Duplex ultrasound for the diagnosis of symptomatic deep vein thrombosis in the lower limb

  • Protocol
  • Diagnostic

Authors


Abstract

This is the protocol for a review and there is no abstract. The objectives are as follows:

To estimate the sensitivity and specificity of duplex ultrasound for the detection of distal and proximal DVT in symptomatic patients with prior testing by a clinical prediction rule (with or without additional D-dimer testing).

The accuracy of ultrasound for DVT may be affected by body mass index and whether the patient has had a previous DVT. If possible, we shall investigate both previous DVT and body mass index as possible sources of heterogeneity, but recognise that these are patient-specific rather than study-specific characteristics. This means that results reported at the study level, for example average body mass index, may not be informative in an analysis and are more appropriately investigated with individual patient data. However, such an analysis may be possible if studies report results stratified by previous DVT status or body mass index. Two study-specific characteristics we shall investigate are the generation of the technology of the ultrasound scanner and type of reference standard: ascending venography, CT venography, or MR venography. Moreover, as the accuracy of duplex ultrasound depends on whether the DVT is distal or proximal, we shall perform separate meta-analyses according to site.

Background

Venous thromboembolism (VTE) is a single disease affecting the venous circulation. It has two distinct presentations, deep vein thrombosis (DVT) and pulmonary embolism (PE) (Anand 1998). A DVT is a venous thrombosis (blood clot) commonly found in the deep veins of the lower limb or pelvis. Thromboses may break off, or embolise, and travel through the veins to other parts of the body. If a thrombosis reaches the lungs it can cause PE by blocking one of the pulmonary arteries. DVT is more common than PE, affecting around 70,000 people each year in the UK, and it is potentially fatal if left untreated.

There are many risk factors associated with VTE but the incidence of the condition rises dramatically in people over the age of 60 years (White 2003). Major surgery, trauma, fractures or surgery to the hip or knee, malignancy, cardiac or respiratory failure, prolonged immobility, pregnancy, oestrogen therapy, previous DVT or PE, and obesity are all known to increase the likelihood of VTE (Anderson 2003). Affected individuals may have no symptoms but have a history of one or more risk factors, which creates a suspicion that a DVT is present, with subsequent testing confirming the presence of the blood clot (Goodacre 2006). Those with symptoms may have calf pain and swelling.

In general, symptomatic DVT is usually found proximally in the large veins of the thigh whereas non-symptomatic DVT is found more distally in the smaller veins of the calf. DVT in the large proximal veins is associated with poorer patient outcomes as these thromboses are more likely to embolise to the lungs and result in fatalities (NCGC 2012). Uncertainty exists about the most accurate method to diagnose DVT and this creates a great deal of variation in clinical practice (Goodacre 2006). Non-invasive tests to diagnose DVT include tests to detect D-dimer in plasma and duplex ultrasound. Clinical algorithms (clinical prediction rules) such as the Wells test are also recommended in clinical guidelines and are an important part of the patient pathway in routine clinical practice (NCGC 2012).

A systematic review of the clinical and cost effectiveness of non-invasive tests for DVT found that substantial heterogeneity was present in the studies included in the meta-analyses. The authors used meta-regression techniques to investigate sources of heterogeneity, but missing data prevented the analysis of covariates that could have explained some of this heterogeneity. The reviewers concluded that the observed heterogeneity is likely to have arisen from unreported differences between patient groups and the clinical probability of DVT (Goodacre 2006).

Non-invasive tests are generally considered to be less accurate than ascending venography, which has traditionally been the reference standard test for DVT (Goodacre 2006). For venography, the patient receives an injection of a contrast agent in one of the veins of the foot and then undergoes an x-ray. Venography allows the proximal and distal venous system to be clearly visualised but it is invasive, requires the use of an intravenous contrast agent, and delivers a dose of radiation. It is therefore contraindicated for certain groups of people such as pregnant women and those with renal failure, or an allergy to contrast. A further factor limiting availability is that its accuracy depends on the skill of the person conducting the test, and ascending venography should only be done by skilled radiologists with appropriate training, or similar. As ascending venography is not always readily available, other reference standards are also used.

Other acceptable reference standards include computed tomography (CT) venography and magnetic resonance (MR) venography. CT venography requires the administration of a nephrotoxic contrast agent to the patient who then undergoes CT scanning to produce images of the pelvic and lower limb veins. The CT scan uses x-rays and therefore delivers a dose of radiation to the patient. Like ascending venography, CT venography cannot be used on patients who are pregnant, have renal failure, or a contrast allergy. MR venography may or may not involve the use of a contrast agent but does not use additional radiation. Instead, MR imaging uses the magnetic properties of water particles in the body to form images. However, MR venography will not be suitable for patients who have metallic implants or suffer from claustrophobia. It is also not recommended for pregnant patients. If a contrast agent is used, then MR venography will also be contraindicated for patients with renal failure or a contrast allergy.

DVT is treated with anticoagulant therapy, which can be given for a period of between six weeks to six months during which time the patients require regular blood chemistry monitoring (Cundiff 2006; Watson 2004). The purpose of the treatment is to prevent recurrent episodes of DVT, post-thrombotic syndrome and the development of a PE. The accurate diagnosis of DVT is essential to avoid morbidity and mortality, and this review forms part of a programme of high priority systematic reviews in the diagnosis and treatment of peripheral vascular disease that is funded by the National Institute of Health Research (NIHR).

This protocol is complementary to the Cochrane protocol on the diagnosis of DVT in the upper extremity by ultrasound (Di Nisio 2011), however there is no overlap between the two protocols as they are concerned with different limbs.

Target condition being diagnosed

Deep venous thrombosis (DVT) of both proximal (thigh) and distal (calf) veins.

Index test(s)

Duplex ultrasound is a non-invasive test which measures the flow of blood in the arteries and veins, and also allows the blood flow to be seen in real-time. It is considered to be the standard diagnostic test for clinically suspected DVT and is inexpensive but time-consuming to perform (Sampson 2005). Duplex ultrasound combines greyscale ultrasound with another Doppler ultrasound modality.

Traditional or greyscale ultrasound produces images that reveal the structure of the veins as two-dimensional (2-D) images. Other ultrasound modalities combined with greyscale ultrasound in the diagnosis of DVT include pulsed wave Doppler, colour Doppler ultrasound and power Doppler.

Pulsed wave Doppler ultrasound measures the velocity of blood flow and other flow features, and provides a waveform showing various attributes of the blood flow. Both colour and power Doppler are used to detect the patency of blood vessels. Colour Doppler ultrasound adds blue or red on top of the greyscale image to indicate the presence and direction of blood flow. Power Doppler is a version of colour Doppler that overcomes some of the technical difficulties that sometimes occur with directional colour Doppler, for example in distinguishing slow blood flow from background noise, but does not provide information on the direction of blood flow. Modern ultrasound scanners generally have all these modalities and the person testing for DVT may switch between modalities, as appropriate, to make a diagnosis.

The utility of a particular method will depend on the anatomy of the patient and the site of the DVT. Distal DVTs in the calf can be harder to image than proximal DVTs as the veins are smaller and more difficult to identify and image. If a DVT is above the groin in the pelvis, there may be bowel gas blocking the ultrasound beam. In such cases pulsed wave Doppler ultrasound at the groin may be more informative than the other kinds of duplex ultrasound.

The primary assessment of DVT diagnosis with ultrasound includes the use of vein compression. The person conducting the examination uses the ultrasound transducer to image the vein in the standard way with greyscale, and then tries to compress the vein with the transducer. Veins are normally highly compressible, and so any resistance to compression indicates the presence of a clot.

To enhance flow in the veins, manual compression of the calf muscle may also be used with colour Doppler ultrasound to help diagnose the presence of DVT. In this case the calf muscle is held and compressed with one hand while the transducer is used with the other hand to visualise the movement of blood flow, for example, no colour would indicate that a thrombus has occluded the vein.

There are, therefore, no explicit threshold criteria that are used when assessing the presence or absence of a DVT, as it is a judgement made by the person performing the ultrasound examination. They either see evidence for a DVT or they do not. A systematic review of the accuracy of ultrasound for detecting any DVT estimated the sensitivity to be 89.7% (95% confidence interval (CI) 88.8 to 90.5) and the specificity to be 93.8% (95% CI 93.1 to 94.4). However, there was significant heterogeneity and the accuracy of ultrasound varied according to whether DVT included both proximal and distal disease or not, and the type of ultrasound. These data were collected in 99 studies from patients who were managed in a variety of healthcare settings including as inpatients, outpatients and in hospital emergency departments (Goodacre 2005).

For the purpose of diagnosing DVT, the proximal segment includes the popliteal and more proximal veins, distal disease would involve the calf veins below the popliteal level.

Clinical pathway

Symptomatic patients generally undergo scoring by the DVT Wells test prior to any ultrasound (Wells 2006). The Wells test is a clinical prediction rule that gives the patient a point for each criterion met (for example pitting oedema in the symptomatic leg, previous known DVT), and subtracts two points if an alternative diagnosis is considered just as likely. The maximum score is nine points. Current NHS guidance recommends that patients with a score of 2 or more undergo ultrasound testing within four hours, and those with a score of 1 or less or without access to ultrasound within four hours undergo D-dimer testing (NCGC 2012). Patients with a Wells score of 1 or less but a positive D-dimer test are also recommended to undergo ultrasound within four hours of request.

This NHS guidance has been available for less than one year, and so we do not expect that all studies will have dichotomised the Wells score at 1 or less and 2 or more. For example, some studies will have used three subgroups where a patient is graded as low risk if he or she has a score of 0 or less, moderate risk if the score was 1 or 2, and high risk if the score was 3 or more, as this is the categorisation used by the developers of the Wells score (Wells 1997). In addition, the availability of other clinical prediction rules means that studies may have used other scores to screen for DVT.

Symptomatic patients thought to be at low risk by the clinical prediction rule will generally undergo D-dimer testing. Doctors will look for non-DVT causes of the symptoms if a low risk score is combined with a negative D-dimer test. Patients with a low risk score but a positive D-dimer test will be referred for ultrasound imaging. Patients with a high risk score may not receive a D-dimer test but be referred straight to ultrasound imaging. Furthermore, patients may be offered anticoagulation treatment prior to the ultrasound if rapid access to ultrasound is not available and they are either classed as high risk by the clinical prediction rule or have a positive D-dimer test.

In this review, we will consider studies that use a clinical prediction rule with or without D-dimer testing to screen patients for ultrasound testing in secondary care. Specifically, we will use studies that recruited symptomatic outpatients with the following.

1. A low risk score but a positive D-dimer test.

2. A high risk or medium risk score without a D-dimer test.

3. A high risk or medium risk score with a negative D-dimer test.

4. A high risk or medium risk score with a positive D-dimer test.

The probability of DVT will differ between these four groups. All four groups will have a higher probability of DVT than a group of untested patients with symptoms of DVT. This is because the patients who are least likely to have a DVT, that is those with a low risk score and a negative D-dimer test, are not included in any of the four groups.

If possible, we shall also group patients according to whether they received anticoagulant treatment or not, and the specific clinical prediction rule.

Patients with indeterminate or conflicting test results may undergo repeat imaging, but the review will be restricted to studies investigating the diagnostic accuracy of the initial ultrasound only.

Rationale

Duplex ultrasound is commonly used in the detection of DVT and is non-invasive, convenient and readily available. A non-Cochrane systematic review (search date April 2004) found that ultrasound demonstrated good overall sensitivity (89.7%; 95% CI 88.8 to 90.5) and specificity (93.8%; 95% CI 93.1 to 94.4), but these varied according to ultrasound type and other factors (Goodacre 2006). There is a need for non-invasive tests for DVT given the lack of widespread applicability of ascending venography.

Objectives

To estimate the sensitivity and specificity of duplex ultrasound for the detection of distal and proximal DVT in symptomatic patients with prior testing by a clinical prediction rule (with or without additional D-dimer testing).

Secondary objectives

The accuracy of ultrasound for DVT may be affected by body mass index and whether the patient has had a previous DVT. If possible, we shall investigate both previous DVT and body mass index as possible sources of heterogeneity, but recognise that these are patient-specific rather than study-specific characteristics. This means that results reported at the study level, for example average body mass index, may not be informative in an analysis and are more appropriately investigated with individual patient data. However, such an analysis may be possible if studies report results stratified by previous DVT status or body mass index. Two study-specific characteristics we shall investigate are the generation of the technology of the ultrasound scanner and type of reference standard: ascending venography, CT venography, or MR venography. Moreover, as the accuracy of duplex ultrasound depends on whether the DVT is distal or proximal, we shall perform separate meta-analyses according to site.

Methods

Criteria for considering studies for this review

Types of studies

We will include the following.

Cross-sectional studies evaluating the diagnostic test accuracy of duplex ultrasound. Diagnostic cohort studies including both prospective and retrospective designs.

Diagnostic case-control studies (two-gate design) will be excluded because clinically relevant estimates of sensitivity and specificity can only be derived from the clinical population in question. Diagnostic case-control studies for DVT recruit patients already known to have DVT and compare test results with a group who are known not to have DVT. This second group may comprise healthy controls. Case-control designs are known to overestimate the sensitivity and specificity that a diagnostic test has in clinical practice (Rutjes 2005).

Participants

Participants are patients who present with the symptoms and clinical signs of DVT in the hospital outpatient setting. The specific symptoms are: leg pain and possibly swelling, redness and heat. Other signs may include the presence of pitting oedema and fever (Wells 2006). We will include patients who have been previously tested with a clinical prediction rule (CPR) with or without D-dimer assays and who are classed as ‘likely’ to have a DVT after assessment, that is, who are not considered low-risk patients. We recognise that the index tests are likely to perform differently in patients with different prior test results, and will not combine data from these diverse groups. It will only be possible to include studies where these data are presented separately.

Index tests

The index test is duplex ultrasound. We note that the precise form of duplex ultrasound may vary from study to study, and even from patient to patient within a study, as this is clinical practice. This is also true of duplex ultrasound combined with compression. We are therefore not prespecifying that only particular forms of duplex ultrasound are acceptable, or that compression must be used, but will carefully record and report the ultrasound modalities used in each study contributing to the final review.

Target conditions

There is some evidence that the accuracy of duplex ultrasound is site specific (Goodacre 2006). We will therefore have two target conditions: distal DVT and proximal DVT.

Reference standards

The acceptable reference standards are ascending venography, CT venography, and MR venography. Studies can use any or all of these reference standards. If a study uses ascending venography, CT venography, or MR venography in combination with another reference standard, results must be reported separately for ascending venography, CT venography, and MR venography.

Search methods for identification of studies

There will not be any restrictions in terms of date, language of publication or publication status. No diagnostic methodology search filters will be employed.

Electronic searches

We will search MEDLINE (OvidSP) using the search strategy shown in Appendix 1.

We will search EMBASE (Ovid SP) using the search strategy shown in Appendix 2.

In addition, similar search strategies will be designed for:

  • CINAHL via EBSCOhost;

  • LILACS (BIREME);

  • DARE ( Database of Abstracts of Reviews of Effects) and the Health Technology Assessment Database (HTA) in The Cochrane Library;

  • ISI Conference Proceedings Citation Index - Science;

  • British Library Zetoc Conference Search;

using search terms appropriate for each of these databases and based on the MEDLINE search terms.

We will also search MEDION (www.mediondatabase.nl/) using the 'Systematic Reviews of Diagnostic Studies' search filter.

We will request a search of the Cochrane Register of Diagnostic Test Accuracy Studies for further relevant studies.

The following trial databases will be searched for details of ongoing and unpublished studies:

Searching other resources

We will handsearch the reference lists of the primary studies and reviews identified from the electronic searches.

Data collection and analysis

Selection of studies

One review author (FC) will screen the titles and abstracts retrieved by the electronic searches and a second review author (AA) will check a random sample of 10% of the screened titles and abstracts. We will calculate the proportion of references where the review authors agree. Full papers will be obtained for potentially eligible studies including those identified by non-electronic means. Two review authors (FC and AA) will independently apply the exclusion criteria to the full papers and resolve any disagreements by discussion or refer to a third review author as necessary. We will use a PRISMA flow diagram to show the selection process (PRISMA 2009) and will use Endnote version 4 as our reference management system.

The exclusion criteria for abstracts are: ultrasound is not used, patients do not have symptoms consistent with DVT, study uses a case-control design, study is not a diagnostic test accuracy study. We shall report the number of excluded abstracts in the PRISMA flow diagram. Studies that fulfil the abstract criteria will be assessed with the full text.

Given that the reporting in abstracts is necessarily sparse, the full-text articles will be checked for meeting the abstract criteria and also for fulfilling the following: a 2 x 2 contingency table is either supplied or can be back-calculated; a clinical prediction rule is used; all patients are symptomatic; the reference standard is either ascending venography, CT venography, or MR venography. The number of studies failing these criteria will be reported in the PRISMA flow diagram.

Studies that pass the initial screening of the full-text articles will undergo a more in-depth evaluation as described below, and will be fully reported in the review.

Data extraction and management

Data will be independently extracted by two review authors (FC and AA) using a standard form, which will include an assessment of study quality (Whiting 2011). The review authors will corroborate their data extraction and quality assessment decisions and disagreements will be resolved by discussion or referred to a third review author as necessary. Patient level 2 x 2 contingency tables (true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN)) will be extracted as reported. If necessary they will be back-calculated from estimates of sensitivity, specificity, positive and negative predictive values and the total number of patients.

Data on mortality, adverse events and the number of technical failures for duplex ultrasound will be collected along with prognostic factors from the patient history including the proportion with cancer, recent surgical procedures, anticoagulant treatment, and prolonged periods of immobilisation; and time to ultrasound (mean or median, with range if available). We shall report if studies reached diagnoses through consensus (where more than one person tests each patient). Data for use in analyses on the proportion with previous DVT, average body mass index, site of DVT, and generation of technology will also be extracted. Data will be saved in Microsoft Excel 2010.

We will also record data on the technical aspects of duplex ultrasound and the reference standards, for example, type of scanner, contrast agent, specified MR sequences.

Assessment of methodological quality

We will use the Quality Assessment of Diagnostic Test Accuracy Studies-2 (QUADAS-2) to develop a quality assessment tool incorporating the review question, a flow diagram for the study and an assessment of the risk of bias and applicability judgements (Whiting 2011). Review-specific signalling questions (for example, signalling questions on the expertise of the people conducting the tests), appropriate items concerning the applicability of primary studies relative to the review, together with guidance about rating can be found in additional Table 1. We propose that two review authors working independently will pilot the tool and an assessment of agreement will be calculated. Disagreements will be resolved by discussion.

Table 1. QUADAS-2
Domains, signalling questions (SQ) and applicability questions (AQ) Rating criteria (information used to support risk of bias judgements)
Domain 1: Patient selection 
A. Risk of biasDescribe the methods of patient selection as given in the paper:
SQ1: Was a consecutive or random sample of patients enrolled?

Yes: It is stated that the sample was consecutive or a random sample

No: It is stated that the sample was not consecutive or a random sample

Unclear: The method of sampling is ambiguous

SQ2: Did the study avoid inappropriate exclusions?

Yes: The study included all symptomatic outpatients patients who had been screened with a clinical prediction rule (CPR), with or without further testing such as D-dimer, but not prior ultrasound

No: The study excluded patients who had been assessed using a CPR or who had undergone D-dimer testing or who had symptoms of DVT

Unclear: The study's exclusion criteria allow for inappropriate exclusions

SQ3: Did the study avoid inappropriate inclusions?

Yes: The study included only symptomatic outpatients patients who had been screened with a clinical prediction rule (CPR), with or without further testing such as D-dimer, but not prior ultrasound

No: The study also recruited some patients with prior ultrasound (for example, in cases where previous ultrasound testing had been inconclusive), or hospital inpatients, or asymptomatic patients

Unclear: The study's inclusion criteria allow for inappropriate inclusions

AQ1: Could the selection of patients have introduced bias?

Risk:

High: The patient sample is a non-random sample or a non-consecutive sample such as a convenience sample, or includes asymptomatic patients, hospital inpatients, or patients who have had a prior ultrasound

Low: The paper explicitly states that the sample was consecutive or random, that only symptomatic outpatients were recruited, and that patients were screened with a CPR with or without D-dimer, but not ultrasound

Unclear: The method of sampling is ambiguous

B. Concerns regarding applicabilityGive the paper's description of the inclusion and exclusion criteria, including setting, prior tests, symptoms here:
AQ2: Are there concerns that the included patients and setting do not match the review question?

Concern:

High: The patient population and setting are different from the review question setting with regard to prior testing, previous treatments or the way in which the sample was recruited (not consecutive or a random sample). This can occur when the study uses a mixed group of patients, some of whom do not meet the review criteria. Studies that recruit patients from the inpatient setting or selected patients on the bias of atypical presentation fall into this category.

Low: The study population meets the eligibility criteria exactly as described in the Criteria for considering studies for this review section

Unclear: Not enough information is given about the study population

Domain 2: Index test 
A. Risk of biasGive the paper's description of the duplex ultrasound with or without compression and how it was conducted and interpreted, including the background of the person who carried out the test
SQ4: Were the duplex ultrasound results interpreted without knowledge of the results of the ascending venography?

Yes: It is stated that the duplex ultrasound results were interpreted without knowledge of the results of the ascending venography, or the duplex ultrasound was always given and interpreted before ascending venography

No: The results of the ascending venography were known to the readers of the duplex ultrasound

Unclear: The study does not say that the duplex ultrasound was conducted blind to the results of the ascending venography or that the duplex ultrasound was always carried out prior to the ascending venography

SQ5: Did the person conducting the duplex ultrasound have expertise comparable to a radiologist?

Yes: The person conducting the duplex ultrasound was a radiologist or had extensive experience and training (e.g. ultrasonographer, vascular scientist)

No: The person conducting the duplex ultrasound was not a radiologist and did not have sufficient experience and training

Unclear: The expertise and background of the duplex ultrasound readers is not clear

AQ3: Are there concerns that the duplex ultrasound in its conduct or its interpretation have introduced bias?

Concern:

High: The duplex ultrasound was not read blind to the results of the ascending venography by a person with expertise comparable to a radiologist

Low: The duplex ultrasound was read blind to the results of the ascending venography by a person with expertise comparable to a radiologist

Unclear: The study does not give sufficient information to conclude that the duplex ultrasound was read blind to the ascending venography or by a person with expertise comparable to a radiologist

B. Concerns regarding applicability 
AQ4: Is there concern that the duplex ultrasound in its conduct, or interpretation differ from the review question?

Concern:

High: The duplex ultrasound was not read blind to the results of the ascending venography by a person with expertise comparable to a radiologist

Low: The duplex ultrasound was read blind to the results of the ascending venography by a person with expertise comparable to a radiologist

Unclear: The study does not give sufficient information to conclude that the duplex ultrasound was read blind to the ascending venography or by a person with expertise comparable to a radiologist

Domain 3: Reference standard 
A. Risk of biasGive the paper's description of the reference standard including how it was conducted and interpreted and the background of the reader
SQ5. Is the reference standard likely to classify correctly the target condition?

Yes: The study reports that either ascending venography, CT venography, or MR venography was performed for all participants

No: Not all patients received ascending venography, CT venography, or MR venography. For example, a definite diagnosis may have been decided by a repeated ultrasound scan

Unclear; It is unclear whether all patients underwent either ascending venography, CT venography, or MR venography

SQ6: Were the reference standard test results interpreted without knowledge of the duplex ultrasound results?

Yes: The person classifying the reference standard results was unaware of the duplex ultrasound results

No: The person classifying the reference standard results was aware of the duplex ultrasound results

Unclear: Details about the person classifying the reference standard results are not reported

SQ7: Did the person conducting the reference standard have expertise comparable to a radiologist?

Yes: The person conducting the reference standard was a radiologist, sonographer or vascular scientist with extensive experience in performing duplex ultrasound scans

No: The person conducting the reference standard was not a radiologist, sonographer or vascular scientist and did not have expertise in performing duplex ultrasound

Unclear: The expertise and background of the reference standard readers is not clear

AQ5: Could the reference standard, its conduct, or its interpretation have introduced bias?

Risk:

High: The reference standard was not carried out to the required standard by a person with expertise, or it was carried out with knowledge of the duplex ultrasound results and/or the period for clinical follow-up is longer than 3 months

Low: The reference standard was carried out by a radiologist or a person with sufficient expertise without knowledge of the duplex ultrasound results

Unclear: Whether the reference standard was read blind to the results of the duplex ultrasound, or the expertise of the reader is unknown

B. Concerns regarding applicability 
AQ6: Are there concerns that the target condition does not match the question (as defined by the review question)?

Concern:

Yes: The way the reference standard was conducted meant that a patient with DVT could have been missed, or a patient with no DVT could have been diagnosed with DVT

No: The reference standard was conducted appropriately by the appropriate person

Unclear: The conduct of the reference standard is not clear enough to conclude whether DVT was accurately diagnosed

Domain 4: Flow and timing 
A. Risk of biasDescribe the reasons why any patients recruited into the study did not contribute to the 2x2 table (i.e. patients who did not undergo the reference standard and/or duplex ultrasound) referring to the flow diagram:
SQ8: Was there an appropriate interval between the index test and the reference standard?

Yes: The time interval between ultrasound and reference standard was less than 7 days

No: The time interval between ultrasound and reference standard was more than 7 days

Unclear: The time interval between ultrasound and reference standard was not reported, or reported ambiguously (for example, only the mean or median time is reported, without an indication of the maximum time between tests)

SQ9. Did all patients receive the reference standard?

Yes: All patients received the reference standard regardless of the results of prior tests

No: Patients were referred for the reference standard according to their ultrasound results

Unclear: It is not clear whether all patients received the reference standard

SQ10: Did all the patients receive the same reference standard?

Yes: All patients underwent either ascending venography, or all underwent CT venography, or all underwent MR venography

No: The study includes patients who did not undergo ascending venography, or CT venography, or MR venography, and may have had a repeat ultrasound or follow-up instead

Unclear: The study may have included patients who had a repeat ultrasound or follow-up instead

SQ11: Were all patients included in the final analysis?

Yes: All patients enrolled contributed to the 2x2 table

No: There are some patients enrolled who did not contribute to the 2x2 table

Unclear: It is not clear whether there were patients recruited, but not included in the study report or 2x2 table

SQ12: Did the patients receive anticoagulant therapy between the index test and reference standard?

Yes: Some patients started anti-coagulant therapy between the ultrasound and the reference standard

No: Patients did not start anticoagulant therapy until after both ultrasound and the reference standard tests had been completed

Unclear: It is not reported when patients started any anticoagulant therapy

AQ7: Could the patient flow have introduced bias?

Risk:

Low: All patients recruited underwent both ultrasound and the reference standard in a timely manner and did not start anticoagulant therapy until after both tests had been completed

High: Not all patients recruited received both tests, or the time between tests allowed for disease progression to occur, or patients started anticoagulant therapy between tests

Unclear: The time interval between tests is not clear, the treatment status of the patients is not clear, or whether all patients received both tests is not clear

Since we will exclude case-control studies from our review, we shall not use the QUADAS-2 signalling question "Was a case-control design avoided?" as these studies will not reach the full-text stage, nor is the domain 2 SQ2 relating to duplex ultrasound thresholds relevant.

Statistical analysis and data synthesis

We will present and use the 2 x 2 contingency tables to estimate sensitivity and specificity with confidence intervals for each study. The unit of analysis in studies included in the meta-analyses will be patients rather than lower limbs as the method for calculating within-study variance is often unclear in studies which have not used the patient as the unit of analysis. These estimates will be used to create receiver operator characteristic (ROC) curves and forest plots. If the data are adequate, we will perform a bivariate random-effects model meta-analysis of sensitivity and specificity, with within-study variance modelled as binomial. The programs used to fit the bivariate model must converge. However, convergence is a necessary but not sufficient measure of a model having a satisfactory fit, and further checks are necessary. In particular, the covariance parameter can be poorly estimated even when convergence is achieved. A small number of studies is one but not the only common reason for non-convergence (and we would not attempt to fit the five parameter bivariate model with fewer than six studies). It can also happen when there is a lack of variation in threshold between studies. We note that there are many possible reasons, both statistical and clinical, why we would not undertake a summary receiver operator characteristic (SROC) curve method of meta-analysis and we would want to consider both clinical and statistical sources of heterogeneity.

Although in this context explicit thresholds are not available, we are aware that the data may still be affected by the threshold effect due to differences in the use of equipment and interpretation of imaging by the people conducting the ultrasound tests for DVT. We shall therefore assess the presence or absence of the threshold effect in the studies by examination of ROC plots and careful consideration of the bivariate model diagnostics for the correlation coefficient parameter, as it is this parameter which models the threshold effect in the bivariate model. If there is evidence of a threshold effect, we shall consider presenting SROC curves.

We are focusing on the bivariate model, which provides summary estimates of sensitivity and specificity, rather than the hierarchical summary receiver operator characteristic (HSROC) model (Harbord 2007) which uses threshold and accuracy parameters, as in this context explicit thresholds are not applicable. Separate meta-analyses will be undertaken for each patient group. Patient groups will be defined by their prior clinical prediction rule and D-dimer result, and site of DVT (proximal versus distal).

If the data are not amenable to the bivariate model, we shall conduct univariate meta-analyses, one each for sensitivity and specificity. Performing univariate analyses, one each for sensitivity and specificity, is not always recommended. However, where studies do not exhibit heterogeneity in the threshold parameter, the univariate meta-analyses could be informative and would also allow meta-regression in a situation where use of the bivariate model may not be valid.

There are eight primary meta-analyses, where patients are grouped according to the results of prior testing (see Clinical pathway) and whether the DVT is distal or proximal. All analyses will be performed in R 7.1 (cran.r-project.org) and SAS 9.3 (www.sas.com).

Investigations of heterogeneity

Estimates will be grouped according to all the items listed as potential sources of heterogeneity and presented in forest plots and ROC curves for visual assessment of heterogeneity. Formal investigation of a these potential sources will be undertaken using meta-regression if there are sufficient data and studies. We prespecify these sources: body mass index, previous DVT, generation of technology, and type of reference standard. We note that the first two sources are patient specific rather than study specific, and this may preclude an informative analysis. Each meta-regression will be carried out separately for each patient group by adding the items as covariates to either the bivariate model or univariate models depending on which are used for the main analyses.

If and only if the data for these prespecified analyses are lacking, we may consider analysing other items if the data are available and they are considered to be clinically relevant. We do not list the other items here but may choose items if the results of the data extraction suggest that they could be important. Any such analysis would be post hoc and we would report the results as not prespecified in the protocol with suitable precautions with regard to interpretation.

Sensitivity analyses

If the data are reported in the primary studies, we shall repeat the meta-analyses restricted to studies recruiting only cancer patients and to studies recruiting only recent surgical patients as sensitivity analyses. We will also produce ROC curves and forest plots where studies are grouped according to anticoagulant status and the specific clinical prediction rule.

Assessment of reporting bias

Methods for dealing with publication bias in reviews of diagnostic accuracy studies are relatively underdeveloped. Consequently, we shall interpret our results cautiously, and with an awareness of the likelihood of publication bias. We shall consider using a funnel plot of the log of the diagnostic odds ratio (lnDOR) (Deeks 2005) providing there is low heterogeneity in the lnDOR.

Appendices

Appendix 1. MEDLINE search strategy

1. Thromboembolism/

2. Venous Thromboembolism/

3. Thrombosis/

4. Venous Thrombosis/

5. Upper Extremity Deep Vein Thrombosis/

6. (vein$ adj3 (thromb$ or clot$)).ti,ab.

7. (ven$ adj3 (thromb$ or clot$)).ti,ab.

8. (calf adj3 (thromb$ or clot)).ti,ab.

9. (thigh adj3 (thromb$ or clot)).ti,ab.

10. (proximal adj3 (thromb$ or clot)).ti,ab.

11. (limb adj3 (thromb$ or clot$)).ti,ab.

12. (leg adj3 (thromb$ or clot$)).ti,ab.

13. (DVT or VTE).ti,ab.

14. or/1-13

15. diagnostic imaging/

16. Femoral Vein/us [Ultrasonography]

17. Lower Extremity/us [Ultrasonography]

18. Popliteal Vein/us [Ultrasonography]

19. exp ultrasonography, doppler/

20. ultrasonography/

21. ultrasonics/

22. ultrasound.ti,ab.

23. ultrasonogra$.ti,ab.

24. ultrasonic$.ti,ab.

25. echograph$.ti,ab.

26. echotomograph$.ti,ab.

27. (USS or DUS or CDUS or CEUS).ti,ab.

28. (doppler or duplex).ti,ab.

29. sonograph$.ti,ab.

30. sonogram$.ti,ab.

31. (contrast adj4 US).ti,ab.

32. or/15-31

33. 14 and 32

Appendix 2. EMBASE search strategy

1. thromboembolism/

2. venous thromboembolism/ or deep vein thrombosis/ or lower extremity deep vein thrombosis/ or upper extremity deep vein thrombosis/

3. thrombosis/ or vein thrombosis/ or leg thrombosis/

4. (vein$ adj3 (thromb$ or clot$)).ti,ab.

5. (ven$ adj3 (thromb$ or clot$)).ti,ab.

6. (calf adj3 (thromb$ or clot)).ti,ab.

7. (leg adj3 (thromb$ or clot)).ti,ab.

8. (thigh adj3 (thromb$ or clot)).ti,ab.

9. (proximal adj3 (thromb$ or clot)).ti,ab.

10. (limb adj3 (thromb$ or clot$)).ti,ab.

11. (DVT or VTE).ti,ab.

12. or/1-11

13. diagnostic imaging/

14. color ultrasound flowmetry/

15. echography/

16. Doppler echography/

17. ultrasound/

18. ultrasound.ti,ab.

19. ultrasonogra$.ti,ab.

20. ultrasonic$.ti,ab.

21. echograph$.ti,ab.

22. echotomograph$.ti,ab.

23. (USS or DUS or CDUS or CEUS).ti,ab.

24. (doppler or duplex).ti,ab.

25. sonograph$.ti,ab.

26. sonogram$.ti,ab.

27. (contrast adj4 US).ti,ab.

28. or/13-27

29. 12 and 28

Contributions of authors

This review will be conducted by a team of individuals who collectively possess all the skills necessary to conduct the review. FC and AA will be responsible for applying the eligibility criteria to the studies identified by the search strategy. They will both extract the data and undertake assessment of study quality. KW will design the search strategy, JMcC and SG will provide clinical advice and contribute to the interpretation of the data analysis. FMC will be responsible for the presentation of the data, any meta-analysis and the investigations of heterogeneity. All will contribute to the final written review.

Declarations of interest

FMC: none known
FC: none known
AA: none known
SG: none known
JM: I received sponsorship from Medtronic as part of the Medtronic University program. This is an educational program, and includes registration and attendance at the European Vascular Course 2012. No financial remuneration was received by myself, other than costs of travel, accommodation, course fees and meals. I received sponsorship from Medtronic to attend the Vascular Society annual meeting 2012 in the form of registration fees and accommodation costs. I received sponsorship from Gore to attend a stenting masterclass and the Verve clinical meeting in 2013. This was in the form of travel, accommodation and meals. No other financial remuneration was received. I am a co-founder of UKETS, a not-for-profit trainee initiative which receives funding through sponsorship from endovascular technology and simulation companies (major funding from Medtronic and Mentice). The majority of this sponsorship is non-financial (that is the companies supply trainers on the courses or allow use of their simulators), although some direct financial input is received from Medtronic and Mentice and is used to run events. No profit is derived from this initiative.
KW: none known
CO: Royalties from book: Oates CP. Cardiovascular Haemodynamics and Doppler Waveforms Explained. Cambridge: Cambridge University press, 2001. I was sole author of this standard textbook on blood flow and the interpretation of Doppler waveforms.

This review forms part of a National Institute of Health Research (NIHR) Cochrane programme grant. The review is being conducted independently of our funders, the NIHR. The NIHR have no input on the conduct or results of the review.

Sources of support

Internal sources

  • No sources of support supplied

External sources

  • National Institute of Health Research (NIHR), UK.

    This project was funded by the National Institute for Health Research. The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the NIHR, NHS or the Department of Health

  • Chief Scientist Office, Scottish Government Health Directorates, Scottish Government, UK.

    The PVD Group editorial base is supported by the Chief Scientist Office.

  • National Institute of Health Research (NIHR), UK.

    The PVD Group editorial base is supported by a programme grant from the NIHR.

Ancillary