Community Medicine — Epidemiology and Screening, NMC MBBS licence examination syllabus (Nepal Medical Council).
Epidemiology: the arithmetic behind every clinical decision
The same test result means different things in different patients. That is not a paradox — it is arithmetic.
Community medicine is often revised last and least, on the assumption that it is a collection of definitions to be memorised the night before. It is the opposite. The reasoning in this chapter is what tells you whether a positive test in front of you means anything, whether a study you have read supports what it claims, and whether a screening programme helps the people it enrols.
Nearly all of it comes out of a single four-cell table. Learn to build that table and read it in two directions, and the definitions stop needing to be memorised — they can be derived.
🩺 Where this lives: A test with 99% sensitivity and 99% specificity sounds close to perfect. Apply it to a population where the disease affects 1 in 10,000, and among those who test positive, the great majority will not have the disease — because the tiny false-positive rate acts on a very large number of healthy people, while the true positives come from a very small number of sick ones. That is why screening the general population and testing a symptomatic patient are different activities with different answers, and why "the test was positive" is never by itself a diagnosis.
The table everything comes from
The direction you read the table is the whole distinction. Reading down the columns starts from the truth and asks how the test performed — that gives sensitivity and specificity, which belong to the test. Reading across the rows starts from the test result and asks what it means for this patient — that gives predictive values, which depend on how common the disease is in the population you are testing.
A WORKED EXAMPLE — why prevalence matters
A test with 99% sensitivity and 99% specificity.
IN A HIGH-PREVALENCE SETTING (say 50% of those tested
actually have the disease), out of 1000 people:
500 diseased → 495 true positives
500 healthy → 5 false positives
PPV = 495 / 500 = 99%. The test is trustworthy.
IN A LOW-PREVALENCE SETTING (say 1 in 1000), out of
100,000 people:
100 diseased → 99 true positives
99,900 healthy → 999 false positives
PPV = 99 / 1098 ≈ 9%. The SAME test, and now most
positives are wrong.
Nothing about the test changed. Only the population did.
This is why a screening test needs a confirmatory test
behind it, and why applying a test outside the population
it was validated in produces nonsense.
💡 Exam angle: SnNout and SpPin are asked constantly, and both are derivable rather than memorisable. A highly Sensitive test has few false negatives, so a Negative rules the disease out. A highly Specific test has few false positives, so a Positive rules it in. That is why screening uses sensitive tests — missing a case is worse than a false alarm you can later disprove — and confirmation uses specific ones.
Measuring disease in a population
INCIDENCE and PREVALENCE
INCIDENCE NEW cases in a period ÷ population at risk
A measure of RISK and of how fast disease is
appearing. The right measure for causation
and for outbreak work.
PREVALENCE ALL existing cases at a point ÷ population
A measure of BURDEN — how much disease is
there to be managed.
PREVALENCE ≈ INCIDENCE × DURATION
That relationship explains a result that confuses people:
a treatment that keeps patients alive without curing them
INCREASES prevalence while leaving incidence unchanged.
Rising prevalence can mean better survival, not a worsening
epidemic — and the two require completely different
responses.
MEASURES OF ASSOCIATION
RELATIVE RISK risk in exposed ÷ risk in unexposed
Requires a cohort design.
ODDS RATIO from case-control studies; approximates
relative risk when the disease is rare.
ATTRIBUTABLE RISK risk in exposed − risk in unexposed
How much of the risk is due to the
exposure — the number that matters for
public health action.
NUMBER NEEDED TO TREAT
1 ÷ absolute risk reduction. The most
clinically honest way to express a
benefit, because it uses ABSOLUTE rather
than relative change.
Relative and absolute risk are the commonest way statistics mislead. "Reduces risk by 50%" sounds decisive — but if the risk falls from 2 in 10,000 to 1 in 10,000, that is a relative reduction of 50% and an absolute reduction of 0.01%, with a number needed to treat of 10,000. Both figures are true. Only one of them helps a patient decide.
Study designs
WHAT GOES WRONG IN STUDIES
SELECTION BIAS the people studied differ
systematically from the population you want to describe.
Includes healthy worker effect, volunteer bias, loss to
follow-up.
INFORMATION BIAS the measurement itself is wrong.
RECALL bias — people with a disease remember exposures
more thoroughly. A major weakness of case-control
studies.
OBSERVER bias — the assessor knows the group.
Addressed by BLINDING.
CONFOUNDING a third factor is associated with both
exposure and outcome and explains the apparent link.
Classic example: an association between coffee and lung
cancer that is really smoking.
Handled by randomisation (best), restriction, matching,
stratification, or statistical adjustment.
CHANCE addressed by adequate sample size,
confidence intervals and p-values.
A CONFIDENCE INTERVAL is more informative than a p-value
alone: it shows both whether an effect is likely real and
how large it might plausibly be. An interval for a ratio
that crosses 1 is compatible with no effect.
Screening
💡 Exam angle: lead time bias is the one most often asked and least often understood. Diagnosing a disease earlier automatically lengthens the time between diagnosis and death, even if the date of death is completely unchanged. So improved five-year survival after introducing screening is not evidence that screening works — the only convincing endpoint is a fall in mortality, which is why screening programmes are judged on that rather than on survival statistics.
Prevention
🔍 The levels, and where each acts
PrimordialPreventing the risk factor from arising at all — policy on tobacco, air quality, urban design, food environment.
PrimaryPreventing disease in healthy people who have the risk factor — immunisation, smoking cessation, clean water, sanitation, safe cooking fuel.
SecondaryDetecting disease early, in its latent stage, to alter its course — this is where screening sits.
TertiaryLimiting disability and complications in established disease — rehabilitation, diabetic foot care, secondary prevention after myocardial infarction.
The prevention paradoxA measure that brings large benefit to a population may offer little to each individual — and vice versa. Shifting the whole population's blood pressure down slightly prevents more strokes than treating only the highest-risk few, even though no individual notices much. It also explains why population-level advice can feel pointless to the person receiving it.
HERD IMMUNITY — why it matters and where it fails
When enough of a population is immune, transmission
chains break and even the non-immune are protected.
The threshold depends on how transmissible the organism
is: highly transmissible diseases such as measles need
very high coverage, while less transmissible ones need
less.
IT PROTECTS THOSE WHO CANNOT BE VACCINATED — infants
below the eligible age, the immunosuppressed, those with
genuine contraindications.
IT FAILS LOCALLY BEFORE IT FAILS NATIONALLY. National
coverage figures conceal pockets of low uptake, and
outbreaks begin in those pockets. So coverage is
monitored geographically, not only in aggregate.
It does not apply to diseases without person-to-person
spread — vaccinating against tetanus protects the
individual, not the community.
Clinical reasoning: four presentations
🔍 Case 1 — the positive screening test
PresentationA well 40-year-old with no symptoms or risk factors has a positive result on a screening test with 99% sensitivity and 95% specificity for a condition affecting about 1 in 2000 people. He is distressed and asks whether he has the disease.
Key clueVery low prevalence in the population being tested.
ReasoningWith 5% false positives applied to a large healthy population and only 1 in 2000 truly affected, the great majority of positive results in this setting will be false. The positive predictive value is low despite excellent test characteristics.
AnswerExplain that a positive screening result is not a diagnosis, and arrange a confirmatory test with high specificity. The honest answer to "do I have it?" at this stage is "probably not, and here is how we find out".
🔍 Case 2 — the impressive statistic
PresentationA drug representative reports that a new agent "reduces the risk of stroke by 40%". The underlying trial shows the event rate falling from 0.5% to 0.3% over five years.
TrapA relative reduction presented without its absolute counterpart.
ReasoningThe absolute risk reduction is 0.2 percentage points, giving a number needed to treat of 500 over five years. The 40% figure is accurate and almost useless for a patient deciding whether to take a tablet daily.
AnswerAsk for the absolute risk reduction and the NNT, and weigh them against cost, adverse effects and the patient's own baseline risk. Relative figures without absolute ones should always prompt the question.
🔍 Case 3 — the survival that improved
PresentationAfter a new screening programme, five-year survival for a cancer rises from 40% to 70%. The programme is declared a success. The number of deaths from that cancer each year has not changed.
Key clueSurvival improved; mortality did not.
ReasoningLead time bias — diagnosing earlier lengthens measured survival even if death occurs at exactly the same time. Length time bias compounds it, because screening preferentially detects slower-growing disease with a better prognosis regardless of treatment.
AnswerJudge the programme on mortality, ideally from randomised evidence. Unchanged mortality with improved survival is the signature of these biases, not of benefit.
🔍 Case 4 — the rising prevalence
PresentationDistrict data show the prevalence of a chronic disease has doubled over ten years. Incidence of new diagnoses is unchanged. A report concludes that the disease is spreading rapidly.
TrapReading prevalence as a measure of how fast disease is appearing.
ReasoningPrevalence ≈ incidence × duration. With incidence unchanged, rising prevalence means patients are living longer with the disease — which usually reflects better treatment or earlier diagnosis rather than more transmission.
AnswerInterpret this as an increased burden of care rather than a spreading epidemic. The service implication is chronic disease management capacity, not outbreak control.
Commonly confused
Confusion
The distinction
Why it matters
Sensitivity vs PPV
Columns versus rows of the same table
One belongs to the test, the other to the population.
Incidence vs prevalence
New cases versus all existing cases
Prevalence rises when survival improves.
Relative vs absolute risk
A large relative change can be a tiny absolute one
Only the absolute figure helps a patient decide.
Case-control vs cohort
Backward from outcome versus forward from exposure
Determines whether you get an odds ratio or a relative risk.
Association vs causation
Confounding explains many associations
Only randomisation balances unknown confounders.
Survival vs mortality in screening
Lead time inflates survival without saving lives
Mortality is the only convincing endpoint.
Good test vs good programme
Screening needs treatment and follow-up to exist
Screening without treatment harms without benefit.
Rapid revision
MUST-KNOW FACTS
1. Build the 2×2 table; every test statistic comes from it.
2. Sensitivity = a/(a+c); specificity = d/(b+d) — read DOWN the columns.
3. PPV = a/(a+b); NPV = d/(c+d) — read ACROSS the rows.
4. Sensitivity and specificity are properties of the TEST.
5. Predictive values depend on PREVALENCE.
6. The same test performs differently in different populations.
7. SnNout: a sensitive test, when NEGATIVE, rules OUT.
8. SpPin: a specific test, when POSITIVE, rules IN.
9. Screen with a sensitive test; confirm with a specific one.
10. Moving the cut-off trades sensitivity against specificity.
11. Incidence = new cases; prevalence = all existing cases.
12. Prevalence ≈ incidence × duration.
13. Better survival RAISES prevalence without raising incidence.
14. Relative risk needs a cohort; odds ratio comes from case-control.
15. Absolute risk reduction and NNT are the clinically honest measures.
16. Cross-sectional: prevalence, no temporal sequence.
17. Case-control: backward from outcome; recall bias.
18. Cohort: forward from exposure; gives incidence.
19. Randomised trial: balances UNKNOWN confounders — the causal design.
20. Confounding: a third factor linked to both exposure and outcome.
21. Blinding addresses observer bias.
22. A confidence interval crossing 1 is compatible with no effect.
23. Screening needs a latent stage AND an accepted treatment.
24. LEAD TIME bias inflates survival without delaying death.
25. LENGTH TIME bias favours slow-growing disease.
26. Judge screening on MORTALITY, not survival.
27. Prevention: primordial, primary, secondary (screening), tertiary.
28. Herd immunity protects those who cannot be vaccinated.
29. Herd immunity fails in local pockets before national figures show it.
💡 Exam angle: the reliable threads are (a) predictive value changing with prevalence, (b) SnNout and SpPin, (c) relative versus absolute risk, (d) lead time bias in screening, and (e) prevalence rising because survival improved. Every one of them is a number that appears to say something obvious and says something else — the same habit of mind this stream trains clinically, applied to populations.
Syllabus points
The 2×2 table and reading it in two directions
Sensitivity, specificity and predictive values
Why prevalence changes what a positive result means
SnNout, SpPin and choosing a test
Incidence, prevalence and their relationship
Relative risk, odds ratio, absolute risk and NNT
Study designs and what each can claim
Bias, confounding and how they are addressed
Screening criteria
Lead time, length time and selection bias
Levels of prevention and the prevention paradox
Herd immunity and why it fails locally first
Create a free account to tick topics off, take notes as you read, watch the video lessons and get a day-by-day study plan built around your exam date.