Net treatment benefit
Net Treatment Benefit (NTB) is a statistical measure of treatment effect used in randomized clinical trials. Developed by Marc Buyse for multiple prioritized outcomes, it is derived from Generalized pairwise comparisons (GPC) and summarizes the overall difference between treatments as the net probability that a randomly selected patient receiving the experimental treatment has a better overall outcome than a randomly selected patient receiving the control treatment.[1]
Net Treatment Benefit can summarize treatment effects across multiple clinically relevant outcomes, including efficacy, safety and patient-reported outcomes, while preserving their relative clinical importance through pre-specified prioritization.[2]
Definition
[edit]For two treatment groups, experimental (E) and control (C), the Net Treatment Benefit is defined as
- NTB = P(E > C) − P(C > E)
where P(E > C) is the probability that a randomly selected patient receiving the experimental treatment has a better overall outcome than a randomly selected patient receiving the control treatment, and P(C > E) is the probability of the opposite situation.[1]
The NTB ranges from −1 to +1. A value of zero indicates no overall treatment difference. Positive values favor the experimental treatment, whereas negative values favor the control treatment.
Interpretation
[edit]NTB has a probabilistic interpretation as an absolute measure of treatment effect.
For example, an NTB of 0.15 indicates a net probability of 15% that a randomly selected patient receiving the experimental treatment will have a better overall outcome than a randomly selected patient receiving the control treatment, according to the pre-specified comparison rules.[1]
When multiple outcomes are prioritized, the interpretation remains unchanged. Rather than evaluating a single endpoint, the comparison reflects the overall clinical benefit determined by the outcome hierarchy established before the trial. Patients are first compared on the outcome considered most clinically important, with lower-priority outcomes considered only when higher-priority outcomes do not distinguish between patients.[2]
Because NTB is expressed on an absolute probability scale, its reciprocal has been proposed as a number needed to treat for the comparison of treatment effects on several prioritized outcomes.[3]
Comparison with the Win Ratio
[edit]The Win ratio is another summary measure derived from generalized pairwise comparisons for multiple prioritized outcomes.[4]
The two measures are based on the same pairwise comparisons, therefore tests of statistical significance are identical for the two measures. The win ratio is calculated as the ratio of favorable to unfavorable comparisons after excluding neutral pairs, whereas NTB is calculated as the difference between the probabilities of favorable and unfavorable comparisons.
Because NTB is expressed as an absolute probability difference, it has a direct probabilistic interpretation. Unlike the Win Ratio, NTB does not ignore neutral comparisons, it eliminates them by subtraction.[2]
Authors have discussed limitations of the win ratio as a measure of treatment effect.[5]
Applications
[edit]NTB has been applied in clinical trials evaluating treatments across a range of therapeutic areas, including oncology, cardiovascular disease and rare diseases.[6][2]
Because it can jointly evaluate efficacy, safety and patient-reported outcomes within a single measure of treatment effect, NTB has also been proposed for benefit-risk assessment, patient-centred endpoint design and dose optimization in clinical development.
Limitations
[edit]The interpretation of NTB (and Win Ratio) depends on the comparison rules specified before the analysis, including the choice of outcomes, their order of priority and any thresholds defining clinically meaningful differences. NTB may differ across patient populations with different baseline risks, hence the NTB estimated in a specific sample of patients does not generalize to different patient populations.[3]
As with other measures of treatment effect, estimates may differ across study populations with different baseline risks. Calculation, including inferential statistics, may also become computationally intensive in large studies because every patient in one treatment group is compared with every patient in the other.[7]
References
[edit]- 1 2 3 Buyse, Marc (December 30, 2010). "Generalized pairwise comparisons of prioritized outcomes in the two-sample problem". Statistics in Medicine. 29 (30): 3245–3257. doi:10.1002/sim.3923. ISSN 0277-6715. PMID 21170918.
- 1 2 3 4 Verbeeck, Johan; De Backer, Mickaël; Verwerft, Jan; Salvaggio, Samuel; Valgimigli, Marco; Vranckx, Pascal; Buyse, Marc; Brunner, Edgar (September 2023). "Generalized Pairwise Comparisons to Assess Treatment Effects". Journal of the American College of Cardiology. 82 (13): 1360–1372. doi:10.1016/j.jacc.2023.06.047. hdl:1942/42083. PMID 37730293.
- 1 2 Buyse M, Verbeeck J, Saad ED, De Backer M, Deltuvaite-Thomas V, Molenberghs G (eds.). Handbook of Generalized Pairwise Comparisons: Methods for Patient-Centric Analysis. Chapman & Hall/CRC; 2025.
- ↑ Pocock, S. J.; Ariti, C. A.; Collier, T. J.; Wang, D. (January 2012). "The win ratio: a new approach to the analysis of composite endpoints in clinical trials based on clinical priorities". European Heart Journal. 33 (2): 176–182. doi:10.1093/eurheartj/ehr352. ISSN 0195-668X. PMID 21900289.
- ↑ Butler, Javed; Stockbridge, Norman; Packer, Milton (May 2024). "Win Ratio: A Seductive But Potentially Misleading Method for Evaluating Evidence from Clinical Trials". Circulation. 149 (20): 1546–1548. doi:10.1161/CIRCULATIONAHA.123.067786. ISSN 0009-7322. PMID 38739696.
- ↑ Péron, Julien; Roy, Pascal; Conroy, Thierry; Desseigne, Françoise; Ychou, Marc; Gourgou-Bourgade, Sophie; Stanbury, Trevor; Roche, Laurent; Ozenne, Brice; Buyse, Marc (December 2016). "An assessment of the benefit-risk balance of FOLFIRINOX in metastatic pancreatic adenocarcinoma". Oncotarget. 7 (50): 82953–82960. doi:10.18632/oncotarget.12761. ISSN 1949-2553. PMC 5347744. PMID 27765912.
- ↑ Ozenne, Brice; Budtz-Jørgensen, Esben; Péron, Julien (November 2021). "The asymptotic distribution of the Net Benefit estimator in presence of right-censoring". Statistical Methods in Medical Research. 30 (11): 2399–2412. doi:10.1177/09622802211037067. ISSN 0962-2802. PMID 34633267.