INTRODUCTION

Deep vein thrombosis (DVT) causes thousands of patients to be hospitalized each year, and many more instances of DVT are diagnosed while patients are hospitalized for surgical procedures or medical illness.

Prompt diagnosis of DVT and treatment circumvent the short-term onset of pulmonary embolism and death, the long-term complications of recurring venous thromboembolism and the complications of post-thrombotic syndrome [1].

The objective diagnosis of DVT relies on highly sensitive and specific compression ultrasonography (CUS) or ascending venography. The cost of these modalities and the incidence of negative test findings have led to alternative approaches to diagnosis and decision making in suspected cases of DVT.

These rely on the use of diagnostic information from clinical history, examination and assays to detect D-dimer [2].

Sensitive D-dimer assays have been developed to exclude the diagnosis of DVT and have exhibited great success when used in conjunction with a diagnostic algorithm, including pretest probability scoring and CUS [2,3]. A value of >500 ng/mL has been established by our laboratory as a positive D-dimer result.

In conjunction with a low-to-moderate pretest probability, a CUS is ordered. For those patients with a high pretest probability, a CUS is ordered without the D-dimer assay being done.

With the combination of pretest probability and D-dimer assay, both a sensitivity and negative predictive value greater than >96 % has been established in the literature [3,4] and has been duplicated by our laboratory.

The object of this study, approved by the Kaiser Permanente (KP) Southern California Institutional Review Board, was to test whether the positive predictive value, specificity and Bayesian positive probability of the D-dimer assay were improved by the use of assay T in place of assay B previously used in our laboratory.

We also hoped to discover whether improvement in sensitivity, negative predictive value and Bayesian negative probability occurred when using the new latex-enhanced immunoassay. In addition, we believe that our data will provide a clear understanding of whether assay T enabled a more efficient use of CUS than assay B did.

METHODS

We retrospectively reviewed 1,015 continuous cases for outpatients suspected of DVT who were seen in the KP Fontana Medical Center in Fontana, California, between January 2007 and April 2008. We also included in our data a 3-month follow-up review of all patients to determine whether any patient presented with a delayed positive DVT.

The 3-month follow-up assessment involved chart or electronic medical record review for all 1,015 patients. We excluded inpatients and residents of skilled nursing facilities for whom the false positive rate of the D-dimer was markedly high. The excluded patient population included elderly patients, pregnant patients and those with cancer or autoimmune diseases.

The patients reviewed for this study were referred from primary care clinics, the Emergency Department, surgery clinics and the Obstetrics/Gynecology Department for DVT.

For all 1,015 patients we used a diagnostic algorithm including pretest probability scoring to assess the clinical likelihood of DVT; those with low-to-moderate probability scores were deemed candidates for D-dimer analysis.

Of the 1,015 continuous outpatients suspected of having DVT, 503 patients were tested with assay B and 512 were tested with assay T. The cost per D-dimer test was the same for both assays.

To determine the clinical performance of both assays, we used six statistical assessments: sensitivity, negative predictive value, specificity, positive predictive value and Bayes' theorem for both positive and negative post-test DVT probability [4,5].

RESULTS

Table I shows the parameters used for evaluating the clinical performance of the D-dimer assays used to exclude the diagnosis of DVT.

The 503 patients for whom assay B was used yielded negative findings on 196 D-dimer assays, false negative findings on one assay and positive assays of which 27 patients had DVT confirmed by a positive-findings CUS.

We used these data to calculate sensitivity, negative predictive value, specificity and positive predictive value.

In addition, these four parameters helped us use a Bayesian statistical analysis that estimates the probability of a hypothesis when pretest odds and likelihood ratio are known values [6].

This Bayesian analysis gives us a probability estimate for the presence of DVT when a patient presents with either a positive or a negative D-dimer assay.

TABLE I: Parameters used for clinical performance evaluation

Assay B yielded a sensitivity of 96.4 %, a negative predictive value of 99.5 %, a specificity of 41.3 % and a positive predictive value of 8.8 %.

Using a Bayesian analysis, after a negative assay B result we can estimate the probability of DVT to be 1.8 %; after a positive assay B result we can estimate the probability of DVT to be 25 %.

Using a Bayesian analysis, we can estimate that after negative findings with assay B, the probability of DVT will be 1.8 %; we can estimate the probability after positive findings with assay B to be 25 %.

For the 512 patients for whom assay T was used, there were 316 negative D-dimer results, one false negative result and 195 positive results; 39 of those latter patients had DVT confirmed by positive results on CUS. We used these data to calculate sensitivity, negative predictive value, specificity and positive predictive value.

In addition, these four parameters helped us use a Bayesian statistical analysis that estimates the probability of a hypothesis when pretest odds and likelihood ratio are known values [4]. This gave us a probability estimate for the presence of DVT when a patient presented with either a positive or negative result on a D-dimer assay.

Assay T yielded a sensitivity of 97.5 %, a negative predictive value of 99.7 %, a specificity of 66.9 % and a positive predictive value of 20.0 %. Using a Bayesian analysis, we estimated the probability of DVT after negative findings with assay T to be 0.8 % and after positive findings to be 38 %.

DISCUSSION

The formation of DVT is normally followed by a physiologic fibrolytic response. As a result of this fibrolytic response, plasmin is generated, which causes the release of fibrin-degradation products into the circulation. Because D-dimer is the predominant form of fibrin-degradation product, the absence of a clinically significant rise in circulatory D-dimer implies that thrombosis is not occurring.

This is why negative results on D-dimer assays have played such an important role in excluding the diagnosis of DVT. It is therefore most strategic to employ a D-dimer assay that has both a high sensitivity and a high negative predictive value. The specificity and positive predictive value have not been emphasized and have historically produced variable results. False positive results have become common when testing for D-dimer.

Our study compared two D-dimer methodologies, postulating that one of them, assay T, would maintain a high sensitivity, an excellent negative predictive value, and Bayesian negative probability while greatly improving specificity, positive predictive value and Bayesian positive probability. Improving specificity, positive predictive value and Bayesian positive probability would also significantly decrease the number of false positive D-dimer results and decrease unneeded usage of CUS.

Our study showed that with assay T, all parameters for clinical performance improved, including sensitivity, negative predictive value, negative probability, specificity, positive predictive value and positive probability (Table I).

Greatly significant for our study, using assay T increased specificity from 41.3 % to 66.9 %, positive predictive value increased a statistically impressive 2.3 times from 8.8 % to 20.0 %, the Bayesian probability of DVT after a positive results on a D-dimer test significantly increased 1.5 times from a 25 % probability to a 38 % probability and the Bayesian probability of DVT decreased a statistically impressive 2.3 times from 1.8 % probability to 0.8 % probability (Table I).

In addition to the improved parameters for clinical performance, a significant number of unneeded CUSs were saved. The number of false positive D-dimer results for assay B was 279/503; for assay T it was 156/512. This was a decrease of 123 unneeded CUSs over a 6-month period.

With the billable cost for each CUS being USD 315 [7] and our cost per D-dimer test being the same for both assay B and assay T, we calculated that a potential 6-month CUS savings of USD 38,745 or a more significant annual CUS savings of USD 77,490 could be realized by using assay T [4,7].

Future investigations for D-dimer research might include examining a wider patient source including inpatients and those suspected of having pulmonary embolism. Much work is still needed to improve the standardization for D-dimer assays, and newer technologies for assay methods that help establish clot age and the probability of DVT recurrence will require thorough assessment.

ACKNOWLEDGEMENTS

Bill Paringer, Grace Johnston, Bovi Nielsen and Leah Diaz, (Kaiser Permanente Medical Library, Fontana, CA) provided outstanding research assistance for this study.

AUTHOR

Craig M. Nelson, PhD, CLS
Medical Bioethics Director at Kaiser Permanente, South Bay Medical Center in Harbor City, CA, a Clinical Laboratory Scientist and a Lecturer at California State University, Fullerton.
Address:
Craig M. Nelson
10560 Lemon Avenue
Alta Loma, CA 91737

Geary S. Wright, BA, MLT 

Medical Laboratory Technician at the Kaiser Permanente Medical Center Fontana, CA.

Tom R. Silbaugh, CPT
Certified Phlebotomy Technician at the Kaiser Permanente Medical Center, CA

Louis J. Cota, CLS
Department Administrator of the Clinical Laboratory at the Kaiser Permanente Medical Center, Fontana, CA