Excluding deep vein thrombosis safely in primary care
Validation study of a simple diagnostic rule.
- Our validated primary care rule safely excludes deep vein thrombosis in one quarter of patients suspected of having the disease (A). We recommend the use of this rule by family practitioners.
Most primary care patients with suspected deep vein thrombosis (DVT)—even if suspicion is low—are referred for burdensome and costly tests such as ultrasonography of the legs or venography. How many of these patients end up having DVT? About 25%.1-3
Clearly there is a need for a clinical tool that can help distinguish patients with low risk of DVT from those with high risk. In a previous study, we developed and internally validated a simple diagnostic prediction rule for DVT that includes 7 patient characteristics and the result of a D-dimer test (so-called derivation study).4 In the devivation data set, this rule showed good performance and could safely exclude the presence of DVT in about one quarter of patients, minimizing the number of unnecessary patient referrals.
Prediction rules commonly show good performance with the data from which they were developed, even when bootstrapping techniques are applied to correct them for overoptimism (internal validation). But good accuracy with the development data set is no guarantee the rule will be accurate for future patients.5-10 Testing the accuracy of a prediction rule in new patients is necessary before implementing it in daily patient care—ie, so-called external validation or generalizability. Promising results on such validation studies increase the confidence for applying rules in practice.8,9
The aim of this study was to quantify (externally validate) the accuracy of our previously derived diagnostic rule for DVT in a large sample of new primary care patients suspected of having DVT, thus testing the rule’s ability to safely exclude DVT.
The rule
The diagnostic rule validated in the present analysis, has been derived in a previous study by using multivariable logistic regression analysis, which identified independent diagnostic indicators of DVT: male gender, 6 items from the history and physical examination, and the result of a D-dimer test.4 Combining these items in a prediction rule, we reached the optimal diagnostic accuracy for the diagnosis of DVT (TABLE 1). The formula of the diagnostic rule:
(1×male gender)+(1×oral contraceptive use)+(1×presence of malignancy)+(1×recent surgery)+(1×absence of trauma)+(1×vein distension)+(2×calf difference ≥ 3 cm)+(6×abnormal D-dimer test result)
The numerical value associated with each indicator represents the weight of that indicator. Each indicator is assigned the value 1 if present, and 0 if absent. For example, a man without leg trauma, with a history of malignancy, and a normal D-dimer test result receives a score of (1×1)+(1×0)+(1×1)+(1×0)+(1×1)+(1×0)+(2×0)+(6×0)=3 points. In the derivation study, the area under the receiver operating characteristic (ROC) curve of the rule, after adjustment for overoptimism using bootstrapping, was 0.78 (95% confidence interval [CI], 0.75–0.81).
To enhance the clinical usefulness of the rule, total score results were combined into different categories: very low risk (score 0–3, DVT prevalence 0.7%), low risk (score 4–6, DVT prevalence 4.5%), moderate risk (score 7–9, DVT prevalence 21.7%), and high risk (score 10–13, DVT prevalence 54.3%). Using a score threshold ≤3 (very low risk) and retaining these patients in primary care would result in a 23% reduction in referrals, at the cost of only 0.7% missed DVT cases. The man in the example above has a score of 3 points. The primary care physician could decide not to refer this patient, with a risk of just 0.7% of missing a DVT.
TABLE 1
Distribution of the predictors in the rule in the validation and derivation study set
| PREDICTORS OF DVT | POINTS ATTRIBUTED TO PREDICTOR | PREVALENCE (%) | |
|---|---|---|---|
| VALIDATION SET (N=532) | DERIVATION SET (N=1295)4 | ||
| Male gender | 1 | 40 | 36 |
| Oral contraceptive use | 1 | 9 | 10 |
| Presence of malignancy | 1 | 3 | 6 |
| Recent surgery | 1 | 13 | 14 |
| Absence of leg trauma | 1 | 83 | 85 |
| Vein distension | 1 | 18 | 20 |
| Calf difference ≥3 cm | 2 | 41 | 43 |
| D-dimer abnormal* | 6 | 75 | 69 |
| DVT Presence | 18 | 22 | |
| *D-dimer was considered abnormal if the concentration was ≥500 ng/mL with the Vidas assay and ≥400 ng/mL with the Tinaquant assay. | |||
Patients and methods
Derivation study
The diagnostic rule was derived from 1295 consecutive patients consulting their primary care physician with symptoms suggestive of DVT. The study was performed among 110 primary care physicians affiliated with 3 nonacademic hospitals. The characteristics of the derivation study and the rule are described in detail elsewhere.4
In short, suspicion of DVT was based on the presence of at least 1 of the following signs and symptoms: swelling, redness, or pain of the lower extremities. Patients were included if the primary care physician decided that the diagnosis of DVT should be confirmed or excluded by objective diagnostic testing (ultrasonography) in the hospital. Exclusion criteria were symptoms or signs existing for more than 30 days or suspicion of pulmonary embolism.
