5 Epidemiology Module - Study Design

Alison Grant 28/8/18

Compare and contrast the main types of study.

Remember to copy the notes from moodle.

Observational vs interventional studies O: Descriptive studies, describe what’s there O: Analytic studies, explain what’s happening (why has someone got this illness)

I: do something, see what happens

exposure: something you’re exposed to that affects your risk of disease (either protective or harmful) outcome: disease state, or event

5.1 Types of studies

5.1.1 Case Report

The experience of single patient or series of patients. Allows you to spread new info, individuals in case series may allow generation of hypothesis of disease causation.

An example in 1981 was a case series of PCP in los angeles in 5 young men. This was an unusual series of cases in people who hadn’t had this disease before, who were all noted to be MSM.

Problem is that it’s one researchers role, and any association observed could be coincidence. The problem is a lack of comparison group.

5.1.2 Cross-sectional studies

You are taiking a representative sample of a population (with disease, or expusire, or both). It’s a snapshot of the population.

Any survey is usually cross-sectional. To identify patterns over time needs multiple cross sectional studies.

Opinon polls should be a cross-sectional studies, or descriptive studies of disease prevalence, or an analytical study of assessing exposures to disease. Or assessing the spectrum of disease.

Cross sectional studies are quick and easy, often what you used first. You can also use these to generate hypotheses.

The problem is that you’re looking at exposure and disease at same time, you cannot discuss anything about timescale. You cannot show that exposure came before disease.

5.1.3 Case control

You take people with outcome of interest (cases) compare with folk without outcome of interest (controls)

Retrospective, look back from outcome to cause, best to call them case-control rather than retrospective.

Useful to investigate rare diseases, or diseases with a long incubate.

They are relatively quick and inexpensive, good for rare diseases, you don’t have to wait for the disease to happen. You can use them to investigate multiple exposures

Disadvantages Not useful for a rare exposure though, and vulnerable to bias, as you’re recording backwards, can be difficult to work out if exposure or outcome came first. You can’t use them to measure disease incidence.

As an aside, “Nested” case control, you can nest a case control into a cohort study. At the end of a cohort study you can select those with the outcome, and look back to investigate for additional exposures. Useful for looking back on those with additional expensive analyses.

5.1.4 Cohort

Simplest cohort is natural history, you have a bunch of people with an exposure, and you follow them up to see if they develop the outcome of interest. This is called a descriptive cohort.

But usually you want analytic cohort, where you get folk with and without exposure, and then you follow them up to see if there are different rates in outcome of interest.

Cohort studies give you good info about: incidence, prognosis, causation.

Good for rare exposures, and can examine multiple outcomes. You definitely know the timescale, and that exposure precedes disease. You get accurate info about exposure and outcome

But they are less useful for rare outcomes like cancer. And can be vulnerable to losses to follow up. They can be expensive and time consuming,

Prospective cohort, people are followed up in real time.This is ideal, but expensive.

Retrospective cohort, you reconstruct this from existing records. But really relies on accurate data. This can work with standardised data collection, but more difficult without. V poor for recording events that are recorded poorly.

Cohort studies don’t necessarily provide a representative sample of the community, that’s what cross sectional studies should do.

5.1.5 RCTs

The gold standard. They are v similar to cohort. But the investigator allocates exposure. You can have multiple outcomes but you need to state them in advance.

They advantage is that the exposure should be the only difference really between the two groups. Therefore the outcome difference should be attributable to the exposure.

So one of the big advantages is should eliminate internal biases. But they are expensive, difficult, can be too small or too short. They may be unethical (e.g. smoking exposure), or require too many patients. They are also difficult to compare study practices to real life medicine.

Interventional

5.1.6 Evaluating studies

PICO Population Intervention/Exposure Comparator Outcome

5.1.7 Bias

Systematic error in a study, that results in a colclusion which is different from the truth.

It aries because uou design a study badly. Clever analysis cannot correct bias! Studying the wrong people or collect information badly can’t be fixed.

Two main types are Selection bias and Information bias.

Selection bias - error in selection the study population. Or in analytic studies the comparison group are not comparable.

Information bias - error in the measurement of the exposure or the outcome. But it needs to be differing levels of inaccurate measurements between groups, to be a true bias.

An example of “social desirability bias” - patients lying to say what they think the researcher wants to hear, or to say the socially acceptable thing.

Checking for selection bias:

READ NOTES ON MOODLE FOR KEY QUESTIONS TO ASK

Bias versus measurement error. Precision is measurement error. If i measure the same thing lots of time, do I get similar answers? Accuracy is measurement error. If I measures something, do I get close to the true answer?