Number needed to treat

Last updated

The number needed to treat (NNT) or number needed to treat for an additional beneficial outcome (NNTB) is an epidemiological measure used in communicating the effectiveness of a health-care intervention, typically a treatment with medication. The NNT is the average number of patients who need to be treated to prevent one additional bad outcome. It is defined as the inverse of the absolute risk reduction, and computed as , where is the incidence in the control (unexposed) group, and is the incidence in the treated (exposed) group. [1] [2] This calculation implicitly assumes monotonicity, that is, no individual can be harmed by treatment. The modern approach, based on counterfactual conditionals, relaxes this assumption and yields bounds on NNT.

Contents

A type of effect size, the NNT was described in 1988 by McMaster University's Laupacis, Sackett and Roberts. [3] While theoretically, the ideal NNT is 1, where everyone improves with treatment and no one improves with control, in practice, NNT is always rounded up to the nearest round number [4] and so even a NNT of 1.1 becomes a NNT of 2 [5] . A higher NNT indicates that treatment is less effective. [6]

NNT is similar to number needed to harm (NNH), where NNT usually refers to a therapeutic intervention and NNH to a detrimental effect or risk factor. A combined measure, the number needed to treat for an additional beneficial or harmful outcome (NNTB/H), is also used.

Relevance

The NNT is an important measure in pharmacoeconomics. If a clinical endpoint is devastating enough (e.g. death, heart attack), drugs with a high NNT may still be indicated in particular situations. If the endpoint is minor, health insurers may decline to reimburse drugs with a high NNT. NNT is significant to consider when comparing possible side effects of a medication against its benefits. For medications with a high NNT, even a small incidence of adverse effects may outweigh the benefits. Even though NNT is an important measure in a clinical trial, it is infrequently included in medical journal articles reporting the results of clinical trials. [7] There are several important problems with the NNT, involving bias and lack of reliable confidence intervals, as well as difficulties in excluding the possibility of no difference between two treatments or groups. [8]

NNT may vary substantially over time, [9] [10] and hence convey different information as a function of the specific time-point of its calculation. Snapinn and Jiang [11] showed examples where the information conveyed by the NNT may be incomplete or even contradictory compared to the traditional statistics of interest in survival analysis. A comprehensive research on adjustment of the NNT for explanatory variables and accommodation to time-dependent outcomes was conducted by Bender and Blettner, [12] Austin, [13] and Vancak et al. [14]

Explanation of NNT in practice

There are a number of factors that can affect the meaning of the NNT depending on the situation. The treatment may be a drug in the form of a pill or injection, a surgical procedure, or many other possibilities. The following examples demonstrate how NNT is determined and what it means. In this example, it is important to understand that every participant has the condition being treated, so there are only "diseased" patients who received the treatment or did not. This is typically a type of study that would occur only if both the control and the tested treatment carried significant risks of serious harm, or if the treatment was unethical for a healthy participant (for example, chemotherapy drugs or a new method of appendectomy - surgical removal of the appendix). Most drug trials test both the control and the treatment on both healthy and "diseased" participants. Or, if the treatment's purpose is to prevent a condition that is fairly common (an anticoagulant to prevent heart attack for example), a prospective study may be used. A study which starts with all healthy participants is termed a prospective study, and is in contrast to a retrospective study, in which some participants already have the condition in question. Prospective studies produce much higher quality evidence, but are much more difficult and time-consuming to perform.[ citation needed ]

In the table below:

DescriptionNNTInterpretation
Perfect treatment, previously untreatable condition with no placebo effect involved0.01.01Half of participants receive the treatment, and half receive a control (which may be simply a placebo, or may be an existing treatment with a known effectiveness). Every person that receives the treatment shows improvement, which may be a reduction or halt in worsening of the condition, an improvement in the condition, or an outright cure of the condition. Every person in the control group shows no improvement, therefore the condition never improves on its own and the control is never effective. NNT is 1/(1.0-0.0), which is 1.
Very effective treatment with large improvement over control0.10.91.25For simplicity, a low number of participants will be used, though scientific studies almost always require many more. Ten people receive the treatment, and ten receive a control. Of the ten in the treated group, nine show improvement, and one shows no improvement. In the control group, one person shows improvement and nine show none. Since one of those who received the control showed improvement without the treatment, it is said that one of the nine from the treated group would have improved without receiving the treatment. Therefore, one person’s outcome does not represent evidence that the treatment is better than the control. NNT is 1/(0.9-0.1), which is 1.25. The absolute risk reduction is 0.9-0.1, equal to 0.8.
Effective treatment with moderate improvement over control0.30.72.5Ten receive the treatment, and ten receive a control. In the treatment group, seven show improvement and three show none. In the control group, three show improvement and seven show none. Therefore, the treatment was more helpful than the control in four of ten cases (7 treated improved minus 3 controls improved), and was not any more helpful in six of ten cases (3 not improved despite treatment, 3 that would have improved anyway as seen in the control group). NNT is 1/(0.7 – 0.3), which is 2.5.
Effective treatment, but little improvement over control0.40.510Ten receive the treatment, and ten receive a control. In the treatment group six improve with the treatment, and four do not. In the control group, five improve and five do not. Therefore, the treatment was more helpful than the control in only one of ten cases (6 treated improved minus 5 controls improved), and was not helpful is nine of ten (4 not improved despite treatment, 5 that would have improved anyway as seen in the control group). NNT is 1/(0.5 – 0.4), which is 10.
Not very effective treatment with little improvement over control0.80.910Ten receive the treatment, and ten receive a control. Two improve with the treatment and eight do not. In the control group, one improves and nine do not. Therefore, the treatment was more helpful than the control in only one of ten cases, and was not helpful is nine of ten. NNT is 1/(0.9 – 0.8), which is 10.
Apparently very effective treatment, but with little real improvement over control0.10.210Ten receive the treatment, and ten receive a control. Nine improve with the treatment and one does not. In the control group, eight improve and two do not. Therefore, the treatment was more helpful than the control in only one of ten cases, and was not helpful is nine of ten. NNT is 1/(0.2 – 0.1), which is 10.
Treatment is very effective but worse than control0.20.1−10Ten receive the treatment, and ten receive a control. Eight improve with the treatment and two do not. In the control group, nine improve and one does not. Therefore, the treatment was less helpful than the control in one of ten cases. NNT is 1/(0.1 – 0.2), which is -10. Notice that, even though the treatment was effective in eight of ten cases (only one less than the previous example) the NNT has shifted from 10 to -10. This is because NNT measures how many patients must be given the treatment instead of the control in order to see improvement in one person. Since giving the treatment to ten people would cause one of those people to be worse than if they had received the control instead, the NNT is -10. If control is placebo, giving placebo appear better than to give treatment.

Real-life example

ASCOT-LLA manufacturer-sponsored study addressed the benefit of atorvastatin 10 mg (a cholesterol-lowering drug) in patients with hypertension (high blood pressure) but no previous cardiovascular disease (primary prevention). The trial ran for 3.3 years, and during this period the relative risk of a "primary event" (heart attack) was reduced by 36% (relative risk reduction, RRR). The absolute risk reduction (ARR), however, was much smaller, because the study group did not have a very high rate of cardiovascular events over the study period: 2.67% in the control group, compared to 1.65% in the treatment group. [15] Taking atorvastatin for 3.3 years, therefore, would lead to an ARR of only 1.02% (2.67% minus 1.65%). The number needed to treat to prevent one cardiovascular event would then be 98.04 for 3.3 years. [16]

Numerical example

Example of risk reduction
QuantityExperimental group (E) Control group (C)Total
Events (E)EE = 15CE = 100115
Non-events (N)EN = 135CN = 150285
Total subjects (S)ES = EE + EN = 150CS = CE + CN = 250400
Event rate (ER) EER = EE / ES = 0.1, or 10% CER = CE / CS = 0.4, or 40%—
Variable Abbr. FormulaValue
Absolute risk reduction ARRCER−EER0.3, or 30%
Number needed to treat NNT1 / (CER−EER)3.33
Relative risk (risk ratio)RREER / CER0.25
Relative risk reduction RRR(CER−EER) / CER, or 1 −RR0.75, or 75%
Preventable fraction among the unexposed PFu(CER−EER) / CER0.75
Odds ratio OR(EE / EN) / (CE / CN)0.167

Modern Approach to NNT

The above calculations for NNT are valid under monotonicity, where treatment can't have a negative effect on any individual. However, in the case where the treatment may benefit some individuals and harm others, the NNT as defined above cannot be estimated from a Randomized Controlled Trial (RCT) alone. The inverse of the absolute risk reduction only provides an upper bound, i.e., .

The modern approach defines NNT literally, as the number of patients one needs to treat (on the average) before saving one. However, since "saving" is a counterfactual notion (a patient must recover if treated and not recover if not treated) the logic of counterfactuals must be invoked to estimate this quantity from experimental or observational studies. [17] The probability of "saving" is captured by the Probability of Necessity and Sufficiency (PNS), where . [18] Once PNS is estimated, NNT is given as . However, due to the counterfactual nature of PNS, only bounds can be computed from an RCT, rather than a precise estimate. Tian and Pearl have derived tight bounds on PNS, based on multiple data sources, and Pearl showed that a combination of observational and experimental data may sometimes make the bounds collapse to a point estimate. [19] [20] Mueller and Pearl provide a conceptual interpretation for this phenomenon and illustrate its impact on both individual and policy-makers decisions. [21]

See also

Related Research Articles

<span class="mw-page-title-main">Clinical trial</span> Phase of clinical research in medicine

Clinical trials are prospective biomedical or behavioral research studies on human participants designed to answer specific questions about biomedical or behavioral interventions, including new treatments and known interventions that warrant further study and comparison. Clinical trials generate data on dosage, safety and efficacy. They are conducted only after they have received health authority/ethics committee approval in the country where approval of the therapy is sought. These authorities are responsible for vetting the risk/benefit ratio of the trial—their approval does not mean the therapy is 'safe' or effective, only that the trial may be conducted.

<span class="mw-page-title-main">Scandinavian Simvastatin Survival Study</span>

The Scandinavian Simvastatin Survival Study, was a multicentre, randomized, double-blind, placebo-controlled clinical trial, which provided the initial data that supported the use of the cholesterol-lowering drug, simvastatin, in people with a moderately raised cholesterol and coronary heart disease (CHD); that is people who had previously had a heart attack or angina. The study was sponsored by the pharmaceutical company Merck and enrolled 4,444 people from 94 centres in Scandinavia.

In survival analysis, the hazard ratio (HR) is the ratio of the hazard rates corresponding to the conditions characterised by two distinct levels of a treatment variable of interest. For example, in a clinical study of a drug, the treated population may die at twice the rate of the control population. The hazard ratio would be 2, indicating a higher hazard of death from the treatment.

<span class="mw-page-title-main">Duloxetine</span> Antidepressant medication used also for treatment of anxiety and chronic pain

Duloxetine, sold under the brand name Cymbalta among others, is a medication used to treat major depressive disorder, generalized anxiety disorder, obsessive-compulsive disorder, fibromyalgia, neuropathic pain and central sensitization. It is taken by mouth.

<span class="mw-page-title-main">Fluvastatin</span> Chemical compound

Fluvastatin is a member of the statin drug class, used to treat hypercholesterolemia and to prevent cardiovascular disease.

Postherpetic neuralgia (PHN) is neuropathic pain that occurs due to damage to a peripheral nerve caused by the reactivation of the varicella zoster virus. PHN is defined as pain in a dermatomal distribution that lasts for at least 90 days after an outbreak of herpes zoster. Several types of pain may occur with PHN including continuous burning pain, episodes of severe shooting or electric-like pain, and a heightened sensitivity to gentle touch which would not otherwise cause pain or to painful stimuli. Abnormal sensations and itching may also occur.

In the design of experiments, hypotheses are applied to experimental units in a treatment group. In comparative experiments, members of a control group receive a standard treatment, a placebo, or no treatment at all. There may be more than one treatment group, more than one control group, or both.

<span class="mw-page-title-main">Relative risk</span> Measure of association used in epidemiology

The relative risk (RR) or risk ratio is the ratio of the probability of an outcome in an exposed group to the probability of an outcome in an unexposed group. Together with risk difference and odds ratio, relative risk measures the association between the exposure and the outcome.

<span class="mw-page-title-main">Number needed to harm</span> Measure in epidemiology

In medicine, the number needed to harm (NNH) is an epidemiological measure that indicates how many persons on average need to be exposed to a risk factor over a specific period to cause harm in an average of one person who would not otherwise have been harmed. It is defined as the inverse of the absolute risk increase, and computed as , where is the incidence in the treated (exposed) group, and is the incidence in the control (unexposed) group. Intuitively, the lower the number needed to harm, the worse the risk factor, with 1 meaning that every exposed person is harmed.

<span class="mw-page-title-main">Cladribine</span> Pharmaceutical drug

Cladribine, sold under the brand name Leustatin, among others, is a medication used to treat hairy cell leukemia and B-cell chronic lymphocytic leukemia. Cladribine, sold under the brand name Mavenclad, is used for the treatment of adults with highly active forms of relapsing-remitting multiple sclerosis.

<span class="mw-page-title-main">Confounding</span> Variable or factor in causal inference

In causal inference, a confounder is a variable that influences both the dependent variable and independent variable, causing a spurious association. Confounding is a causal concept, and as such, cannot be described in terms of correlations or associations. The existence of confounders is an important quantitative explanation why correlation does not imply causation. Some notations are explicitly designed to identify the existence, possible existence, or non-existence of confounders in causal relationships between elements of a system.

<span class="mw-page-title-main">Dronedarone</span> Medication

Dronedarone, sold under the brand name Multaq, is a class III antiarrhythmic medication developed by Sanofi-Aventis. It was approved by the US Food and Drug Administration (FDA) in July 2009. Besides being indicated in arrhythmias, it was recommended as an alternative to amiodarone for the treatment of atrial fibrillation and atrial flutter in people whose hearts have either returned to normal rhythm or who undergo drug therapy or electric shock treatment i.e. direct current cardioversion (DCCV) to maintain normal rhythm. It is a class III antiarrhythmic drug. The FDA label includes a claim for reducing hospitalization, but not for reducing mortality, as a reduction in mortality was not demonstrated in the clinical development program. A trial of the drug in heart failure was stopped as an interim analysis showed a possible increase in heart failure deaths, in people with moderate to severe congestive heart failure.

<span class="mw-page-title-main">Risk difference</span>

The risk difference (RD), excess risk, or attributable risk is the difference between the risk of an outcome in the exposed group and the unexposed group. It is computed as , where is the incidence in the exposed group, and is the incidence in the unexposed group. If the risk of an outcome is increased by the exposure, the term absolute risk increase (ARI) is used, and computed as . Equivalently, if the risk of an outcome is decreased by the exposure, the term absolute risk reduction (ARR) is used, and computed as .

Medical statistics deals with applications of statistics to medicine and the health sciences, including epidemiology, public health, forensic medicine, and clinical research. Medical statistics has been a recognized branch of statistics in the United Kingdom for more than 40 years, but the term has not come into general use in North America, where the wider term 'biostatistics' is more commonly used. However, "biostatistics" more commonly connotes all applications of statistics to biology. Medical statistics is a subdiscipline of statistics.

It is the science of summarizing, collecting, presenting and interpreting data in medical practice, and using them to estimate the magnitude of associations and test hypotheses. It has a central role in medical investigations. It not only provides a way of organizing information on a wider and more formal basis than relying on the exchange of anecdotes and personal experience, but also takes into account the intrinsic variation inherent in most biological processes.

<span class="mw-page-title-main">Esketamine</span> Medication

Esketamine, sold under the brand names Spravato and Ketanest among others, is the S(+) enantiomer of ketamine. It is a dissociative hallucinogen drug used as a general anesthetic and as an antidepressant for treatment of depression. Esketamine is the active enantiomer of ketamine in terms of NMDA receptor antagonism and is more potent than racemic ketamine.

Population impact measures (PIMs) are biostatistical measures of risk and benefit used in epidemiological and public health research. They are used to describe the impact of health risks and benefits in a population, to inform health policy.

<span class="mw-page-title-main">Nintedanib</span> Chemical compound

Nintedanib, sold under the brand names Ofev and Vargatef, is an oral medication used for the treatment of idiopathic pulmonary fibrosis and along with other medications for some types of non-small-cell lung cancer.

The Effect Model law states that a natural relationship exists for each individual between the frequency (observation) or the probability (prediction) of a morbid event without any treatment and the frequency or probability of the same event with a treatment . This relationship applies to a single individual, individuals within a population, or groups. This law enables the prediction of the (absolute) benefit of a treatment for a given patient. It has wide-reaching implications in R&D for new pharmaceutical products as well as personalized medicine. The law was serendipitously discovered in the 1990s by Jean-Pierre Boissel. While studying the effectiveness of class-I antiarrhythmic drugs in the prevention of death after myocardial infarction, he stumbled upon a situation which contradicts one of the basic premises of meta-analysis theory, i.e. that the heterogeneity test was significant at the same time for the assumption “the relative risk is a constant” and “ is a constant”.

<span class="mw-page-title-main">Dulaglutide</span> Diabetes medication

Dulaglutide, sold under the brand name Trulicity among others, is a medication used for the treatment of type 2 diabetes in combination with diet and exercise. It is also approved in the United States for the reduction of major adverse cardiovascular events in adults with type 2 diabetes who have established cardiovascular disease or multiple cardiovascular risk factors.

<span class="mw-page-title-main">Dextromethorphan/bupropion</span> Combination medication

Dextromethorphan/bupropion (DXM/BUP), sold under the brand name Auvelity, is a combination medication for the treatment of major depressive disorder (MDD). Its active components are dextromethorphan (DXM) and bupropion. Patients who stayed on the medication had an average of 11% greater reduction in depressive symptoms than placebo in an FDA approval trial. It is taken as a tablet by mouth.

References

  1. ↑ Porta M, ed. (2016-07-21). "A Dictionary of Epidemiology". Dictionary of Epidemiology - Oxford Reference. Oxford University Press. doi:10.1093/acref/9780199976720.001.0001. ISBN   9780199976720 . Retrieved 2018-05-09.
  2. ↑ Vancak, V., Goldberg, Y., Levine, S. Z. (2020). "Systematic analysis of the number needed to treat". Statistical Methods in Medical Research. 29 (9): 2393–2410. doi:10.1177/0962280219890635. PMID   31906795. S2CID   210041962.{{cite journal}}: CS1 maint: multiple names: authors list (link)
  3. ↑ Laupacis A, Sackett DL, Roberts RS (1988). "An assessment of clinically useful measures of the consequences of treatment". N. Engl. J. Med. 318 (26): 1728–33. doi:10.1056/NEJM198806303182605. PMID   3374545.
  4. ↑ Richard T, Vanhaeverbeek M, Van Meerhaeghe A (September–October 2011). "The number needed to treat (NNT)". Revue Médicale de Bruxelles. 32 (5): 453–458. PMID   22165523.
  5. ↑ Citrome L (2011). "Number Needed to Treat: What It Is and What It Isn't, and Why Every Clinician Should Know How to Calculate It". The Journal of Clinical Psychiatry. 72 (3): 412–413. doi:10.4088/JCP.11ac06874. PMID   21450157.
  6. ↑ "Number Needed to Treat". Bandolier. Archived from the original on 2020-10-19. Retrieved 2017-04-21.
  7. ↑ Nuovo J, Melnikow J., Chang D. (2002-06-05). "Reporting number needed to treat and absolute risk reduction in randomized controlled trials". JAMA. 287 (21): 2813–4. doi:10.1001/jama.287.21.2813. PMID   12038920.
  8. ↑ Hutton JL (2010). "Misleading Statistics: The Problems Surrounding Number Needed to Treat and Number Needed to Harm". Pharm Med. 24 (3): 145–9. doi:10.1007/BF03256810. ISSN   1178-2595. S2CID   39801240.
  9. ↑ Palle Mark Christensen, Kristiansen IS (2006). "Number-Needed-to-Treat (NNT) – Needs Treatment with Care". Basic & Clinical Pharmacology & Toxicology. 99 (1): 12–16. doi:10.1111/j.1742-7843.2006.pto_412.x. PMID   16867164. Archived from the original on 2013-01-05.
  10. ↑ Vancak, V., Goldberg, Y., & Levine, S. Z (2021). "Guidelines to understand and compute the number needed to treat" (PDF). Evid Based Ment Health. 24 (4): 131–136. doi:10.1136/ebmental-2020-300232. PMC   10231569 . PMID   33619181. S2CID   231992303.{{cite journal}}: CS1 maint: multiple names: authors list (link)
  11. ↑ Snapinn S, Jiang Q (2011). "On the clinical meaningfulness of a treatment's effect on a time-to-event variable". Stat Med. 30 (19): 2341–2348. doi:10.1002/sim.4256. PMID   21520457. S2CID   21986412.
  12. ↑ Bender R, Blettner M (2002). "Calculating the "number needed to be exposed" with adjustment for confounding variables in epidemiological studies". J Clin Epidemiol. 55 (5): 525–530. doi:10.1016/S0895-4356(01)00510-8. PMID   12007557.
  13. ↑ Austin PC (2010). "Absolute risk reductions, relative risks, relative risk reductions, and numbers needed to treat can be obtained from a logistic regression model". J Clin Epidemiol. 63 (1): 2–6. doi:10.1016/j.jclinepi.2008.11.004. PMID   19230611.
  14. ↑ Vancak V, Goldberg Y, Levine SZ (2022). "The number needed to treat adjusted for explanatory variables in regression and survival analysis: Theory and application". Stat Med. 41 (17): 3299–3320. doi:10.1002/sim.9418. PMC   9540555 . PMID   35472818.{{cite journal}}: CS1 maint: multiple names: authors list (link)
  15. ↑ Sever PS, Dahlöf B, Poulter NR, et al. (2003). "Prevention of coronary and stroke events with atorvastatin in hypertensive patients who have average or lower-than-average cholesterol concentrations, in the Anglo-Scandinavian Cardiac Outcomes Trial—Lipid Lowering Arm (ASCOT-LLA): a multicentre randomised controlled trial". Lancet. 361 (9364): 1149–58. doi:10.1016/S0140-6736(03)12948-0. PMID   12686036. S2CID   9409142.
  16. ↑ John Carey. "Do Cholesterol Drugs Do Any Good?". Business Week. Archived from the original on December 28, 2014. Retrieved 2008-03-31.
  17. ↑ Vancak V, Sjölander, A (2024). "Estimation of the number needed to treat, the number needed to be exposed, and the exposure impact number with instrumental variables". Epidemiologic Methods. 13 (1): 20230034. arXiv: 2307.09319 . doi:10.1515/em-2023-0034.
  18. ↑ Pearl J (1999). "Probabilities of Causation: Three Counterfactual Interpretations and their identification". Synthese. 121: 93–149. doi:10.1023/A:1005233831499. S2CID   7019552.
  19. ↑ Tian J, Pearl J (2000). "Probabilities of causation: Bounds and identification". Annals of Mathematics and Artificial Intelligence. 28: 287–313. doi:10.1023/A:1018912507879. S2CID   150352.
  20. ↑ Pearl J (2009). Causality: Models, Reasoning and Inference. Cambridge University Press. doi:10.1017/CBO9780511803161. ISBN   9780511803161.
  21. ↑ Mueller S, Judea Pearl (2022). Personalized Decision Making -- A Conceptual Introduction (PDF) (Technical report). UCLA.
This page is based on this Wikipedia article
Text is available under the CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.