By Bertrand Janne d’Othée
The introduction of new optionally retrievable filters approximately a decade ago generated lots of initial excitement. Hopes were rising high that many of these filters could be easily removed once they were no longer necessary. Much of the early literature focused on success rates of attempted filter retrieval, or stressed that these devices can be removed after increasingly longer dwelling times—often well beyond manufacturers’ instructions. Most filter removals are indeed uneventful, with a high degree of success, and this may have encouraged even further the adoption of these devices in practice.
About five years later, however, new reports started voicing concerns that too few optionally retrievable filters ended up being actually removed (1, 2). Real life experience had shown that removal rates were quite, in fact, low and most likely removed in less than half of patients (the actual rate is unknown). This is unfortunate as technical success of attempted removal is high (3). Low removal rates are expected to increase the risks of long-term filter implantation.
Causes for non-removal of implanted optional filters are multiple and may include technical, clinical, patient management or follow-up reasons. The major causes seem to be other than technical and several studies have attempted to better understand the reasons behind the problem (4, 5), but the overall picture from the existing literature, taken as a whole, is confusing.
Many factors of undetermined importance have been identified among these studies and with incompletely overlapping findings among studies it becomes very difficult to know which factors are the most important predictors that lead to the eventual removal of the filter. For example, some predictors have been found in more than one published series, including age, malignancy, filter tilting or malpositioning, caval thrombosis or presence of large clots trapped in filters, prolonged indwelling time after implantation, patient non-compliance, loss to follow-up, the absence of patient follow-up by the procedural service performing filter implantation, and the absence of a longitudinal follow-up programme after filter implantation supported by dedicated personnel.
However, other significant predictors have been mentioned only once in a single study; this does not necessarily imply that they might be less valid or important. They include, for example, the onset of a new clinical need for permanent interior vena cava filtration, cases of prolonged filtration extending beyond the filters’ time window for retrievability, patients not being discharged on anticoagulants, insufficient patient education, deep vein thrombosis, and the involvement of a large number of healthcare providers in a given patient. Still, other predictors have been found that may not always have an obvious explanation but might still be important clinically (eg. female gender, or various settings in which the discharge from the hospital or intensive care unit happened.)
Hence, all these publications have shown so much between-study variability and incompletely overlapping conclusions that one might wonder: what predictors are really the important ones, and which one(s) should we go after primarily? None of these predictors seem to be the best single explanation to the problem of low removal rates, and it appears there is likely no single miracle solution. An unknown combination of some predictors probably accounts better for the overall picture (not to mention other possible predictors that have yet to be identified). Finding the correct combination (the “best model”) is crucial to better direct efforts and resource utilisation to solve the problem of low filter removal rates.
If we had to pick one best predictor, most studies would agree that the first choice is the absence of a dedicated longitudinal follow-up programme. Such programmes lead to higher removal rates, more retrieval attempts, and decreased loss to follow-up. Although the benefits of these programmes are undeniable and should not be underestimated, these initiatives alone have not been able to fully solve the problem yet—that factor alone cannot explain everything. Further efforts are needed to find additional complementary solutions that can be implemented in parallel to these longitudinal follow-up programmes. For example, other ideas have also been launched that seem reasonable too, including (a) sensitisation of referring physicians to the clinical advantages of filter removal, (b) sensitisation of interventionalists to the cost advantage of filter removal (2), and (c) better identification of patients at risk of no future filter removal. The latter is where the role of predictive models comes into play.
The problem with the existing models is that many of the published studies reported their observations in a single patient cohort, built a model based on it, and then stopped short of validating their model. Therefore the question arises—how well does the model describe reality when used in a set of patients that is comparable but different from the study cohort? In the absence of validation, the models’ equation could simply describe what was observed in a given study sample, and not be generalisable to other similar patient groups (6).
This validation step is often forgotten in the medical literature, even in so-called clinical prediction rules. The time is ripe for future, second-generation studies to attempt to verify whether the aforementioned candidate predictors are indeed significant predictors and what their relative importance is. Model validation can be done by splitting upfront any new patient cohort into a training set and a testing set prior to data analysis, or by using bootstrapping (a specific statistical technique). The quality of these second-generation studies might also benefit from being integrated in new research reporting standards for studies focusing on optionally retrievable filters.
These questions are gaining interest as filter placement is increasingly performed nationwide in the USA, in part due to the perceived harmlessness of retrievable filters. A few well-designed studies (even small retrospective ones) are needed to elucidate convincingly the causes of low removal rates and translate their findings into improvements in patient care.
Bertrand Janne d’Othée is associate professor of Radiology, University of Maryland Medical Center, USA, and chair of the Evidence-Based Interventional Radiology (EBIR) Committee of the SIR Foundation.
1. Silberzweig JE. Successful clinical follow-up for trauma patients with retrievable inferior vena cava (IVC) filters can be challenging to achieve. J Trauma 2007; 63:1193.
2. Janne d’Othée B, Faintuch S, Reedy AW, Nickerson CF, Rosen MP. Retrievable versus permanent caval filter procedures: when are they cost-effective for interventional radiology? J Vasc Interv Radiol 2008; 19:384-392.
3. Ray CE, Jr., Mitchell E, Zipser S, Kao EY, Brown CF, Moneta GL. Outcomes with retrievable inferior vena cava filters: a multicenter study. J Vasc Interv Radiol 2006; 17:1595-1604.
4. Dinglasan LA, Oh JC, Schmitt JE, Trerotola SO, Shlansky-Goldberg RD, Stavropoulos SW. Complicated Inferior Vena Cava Filter Retrievals: Associated Factors Identified at Preretrieval CT. Radiology 2013; 266:347-354.
5. Eifler AC, Lewandowski RJ, Gupta R, et al. Optional or Permanent: Clinical Factors that Optimize Inferior Vena Cava Filter Utilization. J Vasc Interv Radiol 2013; 24:35-40.
6. Janne d’Othée B. Predicting the removal of optionally retrievable caval filters: are we there yet? J Vasc Interv Radiol 2013; 24:40-42.