Excluding deep vein thrombosis safely in primary care
Validation study of a simple diagnostic rule.
From the literature, 16 history findings and physical examination items were selected as potential diagnostic indicators. After standardized history taking and physical examination, all patients were referred to the hospital to undergo D-dimer testing. Finally, repeated compression ultrasonography of the symptomatic leg was used as a reference test for all patients. D-dimer level was measured by either the ELISA method (VIDAS, Biomerieux, France) or a latex assay method (Tinaquant, Roche, Germany), depending on the lab routine of the participating hospital.
In an earlier study, the optimal thresholds of the D-dimer tests were determined; the test was considered abnormal if the latex assay yielded a D-dimer level ≥400 ng/mL (Tinaquant) or ≥500 ng/mL for the ELISA assay (VIDAS).11 DVT was considered present if (one of) the deep veins of the legs were not or not completely compressible, as determined with a 5–7.5 MHz linear-array sonographic scanner (system VGE/Sonotion).12 In patients with a normal ultrasound, the test was repeated within 7 days to exclude DVT.4 All patients with a positive ultrasound were treated with anticoagulants.
Validation study
After completing the data sampling of the derivation study, we began collecting data on the next 532 consecutive patients visiting a general practitioner with symptoms suggestive of DVT. The validation study was conducted between June 1, 2003, and June 1, 2005, among the same 110 general practitioners who participated in the derivation study. The protocol of the validation study was similar to that of the original study, using the same inclusion and exclusion criteria, D-dimer assays, and definition of presence and absence of DVT.
The 7 items in the rule were obtained from the standardized history taking and physical examination, and D-dimer testing and ultrasonography were performed in the hospital.
The study protocol was approved by the Medical Ethical Committee of the University Medical Centre Utrecht, and informed consent was obtained from all patients.
Statistical analysis
To quantify external validity, we calculated each patient’s total score using the rule (FORMULA and TABLE 1). First, the overall ability of the rule to discriminate between patients with and without DVT was assessed by the ROC area. Perfect discrimination is represented by an ROC area of 1.0; an ROC area of 0.5 equals the discrimination of a coin flip.13
All patients were assigned to 1 of 4 risk groups, based on total points scored on the rule (TABLE 2). Patients were again considered at very low risk if they received 3 points or less. For this threshold, we calculated corresponding sensitivity, specificity, negative predicted value, and likelihood ratio of a negative test result (negative likelihood ratio=[1–sensitivity]/specificity) with their 95% CIs. Because ruling out DVT is the main purpose of applying the rule, the positive predictive value and the likelihood ratio for a positive test result are not presented.
One hundred fifty-three of the 532 subjects had missing values for one or more predictors in the rule. Missing values ranged from 1.3% in gender to 11.8% for D-dimer test result (1.5% in oral contraceptive use, 2.3% trauma, 7.0% calf difference ≥3 cm, 7.5% presence of malignancy, 8.3% recent surgery, 11.7% vein distension). Data seldom are missing completely at random. Deleting subjects with a missing value not only leads to a loss of statistical power, but also to biased results. Therefore, imputing missing values is generally preferred to complete case analysis.14,15 Missing data were thus (single) imputed, using the linear regression method available in SPSS version 12.0.1 (SPSS, Inc, Chicago, Ill, USA). For comparison purposes, a complete case analysis was also performed.
TABLE 2
Prevalence of DVT across the 4 risk categories (with diagnostic accuracy measures)
| DVT PRESENT N (%) | DVT ABSENT N (%) | PATIENTS N (%) | SN % (95% CI) | SP % (95% CI) | PV– (95% CI) | LR– (95% CI) | |
|---|---|---|---|---|---|---|---|
| Validation study (n=532) | |||||||
| Very low (0–3) | 0 (0) | 112 (100) | 112 (21) | 100 (98.6–100) | 25.7 (21.6–29.8) | 100 (98.8–100) | 0.0 (0.00–0.32) |
| Low (4–6) | 2 (6.3) | 30 (93.8) | 32 (6) | 97.9 (95.1–100) | 32.6 (28.1–37.0) | 98.6 (96.7–100) | 0.06 (0.00–0.53) |
| Moderate (7–9) | 53 (19.2) | 223 (80.8) | 276 (52) | 42.7 (32.8–52.6) | 83.7 (80.2–87.2) | 86.9 (83.7–90.1) | 0.68 (0.57–0.82) |
| High (10–13) | 41 (36.6) | 71 (63.4) | 112 (21) | ||||
| Derivation study (n=1295) | |||||||
| Very low (0–3) | 2 (0.7) | 291 (93.3) | 293 (23) | 99.3 (98.4–100) | 28.9 (26.1–31.7) | 99.3 (98.4–100) | 0.02 (0.00–0.10) |
| Low (4–6) | 3 (4.5) | 63 (95.5) | 66 (5) | ||||
| Moderate (7–9) | 144 (21.7) | 519 (78.3) | 663 (51) | ||||
| High (10–13) | 140 (51.3) | 133 (48.7) | 273 (21) | ||||
| DVT, deep vein thrombosis; SN, sensitivity; SP, specificity; PV–, negative predictive value; LR–, negative likelihood ratio. | |||||||
Results
The frequencies of the diagnostic indicators in the validation (and derivation) set are presented in TABLE 1. Patient age varied between 18 and 98 years (mean±standard deviation: 60±17 years), and 40% of the patients were male. The distribution of variables in the validation and derivation set were comparable, except for the prevalence of malignancy (3% and 6%, respectively) and DVT (18% and 22%, respectively).
