While drug-coated balloons (DCBs) have been “consistently efficacious” across various studies, the mortality signal has not, Peter Schneider (University of California San Francisco, San Francisco, USA) noted at this year’s Leipzig Interventional Course (LINC 2020; 28–31 January, Leipzig, Germany). He concluded that this is partly because of ascertainment bias, but also “probably” because of treatment bias across different geographies.
Schneider began his talk in the paclitaxel update session at LINC 2020 by stating that he wanted to focus on the consistency of the mortality signal. “If something is dangerous, it should be dangerous everywhere,” he hypothesised.
Referring to the US Food and Drug Administration (FDA) panel on paclitaxel safety, held in June 2019, he quoted the findings directly: “There is no clear evidence of a dose effect on mortality and no identifiable pathophysiological mechanism for late deaths,” adding that these words were subsequently confirmed by the Vascular and Interventional Advances (VIVA) individual patient data meta-analysis conducted in November of the same year.
Schneider noted that efficacy results have been consistent across devices. The In.Pact investigational device exemption (IDE) trial—carried out in the USA and in the European Union (EU)—and the In.Pact Japan trial, both show a “dramatic, very consistent” improvement in patency at three years.
In contrast, Schneider remarked that the mortality risk has not been steady. “The hazard ratio seems to be disappearing before our eyes, and it is a little hard to tell what is going on,” he commented. “Why are we having so much trouble finding this signal in our real-world experience, both in our individual practices but also in these accumulations of now tens of thousands of patients, and why is there this dropping signal each time the RCT data is re-evaluated?” he questioned.
In search of an explanation, Schneider first considered patients lost to follow-up. He noted that the FDA did their own analysis of US RCTs, which they repeated after some of the patients lost to follow-up were found. This data was presented at the FDA panel in June and showed that, at five-year follow-up, initial relative risk of 1.72 is reduced to 1.57 after finding some of the patients that were lost to follow-up. This is a decrease in the mortality signal of 21%.
Schneider went on to reference the In.Pact series where the follow-up is now at 97%, in which “the same thing happened”: the hazard ratio decreased from 1.63 to 1.39—a decrease of 38%. “This indicates that patients are randomised on the way into the study, but they are not randomised in the way they are lost to follow-up. As we are finding them, we are revealing an ascertainment bias that influenced mortality results”.
Schneider also pointed out that the drop in mortality signal between the FDA and the VIVA analysis can be explained by considering geography. While the FDA included trials that were performed “to receive approval in the USA” and were therefore mostly conducted in the USA, in the VIVA analysis, trials were included that were performed outside the USA on devices that were approved in the USA. “If you look at it another way,” he explained, “the studies conducted outside the USA have hazard ratios that hovered around one, whereas most of those conducted inside the USA had hazard ratios of up to 1.8”.
Considering this on a “granular” level, patients who were entered into the In.Pact IDE trial from the USA had substantially higher hazard ratio for mortality than patients who were from outside the USA, Schneider commented. The same thing can be observed in the Levant II trial, he added.
“To evaluate whether there was actually some treatment bias going on here, the attendance to follow-up visits was evaluated; it turned out to be lower among the PTA [percutaneous transluminal angioplasty] patients than the DCB patients across all regions [USA, EU and Japan], but the largest gap was in the USA, and that is where the increased hazard ratio was,” Schneider explained.
Schneider concluded that although DCBs have been “consistently efficacious,” the mortality signal has not. “It has not been apparent in real-world data and it is not consistent within the randomised controlled trials, which is where we found [the signal] in the first place”. He posited that this is partly because of ascertainment bias, but may also be because of a treatment bias across different geographies.