Where We Left Off
A week had passed since Dr. Junaid Rashid walked out of the last session with a cold cup of tea and a question he finally knew how to answer.
He had his cross-sectional study mapped out. Hypertensive patients in his OPD. A validated questionnaire. A sampling strategy. Three months, he had said. Doable.
He arrived this evening with his notebook open to a new page and a new question written at the top in the same careful handwriting.
“Is non-compliance linked to the hypertensive crises I keep admitting?”
He read it out before he had even sat down.
“Every month I admit patients with a hypertensive crisis,” he said. “Almost all of them tell me they had stopped their tablets. But I also have many stable patients who miss doses. Am I seeing a real link, or only what I expect to see? My frequency study will not tell me that.”
Dr. Sumaira Talib, already seated, looked up with an expression that suggested she had been thinking about the same problem.
Dr. Hammad Ali was drawing something in his notebook that may or may not have been a flowchart.
Dr. Hassan Raza had his arms crossed in the way he does when he is waiting to see where something is going.
Dr. Muhammad Yaqoob let Dr. Junaid’s question sit in the air for a moment.
“Last time,” Dr. Yaqoob said, “I told you the cross-sectional study has limits. You found one of them yourself, if I remember correctly.”
“Association, not causation,” Junaid said immediately.
“Good. You remembered. Tonight’s design is built for exactly your new question. You already have the outcome, the crisis, walking into your ward. The case-control study starts from that outcome and looks back at exposure. But I will warn you now: it does not solve the time problem. You reconstruct the exposure after the outcome has already happened, so a case-control study is weak on temporality. It tells you which factors are associated with an outcome. It never tells you, on its own, what caused it.”
The Detective Who Works Backwards
Dr. Yaqoob walked to the whiteboard and drew two columns.
On the left: CASES. On the right: CONTROLS.
“Here is the logic,” Dr. Yaqoob said. “You already know who got sick. Your cases are the patients who have the outcome. In Dr. Junaid’s scenario, the cases are hypertensive patients admitted with a hypertensive crisis. Your controls are hypertensive patients from the same medicine OPD whose blood pressure is controlled and who have never had a crisis. Same disease, same hospital, same catchment. The only thing missing is the outcome.”
Dr. Yaqoob drew arrows pointing backwards from both columns.
“Now you go back in time. You ask both groups: What were you exposed to before this happened? Were you non-compliant with medication? Did you have high salt intake? Did you have uncontrolled stress? Were you a smoker? Your main exposure is non-adherence. The others are factors you will need to account for.”
Dr. Yaqoob turned around.
“You are starting from the outcome and tracing backwards to find the exposure. Most study designs progress over time. The case-control study moves backwards. It is the detective’s design. You already have the crime. You are now looking for clues.”
Dr. Hammad stopped drawing and looked up. “So you already know who got sick and who didn’t, and then you look at what they were exposed to in the past?”
“Exactly.”
“That’s… actually clever,” he said, with the mild surprise of someone who had expected to be bored.
The Question the Cases and Controls Can Answer
Dr. Sumaira raised her hand. “Sir, what does the analysis actually give you? What number do you calculate?”
This was the right question to ask, and she knew it.
“The Odds Ratio,” Dr. Yaqoob said. He wrote it on the board: OR: Odds Ratio.
“Not relative risk. Not prevalence. The Odds Ratio. Because in a case-control study, you are not following a population through time to see who develops the disease; you are selecting people who already have it. Which means you cannot calculate incidence. You cannot calculate risk. What you can calculate is the odds of exposure among your cases compared to the odds of exposure among your controls.”
Dr. Yaqoob wrote the formula simply:
OR = (Odds of exposure in cases) ÷ (Odds of exposure in controls)
“An OR of 1 means no association. An OR greater than 1 indicates the exposure is more common among cases, suggesting it may be a risk factor. An OR less than 1 suggests it may be protective.”
Dr. Junaid was writing carefully. “So if my OR for medication non-compliance is 4.5…”
“Then the odds of having been non-compliant are 4.5 times higher among patients who had a crisis compared to those who did not. That is a meaningful signal. Not proof of causation, but a meaningful signal that demands further investigation. Always report it with its 95% confidence interval.”
He underlined something in his notebook.
Why This Design Exists, and When It Is the Only Sensible Choice
Dr. Yaqoob sat on the edge of the desk, which is where he tends to end up when the important part of an explanation arrives.
“Tell me,” Dr. Yaqoob said to the group, “what would happen if you tried to study a rare disease using a cohort design? Say, a condition that affects one in every ten thousand patients?”
Dr. Hassan answered immediately. “You would need an enormous sample. Tens of thousands of participants, followed for years, before you saw enough cases to analyse.”
“And the cost?”
“Enormous. The time? Even worse.”
“Exactly. This is why the case-control study exists. It was built for exactly this problem. Rare diseases. Conditions with long latency periods, such as diseases that take 10 or 20 years to develop after exposure. You cannot follow people for twenty years and wait. But you can find the people who already have the disease, find a comparable group who do not, and look back.”
Dr. Yaqoob listed the situations on the board:
Use a case-control study when:
- The disease is rare
- The latency period between exposure and outcome is long
- You want to study multiple risk factors for a single outcome
- Time and resources are limited
- You need an answer faster than a cohort study can provide
“One of the most famous early case-control studies linked smoking to lung cancer,” Dr. Yaqoob said. “Richard Doll and Austin Bradford Hill, 1950. They were not the first. Franz Müller in Germany published a case-control study on smoking and lung cancer in 1939, and Ernst Wynder and Evarts Graham in the United States published theirs in the same year as Doll and Hill. Doll and Hill did not follow thousands of smokers for twenty years and watch them develop cancer. They took lung cancer patients, the cases, and compared them to patients without lung cancer, the controls, and asked about their smoking histories. The association was strong and consistent. They did not report an odds ratio, because the odds ratio was only introduced by Jerome Cornfield in 1951. Together with the cohort studies that came later, this work changed medicine.”
The room was quiet.
Dr. Junaid looked at the board for a long moment. “And we can do that. In Pakistan. With the diseases that are here.”
“Yes,” Dr. Yaqoob said. “You can.”
The Limitations Nobody Warns You About, Until It Is Too Late
“Now,” Dr. Yaqoob said, “the part that will save you from a rejected manuscript.”
“Recall bias.”
Dr. Yaqoob wrote it on the board.
“You are asking people to remember what they were exposed to in the past. Your cases, the ones who had the crisis, have spent days or weeks in a hospital, thinking about why this happened to them. They are highly motivated to remember everything they did wrong. They will dig deep. They will reconstruct. Sometimes they will over-report exposures because they are looking for an explanation.”
Dr. Yaqoob paused.
“Your controls, stable patients attending the same OPD who have had no crisis, have no such reason to search their memory. They are less motivated to reconstruct their dietary and medication history from eight months ago. They may under-report. The result? Your cases appear more exposed than they actually were, not because they truly were, but because they remembered better.”
Dr. Sumaira nodded slowly. “So the bias isn’t in the disease. It’s in the memory.”
“It is always in the memory,” Dr. Yaqoob said. “This is why validated questionnaires, structured data collection tools, and, where possible, objective records from hospital files matter more in case-control studies than almost any other design. You do not want to be measuring memory. You want to be measuring exposure.”
Dr. Yaqoob moved to the second limitation.
“Selection of controls.” He underlined it twice. “This is where most case-control studies quietly fall apart. Your controls must come from the same source population as your cases. If your cases are hospital inpatients, your controls cannot be healthy community volunteers, because healthy community volunteers are, by definition, different from people who end up in hospitals. The comparison becomes meaningless.”
Dr. Hammad looked up. “So where do you get controls?”
“From the population your cases came from. For Dr. Junaid, that means hypertensive patients registered with the same medicine OPD who have not had a crisis. Not their attendants sitting in the waiting room, and not healthy staff. In other studies, controls may be patients admitted to the same hospital for unrelated conditions, or community controls if the cases came from the community. The rule is: your control, had they developed the outcome, would have been recruited as a case. That is the standard. Hold yourself to it.”
Dr. Yaqoob looked at Dr. Junaid. “What else cannot the case-control study give you?”
He thought for a moment. “Incidence. Prevalence.”
“Correct. You selected your sample based on outcome status. You cannot calculate how many people in the population develop the disease. For that, you need the cohort.”
The Moment Hammad Asked the Wrong Question Correctly
“Sir,” said Hammad, “can’t the OR be confused with relative risk? Like, can’t you just report OR and people will think you’re reporting risk?”
It was the wrong question, framed correctly enough to deserve a full answer.
“Yes,” Dr. Yaqoob said. “And this is a real problem in published literature. The Odds Ratio is not the same as the Relative Risk. When the outcome is rare, less than 10% in the population, the OR approximates the Relative Risk reasonably well, and many researchers use them interchangeably in that context [1]. But when the outcome is common, the OR exaggerates the association compared to the relative risk. Reporting an OR of 3.2 for a common outcome and calling it ‘three times the risk’ is technically incorrect and inflates the apparent effect size.”
Dr. Yaqoob wrote on the board:
Rare outcome → OR ≈ RR (acceptable approximation)
Common outcome → OR overestimates RR (do not conflate)
“Know your formula. Know what it measures. Know when the approximation holds and when it does not. Reviewers know. Editors know. If your paper conflates these and the reviewer catches it, and they will, the manuscript comes back.”
What Happened After the Session
Dr. Junaid stayed behind again.
This was becoming a habit, and Dr. Yaqoob did not mind it.
“Sir,” he said, “let me say my design back to you. Cases: hypertensive patients admitted with a hypertensive crisis. Controls: hypertensive patients from the same OPD whose blood pressure is controlled and who have never had a crisis. Two controls for every case. Exposure: non-adherence in the three months before, from a structured questionnaire and checked against the OPD records. Adjust for age, duration of disease, salt intake and smoking. Result: an odds ratio, with its confidence interval.”
He said it without prompting. Correctly. Completely.
“Then you have a case-control study,” Dr. Yaqoob said.
He was quiet for a moment.
“Fifteen years,” he said, not to Dr. Yaqoob exactly, but more to the room, or to himself. “Fifteen years I’ve been watching this happen and not knowing how to study it.”
“You know now,” Dr. Yaqoob said.
“I know now,” he agreed.
He closed his notebook and picked up his bag. At the door, he turned back.
“I will write down exactly where my controls will come from before I write anything else.”
“Good,” Dr. Yaqoob said. “That is where most case-control studies are won or lost.”
He always thinks before he comes, now. That is what fifteen years of unasked questions does when it finally finds the right room.
References:
- Dettori JR, Norvell DC, Chapman JR. Risks, rates and odds: What’s the difference and why does it matter? Global Spine Journal. 2021 Sep;11(7):1156-8.