D-Dimer Testing to Determine the Duration of Anticoagulation Therapy – The Prolong Study November 28, 2006
Posted by rajkmd in Hematology, Uncategorized.trackback
A D-dimer molecule, abutting ends at the D:D junction as part of the fibrinogen gallery at UCSD.
Background
- The optimal duration of warfarin treatment after deep venous thromboembolism (DVT) for the purpose of preventing future DVTs in unclear
- The risk of recurrance is greatest in the first 6 to 12 months after the initial episode and then gradually decreases.
- Previous studies have suggested that elevated D-dimer levels may predict the risk of developing DVTs in the future
- To test this hypothesis, patients with unprovoked DVTs and a positive D-dimer who were treated with vit. K antagonist for three months were randomized to recieve anticoagulation or to discontinue anticoagulation. Patients that had a normal positive D-dimers test after three months of anticoagulation therapy did not receive further treatment.
Methods
- All patients were treated with a minimum of 3 months of vitamin K antagonist therapy after an unprovoked DVT.
- Inclusion Criteria – Patients who are between the ages of 18 and 85 who had a first episode of symptomatic, unprovoked venous thromboembolism, defined as proximal DVT of the lower legs, pulmonary embolism, or both, all patients needed to be treated with a vit. K antagonist for three months with a target INR of 2.5 (range of 2-3)
- Unprovoked DVT was defined as a DVT not associated with pregnancy, fracture, immobilization, surgery, cancer, antiphospholipid antibody syndrome, or antithrombin deficiency.
- Other exclusion criteria inculded patients who had serious liver disease, renal insufficiency, other indications or contraindications for anticoagulation, or limited life expectancy.
- Pt’s with Factor V Leiden deficiency were allowed in the study.
- Finally, all patients underwent ultrasound to assess the proximal deep veins after three months of treatment with vit. K antagonists. If they had a recurrent DVT, during this time period, then they were excluded from the study.
- All patients were studied for a period of 18 months and were seen at the clinic at intervals of 3 to 6 months.
- This is an intention-to-treat analysis.
- This study was not blinded, though committee members were unaware of the results of the D-dimer test.
Results
- A total of 619 patients underwent D-dimer testing and a thrombophilia workup. Of these, 11 patients were ultimately excluded because they were positive for lupus anticoagulant or antithrombin deficiency.
- Of the 608 patients that were included, 63% had a normal D-dimer level.
- Of the remaining 223 patients with an abnormal D-dimer level , 103 were randomly assigned to receive anticoagulation and 120 were assigned to receive no anticoagulation.
- Of the pts with positive D-dimers
- Of the 120 pts who d/c’d anticoagulation –> 15% developed recurrent venous thromboembolism
- Of the 103 pts who resumed anticoagulation, one had a major bleeding event and 2 had a recurrent venous thromboembolism.
- Adjusted hazard ratio 4.26 (95% CI: 1.23-14.6) with p 0.007
- There was no significant difference in the rate recurrent venous thrmboembolism between the group with normal d-dimers compared to the group with abnormal d-dimers who were anticoagulated (adjusted hazard ratio, 2.46; 95% CI, 0.71 – 8.46), but the absolute difference (6.2% vs 2.9%) may be clinically significant.
Discussion
- The PROLONG study shows that patients with abnormal D-dimer test who were not anticoagulated had high rates of recurrance venous thromboembolism (15%), while pts with normal D-dimer had a significantly lower chance of developing a recurrent thromboembolism (6.2%). The adjusted hazard ratio comparing the rates of recurrance was 2.27 (95% CI 1.15-4.46, p 0.02)
- In patients with a positive D-dimer, when comparing those that were anticoagulated to the those that were not, the group that was anticoagulated had a low rate of bleeding and recurrence of thromboembolism (combined endpoint of 2.9%, p 0.005)
- Patients with abnormal D-dimer were significantly older than patients with a normal D dimer test (average age 70 yo versus 59 yo, p<0.001)
- The study was not powered to make a definitive assessment of the risk of bleeding, since the risk increases with time, the risk/benefit ratio may change as anticoagulation is prolonged.
THE BOTTOM LINE: In patients who have completed at least 3 months of anticoagulation therapy after an unprovoked venous thromboembolism, a positive D-dimer one month after discontinuation of anticoagulation therapy is correlated with the risk of recurrence. In patients with a positive D-dimer test, continuing anticoagulation therapy decreases the risk of recurrent venous thromboembolism when compared to patients who are not anticoagulated regardless of their D-dimer. In sum, this study shows that the D-dimer assay may help guide clinicians on the length of anticoagulation therapy, but it does not definitively establish the optimal course of therapy.
I have been recently approached by several residents asking for my opinion on the recent paper in NEJM discussing the use of D-dimer test in stratifying patients with venous thromboembolism (VTE) after the discontinuation of anticoagulation. Thank you, Raj, for opening this discussion. Here are some of my thoughts on the issue.
1. As pointed out in Raj’s summary, the paper demonstrated correlation between DD status after 3m of anticoagulation and the rate of recurrence. The question, however, is not whether or not there is a correlation, but rather how good the test is in risk stratification.
2. When assessing utility of a prognostic test, reasoning very similar to that used in assessing utility of a diagnostic test may be used. The math is complex(1), but in simplified terms, the only difference between diagnosis and prognosis is that the index event is coincidental with testing (in case of diagnosis) and is removed in time (in case of prognosis). Consequently, estimates of test accuracy similar to those used in diagnosis, may be used in prognosis (sensitivity, specificity, likelihood ratios). Perhaps more importantly, in direct parallel to the model used in diagnosis, Bayesian logic is useful in considering prognostic issues. In diagnosis, probability of an event after the test depends on the probability of that event prior to the test (baseline probability), and the properties of the test itself. Similarly, when prognosis is concerned, probability of an outcome after a risk-stratifying test will depend on the probability of that outcome prior to the test (baseline risk), and the test itself. The major advantage of this Bayesian approach in diagnosis is that we can apply known likelihood ratios of a test to each individual patient with individual risk of a condition in question. For example, a negative DD test in a patient at high risk for VTE does not have the same implications as the same DD test result in a patient at a low risk for VTE. The same logic applies to prognostication: the final risk of an event for each individual patient will depend not just on the properties of the test itself, but also on their baseline risk. For those not quite comfortable with such an extrapolation, I’d like to point out that diagnostic techniques and terminology are already being extensively used (frequently inappropriately). Recall all the outcome studies, using DD to rule out VTE. In a typical study, patients presenting with suspected VTE would have DD checked and those with a negative DD and a low baseline risk are followed for recurrence. By technical definition, all such studies are prognostic, not diagnostic, since the test under study and the reference test are significantly separated in time. Yet, these studies frequently report their results as specificity of the DD test.
3. Because individual studies always enroll populations at different (at least somewhat different) risk, direct comparisons of results of these studies may be problematic. This is usually well appreciated in case of diagnostic studies. Consider two studies evaluating the same DD test in the diagnosis of DVT without knowing that the tests under study are indeed one and the same with a negative LR of 0.1. The reference standard is Duplex ultrasonography in both studies, which are otherwise identically designed but conducted in 2 different hospitals. Study 1 enrolls patients with an average risk of VTE of 5%, Study 2 – 15%. The results show that among those with negative DD in study 1- 0.5% had DVT, while in study 2, 1.7% had the condition. Needless to say, the conclusion that DD test 1 is 3.3 times better in ruling out DVT is erroneous. All the difference between the post-test probabilities in this example comes from the difference in baseline risk. Similarly, when comparing prognostic studies, event rates after a prognostic test has been applied (and follow-up time has elapsed) are not informative when not coupled with baseline risk estimations.
4. As I mentioned above determination of prognostic test characteristics may be difficult. The simplest way of estimating the LR of prognostic test is to use Byes theorem in reverse. LR = PostTest Odds/ PreTest Odds. In the study of Palareti et al. the odds can be obtained from the probabilities (event rates) given in table 2.
Probability, Odds
per year
Pre-test 5.9% 0.062
Post-negative DD 4.4% 0.045
Post-positive DD 10.9% 0.129
Post-test probability is estimated by adding up the events (24+18=42) and dividing it by total number of patient years accrued (545.45 + 165.14 = 710.59).
Then NLR = 0.72 and PLR = 2.07.
It is important to realize that the above calculations assume a uniform rate of events over time. This certainly is not always correct, and specifically is not correct for VTE recurrence, as seen in fig 2, but this is the best estimate one can obtain from the presented data.
5. Both NLR and PLR of DD test in risk stratifying appear to be very poor. And therefore despite the apparent correlation between the test result and outcome, performance of the test is poor.
6. The study in question was a multi-center trial, using different DD test at each hospital. The results were classified as “normal” or “abnormal” as pertaining to each of the tests used. This is obviously, arbitrary. The “normal” cut-off has been identified for each test from ROC analysis for diagnosis of VTE, and there is no inherent reason why the same cut-off should be the best for prognostication purposes.
7. ROC analysis also allows for the quantification of the discriminating power of a test “overall”, regardless of the specific cut-off value used. Such analysis shows that AUC for DD-test is close to 0.5 (a useless test) (2, 3).
8. The baseline risk of recurrence in the study 5.9% per year is consistent with previously reported recurrence rates among patients with idiopathic VTE after anticoagulation is discontinued: 5-7.2% per year (4-6). It is lower in patients with a transient risk factor and higher in patients with non-removable risk factors. The study thus seems to be in sync with all the previously reported data.
9. Here are more papers investigating the utility of DD in predicting recurrence(6-8).
10. In summary, there is a correlation between DD-level and recurrence risk. Overall performance of the test is very poor. This is consistent with previous studies. I strongly urge you not to use the test for risk stratification in your patients.
Refs.
1. Cai T, Pepe MS, Zheng Y, Lumley T, Jenny NS. The sensitivity and specificity of markers for event times. Biostatistics 2006;7(2):182-97.
2. Le Gal G, Bounameaux H. d-Dimer testing to predict recurrence risk in venous thromboembolism: looking for a useful threshold: a rebuttal. Journal of Thrombosis & Haemostasis 2004;2(9):1670-2.
3. Cosmi B, Lebnani C, Palareti G. D-dimer testing to predict recurrence risk in venous thromboembolism: looking for a useful threshold: rely to a rebutal. Journal of Thrombosis and Haemostasis 2004;2:1673–5.
4. Ridker PM, Goldhaber SZ, Danielson E, et al. Long-Term, Low-Intensity Warfarin Therapy for the Prevention of Recurrent Venous Thromboembolism. New England Journal of Medicine 2003;348(15):1425-34.
5. Agnelli G, Prandoni P, Santamaria MG, et al. Three Months versus One Year of Oral Anticoagulant Therapy for Idiopathic Deep Venous Thrombosis. New England Journal of Medicine 2001;345(3):165-9.
6. Palareti G, Legnani C, Cosmi B, Guazzaloca G, Pancani C, Coccheri S. Risk of venous thromboembolism recurrence: high negative predictive value of D-dimer performed after oral anticoagulation is stopped. Thrombosis & Haemostasis 2002;87(1):7-12.
7. Eichinger S, Minar E, Bialonczyk C, et al. D-dimer levels and risk of recurrent venous thromboembolism. JAMA 2003;290(8):1071-4.
8. Palareti G, Legnani C, Cosmi B, et al. Predictive value of D-dimer test for recurrent venous thromboembolism after anticoagulation withdrawal in subjects with a previous idiopathic event and in carriers of congenital thrombophilia. Circulation 2003;108(3):313-8.
[...] This is relation to the following summary. [...]