NHPC BPH common mistakes
NHPC BPH · Understand the traps and how to avoid them.
When a question gives a 2x2 table and asks for the measure of association, candidates compute the risk ratio by habit even when the study design is case-control, producing a number above 1 that they then interpret as 'cases were twice as likely to be exposed'.
Textbook practice drills cohort-style risk ratios, and the 2x2 layout looks identical regardless of design. Case-control sampling fixes the number of cases and controls, so the incidence and therefore the risk ratio cannot be estimated from the table; only the exposure odds ratio is valid.
What to do instead
Before touching the numbers, write the design on your scratch paper: 'cohort/cross-sectional = risk ratio or prevalence ratio; case-control = odds ratio only'. Then state in one sentence what the denominator represents. If the stem says participants were selected by outcome status, the answer must be an odds ratio, and you should note that it approximates the risk ratio only when the outcome is rare.
Candidates answer screening and diagnostic-accuracy questions by reasoning from sensitivity and specificity alone, then declare a test 'good for ruling out disease' without checking the prevalence or the predictive values in the scenario.
Sensitivity and specificity feel like fixed quality labels, and candidates memorise 'high sensitivity rules out' as a slogan. They forget that positive and negative predictive values depend on prevalence, so the same test behaves very differently in a district hospital versus a community screening camp.
What to do instead
For every accuracy question, draw the 2x2 with a hypothetical 1,000 people and the stated prevalence, then compute PPV and NPV explicitly. Attach the rule to the correct cell: a highly sensitive test with a negative result is useful for exclusion (high NPV), a highly specific test with a positive result is useful for confirmation (high PPV). Recompute once with a different prevalence to see the shift.
Faced with a question about which measure of central tendency or dispersion to report, candidates pick the mean and standard deviation reflexively, even when the data are skewed, income-like, or described as heavily right-tailed with outliers.
Mean and SD dominate teaching examples and software defaults. Candidates also conflate 'which measure is most appropriate' with 'which measure is most familiar', and they do not connect skewness to the distortion of the mean by extreme values.
What to do instead
Train a two-question routine: (1) Is the variable categorical, ordinal, or continuous? (2) Is the distribution symmetric or skewed, and are there outliers? Skewed continuous data go to median and interquartile range, with a one-line justification about resistance to outliers. When a question asks for the best summary of a skewed distribution, eliminate any option containing mean or SD before comparing the rest.
In hypothesis-testing items, candidates read a p-value of 0.03 as 'a 97% chance the intervention worked' or as proof of a large effect, and they treat a non-significant result as proof of no difference.
The p-value is taught procedurally as a threshold to cross rather than as the probability of data this extreme under the null. Statistical significance also gets confused with clinical or programmatic importance, especially when the sample is very large.
What to do instead
Recite the definition before answering: the p-value is the probability of observing data at least as extreme as these, assuming the null hypothesis is true. It says nothing about the size of the effect or the probability the null is true. For any item with a p-value, also look for the confidence interval and effect size; a narrow CI far from the null supports a real effect, a wide CI crossing the null supports uncertainty rather than 'no effect'.
When asked to interpret a confidence interval, candidates say 'there is a 95% probability that the true value lies in this interval' or judge significance purely by whether the interval looks wide, ignoring whether it crosses the null value.
The frequentist meaning of coverage is counterintuitive, and candidates default to the intuitive but incorrect probabilistic reading. They also focus on width as a proxy for precision without checking where the interval sits relative to 1 or 0.
What to do instead
Use the operational rule: a 95% CI for a ratio that excludes 1, or for a difference that excludes 0, corresponds to a two-sided p below 0.05. State the correct interpretation as 'if we repeated this study many times, 95% of intervals constructed this way would contain the true value'. Then read the interval for both precision (width) and direction (position relative to the null).
Candidates classify a prevention scenario as primary, secondary, or tertiary prevention by the setting or the provider rather than by the stage of disease, so they label mammography screening as primary prevention because it happens in healthy women.
The words 'prevention' and 'screening' are used loosely in programme documents, and the three-level framework is memorised as a list without an anchor. The presence of apparently healthy people makes primary prevention feel correct.
What to do instead
Anchor each level to a single test question. Primary: does this stop disease before it starts (immunisation, safe water, tobacco taxation, condoms)? Secondary: does this detect existing but asymptomatic disease (screening, contact tracing, early case finding)? Tertiary: does this limit disability or restore function in established disease (rehabilitation, antiretroviral therapy, physiotherapy)? Apply the test to the verb in the stem, not the venue.
In outbreak-investigation items, candidates jump from a cluster of cases straight to a control measure, skipping the stepwise sequence, and they cannot distinguish the roles of descriptive epidemiology, hypothesis generation, and analytical testing.
Real outbreak stories are told as narratives that compress the steps, and candidates remember the dramatic interventions. The exam tests the order and purpose of each step, which is less memorable than the outcome.
What to do instead
Memorise the sequence as a checklist: verify the diagnosis and confirm the outbreak exists, define and count cases (person, place, time), generate hypotheses from interviews and descriptive curves, test hypotheses with a suitable analytical design, then implement control and communicate. When a question asks what to do next, choose the earliest uncompleted step rather than the most satisfying intervention.
Candidates confuse incidence and prevalence, using them interchangeably, and then mis-answer questions about duration, survival, or the effect of a treatment that prolongs life without curing.
Both are expressed as proportions or rates per population, and routine reporting often presents them side by side. The relationship prevalence approximately equals incidence times duration is stated once in teaching and rarely practised.
What to do instead
Define each in a phrase: incidence is new cases per population at risk per unit time; prevalence is all existing cases at a point or over a period. Apply the duration rule to every treatment question: a drug that extends survival without curing raises prevalence while leaving incidence unchanged. When a stem mentions 'burden in the community', suspect prevalence; when it mentions 'risk of developing', suspect incidence.
On health-system and financing questions, candidates blur the functions of financing, pooling, purchasing, and service delivery, and they mislabel out-of-pocket spending as a form of prepaid pooling or describe user fees as social health insurance.
The functions overlap in real systems and the vocabulary is borrowed from economics. Candidates also mix up the goals of universal health coverage (population coverage, service coverage, financial protection) with the mechanisms used to reach them.
What to do instead
Draw the four functions as separate boxes and place each example in one box only: collecting revenue is financing, spreading risk across people is pooling, allocating funds to providers is purchasing, and actually treating patients is delivery. For any payment mechanism, ask who bears the risk at the point of care; out-of-pocket payment shifts risk to the household and provides no pooling, which is why it is regressive and a barrier to financial protection.
In maternal and child health items, candidates apply a single global risk threshold or a developed-country referral rule without checking the local level of the health system, so they recommend caesarean section or specialist referral where the question expects community-level triage and referral to the next tier.
Clinical training emphasises definitive management, while public health exams test the organisation of care across levels. Candidates also treat risk as binary rather than graded by available resources and referral pathways.
What to do instead
For every maternal or newborn scenario, first identify the level of the system in the stem (community, primary facility, district hospital, referral hospital) and the skills and supplies available there. Match the action to that level: recognise danger signs and stabilise, then refer with communication and transport arranged. Reserve definitive obstetric or neonatal interventions for the level that can safely provide them.