comment on the three general types of biases (internal validity):
Selection bias
Information bias
Confounding
Have to talk about this when talking about external validity, cannot comment on generalisability without commenting on these.
how to tell if something is not an intermediate in the pathway?
Is this variable being caused by the exposure? If yes, it might be an intermediate.
Does this variable affect the outcome independently of the exposure? If yes, it could be a confounder instead of an intermediate.
what to include when interpreting the conclusion criteria?
- if observational. what other studies?
- Bradford hill criteria
- comment on study design
- comment on causality
what to look for when analysing trends in tables?
General Tips for Analyzing Trends in Tables (Memorize These!)
Here’s a simple checklist you can repeat to yourself when looking at any table to spot trends and analyze them like a pro:
What’s Changing?
Look across columns: Does the number go up, down, or stay flat as the main variable (e.g., coffee intake) increases?
Tip: “Is there a clear direction—higher, lower, or mixed?”
How Big Is the Change?
Quantify it: Calculate differences (e.g., 142.7 - 135.4 = 7.3 mmHg) or percentages (e.g., 40.2% vs. 32.5% is a 23% relative increase).
Tip: “Is this a tiny shift or a big jump? Numbers tell the story.”
Does It Match the Outcome?
Check if the trend aligns with the study’s focus (e.g., hypertension rising with coffee). If not, why? (e.g., lower BMI but higher BP—odd!)
Tip: “Does this support or confuse the main finding?”
Could It Be a Confounder?
Ask: Does this trait (e.g., smoking, age) affect the outcome (e.g., cancer, BP) on its own? Could it mix up the main variable’s effect?
Tip: “What else might explain this—age, habits, health?”
Is There a Pattern or Surprise?
Look for linear trends (steady rise/fall) or quirks (e.g., middle group peaks, like moderate drinkers’ activity). Surprises need explaining.
Tip: “Is it a straight line, or does it zigzag? Why?”
Why Does It Matter?
Link it to the study: Does this strengthen/weaken the results? Suggest bias? Need adjustment?
Tip: “So what? How does this change the conclusion?”
Memorize this mantra: “Direction, size, outcome fit, confounding, pattern, impact—check them all!”
Repeat that to yourself, and you’ll cover the bases for a solid analysis.
scentences to analyse table 1?
stating the trend: across the groups, (variable) shows a clear increase, decrease, no consistent trend.
variable varies with (exposure)
as main factor increases, (variable) trends to rise, fall, shift unevenly
quantify it
link to outcome: this trend ties to (outcome) as (group with trend) has (higher/lower outcome measure) since (variable) influences (outcome) this could explain why (group) shows (result)
assessing confounding: (variable) may act as a confounder as it effects both ….
the difference could skew the exposure-outcome link, given (variable’s) known impact on (outcome)
if (variable) drives (outcome) independently, it might exaggerate/reduce the effect of (exposure)
explain impact; thus (variable) needs adjusment to clarify (exposure) true role in (outcome) adjusting for (variable) is key to isolating (exposure) impact on (outcome)
crtiique: Add a critical point (e.g., missing data, unmeasured factors)—this ups your grade.
spotting selection bias?
Other Ways to Spot Selection Bias (Beyond Exclusions)
Here’s how else selection bias sneaks in—use these to double-check any study:
Where the Study’s Done (Source Population):
Clue: Single center, specific region, or unique setting (e.g., hospital vs. community).
Qi et al. Example: One high-volume thyroid surgery center in China (>6000 cases/year). Might overrepresent severe PTC cases needing surgery, unlike milder cases managed elsewhere.
Impact: Results might not fit smaller clinics or diverse regions—recurrence could be lower elsewhere.
Who Volunteers or Enrolls:
Clue: Self-selection (e.g., only willing participants).
Qi et al. Example: Retrospective, so not voluntary, but all had surgery—selects for patients healthy enough for it (Methods).
Impact: Excludes frail patients who avoided surgery, possibly underestimating recurrence in that group.
Loss to Follow-Up:
Clue: High dropout rates or uneven follow-up.
Qi et al. Example: “Irregular follow-up” excluded—unknown how many dropped out post-inclusion (Table 1: 38.2 months recurrence vs. 66.3 non-recurrence).
Impact: If dropouts had less recurrence, study overestimates it; if more, underestimates.
Comparison Groups Uneven:
Clue: Cases/controls or exposed/unexposed differ in ways unrelated to the study question.
Qi et al. Example: Recurrent (n=121) vs. non-recurrent (n=17,874) differ in age, metastasis (Table 1)—some due to exclusions (e.g., persistent disease).
Impact: Hard to tell if location drives recurrence or if sample skew does.
Time Period or Context:
Clue: Long time span with changing practices.
Qi et al. Example: 2009–2022—surgery, ultrasound tech evolved (Discussion notes protocol shifts).
Impact: Early patients might differ (e.g., less precise imaging), selecting for later, better-diagnosed cases.
what does it mean when data comes from a biobank?
If a study’s results come from a biobank, it’s a clue that selection bias might be at play, but it also hints at other strengths and limitations tied to the study design and data quality. As a beginner, spotting this is a great step—it’s one of those sneaky details that can shape how you interpret findings! Let’s break down what a biobank suggests, how it could point to selection bias (and other biases), and apply it to a hypothetical tweak of the Qi et al. (2025) PTC study. I’ll keep it clear and tie it to what you’ve learned.
What’s a Biobank?
Definition: A biobank is a collection of biological samples (e.g., blood, tissue) and linked data (e.g., health records, demographics) from a specific group of people, often stored for research. Think of it like a library of human data—researchers borrow from it to study diseases like PTC recurrence.
Example: UK Biobank—500,000 people’s DNA, medical records, etc., used for tons of studies.
What Does It Suggest About the Study?
When results come from a biobank, it flags a few things:
Pre-Existing Data:
What: The study uses samples/data collected before the research question was even asked (retrospective-like).
Implication: Researchers can’t control who’s in the biobank or what data’s there—it’s a grab bag.
Defined Population:
What: Biobank participants are often from a specific region, time, or group (e.g., volunteers, hospital patients).
Implication: Might not match the broader population you care about (e.g., all PTC patients).
High-Quality Data (Sometimes):
What: Biobanks often have standardized, detailed records (e.g., tumor samples, ultrasound reports).
Implication: Can reduce measurement error, but only if the data fits the study’s needs.
Does It Point to Selection Bias?
Yes, Often: Selection bias happens when the study sample doesn’t represent the target population, skewing results. Biobanks can trigger this because:
Who’s In: Participants might be healthier (volunteers), sicker (hospital-based), or from one area— not a random slice of everyone with the condition.
Who’s Out: People who didn’t join (e.g., too sick, no consent) or weren’t sampled (e.g., no tissue stored) get excluded, systematically altering the group.
Everyday Example: A biobank of marathon runners’ blood tests studies heart disease. Runners are fitter than average, so heart disease looks rarer—bias from selecting a non-typical group.
Other Biases It Might Suggest
Differential Misclassification: If biobank data quality varies (e.g., better records for sick participants), errors might differ by group (cases vs. controls).
Nondifferential Misclassification: If data’s old or incomplete (e.g., no location for some tumors), errors might be random across all participants, diluting effects.
Confounding: Biobanks might lack key variables (e.g., lifestyle data), leaving unadjusted confounders.
Types of Biobanks and Participant Health
Population-Based Biobanks (e.g., UK Biobank):
Who: General public, often volunteers via self-referral (e.g., respond to invites).
Health: Tend to be healthier than average—why?
Self-Selection: Sick, frail, or busy people are less likely to join.
Example: A 50-year-old runner signs up; a bedridden diabetic skips it.
Bias: “Healthy volunteer effect”—underrepresents sick people, skewing results (e.g., less disease than expected).
Disease-Specific Biobanks (e.g., Cancer Tissue Bank):
Who: Patients with a condition (e.g., PTC), often recruited via clinics, not self-referral.
Health: Sicker than the general population—why?
Clinical Source: Enrolled during treatment (e.g., surgery), so they’re already diagnosed.
Example: PTC patients donate tumor tissue post-thyroidectomy—already have cancer.
Bias: Overrepresents severe cases needing care, missing mild ones (e.g., no surgery).
Hospital/Clinic-Based Biobanks:
Who: Patients at a specific facility (e.g., Qi et al.’s center if it had a biobank).
Health: Mix—depends on the hospital:
High-Volume Specialty (like Qi’s): More severe cases (e.g., surgical PTC).
General Clinic: Wider range (e.g., mild to severe).
Bias: Reflects that hospital’s patients, not all PTC cases.
Self-Referral Specifics
When It Happens: Mostly in population biobanks (e.g., UK Biobank mails invites, people opt in).
Healthier?: Yes, often:
Why: Healthier people have time/energy to join; sicker ones don’t.
Evidence: Studies show biobank volunteers have lower mortality, less smoking than non-joiners.
But: If it’s a disease biobank (e.g., cancer patients self-referring to a registry), they’re not healthier—they’re sick but motivated (e.g., want research to help).
