The Photograph That Cannot Lie, and Cannot Explain

The Question That Started It All

Dr. Junaid Rashid arrived that evening carrying two things: a cup of tea he had not yet touched, and a question that had clearly been bothering him for longer than he wanted to admit.

Fifteen years. That is how long he had been practising medicine. Fifteen years of ward rounds, outpatient clinics, and emergency calls at two in the morning. He had seen more patients than most researchers ever recruit into a study. He knew what diseases looked like. He knew what they did to families.

What he did not know, not yet, was how to put any of that into a research question with a study design behind it.

He sat down, finally took a sip of his tea, and said, “Sir, I want to find out how many patients with hypertension in our OPD are actually compliant with their medication. I see non-compliance everywhere. I just want to know how common it is. What kind of study do I do?”

The room went quiet in that particular way it does when someone asks a question that turns out to be the exact right question for the evening.

Dr. Muhammad Yaqoob set down his marker: “You want a photograph,” he said. “Not a film.”

What a Cross-Sectional Study Actually Is

Dr. Sumaira Talib looked up from her notebook. She had been writing, as always, the kind of focused writing that suggested she was already thinking two steps ahead.

“A photograph?” she said.

“Think of it this way,” Dr. Yaqoob said. He walked to the whiteboard and drew a single vertical line. One moment in time. “You walk into your OPD on a Monday morning. You look at every hypertensive patient sitting in the waiting room. You measure their blood pressure. You ask them about their medication. You record their answers. Then you go home.”

He turned around.

“That is a cross-sectional study. You take one measurement, at one point in time, across a cross-section of your population. No follow-up. No before and after. No watching what happens next. Just a snapshot: clear, clean, and exactly as it is on that particular day.”

Dr. Junaid nodded slowly. “So it tells me the prevalence.”

“Exactly,” Dr. Yaqoob said. “Prevalence. How many people in a defined population, at a specific moment, have a condition, a behaviour, or an exposure? Not how many develop it over time; that would require something else, something we will come to. But right now, if your question is, “How common is this thing?”, the cross-sectional study is your answer.”

He wrote it on the board:

Cross-Sectional Study = Prevalence = Snapshot in Time

Why Pakistani Doctors Need This Study More Than They Realise

Dr. Hammad Ali had been quiet for approximately four minutes, which for Hammad was a record. He broke it.

“Sir, but isn’t this a bit… simple? You just go and ask people things? Where is the science?”

Dr. Yaqoob had been expecting this.

“Simple to design,” he said. “Not simple to execute, and not simple to interpret. And before you dismiss it, let me tell you what this so-called simple study has told the world.”

He turned to the board again.

The Pakistan Demographic and Health Survey. The WHO STEPwise Approach surveys that map the prevalence of non-communicable diseases across entire nations. These are national cross-sectional surveys. Large projects such as the Global Burden of Disease study are different: they are modelling syntheses that combine data from many sources, including surveys like these. Much of national health planning, including decisions about where hospitals are built, where funds are allocated, and which conditions get public health campaigns, relies on data from cross-sectional surveys.

“When the government of Pakistan needs to know how many adults in rural Sindh have diabetes,” he said, “they do not follow people for ten years and wait for them to develop it. They go out, they test a representative sample, and they record the numbers. Cross-sectional. Done in months. Actionable immediately.”

Dr. Hassan Raza, who has the particular expression of someone running cost-benefit calculations in real time, nodded. “Fast and affordable,” he said.

“Fast, affordable, and ethical,” Dr. Yaqoob said. “You are not withholding treatment from anyone. You are not following vulnerable patients into unknown outcomes. You are observing what already exists. For a health system like ours, where resources are limited and much of the burden of disease is still poorly mapped, this is not a small thing.”

The Day Dr. Junaid Saw the Limitation

Thirty minutes in, Dr. Yaqoob turned to Dr. Junaid.

“Tell me your question again,” he said.

“How many of my hypertensive patients are non-compliant with medication?”

“Good. You run your study. You find that 60% report non-compliance. What do you do with that number?”

“I… publish it? Use it to make a case for a compliance programme?”

“Yes,” Dr. Yaqoob said. “You can absolutely do that. Now tell me something else. You also notice in your data that non-compliant patients tend to have lower levels of education. You want to say in your paper that low levels of education cause non-compliance. Can you?”

Dr. Junaid opened his mouth, then stopped.

Something shifted behind his eyes: the particular flicker of a clinician who has just encountered a methodological boundary for the first time and recognises it, even before he has the language for it.

“I don’t think so,” he said slowly. “But not because of timing. A person’s schooling is finished long before they get hypertension. So education clearly came first.”

The room sharpened.

“Go on,” Dr. Yaqoob said.

“I measured both things at the same time. Education level and compliance. So I know they’re related; they appear together in my data. But my less educated patients are also often poorer and older. Some cannot afford the tablets. Maybe it is the money or the age, some third factor, and not the schooling, that stops them taking their medicine. I can see the association. I can’t explain it.”

Dr. Yaqoob said nothing for a moment.

“That,” he said finally, “is the most important thing you will learn this evening. And you figured it out yourself. It has a name: confounding. Income and age are linked to education and also to adherence, so they can create or hide an association. You deal with it by measuring those factors and adjusting for them in the analysis. But you can never adjust for something you did not measure.”

Association, Not Causation

He wrote it on the board in large letters:

ASSOCIATION ≠ CAUSATION

“The fundamental limitation of the cross-sectional study,” Dr. Yaqoob said. “You measure exposure and outcome simultaneously. For many pairs of variables, you cannot know which came first. You cannot establish a causal pathway. You can say: these two things exist together in this population at this moment. You cannot say: one caused the other.”

Dr. Sumaira looked up. “Like the chicken and the egg.”

“Exactly like the chicken and the egg. Did the depression come before the chronic pain, or did the chronic pain lead to the depression? Each can cause the other. In a cross-sectional study, you cannot answer that. You find them together. Their sequence is invisible to you. With education and adherence, time was not the problem, because education is fixed early in life. With depression and pain, time is exactly the problem.”

Dr. Yaqoob watched the group absorb this. Hammad, who had dismissed the design as simple twenty minutes ago, was now writing at some speed.

“There is another limitation,” Dr. Yaqoob said. “Prevalence-incidence bias. Or what some call survivor bias. Think about it: you are recruiting patients who are currently in your OPD with hypertension. The patients who died of a hypertensive crisis last year are not in your waiting room. The ones who moved away are not there. The ones who recovered and never came back are not there. Your sample is, by definition, the survivors. The ones still present. This means your prevalence estimates can be skewed; you may be over-representing patients with milder, more manageable disease, simply because they are the ones still alive to participate.”

The room was quiet.

Dr. Hassan said: “So the study is honest about what it sees, but what it sees isn’t the whole picture.”

“The photograph,” Dr. Yaqoob said, “captures exactly what is in front of the camera. It does not capture what happened before the shutter clicked, or what happens after. That is not a flaw in the photograph. It is simply what a photograph is. The mistake is in asking a photograph to be a film.”

What It Is Good For, and When to Stop

Dr. Yaqoob put the marker down and sat on the edge of the desk.

“So when should you use it?”

He listed them one by one.

When you want to estimate the prevalence or burden of a disease, behaviour, or risk factor in a population. When you are doing a needs assessment: finding out what a community lacks, or what conditions are going untreated. When you are generating hypotheses: noticing patterns that you can then investigate more deeply with a stronger design. When you are working under time and funding constraints and need to produce actionable data quickly.

“And when should you stop and choose a different design?” he continued.

When your question is about causality: what caused this condition, what will reduce it, does this intervention work? When you are studying diseases with very short durations: conditions so brief that by the time you arrive to measure them, they have already resolved. When your outcome is rare, because a cross-sectional study of a rare disease will struggle to find enough cases in a single snapshot. For all of these, you need something else. A cohort study, perhaps. Or a case-control.

“Those,” he said, looking around the room, “each deserve their own evening. Each design is a different kind of lens. The cross-sectional study is your widest lens: broad, fast, honest about what it can see. What it cannot see is time. And in medicine, time is often where the answer lives.”

What Happened Before He Left

Dr. Junaid stayed at his seat for a moment after the others had started gathering their things.

He was looking at his notes. The question he had arrived with (how common is non-compliance in my OPD?) was still there at the top of the page. But beneath it, in his own handwriting, was a new line he had added without Dr. Yaqoob noticing:

Snapshot. Not film. Association, not causation. Ask the right question first.

He looked up. “Sir, this is a doable study. I can run this in my own OPD. In three months.”

“Yes,” Dr. Yaqoob said. “You can.”

“And it will be publishable?”

“If your sampling is sound, your questionnaire validated, your ethics clearance in place, and your analysis honest, yes. Well-done frequency studies from Pakistani hospitals are useful, because local data on adherence is still limited. You would not be reinventing the wheel. You would be filling a gap in what we know about our own patients.”

Dr. Junaid nodded, closed his notebook, and picked up the cup of tea that had gone completely cold without him noticing.

“I’ll start with the questionnaire,” he said.

He walked out.

Fifteen years of clinical experience. Three months to a well-designed study.

That is what the right question, and the right design, can do.