In Statistics we Trust: Towards the Careful Derivation and Interpretation of Meaningful Survival Estimates in Cystic Fibrosis

Advances in therapies have not only improved the quality but increased the longevity of life for people with cystic fibrosis (CF). Understandably, survival projections are of critical importance to individuals with CF and their families. Over the years, significant efforts have gone towards developing survival models to identify patient factors associated with higher risk of mortality and predict the probability of death for individual patients (1–6). In parallel, annual estimates of median survival among large cohorts followed through national patient registries have been reported (7–9). In both of these undertakings, it is imperative that CF researchers go beyond elegant statistical modeling and estimation to develop pragmatic measures that can be clearly communicated to patients and families.

In this issue of the JCF, we are presented with two key articles that together add to our understanding of complex survival metrics. Taking the mystery out of the calculation of survival metrics is the focus of the commentary provided by Keogh and Stanojevic (10), which provides a welcome clarity in outlining a statistic that is so often publicized but not well understood, the median age of survival. Two pivotal estimation approaches are explained, the birth cohort and period approaches; the former represents a measure of trends across birth cohorts, whereas the latter provides a contemporary prediction of survival for babies born in a recent period. Recently, it was reported in the US CF Foundation Patient Registry that in 2016, the median predicted survival age of a newborn with CF was 47.7 years (95% confidence interval [CI] 45.6–51.1 years) as compared to 41.2 in 2015 (95 %CI 38.2–43.9). This change shows clear improvements in survival that could reflect improvements in care, potentially attributable to the introduction of CFTR modulators or a marked increase in transplants (8). Using the period approach between 2012 to 2016, the median predicted survival age was 42.7 years (95% CI 41.7– 43.9 years) reflecting that 50% of babies born in the US between 2012 and 2016 are estimated to live beyond 42.7 under assumptions that there will be no improvements in care to impact future mortality rates. Despite differences in these estimates, it is clear that each of these statistics provide clear purpose for different intended uses. Moreover, Keogh et al discuss yet another survival metric that may be even more meaningful when it comes to clinical care – conditional survival, which enables a more customized estimation of median survival age relevant to cohorts of patients of various ages, and not just babies. If a practitioner were discussing CF survival with parents of a 14-year-old in 2018, a survival prediction from when they were born in 2004 would need to be heavily caveated and thus prone to misunderstanding. Conditional survival offers a more pragmatic approach to conveying meaningful survival projections to our patients.

It cannot be emphasized enough that these survival metrics are population level predictions rather than predictions for an individual patient. In fact, there has been significant effort towards developing prediction models for individual patients resulting in mixed accuracy (2–4). While forced expiratory volume in one second (FEV1) as a percentage of predicted less than 30% has been prominently associated with increased risk of mortality, there is significant heterogeneity in survival outcomes among cohorts with lung function below this threshold, revealing the difficult use of this variable on an individual patient basis (11). Thus, we stand on our most solid ground when interpreting survival metrics on a population rather than individual level, and efforts to refine these population level estimates utilizing our newfound understanding of survival metrics is the focus of the second paper by Keogh et al. (12).

Using the UK CF registry, Keogh et. al. provide a new framework for the estimation of informative survival metrics through novel survival analysis approaches. Their approach enables more refined estimates of median survival for key subgroups of patients which do not force the associations between covariates and survival to be linear, such as age of diagnosis. As such, this is the first paper to report detailed survival statistics in the population of individuals with CF in the United Kingdom by key patient characteristics. One striking finding in this paper is the difference between median predicted survival age from birth versus median survival age conditional on the survival to a specific age. For instance, the median survival age from birth among F508del homozygous patients was 46 as compared to 52 years among F508del homozygous patients who had survived to age 30.This is comparable to that observed in the US CF population, with 50% of individuals with CF living to age 30 predicted to live beyond 50 years of age (8). In clinical applications, care providers may find the utility of conditional survival estimates to be much more informative when engaging in patient discussions. The methods proposed by Keogh et. al. provide a valuable new statistical approach which tailors survival estimates towards clinically relevant cohorts defined by known predictors of survival, and which enables estimation of changes that may occur with future changes in mortality rates. Overall, this approach affords a new perspective to predict the future of an ever-growing population: those living with and surviving CF.

At the end of the day, statistics are just that - estimates, with error, of our current outlook or future predictions. The ability to standardize these estimates must continue to be a key effort both within and across patient registries, and the newly formed International CF Registry Harmonization Group will be well positioned to tackle this effort in their work to standardize data and methodology globally across several CF patient registries. As we continue to grasp for our own understanding of survival metrics, we can ground ourselves by using them as indicators of the tremendous progress in care and therapeutics we will continue to witness, and as ever important milestones to beat when caring for individuals with CF.