The most important properties of a drug are: Efficacy, Quality & Safety, Tolerability, Cost-effectiveness. ^[1]

Think of it this way: efficacy estimates whether and how much a drug can work in a specified setting; safety characterises harms and uncertainty (no drug is guaranteed harmless); tolerability affects whether a patient can continue it; and cost-effectiveness compares incremental health benefit with incremental cost. These are separate dimensions, so there may be no single “winner” for every patient.

What the opening images are doing: the Olympic motto and finish-line photograph frame “best” as faster/higher/stronger, while the consumer comparison image asks “best by which criteria?” A drug can lead on one outcome yet lose on adverse effects, convenience, or value. The comparison must therefore specify the patient, comparator, outcome, and time horizon ^[1].

New conditions, new pathogens, new understanding of disease mechanisms (e.g. new receptor, enzyme or pathway identified), new mutations. ^[1]

More choices, to differ from generics, resistance, more specificity. ^[1]

Better efficacy, better safety, better tolerability, better compliance. ^[1]

The lecture groups reasons into three categories:

Assessment of a method of treatment in a planned experiment involving patients. ^[1]

This distinguishes a clinical trial from: (a) laboratory studies (not involving patients), (b) anecdotes or case reports (not planned experiments), (c) clinical experience (not controlled).

Traditional recipe used by generations, anecdotes, clinical experience of a lifetime, expert opinion — "the man in the white coat recommends." ^[1]

Before the era of evidence-based medicine, treatments were based on tradition, authority, and uncontrolled observation. The problem: humans are subject to confirmation bias, placebo effects, regression to the mean, and confounding. Clinical trials exist to overcome these biases.

Phase I: 20-80 people, evaluate safety, determine safe dosage range, identify side effects. ^[1] Phase II: 100-300 people, evaluate effectiveness and further safety. ^[1] Phase III: 1,000-3,000 people, confirm effectiveness, monitor side effects, compare to standard treatment. ^[1] Phase IV: post-marketing studies, additional information on risks, benefits, and optimal use. ^[1]

The phase labels and sample sizes are typical, not rigid rules: objectives can overlap, adaptive designs may span phases, and early oncology studies often enrol affected patients rather than healthy volunteers ^[11].

Before Phase I, laboratory and pre-clinical studies inform an Investigational New Drug (IND) submission describing the proposed human studies. In the US, an IND is not a marketing approval: FDA has 30 days to review the original submission and may place the study on clinical hold; otherwise the proposed clinical investigation can proceed ^[11].

The attrition rates across phases are striking:

These percentages come from the historical GAO/Biomedtracker pipeline shown on the slide and are illustrative, not timeless constants; success varies by therapeutic area, modality, period, and denominator. In that displayed dataset, the largest attrition occurred at Phase II → III ^[1].

Subject characteristics are important. Defined by inclusion criteria and exclusion criteria. Affect generalisability of the conclusions. ^[1]

This is a crucial concept. Eligibility criteria may under-represent older adults, children, pregnant people, multimorbidity, or polypharmacy, although this varies by trial. This means:

• Internal validity depends on randomisation, concealment, blinding where possible, follow-up, outcome measurement, and analysis—not merely the label “trial.”
• External validity / generalisability asks whether the participants, intervention delivery, comparator, outcomes, and setting match the patient and decision at hand.

Usually calculated by a statistician, who needs to know: the P value required (usually < 0.05), the anticipated difference between treatments relative to the SD (effect size), drop out rate. Usually at least a power of 80% to detect a difference between treatments at the 5% level of significance is expected. ^[1]

Power = 1 − β (where β = probability of Type 2 error). Power of 80% means there is an 80% chance of detecting a true difference if one exists.

Why this matters: An underpowered study may miss a clinically important effect (Type 2 error). In the slide's paired plots, the estimated female and male mean heights are similar in both datasets, but N=500 has much narrower 95% confidence intervals than N=16. Sample size improves precision; it does not remove systematic bias.

Range, median, Mean, SD, SE, 95%CI, Null hypothesis, P value, T-tests, chi-square test, Correlation coefficient, Odds ratio, Relative risk, relative risk reduction, Absolute risk reduction, NNT. ^[1]

Mean ± 1.96 × SE → 95% CI ^[1] SE = SD / √N ^[1]

Correct interpretation: under repeated sampling with the same method, 95% of intervals constructed this way would contain the true parameter. For the single interval already observed, frequentist analysis does not assign a 95% probability to a fixed parameter. The SE shrinks as N increases, so larger studies usually give more precise estimates. For a conventional two-sided 5% test, an effect-difference CI containing 0 or a ratio CI containing 1 is compatible with no effect.

Null hypothesis: any difference observed is due to chance. ^[1] P value: under a specified statistical model, and assuming the null hypothesis and model assumptions are true, the probability of a test statistic at least as extreme as the observed value. ^[13] Type 1 error: null hypothesis wrongly rejected when it is true (false positive). ^[1] Type 2 error: null hypothesis not rejected despite being false (false negative). ^[1]

Continuous data → T-test (means) or Mann-Whitney (rank order). Categorical data → Chi-square (observed vs expected frequencies). Time-to-event ("survival data") → Log-rank (observed vs expected events). ^[1]

If outcome depends on several factors (multivariate): Continuous → Multiple regression. Categorical → Logistic regression. Time-to-event → Cox proportional hazard. ^[1]

Why multivariable methods? Outcomes may depend on age, sex, comorbidities, and baseline severity. Regression can estimate associations conditional on included covariates, but it does not automatically eliminate residual confounding or prove causation. The correct test also depends on distributional assumptions, pairing/repeated measures, censoring, and the pre-specified analysis plan.

Case report (−), Case-control study (+), Cohort study (++), RCT (+++), Meta-analysis (++++), N=1 trial (+++++). ^[1]

The retrospective/prospective labels are common patterns, not definitions: cohort studies may be retrospective, and systematic reviews/meta-analyses are secondary syntheses rather than simply “retrospective trials.” Evidence quality depends on design and execution, and the most informative design depends on whether the question is about benefit, harm, prognosis, diagnosis, or an individual treatment response.

There may be a number of trials addressing the same research question. The results may vary. There is a need to summarise the results. ^[1]

Meta-analysis: a statistical way of combining compatible effect estimates from different studies. Weight is usually driven by statistical precision (often inverse variance), not size alone ^[16].

The lecture shows a forest plot of statin trials for major coronary events ^[1]. Key features of reading a forest plot:

Reading the statin plot shown: the effect ratio is to the left of 1 when treatment is favoured. LIPS has a CI crossing 1 and is individually inconclusive, while the pooled diamond is entirely left of 1 at 0.73 (95% CI 0.70–0.77). The diamond is a relative pooled effect, not an absolute benefit; baseline risk and follow-up are still needed for ARR/NNT. A precise pooled estimate can still mislead if studies are biased, clinically heterogeneous, selectively reported, or inappropriately combined ^{[1] [16]}.

Definition: any untoward event (does not need to be related to treatment). All adverse events must be documented. Management (no action, observation, treatment, hospitalisation, surgery). Outcome (fully or partially recovered, persistent, death). ^[1]

Key distinction: An adverse EVENT is anything bad that happens during the trial. An adverse DRUG REACTION is an adverse event judged to be causally related to the drug. All events must be documented regardless of suspected causation.

Voluntary in HK to Drug Office, Dept of Health. Voluntary in Mainland China, but there are targets. Yellow card system in the UK. In USA, reports from professionals, consumers, and manufacturers are submitted to the FDA Adverse Event Reporting System. ^[1]

• Look for RCT evidence (Phase III trials)
• Check effect size (ARR, NNT), not just P-values
• Is the endpoint clinically meaningful (hard endpoints like mortality/MI, not just surrogate markers)?

• Review Phase III trial ADR data
• Check Phase IV / post-marketing data for rare ADRs
• Consider the specific patient's risk factors for ADRs

• Side effect profile relevant to THIS patient
• Dosing convenience (once daily > multiple daily; oral > injection for most)
• Drug interactions with current medications

• Compare NNT with drug cost
• Generic alternatives available?
• Does incremental benefit justify incremental cost?

• Was the patient's demographic represented in the trials?
• Pharmacogenomic considerations?
• Comorbidities that alter benefit-risk balance?

GC 029 focuses on the practical prescribing process (correct drug, dose, route, duration, avoiding errors). GC 070 is the evidence evaluation that precedes prescribing — you must first determine the BEST drug before worrying about prescribing it correctly.

The Beers Criteria and STOPP/START criteria ^[5] operationalise the concept from GC 070 that subgroups may have different benefit-risk profiles. Drugs that are appropriate for younger adults may be potentially inappropriate in older adults due to altered pharmacokinetics, increased susceptibility to ADRs, and polypharmacy interactions.

GC 035 explains the pharmacokinetic parameters (absorption, distribution, metabolism, excretion) that influence both efficacy and safety. Understanding CYP450 enzymes (especially 3A4, 2D6, 2C9) ^[2] helps explain why some patients respond differently — directly relevant to the personalised medicine theme of GC 070.

The statin meta-analysis shown in the forest plot ^[1] is a practical example of how meta-analyses guide prescribing decisions. Similarly, the evidence base for antiplatelet therapy in ACS comes from large RCTs with NNT calculations.