19.2 Research Methodology & Biostatistics
Key Takeaways
- Research Methodology & Biostatistics is a named NCH subject (HomUG-RM-Stat) in the fourth professional year and is examinable AIAPGET content
- Evidence hierarchy runs systematic review and meta-analysis above RCT, then cohort, case-control, cross-sectional, and finally case series and case reports
- Type I error rejects a true null hypothesis (false positive, α); Type II error fails to reject a false null (false negative, β); power equals 1 − β
- Use the median rather than the mean when data are skewed or contain extreme outliers; in a normal distribution mean, median and mode coincide
- Indian clinical research must follow ICMR National Ethical Guidelines with Institutional Ethics Committee clearance and prospective CTRI registration
19.2 Research Methodology & Biostatistics
Quick Answer: Know the evidence hierarchy (meta-analysis > RCT > cohort > case-control > cross-sectional > case series > case report), the error pair (Type I = α = false positive; Type II = β = false negative; power = 1 − β), when the median beats the mean, the 68-95-99.7 rule, and the Indian ethics chain: ICMR guidelines → Institutional Ethics Committee → CTRI registration.
NCH gives this its own subject code (HomUG-RM-Stat) in the fourth professional year, which puts it squarely inside the AIAPGET syllabus reference. It is also the most directly useful chapter for anyone actually heading into MD Homoeopathy, because every dissertation begins with a protocol, a sample-size justification and an ethics submission.
Study Designs
| Family | Design | What it does |
|---|---|---|
| Descriptive observational | Case report | Single instructive patient |
| Case series | Several similar cases, no comparison group | |
| Cross-sectional | Exposure and outcome measured at one time point — gives prevalence | |
| Analytical observational | Case-control | Starts from outcome, looks backward for exposure; efficient for rare diseases; yields an odds ratio |
| Cohort | Starts from exposure, follows forward to outcome; gives incidence and relative risk | |
| Experimental | Randomised controlled trial | Investigator allocates the intervention at random; strongest single-study design for causation |
| Synthesis | Systematic review / meta-analysis | Pools results across studies under a pre-specified protocol |
Designs specific to homoeopathy
- The Homoeopathic Pathogenetic Trial (drug proving) is the discipline's own experimental design — administering a substance to healthy volunteers under protocol to record its symptom totality.
- Clinical trials split into individualised (each participant receives the similimum chosen for them — externally valid, harder to standardise) and non-individualised (a fixed remedy or complex for all participants — easier to blind, but arguably not testing homoeopathy as practised). Recognising this distinction is a favourite discussion-style MCQ.
- Randomisation, allocation concealment, blinding (single, double, triple) and placebo control are the bias-control tools; blinding is straightforward in homoeopathy because unmedicated globules are indistinguishable from medicated ones.
Hypothesis Testing
| Concept | Meaning |
|---|---|
| Null hypothesis (H₀) | No difference / no association |
| Alternative hypothesis (H₁) | A difference exists |
| Type I error (α) | Rejecting a true null — a false positive |
| Type II error (β) | Failing to reject a false null — a false negative |
| Power | 1 − β; conventionally targeted at 80% or more |
| p-value | Probability of the observed result (or more extreme) if H₀ were true; p < 0.05 conventionally "significant" |
| Confidence interval | Range of plausible values; a 95% CI for a difference that includes zero signals non-significance |
Trap: p < 0.05 does not mean the finding is clinically important, nor that the probability of H₀ is 5%. Statistical significance and clinical significance are separate judgements — a trivial effect can be significant in a huge sample.
Sampling
| Probability sampling | Non-probability sampling |
|---|---|
| Simple random | Convenience |
| Systematic (every kth unit) | Purposive / judgemental |
| Stratified (sample within strata) | Quota |
| Cluster and multistage | Snowball |
Only probability methods permit valid generalisation to the population. Stratified sampling is the answer when a stem stresses guaranteed representation of subgroups; cluster is the answer when whole intact groups such as villages or colleges are sampled.
Data and Scales
- Qualitative: nominal (blood group, religion — categories with no order) and ordinal (pain grade, socio-economic class — ordered but unequal intervals)
- Quantitative: discrete (number of episodes) and continuous (height, blood pressure)
- Measurement scales in ascending order of information: nominal → ordinal → interval → ratio. Only a ratio scale has a true zero, which is why temperature in Celsius is interval while weight in kilograms is ratio.
Descriptive Statistics
| Measure | Use it when |
|---|---|
| Mean | Data roughly symmetric, no extreme outliers |
| Median | Data skewed or containing outliers — the exam's standard right answer for income, hospital stay, parasite counts |
| Mode | Most frequent value; the only usable centre for nominal data |
| Range | Crude spread — highest minus lowest |
| Standard deviation | Standard measure of dispersion around the mean |
| Coefficient of variation | SD ÷ mean × 100 — compares variability across different units |
The normal distribution
Symmetric and bell-shaped, with mean = median = mode. Approximately 68% of observations fall within ±1 SD, 95% within ±2 SD, and 99.7% within ±3 SD. That 68-95-99.7 rule is directly examinable and also underlies the construction of reference ranges.
Choosing a test
| Question | Test |
|---|---|
| Compare two means | Student's t-test (paired if the same subjects measured twice) |
| Compare more than two means | ANOVA |
| Association between categorical variables | Chi-square |
| Strength and direction of a linear relationship | Correlation coefficient r, ranging −1 to +1 |
A correlation near zero means no linear relationship — not necessarily no relationship at all — and correlation never by itself establishes causation.
Diagnostic Test Statistics
| Measure | Definition | Plain meaning |
|---|---|---|
| Sensitivity | True positives ÷ all diseased | Ability to detect disease; a highly sensitive test is good for ruling out |
| Specificity | True negatives ÷ all non-diseased | Ability to exclude the healthy; a highly specific test is good for ruling in |
| Positive predictive value | True positives ÷ all test positives | Depends strongly on prevalence |
| Negative predictive value | True negatives ÷ all test negatives | Also prevalence-dependent |
Sensitivity and specificity are properties of the test; predictive values change with the population you apply it to. That prevalence dependence is the single most tested nuance here.
Research Ethics in India
- Informed consent — voluntary, documented, with the right to withdraw at any time without penalty to care
- Institutional Ethics Committee (IEC) approval before enrolment begins
- ICMR National Ethical Guidelines for Biomedical and Health Research Involving Human Participants as the governing national framework
- Clinical Trials Registry – India (CTRI) — prospective registration of clinical trials before the first participant is enrolled
- Declaration of Helsinki as the international reference for research on human subjects
- Confidentiality, data protection, and special protections for vulnerable groups including children, pregnant women and those unable to consent
Trap: CTRI registration is required prospectively. Registering after recruitment has begun does not satisfy the requirement, and retrospective registration is a standard MCQ distractor.
Compact Revision Card
- Hierarchy: meta-analysis > RCT > cohort > case-control > cross-sectional > case series > case report
- Case-control = backward, odds ratio, rare disease; cohort = forward, incidence, relative risk
- Type I = α = false positive; Type II = β = false negative; power = 1 − β
- Skewed data → median; normal distribution → 68/95/99.7
- t-test two means, ANOVA more than two, chi-square categorical, r from −1 to +1
- Sensitivity rules out; specificity rules in; predictive values follow prevalence
- Ethics chain: consent → IEC → ICMR guidelines → prospective CTRI registration
A researcher studying a rare form of hepatic carcinoma identifies affected patients and matched unaffected controls, then reviews their past exposures. Which design is this, and what measure of association does it yield?
A trial concludes that a remedy is effective when in reality it is not. Which error has occurred and which probability governs it?
Length of hospital stay in a ward is recorded as 2, 3, 3, 4, 4, 5, and 61 days. Which measure of central tendency best summarises this data set?
In a normal distribution, approximately what proportion of observations lies within two standard deviations of the mean?
Which statement about ethical requirements for a clinical trial conducted in India is correct?