Blog 8: Understanding Association, Confounding, and What The Results Really Mean

Blog 7: Understanding Association, Confounding, and What The Results Really Mean

Welcome back.

If you’ve been following this series, you’ve already done something many first-time researchers never fully do: you’ve taken a clinical idea, turned it into a measurable outcome, and started identifying realistic exposure variables. That’s real progress.

Now we move into a new phase — the point where your study stops being an idea and starts becoming evidence.

But to do this well, you need one mindset shift:

Observational research is about association — not conclusions.

Let’s unpack what that means using our anchor question.

Returning to our working research question

In elderly postoperative patients admitted to our ward, what postoperative clinical, functional, and system-related factors are associated with delayed mobilisation after surgery?

Notice the wording.
It does not ask:

  • What causes delayed mobilisation?
  • Who or what is to blame?

It asks what is associated.

That single word keeps your research cautious, and scientifically sound.

When an association looks like a cause — but may not be

In the previous blog, we discussed the range of variables that could influence delayed mobilisation, grouped under three broad categories: postoperative clinical factors, functional factors, and system-related factors.

Let us now assume your data collection is complete.

The next step is to understand and analyse the data. However, it is worth emphasising that this step should not begin only after data collection. Ideally, you should have written a clear analysis plan at the very beginning of the study, outlining what you intended to test and how.

Now, imagine that while exploring your results, you find the following pattern:

Patients who received higher doses of opioids were more likely to have delayed mobilisation.

“Opioids delay mobilisation.”

But pause.

In real clinical practice, opioids are more commonly given to patients who:

  • Have higher postoperative pain
  • Underwent more extensive surgery

So opioids may not be the true reason for delayed mobilisation.
They may simply be a marker of more severe pain.

To make this clearer, think of a non-clinical analogy.

The ice-cream analogy: when correlation misleads

Ice-cream sales are associated with higher drowning rates. But ice-cream does not cause drowning. The real reason is a third factor — hot weather:

  • Hot weather → more ice-cream consumption
  • Hot weather → more swimming → more drowning

Ice-cream and drowning look linked only because both rise in summer. The same logic applies in clinical research.

Confounding — the hidden third factor

When a third variable makes two things look connected, we call it confounding.

In our mobilisation study:

  • Exposure: Opioid use
  • Outcome: Delayed mobilisation
  • Confounder: Pain severity

Pain severity:

  • increases the likelihood of opioid use
  • increases the difficulty of early mobilisation

So opioids may appear responsible —when pain severity is actually influencing both.

Why this matters?

Confounding is not a mistake. It is expected in real-world clinical data. Your role as a researcher is not to eliminate confounding, but to recognise it and adjust for it,
so your conclusions reflect reality — not illusion.

Why analysing one factor at a time can mislead

Researchers often:

  • test each variable separately
  • compare delayed vs non-delayed groups
  • report p-values

This is called unadjusted (univariable) analysis.
It can show patterns — but clinical reality is never one factor at a time.

Delayed mobilisation may involve:

  • pain
  • opioids
  • drains
  • hypotension
  • frailty
  • staffing
  • physiotherapy delays

These influences overlap. So the real question becomes: Does this factor still matter after accounting for other relevant factors? That is the purpose of adjusted analysis.

The key tool for a binary outcome: Logistic regression

Because delayed mobilisation is a yes/no outcome, the most commonly used method to study adjusted associations is:

Binary logistic regression

In simple terms, it:

  • looks at multiple factors together
  • estimates how much each one matters
  • while accounting for the others

It produces two key numbers:

  • an Odds Ratio (OR)
  • a 95% Confidence Interval (CI)

You don’t need to be a statistician — but you do need to understand what these numbers mean.

What logistic regression is really doing — a simple analogy

Think of delayed mobilisation like the final scene of a movie.

The impact of the scene depends on:

  • the actor
  • the lighting
  • the music
  • the camera angle

If the scene feels powerful, we might ask:

How much of that impact came from the actor alone — separate from the other elements?

We cannot replay the scene without music or lighting —
but we can estimate how much each element contributed.

Logistic regression does the same thing with clinical factors.

It helps separate overlapping influences so we can estimate the independent role of each factor.

What “adjusted” actually means

Adjusted does not mean:

  • patients were made identical
  • real-world differences disappeared
  • other factors were ignored

It means:

We mathematically account for other influences so we can estimate the effect of one factor more fairly.

Nothing changes in real life — only in the statistical comparison.

What an Odds Ratio (OR) really tells you — in plain language

An Odds Ratio answers this question:

How much more (or less) likely is delayed mobilisation when this factor is present?

Example:

  • OR = 1 → No difference
  • OR = 2 → About twice the odds
  • OR = 0.5 → About half the odds

So if your results show:

  • Catheter → OR 2.1
    This means:

Patients with a catheter had about twice the odds of delayed mobilisation compared to those without — after adjusting for other factors.An odds ratio of 2 means the event is roughly twice as likely to occur in one group compared to another — it does not mean it will happen in everyone.

The confidence interval (CI) tells you how precise that estimate is.

A common beginner mistake: “Let’s add everything into the model”

Many new investigators think:

“If I include every variable, confounding is solved.”

Not always.

Too many variables can:

  • make results unstable
  • create overlapping predictors (collinearity)
  • adjust for consequences instead of causes
  • reduce clarity instead of improving it

More variables ≠ better science.

Choosing variables wisely: think clinically first

A practical beginner rule:
Include variables that:

  • could realistically bias the relationship
  • could mislead the interpretation
  • could falsely inflate or weaken the association
  • could confuse the true cause of the outcome

Your model should reflect clinical reasoning — not random inclusion.

A simple self-check: the “time sequence” test

Ask yourself: Did this happen before delayed mobilisation?

Include:

  • type of surgery
  • anaesthesia type
  • baseline frailty
  • early pain scores
  • catheter or drains immediately post-op

Be cautious with:

  • length of stay
  • late complications
  • physiotherapy delays that occurred after mobilisation was already delayed

Avoid mixing causes, confounders, and consequences.

Don’t forget system factors — they’re often the most actionable

Delayed mobilisation is not only a patient issue — systems matter.

Instead of vague terms like:

  • “poor staffing”
  • “physio delay”

Define measurable exposures:

  • time to first physio review
  • first mobilisation attempt after day 2
  • weekend vs weekday surgery
  • shift timing of first mobilisation note

System variables often lead to the most practical and policy-relevant findings.

What your results can responsibly say

If your analysis shows:

  • Catheter → OR 2.1
  • High pain → OR 1.8
  • Weekend surgery → OR 1.6

You can say:
These factors were independently associated with delayed mobilisation
Patients with these characteristics had higher odds

You should not say:
X These factors caused delayed mobilisation

Observational studies prioritise risks — they don’t prove causality.

A moment of reflection — apply this to your own question

Ask yourself:

  • What is my outcome?
  • What exposures truly occur before it?
  • Which factors might confound the associations I am seeing?
  • If I report an Odds Ratio, can I explain what it means in plain language to a colleague?

If you can answer these, you’re no longer just collecting data —
you’re thinking like a researcher.

Coming up in Blog 9…

So far, we’ve talked about what to analyse and how to interpret it.

Next, we’ll tackle something just as important:

How to present your results clearly — in tables, text, and figures — without confusing or misleading your readers.

We’ll break down how to write results that are accurate, readable, and publication-ready.

Because observational studies can suggest and prioritise, but they do not prove causality.

Gomathi Sundar is a research consultant and medical writer with over 25 years in healthcare. She specializes in study design, protocol writing, research reports, and manuscripts. She founded Right Research Academy to support postgraduates, novice researchers, and medical companies.
Gomathi Sundar
    Leave a Reply

    Your email address will not be published. Required fields are marked *

    Blogs

    Other Blogs