In pursuit of the normal progressor: the holy grail for ALS clinical trial design?

The long list of futile clinical trials in amyotrophic lateral sclerosis (ALS) signifies the detrimental consequences of disease heterogeneity for effective drug development. The wide range of pathophysiological mechanisms underlying ALS results in a variety of clinical manifestations. As a consequence, trial populations consist of patients with varying degrees of disease severity, rates of decline and distributions of symptoms. This inflates the noise in efficacy endpoints and could disguise critical treatment clues. In order to improve the design of future clinical trials, defining a favorable patient population may benefit trial efficiency and increase the probability to detect effective compounds. In this issue of the Journal, Mora et al. illustrate the potential of excluding fast-progressing patients from clinical trials (1). The authors present a large randomized, placebo-controlled trial to investigate the efficacy and safety of Masitinib over a 48-week treatment period. Patients were classified into normal and fast progressing subgroups based on their baseline delta ALS functional rating scale (i.e. DFRS, defined as 48-ALSFRS-R score/ duration of symptoms (2)). Due to the lack of a molecular biomarker, the DFRS was used as indirect marker of disease aggressiveness. Interestingly, by excluding the fast progressors (defined as DFRS > 1.1) a treatment response appears in an otherwise futile trial population. Notwithstanding the inherited risks of subgroup analyses and the need for a confirmatory trial, Mora et al. report an intriguing observation that could be a valuable lesson for future clinical trials. The removal of patients with higher DFRS scores may have resulted in a less heterogeneous trial population, one that was more sensitive to detect a treatment response. Based on the PROACT database, we evaluated this hypothesis in Figure 1 where population heterogeneity is defined as the between-patient variability in ALSFRS progression rates after randomization. As can be seen, by including patients with higher DFRS scores, the between-patient variability in progression rates after randomization (red) increases and the trial population becomes more heterogeneous. Thus, by selecting patients with lower DFRS scores, one is able to make a more homogenous trial population, a phenomenon that may have benefited the Masitinib trial. However, Figure 1 also indicates that the removal of patients with higher DFRS scores additionally reduces the mean rate of decline after randomization (green). This is a critical observation and, in case one solely excludes patients with higher DFRS scores, the reduction in the mean rate of decline is larger than the reduction in the between-patient variability. As a result, the net effect of excluding patients with higher DFRS scores is a paradoxical worse signal-to-noise ratio, which could lower statistical power when implemented in future settings. This paradoxical result is partly driven by the relatively low predictive value of baseline DFRS for the progression rate after randomization (3). Both prognosis and progression rate are driven by a multivariate combination of patient characteristics (e.g. DFRS, age, site of onset, El Escorial classification and lung function); a single variable, therefore, tells only part of the story (4,5). Consequently, solely selecting patients based on