A prescription, a relative reduction and the probability of benefit for one patient are three different figures. A study may be methodologically sound, yet its average transfers to a particular person only with regard to diagnosis, baseline risk, other illnesses, concurrent medicines and the follow-up period. A large relative difference may moreover correspond to a small absolute benefit.
What matters is to establish for whom we have evidence, which outcome the treatment changes, and how both benefit and possible harms will be monitored.
What we know: Randomised trials estimate an average difference between groups, not a certain outcome for an individual. For cardiovascular prevention NICE recommends discussing absolute risk and absolute benefits and harms, and starting a statin after an informed discussion.[5] For depression it distinguishes less and more severe states, offers several treatment options, and when antidepressants are stopped it recommends a gradual dose reduction agreed with the prescribing clinician.[22]
What we do not know: Without a diagnosis, examination, a specific product, dose, other medicines and health status, neither a personal benefit nor the cause of a new symptom can be determined. An improvement or a difficulty cannot at a distance be labelled placebo, nocebo or an adverse effect. This text is not a guide to starting, stopping or changing a dose.
Safety: Do not change prescribed treatment, or stop it abruptly, on the basis of an article; agree the next step with a doctor or pharmacist. Abrupt withdrawal of antidepressants can trigger even serious discontinuation symptoms.[22] In a severe or rapidly worsening condition, unconsciousness, breathing difficulty, chest pain or an immediate risk of self-harm, call 155 or 112.
1. A prescription that promises more than a study can carry
Imagine a morning ward round in a Czech hospital. The cardiologist wants a statin because a patient after a heart attack has a high absolute risk of a further event. The general practitioner hesitates with another person of the same age, an athlete with mildly raised cholesterol, because they do not want to treat a laboratory number. The clinical pharmacologist asks about other tablets, about the liver and about which enzyme even activates the medicine. The nurse reports that patients complain about their muscles and about the package leaflet. The quality manager wants it documented that care is “lege artis”. The patient, who is only half present in the room, wants to hear the sentence: this will protect you.
At first glance this is a dispute about whether “statins work”. In fact each person is describing a different object. The cardiologist measures population benefit at high risk. The GP weighs the benefit and burden of primary prevention. The pharmacologist assesses metabolism and interactions. The nurse catches a muscle problem whose cause cannot be determined without clinical assessment. The manager measures a process. The patient needs to understand the degree of uncertainty in plain language.
Each member of the round may therefore work with a different, legitimate figure, and a shared conclusion for the patient may still be missing. “Efficacy” is not one unchanging property of a tablet; it includes relative and absolute difference, baseline risk, possible harms, interactions and the clinical importance of the outcome being followed.
The first task is therefore to determine which treatment outcome is being assessed: a laboratory change, relief of a symptom, an averted event, or an adverse effect. These outcomes are not interchangeable.
A relative miracle may be an absolute drop. The patient does not sit in the study average. They sit in their own body.
— Jiný Kontext
2. What an effect estimate is made of
When the screen shows “the medicine reduces risk by 30 %”, it is easy to get the impression that the tablet switches the illness off in thirty cases out of a hundred. The result, however, arises in a chain. Each layer may enlarge the benefit, shrink it, or change the type of error. Two patients with the same prescription therefore need not receive the same treatment. The same medicine may behave differently according to baseline risk, the liver, concurrent tablets and whether we are even measuring the right endpoint.
Five layers of one effect
- Molecule and mechanism: what the medicine does in the body if it reaches its target.
- Population evidence: a randomised trial, the relative and absolute difference against the comparator.[1][7]
- The patient’s baseline risk: the same relative reduction gives a different person a different absolute benefit.
- The fate of the medicine in the body: absorption, enzymes, interactions, age, kidneys, adherence.
- Context and follow-up: placebo and nocebo, the natural course, monitoring, withdrawal, reporting of harm.[4][10]
The pivotal trial determines an important part of the certainty, but it is not the only input. Expectations can change how a symptom is experienced, yet a new difficulty cannot without examination be attributed to the leaflet rather than the medicine or another cause. The result is also affected by how the patient differs from the study population, by drug interactions, by dose and by the actual pattern of use.
This is where the first error of public debate arises. The headline announces a duel of “medicines save lives” and “medicines do not work”. In fact it compares two layers: a relative reduction in an ideal cohort and the absolute benefit in a person at low risk. Or, conversely, it adds relief from the visit, the retreat of a viral illness and the effect of the molecule into one “the medicine worked”. An honest comparison must state not only the name of the active substance, but also the population, the period, the endpoint, the absolute difference and what the study did not measure.
Distinguishing these influences determines the next clinical step. When a laboratory value does not change, the clinician addresses, among other things, use, dose, interactions and the biological response. When a difficulty appears, they must assess possible adverse effects as well as other causes; nocebo cannot safely be assumed from timing alone. And the occurrence of an event does not by itself prove that preventive treatment had no population effect.
3. The study average is not the body in the consulting room
Randomised controlled trials are an important tool for assessing the efficacy of medicines; the appropriateness of blinding and the further form of the evidence depend on the question and the type of intervention.[15] As early as 1996 Sackett and colleagues recalled that evidence-based medicine links the best available evidence with clinical experience and the patient’s values.[1] The average result of a group is important, but it is not enough for an individual decision.
The result of a trial is a difference between groups. The group with the medicine fared, on average, better than the group without it. Inside the average, however, lies heterogeneity that the consulting room does not see until it looks for it:
- people with a large benefit,
- people with a small benefit,
- people with no measurable response,
- people with harm and no benefit.
At the moment of signing the prescription the clinician does not know which of these four groups the particular person belongs to. They know only how groups that resemble the patient in a few baseline features fared — and that often do not resemble them in age, in the number of diagnoses and in the number of other boxes.
CONSORT and GRADE therefore require that evidence also be described by its limits: who was included, who was excluded, what the interval of uncertainty was, how serious the outcome was, how direct the analogy to the consulting room is.[13][14] That is not bureaucracy. It is an admission that statistical significance may be a precise answer to a question this patient did not ask.
For a decision the size of the average effect is not enough. One needs to know whether the effect can be monitored for the given indication, how quickly harm can be caught, and whether the aim is relief of a symptom, a laboratory marker, or prevention of a serious event. The choice therefore rests on the whole balance of benefits and risks, not on a single ranking on a poster.
4. Relative and absolute risk can read differently
Pharmaceutical communication and media headlines often speak in relative risk reduction. It sounds impressive. Gigerenzer and colleagues showed that patients, journalists and clinicians systematically overestimate this figure if they do not know the baseline risk and the absolute difference. Informed consent then takes place in a fog.[2]
A didactic example that is not a citation of one particular study, but the arithmetic of the same illusion:
- in the group without treatment, 2 people out of 100 have the event being followed (risk 2 %),
- in the group with the medicine, 1 person out of 100 (risk 1 %).
The report sounds: the medicine reduces risk by 50 %. The fact for the person in the waiting room sounds: the absolute reduction is one percentage point. Of a hundred people treated, the event being followed fails to occur in one extra person. Ninety-nine of the others would either not have had the event, or will have it even with the medicine. The relative miracle and the absolute drop are the same data. They differ in the frame.
Gigerenzer therefore asks for a different language: absolute risks, natural frequencies, a clear reference class. A “25 % reduction” in screening may mean that one fewer woman dies out of a thousand. Without a denominator the percentage is an advertising word.[2] Welch, Schwartz and Woloshin in their book on overdiagnosis show the same trap in early detection: higher “survival” need not mean a longer life if only the moment at which the illness is named is shifted.[9]
An honest figure therefore needs five items, without which “30 % less” is only a poster: which event, which population, which period, which baseline risk, which absolute difference. If a single one is missing, an impression remains.
5. What NNT says and what it does not
To get medicine out of the trap of relative figures it uses NNT (number needed to treat) and NNH (number needed to harm). NNT is the reciprocal of the absolute risk reduction: how many people need to be treated for a given period for one of them to gain the benefit being followed. NNH says the same sentence about harm.[2]
NNT is not a property of a molecule like a melting point. If baseline risk changes, NNT changes, even if the relative effect remains similar. A person after a heart attack and a healthy person with a mildly off laboratory number therefore cannot share “the NNT of statins” as a single score. They can share the same active substance and receive an entirely different account of benefit and harm.
An ideal medicine would have NNT = 1 and an NNH so large that harm vanishes in practice. Real medicine moves in a compromise. In therapies where baseline risk is high, the absolute benefit is usually larger and NNT lower. In prevention in people at low risk, NNT is usually higher: many are treated so that one gains the endpoint being followed. That is not evidence that prevention “does not work”. It is evidence that the account should be read in absolute figures.
NNT inherits the limits of the source study: the follow-up period, the selection of the population, the definition of the outcome and the uncertainty of the estimate. Without these items and without knowledge of the patient one cannot derive from it whether treatment is “worth it”. In the consulting room it can help formulate two questions: how many comparable people need to be treated for a given period for one additional benefit, and how often the harm being followed appears.
The Cochrane review of statins in primary prevention (Taylor et al., 18 studies, 56 934 participants) reports for all-cause mortality OR 0.86 and for composite cardiovascular events RR 0.75.[3] From these relative measures one cannot, without baseline risk and follow-up time, determine a personal absolute benefit or a universal NNT. Converting OR 0.86 by itself into the slogan “14 % fewer deaths” would moreover mistake an odds ratio for a simple relative risk.
6. Placebo, nocebo and the natural course of illness
Why does a person often feel better after a visit and after starting a medicine, even when the absolute benefit of the molecule is not large? Three processes add up in the consulting room and are rarely separated.
The first is the natural course and regression to the mean. People go to the doctor when a symptom is at its worst. Many acute states recede on their own. A tablet that arrives on the right day takes credit that belongs to time.
The second is placebo. It is not “nothing”. Benedetti describes a measurable neurobiology of expectation, ritual and relationship: endogenous opioids, dopamine, conditioned responses.[10] The context of treatment can change the experience of pain and anxiety. It can also confuse the evaluation of a medicine if a study does not separate the context.
The third is nocebo: negative expectation can contribute to a real experience of a symptom even without a specific pharmacological mechanism. The same headache, nausea or muscle problem can, however, be caused by the medicine, the illness, an interaction or another state. From a symptom or from its timing one therefore cannot determine the cause at a distance. Suspicion of an adverse effect belongs to clinical assessment and can be reported to SÚKL; a report is a safety signal, not automatic proof of causality.[7]
A pleasant improvement can conceal the same old problem
Well-timed relief does not by itself prove what share of the change was caused by the specific effect of the medicine, the natural course or the context of care. In acute and fluctuating states it is therefore important to decide in advance what will be assessed and when. That does not, however, justify attributing a persistent or serious symptom to placebo or nocebo without examination.
Separating these layers is not cynicism. It is a condition for not betting on a tablet where time will suffice — and conversely for not betting on time where the absolute risk does not allow waiting.
7. Pharmacogenomics: the same tablet, a different liver
A registration trial gives an average dose to an average body. Livers are not average.
Clopidogrel is a documented example, not a slogan. It is a prodrug: it acts against platelets only after activation by cytochrome P450 enzymes, especially CYP2C19. Mega et al. in the New England Journal of Medicine showed that carriers of at least one reduced-function CYP2C19 allele — in their healthy cohort roughly 30 % — had lower exposure to the active metabolite and weaker platelet inhibition. In a cohort of patients with acute coronary syndrome these carriers had a higher occurrence of cardiovascular events (12.1 % versus 8.0 %) and a higher risk of stent thrombosis.[6]
That is not evidence that “genetics refutes EBM”. It is evidence that the study average hides a subgroup for whom the same tablet name is a different chemistry. The share of homozygous slow metabolisers in the European population is lower than the share of all allele carriers; clinically, however, it is enough that the clinician does not know whether they have a carrier in front of them until a test is done or until the effect fails.
The same logic applies to other CYP substrates. There is no need to invent that every other antihypertensive has data as hard as clopidogrel. It is enough to know that an identical prescription is not an identical fate of the molecule. Where a validated test exists and where failure costs a stent, genetics is a layer of effect. Where there is no test and the medicine has a wide therapeutic window, genetics is speculation. To confuse the two is to read one study as a law about all livers.
8. Multimorbidity: a trial without the person in the waiting room
Clinical trials use inclusion and exclusion criteria, so people with several illnesses and extensive medication may be represented less than in an ordinary consulting room. For a particular patient, therefore, interactions, kidney and liver function, fall risk, frailty and other treatment decide alongside the trial result. One cannot, however, claim without checking that none of their combinations has ever been studied.
ÚZIS records consumption of medicines by anatomical-therapeutic groups.[11] The consumption curve says that chronic pharmacotherapy is a mass phenomenon. It does not say whether every combination in the home medicine cabinet was ever tested. SÚKL collects reports of suspected adverse effects.[7] A report is a signal, not automatic proof of causality. Even so it is one of the few maps that sees the person outside the protocol.
Welch’s book on overdiagnosis belongs here as a warning about another layer: the more one treats risks rather than illnesses, the more people enter treatment whose absolute benefit is small and whose harm is real.[9] It is not a call to stop medicines after a heart attack. It is a call not to apply the logic of a heart attack to a laboratory deviation in an otherwise fit person.
The NNT of one treatment does not by itself capture the overall burden of several products. That must be assessed in a regular medication review, including the indication of each medicine, interactions, duplications and symptoms that may be related to treatment.
The Medicines Act and pharmacovigilance rules govern, among other things, the handling of medicines and the reporting of suspected adverse effects.[7][17] The summary of product characteristics (SPC) describes indications, dosing, contraindications, interactions and known risks of a particular product.[21] In a person with several medicines, however, the combination and health state as a whole must also be assessed.
9. How treatment results can miss the goal
The choice of treatment does not rest only on the probability of average benefit. The seriousness of possible harm, the possibility of timely review and the consequences of ineffective or interrupted treatment also decide. Short-term failure of cough relief, occult bleeding and arbitrary interruption of secondary prevention after a heart attack therefore require a different safety procedure.
| Type of failure | What it looks like | How to test it | What limits the harm |
|---|---|---|---|
| Frame illusion | A relative 50 % conceals an absolute 1 point | Ask for frequencies: how many out of a hundred | ARR, NNT, the same period |
| Context illusion | Relief from time and placebo is credited to the molecule | Compare with the natural course | Time, placebo control, a real target |
| Body illusion | The study average misses the liver, kidneys, interactions | Medication history, possibly genotype | Start-stop, review, an estimate of NNH |
| Endpoint illusion | A number is treated, not the event that matters | Ask which event is meant to fall | The endpoint being followed, not only a laboratory shift |
An error in dose can be acutely dangerous and requires immediate verification with a health professional; its consequences cannot in advance be labelled reversible. The dispatched impression “I am already protected, I can go on smoking” is a different, long-term type of risk. It therefore makes sense to separate a medicine that changes a laboratory marker from a medicine that, in a given group, changed heart attacks and deaths. The first can be a useful tool. It is not automatically the second.
10. Statins: the same sentence, two different accounts
Statins show well the difference between relative and absolute benefit. The Cochrane review of primary prevention found in the included studies a fall in total mortality and in cardiovascular events.[3] Current NICE NG238, for people without known cardiovascular disease, works with an estimate of ten-year risk and requires an informed discussion of absolute benefit and harms, preferences, comorbidities, polypharmacy, frailty and life expectancy.[5] This British threshold is not an automatic Czech prescribing rule, but the methodological principle of communicating risk is transferable.
It does not follow that “statins are pointless”. It follows that secondary prevention after a previous event and primary prevention in a person without known disease are not one situation. As baseline risk rises, absolute benefit usually rises at the same relative effect. A personal NNT or NNH cannot, however, be determined from age, one LDL figure and a general mention of muscle problems.
The Cholesterol Treatment Trialists’ Collaboration, in a meta-analysis of individual data from 28 randomised trials, found a reduction in major vascular events with statin treatment across age groups, with less direct evidence for primary prevention in people older than 75 years.[16] Even this result does not make NNT a universal figure. NICE and European recommendations link the decision to overall risk, the indication and an informed choice, not to a single laboratory column.[5][18]
A precise NNT cannot be calculated from one sentence about cholesterol. It can be calculated only when these are known: the event, the time, the baseline risk, the absolute difference. Anyone who does not have these four figures does not have an NNT. They have an impression.
A laboratory fall in LDL is a useful intermediate. It is not the end for which a person takes the medicine. The end is the event they want to avoid. If the consulting room discusses only a laboratory number, it is discussing the dashboard, not the road.
11. Antidepressants: the average depends on the population and the scale
Kirsch et al. in PLOS Medicine processed data submitted to the US FDA for the registration of four newer-generation antidepressants, including then unpublished studies. The difference against placebo grew with baseline severity. In milder and moderate states it was small; it reached the conventional threshold of clinical importance only at the upper end of very severe depression. By meta-regression the authors attributed this more to a falling placebo response in the most severe states than to a rising strength of the molecule.[4]
That is not evidence that antidepressants “do not work”. Kirsch’s work shows that the average difference against placebo in this set depended on baseline severity and that a change in score is not the same as a certain clinical outcome for a particular person. The studies did not measure whether treatment delayed hospitalisation, and such a conclusion cannot be drawn from them.
Cipriani et al., in a network meta-analysis of 21 antidepressants, found that all the medicines examined were in the acute treatment of adults with major depressive disorder more effective than placebo, while they differed in efficacy and acceptability.[20] The results cannot simply be set against Kirsch as a vote: the works have different sets of studies, methods and outcomes followed.
NNT cannot honestly be lifted from Kirsch’s work as a single figure for all depressions. The studies measured a severity score, not the share of people cured in a Czech outpatient clinic. Anyone who makes from it a poster “antidepressants are placebo” reads as dishonestly as anyone who turns a relative difference in severe depression into an obligation of a tablet for every sadness.
Current NICE, for a new episode of less severe depression, does not recommend an antidepressant routinely as first choice unless it is the informed preference of the patient; for more severe depression it offers several psychological, pharmacological and combined options.[22] The choice depends on clinical assessment, risk, previous response, comorbidities and preferences. The medicine should not be stopped arbitrarily: the dose is usually reduced gradually with the prescribing specialist.[22]
12. Antibiotics where the benefit disappears
In a serious bacterial infection, timely antibiotics can be decisive. Against viruses, however, antibiotics have no effect, and in a particular respiratory infection the clinician must assess the diagnosis, severity and the risk of complications. Possible adverse effects and the population risk of resistance remain even when a bacterial benefit is not confirmed.
The 2017 Cochrane review of acute bronchitis or acute productive cough in people without underlying lung disease found limited clinical benefit and more adverse effects; the authors also noted little data in frail older people with several illnesses.[19] This result is not a blanket ban on antibiotics. It shows why it is necessary to decide according to the probable cause, the clinical state and the balance of benefits and risks.
What decides is the diagnosis and the sensitivity of the pathogen, not a general sentence that antibiotics “work” or “do not work”. A fall in fever during treatment does not by itself prove that the cause was a sensitive bacterium; the clinical course must be assessed by a specialist.
NNT makes sense only for a precisely defined population, treatment, outcome and period. For “an antibiotic against a virus” such a figure is not a clinical argument, because a corresponding mechanism of action is missing.
13. Shared decision-making is not a loss of authority
Elwyn et al. described a practical model of shared decision-making in three steps: announce that a choice exists; describe the options; help the person say what benefit and harm mean for them.[8] The ethical code of the Czech Medical Chamber places informed consent and the patient’s autonomy at the core of the profession.[12] Neither text portrays the clinician as a statistician with a calculator. It portrays someone who can bear to say the figure.
What needs to be said before a prescription is written
The conversation should, according to the situation, cover baseline risk, the expected absolute benefit, common and serious harms, reasonable alternatives and a review plan. In preventive treatment one cannot directly know, in an individual, whether a particular event was averted; one can follow, for example, risk factors, the laboratory response, use and safety. Any change or end of treatment is determined by a professional plan, not by a patient’s solo experiment.
The sentence “it will cut the risk by half” without a denominator and a time is not enough for an informed choice. The patient should receive understandable absolute figures if they are available for a comparable population, and space to state their priorities. The clinician at the same time explains the seriousness of the indication and the risks of refusing or interrupting treatment.
Authority is not lost by this. What is lost is only the authority that stood on a certainty the data did not have.
14. Primary and secondary prevention are not the same situation
A person with already proven cardiovascular disease and a person without a known event, in whom future risk is only being estimated, do not enter the same decision. NICE therefore uses, in people without proven disease, an estimate of ten-year risk, while in secondary prevention it does not require a risk calculation for the decision to treat.[5]
In neither situation can a personal NNT or NNH be manufactured from a general sentence about statins. For an absolute estimate, data on a comparable population, a specific event and a follow-up period are needed. In an individual, moreover, we do not observe the counterfactual result: if the event does not occur, we do not know whether the medicine prevented it, or whether it would not have occurred without it either.
A practical decision should therefore rest on the indication, overall risk, comorbidities, other medicines, possible interactions, expected absolute benefit and the patient’s preferences. New muscle problems or other symptoms should not automatically be labelled nocebo or a reason for unprompted withdrawal; they belong to assessment by the prescribing clinician.[5]
15. What to choose for which treatment
A reasonable choice does not start with the name of the molecule. It starts with the kind of work the medicine is meant to do, and with the review the consulting room has available.
| Need | First question | Decisive test | Unacceptable shortcut |
|---|---|---|---|
| Secondary prevention | Which event, and over what horizon, does treatment reduce? | Clinical guidance, absolute benefit and safety reviews | A decision from a single laboratory number alone |
| Primary prevention | What is the baseline absolute risk? | An informed discussion of benefits, harms and preferences | A relative percentage without a denominator |
| Treatment of depression | What is the severity, the risk and the patient’s preference? | Clinical assessment, recommended options and a review plan | An automatic tablet or an automatic refusal of medicine |
| Infection | Bacterium, or virus? | Mechanism, not fever itself | An antibiotic “just in case” |
| Many products | Is each medicine still indicated and safe in combination? | Medication history and a professional medication review | Adding or stopping a product on one’s own |
| Suspicion of genetics | Is the medicine a prodrug with a narrow window? | A validated test where failure costs an event | A gene poster for every antihypertensive |
For Czech use a separate language test is necessary. A “30 % risk reduction” in a press release is not the same as 30 out of 100 people. A decimal point in an English study is not a decimal comma in a Czech laboratory. “Statistically significant” is not “significant for this life”. A test set should contain: the denominator, the period, the endpoint and the sentence “the source is not enough”, if a leaflet offers only a relative frame.
Some people need a combination of pharmacotherapy, lifestyle measures, psychological support or a specialist adjustment of other treatment. Individual interventions should have their own indication and review plan; no prescribed product should be changed without agreement with a specialist.
16. A review plan instead of a universal test
There is no safe fourteen-day test for all medicines. The time to onset of effect, the appropriate review and the manner of any discontinuation differ by product and indication. In prevention, moreover, an averted event cannot be observed directly in an individual: the fact that nothing happened in a short period neither confirms nor refutes a long-term benefit.
A safe plan arises together with the prescribing specialist and contains:
- Indication and aim: which symptom, laboratory marker or clinical event the treatment is meant to influence.
- Baseline state: important values, other medicines, allergies, organ function and risk factors relevant to the given product.
- Time horizon: when a change can be expected and when a review should take place according to guidance and the SPC. For example NICE, for statins, states a check of lipids and liver transaminases at 2 to 3 months after starting or changing treatment, while for antidepressants it works with other intervals and regular reviews.[5][22]
- Safety signals: which problems to raise at a routine review, which to report without delay and for which to call emergency help.
- Further decisions: who will adjust the dose, change the treatment or plan gradual withdrawal, and on what basis. The patient should not change a prescribed medicine on their own or stop it abruptly.
- A record of uncertainty: what is documented for a similar group of patients and what we do not yet know in the particular person.
Three questions for the follow-up review
Can the benefit be expressed in absolute terms for a comparable population and the right time horizon? Does the treatment have a measurable interim target, or is it long-term prevention whose personal counterfactual result we will not see? And does the patient know which symptoms to tell a doctor or pharmacist, and that a suspected adverse effect can also be reported to SÚKL?[7]
The review should not prove that a symptom is “only nocebo”, nor experiment with withdrawal. It should compare expected benefit, the actual course, new difficulties and current preferences, and then adjust the plan safely.
17. Uncertainty must be reviewed as one goes
Treatment can affect a symptom, a laboratory value or the risk of a future event. None of these results by itself means that overall health risk is “dealt with”. Pharmacotherapy does not replace reviews or other recommended parts of care.
The way figures are presented affects the decision. A relative reduction without baseline risk can overstate the benefit; the experience of one patient cannot in turn replace a comparison of groups. Population evidence therefore needs to be joined to the clinical state and preferences of the particular person.
A randomised trial is a basic tool for estimating the causal effect of treatment in the study population.[1][13] For an informed decision one needs to add which event is being followed, what the baseline risk is, the absolute difference, possible harms and a review plan.
ÚZIS publishes statistics on medicine consumption in Czechia and SÚKL runs the system for reporting suspected adverse effects.[7][11] These systems remind us that knowledge of safety continues after registration. In long-term treatment a regular professional review of indication, effect, risks and the patient’s preferences therefore matters.
Suitable treatment is not determined by the highest relative percentage. It should match the patient’s indication and risk, have an acceptable balance of benefits and harms and a clear monitoring plan. Some harms, moreover, are not reversible, so wording about safety must not create the opposite impression.
Different criteria apply to short-term relief, different ones to the treatment of depression and different ones to the long-term prevention of a heart attack. In every case, however, it must be clear what the treatment changes, how quickly that can be assessed and how possible harms are handled.
Let us ask: For whom do we have evidence, how large is the expected absolute benefit, what remains uncertain, and how will the treatment be safely reviewed?
— Jiný Kontext