Research Methods:

1.1 The Two Foundational Questions That Define Every Study

Before we can even name a study design like "cohort" or "case-control," we must understand that these designs are simply the logical outcomes of answering two fundamental questions. Every single study, no matter how complex, is built upon the answers to these two questions.

Think of a psychiatrist trying to understand the origins of a complex disorder. The "outcome" is the diagnosis (e.g., schizophrenia). The "exposure" is a potential contributing factor (e.g., cannabis use). The psychiatrist's entire investigation will be shaped by how they answer these two questions:

1. Where do I start my investigation? (The Grouping Strategy)
2. Which direction in time do I look? (The Arrow of Time)

Question 1: Where Do You Start? (The Grouping Strategy)

This is the most important strategic decision a researcher makes. It dictates the entire structure of the study. The psychiatrist has two possible starting points for their investigation.

Option A: The Outcome-First Approach

Imagine a psychiatrist working in a clinic for patients with established schizophrenia. The diagnosis (the "outcome") is already known. The psychiatrist's job is to work backwards from this known outcome to understand potential causes. They start by identifying the patients with the diagnosis (the "cases"). Then, to understand what might be different about them, they need a comparison. They need to know what the past looks like for people who did not develop schizophrenia. So, they find a similar group of people from the community who are mentally well (the "controls").

The psychiatrist then investigates the past of both the cases and the controls, looking for potential risk factors (exposures). Did the patient with schizophrenia use cannabis heavily in their teens? Did they have birth complications? Did the healthy control person also have these experiences? The psychiatrist is comparing the frequency of past exposures between the "cases" and the "controls."

This is the absolute core logic of a Case-Control Study.

Option B: The Exposure-First Approach

Now, imagine a different clinical question. A psychiatrist is interested in the long-term mental health effects of a specific genetic marker. They don't start with people who are already ill. Instead, they start with a large group of healthy young people. They identify those who carry the genetic marker (the "exposed" group) and those who do not (the "unexposed" group). The psychiatrist's job is to follow both groups forward in time, perhaps for decades, to see who goes on to develop a psychiatric disorder.

The psychiatrist isn't starting with a disorder that has already happened. They are starting with a group defined by their biological characteristics ("exposed" vs. "unexposed") and waiting to see what outcome develops.

This is the absolute core logic of a Cohort Study.

Question 2: Which Way Do You Look?

This second question is inextricably linked to the first. Once you've chosen your starting point, it largely determines the direction of your investigation.

• Retrospective (Looking Backwards): If you choose the "Outcome-First" approach (Case-Control), you have no choice but to look backwards in time. The outcome has already occurred. Your entire investigation is about reconstructing the past. The arrow of time for your data collection points from the present to the past.
  • Strength: Fast and cheap.
  • Weakness: You are at the mercy of memory and old records. This is called recall bias. A person with a serious illness (a case) might search their memory much harder for a cause and be more likely to remember an exposure than a healthy person (a control).
• Prospective (Looking Forwards): If you choose the "Exposure-First" approach (Cohort), your investigation is designed to move forwards. You establish your groups in the present and follow them into the future. The arrow of time for your data collection points from the present to the future.
  • Strength: This is scientifically more powerful. You measure the exposure before the outcome happens. This eliminates recall bias and establishes a clear temporal relationship (the cause came before the effect), which is a vital piece of evidence for causation.
  • Weakness: Slow and very expensive.
• Cross-Sectional (A Snapshot in Time): There is a third option. What if a psychiatrist wants to understand the relationship between burnout and working hours among trainees? They send out a single survey. The exposure (working hours) and the outcome (burnout score) are measured at the exact same time. The psychiatrist is measuring everything in a single snapshot.
  • How it works: You take a defined population and measure both the exposure and the outcome simultaneously.
  • Clinical Example: You survey a group of 500 junior doctors. In the same questionnaire, you ask them about their current work hours (exposure) and their current score on a burnout scale (outcome). You might find that those with longer hours have higher burnout scores.
  • The Critical Weakness: You cannot tell which came first. This is the problem of temporality. Did the long hours cause the burnout? Or did the doctors who were already burning out become less efficient and therefore have to work longer hours? A cross-sectional study can show an association is present, but it is very weak at explaining why it is present.

1.2 The Three Families of Study Design

Now that we understand the two foundational questions—Where do you start? and Which way do you look?—we can see how the answers logically combine to create three distinct "families" of research. Each family has a different purpose, a different level of strength, and answers a different kind of clinical question.

1.3 The Descriptive Family

Descriptive studies are the bedrock of medical inquiry. They provide the initial observations that spark more complex research. Their function is to describe the characteristics of a sample or population, and they do not have a comparison or control group in the formal sense.

1.3.1 Case Report and Case Series

This is the simplest and most traditional form of medical writing.

• Case Report: A detailed, narrative account of a single, interesting patient case. It might describe an unusual presentation of a common illness, a rare side effect of a medication, or a novel treatment approach for one person.
• Case Series: A collection of individual case reports involving several patients who share similar characteristics.
• Purpose: To be a "red flag" or an "idea generator." They alert the medical community to a potential new phenomenon.
• Psychiatric Example: A psychiatrist working in an early intervention service notices that five young men have presented in the last month with their first episode of psychosis. During history taking, she discovers that all five had been heavily using a new, illegally-marketed vaping liquid. She writes up a paper describing the common features of these five cases: the specific psychotic symptoms, the temporal link to the vaping, and their response to antipsychotics. This is a case series.
• Strengths:
  • Excellent for identifying rare conditions or novel adverse effects.
  • Can generate hypotheses for future, more rigorous studies (e.g., "Does this vaping liquid cause psychosis?").
• Weaknesses:
  • There is no comparison group. We don't know how many people used the liquid and didn't get psychosis.
  • It is essentially a collection of anecdotes and cannot prove association, let alone causation. It is highly susceptible to bias and coincidence.

1.3.2 Cross-Sectional Studies (Prevalence Studies)

This is a more structured descriptive design that provides a "snapshot" of a population at a single point in time.

• How it works: The researcher defines a population of interest and measures both the exposure(s) and the outcome(s) simultaneously. Think of it as taking a single photograph of a crowd; you can see who is tall (exposure) and who is wearing a hat (outcome) at that exact moment, but you don't know if they were born tall or if wearing a hat made them tall.
• Purpose: Its primary purpose is to measure the prevalence of a condition. This is why they are often called prevalence studies.
• Psychiatric Example: A public health team wants to determine the prevalence of Post-Traumatic Stress Disorder (PTSD) among firefighters in a large city. They send a one-time survey to all 2,000 firefighters. The survey contains a validated questionnaire to screen for current PTSD symptoms (the outcome) and also asks about their years of service (an exposure). The results show that 300 firefighters (15%) currently meet the criteria for PTSD. This 15% is the point prevalence. The study also shows that firefighters with more years of service have higher rates of PTSD, suggesting an association.
• Strengths:
  • Relatively quick and inexpensive to conduct.
  • The best design for determining the prevalence of a condition.
  • Good for generating hypotheses about associations.
• Weaknesses:
  • The cardinal weakness is the inability to determine temporality (the "chicken and egg" problem). In our example, did more years of service (exposure) lead to PTSD (outcome)? Or did firefighters who were predisposed to PTSD stay in the job longer, while others left? The snapshot can't tell us the direction of the relationship.
  • Because of this, it is very weak for testing hypotheses about causation.

1.4 The Analytical Family

This is where true hypothesis testing begins. Analytical studies always involve a comparison between groups to quantify the statistical relationship between an exposure and an outcome. The researcher is still an observer—they do not assign the exposure.

1.4.1 Case-Control Studies (The Retrospective Detective)

As we discussed, this design starts with the "disorder" (the outcome) and works backwards to find the "clues" (the exposure).

• Design Recap:
  1. Start with the Outcome: Identify a group with the disease (Cases) and a group without the disease (Controls).
  2. Look Backwards: Compare the past exposure rates between these two groups.
• Psychiatric Example: Researchers want to test the hypothesis that prenatal maternal stress is a risk factor for Autism Spectrum Disorder (ASD) in children.
  • Cases: They recruit 100 children with a confirmed diagnosis of ASD.
  • Controls: They recruit 100 neurotypical children, matched for age and socioeconomic status.
  • Look Backwards: They conduct detailed interviews with all the mothers, using a validated scale to quantify the level of psychosocial stress (e.g., bereavement, financial hardship, relationship distress) they experienced during their pregnancy.
  • Result: If the average stress score is significantly higher in the mothers of the ASD children compared to the mothers of the control children, it supports the hypothesis of an association.
• Strengths:
  • The best design for studying rare diseases like ASD or schizophrenia.
  • Relatively quick and inexpensive.
• Weaknesses:
  • Recall Bias: This is a major flaw. A mother of a child with ASD (a case) may have spent years searching for a reason and might remember the stress of her pregnancy more vividly and report it more thoroughly than the mother of a neurotypical child (a control).
  • Selection Bias: Finding an appropriate control group is extremely difficult and can heavily bias the results.
  • Cannot calculate the incidence or absolute risk of the disease.

1.4.2 Cohort Studies (The Prospective Observer)

This is the most powerful type of observational study because it moves forward in time, mirroring the natural course of events.

• Design Recap:
  1. Start with the Exposure: Identify a group of people free from the outcome and classify them as Exposed or Unexposed to a risk factor.
  2. Look Forwards: Follow both groups over time to see who develops the outcome.
• Psychiatric Example: A research team wants to investigate if social isolation in older adults (exposure) is a risk factor for developing dementia (outcome).
  • The Cohort: They recruit 5,000 people aged 65 and over who are all dementia-free at the start of the study.
  • Grouping by Exposure: Using a social network index, they classify the participants into two groups: 1,000 who are socially isolated (the Exposed group) and 4,000 who are socially active (the Unexposed group).
  • Follow Forwards: They follow the entire cohort for 10 years, conducting cognitive assessments every two years to identify any new (incident) cases of dementia.
  • Result: If, after 10 years, 15% of the isolated group has developed dementia compared to only 5% of the socially active group, this provides strong evidence that social isolation is a risk factor.
• Strengths:
  • Establishes temporality—the exposure is measured before the outcome develops. This is a critical step towards inferring causality.
  • Can measure the incidence (rate of new cases) of the disease in both groups.
  • Minimizes recall bias.
• Weaknesses:
  • Very expensive and time-consuming (can take decades).
  • Prone to attrition bias (participants dropping out over time), which can skew results.
  • Inefficient for studying rare diseases—you would need an enormous cohort to get enough cases.

1.5 The Experimental Family

In the hierarchy of evidence, experimental studies sit at the very top for questions of causation. Unlike descriptive studies that merely paint a picture, or analytical studies that observe associations, experimental studies are designed to prove that an intervention or exposure causes a specific outcome.

The defining characteristic of an experimental study is intervention. The researcher is no longer a passive observer; they actively manipulate one variable (the "exposure" or "intervention") and then measure its effect on another variable (the "outcome"). This active manipulation allows for the strongest inferences about cause and effect.

1.5.1 Randomized Controlled Trials (RCTs)

The Randomized Controlled Trial (RCT) is the cornerstone of evidence-based medicine and the most robust design for evaluating the efficacy and safety of new treatments. It is essentially a prospective cohort study where the researcher controls the exposure by randomly assigning participants to different groups.

Core Principle: Randomization

• What it is: Randomization is the process by which participants are assigned to either the intervention group (receiving the new treatment) or the control group (receiving a placebo or standard treatment) purely by chance.
• Why it's crucial: This is the single most important feature of an RCT. Its purpose is to create groups that are, on average, identical in every way except for the intervention they receive. By distributing both known and unknown confounding factors evenly between the groups, randomization minimizes selection bias.

Core Principle: Control Group

• What it is: The control group serves as the baseline for comparison. Without it, you cannot know if the observed changes in the intervention group are due to the treatment itself, the natural course of the illness, or other factors (like the placebo effect).
• Types of Control:
  • Placebo Control: An inert substance (e.g., a sugar pill).
  • Active Control: The current standard treatment. Used when it would be unethical to withhold treatment.
  • Wait-list Control: Participants are told they will receive the treatment after the study period.

Core Principle: Blinding (Masking)

• What it is: Blinding refers to keeping participants, researchers, and/or outcome assessors unaware of which treatment arm a participant is in.
• Levels of Blinding:
  • Single-blind: Only the participants are unaware.
  • Double-blind: Both the participants and the researchers/clinicians are unaware. This is the most common and preferred method.
  • Triple-blind: Participants, researchers, and the data analysts are all unaware.
• Why it's crucial: Blinding prevents observer bias and participant bias.

1.5.2 Crossover Trials

A crossover trial is a specialized type of RCT that is particularly efficient for certain types of clinical questions, especially when comparing two active treatments for chronic, stable conditions.

• Core Principle: Within-Subject Comparison: In a crossover trial, each participant receives all the interventions being tested, but in a randomized sequence. For example, in a two-treatment trial (Treatment A and Treatment B), half the participants might receive A then B, and the other half receive B then A.
• Washout Period: Crucially, there is usually a "washout period" between the two treatments. This prevents carryover effects, where the residual effect of the first treatment influences the response to the second.
• Strengths: High Statistical Power/Efficiency (each participant serves as their own control).
• Weaknesses: Only suitable for chronic, stable conditions. Risk of carryover effects.

1.5.3 Uncontrolled Trials

While technically "experimental" because an intervention is applied, uncontrolled trials lack a comparison group. Often used in early-phase drug development (Phase I or II) to assess safety.

2.0 Sampling and Randomization

2.0 Introduction

A study's credibility rests on two pillars: who you study, and how you compare them. These are governed by the distinct processes of sampling and randomization. It is absolutely critical to understand that these are not the same thing.

2.1 Sampling

In an ideal world, to find out the prevalence of depression in the UK, we would assess every single person in the UK (a census). This is impossible. Therefore, we study a smaller, manageable group—a sample.

2.1.1 Probability (Random) Sampling

In probability sampling, every single individual in the population has a known, non-zero chance of being selected. This is the "gold standard."

• Simple Random Sampling: Every person has an equal chance of being picked (names in a hat).
• Systematic Sampling: You select individuals at regular intervals from a list (e.g., every 10th person).
• Stratified Sampling: You first divide the population into "strata" (subgroups) based on a key characteristic (e.g., age, gender) and then perform simple random sampling within each stratum. This guarantees specific subgroups are properly represented.
• Cluster Sampling: The population is divided into pre-existing "clusters" (e.g., different hospitals), and a random sample of the clusters is chosen. Then, every individual within the selected clusters is studied.

2.1.2 Non-Probability Sampling

Selection is not random. Easier and cheaper, but higher risk of bias.

• Convenience Sampling: Selecting whoever is easiest to reach.
• Quota Sampling: Aiming to include a certain number of participants with specific characteristics, but selecting them non-randomly.
• Snowball Sampling (Chain-Referral): Used when the population is hard to reach, hidden, or rare. The researcher starts with one individual (the "seed") and asks them to refer others.

2.2 Randomization

Once you have your sample, if you are conducting an RCT, you must allocate them to the treatment and control groups. Randomization ensures this allocation is free from bias.

2.2.1 Simple Randomization

Like flipping a coin. Weakness: Can lead to unequal group sizes in smaller studies.

2.2.2 Block Randomization

Participants are randomized in small "blocks" (e.g., groups of 6). This guarantees that after every block, the groups are perfectly balanced. Crucial if a trial is stopped early.

2.2.3 Stratified Randomization

Used when there are specific baseline characteristics (prognostic factors) that are known to strongly influence the outcome. The goal is to ensure these crucial factors are perfectly balanced.

• How it works: First, divide participants into strata (e.g., "Mild" vs "Severe"). Then, perform randomization within each stratum.
• Benefit: Guarantees that the number of participants with that key characteristic is equal in both groups.

2.3 Ecological Studies and the Ecological Fallacy

Within the family of analytical studies, there is a unique design that is important to understand, primarily because of the specific and common error in interpretation associated with it.

• What is an Ecological Study? An observational study where the unit of analysis is a population or a group, not an individual. It compares groups with different levels of exposure using aggregated data (e.g., comparing suicide rates in different cities).
• The Critical Flaw: The Ecological Fallacy. The logical error of assuming that an association observed at the group level must also be true at the individual level.

2.4 Matching the Clinical Question to the Best Study Design