External validation of a quantitative model to guide patient selection for placement of an optional or permanent inferior vena ca va filter

INTRODUCTION

Inferior vena cava filter (IVCF) placement has been performed in various clinical situations in patients with venous thromboembolism (VTE).^[1] In addition, filtration was previously recommended prophylactically in patients with high risk of VTE,^[2-4] although the majority of current guidelines discourage prophylactic use.^[1,5,6] The Food and Drug Administration (FDA) approval of optional, or retrievable, filters in 2003 contributed to the increase in use of IVCFs in the US from 2000 to 2009.^[7] Growing awareness of potential risks associated with IVCF placement resulted in 2010 and 2014 FDA warnings encouraging physicians to ensure proper follow-up for patients who have received an optional IVCF.^[8,9] Over the past decade, multiple reports were published illustrating the efficacy of dedicated programs in improving filter retrieval rates. Despite that, a significant proportion of filters remain unretrieved, with reported retrieval rates ranging from 28% to 60%.^[10-14] A significant proportion of patients either die during the follow-up period or their filters are deemed permanent on follow-up.^[14] This subgroup of patients could benefit from placement of permanent IVCFs, which are associated with less reported complications,^[15-17] less cost, and do not require dedicated follow-up.^[17] In 2013, Eifler et al. published a study utilizing logistic regression analysis for modeling optional IVCF permanence, based on multiple patient parameters correlating to retrievability of these filters, and formulated an equation that would aid clinicians in the decision between placing an optional versus a permanent filter. The study yielded a c-statistic of the formula of 0.8, indicating a good predictive ability.^[18] This formula has never been validated on an external patient population. Thus, our study aims to externally validate this quantitative model.

MATERIAL AND METHODS

The study was approved by the Institutional Review Board and was Health Insurance Portability and Accountability Act compliant. IVCF placement was performed in accordance with published guidelines.^[3]

RESULTS

A total of 270 patients with removable IVCFs were identified during the study period. Seventy-one filters (26.3%) were categorized as “removed” and 199 (73.7%) were “kept permanent.” The patient parameters are summarized in Table 1. The age at placement was statistically significant, with the patients in “removed” category being younger in comparison to “kept permanent” (mean age 63.8 vs. 50.6 years; P < 0.001). A statistically higher proportion of patients with history of malignancy were in the “kept permanent” category in comparison to the “removed” category (34.2% vs. 16.9%; P = 0.0062). No other statistically significant differences were identified between the two groups.

The univariate logistic regression analysis of the parameters was subsequently performed. Given that only one patient with high risk VTE was present in the studied cohort, a logistic regression analysis was performed with exclusion of the high-risk VTE patients and summarized in Table 2. Of those, advanced age was associated with the IVCF being kept permanently versus retrieved (OR 1.05; 95% CI 1.03–1.07; P < 0.001) as well as the presence of malignancy (OR 2.55, 95% CI 1.29–5.07; P = 0.0075). None of the other parameters, including sex, history of VTE, presence of neurologic disease, indication for IVCF placement due to VTE/anti-coagulation complication, or indication for IVCF placement due to VTE/anti-coagulation failure was statistically significant.

The receiver operator characteristic curve for the prediction model based on the previously published formula yielded an area under the curve of 0.66, compared to the previously published value of 0.8. The receiver operator curve illustrated in Figure 1. The sensitivity and specificity of the prediction model were estimated 64.3% and 62.0%, respectively.

Close

Figure 1:

Receiver operating characteristic curve for reference predicted probability. Area under the curve = 0.66.

Export to PPT

DISCUSSION

Despite the availability of optional IVCFs, there are still several potential advantages of placing a permanent device. First, current evidence shows that indwelling optional filters are associated with a higher complication rate.^[15-17,19] For example, in a retrospective study evaluating 449 retrievable versus 785 permanent filters, Desai et al. showed statistically significant higher overall rates of complications with retrievable filters (9% vs. 3%).^[15] These included both thrombotic complications, such as lower extremity deep venous thrombosis, pulmonary embolism, and inferior vena cava thrombosis, and device-related complications, such as filter penetration, migration, or device fracture.^[15] Likewise, a review of the Manufacturer and User Facility Device Experience database demonstrated that 86.8% of all reported adverse events for IVCFs were associated with optional IVCFs.^[16] Second, there may be cost benefits to placing permanent devices. For instance, some authors have previously shown that the predicted cost of a permanent IVCF is lower than that of an optional device, in addition to a higher predicted rate of clinical efficacy.^[17] Finally, the utilization of a permanent IVCF obviates the need for routine clinical follow-up. Without a dedicated system in place, IVCF retrieval rates are poor. However, despite an effort and success in implementation of such programs, optional filters retrieval rates are repeatedly reflected in literature to be below 60%.^[10-14] One analysis of a filter retrieval program showed that despite a low percentage of patients lost to follow-up, only 28% of filters were retrieved. Almost 50% of the patients with optional filters had their filters either deemed permanent or died before a decision regarding filter retrieval.^[14] These group of patients could benefit from placement of permanent IVCFs or optional IVCFs with clear intent of acting as permanent devices with no scheduled follow-up. This approach represents a more cost-effective option, emphasizing a need for improved patient selection tools.

Eifler et al. conducted a study using logistic regression analysis for quantitative prediction of optional IVCF permanence.^[18] They analyzed 265 optional IVCFs placed between December 2008 and July 2011 in a retrospective manner. The authors observed patient parameters correlating with retrievability of these filters and formulated an equation that would aid clinicians in a decision between placing an optional versus a permanent filter. This equation is available as an easy-to-use online-based calculator.^[20] In their multivariable model predicting filter disposition, significant factors associated with permanence were age (OR 1.03; 95% CI 1.01–1.05), male sex (OR, 3.01; 95% CI, 1.64–5.54), underlying malignancy (OR, 3.27; 95% CI, 1.77–6.03), and an indication of anticoagulation failure (OR, 8.12; 95% CI, 1.83–36.0). Significant factors associated with removal were history of VTE (OR, 0.39; 95% CI, 0.21–0.74), prophylactic filter placement indication (OR, 0.14; 95% CI, 0.04–0.43), and high-risk VTE (OR, 0.37; 95% CI, 0.15–0.94). This analysis yielded a c-statistic of the formula of 0.8, indicating a good ability to predict filter permanence.

The current data aim to externally validate this quantitative model in a different patient population. Similar to prior reports, age and presence of malignancy were significantly associated with retrievable IVCFs being kept as permanent devices. It is intuitive, that advanced age implies more comorbidities, higher chance of dying, as well as limited opportunities to follow-up for a filter retrieval in comparison with a younger patient population. Furthermore, this analysis supports the conclusions of prior studies that the presence of malignancy decreased the probability for retrieval. Sex was shown as a predictor of filter retrieval by Eifler et al.,^[18] which, however, was not a statistically significant predictor in study by D’Othée et al.,^[21] similarly to our study. Indications for filter placement in our patient population were not statistically significant for filter retrieval prediction. Patients with an indication for placement of anticoagulation failure, which was predictive of filter permanence in the Eifler et al.^[18] cohort, were presented in very small numbers in both the reference study and the present study, which could explain discordance in the significance of this finding. Only one high-risk VTE patient was present in our cohort and was, therefore, excluded from this analysis. The number of IVCFs placed for prophylaxis at our institution was markedly lower in comparison to the study by Eifler et al,^[18] which might explain the discordance between the current data and reference report. The differences in patient numbers between the two data sets are likely due to evolving standards with regard to prophylactic filter placement. The receiver operator characteristic curve based on the prediction model by Eifler et al.^[18] yielded the area under the curve of 0.66 for this external patient population, compared to the previously published value of 0.8. The sensitivity and specificity of the prediction model were 64.3% and 62.0%, respectively. These findings indicate poor performance of the formula in this external patient population.

The study is limited by several factors. First, it is retrospective in nature, which limits data collection to that which is available. Second, there is no standardized process at the authors’ institution for the selection of a permanent or optional IVCF. This means that patient selection was subject to operator preferences and bias as well as filter availability at the time of placement. Third, the decision to attempt filter retrieval is subject to physician bias and patients’ preferences. In the authors’ anecdotal experience, patients’ preferences have a significant role in the decision to attempt retrieval. Fourth, the two studies were performed at different time points. Even though the purpose of the paper was to externally validate the previously published quantitative model, evolving guidelines regarding filter placement may have had a role in discrepancies between the two populations. Regardless of these limitations, this analysis shows that the application of this quantitative model to a different patient population resulted in a relatively poor predictive performance. Nevertheless, placement of permanent IVCFs in patients with advanced age and presence of malignancy should be considered, as these factors were repeatedly associated with lower likelihood of retrieval in va rious patient populations. Larger scale studies to derive a better model for selection of candidates for permanent filter placement should be considered.

CONCLUSION

Advanced age and presence of malignancy are associated with optional IVCFs being kept as permanent devices. However, the previously published formula did not yield an accurate quantitative prediction in this external patient population.