New model predicts gain in healthy life expectancy for individual patientLiterature - Dorresteijn JAN et al, BMJ 2016
How to translate clinical trial results into gain in healthy life expectancy for individual patients
Dorresteijn JAN, Kaasenbrood L, Cook NR et al.,
BMJ 2016;352:i1548 doi: http://dx.doi.org/10.1136/bmj.i1548 (Published 30 March 2016)
Lifetime prediction modelsExtrapolation of risk predictions beyond the duration of follow-up of the studies is precarious. As compared with traditional survival models that use follow-up as the underlying timescale, lifetime model use age. Entry into the study, and time to event or censoring is defined by age. This way, predictions of lifetime models are not limited by the duration of follow-up, but by the age distribution of participants.
Traditional survival analyses inappropriately assume that people will not die from other causes than the one studied, by censoring patients who died from competing events. The disease of interest and competing events may, however, not be independent, since they may share mutual risk factors.
Disregarding competing risks can lead to erroneous conclusions about the relative effect of risk factors and treatment. Moreover, the cumulative incidence of disease may be overestimated, since patients who die from other causes are no longer at risk for the disease. Thus, the population at risk gradually declines over time, a factor not accounted for in traditional survival analyses.
Adjustment for competing risk is therefore essential when predicting the lifetime risk of disease.
Lifetime predictions are based on combining the techniques of adjustment for competing risk and an age-based time axis. The Joint British Societies recommendations on the prevention of Cardiovascular Disease (JBS3) is a good example. Such models use separate cause-specific models for the event of interest and the competing event. It is recommended to choose readily available patient characteristics and the same set of variables is used for prediction of the event of interest and of the competing event. Models based on trial datasets should include a covariate for the effect of treatment.
Competing risk adjusted lifetime prediction models
Estimation of (gain in) disease-free life expectancyBased on lifetime prediction models adjusted for competing risk, a life table for estimation of disease-free survival of individual patients can be composed. For a given age interval, the life table considers:
- the probability of being healthy and alive at the start of the interval,
- the probability of experiencing a disease event during that interval (given disease-free survival up to start of interval),
- the probability of experiencing a competing event during the interval (given disease-free survival up to start of interval).
Effect of lifelong treatment can be obtained by computing the survival difference between with and without treatment and is equal to the difference in area under the survival curves, plotted based on the life tables in both scenarios.
Real data example: aspirinAspirin is effective for primary prevention of cardiovascular (CV) disease, predominantly in high-risk individuals. Age is the most important determinant of 10-year CV disease risk, but also of death from non-CV causes. Thus, non-CV death may occur before a CV disease event, and should thus be considered a competing risk. Patients who die from non-CV causes may not benefit from aspirin at all, despite a high 10-year CV risk. That makes aspirin a classic example of a scenario in which doctors need to consider competing risks when estimating the treatment effect for an individual.
A competing risk adjusted lifetime model was developed based on data from the Women’s Health Study, which illustrates these concepts. The gain in CV disease-free life expectancy obtained with aspirin treatment in a hypothetical patient scenario is described, as well as the prediction without adjustment for competing risks (see full text article).
Notably, individual lifelong aspirin treatment effect predictions reveal an inverse association between treatment effect and age. The highest treatment effect is seen in younger patients with otherwise high risk factor levels, as a consequence of the fact that these individuals are at lower risk for competing non-CV disease mortality. Thus, this group will benefit more from preventive treatment.
Validation of lifetime modellingThe Women’s Health Study invited participants for further observational follow-up, after the end of the randomised treatment study period. This extension resulted in a total of 17.5 years of follow-up. Only the first 10.1 years of follow-up data were used for model development, thus the remaining time frame could be used for temporal validation. This showed that the model’s long term survival probabilities closely match those of the observed Kaplan-Meier survival for the example calculations.
Additional considerationsA better comparison of the efficiency of treatment may be obtained when treatment duration is standardised, for instance by truncating to predict average gain in disease-free life expectancy in the next 10 years. Also changes in risk factor levels over time, or age-related fluctuations in biomarkers, may be considered in similar models.
Model predictions only apply to patients who would have been eligible to participate in the trial of which the data were used. Thus, stringent enrolment criteria may exclude certain patient subgroups, for instance very elderly patients. But named models can also be developed based on observational cohort studies, which usually enrol a more heterogeneous patient population.
Find this article at The BMJ