Designer medicine for suspected DVT: the 4D study. St Emlyn’s

Do you want to halve your ultrasound referral rate for suspected Deep Vein Thrombosis?

You do? How marvellous. Reducing referral rates from ED for a specialist test like vascular sonography can have multiple benefits for organisations and patients, including cost savings, reduced waiting times, reduced incidental diagnoses, less risk of iatrogenic infection (from multiple hospital visits) and increased convenience. However – there are usually some caveats to less testing; less early pick up of developing pathology, less information to  explain symptomatology and less reassurance for patients. For these reasons, we need to be very confident in a strategy that recommends less testing. 

Enter the 4D (Designer D-Dimer DVT Diagnosis) algorithm, essentially a d-dimer threshold for exclusion of DVT adjusted to initial pretest probability. A low risk Wells score = an automatic threshold of 1000ng/ml (remember this will need potential adjustment dependent on your assay, reporting units and conventional threshold – see here). The strategy also incorporates variable thresholds for recommending repeat proximal songpraphy at 1 week. The abstract headline reports a substantial reduction in the need for ultrasound imaging with this strategy.

We have seen similar work from the same team previously with regard to investigation of pulmonary embolism.1 The concept certainly has face validity and results for exclusion of PE were very promising. Like me, I am sure you have all been eagerly waiting to see how the strategy performs in patients attending the Emergency Department with suspected DVT. Well, I am pleased to report that not only do we now have attempted validation of this algorithm, but it has been delivered by the same team using the extensive CanVECTOR thrombosis research network and published in the BMJ this week. Good stuff. As always, we would highly recommend that you read this article and critically appraise it yourself, taking into account your current practice and challenges. But the abstract is below and here is our hot take. 

Kearon C, de Wit K, Parpia S, Schulman S, SpencerF A, Sharma S et al. Diagnosis of deep vein thrombosis with D-dimer adjusted to clinical probability: prospective diagnostic management study BMJ  2022;  376 :e067378 doi:10.1136/bmj-2021-067378

What type of study is this?

This is a prospective management (or implementation) study across 10 centres in Canada. Patients presenting to the emergency department or outpatient clinics with symptoms or signs of DVT were potentially eligible. Consent is interesting – in some centres patients provided written informed consent prior to diagnostic testing, in other centres patients underwent diagnostic testing according to the novel study algorithm, but were then approached for deferred consent within days. All sites had formal research ethics board approval. The only intervention studied was the novel diagnostic algorithm, otherwise all care was conducted in accordance with local practice/standards. 

Patients were followed up at 90 days by telephone, following initial diagnostic evaluation. The primary outcome was symptomatic, objectively verified, venous thromboembolism. All outcomes were adjudicated centrally, blinded to diagnostic work up and treatment. 

Sample size was calculated at 1500 patients, based on an expected false negative algorithm rate of approximately 0.6% and a high power to exclude VTE at an incidence of <2%. 

Who were the patients?

Patients with symptoms or signs of DVT. This is a fairly pragmatic and realistic inclusion. The authors excluded the typical groups, such as those patients receiving anticoagulation for >24h prior to diagnostic testing, those under 18 years of age and all pregnant patients. They also excluded inpatients and those with previous DVT (to avoid chronic thrombi masquerading as positive results, which has implications for generalisability) and those on any form of long term anticoagulation, or palliative pathways. 

Explain the algorithm to me again?

The key acute change in their diagnostic algorithm was to use the original 3 level Wells score and implement a higher fixed cut point (>1000ng/mL) for a positive d-dimer result in those patients identified as low risk. Patients at moderate risk used a cut point of >500ng/mL and all high risk patients were referred for vascular compression ultrasound (CUS), as per standard practice. It’s worth highlighting here that current NICE guidance in the UK uses a dichotomised Wells score of unlikely (<2) and likely (>/=2); this evidence cannot therefore be immediately generalised to ‘unlikely’ patients, as you need a Wells score of <1 (not <2) to be classed as low risk using the three part score. 

The authors had a second intervention in the algorithm, of raising the bar fairly substantially for a repeat CUS in the event of a negative test through implementation of a higher d-dimer cut point. In this study, following an initial negative CUS, you got a repeat CUS in a week if you were low/moderate risk Wells and had a d-dimer </=3000ng/mL or if you were high risk Wells and had a d-dimer </=1500ng/mL. Again, this is very different to what NICE currently recommends, which is that only patients who are high risk Wells and have a positive d-dimer require a repeat CUS after a week.2

What were the results?

Over 6 years, participating centres assessed 3726 potentially eligible patients. 1894 met exlcusions and 309 did not provide consent. Most of the exclusions were around protocol violation; 223 patients had a CUS prior to any assessment of clinical pretest probability, 385 patients had a CUS despite a negative d-dimer (by algorthim definition) and 206 had already had a CUS prior to d-dimer estimation. Approximately 1:7 patients evaluated had a previously confirmed episode of DVT; all these patients were excluded from the study and this is an important point to note. A further 15 were considered ineligible by the central data monitoring team. The authors analysed 1508 patients.

Baseline demographics were fairly typical and in keeping with other VTE diagnostic studies, such as a mean age of 60 and median duration of symptoms for 7 days. Overall pretest probability was 11% (173 patients had a positive CUS using the algorithm), although we need to remember that this study used proximal CUS only and sonography for evaluation of isolated distal DVT was actively discouraged. 

Regarding the primary outcome, 8 patients out of 1275 who had DVT excluded using the novel 4D algorithm were diagnosed with acute VTE during 90 day follow up, for an event rate of 0.6% (95% CI 0.3 to 1.2). It looks from the supplementary data like at least 3 of these events were pulmonary emboli. It’s hard to tell what type of DVTs were diagnosed and days to diagnosis ranged from 2 to 105 (investigations were commenced within 90 days). There were 18 deaths during the study period; no deaths were attributed to VTE events. 

Of more direct relevance to Emergency Medicine, 529/1508 (35%) of patients in this study were identified as low risk using the Wells score. 377 of these patients had a d-dimer result <1000ng/mL and received no further diagnostic assessment. 3 patients were subsequently lost to follow up in this group. A single patient was diagnosed with a VTE within 90 days, for an event rate of 0.3% (95% CI 0.1 to 1.5%). This patient was diagnosed with a DVT at 39 days, having had an original d-dimer level well below even the conventional cut point (400ng/mL). 

What about sonography? In the low risk Wells group, use of the 4D strategy resulted in a reduction of 97 scans during the study period, or a 39% reduction in referral for imaging. It gets a bit murky after that – the authors present their analysis as a mean number of CUS examinations per patient and report an overall difference of -0.64 (95% CI -0.68 to -0.60) using a 4D strategy. This is transparent, but it assumes several aspects of the conventional strategy that are not generalisable to UK practice. First, that everyone at moderate risk of DVT by Wells criteria is referred for CUS. Second, that the majority of patients at moderate or high risk by Wells criteria get a repeat CUS in a weeks time. Recently updated NICE guidance differs fairly significantly from this protocol, by dichotomising the Wells criteria to unlikely (which incorporates some patients at moderate risk) and likely, and by only recommending repeat CUS in patients at likely risk of DVT with a raised D-dimer. 

Are there any limitations here?

The above issues around lack of generalisability to UK practice raise some concerns about how we interpret these findings. In addition, the exclusion of patients with previously diagnosed DVT is an alarm bell – this is a specific criterion within the Wells score. By excluding these patients the authors have also therefore excluded the cohort of patients that could be low risk by Wells with a previous DVT (1 point for previous, -2 points for other diagnosis more likely). This is another concern about generalisability and further study of the algorithm is needed in these patients, which the authors highlight. I would also like to know a bit more about the previous DVT exclusions – are these provoked clots within the last year or >5 years ago? 

These findings do not apply to inpatients, pregnant patients, children or those patients on any form of anticoagulation. Also, the algorithm does not evaluate patient orientated outcomes (as the authors highlight) and assumes that neither clinicians or patients have concerns about the diagnosis of isolated distal DVT. This is always very challenging to unpick. I have an interest in this area and so am likely to present a biased viewpoint, but I would suggest that the recent Cochrane review and previous data suggest a potential benefit to identifying and treating these clots early.3-5 Indeed, during the initial CUS work up within this study 8 cases of isolated distal DVT were diagnosed, in spite of specific study advice to avoid imaging this area. This suggests to me that clinicians and sonographers still care very much about this disease. Whereas the 4D algorithm assumes the opposite. 

Is this ready for implementation?

This is an excellent study, conducted by a world class team and reported transparently. It is a brilliant addition to the literature and the authors should be thoroughly congratulated. But whether it is ready for implementation I think depends on your current practice.

For example, this will be hard to deliver as a full algorithm in the UK. Many centres already use whole leg CUS in attempt to avoid repeat imaging/attendance and to identify isolated distal DVT early, with very low rates of repeat CUS, return or complication.6 It is uncertain whether a change to the 4D algorithm strategy would outperform this current practice. Using a 3 level Wells score when NICE continues to recommend a 2 level simplified version would also add confusion, as would the use of different algorithms for inpatients, outpatients, pregnant patients and those with previous DVT. Finally, the authors report use of variable assays within this study, some of which were initially point of care. Whenever you decide to change your algorithm, it’s essential to understand how the assay evaluated in the evidence compares to your current standard. NICE guidelines highlight the limitations in evidence supporting point of care D-dimer assays, and the variation in manufacturer and reporting means everyone should probably consider an evaluation of novel algorithm use in their population, prior to widespread implementation.

Any final comments?

The concerns above do not constrain this research. In particular for me, I am very interested to see the additional efficiency created through use of an elevated cut point in low risk patients. This fits with the trend of practice; we have been using age adjusted D-dimers for several years locally to good effect7, and I think we could add the 4D strategy into the mix now to perhaps refine referral for CUS even further in low risk groups. And I think this work is crying out for validation in an unlikely/likely population, to evaluate the potential impact and generalisability. 

Interesting stuff, and a real pleasure to read such a well crafted manuscript. Hats off (again) to the CanVECTOR network and in particular to Virchester alumnus Kerstin de Wit who is racing to international fame and fortune with a paper in the NEJM and the BMJ within the last few years. It’s hugely impressive work. 

Best wishes



1. Kearon C, de Wit K, Parpia S, et al. Diagnosis of Pulmonary Embolism with d-Dimer Adjusted to Clinical ProbabilityN Engl J Med 2019;381(22):2125-34. doi: 10.1056/NEJMoa1909159 [published Online First: 2019/11/28]

2. McCormack T, Harrisingh MC, Horner D, et al. Venous thromboembolism in adults: summary of updated NICE guidance on diagnosis, management, and thrombophilia testingBMJ 2020;369:m1565. doi: 10.1136/bmj.m1565 [published Online First: 2020/05/21]

3. Horner D, Hogg K, Body R. Should we be looking for and treating isolated calf vein thrombosis? Emerg Med J 2016;33(6):431-7. doi: 10.1136/emermed-2014-204230 [published Online First: 2015/06/24]

4. Horner D, Hogg K, Body R, et al. The anticoagulation of calf thrombosis (ACT) project: results from the randomized controlled external pilot trialChest 2014;146(6):1468-77. doi: 10.1378/chest.14-0235 [published Online First: 2014/07/11]

5. Kirkilesis G, Kakkos SK, Bicknell C, et al. Treatment of distal deep vein thrombosis. Cochrane Database Syst Rev 2020;4:CD013422. doi: 10.1002/14651858.CD013422.pub2 [published Online First: 2020/04/10]

6. Horner D, Hogg K, Body R, et al. Single whole-leg compression ultrasound for exclusion of deep vein thrombosis in symptomatic ambulatory patients: a prospective observational cohort studyBr J Haematol 2014;164(3):422-30. doi: 10.1111/bjh.12642 [published Online First: 2013/11/06]

7. Takach Lapner S, Stevens SM, Woller SC, et al. Age-adjusted versus clinical probability-adjusted D-dimer to exclude pulmonary embolismThromb Res 2018;167:15-19. doi: 10.1016/j.thromres.2018.05.003 [published Online First: 2018/05/14]

Thanks so much for following. Viva la #FOAMed

Scroll to Top