Purpose
Treatment value is not one number. It is a structured judgement about benefit, harm, burden, cost, uncertainty, and patient goals.
Educational use only
This page explains common concepts used when discussing cancer treatment evidence. It is not a clinical recommendation engine and it is not a formal health-economic assessment.
Treatment decisions require original trial data, local guidelines, reimbursement rules, MDT interpretation, clinical judgement, and patient preference.
Key terms
These terms often appear together, but they answer different questions.
Overall survival, OS
Meaning: Time from a defined starting point, often randomisation, until death from any cause.
Example: Median OS is 18 months with the new treatment and 14 months with control. The median OS gain is 4 months.
Median OS does not show the full survival curve. It may miss delayed benefit, long-term survivors, cure fraction, or early harm.
Hazard ratio, HR
Meaning: A relative comparison of event rates over time between two groups.
Example: HR 0.70 means the event rate is estimated to be 30% lower in the treatment group during follow-up.
HR is not the same as 30% more patients surviving. It must be interpreted with the absolute survival difference and the shape of the survival curves.
Absolute benefit
Meaning: The actual difference in outcome between groups at a defined time point.
Example: At 2 years, 45% are alive with treatment and 35% with control. The absolute survival benefit is 10 percentage points.
Absolute benefit is usually easier to discuss with patients than relative benefit.
Number needed to treat, NNT
Meaning: Number of patients who need the treatment for one additional patient to benefit at a defined time point.
Example: If survival improves from 35% to 45%, the absolute benefit is 10 percentage points. NNT = 1 / 0.10 = 10.
Number needed to harm, NNH
Meaning: Number of patients treated for one additional patient to experience a harmful outcome.
Example: If severe toxicity rises from 10% to 20%, the absolute increase is 10 percentage points. NNH = 10.
Quality-adjusted life year, QALY
Meaning: A measure combining length of life and health-related quality of life.
Example: 1 year lived with utility 0.75 equals 0.75 QALYs.
Direct treatment cost
Meaning: Costs directly related to treatment delivery and consequences.
Examples: Drug cost, infusion visits, scans, laboratory tests, hospitalisation, toxicity management, and later healthcare use.
A low drug price does not always mean low total cost. A high drug price does not always mean poor value.
Evidence certainty
Meaning: How confident we are that the observed effect is real, relevant, and applicable.
Consider: Trial phase, randomisation, sample size, confidence intervals, follow-up maturity, crossover, missing data, endpoint choice, and external validity.
Certainty should modify interpretation. It should not be treated as just another arithmetic point.
Applicability
Meaning: Whether the trial result is relevant to the patient or population being discussed.
Consider: Performance status, age, comorbidity, molecular subgroup, line of therapy, previous treatments, geography, access, and patient goals.
A valid trial can still be poorly applicable to an individual patient.
Absolute and relative benefit are different
Relative effects can sound large even when the absolute benefit is small. Absolute effects usually make the clinical meaning clearer.
| Scenario | Control survival | Treatment survival | Absolute benefit | Simple interpretation |
|---|---|---|---|---|
| A | 40% | 50% | 10 percentage points | 10 more patients alive per 100 treated. |
| B | 4% | 5% | 1 percentage point | 1 more patient alive per 100 treated. |
Both scenarios may have a similar relative improvement, but their practical meaning is different.
Worked examples
The examples are simplified and fictional. They are designed to teach interpretation, not to recommend treatment.
Example 1: Moderate survival gain, acceptable harm
Trial result: Median OS improves from 14 to 18 months. HR 0.72. Two-year survival improves from 35% to 45%.
NNT: Absolute benefit is 10 percentage points. NNT = 10.
Harm: Severe toxicity increases from 12% to 17%. Absolute harm increase is 5 percentage points. NNH = 20.
Teaching interpretation: Benefit appears clinically relevant, but the discussion should include toxicity severity, patient priorities, and whether the trial population matches local patients.
Example 2: Good HR, small absolute benefit
Trial result: HR 0.65. One-year survival improves from 94% to 96%.
NNT: Absolute benefit is 2 percentage points. NNT = 50.
Harm: Grade 3 to 4 toxicity increases by 8 percentage points. NNH = 13.
Teaching interpretation: The relative effect looks favourable, but the absolute benefit is small. Toxicity and patient selection become central.
Example 3: Delayed benefit and uncertainty
Trial result: Median OS differs by only 1 month, but the survival curves separate after 18 months.
Issue: Median OS may underestimate benefit if a subgroup has durable response.
Evidence question: Are biomarkers, subgroup analyses, and follow-up mature enough to identify who benefits?
Teaching interpretation: A single median value is inadequate. Survival curves, landmark survival, subgroup validity, and long-term follow-up are needed.
How the concepts connect
A treatment can look valuable in one dimension and weak in another.
- OS asks: Do patients live longer?
- HR asks: How different are the event rates over follow-up?
- Absolute benefit asks: How many more patients benefit at a specific time point?
- NNT asks: How many patients must be treated for one additional benefit?
- NNH asks: How many patients must be treated for one additional harm?
- QALY asks: How much quality-adjusted survival is gained?
- Cost asks: What resources are required, and what else could those resources provide?
- Certainty asks: How much trust should we place in the estimate?
- Applicability asks: Does the evidence fit this patient or population?
Teaching score exercise
This optional exercise helps learners make assumptions visible. The score is a prompt for discussion, not a decision.
How to use the score
- 6 to 10: weak or uncertain value signal. Recheck assumptions and evidence quality.
- 11 to 15: mixed value signal. Discuss absolute benefit, toxicity, uncertainty, and patient goals.
- 16 to 21: stronger value signal. Interpret only if evidence certainty and applicability are acceptable.
Teaching score
The score is only a discussion aid. Review uncertainty, absolute effects, toxicity severity, patient preference, and local costs.
Interpretation checklist
Before calling a treatment high value or low value, make the assumptions explicit.
- What is the absolute survival benefit at a clinically meaningful time point?
- Does the hazard ratio match the visual impression of the survival curves?
- Are hazards proportional, or do curves cross or separate late?
- What is the NNT for the endpoint that matters most?
- What is the NNH for severe, persistent, irreversible, or fatal toxicity?
- How does treatment affect symptoms, function, and quality of life?
- Are QALY estimates based on transparent utility assumptions?
- Which costs are included, and which are excluded?
- Are subgroup findings prespecified, powered, and biologically plausible?
- Does the trial population resemble the patient or local population?
- How mature are the data, and how wide are the confidence intervals?
- What would change the conclusion?
Common interpretation traps
A single metric can be misleading when separated from context.
- A favourable HR can coexist with small absolute benefit.
- Median OS may miss long-term responders or delayed treatment effects.
- NNT and NNH change with time point, baseline risk, endpoint definition, and follow-up duration.
- Progression-free survival does not automatically imply better overall survival or quality of life.
- QALY estimates depend on utility values and model structure.
- Cost thresholds differ across countries and reimbursement systems.
- Exploratory subgroup effects are hypothesis-generating unless confirmed.
- Patient goals may outweigh a population-level value judgement.