Treat each cancer route as a measured control system: the disease has hidden state, tests are imperfect sensors, treatments are actuators, trials estimate what the actuator changes, and delivery determines how much of that result reaches the population.
Medical terms on this page26
- Resectable
- Judged removable by an operation. It describes what can be cut out, not whether invisible spread is absent.The visible instance can be deleted; hidden replicas may remain elsewhere.
- Metastatic
- Cancer that has formed new tumours away from its original site.A distributed system, not one local process.
- Systemic treatment
- Treatment that travels through the body, such as an injected or swallowed drug.A network-wide action rather than work on one physical node.
- Neoadjuvant
- Treatment given before the main operation.Work before the planned cutover.
- Adjuvant
- Treatment given after the main operation to reduce recurrence risk.Cleanup after removing the visible system.
- ctDNA
- Circulating tumour DNA: small fragments of DNA shed by tumour cells into blood.Sparse telemetry packets from a hidden process; absence can mean no process or a sensor below its detection limit.
- Minimal residual disease
- Evidence of tumour material remaining below ordinary scan detection after treatment.A process inferred from low-level telemetry after the visible service appears shut down.
- Sensitivity
- Among people who truly have the condition, the fraction the test detects.Recall: true alerts divided by all real faults.
- Specificity
- Among people who do not have the condition, the fraction the test correctly clears.True rejection rate for normal traffic.
- Positive predictive value
- Among positive test results, the fraction that are real. It changes with disease prevalence.Alert precision, controlled by both classifier performance and the base rate.
- Overall survival
- Time from a defined start until death from any cause.The final output; direct, but slower to observe.
- Progression-free survival
- Time until measured cancer growth or death.Time until a monitored failure flag; it can move without changing total runtime.
- Disease-free or event-free survival
- Time after a defined start without recurrence, progression, death or another stated event.A composite uptime measure; inspect exactly which events count.
- Response rate
- The fraction of patients whose measured tumour shrinks by a stated amount.A change in one monitored state, not proof of longer life or cure.
- Hazard ratio
- A comparison of event rates through follow-up. It is not the chance that one person benefits.A ratio between failure-rate curves, not a binary user-level outcome.
- Confidence interval
- A range showing the precision of an estimated effect under the study model.An error bar. Wide intervals permit several materially different realities.
- P value
- A compatibility calculation under a stated null model. It does not measure effect size or the probability that a claim is true.A test alarm threshold, not business value or posterior truth.
- Randomised trial
- People are assigned by chance to comparison groups to reduce systematic differences between them.A controlled A/B allocation; later loss or switching can still corrupt the estimate.
- Prospective study
- The study defines data collection and then follows people forward.Instrument the system first, then observe it.
- Retrospective study
- The study analyses records that already exist.Query production logs built for another purpose; missing data and selection are part of the problem.
- Intention to treat
- Analyse people in the groups they were originally assigned to, whether or not treatment was completed.Preserve the A/B assignment contract.
- Single-arm study
- Everyone receives the studied intervention; there is no concurrent comparison group.One production trace without a control cannot isolate the cause of change.
- Organoid
- A small laboratory-grown model made from a patient tumour sample.A sandbox instance: useful only if it builds in time and predicts the live system.
- Endpoint
- The outcome a study is designed to measure at a stated time.The output variable in the test contract; changing it changes the question.
- Median
- The middle observed value: half the group is above it and half below it.The 50th percentile, not an individual forecast and not the mean.
- Recurrence
- Cancer detected again after a period when it was not detectable.A failure returning after apparent clearance; it may be regrowth of an unseen earlier state.
Start with the unit of analysis. A cancer is not one object and a treatment is not one command. The useful model is a distributed system with changing clones, different organs, a supporting tissue environment and a patient whose reserve constrains every action.
A scan sees macroscopic structure. A blood marker samples a noisy output. A biopsy samples one place at one time.
Each sensor has a detection limit and a false-alarm rate. No sensor result is useful until it is connected to a decision that improves survival, function or suffering.
Read treatment results as a chain, not a headline. Biological efficacy asks whether the intervention can change the target. Eligibility asks how many patients have that target and are fit enough to receive it.
Route completion asks how many actually receive the intended test, referral, drug, operation and follow-up on time. Population effect is approximately efficacy × eligibility × completion. For example, 60% efficacy × 40% eligibility × 50% completion = 12% of the starting population affected.
This is why a strong drug result in a narrow, selected group can coexist with little population change.
Read endpoints as different outputs. Overall survival is time until death from any cause and is the hardest patient outcome. Progression-free survival is time until measured growth or death; it can improve without extending life.
Response rate is the fraction whose measured tumour shrinks by a stated rule; shrinkage can be brief and is not cure. Disease-free survival after local treatment is time without detected recurrence or death. Patient-reported outcomes measure how people feel or function.
The atlas does not exchange an earlier output for a later one without direct evidence that the substitution holds in that disease and setting.
Read study design like system identification. A prospective study declares what it will collect and follows forward. A retrospective study queries records already produced by care, so missing fields and selection can be part of the result.
Random assignment is an A/B allocation intended to balance known and unknown causes. It does not repair loss after assignment, treatment switching or selective reporting. Intention-to-treat analysis keeps people in their assigned group and protects the comparison created by randomization.
A single-arm study can show that something happened after treatment but usually cannot identify what would have happened without it.
Read numbers in absolute as well as relative units. A hazard ratio compares event rates through follow-up; it is not the probability that one person benefits. A 95% confidence interval describes the range of effects compatible with the data and model at that confidence rule; width is the precision signal.
A p value does not measure importance and does not give the probability that the claim is true. Median survival is the time by which half the group has had the event; it is not an individual forecast or the mean lifespan. Always ask for the denominator, the absolute difference, the follow-up time and who was missing.
Read diagnosis through base-rate maths. Sensitivity is the true-positive catch rate among people who have the condition. Specificity is the true-negative rejection rate among people who do not.
Positive predictive value asks what fraction of positive results are real, and it changes sharply with prevalence. With prevalence p, sensitivity s and specificity c: PPV = s×p / (s×p + (1-c)×(1-p)). At low prevalence, a small false-positive fraction can overwhelm the true cases.
A diagnostic claim is therefore incomplete without the intended population and the action caused by a positive result.
Read common clinical route words literally. Resectable means surgeons judge that visible disease can be removed; it does not mean systemic disease is absent. Neoadjuvant treatment occurs before an operation.
Adjuvant treatment occurs after it. Metastatic disease has established new tumours away from the original site. Minimal residual disease means tumour material is inferred below ordinary imaging, often through circulating tumour DNA, or ctDNA: small DNA fragments shed into blood.
An organoid is a small laboratory-grown model made from a patient's tumour. It is a sandbox instance, not the live system; build success, time, sampling and proof that its recommendation helps patients all matter.
Read the investment pages as staged capital allocation. A release gate states what evidence must exist before more money moves. A stop rule states what failure ends or changes the work.
A programme that cannot name its next decision, measurement and stopping condition is activity, not a decision system. Costs and probabilities in the atlas are planning judgements unless a source states them; clinical effects and trial counts remain tied to their source records.