Study Design & Sampling

Choosing the Right Research Approach

King Abdulaziz University · Faculty of Medicine

MSTA112 · Week 2

Connecting: Stats (Week 5) & Viz (Week 3) → Design (Today)

1 / 28

Why PURE-Saudi Results Are Trusted

Good Design = Trusted Results

You already know how to describe data (Mean/SD). Today we learn how to collect it so your mean and SD actually mean something.

What's NEW Today:

Study Designs: When to use RCT vs Cohort?
Advanced Sampling: Avoiding selection bias.
PICO: How to write a research question.

2 / 28

Quick Review: Variable Types

You mastered this in Week 1. We review it because Type determines Design.

Type	Example	Statistical Test (Week 5)
Nominal	Blood type	Chi-square
Ordinal	Pain scale (0-10)	Mann-Whitney U
Discrete	Number of hospitalizations	Counts
Continuous	Blood pressure, BMI	Mean ± SD

Click here: Which type is "Pain Scale"? (Hint: It's not continuous!)

3 / 28

Quick Review: Roles of Variables

Independent (X)

The variable we manipulate or study.

Example:
Exercise Program

Dependent (Y)

The outcome we measure.

Example:
Weight Loss

NEW Today: Study design limits the relationship we can claim (Association vs Causation).

4 / 28

NEW Concept: Confounding

The Hidden Third Variable

Just because two things correlate, doesn't mean one causes the other.

Observation: Coffee drinkers have less heart disease.

Hidden Reality: Age affects both!

☕ Coffee

❤️ Heart Disease

⏰ AGE

5 / 28

Not All Research Is Equal

Meta-Analysis
RCTs
Cohort Studies
Case-Control
Cross-Sectional

Association (Bottom)

Causation (Top)

6 / 28

The Big Divide

Feature	Observational	Experimental (RCT)
Researcher Role	Observes Nature	Manipulates Exposure
Assignment	Natural	Random Assignment
Causation?	Difficult	YES
Cost & Time	Lower	Higher

Key Insight: Observational studies can suggest, but only experiments can prove.

7 / 28

Key Takeaways: Module 1

Hierarchy of Evidence: Not all studies are equal. RCTs and Meta-Analyses provide the strongest evidence for causation.

Confounding is Key: Always ask "What's the hidden variable?" A good study design aims to control for confounders.

Variables Matter: The type of variable (Nominal, Ordinal, Continuous) and its role (Independent, Dependent) dictate your entire research plan.

8 / 28

Module 2

The Four Study Designs

Cross-Sectional · Case-Control · Cohort · RCT

9 / 28

Design #1: The Snapshot

Cross-Sectional Study

Measure exposure + outcome at the same time.

Answers: Prevalence ("How common is it?")
Cannot Answer: Causation (No time sequence)

Saudi Health Interview Survey (SHIS)

Surveyed 10,000 households once.

Found: Mean BMI = 29.4 kg/m² · 23.9% have diabetes

Did obesity cause diabetes, or did diabetes cause weight change? We don't know.

📊 Key Metrics: Prevalence · Mean ± SD · Proportions · Chi-square (χ²)

Cannot calculate: Risk Ratio, Odds Ratio, Incidence Rate (no time component)

10 / 28

Design #2: Working Backwards

The Detective Approach

Find people WITH disease (Cases).
Find people WITHOUT disease (Controls).
Look BACK at past exposures.

Best For: Rare diseases (MERS-CoV).

Group	Camel Exposure
MERS Cases (100)	75%
Healthy Controls (100)	15%

Odds Ratio = 17.0 (17× higher odds if exposed)

📊 Key Metric: Odds Ratio (OR)

OR = (a/c) ÷ (b/d) · OR > 1 = higher risk · OR = 1 = no association · 95% CI must not cross 1

11 / 28

Design #3: Following Forward

The Time Machine

Start with healthy people.
Measure exposure (Diet).
Follow FORWARD (10 years).
Count outcomes (Heart Attacks).

Strength: Proven time sequence.

PURE-Saudi Study

Baseline (2009): Measured Diet. Follow-up (2019):

High Carb: 25% events

Low Carb: 11% events

Risk Ratio = 2.3x

📊 Key Metrics: Risk Ratio (RR) · Hazard Ratio (HR) · Incidence Rate

RR = Risk(exposed) ÷ Risk(unexposed) · HR from Cox regression · Can also compute Attributable Risk (AR)

12 / 28

Randomized Controlled Trial illustration

Design #4: The Gold Standard

Randomized Controlled Trial (RCT)

Randomization is the magic ingredient. It balances Age, Gender, Wealth, AND unknown factors between groups.

If Group A does better, it MUST be the treatment.

Saudi Vaccine Trial

1,200 Participants Randomized

Vaccine
5 COVID Cases

Placebo
95 COVID Cases

Efficacy = 95%

Can we claim Causation? YES. ✓

📊 Key Metrics: RR / OR · ARR · NNT · p-value

ARR = |Risk₁ − Risk₂| · NNT = 1/ARR · "Treat 20 patients to prevent 1 event" · ITT vs Per-Protocol analysis

13 / 28

Summary: Choosing a Design

Design	Direction	Causation?	Key Metric	Saudi Example
Cross-Sectional	📷 Snapshot	NO	Prevalence, Mean ± SD	SHIS
Case-Control	⏪ Backward	NO	Odds Ratio (OR)	MERS
Cohort	⏩ Forward	Suggests	RR, HR, Incidence	PURE-Saudi
RCT	⏩ Forward	YES	ARR, NNT, RR	Vaccine Trial

14 / 28

Design #5: Meta-Analysis

Combining Power

Instead of doing one study, we find ALL studies on a topic and combine their statistics.

Larger sample size = More precise estimate.

🏆 Top of the Evidence Pyramid.

Gulf Diabetes Meta-Analysis

Combined: 47 distinct studies (including SHIS).

Total Sample: 180,000 people.

Pooled Prevalence: 20.8%

(95% CI: 18.1 – 23.5%)

📊 Key Metrics: Pooled Effect Size · I² (Heterogeneity) · Forest Plot · Funnel Plot

I² < 25% = low heterogeneity · Fixed vs Random effects model · Funnel plot asymmetry → publication bias

15 / 28

Study Design Selector

Can you ETHICALLY randomize?

YES

🏆 RCT

NO (Observational)

Has outcome happened?

YES

🔍 Case-Control

⏩ Cohort

Practice: "Does air pollution cause asthma?" → Cohort (Unethical to randomize).

16 / 28

Key Takeaways: Module 2

Time is the Decider: Cross-sectional (snapshot), Case-control (backwards), Cohort/RCT (forwards). The direction determines what you can conclude.

Randomization is Power: Only RCTs, through random assignment, can reliably control for both known and unknown confounders to establish causality.

Fit the Design to the Question: Use Case-Control for rare diseases, Cohort for risk factors, Cross-sectional for prevalence, and RCTs for interventions.

17 / 28

Module 3

Advanced Sampling

From Population to Sample — Without Bias

18 / 28

Population → Sample

Population
All Saudi Diabetics
↓
Sample
1,000 Studied

Selection Bias: If your sample doesn't look like the population, your results are wrong.

19 / 28

Sampling Methods & Bias Risk

Method	How It Works	Bias Risk	When to Use
Simple Random	🎲 Lottery	LOW	Homogeneous populations
Stratified	📊 Subgroups first	LOW	Ensure subgroup rep
Cluster	🏘️ Pick entire groups	MEDIUM	Large areas
Convenience	🚶 Whoever is available	HIGH	Pilot studies ONLY

20 / 28

Why Convenience Sampling is Dangerous

Study A: Convenience

Method: Survey patients at KAU Hospital.

Result: 65% Obesity

Mean BMI: 32.4 kg/m²

Why? Hospital patients are sicker than average.

Study B: Random

Method: Random national ID sample.

Result: 35% Obesity

Mean BMI: 29.4 kg/m²

Why? Represents everyone.

Selection Bias = 30% Overestimate!

21 / 28

Ensuring Representation: Stratified Sampling

PURE-Saudi split the population into strata (layers) to ensure fair representation.

Stratum	Population %	Sample Size	Why?
Urban	83%	1,699	Most Saudis live here.
Rural	17%	348	Different lifestyle.
Gender	50/50	1,044 M / 1,003 F	Balanced.

22 / 28

Key Takeaways: Module 3

Goal is Generalizability: A good sample accurately reflects the entire population, allowing you to generalize your findings.

Convenience Kills Validity: Convenience sampling is the easiest method but introduces severe selection bias, making results unreliable.

Stratify for Accuracy: Use Stratified Sampling when you have important subgroups (like urban/rural) to ensure each is fairly represented in your sample.

23 / 28

Module 4

Bias & Confounding

The enemies of valid research

🎯

Selection Bias

Sample ≠ Population

Ex: Hospital study

🧠

Recall Bias

Bad memory

Ex: Case-Control studies

📏

Measurement Bias

Bad tools/procedure

Ex: Uncalibrated BP cuff

👻

Confounding

Hidden variable

Ex: Age affects result

24 / 28

Deep Dive: Confounding

Coffee & Heart Disease

Naive Conclusion: Coffee protects the heart.

Reality: The "Coffee Drinkers" group was different.

It wasn't the coffee. It was the Age, Exercise, and Wealth.

Confounder	☕ Coffee	🚫 Non-Drinkers
Mean Age	35 Years	55 Years
Exercise	65%	30%
Wealth	High	Low

25 / 28

Bias Prevention Checklist

Strategy	Prevents Which Bias?	Study Type
🎲 Randomization	Confounding	RCTs
🙈 Blinding	Observer/Participant Bias	RCTs
📐 Standardization	Measurement Bias	All Studies
📋 Medical Records	Recall Bias	Case-Control

26 / 28

PICO: Structuring Your Research

Before you choose a design, fill in the blanks.

P

Population

Saudi adults (40-65) with Pre-diabetes

I

Intervention

Supervised Exercise

C

Comparison

Standard Care

O

Outcome

Diabetes Incidence

Can we randomize? YES → Design: RCT · Sampling: Stratified Random

27 / 28

Final Summary: Design Before Data

PICO is your Blueprint: A well-defined PICO question makes choosing the right study design and sampling method straightforward.

Anticipate and Neutralize Bias: Good research isn't about avoiding bias entirely (that's impossible), but about recognizing potential biases (Selection, Recall, etc.) and actively using strategies like randomization and blinding to minimize their impact.

The Chain of Validity: Your final conclusion is only as strong as the weakest link in your research chain: PICO → Design → Sampling → Bias Control. A flaw in any one part compromises the entire study.

28 / 28

1 / 28

↑ ↓ Navigate · Home End

🧬 Aqrab Lab

Study Design & Sampling

Why PURE-Saudi Results Are Trusted

Good Design = Trusted Results

Quick Review: Variable Types

Quick Review: Roles of Variables

Independent (X)

Dependent (Y)

NEW Concept: Confounding

The Hidden Third Variable

Not All Research Is Equal

The Big Divide

Key Takeaways: Module 1

The Four Study Designs

Design #1: The Snapshot

Cross-Sectional Study

Saudi Health Interview Survey (SHIS)

Design #2: Working Backwards

The Detective Approach

Design #3: Following Forward

The Time Machine

PURE-Saudi Study

Design #4: The Gold Standard

Randomized Controlled Trial (RCT)

Saudi Vaccine Trial

Summary: Choosing a Design

Design #5: Meta-Analysis

Combining Power

Gulf Diabetes Meta-Analysis

Study Design Selector

Key Takeaways: Module 2

Advanced Sampling

Population → Sample

Sampling Methods & Bias Risk

Why Convenience Sampling is Dangerous

Study A: Convenience

Study B: Random

Ensuring Representation: Stratified Sampling

Key Takeaways: Module 3

Bias & Confounding

Selection Bias

Recall Bias

Measurement Bias

Confounding

Deep Dive: Confounding

Coffee & Heart Disease

Bias Prevention Checklist

PICO: Structuring Your Research

P

I

C

O

Final Summary: Design Before Data

Table of Contents