MFPHMI Part I common mistakes
MFPHMI Part I · Understand the traps and how to avoid them.
Choosing the measure of association by habit (always reporting an odds ratio) rather than by study design and outcome type, so a cohort study with a common outcome is summarised with an OR that overstates the risk ratio.
Candidates memorise 'OR = case-control, RR = cohort' as a slogan and stop at that, missing that the OR is only an approximation of the RR when the outcome is rare, and that risk difference, rate ratio, or attributable measures may be the required answer.
What to do instead
Before answering, state the design, the outcome type, and the comparison being made, then select the measure whose definition matches that structure. Practise converting between OR, RR, and risk difference on the same 2x2 table so the numeric consequences are visible.
Confusing incidence with prevalence in a calculation or interpretation, for example using prevalent cases as the denominator for a risk, or describing a prevalence difference as a measure of association.
Both are proportions expressed as a number of cases, and question stems often supply only one of the two counts, so the candidate reaches for whichever number is present rather than checking what the denominator represents.
What to do instead
Write the denominator explicitly in words (persons at risk at the start of follow-up versus all persons examined at a point in time) before doing arithmetic. Check that any stated time frame is compatible with an incidence measure.
Interpreting a confidence interval as a probability that the true value lies in the interval, or as a statement about the distribution of individual observations, instead of a statement about the long-run coverage of the estimation procedure.
The frequentist definition is counterintuitive and the intuitive misreading is reinforced by everyday language about '95% chance', so candidates default to the intuitive phrasing under time pressure.
What to do instead
Memorise one precise sentence about repeated sampling and coverage, and rehearse rejecting the three classic wrong phrasings (probability the parameter is in the interval, 95% of the sample lies in the interval, 95% chance of reproducing the result).
Reading a p-value as the probability that the null hypothesis is true, or treating a non-significant result as evidence that there is no effect, rather than as absence of evidence against the null at the chosen threshold.
The p-value is taught alongside hypothesis-testing language ('accept' or 'reject') that encourages a binary verdict, and the asymmetry of the logic is easy to forget when the stem asks for a clinical conclusion.
What to do instead
Restate every p-value answer as 'the probability of data at least this extreme if the null were true'. When a result is non-significant, write the confidence interval and comment on whether the study was powered to exclude a clinically important effect.
Mixing up type I and type II error, or power and significance level, when asked what happens to a study's error rates after changing sample size, effect size, or the alpha threshold.
The four quantities move together in a way that is easy to memorise as a table but hard to reason about, and candidates recall only 'power goes up with n' without tracking which error probability is affected and in which direction.
What to do instead
Reconstruct the relationships from first principles each time: alpha fixes the type I error, power is 1 minus type II error, and increasing n, effect size, or alpha increases power. Draw the two sampling distributions and shift them to check the direction.
Answering a health protection scenario with generic clinical management (treat the patient, advise rest, follow up in clinic) rather than the specific public health actions: notification, isolation or exclusion periods, contact tracing, post-exposure prophylaxis, and environmental or food-chain investigation.
Candidates with a clinical background default to the individual patient in front of them, and the statutory and operational steps are the least familiar part of the syllabus, so they are the first to be dropped under time pressure.
What to do instead
For every communicable disease scenario, run a fixed checklist: is it notifiable, what is the exclusion or isolation period, who are the contacts, what prophylaxis or vaccination is indicated, and what source investigation is needed. Learn the legal notification duty and the typical incubation and infectious periods for the high-frequency organisms.
Confusing outbreak investigation sequence and purpose, for example jumping to analytical studies or control measures before verifying the diagnosis and establishing a case definition, or failing to describe a hypothesis-generating step.
The steps are usually memorised as a list without their rationale, so the candidate cannot tell which step a scenario is testing and picks the most clinically satisfying action rather than the next correct one.
What to do instead
Learn the steps with the question each one answers (verify diagnosis, confirm outbreak, define cases, describe by person place time, generate hypotheses, test hypotheses, control, communicate). For each practice scenario, name the step the stem is currently at before choosing an option.
Treating screening as automatically beneficial, so a question about a screening programme is answered by citing earlier detection, without addressing lead-time bias, length bias, overdiagnosis, or the need for evidence from randomised trials of outcome rather than survival from diagnosis.
Screening feels intuitively good, and the biases that inflate apparent survival are conceptually slippery, so candidates answer from values rather than from the evaluative criteria.
What to do instead
Memorise the classic biases with a one-line mechanism each (lead time, length, overdiagnosis, selection) and the Wilson and Jungner style criteria for judging whether a programme is justified. Anchor every screening answer on evidence of reduced mortality in a trial, not on survival after diagnosis.
Confusing the main categories of epidemiological study design when asked about strengths and weaknesses, for example claiming a cross-sectional survey can establish temporality or that a case-control study yields incidence directly.
Designs are learned as definitions rather than as structures with a specific sampling scheme and a specific vulnerability to bias, so the candidate cannot derive which inferences each design supports.
What to do instead
For each design, write the sampling frame, the timing of exposure and outcome measurement, the measure obtained, and the principal bias. Derive the strengths and weaknesses from that structure instead of memorising lists.
Misreading the direction of a comparison of need, uptake, or outcome across groups, for example attributing a difference to the intervention when it reflects the inverse care law or differences in case mix, or reporting relative differences when the question asks about absolute inequality.
Health improvement questions often give several plausible numbers and the candidate picks the largest or most striking rather than the one that answers the specific question about inequality, equity, or access.
What to do instead
Before answering, identify the comparison group, the metric (relative versus absolute), and whether the stem asks about need, access, uptake, or outcome. Convert between relative and absolute measures on the same data to see how differently they rank groups.