Interpreting the Net Treatment Benefit (NTB)
One advantage of NTB, as compared to the win ratio, is its interpretation in terms of a difference between two probabilities:
- considering a single single outcome, the NTB represents the net probability that a random patient has a better outcome in the experimental group than in the control group (Buyse 2010). The net probability is the difference between the probability of a better outcome in the experimental group than in the control group minus the probability of a better outcome in the control group than in the experimental group.
- considering a single outcome with a threshold of clinical similarity, the NTB represents the net probability that a random patient has an outcome better by at least the threshold of clinical similarity in the experimental group than in the control group (Péron et al. 2016).
- for multiple prioritized outcomes, the NTB represents the net probability that a random patient has a better outcome in the experimental group than in the control group, either for the outcome of highest priority, or, in case of a tie for the outcome of highest priority, for the outcome of next priority, and so on.
- for multiple non-prioritized outcomes, the NTB represents the net probability that a random patient has a better outcome, on average, in the experimental group than in the control group.
NTB ranges from -1 to +1, with a value of 0 indicating no difference between the treatment groups. NTB being a difference between two probabilities, its reciprocal is the number needed to treat (NNT), a measure of effect familiar in health technology assessment:
In contrast to NTB, which is a difference between two probabilities (an absolute measure of effect), the win ratio is a ratio of probabilities (a relative measure of effect). The win ratio does not have a straightforward interpretation, except for a single outcome under proportional hazards, in which case Oakes (2016) has shown that the win ratio is the reciprocal of the hazard ratio. An important advantage of the hazard ratio is that it is likely to be similar across populations having different baseline risks. This property does not hold for the win ratio when multiple outcomes are considered, not does it hold for NTB.
Based on the above definitions, the interpretation of the NTB in this specific example, is that there is a net probability of 9% that a random patient has a better outcome in the experimental group than in the control group, either in Time to death, or in case of a tie in time to death, in time to progression, or in case of a tie in time to progression, in toxicity grade 4.
This also means that on average, 11 patients (1/0.09= 11.11) need to be treated in order to see at least one patient benefiting from the experimental treatment in terms of time to death, time to progression or toxicity grade 4 (in this specific order).
If we decompose the overall NTB, we can also see that there is a net probability of 2% that a random patient has a better outcome in the experimental group than in the control group in Time to death. If there is no difference in time to death, there is an additional 10% net probability that a random patient has a better outcome in the experimental group than in the control group in terms of Time to progression. An in case of tie in time to death and time to progression there is a net probability of 3% for a random patient to have a better outcome in the control group than in the experimental group in toxicity grade 4.
Overall, there is then a 9% net probability that a random patient has a better outcome in the experimental group than in the control group, either in Time to death, or in case of a tie in time to death, in time to progression, or in case of a tie in time to progression, in toxicity grade 4.