Lost in translation: Why ‘lost to follow-up’ matters

Lower extremity peripheral vascular interventions (PVI) are important for symptom treatment and limb salvage in patients with symptomatic peripheral artery disease (PAD). After PVI, short-term follow-up with the proceduralist allows for the opportunity to ensure adherence with medications, assess for clinical improvement, and obtain post-procedure noninvasive testing of the limb. The patient’s symptom status, clinical exam, and noninvasive test results all reflect a primary outcome of interest: vessel patency. This focus on vessel patency in clinical practice parallels the design of peripheral device trials, in which vessel patency is almost always the primary endpoint. However, following a peripheral revascularization, patients with PAD are at high risk for long-term cardiovascular and limb ischemic events and mortality.^1,2 Therefore, optimizing post-PVI outcomes and understanding the safety of new therapies in this population require longer-term follow-up past the traditional peri-procedural timeframe.

In this issue of Vascular Medicine, Wang and colleagues examine the phenomenon of loss to follow-up among patients with PAD undergoing PVI using data from the PVI Registry of the Vascular Quality Initiative. ³ Among the 39,342 patients included in this study, the rate of follow-up at 1 year was 91.6%. Compared with participants who had complete follow-up data, patients who were lost to follow-up were more often male, of non-white race, had a greater burden of comorbidities, and more frequently presented with critical limb ischemia. Patients lost to follow-up also were more likely to have post-procedure complications requiring admission and were less likely to be discharged on aspirin, P2Y12 inhibitor, or statin. After multivariable-adjusted modeling, older age, male sex, use of dialysis, more severe American Society of Anesthesiology class, and longer fluoroscopy time were associated with higher risk of loss to follow-up.

The authors additionally examined survival in this study. They demonstrated that over a period of up to 8 years, patients remaining in follow-up had significantly better survival than those lost to follow-up (83.5% vs 43.2%, p < 0.001). In particular, follow-up in the office was associated with greater survival than phone follow-up. After multivariable adjustment, loss to follow-up at 1 year was associated with a 6.56-fold greater risk of death. Although patient follow-up is important for many reasons, causality regarding post-procedural follow-up and its association with increased survival cannot be inferred from this observational study due to a high risk of unmeasured confounding. Consistent with the baseline characteristics of those lost to follow-up, it is likely that these patients were sicker and had more severe disease, resulting in a higher risk of mortality.

Nevertheless, these observations contribute to efforts to better understand outcomes and to optimize care for patients with PAD. In general, patients with PAD represent an understudied population. One of the challenges in caring for this population is the fragmentation of care provided, as patients may receive vascular treatment from multiple medical specialties, including cardiology, vascular medicine, vascular surgery, and interventional radiology. Patients with PAD also typically have comorbidities requiring additional visits, not only with their internist or other primary care provider but also with other specialists. Given the multiple clinicians caring for patients with PAD, there may be faulty provider assumptions regarding who is following and optimizing the patient’s medical treatment, resulting in underutilization of guideline-recommended therapies. The patient also may not understand the role of each provider, especially post-PVI, which may increase the risk of missing appointments.

Beyond the implications for patient care and improving outcomes, complete patient follow-up is critically necessary to fully understand the risks and benefits of both investigational and established therapies. Clinical trials typically define a primary efficacy endpoint for which the sample size and number of efficacy events is pre-specified to achieve statistical significance. In contrast, safety assessments are descriptive and limited to the sample size, duration of follow-up of the study, and number of safety events of interest. The understanding of safety and tolerability of a new agent or device may be incomplete in individual trials or even when a series of studies of the test agent are analyzed together, particularly if the safety events of concern are relatively rare but hold the potential for significant harm.

In any randomized clinical trial, patients who were enrolled and actively participating who then become unreachable are considered lost to follow-up. Outcomes, including death, in these patients become difficult, if not impossible, to ascertain. Loss to follow-up is likely occurring in a non-random fashion such that patients randomized to the experimental therapy who go missing may be at higher risk for adverse events than patients in the control group due to measured or unmeasured side effects of the experimental therapy. This potential for non-random removal (censoring) of data is called informative censoring because the censoring is not random, and missing patients still retain critical information for the outcomes of the study.

The standard for a well-run cardiovascular outcomes trial evaluating a drug therapy is to have very little loss to follow-up. For example, in the EUCLID trial examining ticagrelor versus clopidogrel in symptomatic PAD, there were five patients (0.04%) out of 13,885 patients randomized that were lost to follow-up. ⁴ An additional nine participants withdrew consent, resulting in a total of seven in each group who had unknown vital status at the end of the study. When these trials are reviewed by regulatory authorities, the US Food and Drug Administration (FDA) will perform a sensitivity analysis that adopts the most conservative approach: patients randomized to study drug and lost to follow-up are assumed to have the worst outcome (presumed dead), while patients randomized to placebo and lost to follow-up are assumed to have survived without any events. If a trial has more subjects with unknown vital status than the number of subjects with a primary efficacy event, the results are essentially uninformative.

A recent publication highlighting these concerns is the meta-analysis of paclitaxel-eluting devices, including drug-eluting stents and drug-coated balloons, used in PVI, in which the authors conclude that use of these devices was associated with greater risk for long-term mortality. ⁵ Consequently, the FDA issued a letter advising providers to consider alternative treatment options for most patients, citing concern over the signal for increased mortality seen with ‘paclitaxel-coated products.’ ⁶ However, substantial missing data in the studies included in this meta-analysis may limit interpretation of the results. For example, the 5-year results for treatment of femoropopliteal arteries with the Zilver PTX drug-eluting stent demonstrated a combined rate of loss to follow-up plus withdrawal of consent of 6.4% per year, potentially resulting in missing data from 32% of the study population. ⁷ Similarly, in a report of 5-year results after treatment of femoropopliteal arteries with paclitaxel-coated balloons, only 31 (30.4%) of 102 patients randomized were available for 5-year follow-up. ⁸ These examples highlight that missing data may lead to an incomplete understanding of safety concerns (mortality); therefore, it is impossible to know whether the results of the meta-analysis represent a real safety signal.

Patients with PAD undergoing PVI are a high-risk population in whom there is ongoing interest in developing new therapeutic drugs and devices. Ensuring complete follow-up in PAD trials is critical for generating high-quality data that allow for a complete understanding of the risks and benefits and net clinical benefit of therapies. Importantly, Wang and colleagues have demonstrated that patients lost to follow-up after PVI in ‘the real world’ are sick and have poor outcomes. Their data highlight the importance of missing data that lead to a lost opportunity in translation of results from clinical trials into clinical practice.