Graduated compression stockings for the prevention of deep vein thrombosis in postoperative surgical patients; a systematic review and economic model with a value of information analysis

Consequences of deep-vein thrombosis

Deep-vein thrombosis on its own does not frequently result in death, but left untreated it can result in PE. 7 The number of hospitalised patients dying each year from PE following DVT in the UK has been estimated to be 25,000. 4 PE is the immediate cause of death in 10% of all patients who die in hospital. 8 Those who do survive DVT or PE are at increased risk of recurrence, particularly within the first 2 years. 9

Untreated patients may also be at risk of post-thrombotic syndrome (PTS) which can occur immediately or within 10–20 years of the initial episode. 4 Signs and symptoms of PTS include pain, swelling, oedema and ulcers. 10 These conditions can also have a significant impact on an individual’s quality of life. 9

Other long-term complications of VTE include pulmonary hypertension (PHT), abnormally elevated blood pressure within the pulmonary artery and stroke. 4 These long-term consequences have implications on extended prophylaxis and the costs arising from treatment, which will be discussed further below (see Chapter 4).

Thromboprophylaxis

There is evidence that routine prophylaxis reduces morbidity, mortality and health-service costs in hospitalised patients at risk of DVT and VTE. 11 Prophylaxis can be pharmacological [fondaparinux sodium, low-molecular-weight heparin (LMWH) or unfractionated heparin] and/or mechanical. Mechanical methods of prophylaxis include graduated compression stockings (GCSs), intermittent pneumatic compression devices (IPCDs) and pneumatic foot pumps (FPs). GCSs have been shown to reduce the incidence of postoperative DVT in surgical patients to approximately 11%, whereas low-dose heparin (LDH) administered via subcutaneous injection reduces the rate of DVT to around 9%; used together, the rate of DVT is reduced further. 12

Graduated compression stockings/antiembolism stockings

There are two different types of compression hosiery: antiembolism stockings and GCSs. Both products offer graduated compression and the two terms are often used interchangeably, although antiembolism stockings are designed for the prevention of VTE in immobile patients, whereas GCSs are designed for the management and treatment of conditions such as venous leg ulcers and lymphoedema in the ambulant patient. For consistency with the Health Technology Assessment (HTA) scope, we will use the more commonly used term GCS.

Graduated compression stockings exert graded pressure at a decreasing gradient from the ankle to the thigh, which increases blood flow velocity and promotes venous return. In addition, preventing passive venous distension is thought to prevent subendothelial tears and the activation of clotting factors. 4

The Sigel profile, which equates to a graduated compression pressure profile of 18 mmHg at the ankle, 14 mmHg at the mid-calf, 8 mmHg at the knee (popliteal break), 10 mmHg at the lower thigh and 8 mmHg at the upper thigh was found to increase deep venous flow velocity by 75%. 13 The current British and European Standards for antiembolism stockings [BS7672 (1); ENV 12719 (70)] do not replicate the Sigel profile and the British Standard requires pressure to be measured at only three points rather than the five specified by Sigel. 4

Graduated compression stockings are available as knee-length or thigh-length stockings. Patients report that both knee-length and thigh-length stockings are difficult to use, but knee-length stockings wrinkle less than thigh-length, and fewer patients report discomfort when using them. 14 Patient adherence is reported to be higher with knee-length stockings, and thigh-length stockings are more likely to be worn incorrectly. 15,16 Incorrectly worn stockings can be unsafe: thigh-length stockings that are fitted incorrectly or roll down can create a tourniquet effect. In addition, for some patient subgroups, one length of stocking may be more appropriate than the other; for example, it is widely believed that knee-length stockings are more likely to induce wound complications in patients undergoing knee replacement surgery. There are also some patients for whom GCSs are contraindicated, such as those who have peripheral arterial disease.

National Institute for Health and Care Excellence guidelines

In January 2010, NICE published Clinical Guideline (CG) 92 on reducing the risk of VTE (DVT and PE) in patients admitted to hospital (updating previous CG46). 4 The key recommendations relating to thromboprophylaxis in surgical patients are detailed below.

If using mechanical VTE prophylaxis, base the choice of mechanical VTE prophylaxis on individual patient factors including clinical condition, surgical procedure and patient preference. Choose any one of thigh- or knee-length GCSs, foot impulse devices or IPCDs (thigh or knee length).

Further recommendations are made, for example regarding correct sizing and fitting of stockings. The guideline states that patients should be encouraged to wear their stockings day and night until they no longer have significantly reduced mobility.

Pharmacological VTE prophylaxis is also recommended for surgical patients at a low risk of major bleeding, taking into account individual patient factors and according to clinical judgement. Pharmacological VTE prophylaxis should also be continued until the patient no longer has significantly reduced mobility (generally 5–7 days), although for patients with hip fracture or undergoing elective hip replacement surgery, pharmacological VTE prophylaxis should be continued for 28–35 days (according to the summary of product characteristics for the individual agent being used) and, for patients undergoing knee replacement surgery, pharmacological prophylaxis should be continued for 10–14 days.

The NICE guideline states that the length of stockings is a controversial issue and there is no clear randomised evidence that one length of stocking is more effective than another. Clinical judgement, patient preference, concordance and surgical site are all important issues when deciding on stocking length.

Contraindications to GCS use are suspected or proven peripheral arterial disease; peripheral arterial bypass grafting; peripheral neuropathy or other causes of sensory impairment; any local conditions in which stockings may cause damage, such as gangrene or dermatitis; known allergy to material of manufacture; cardiac failure; severe leg oedema or pulmonary oedema from congestive heart failure; unusual leg size or shape; or major limb deformity preventing correct fit.

In February 2012, NICE published an evidence update to CG92. 18 New evidence was found (a Cochrane review by Sachdeva et al. 17) that supported the use of GCSs in surgical patients with or without other methods of thromboprophylaxis, which is in line with current recommendations in CG92. The evidence update stated that the review was not able to answer the question of the efficacy of thigh-length versus knee-length GCSs.

A decision-analytic model was also developed in CG92 to determine the most cost-effective thromboprophylaxis strategy for different hospital population subgroups [hip fracture surgery, total hip replacement (THR), total knee replacement (TKR), general surgery (GS) and general medical admissions]. VTEs and major bleeding events were modelled for the acute period [determined by the randomised controlled trial (RCT) follow-up, typically only 10–14 days) but quality-adjusted life-years (QALYs) and health-service costs arising from these events were modelled over the patient’s lifetime, including treatment of PTS and PHT. Results differed across the different population subgroups, although GCSs either alone or combined with pharmacological prophylaxis was consistently found to be the most clinically effective and cost-effective approach for the prevention of VTE. The different results were largely driven by population differences in terms of the baseline risks of major bleeding and PE. The cost of GCSs was assumed to be £6.36 per pair (2009 prices) but the length was not specified. In addition, no attempt was made to formally model the relative cost-effectiveness of different GCSs lengths.

Scottish Intercollegiate Guidelines Network guidelines

The SIGN published guideline 122 on the prevention and management of VTE in December 2010 (updating previous guidelines 62 and 36). 11 The key recommendations relating to thromboprophylaxis in surgical patients are detailed below.

Patients undergoing abdominal surgery who are at risk as a result of the procedure or personal risk factors should receive thromboprophylaxis with mechanical methods unless contraindicated and either subcutaneous LMWH, unfractionated heparin or fondaparinux.

Patients undergoing THR or TKR surgery should receive pharmacological prophylaxis (with LMWH, fondaparinux, rivaroxaban or dabigatran) combined with mechanical prophylaxis unless contraindicated. Extended prophylaxis should be given.

The SIGN guideline states that studies comparing above-knee with below-knee stockings have been too small to determine whether or not they are equally effective, although a meta-analysis suggested no major difference in efficacy in surgical patients. 19 The guideline recommends that above-knee or below-knee GCSs may be used for prophylaxis of DVT in surgical patients, provided that there are no contraindications and that attention is paid to correct fitting and application. Contraindications are massive leg oedema; pulmonary oedema (e.g. heart failure); severe peripheral arterial disease; severe peripheral neuropathy; major leg deformity; and dermatitis.

Cochrane review: knee-length versus thigh-length graduated compression stockings

A Cochrane review undertaken by Sajid et al. 1 included three small RCTs12,20,21 that compared the effectiveness of thigh-length versus knee-length GCSs in hospitalised postoperative surgical patients. There was no statistically significant difference in clinical effectiveness between the two stocking lengths in terms of reducing the incidence of DVT; however, there was significant heterogeneity among the trials and considerable methodological limitations. The authors concluded that there was insufficient high-quality evidence to determine whether or not thigh-length or knee-length stockings differ in their effectiveness in terms of reducing the incidence of DVT in hospitalised patients. They recommended that a large multicentre RCT be conducted to address this issue.

Cochrane review: elastic compression stockings for prevention of deep-vein thrombosis

A Cochrane review undertaken by Sachdeva et al. 17 included 18 RCTs that compared the effectiveness of GCSs, with or without another method of DVT prophylaxis, versus no stockings in hospitalised patients. Eight RCTs compared GCSs alone with no stockings; the incidence of DVT was statistically significantly lower in the stocking group than in the no stockings group. 22–29 Ten RCTs compared GCSs plus another prophylactic method versus the prophylactic method alone; the incidence of DVT was statistically significantly lower in the stocking plus other prophylactic method group than in the other prophylactic alone group. 30–39 The authors concluded that GCSs are effective at diminishing the risk of DVT in hospitalised patients. However, where stated, all of the included RCTs used thigh-length stockings. The authors of this review also recommended a RCT comparing thigh-length with knee-length GCSs.

The two previous Cochrane reviews did not answer our specific research question. The review by Sajid et al. 1 included only three RCTs and did not seek indirect evidence. The review by Sachdeva et al. 17 did not restrict the inclusion criteria to surgical patients or compare the clinical effectiveness of thigh- versus knee-length stockings; the length of stocking used in some of the included studies was unclear.

Search strategy

A systematic search of the relevant guideline and systematic review databases was undertaken between 31 July 2013 and 14 August 2013, and records were inserted into an EndNote^® version 7.2 (Thomson Reuters, CA, USA) library. The following databases were searched: Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects (DARE), NHS Economic Evaluations Database (NHS EED), Cochrane Methods Register, PROSPERO, HTA Database, National Guidelines Clearinghouse, Turning Research Into Practice (TRIP), Clinical Evidence, NHS Evidence and NICE Clinical Knowledge Summaries. During protocol development, scoping searches identified two up-to-date relevant reviews; a Cochrane review of knee-length versus thigh-length GCSs for prevention of DVT in postoperative surgical patients1 and a Cochrane review of GCSs for prevention of DVT. 17 These reviews searched the Cochrane Central Register of Controlled Trials (CENTRAL) and the Cochrane Peripheral Vascular Diseases Group Specialised Register, which is constructed from weekly electronic searches of MEDLINE, EMBASE, Cumulative Index to Nursing and Allied Health (CINAHL), The Allied and Complementary Medicine Database (AMED) and through hand searching relevant journals. The included and excluded studies lists of these two reviews, and other relevant reviews identified by searching the guideline and systematic review databases, were added to the EndNote library for screening.

In order to bring the searches undertaken in these two Cochrane reviews up to date, systematic searches of electronic sources for RCTs published since January 2010 (the date of the search in the earlier Cochrane review17) were undertaken on 19 February 2014 and added to the EndNote library. The following databases were searched: MEDLINE, MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, CINAHL, AMED and CENTRAL. In addition, information on studies in progress, unpublished research or research reported in the grey literature was sought by searching relevant databases including ClinicalTrials.gov and Current Controlled Trials.

The search strategy used for trials of effectiveness developed for Ovid MEDLINE can be found in Appendix 1. This strategy was modified to run appropriately on the other databases searched. The strategy combines terms for GCS, terms for thrombosis and terms for RCTs. No language restrictions were applied to the search strategies.

In addition, clinical advisors were consulted for additional potentially relevant studies, and reference lists of all included studies and relevant reviews and guidelines were also manually searched.

Titles and abstracts of studies identified by the searches were independently assessed for inclusion by two reviewers using the criteria outlined below. Disagreements were resolved through discussion and, where necessary, by consultation with a third reviewer. For studies of potential relevance, full papers were assessed independently by two reviewers, with disagreements resolved by the same procedure.

Inclusion criteria

Data extraction

Data extraction was conducted by one reviewer using a piloted and standardised data extraction form in EPPI-Reviewer 4.0 (Evidence for Policy and Practice Information and Co-ordinating Centre, London, UK) and independently checked by a second reviewer. Discrepancies were resolved by discussion, with involvement of a third reviewer when necessary. In cases where the same study was reported in multiple publications, the most up-to-date or comprehensive publication was used for data extraction. Data were extracted on study details (e.g. author, year, location of study), patient characteristics (e.g. age, sex, type of surgery, baseline risk factors for VTE), details of intervention (e.g. type of stocking, duration of use, co-interventions, including pharmacological thromboprophylaxis) and reported outcomes (e.g. method of assessment and results).

Quality assessment

The quality of the individual trials was assessed by one reviewer and independently checked by a second reviewer. No primary study was excluded based on the result of the quality assessment; disagreements were resolved by consensus and, if necessary, a third reviewer was consulted. The quality of included trials was assessed using the Cochrane risk of bias tool. 42

Data analysis

Data on the incidence of DVT or VTE in the treatment and comparison groups were extracted into 2 × 2 tables in Microsoft Excel^® 2010 (Microsoft Corporation, Redmond, WA, USA) by one reviewer. Data were checked for accuracy by a second reviewer. An analysis of the data was performed using odds ratios (ORs) along with 95% confidence intervals (CIs). A random-effects model was used owing to the clinical and methodological variation between trials. The statistical package used for analysis was RevMan 5.2 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark). Comparisons of results are presented in forest plots by type of comparison and separately by unit of analysis (patient or leg). Subtotals for comparisons and an overall effect estimate are presented for display purposes only. These results should be interpreted with caution because of methodological differences between the included trials, such as the use of the opposite limb as the control and the range of publication dates of the trials. The I²-statistic was used to quantify statistical heterogeneity.

Data were insufficient to assess the effect of duration of stocking use and baseline risk of DVT on the outcome.

Where meta-analysis of the data was considered inappropriate, data were tabulated. Data on the incidence of PE, mortality and adverse events related to the use of GCSs were tabulated and synthesised narratively.

Network meta-analysis

A network meta-analysis (NMA) was planned, first, to produce consistent effect estimates across different comparisons for the cost-effectiveness and VOI analyses (see Chapter 5, Interventions) and, second, to investigate if the utilisation of indirect evidence may increase the precision of the relative effect estimate for knee-length GCSs versus thigh-length GCSs. As stated previously (see Inclusion criteria), trials were included whether patients were day-case patients or inpatients and regardless of the participants’ level of risk of DVT. If either of these factors was an effect-modifier, then effect estimates from a NMA could potentially be biased if these factors are not controlled for. If there is sufficient evidence, then these will be controlled for in the analysis. If there is insufficient evidence to do so then a NMA consisting of loops may still be of value to a cost-effectiveness or VOI analysis. A high level of inconsistency between direct and indirect evidence, which suggests clinical or methodological heterogeneity, will increase the uncertainty in the effect estimates, which will better inform the decision uncertainty where there are multiple treatments.

In addition, the DVT risk in the trials may not reflect the DVT risk in the decision population in the cost-effectiveness analysis and VOI analysis, and including all the studies maximises the use of available evidence. Where the distribution of study characteristics does not represent the setting of the cost-effectiveness analysis then predictive distributions can be used to represent the uncertainty in the generalisability of the effect estimates to the cost-effectiveness analysis setting.

A NMA is an extension of meta-analysis, but where a meta-analysis includes only direct evidence, a NMA can draw on both direct and indirect evidence. The results from studies that compare interventions A and C are considered to be direct evidence for the treatment effect d_AC. If a study X compares treatments A and B, and a study Y compares treatments B and C, and a treatment effect d_AC is calculated from these two studies, then this result is referred to as indirect evidence.

A standard meta-analysis combines the results from two or more studies that have comparable populations, interventions, comparators and outcomes. Study quality and other study characteristics are also assumed to be similar. Similarly, to make indirect comparisons, it is assumed that the study characteristics are comparable. This is known as exchangeability, which can be investigated through the consistency of the direct and indirect evidence. 43–45 It assumes that, had treatment C been included in the study comparing A and B, then the treatment effect d_AC would be the same as that found from the study of A and C. 46 Assuming consistency, the treatment effect d_AC is the sum of the treatment effects d_AB and d_BC:

A NMA can combine both the direct evidence and the indirect evidence for d_AC. 46 As in a meta-analysis, it is the summary treatment effect from each study that is utilised in the NMA; hence the benefit of randomisation in each study is retained.

Although several outcomes were investigated in this review, there was sufficient evidence to perform a NMA only for the outcome DVT. The systematic review of effectiveness included all trials that evaluated the effectiveness of a stocking treatment. The included trials evaluated the effectiveness of many interventions that did not include a stocking treatment when compared with a stocking treatment. The criterion for developing the networks for the NMAs was that interventions were included in the network only if the effectiveness evaluated in the corresponding study informed directly or indirectly the relative effectiveness of thigh-length versus knee-length stockings.

Full details of the trials included in the network are reported in Chapter 3, Network meta-analysis results. To create the network, interventions that were considered sufficiently similar relative to the interventions of interest were lumped together. The effectiveness of different drugs, LMWH, LDH and fondaparinux were assumed to be the same, and these were therefore lumped together in the network and will be referred to collectively as ‘heparin’. The interventions that formed part of comparisons that informed the effect estimates of interest are shown in Figure 1.

FIGURE 1.

The network of treatments included in the NMA. In the base case, every relative effect measure can be described in terms of the additive effects of thigh-length stockings (T), knee-length stockings (K) and heparin (H).

Based on the advice of the clinical advisors, it was assumed that there was no stocking–heparin interaction in the base case. This implies that the effect of thigh-length compared with knee-length stockings is the same as thigh-length stockings plus concomitant heparin compared with knee-length stockings plus concomitant heparin. The simplest way to model this is to lump thigh-length stockings and thigh-length stockings plus heparin together, knee-length stockings and knee-length stockings plus heparin together, and no treatment and heparin together. This approach potentially loses indirect evidence for thigh-length stockings compared with knee-length stockings from trials that compare thigh-length stockings with heparin and knee-length stockings plus heparin compared with knee-length stockings alone. Instead of lumping those interventions, another approach, which also assumes no interaction and is therefore almost an identical approach, is to assume that the effectiveness of each of the six interventions compared with each other can be described in terms of the additive treatment effects of thigh-length stockings, knee-length stockings and heparin. This follows the NMA methods for a no interaction assumption taken in Wolf et al. 47 In a sensitivity analysis, this assumption was relaxed and the relative effectiveness of each treatment compared with every other treatment was estimated.

The model, written in WinBUGS (MRC Biostatistics Unit, Cambridge, UK), was based on code presented in the NICE Technical Support Document 2. 48 The full code can be found in Appendix 5. DVT was a binary outcome and, therefore, a model with a binomial likelihood was adopted.

Assuming that there may be an interaction between stocking treatment and heparin, the model of the probability of a DVT p_ik for trial i and trial arm k on the log scale uses a log-link function.

where I is the baseline risk, d_t is the effect of the treatment compared with the baseline treatment (no treatment), d_c is the effect of the comparator compared with the baseline treatment and I_{{k ≠ 1}} ensures that the treatment effect does not apply to the baseline trial arm.

Assuming that there is no interaction between stocking and heparin treatment, each treatment is coded as no treatment, thigh-length stocking or knee-length stocking with or without an additive heparin effect. So, for a trial comparing thigh-length stocking with heparin to knee-length stocking alone, the model of the probability of a DVT p_ik on the log-odds scale is:

The model produced ORs for every pairwise comparison between the interventions in the network. Given that the objective of this project was to evaluate the expected VOI of doing further research, the probability that each treatment would be the most effective given the results of a new trial is calculated, using the predictive distributions for each treatment effect. Predictive distributions are used because of the unexplained heterogeneity and the true treatment effect from a new trial may arise from anywhere in the random-effects distribution. This predictive distribution is broader than the posterior distribution of the average treatment effect estimate because of the trial heterogeneity. The probability of error in identifying the most effective treatment can be derived from these probabilities.

Finally, for different risks of DVT for patients on heparin the risk of DVT was calculated for patients on each treatment, and the number of patients needed to treat (NNT) with a selection of treatments to avoid an extra case of DVT beyond that achieved by the selected comparator was calculated. The selection of comparisons and baseline risks for DVT for patients on heparin for the NNT calculations was designed to illustrate the incremental benefits of the heparin combination treatments, which are evaluated in the cost-effectiveness analyses below (see Chapters 4–6).

The duration of stocking use and baseline DVT risk were both considered potential effect modifiers for the analysis, but the effect of these on the results was not investigated as there were insufficient data across the trials. The only potential effect modifiers for which there was evidence across the trials were publication in year 2003 or later, and whether or not patients had undergone orthopaedic surgery. Publication in or after 2003 was considered important owing to changes in clinical practice over the past decade, such as less invasive surgery and shorter recovery times, meaning that older trials may not be applicable to current NHS practice. This was on the advice of our clinical advisors. Orthopaedic surgery was considered because orthopaedic surgery carries a high risk of DVT and there were sufficient data available from the trials to compare orthopaedic with non-orthopaedic surgeries. Subgroup analyses were conducted for trials published before and after 2003, and for orthopaedic and non-orthopaedic surgery patients where a connected network existed that informed the relative effectiveness of thigh-length stockings versus knee-length stockings.

Random- and fixed-effects models were both considered. The random-effects analysis was chosen if there were adequate data to estimate the between-study variance and the model was a better fit according to the residual deviance statistic. For the random-effects model, the between-study variance was assumed to be common across all of the comparisons in the analysis. A uniform distribution was the primary choice for the prior distribution of the between-study standard deviation (SD) in the random-effects model. This had the range 0–10. An inverse gamma distribution was also tested for the prior distribution for the between-study variance as an alternative to the uniform distribution. The choice of prior distribution was based on model fit and sensitivity to the variance of the prior distribution. The results were not sensitive to different uniform prior distributions with upper limits 3–10, and the uniform distribution produced a better-fitting model, so the uniform prior distribution of 0–10 was used. The between-study variance was also compared with the between-study variance estimated for the nine trials that provided estimates for the effectiveness of thigh-length stockings versus no stockings using the Mantel–Haenszel method. 22,24,25,30,33,35–37,49

Model convergence was evaluated by reviewing the posterior densities of the model parameters and the Ruben–Gelman statistic. 50 In total, 20,000 iterations were discarded and the results were based on a further 50,000 iterations.

Inconsistency in the evidence was explored using the node-split method on appropriate edges of loops in the network. 43 The between-study SD for the network was compared with the between-study SD of the comparison of thigh-length stockings versus no stockings using a Bayesian analysis to see if any inconsistency increased its estimate. When assessing consistency for the base-case analysis where no stocking–heparin interaction was assumed, edges in the network with the same treatment effect were considered the same edge, for example, thigh-length stockings versus no stockings and thigh-length stockings plus heparin versus heparin.

Search strategy

The RCTs that met the criteria for inclusion in the review of effectiveness of thigh-length and knee-length GCSs were screened independently by two reviewers to identify those that also provided data on patient adherence and/or preference. The Endnote library of records identified for the review of effectiveness was checked for additional relevant studies (not limited to RCTs) on patient adherence and preference. An additional search of the literature for studies on patient adherence and preference was not undertaken. The findings from studies already identified from the Endnote library of records were consistent in terms of patient experiences with wearing different lengths of GCSs. It was therefore concluded that further searches would not have substantially added to the evidence base.

Data extraction

Data on patient adherence and preference were extracted from the RCTs into Eppi-Reviewer 4.0 by one reviewer. Data from the observational studies were extracted into a Microsoft Word 2010 (Microsoft Corporation, Redmond, WA, USA) document by one reviewer. A second reviewer checked all data for accuracy.

Quality assessment

The quality of the RCTs was assessed using the Cochrane risk of bias tool, as described above (see Quality assessment). 42 The quality of the observational studies was not formally assessed. The observational studies were generally small, provided limited data and the methods used to conduct the studies were not rigorous. These studies were therefore considered to be at high risk of bias and formal quality assessment was not deemed necessary.

Data analysis

Given the heterogeneity between the studies and the limited outcome data reported, the data are presented in tables and as a narrative synthesis.

Flow of studies through the review of effectiveness

The electronic search of the review and guideline literature, undertaken to inform the protocol, identified the NICE guidelines for the prevention of VTE,4,18 and two particularly relevant Cochrane reviews. 1,17 Therefore, because of the existence of these systematic reviews (among others), many relevant trials were identified from their included and excluded studies lists, prior to running the update searches for primary studies.

The electronic search of the relevant systematic review and guideline databases identified 307 records, which were inserted into an EndNote library. From this EndNote library, 12 potentially relevant systematic reviews that appeared to assess GCSs in postoperative surgical patients (including the two reviews identified during the protocol development stage1,17) were obtained so that their lists of included and excluded studies could be systematically searched for potentially relevant primary studies. 1,17,19,51–59 A total of 137 records were added to the EndNote library from the included and excluded studies lists of the 12 relevant systematic reviews (after removal of duplicates).

Once the searches of existing systematic reviews and guidelines were completed, more recent primary studies were searched for. These update searches of electronic databases (from 2010 to 19 February 2014) identified an additional 330 records, which were also added to the EndNote library.

The full papers of 68 potentially eligible primary studies were screened for inclusion in the review. A total of 23 RCTs were included in the review of effectiveness of thigh-length versus knee-length stockings. 12,20–25,30–37,49,60–66

Of these 23 RCTs, 21 reported data for the outcome DVT. 12,20–25,30–37,49,60,61,63,65,66 However, one trial did not report sufficient data to be included in the meta-analysis or NMA, as total numbers of patients in treatment groups were not reported. 65 An additional seven trials did not add to the network of evidence comparing thigh-length with knee-length GCSs. 23,31,32,34,61,63,66 Therefore, 13 RCTs contained data that directly or indirectly informed the relative effectiveness of thigh-length versus knee-length stockings and were included in the NMA. 12,20–22,24,25,30,33,35–37,49,60

Figure 2 presents the flow of studies through the study selection process. Details of studies excluded at the full publication stage, with the reason for their exclusion, are provided in Appendix 2.

FIGURE 2.

Flow diagram of the study selection process.

Characteristics of studies included in the review of effectiveness

Detailed study characteristics for the 23 included trials are presented in Appendix 3.

There was substantial variation among the 23 included RCTs in terms of the patient characteristics, suggesting that the participants had a different baseline risk for DVT. Some trials included only patients with at least one VTE risk factor (usually age over 40 years),20,21,31,33,36,37,61–65 whereas others excluded patients with certain risk factors (usually history of prior VTE). 21,22,25,33,60 The majority of trials did not report the proportion of patients with known VTE risk factors, such as history of prior VTE, malignancy and obesity. The type of surgery and anaesthesia also varied between trials, which also alters the baseline risk for DVT, with orthopaedic surgery being associated with the highest risk.

There was also variation in the interventions used in the RCTs; in some trials, a GCS was worn on only one leg rather than on both legs,22,31,34,35,60,63 and the duration of use varied between trials. For most trials, GCSs were worn until full mobilisation or discharge from hospital, where reported. Patients received different brands of thigh-length or knee-length GCS, including Thrombex (medi GmbH & Co. KG, Bayreuth, Germany), Brevet TX (Mölnlycke Health Care Ltd, Dunstable, UK), Kendall TED [Kendall Company (UK) Ltd, Basingstoke, UK] and SaphenaMedical (Griffiths and Nielsen Ltd, Horsham, UK) antiDVT GCSs. Only four RCTs reported the pressure index of the stockings,21,22,31,60 which varied from 11.3 mmHg at the ankle,21 which is outside the British Standard range, to between 30 mmHg and 40 mmHg at the ankle. 60 Concomitant pharmacological prophylaxis also varied between trials; some trials used dextran, which is no longer used in NHS practice, and, therefore, the DVT results for these trials have been reported separately from those using LMWH, LDH or fondaparinux. 31,32,66 None of the included trials used the NOACs (dabigatran, rivaroxaban or apixaban).

The methods of assessing outcomes also varied between trials, with some trials assessing certain outcomes, such as PE, only if signs or symptoms were present. 30,36,37,60,66 The timing of outcome assessments was generally short, where reported; DVT was assessed up to the seventh postoperative day in nine RCTs,12,23,25,31,34–37,65 one RCT assessed DVT up to the ninth postoperative day,63 one RCT assessed DVT for 10 days,32 one assessed DVT up to the 12th postoperative day33 and two assessed DVT up to the 14th postoperative day. 24,66 The included trials assessed all DVTs, not just symptomatic DVTs. Where reported, the majority of DVTs were asymptomatic, the clinical consequences of which are unknown. Two RCTs that were included in the review for the outcomes PE and mortality also assessed DVT using the ^99mTc-labelled plasmin test. However, owing to the unreliability of this test, our inclusion criteria stated that DVT data were included only if diagnosed using radioiodine (¹²⁵I) fibrinogen uptake, venography, Doppler ultrasound or MRI. Therefore, the DVT results for these two RCTs are not reported. 62,64 Some trials also reported results relating to adverse events, quality of life and patient preference and adherence.

Summary study details are presented in Tables 2–6, categorised by the intervention and comparator assessed, with the most informative to the review question and most clinically relevant interventions presented first (corresponding to the groupings for meta-analysis presented in Results of studies included in the review). The tables demonstrate the clinical heterogeneity between the included trials for each of the meta-analyses.

Quality of studies included in the review

All of the included studies were RCTs. Results of the quality assessment, using the Cochrane risk of bias tool, are presented in Table 7. Each trial has been given an overall risk of bias judgement; trials that have a low risk of bias for all key domains (i.e. have a ‘yes’ response for each key domain) are judged to have a low overall risk of bias, trials that have a high risk of bias for one or more key domains (i.e. have a ‘no’ response) are judged to have a high overall risk of bias, and trials that have an unclear risk of bias for one or more key domains are judged to have an unclear overall risk of bias. The domains relating to allocation concealment and ‘other sources of bias’ were not judged to be ‘key domains’.

Generally, methods were poorly reported, with a high proportion of assessments for each domain having to be recorded as unclear (see Table 7). It was clear that the allocation sequence was adequately generated in eight RCTs; methods of sequence generation were inadequate in two RCTs and methods were unclear in 13 RCTs. Concealment of allocation was poorly reported; only two RCTs reported adequate methods, three RCTs reported inadequate methods and methods of concealment of allocation were unclear in 18 RCTs. Study groups were similar at baseline in 18 RCTs, there were differences between groups in important prognostic characteristics in three RCTs and insufficient data were available to assess similarity of baseline characteristics in two RCTs. Seven RCTs reported blinding of outcome assessors; the remainder of RCTs did not report whether or not outcome assessors were blinded to treatment group. Outcome data were either complete, or incomplete outcome data were adequately addressed, in 18 RCTs; only one RCT did not adequately address missing outcome data, and in four RCTs it was unclear whether or not missing outcome data were adequately addressed. Nineteen RCTs appeared to be free of the suggestion of selective outcome reporting and it was unclear whether or not three RCTs were free of the suggestion of selective outcome reporting. One RCT did not report results for one outcome; there were a large number of false-positive fibrinogen uptake test results in patients with stockings; therefore, the paper did not report results of fibrinogen uptake tests for any of the patients. None of the RCTs clearly contained other sources of bias.

Overall, three RCTs can be considered to have a low risk of bias,31,49,60 five have a high risk of bias20,25,35,37,66 and for 15 RCTs the reporting was inadequate to judge the risk of bias. 12,21–24,30,32–34,36,61–65

Eighteen of the RCTs randomised by patient. 12,20,21,23–25,30,32,33,36,37,49,60–62,64–66 Five RCTs randomised by leg;22,31,34,35,63 GCSs were worn on only one leg, rather than both legs in these trials. In addition to the RCTs that randomised by leg, one of the RCTs that randomised by patient applied GCSs to the operated leg only rather than to both legs. 60 In current practice, patients are advised to wear GCSs on both legs; therefore, these trials are not representative of clinical practice. In addition, outcomes such as PE cannot be adequately assessed in trials in which patients wore stockings only on one leg, with the other leg serving as the control, as it may not be clear whether the embolism arose from the stockinged leg or the unstockinged leg.

Many of the included RCTs date back to the 1970s23,30 and 1980s and20–22,24,31,32,34,35,37,61–64,66 therefore, their results may not be generalisable to current practice; surgical practice has changed over time with surgical procedures that are less invasive, shorter duration of hospitalisation and earlier mobilisation after surgery.

In addition, some trials excluded high-risk patients (usually patients with a history of prior VTE);21,22,25,33,60 therefore, the patients in the included trials may not be representative of patients likely to be seen in practice.

Results of studies included in the review

Deep-vein thrombosis

A total of 21 of the included trials reported rates of DVT, and a total of 20 trials provided sufficient data to be included in meta-analyses. The analyses and results of the included trials are reported below by specific treatment comparison.

Thigh-length graduated compression stockings (with or without pharmacological prophylaxis) versus knee-length graduated compression stockings (with or without pharmacological prophylaxis)

The trials by Cohen et al. 49 and Howard et al. 12 were the two most informative trials to answer the review question. Both trials used a combination of GCSs plus pharmacological prophylaxis, reflecting current practice for the treatment of patients at high risk of DVT but who are not at high risk of bleeding. The primary outcome of interest in both trials was VTE. However, all patients experienced DVTs and no PE events occurred. To avoid confusion, we have therefore reported outcomes as DVT events.

Cohen et al. 49 and Howard et al. 12 present inconsistent findings in terms of direction of effect (Figure 3), the first trial favouring knee-length GCSs for the prevention of DVT (OR 0.75, 95% CI 0.26 to 2.13) and the second favouring thigh-length GCSs (OR 2.92, 95% CI 1.14 to 7.52). Reasons for the inconsistent findings between the two trials were unclear and may be attributable to chance. The relatively small number of DVT events and variances in surgical procedure, type of pharmacological prophylaxis, patient risk factors and quality of the trials may have contributed to the inconsistent findings (see Table 2). Patients in the Howard et al. trial12 were slightly younger than patients in the Cohen et al. trial49 (mean age 58 years and 65 years, respectively). Patients in the Cohen et al. trial49 were undergoing THR or standard fracture surgery, whereas patients in the Howard et al. trial12 were recruited from general wards and were, therefore, undergoing various different types of surgery, including orthopaedic and abdominal surgery. Overall, the impact of the clinical heterogeneity is unclear and differences between the trials do not readily predict different treatment effects. Therefore, despite the substantial statistical heterogeneity, the pooled estimate of effect is presented here (see Figure 3).

FIGURE 3.

Rates of DVT comparing thigh-length GCSs (with pharmacological prophylaxis) vs. knee-length GCSs (with pharmacological prophylaxis). M–H, Mantel–Haenszel.

The summary estimate of effect indicated fewer DVT events in patients receiving thigh-length GCSs plus pharmacological prophylaxis than in patients receiving knee-length GCSs plus pharmacological prophylaxis (OR 1.51, 95% CI 0.40 to 5.73), but the result was not statistically significantly. As discussed above, there was substantial statistical heterogeneity (I² = 72%).

Four additional RCTs that compared thigh-length with knee-length GCSs were identified, but these trials did not include additional pharmacological prophylaxis. 20,21,25,65 Unfortunately, the trial by Ayhan et al. 65 was reported only as an abstract and did not provide details on the number of patients in each treatment group; the abstract reported only that no DVTs occurred in either treatment arm. This trial was therefore excluded from meta-analyses. Few events occurred in the Porteous et al. 21 trial. Slightly more DVTs were observed in the Williams and Palfrey trial20 In contrast, Hui et al. 25 reported high DVT event rates in all three treatment arms.

Based on clinical advice, it was considered that the addition of pharmacological prophylaxis to GCSs was unlikely to affect the relative effectiveness of knee-length or thigh-length GCS. The five available RCTs12,20,21,25,49 comparing thigh-length with knee-length GCSs with or without additional pharmacological prophylaxis were therefore combined using meta-analysis (see Figure 4). The potential for a stocking drug interaction will be explored in the NMA, discussed below (see Network meta-analysis results).

FIGURE 4.

Rates of DVT (or VTE) comparing thigh-length GCSs (with or without pharmacological prophylaxis) with knee-length GCSs (with or without pharmacological prophylaxis). a, Number of events calculated from % of DVTs reported and it is unclear if any were bilateral and therefore whether or not double counting of patients has occurred. M–H, Mantel–Haenszel.

The summary estimate of effect for all five trials indicated a trend favouring thigh-length GCS, but, again, the findings were not statistically significant (OR 1.48, 95% CI 0.80 to 2.73) and some trials lacked precision (Figure 4). There was some evidence of statistical heterogeneity for the summary estimate of effect (I² = 33%), which was greater when testing for differences between the trial groupings presented (I² = 44.4%). There was some inconsistency in the direction of effect for trials assessing patients in similar surgical groups. The Cohen et al. 49 and Hui et al. 25 trial were in orthopaedic patients, and the Porteous et al. 21 and Williams and Palfrey20 trials were in patients undergoing abdominal surgery. The reasons for the inconsistency were unclear and may be attributable to chance.

Thigh-length or knee-length graduated compression stockings plus pharmacological prophylaxis versus pharmacological prophylaxis alone

Seven RCTs, comprising 928 surgical patients, compared thigh-length GCSs plus pharmacological prophylaxis with pharmacological prophylaxis alone and reported rates of DVT. 31–33,35–37,66 For study characteristics, see Characteristics of studies included in the review of effectiveness. No trials comparing knee-length GCSs plus pharmacological prophylaxis with pharmacological prophylaxis alone met criteria for inclusion in the review.

Dextran is no longer used in clinical practice; therefore, the three trials administering dextran (Bergqvist and Lindblad,31 Fredin et al. 32 and Ishak and Morley66) are presented for completeness only (see Figure 6) and are considered separately from more recent publications assessing pharmacological prophylaxis currently in use. These comparisons are also omitted from the NMA.

The summary estimate of effect indicated statistically significantly fewer DVT events in patients receiving thigh-length GCSs plus pharmacological prophylaxis than in those receiving pharmacological prophylaxis alone (OR 0.31, 95% CI 0.16 to 0.61), as shown in Figure 5. All trials were consistent for direction of effect. The test suggested no statistical heterogeneity (I² = 0%) but some evidence for subgroup differences (I² = 21.6%).

FIGURE 5.

Rates of DVT comparing thigh-length GCSs plus currently used pharmacological prophylaxis with pharmacological prophylaxis alone. M–H, Mantel–Haenszel.

The summary estimate of effect indicated statistically significantly fewer DVT events in patients receiving thigh-length GCSs plus dextran than in those receiving dextran alone (OR 0.27, 95% CI 0.11 to 0.65), as shown in Figure 6. All trials were consistent for direction of effect. The test suggested some statistical heterogeneity (I² = 33%), which was greater when testing for differences between the trial groupings presented (I² = 40.7%).

FIGURE 6.

Rates of DVT comparing thigh-length GCSs plus dextran pharmacological prophylaxis with pharmacological prophylaxis alone. a, 26 patients (74%) in the stocking group and 33 (80%) in the control group received dextran 70. M–H, Mantel–Haenszel.

Thigh-length or knee-length graduated compression stocking versus no active treatment control

Six RCTs, comprising 710 surgical patients, compared thigh-length or knee-length GCSs with no stockings or a drug placebo (i.e. no active treatment control)22–25,30,33 for study characteristics.

Five trials compared thigh-length GCSs with no active treatment control22,24,25,30,33 (see Characteristics of studies included in the review of effectiveness for details). There are clinical and methodological differences between trials, including that trials were conducted in different surgical specialities and cover a wide range of publication dates (1978–96). Interpretation of the findings should, therefore, take into consideration the changes that have taken place in clinical practice in the intervening years, for example, the use of surgical procedures that are less invasive and reduction in length of patient hospitalisation.

As it is unclear how this clinical heterogeneity might impact on the treatment effect, the summary treatment effect for trials comparing thigh- or knee-length GCSs with no active treatment are presented separately. Comparisons between thigh-length GCSs and no active treatment (Figure 7) show statistically significantly fewer DVT events in patients wearing thigh-length stockings (OR 0.25, 95% CI 0.12 to 0.52). There was, however, some evidence of statistical heterogeneity (I² = 41%) and it was higher when testing for differences between the trial groupings presented (I² = 61.7%).

FIGURE 7.

Rates of DVT comparing thigh-length GCSs vs. no treatment control. a, Number of events calculated from% of DVTs reported and it is unclear if any were bilateral and therefore whether or not double counting of patients has occurred. M–H, Mantel–Haenszel.

Two trials compared knee-length GCSs with no active treatment control. 23,25 As for the thigh-length trials versus no active treatment, there are clinical and methodological differences between trials, and both trials are old (1971 and 1996). Pooling the results of these two trials generated a non-significant result for patients wearing knee-length GCSs compared with no active treatment (by patient) (OR 0.57, 95% CI 0.21 to 1.57). There was some evidence of statistical heterogeneity (I² = 28%) (Figure 8). This non-significant effect for knee-length GCSs versus no active treatment control is much smaller than the statistically significant effect reported for thigh-length GCSs versus no active treatment.

FIGURE 8.

Rates of DVT comparing knee-length GCSs with no treatment control. a, number of events calculated from percentage of DVTs reported and it is unclear if any were bilateral and therefore whether or not double counting of patients has occurred. M–H, Mantel–Haenszel.

Graduated compression stockings alone versus pharmacological therapy alone

One RCT of patients undergoing orthopaedic day surgery compared GCSs alone with pharmacological therapy alone, specifically thigh-length GCSs worn on the operated-on leg only to LMWH administered for 7 days or 14 days. 60 This trial found that the rates of DVT were statistically significantly higher with thigh-length GCSs alone than with LMWH alone (OR 2.62, 95% CI 1.46 to 4.71), ignoring the differences in treatment duration for LMWH.

Other comparisons

Three RCTs compared thigh-length GCSs with an alternative mechanical prophylaxis [i.e. pneumatic compression device (Caprini et al. ,61 Mellbring and Palmer,63 Scurr et al. 34)]. Two RCTs randomised patients by leg, with the other leg serving as the control;34,63 the remaining RCT randomised by patient. 61 These comparisons are not included in the NMA as they do not inform the thigh-length versus knee-length GCSs question.

These three RCTs showed no statistically significant difference in DVT event rates between treatment groups. However, the direction of effect differed between trials; Mellbring and Palmer63 and Caprini et al. 61 had fewer DVT events in the pneumatic compression group, whereas Scurr et al. 34 reported fewer DVT events in the GCSs group (Figure 9).

FIGURE 9.

Rates of DVT comparing thigh-length GCSs with pneumatic compression device. M–H, Mantel–Haenszel.

Network meta-analysis results

Inclusion of trials in the network meta-analysis

Figure 10 presents the network of 20 trials that reported data for the outcome DVT and provided sufficient data to be included in the direct meta-analyses. The two trials that did not report DVT outcome data are not included in this figure or in the NMA. 62,64 In addition, the trial by Ayhan et al. ,65 which was only reported as an abstract, did not report sufficient data to be included in the direct meta-analyses or NMA, as total numbers of patients in treatment groups were not reported; therefore, this trial is also excluded from Figure 10.

FIGURE 10.

Network of trials presenting data on DVT. a, not current NHS practice. Shaded boxes represent interventions not included in the NMA as they do not add to the network of evidence comparing thigh-length with knee-length GCSs.

The shaded boxes in Figure 10 represent interventions that are not included in the NMA, as they do not add to the network of evidence comparing thigh-length with knee-length GCSs. After removing these seven trials that did not add to the network of evidence,23,31,32,34,61,63,66 13 RCTs were included in the NMA. 12,20–22,24,25,30,33,35–37,49,60 There is significant clinical heterogeneity across these trials in terms of the risk of DVT and whether patients are day-case patients or inpatients. The degree to which these factors are effect-modifiers is unclear and there are insufficient data to control for them in the analysis. The results of a NMA including these trials may still be of value to cost-effectiveness and VOI analyses because the network consists of loops and statistical heterogeneity and inconsistency will increase the uncertainty around the effect estimates. It will also allow for consistent estimates across multiple comparisons and the maximum use of available evidence. Furthermore, where the distribution of study characteristics does not represent the setting of the cost-effectiveness analysis, then predictive distributions can be used to represent the uncertainty in the generalisability of the effect estimates to the cost-effectiveness analysis setting.

Base-case analysis: assumption of no stocking–heparin treatment interaction

In Figure 10, the unshaded boxes represent the interventions included in the NMA. However, the effectiveness of different drugs (LMWH, LDH and fondaparinux) are assumed to be the same, and these were therefore lumped together in the network and are collectively referred to as ‘heparin’. The final network for the base-case analysis is presented in Figure 11. A total of 13 trials were included in the NMA, three of which were three-armed trials. The number of trials informing each comparison, including each possible comparison in a three-armed trial, is presented in Figure 11.

FIGURE 11.

The network of all studies comparing the treatments shown with the number of trials informing each comparison shown.

In the base-case analysis, it is assumed that there is no interaction between the effects of stocking and heparin. The result is that each possible pair-wise comparison between the six treatments in Figure 11 can be expressed in terms of three basic parameters as illustrated in Figure 1: thigh-length stockings versus no treatment, knee-length stockings versus no treatment and heparin versus no treatment.

The relative effects for no treatment, thigh-length stockings, knee-length stockings and heparin are presented in Table 11. Results are presented for all comparisons possible given the network. Note that the no interaction model assumes that thigh-length stockings versus no treatment is the same as thigh-length stockings plus heparin versus heparin, etc., and this is shown in the results. Where there was sufficient evidence to estimate the between-study variance, the random-effects models were a better fit than the fixed-effect models. As such, the credible intervals (CrIs) represent the uncertainty around the average treatment effect across trials conducted with unknown clinical and methodological heterogeneity. Given that there is statistical heterogeneity, the predictive distribution for the results from a new trial would be more widely dispersed than the posterior distribution for the random-effects estimates.

TABLE 11 - The no interaction, random-effects analysis. The median ORs and CrIs of no treatment, thigh-length stockings, knee-length stockings and heparin are compared with each other

Med, median.

By definition in the no interaction model, thigh vs. no treatment is equal to thigh + heparin vs. heparin, etc.

Residual deviance: 32.2. Data points: 30. Between-study SD on log-scale: median 0.54 (95% CrI 0.0554 to 1.212).

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	2.60	1.58 to 4.87	1.48	0.72 to 3.69	3.81	1.15 to 10.55	9.88	2.47 to 38.04	5.62	1.44 to 22.75
Thigha	0.38	0.21 to 0.63	–	–	0.57	0.28 to 1.21	1.47	0.41 to 3.75	3.81	1.15 to 10.55	2.16	0.62 to 6.77
Knee	0.68	0.27 to 1.38	1.76	0.82 to 3.53	–	–	2.58	0.53 to 8.40	6.70	1.40 to 25.14	3.81	1.15 to 10.55
Heparin	0.26	0.09 to 0.87	0.68	0.27 to 2.42	0.39	0.12 to 1.89	–	–	2.60	1.58 to 4.87	1.48	0.72 to 3.69
Thigh + heparina	0.10	0.03 to 0.41	0.26	0.09 to 0.87	0.15	0.04 to 0.72	0.38	0.21 to 0.63	–	–	0.57	0.28 to 1.21
Knee + heparin	0.18	0.04 to 0.70	0.46	0.15 to 1.62	0.26	0.09 to 0.87	0.68	0.27 to 1.38	1.76	0.82 to 3.53	–	–

The median OR of heparin compared with no treatment was 0.26 (95% CrI 0.09 to 0.87). The median OR of thigh-length stockings with heparin compared with heparin alone was 0.38 (95% CrI 0.21 to 0.63). The median OR of knee-length stockings with heparin compared with heparin alone was 0.68 (95% CrI 0.27 to 1.38). The median OR of knee-length stockings with heparin compared with thigh-length stockings with heparin is 1.76 (95% CrI 0.82 to 3.53). This estimate favours thigh-length stockings slightly more than the direct estimate of 1.48 (95% CI 0.80 to 2.73) from the direct meta-analysis in Figure 4. These estimates are quite similar, but there appears to be greater heterogeneity across studies in the network than across the studies in the direct meta-analysis.

Given that the objective of this project is to evaluate the expected VOI of doing further research, Table 12 presents the probability that each treatment would be the most effective given the results of a new trial, using the predictive distributions for each treatment effect. Predictive distributions are used because of the unexplained heterogeneity and the true treatment effect from a new trial may arise from anywhere in the random-effects distribution. M1 represents the base case, no interaction model. Taking into account all the interventions simultaneously, there is a 0.73 probability that thigh-length stockings plus heparin is the most effective treatment combination within the population of the included trials.

TABLE 12 - Probability of being most effective in a new trial for each treatment for different models

P, probability.

M1: no interaction model.

M2: no interaction, lumped model.

M3: interaction model.

Intervention	P(Best)
	M1a	M2b	M3c
No treatment	0.00	–	0.00
Thigh	0.04	–	0.00
Knee	0.02	–	0.02
Heparin	0.02	0.03	0.10
Thigh + heparin	0.73	0.82	0.75
Knee + heparin	0.20	0.16	0.12

Although there is little uncertainty that it is the most effective treatment, the marginal benefit of thigh-length stockings plus heparin over heparin alone is less than the marginal benefit of heparin over no treatment, as heparin has already reduced the risk of DVT substantially. Tables 13 and 14 present the estimated median risks of symptomatic DVT and of DVT given different treatments assuming different risk of symptomatic DVT and of DVT while on heparin. The tables also present the NNT to avoid an extra case of DVT beyond that achieved by the comparator for different comparisons. For example, if the risk of DVT while on heparin was 4.94%, implying a risk of 16.4% with no treatment, then nine people would need to be treated with heparin to avoid one case of DVT. Adding thigh-length stockings to heparin would require treating a further 34 patients to avoid one case of DVT over and above the cases of DVT avoided using heparin alone. The NNT with thigh-length stockings in addition to heparin to avoid an extra case of DVT is reasonably low, ranging from 9 to 34, depending on the level of risk for GS patients. The NNT with thigh-length stockings in addition to heparin to avoid an extra case of symptomatic DVT is extremely high, ranging from 133 to 524, depending on the level of risk for GS patients.

TABLE 13 - The no interaction model. The median risk of symptomatic DVT with different treatments and the NNT to avoid a case of symptomatic DVT for different comparisons and assuming different risks of symptomatic DVT while on heparin

H, heparin; K, knee; NT, no treatment; T, thigh.

0.31% risk of DVT taking drugs				0.61% risk of DVT taking drugs				1.23% risk of DVT taking drugs
Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT
No treatment	1.16	H vs. NT	114	NT	2.27	H vs. NT	59	NT	4.49	H vs. NT	30
Knee	0.79	K + H vs. H	785	K	1.56	K + H vs. H	399	K	3.11	K + H vs. H	199
Thigh	0.45	T + H vs. K + H	988	T	0.89	T + H vs. K + H	502	T	1.78	T + H vs. K + H	249
Heparin	0.31	T + H vs. H	524	H	0.61	T + H vs. H	267	H	1.23	T + H vs. H	133
Knee + heparin (K + H)	0.21			K + H	0.42			K + H	0.84
Thigh + heparin (T + H)	0.12			T + H	0.24			T + H	0.48

TABLE 14 - The no interaction model. The median risk of DVT with different treatments and the NNT to avoid a case of DVT for different comparisons and assuming different risks of DVT while on heparin

H, heparin; K, knee; NT, no treatment; T, thigh.

4.94% risk of DVT taking drugs				9.88% risk of DVT taking drugs				19.76% risk of DVT taking drugs
Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT
No treatment	16.40	H vs. NT	9	NT	29.28	H vs. NT	5	NT	48.18	H vs. NT	3
Knee	11.80	K + H vs. H	51	K	22.01	K + H vs. H	26	K	38.79	K + H vs. H	14
Thigh	7.03	T + H vs. K + H	62	T	13.76	T + H vs. K+H	31	T	26.38	T + H vs. K + H	16
Heparin	4.94	T + H vs. H	34	H	9.88	T + H vs. H	17	H	19.76	T + H vs. H	9
Knee + heparin (K + H)	3.41			K + H	6.94			K + H	14.35
Thigh + heparin (T + H)	1.96			T + H	4.04			T+ H	8.64

Inconsistency

An inconsistency analysis was conducted and the results are presented in Figure 12. The inconsistency was considerable. The direct and indirect evidence for the thigh-length stocking compared with heparin comparison was almost completely inconsistent (0.003). Inconsistency would increase the between-study variance estimate if the variation in effect estimates increases as a result. The inconsistency adds to the suggestion of unknown clinical and methodological heterogeneity. The effect of excluding each trial on the consistency of the direct and indirect evidence for the thigh-length stockings compared with heparin effect was evaluated, which showed the highest inconsistency across the network. The trial that has the greatest effect on the inconsistency is Williams and Palfrey,20 a trial that provides direct evidence for the treatment effect of heparin, despite being a small trial. Eliminating the Williams and Palfrey20 trial increases consistency between the direct and indirect evidence for the thigh-length stockings compared with heparin effect from 0.003 to 0.39. There is clearly inconsistency between the heparin effect estimates. Another trial that provides evidence for the heparin effect is Kalodiki et al. 33 Excluding the Kalodiki et al. 33 trial increases consistency to 0.09. The Cohen et al. 49 trial informs the effect of thigh-length stockings with heparin versus knee-length stockings with heparin. Excluding the Cohen et al. 49 trial increases consistency to 0.22.

FIGURE 12.

No interaction NMA network. The direct effects on the log-scale, and direct and indirect effect consistency values in brackets. Key: the solid black lines represent comparisons included in the data set. The arrows indicate which treatment is most effective in the comparison for the direct estimate for that comparison. ‘Thigh → no treatment’ indicates that thigh is more effective than no treatment. The negative numbers are the effect estimates on the log-scale; negative represents more effective. Because of the no interaction assumption, the dashed lines show the implied direct estimate treatment effects. The numbers in brackets represent consistency level between direct and indirect evidence. 0 represents complete inconsistency and 1 represents complete consistency.

To explore inconsistency further, for the no interaction model, the six-node network was collapsed into a three-node network as shown in Figure 13; this represents the ‘No interaction, lumped model’. In this network, because the nodes have been lumped, no heparin effect estimates have been estimated.

FIGURE 13.

No interaction, lumped NMA network. The direct effects on the log-scale, and direct and indirect effect consistency values in brackets.

The ORs for each comparison in the network are presented in Table 15. The results are similar to the no interaction model in which the nodes are not lumped, with a little more precision in the estimates, possibly because of less inconsistency. The probability that thigh-length stockings with heparin would be the most effective treatment were a new trial to be conducted is presented in Table 12 in the M2 column. The probability that thigh-length stockings with heparin is the most effective treatment was slightly higher at 0.82. There is still a high level of inconsistency in this network, that is, approximately 0.11 for this loop. Although different trials have different influences on the inconsistency, there are no outlying trials that can be singled out as likely to be outliers and that would individually explain the inconsistency. For each edge of the network in Figure 13, the trials are fairly consistent. For the thigh-length stocking with heparin versus knee-length stocking with heparin comparison, the estimates of Cohen et al. 49 and Porteous et al. 21 favour knee-length stockings over of thigh-length stockings, whereas the other trials favour thigh-length stockings. However, the CIs overlap the other mean estimates (see Figure 4).

TABLE 15 - The no interaction, lumped, random-effects analysis. The median ORs and CrIs of no treatment, thigh-length stockings, knee-length stockings and heparin compared with each other

Med, median.

By definition in the no interaction, lumped model, thigh vs. no treatment is equal to thigh + heparin vs. heparin, etc.; and no effects are calculated for heparin treatment combinations vs. no heparin treatments.

Residual deviance: 32.2. Data points: 30. Between-study SD on log-scale: median 0.54 (95% CrI 0.0554 to 1.212).

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	2.79	1.77 to 5.18	1.66	0.84 to 4.09	–	–	–	–	–	–
Thigha	0.36	0.19 to 0.57	–	–	0.59	0.31 to 1.21	–	–	–	–	–	–
Knee	0.60	0.24 to 1.19	1.69	0.82 to 3.21	–	–	–	–	–	–	–	–
Heparin	–	–	–	–	–	–	–	–	2.79	1.77 to 5.18	1.66	0.84 to 4.09
Thigh + heparina	–	–	–	–	–	–	0.36	0.19 to 0.57	–	–	0.59	0.31 to 1.21
Knee + heparin	–	–	–	–	–	–	0.60	0.24 to 1.19	1.69	0.82 to 3.21	–	–

Sensitivity analysis: no assumption on stocking–heparin treatment interaction

A sensitivity analysis was conducted where a stocking–heparin treatment interaction was considered to be plausible. The network is shown in Figure 14 and each treatment effect is considered to be potentially different and the results are presented in Table 16.

FIGURE 14.

Interaction NMA network. The direct effects on the log-scale, and direct and indirect effect consistency values in brackets. The solid black lines represent comparisons included in the dataset. The arrows indicate which treatment is most effective in the comparison for the direct estimate for that comparison. ‘Thigh → no treatment’ indicates that thigh is more effective than no treatment. The negative numbers are the effect estimates on the log scale; negative represents more effective. Because of the no interaction assumption, the dashed lines show the implied direct estimate treatment effects. The numbers in brackets represent consistency level between direct and indirect evidence. 0 represents complete inconsistency and 1 represents complete consistency.

TABLE 16 - The interaction, random-effects analysis. The median ORs and CrIs of no treatment, thigh stockings, knee stockings, heparin, thigh + heparin and knee + heparin compared with each other. Stocking–heparin interaction allowed

Med, median.

Residual deviance: 32.57. Data points: 30. Between-study SD on log-scale: median 0.62 (95% CrI 0.083 to 1.42).

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	3.64	1.69 to 9.99	2.78	0.96 to 11.37	7.91	2.45 to 31.50	19.60	5.96 to 95.05	7.61	2.14 to 36.61
Thigh	0.27	0.10 to 0.59	–	–	0.76	0.29 to 2.29	2.18	0.69 to 6.63	5.40	1.67 to 20.16	2.09	0.62 to 7.62
Knee	0.36	0.09 to 1.04	1.31	0.44 to 3.50	–	–	2.86	0.63 to 10.75	7.10	1.65 to 30.45	2.73	0.66 to 11.03
Heparin	0.13	0.03 to 0.41	0.46	0.15 to 1.46	0.35	0.09 to 1.59	–	–	2.48	1.21 to 6.13	0.96	0.33 to 3.15
Thigh + heparin	0.05	0.01 to 0.17	0.19	0.05 to 0.60	0.14	0.03 to 0.61	0.40	0.16 to 0.83	–	–	0.39	0.13 to 1.09
Knee + heparin	0.13	0.03 to 0.47	0.48	0.13 to 1.62	0.37	0.09 to 1.51	1.05	0.32 to 3.03	2.59	0.92 to 7.84	–	–

There is considerable uncertainty in many of the estimates in Table 16. There is greater uncertainty in the between-study SD estimate, which also has a higher median estimate than in the base case. The results of the inconsistency analysis are presented in Figure 14. There is less inconsistency evident in this analysis than the no interaction analyses, but this may be attributable to the same amount of evidence informing more parameters, and the lack of evidence of inconsistency does not prove there is not significant between-comparison heterogeneity. There is still significant inconsistency for two or three comparisons in the network. The median ORs of heparin compared with no treatment was 0.13 (95% CrI 0.03 to 0.41). The median OR of thigh-length stockings with heparin compared with heparin alone was 0.40 (95% CrI 0.16 to 0.83). The median OR of knee-length stockings with heparin compared with heparin alone was 1.05 (95% CrI 0.32 to 3.03). This result indicates there was very little evidence informing this estimate. The median OR of knee-length stockings with heparin compared with thigh-length stockings with heparin is 2.59 (95% CrI 0.92 to 7.84). This effect is significantly larger than that for knee-length stockings compared with thigh-length stockings, which was 1.31 (95% CI 0.44 to 3.50).

Table 12 presents the probability that each treatment would be the most effective given the results of a new trial. M3 represents the interaction model. Taking into account all the interventions simultaneously, there is a 0.75 probability that thigh-length stockings plus heparin is the most effective treatment combination within the population of the included trials.

Although there is little uncertainty that it is the most effective treatment, the incremental benefit of thigh-length stockings plus heparin over heparin alone is less than the incremental benefit of heparin over no treatment, as heparin has already reduced the risk of DVT substantially. Tables 17 and 18 present the estimated median risks of symptomatic DVT and of DVT given different treatments assuming different risk of symptomatic DVT and of DVT while on heparin. Tables 17 and 18 also present the NNT to avoid an extra case of DVT beyond that achieved by the comparator for different comparisons. For example, if the risk of DVT while on heparin was 4.94%, implying a risk of 29.24% with no treatment, then four people would need to be treated with heparin to avoid one case of DVT. Adding thigh-length stockings to heparin would require treating a further 35 patients to avoid one case of DVT over and above the cases of DVT avoided using heparin alone. The NNT with thigh-length stockings in addition to heparin to avoid an extra case of DVT is reasonably low, ranging from 9 to 35, depending on the level of risk for GS patients. The NNT with thigh-length stockings in addition to heparin to avoid an extra case of symptomatic DVT is extremely high, ranging from 137 to 541, depending on the level of risk for GS patients.

TABLE 17 - The interaction model. The median risk of symptomatic DVT with different treatments and the NNT to avoid a case of DVT for different comparisons and assuming different risks of symptomatic DVT while on heparin

H, heparin; K, knee; NT, no treatment; T, thigh.

0.31% risk of DVT taking drugs				0.61% risk of DVT taking drugs				1.23% risk of DVT taking drugs
Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT
No treatment	2.41	H vs. NT	47	No treat	4.65	H vs. NT	25	No treat	9.01	H vs. NT	13
Knee	0.88	T + H vs. H	541	Knee	1.73	T + H vs. H	275	Knee	3.45	T + H vs. H	137
Thigh	0.68	T + H vs. K + H	491	Thigh	1.33	T + H vs. K + H	250	Thigh	2.66	T + H vs. K + H	124
Knee + heparin	0.32			Knee + heparin	0.63			Knee + heparin	1.28
Heparin	0.31			Heparin	0.61			Heparin	1.23
Thigh + heparin	0.13			Thigh + heparin	0.25			Thigh + heparin	0.50

TABLE 18 - The interaction model. The median risk of DVT with different treatments and the NNT to avoid a case of DVT for different comparisons and assuming different risks of DVT while on heparin

H, heparin; K, knee; NT, no treatment; T, thigh.

4.94% risk of DVT taking drugs				9.88% risk of DVT taking drugs				19.76% risk of DVT taking drugs
Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT	Intervention	Risk of DVT, %	Comparison	NNT
No treatment	29.24	H vs. NT	4	No treat	46.58	H vs. NT	3	No treat	66.20	H vs. NT	2
Knee	12.97	T + H vs. H	35	Knee	23.92	T + H vs. H	18	Knee	41.40	T + H vs. H	9
Thigh	10.22	T + H vs. K + H	32	Thigh	19.37	T + H vs. K + H	16	Thigh	35.04	T + H vs. K + H	9
Knee + heparin	5.13			Knee + heparin	10.23			Knee + heparin	20.39
Heparin	4.94			Heparin	9.88			Heparin	19.76
Thigh + heparin	2.07			Thigh + heparin	4.26			Thigh + heparin	9.09

Orthopaedic surgery patients

Whether or not the trial population had undergone orthopaedic surgery was considered a potential effect modifier. A subgroup analysis was undertaken on the orthopaedic and non-orthopaedic groups, assuming both no stocking–heparin interaction and allowing for the interaction.

The network for the orthopaedic subgroup is shown in Figure 15 and included five trials. 25,30,33,49,60 The network for the non-orthopaedic subgroup is shown in Figure 16 and included eight trials. 12,20–22,24,35–37 Fixed-effect analyses were conducted for all of the subgroup analyses owing to a lack of trials with which to estimate the between-study variance. The uncertainty in the effectiveness estimates is likely to be underestimated, as between-study variance and inconsistency has been shown to be significant in this set of trials in the network.

FIGURE 15.

The network for the orthopaedic subgroup with the number of trials informing each comparison shown.

FIGURE 16.

The network for the non-orthopaedic subgroup with the number of trials informing each comparison shown.

No stocking–heparin interaction

The results for the analysis assuming no stocking–heparin interaction for the orthopaedic subgroup of trials are presented in Table 19. The results for the non-orthopaedic group are presented in Table 20. The median ORs are slightly more in favour of both thigh- and knee-length stockings with heparin compared with heparin alone for the non-orthopaedic group compared with the orthopaedic group. The effectiveness of thigh-length stockings compared with knee-length stockings is slightly less in the orthopaedic group than the non-orthopaedic group. The median OR of knee-length stockings compared with thigh-length stockings is 1.55 (95% CrI 0.81 to 2.95) in the orthopaedic group and 1.78 (95% CrI 0.89 to 3.57) in the non-orthopaedic group. This compares with 1.76 (95% CrI 0.82 to 3.53) in the base case.

TABLE 19 - The median ORs and CrIs for the orthopaedic subgroup analysis assuming no stocking–heparin interaction. Fixed-effect analysis

Med, median.

By definition, in the no interaction model, thigh vs. no treatment is equal to thigh + heparin vs. heparin, etc.

Residual deviance: 19.47. Data points: 13.

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	2.05	1.32 to 3.23	1.32	0.72 to 2.46	6.30	3.32 to 12.42	12.92	5.03 to 35.16	8.34	3.24 to 22.20
Thigha	0.49	0.31 to 0.76	–	–	0.64	0.34 to 1.24	3.08	1.77 to 5.49	6.30	3.32 to 12.42	4.07	1.83 to 9.29
Knee	0.76	0.41 to 1.38	1.55	0.81 to 2.95	–	–	4.78	2.10 to 10.90	9.77	3.57 to 27.36	6.30	3.32 to 12.42
Heparin	0.16	0.08 to 0.30	0.33	0.18 to 0.57	0.21	0.09 to 0.48	–	–	2.05	1.32 to 3.23	1.32	0.72 to 2.46
Thigh + heparina	0.08	0.03 to 0.20	0.16	0.08 to 0.30	0.10	0.04 to 0.28	0.49	0.31 to 0.76	–	–	0.64	0.34 to 1.24
Knee + heparin	0.12	0.05 to 0.31	0.25	0.11 to 0.55	0.16	0.08 to 0.30	0.76	0.41 to 1.38	1.55	0.81 to 2.95	–	–

TABLE 20 - The median ORs and CrIs for the non-orthopaedic subgroup analysis assuming no stocking–heparin interaction. Fixed-effect analysis

Med, median.

By definition in the no interaction model, thigh vs. no treatment is equal to thigh + heparin vs. heparin, etc.

Residual deviance: 16.71. Data points: 12.

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	3.83	2.29 to 6.66	2.16	0.90 to 5.21	1.41	0.49 to 4.53	5.41	1.65 to 19.52	3.04	0.89 to 11.39
Thigha	0.26	0.15 to 0.44	–	–	0.56	0.28 to 1.13	0.37	0.11 to 1.31	1.41	0.49 to 4.53	0.80	0.26 to 2.64
Knee	0.46	0.19 to 1.11	1.78	0.89 to 3.57	–	–	0.66	0.14 to 3.11	2.53	0.60 to 10.94	1.41	0.49 to 4.53
Heparin	0.71	0.22 to 2.05	2.72	0.76 to 8.99	1.52	0.32 to 6.99	–	–	3.83	2.29 to 6.66	2.16	0.90 to 5.21
Thigh + heparina	0.18	0.05 to 0.61	0.71	0.22 to 2.05	0.40	0.09 to 1.67	0.26	0.15 to 0.44	–	–	0.56	0.28 to 1.13
Knee + heparin	0.33	0.09 to 1.12	1.26	0.38 to 3.83	0.71	0.22 to 2.05	0.46	0.19 to 1.11	1.78	0.89 to 3.57	–	–

Allowing a stocking–heparin interaction

The results for the analysis assuming no stocking–heparin interaction for the orthopaedic subgroup of trials are presented in Table 21. The results for the non-orthopaedic group are presented in Table 22. Thigh-length stockings plus heparin appears to be significantly more effective in the non-orthopaedic group than in the orthopaedic group. The median ORs are slightly more in favour of both thigh- and knee-length stockings with heparin compared with heparin alone for the non-orthopaedic group compared with the orthopaedic group. The effectiveness of thigh-length stockings plus heparin is significantly higher relative to knee-length stockings plus heparin in the non-orthopaedic group compared with the orthopaedic group. The median OR of knee-length stockings compared with thigh-length stockings is 2.60 (95% CrI 1.11 to 6.34) in the orthopaedic group and 0.99 (95% CrI 0.36 to 2.75) in the non-orthopaedic group. This compares with 1.31 (95% CrI 0.44 to 3.50) when these groups are not distinguished.

TABLE 21 - The median ORs and CrIs for the orthopaedic subgroup analysis allowing for a stocking–heparin interaction. Fixed-effect analysis

Med, median.

Residual deviance: 15.66. Data points: 13.

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	4.02	1.96 to 8.60	1.55	0.63 to 3.77	11.67	5.07 to 28.47	14.58	5.53 to 40.72	17.80	5.15 to 73.66
Thigh	0.25	0.12 to 0.51	–	–	0.39	0.16 to 0.90	2.90	1.66 to 5.19	3.63	1.66 to 8.14	4.38	1.52 to 15.75
Knee	0.65	0.27 to 1.60	2.60	1.11 to 6.34	–	–	7.57	2.80 to 21.14	9.45	3.08 to 30.06	11.58	2.96 to 53.16
Heparin	0.09	0.04 to 0.20	0.34	0.19 to 0.60	0.13	0.05 to 0.36	–	–	1.24	0.70 to 2.26	1.50	0.61 to 4.75
Thigh + heparin	0.07	0.02 to 0.18	0.28	0.12 to 0.60	0.11	0.03 to 0.32	0.80	0.44 to 1.42	–	–	1.21	0.46 to 3.98
Knee + heparin	0.06	0.01 to 0.19	0.23	0.06 to 0.66	0.09	0.02 to 0.34	0.67	0.21 to 1.65	0.83	0.25 to 2.19	–	–

TABLE 22 - The median ORs and CrIs for the non-orthopaedic subgroup analysis allowing for a stocking–heparin interaction. Fixed-effect analysis

Med, median.

Residual deviance: 15.81. Data points: 17.

Intervention	Comparator
	No treatment		Thigh		Knee		Heparin		Thigh + heparin		Knee + heparin
	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI	Med	CrI
No treatment	–	–	3.16	1.60 to 6.62	3.21	0.94 to 11.22	2.02	0.28 to 13.62	10.31	1.85 to 57.64	3.45	0.87 to 14.26
Thigh	0.32	0.15 to 0.62	–	–	1.01	0.36 to 2.81	0.63	0.10 to 3.72	3.23	0.68 to 15.59	1.08	0.33 to 3.76
Knee	0.31	0.09 to 1.06	0.99	0.36 to 2.75	–	–	0.63	0.10 to 3.68	3.22	0.69 to 15.33	1.07	0.33 to 3.65
Heparin	0.49	0.07 to 3.55	1.58	0.27 to 9.66	1.59	0.27 to 9.73	–	–	5.05	2.28 to 13.13	1.72	0.48 to 6.62
Thigh + heparin	0.10	0.02 to 0.54	0.31	0.06 to 1.47	0.31	0.07 to 1.45	0.20	0.08 to 0.44	–	–	0.34	0.12 to 0.88
Knee + heparin	0.29	0.07 to 1.15	0.92	0.27 to 3.00	0.93	0.27 to 3.03	0.58	0.15 to 2.09	2.97	1.13 to 8.06	–	–

Flow of studies through the review of patient adherence and preference

The review of RCTs on the clinical effectiveness of thigh-length and knee-length GCSs identified eight RCTs that also provided some data on patient adherence and/or preference. 21,24,25,32,37,60,63,65 One additional RCT14 and seven observational studies (eight articles) were identified. 15,16,80–85 These additional studies did not meet criteria for inclusion in the review of RCTs of clinical effectiveness, but did provide some data on patient adherence and/or preference.

Another trial included in the review of clinical effectiveness of thigh-length and knee-length GCSs (Cohen et al. 49) that compared knee-length GCSs plus fondaparinux versus thigh-length GCSs plus fondaparinux reported that the level of adherence for the continuous use of GCSs was 85% while patients were hospitalised, but this dropped to 76% once patients were discharged. The authors did not state levels of adherence by stocking length and, therefore, these results will not be discussed further here.

Characteristics of studies included in the review of patient adherence and preference

Detailed characteristics of the RCTs identified by the review of clinical effectiveness of thigh-length and knee-length GCSs have been previously described (see Characteristics of studies included in the review of effectiveness). Brief study characteristics for the observational studies are presented in Table 30.

The objectives of the seven observational studies were specifically to assess the correct use of knee-length and thigh-length GCSs and to elicit patient perspectives about their use. 15,16,80–85 The observational studies reflect adherence only in a hospital setting in which patients are observed by health-care professionals.

Methods for measuring adherence, and definitions for this outcome, were inconsistent across RCTs and observational studies.

Quality of studies included in the review of patient adherence and preference

Detailed quality assessment results for the RCTs identified by the review of clinical effectiveness of thigh-length and knee-length GCSs have been previously described (see Characteristics of studies included in the review of effectiveness). The additional RCT by Benko et al. 14 was assessed for quality and was considered to have an unclear risk of bias. The quality of the observational studies was not formally assessed. The inherent problems associated with observational studies, and the poor reporting in the included observational studies, suggest that they should be considered low quality.

Results of studies included in the review of patient adherence and preference

Methods

Systematic searches of the literature were conducted to identify potentially relevant studies for inclusion in the assessment of the cost-effectiveness of GCSs for the prevention of DVT in postoperative surgical patients.

National Institute for Health and Care Excellence CG924 was published in 2010 and included a review of published cost-effectiveness studies using a comprehensive search strategy designed to find any applied study estimating the cost or cost-effectiveness of any VTE prophylaxis intervention. The searches for this guideline were undertaken up to December 2008; the search identified five non-UK studies, none of which included GCSs as a comparator. The search strategies used in the economics section of the NICE CG924 were rerun with minor amendments in order to update results. The following databases were searched: MEDLINE, EMBASE, Health Economics Evaluations Database and HTA. NICE CG924 searches were originally run in December 2008, so the updated search was limited to 2008–14. Additional searches were run in NHS EED, CENTRAL, EconLit, IDEAS and the NICE website. Results from these were also limited to 2008–14. Following the initial searches as per protocol, the results set was narrowed by conducting more focused searches with a cost-effectiveness study filter, to identify cost-effectiveness-related records within the results set.

Table 34 lists the inclusion criteria for the review of cost-effectiveness. Only full economic evaluations that considered both costs and consequences (including cost-effectiveness, cost–utility and cost–benefit analyses) were included. Full details of the search strategies are reported in Appendix 1. Identified studies were assessed in a stepwise manner; three types of studies in postoperative surgical patients were considered to meet the inclusion criteria for this review: (1) studies with a focus on the cost-effectiveness of different types (lengths) of GCSs, either with or without a background of pharmacological prophylaxis; (2) cost-effectiveness studies that have compared alternative DVT prophylaxis strategies and include either type of compression stockings as a comparator; (3) cost-effectiveness studies comparing alternative DVT prophylaxis strategies but which have not included either type of compression stockings. A UK filter was applied for type (2) and type (3) studies, whereas all studies were considered for type (1) studies. Although study types (2) and (3) do not directly inform questions relating to the cost-effectiveness of the different types of compression stockings, we considered that they may provide useful information in relation to the choice of model structure, key assumptions and source of inputs. Hence, a more focused review of these study types was planned with the intention to report on aspects of these studies which would assist in the conceptualisation and development of a new decision-analytic model.

Titles and abstracts were assessed independently by two reviewers for inclusion and any discrepancies were resolved by consensus. After a detailed review of titles and abstracts, the papers that appeared to meet the inclusion criteria were obtained for a secondary review; this involved the full article being assessed according to the inclusion criteria. Methods and inputs of the included studies were extracted by one reviewer using a standardised data extraction form and checked for accuracy by a second reviewer. This information is summarised within the text of the report, alongside a detailed critique of the main studies and their relevance to the UK NHS. The findings from the review provide the basis for the development of a new model reported in Chapter 5.

Results

A total of 687 records were identified from the systematic literature search of existing cost-effectiveness evidence. In addition, eight previous NICE technology appraisals were identified from a focused search in the NICE website. Figure 17 presents a flow diagram summarising the identification and selection of studies. A total of 659 records were excluded based on the review of titles and abstracts. After reviewing the full papers for the remaining 36 records, only three studies subsequently met the inclusion criteria and were included in the review. 4,73,86 A summary of the included studies is reported in Table 35.

FIGURE 17.

Flow diagram of the cost-effectiveness studies selection process.

No existing economic evaluations were found comparing the different types of GCSs, either with or without a background of pharmacological prophylaxis. The prevention of DVT in postoperative surgical patients, however, has been the subject of a full economic evaluation in two previous NICE CGs, comparing a range of thromboprophylaxis strategies which include compression stockings: NICE CG92, issued in 2010,4 and a 2007 CG produced by the National Collaborating Centre for Acute Care at the Royal College of Surgeons of England. 73

The prevention of DVT has also been the subject of three previous NICE single technology appraisals (STAs): TA157111 (Dabigatran etexilate for the prevention of venous thromboembolism after hip or knee replacement surgery in adults), TA170100 (Rivaroxaban for the prevention of venous thromboembolism after THR or TKR in adults) and TA24586 (Apixaban for the prevention of venous thromboembolism after total hip or knee replacement in adults). All three technology appraisals compared different pharmacological prophylaxis interventions for the prevention of DVT but did not explicitly include either type of GCSs as a comparator. The economic models were not publicly available [either in the manufacturer submission or in the Evidence Review Group (ERG) report] for TA157111 and TA170. 100 Thus, only the manufacturer submission and the ERG critique for TA24586 are reviewed and summarised here.

All included studies evaluated the cost-effectiveness of different thromboprophylaxis strategies from an NHS/PSS perspective. The patient population subgroups differed across studies: the NICE CG924 focused on four surgical patient groups and general medical admission patients; the Royal College of Surgeons guideline73 evaluated four surgical patient groups and the NICE TA24586 focused on adult total hip or knee replacement patients, reflecting the scope for the appraisal (see Table 35). In terms of comparators, the CGs from NICE4 and the Royal College of Surgeons73 included mechanical treatments (i.e. IPCDs, FPs and GCSs) in the comparators list, whereas in NICE TA24586 mechanical prophylaxis (e.g. GCSs) was assumed as background therapy for the anticoagulant treatments and thus not explicitly modelled.

The decision-analytic modelling approach used to estimate the cost-effectiveness of different thromboprophylaxis strategies differed between the three included studies. The CGs from NICE4 and the Royal College of Surgeons73 used a decision-tree structure for the cost-effectiveness analysis. Longer-term costs and consequences of these events captured within the decision tree were subsequently modelled within NICE CG92 by applying different ‘payoffs’ (i.e. longer-terms costs and outcomes) depending on particular pathways and events (see Table 35). However, these longer-term payoffs do not appear to have been modelled using a formal structure and instead simple adjustments appear to have been applied to estimate patients’ remaining life-expectancy and costs conditional on events occurring within the decision-tree structure. The model informing Royal College of Surgeons CG73 included longer-term costs and consequences only as part of a sensitivity analysis, where a separate formal Markov process was used to capture the long-term risk of recurrent VTE and PTS. Within NICE TA245,86 the manufacturer employed a two-stage modelling approach: a decision tree to model treatment in the acute phase (up to 90 days post surgery) and a Markov model for long-term events (90 days post surgery and beyond).

Table 35 provides a structured overview of the three included cost-effectiveness models. The three included studies are discussed in more detail below, focusing on key aspects of the economic modelling approach and the sources for key inputs.

Population

Although there were differences in the patient populations evaluated across the studies, there was also significant overlap in relation to specific patient subgroups. The 2007 Royal College of Surgeons guideline73 included four surgical population subgroups: (1) hip fracture surgery; (2) THR; (3) gynaecological surgery (hysterectomy); and (4) GS. The 2010 NICE CG924 modelled five population subgroups, four of which were surgical: (1) hip fracture surgery; (2) THR; (3) TKR; (4) GS; and (5) general medical patients. The patient population in NICE TA24586 reflected the marketing authorisation for the technology and consisted of adult patients undergoing THR or TKR surgery.

Comparators

The CGs from NICE4 and the Royal College of Surgeons73 included mechanical treatments (IPCDs, FPs and GCSs) in their list of comparators. Compression stockings in the two NICE guidelines were evaluated as a thromboprophylaxis strategy either independently or in combination with a drug intervention (e.g. heparin or warfarin). In the NICE TA245 for apixaban86 mechanical prophylaxis (defined within the manufacturer submission as graduated elasticated compression stockings, intermittent pneumatic foot compression or foot impulse devices) was considered as background therapy and assumed to be used equally in all patients regardless of pharmacological intervention; therefore, GCSs were not explicitly modelled as a comparator in their economic evaluation.

Model structure

The CGs from NICE4 and the Royal College of Surgeons73 employed similar decision-tree model structures. Figures 18 and 19 provide a schematic of each of the two economic models.

FIGURE 18.

Royal College of Surgeons model structure (adapted from CG). 73

FIGURE 19.

National Institute for Health and Care Excellence CG92 model structure (adapted from CG). 4

The events modelled in the NICE CG924 economic model are: DVT, symptomatic DVT, asymptomatic DVT, PE, PTS, fatal PE, PHT, major bleed, fatal bleed and stroke. The Royal College of Surgeons73 economic model includes DVT, symptomatic DVT, asymptomatic DVT, PE, fatal PE, major bleed, fatal bleed and stroke. Recurrent VTE and PTS were modelled only as part of their sensitivity analysis. Therefore, the structural differences between the two economic models are that (1) DVT and PE have been modelled as different events in NICE CG92, whereas in the Royal College of Surgeons73 economic analysis, a combined event is first modelled and then split into symptomatic DVT, asymptomatic DVT and PE; (2) PHT is modelled as a PE consequence in the NICE CG92 economic model only; and (3) VTE recurrence is explicitly modelled only in the Royal College of Surgeons73 economic analysis and only in a sensitivity analysis scenario.

The apixaban appraisal (NICE TA24586) followed a two-stage modelling approach: a decision-tree structure was used to model treatment in the acute phase (up to 90 days post surgery) and a Markov model was developed for long-term events (90 days post surgery and beyond). The rationale was that postsurgery VTE and bleeding would be captured in the decision tree and future events would be modelled over the patients’ lifetime in the Markov model. The model schematic is provided in Figures 20 and 21. The modelled events in the decision tree (acute phase) included PE, proximal symptomatic DVT, distal symptomatic DVT, proximal asymptomatic DVT, distal asymptomatic DVT, intracranial haemorrhage (stroke), other major bleed, non-major clinically relevant bleed, minor bleed and death. The long-term phase (Markov model) included the following states: well, dead, disabled (having experienced intracranial haemorrhage), untreated VTE (after proximal or distal asymptomatic DVT), treated VTE (after PE, proximal or distal symptomatic DVT), mild to moderate PTS and severe PTS.

FIGURE 20.

National Institute for Health and Care Excellence TA24586 decision-tree structure (replicated from manufacturer submission).

FIGURE 21.

National Institute for Health and Care Excellence TA24586 long-term Markov model structure (replicated from manufacturer submission).

Long-term consequences

The three economic models varied in their approach for the inclusion of VTE long-term consequences. The CG from NICE4 modelled the long-term events of PTS and PHT in their base-case analysis but did not address VTE recurrence. The rates for PTS and PHT were sourced from a published meta-analysis of cohort studies (Wille-Jørgensen et al. 112) and a 2006 cohort study, respectively (Miniati et al. 113).

The base-case analysis of the Royal College of Surgeons economic model73 considered only the cost and health effects of events taking place during the observation period of the trials included in their clinical review, and hence did not address longer-term VTE consequences. As a sensitivity analysis, the model estimated events taking place over 5 years (recurrent VTE and PTS) under the hypothesis that strategies that reduce DVT would lead to a similar reduction in the incidence of recurrence and PTS. Symptomatic VTE recurrence rate and PTS rate after symptomatic VTE were derived from a published cohort study by Prandoni et al. 114 and the PTS rate after asymptomatic VTE from a meta-analysis of cohort studies (Wille-Jørgensen et al. 112). A Markov model was used in the sensitivity analysis to estimate long-term costs and effects; its structure is depicted in Figure 22.

FIGURE 22.

Royal College of Surgeons; Markov model structure for long-term events (adapted from CG). 73

In the apixaban appraisal (NICE TA24586), long-term events were part of the two-stage modelling approach that was employed. The long-term Markov model structure has been presented in Figure 21. At 90 days post surgery, patients were assumed to leave the decision-tree model and enter the long-term Markov model. The event or events that patients experienced in the decision tree would define the state in which those patients would enter the Markov model [i.e. well, dead, disabled (after intracranial haemorrhage), untreated VTE (after asymptomatic DVT), treated VTE (after PE or symptomatic DVT)]. The events/consequences included in the long-term Markov model were: DVT (recurrent), PE (recurrent), mild to moderate PTS and severe PTS. Literature reviews were conducted to identify parameter estimates for the long-term risk of recurrent VTE and/or the development of PTS in patients who suffered an initial VTE event. For DVT recurrence rates, the source was Prandoni et al. ;96 for PE recurrence rate a published meta-analysis by Imperiale and Speroff97 was used; and for mild to moderate and severe PTS, Prandoni et al. 96 was referenced.

Risk of venous thromboembolism

The sections below discuss the approach and source of inputs for baseline risk and relative risk (RR) included in the cost-effectiveness analyses of the three identified studies.

Resource use and costs associated with venous thromboembolism

The NICE CG924 cost-effectiveness analysis considered the following categories of cost and resource use: pharmacological prophylaxis costs; mechanical prophylaxis costs; prophylaxis testing and nurse time; VTE diagnosis and treatment costs; and treatment costs for other events (i.e. stroke, PTS, PHT, major bleeding, reoperation). The duration of prophylaxis included in the economic analysis reflected the average duration of prophylaxis in the RCTs. Unit costs were taken from standard NHS sources: NHS reference costs,103 British National Formulary,104 NHS Electronic Drug Tariff,105 NHS Purchasing and Supplies Agency106 and Unit Costs of Health and Social Care 2007. 107 The costs and resources used for diagnosing and treating VTEs were sourced from published guidelines: British Thoracic Society guidelines on the management of PE98 and British Committee for Standards in Haematology guidance on the prophylaxis and treatment of DVT. 99 For patients with stroke, the NICE acute stroke guideline117 was referenced. PTS and PHT were costed using relevant cohort studies and published HTAs and NICE technology appraisals.

The Royal College of Surgeons economic model73 included intervention costs (pharmacological and mechanical prophylaxis), prophylaxis testing and nurse time, VTE diagnosis and treatment costs and other treatment costs (i.e. stroke, PTS, recurrent VTE, major bleeding with or without reoperation). Unit costs were taken from standard NHS sources: NHS reference costs,103 British National Formulary,104 NHS Electronic Drug Tariff,105 NHS Purchasing and Supplies Agency106 and Unit Costs of Health and Social Care 2005. 109 The duration of prophylaxis modelled also reflected the average duration of prophylaxis in the RCTs included in the guideline’s clinical review. For costing VTE diagnosis and treatment, the same published guidelines as in NICE CG924 were considered. PTS, recurrence and stroke were costed using relevant patient cohort studies from the literature.

The economic model in NICE TA24586 for apixaban considered intervention and comparator costs, testing costs, inpatient stay, VTE diagnosis and treatment costs, postdischarge drug administration costs and costs associated with the long-term Markov model states for PE, DVT, mild to moderate and severe PTS. Drug acquisition costs were sourced from Monthly Index of Medical Specialties110 for comparator treatments and from the manufacturer (Bristol-Myers Squibb/Pfizer) for apixaban. Testing unit costs were taken from the rivaroxaban STA submission to NICE (TA170100). Treatment duration in the economic model was derived from the treatment duration in the relevant clinical trials (i.e. ADVANCE-287 and ADVANCE-388 for apixaban, RECORD89–92 and RE-MODEL93 for rivaroxaban and RE-NOVATE94,95 for dabigatran). NHS reference costs (2008/9)103 were used in the analysis for resource use estimation. The Healthcare Resource Group codes were selected based on those used in the NICE guideline (NICE CG924).

Health-related quality of life

In the NICE CG92,4 a literature search was performed for quality-of-life weightings to inform the economic model. For patients with no event, the population average quality of life for England and Wales measured using the European Quality of Life-5 Dimensions (EQ-5D) instrument was used. 118 For other health states, utility value scores from the published literature were used. A very similar approach was followed in the Royal College of Surgeons guideline. 73

In the apixaban NICE technology appraisal (NICE TA24586) utility decrements were used to model utility losses after patients experienced several events within the economic model. A systematic review was conducted to identify quality-of-life values and decrements for VTE-related events and health states. Utility values were identified for the health states of symptomatic distal DVT, symptomatic proximal DVT, PE, major bleed, well/treated VTE, mild to moderate and severe PTS, and intracranial haemorrhage/disability following intracranial haemorrhage. For fully recovered patients following surgery, the value for the health state of ‘well’ derived from EQ-5D UK population norms (Kind et al. 118) was assumed.

Life expectancy

The NICE CG924 and Royal College of Surgeons73 economic models estimated life expectancy using a combination of general population data and subgroup-specific estimates. For the initial postsurgical period, standardised mortality ratios were applied to the relevant age- and sex-specific England and Wales mortality rate. From that period onwards, the relevant age- and sex-specific life expectancy for England and Wales was assumed. In the NICE CG92,4 patients with PHT were assumed to have a life expectancy of 5 years based on evidence from the literature. 119,120 In the case of patients with stroke, a life expectancy of 4.5 years was assumed, based on the NICE acute stroke guideline. 117 In the case of patients dying during the initial hospitalisation, as a result of a fatal PE or fatal bleeding event, the number of expected life-years in the models was zero.

The economic model in NICE TA24586 for apixaban estimated all-cause mortality based on the relevant age- and sex-specific England and Wales mortality rates. 121 For patients dying during the acute period as a result of a fatal PE or fatal bleeding event, the life expectancy was zero. Minor or non-major clinically relevant bleeding events were not associated with higher mortality. Patients with mild/moderate or severe PTS in the long-term model were allowed to transition to the death state only.

Decision problem

This assessment aims to establish the expected value of undertaking additional research comparing the relative effectiveness of thigh-length GCSs with knee-length GCSs, in addition to standard pharmacological thromboprophylaxis, for the prevention of DVT in postsurgical patients. An evidence synthesis was undertaken to estimate clinical effectiveness and inform key clinical parameters for the decision model (see Chapter 2, Network meta-analysis). Here, we discuss the development of a decision-analytic model to formally assess the cost-effectiveness of knee-length versus thigh-length GCSs in addition to standard pharmacological prophylaxis for the prevention of DVT in surgical patients.

Interventions

The thromboprophylaxis strategies being compared in the economic model are:

1. LMWH, which is assumed to be the background pharmacological prophylaxis therapy, administered to all patients in the economic model

2. thigh-length GCSs in addition to standard pharmacological prophylaxis (i.e. LMWH)

3. knee-length GCSs in addition to standard pharmacological prophylaxis (i.e. LMWH).

Although the focus of the assessment was on the relative cost-effectiveness of the different lengths of GCSs as adjunctive treatments to standard pharmacological prophylaxis, owing to the requirements of the model it was necessary to include standard pharmacological prophylaxis alone as an additional comparator. This is because the model requires one of the strategies to function as a source of natural history (or baseline) data to which the relative treatment effects of the comparator strategy or strategies are then applied in order to estimate their predicted event rates. It would have been possible to use either thigh- or knee-length GCSs (in addition to standard pharmacological prophylaxis) as a source of baseline data and then to apply the relative treatment effect of the alternative length GCSs to estimate absolute event rates for both types of stockings. However, we considered that this would introduce additional uncertainty into the analysis, which was not warranted based on the best available evidence and the additional reviews undertaken. Given the relatively small number of trials for GCSs, using these studies to estimate both the baseline event rates and the relative effectiveness of the main strategies would inevitably have resulted in significant uncertainty surrounding both these elements. However, the clinical effectiveness review also identified other external sources of natural history data in relation to the VTE rate itself, which were derived from significantly larger studies and hence provided more precise estimates of the underlying rate of VTE. The external data also allowed variation in the rate of VTE to be explored in relation to specific subgroups. Therefore, it was decided that using external sources of data to provide a baseline in the model was more appropriate than using the RCTs of GCSs identified in our review.

The external sources, however, typically reported estimates for either a no treatment strategy or pharmacological therapy (e.g. LMWH). Consequently, it was necessary to incorporate an additional strategy within the economic model to populate the baseline event rate and then link to the different types of GCSs via the relative effect estimates derived from the synthesis reported above. The justification for including LMWH to function as a baseline in the model is discussed further below (see Baseline risks: acute phase model).

Population

The decision model evaluates the cost-effectiveness of knee-length versus thigh-length GCSs for five surgical population subgroups:

• THR

• TKR

• GS: low-risk patients

• GS: moderate-risk patients

• GS: high-risk patients.

The impact of patient heterogeneity (e.g. as a result of different clinical characteristics) is thus explored in separate analyses. This approach ensures that uncertainty in the decision owing to the imprecision in parameter inputs can be separated from uncertainty over whether or not an intervention is cost-effective for particular subgroups of the population.

Although the population subgroup of hip fracture surgery patients was evaluated in previous NICE guidelines (NICE CG924), we did not identify appropriate evidence to inform key parameter inputs for the current economic analysis (i.e. baseline risk for hip fracture surgery patients on LMWH, patient population numbers for hip fracture surgery). Therefore, the hip fracture surgery subgroup has not been evaluated as part of the current analysis.

Outcomes

The model was developed in accordance with the NICE reference case. 122 The primary outcomes of the analysis are QALYs gained and incremental cost. Costs were estimated from an NHS and PSS perspective. Both costs and QALYs are evaluated over a lifetime horizon and discounted using a 3.5% annual discount rate.

The model

A two-stage modelling approach was adopted to model the VTE pathway, informed by the findings of the cost-effectiveness review. In common with many of the existing model structures, we considered two related elements, reflecting an acute phase post surgery and longer-term consequences. Initial VTE episodes are modelled for the acute period (typically up to 14 days post surgery), but QALYs and health service costs associated with the long-term consequences following an initial VTE episode (i.e. PTS and PHT), as well as consequences of any recurrent VTE event, are modelled in the long-term Markov models, with a lifetime time horizon. Similarly, major bleeding events are modelled for the acute period, but costs and health benefits arising from these events are modelled over the patient’s lifetime.

1. Acute phase: a decision-tree structure has been used to model events taking place in the acute phase (up to 14 days post surgery) and a separate series of Markov models have been designed to model the long-term consequences and the recurrent events that patients can experience after having an initial VTE event. The decision-tree structure is depicted in Figure 23. Depending on the initial VTE event (i.e. symptomatic DVT, asymptomatic DVT, symptomatic PE), patients can experience different long-term consequences and different rates of recurrent VTE events; they are therefore assumed to enter separate Markov models.

2. Long term: conditional on the pathway during the acute phase, a Markov structure is subsequently used to characterise the long-term prognosis over the remainder of a patient’s lifetime. Annual cycles are employed to reflect the annual probability of further VTE events (and associated consequences) and death for each year after the acute period. Therefore, the extent to which the different thromboprophylaxis strategies reduce the risk of VTE events during the acute period will be translated into differences in long-term costs and QALYs on the basis of the long-term model.

FIGURE 23.

Decision-tree model structure.

The Markov model schematics after an initial symptomatic PE, symptomatic DVT and asymptomatic DVT are presented in Figures 24–26. The various health states are represented using circles, and possible transitions between the health states are represented with arrows. Transition to the death state is possible from each of the separate health states and potentially differs depending on the particular state in which a patient resides during each cycle. For the purposes of simplifying the figures, the death state is represented separately for two of these. However, within these particular figures separate transitions are still applied.

FIGURE 24.

Long-term Markov model structure after an initial symptomatic PE event.

FIGURE 25.

Long-term Markov model structure after an initial symptomatic DVT event.

FIGURE 26.

Long-term Markov model structure after an initial asymptomatic DVT event.

The economic model (decision tree and Markov) is populated with different parameter inputs for each prophylaxis strategy (i.e. LMWH prophylaxis alone, thigh-length GCSs plus LMWH prophylaxis, knee-length GCSs plus LMWH prophylaxis). As previously stated, LMWH is incorporated to provide a baseline source of data for the VTE event rates, and the results of the evidence syntheses reported above are used to estimate the equivalent event rates for the two GCSs strategies. Similarly, although the same structure is applied to each of the population subgroups, different parameter inputs are estimated for each of the population subgroups listed in Chapter 5, Population.

The key features of the economic evaluation are presented in Table 36.

Venous thromboembolism events

The clinical effectiveness and cost-effectiveness of the different VTE prophylaxis strategies in the economic model depends not only on the differences between their relative effectiveness, but also on the change in terms of absolute risk for each strategy. Consequently, the baseline risk of VTE is an important determinant of the absolute risk and, hence, cost-effectiveness.

As discussed in Chapter 2, the review of RCTs of effectiveness of thigh-length and knee-length GCSs has provided some data on baseline risk of DVT from the no prophylaxis arms of the trials. To ensure that the best available data are used in the model, searches for existing systematic reviews of risk of DVT in surgical populations were also undertaken, as well as searches for national and international guidelines, for example the guidelines of the ACCP on the prevention of VTE. 123,124 In addition, the approaches used to estimate baseline risk in previous published cost-effectiveness studies were considered.

In terms of the previous cost-effectiveness studies, the NICE CG924 economic model derived estimates of the baseline risk of VTE from the no prophylaxis arms of the studies included in their clinical review. The Royal College of Surgeons CG73 followed a similar approach; the baseline risks of DVT, symptomatic PE and major bleeding in the absence of prophylaxis were also derived from the no prophylaxis arms of the RCTs included in their clinical review.

Although it would have been feasible to use a similar approach within this assessment, the use of ‘no prophylaxis’ as the source of baseline data was rejected on the following grounds:

1. Many of the RCTs that were used to inform the baseline risk estimates within the NICE CG924 date back to the 1970s and 1980s and, therefore, may not appropriately reflect current clinical practice and current surgery techniques. Consequently, there is a significant risk of bias if advances in clinical practice (independent from changes in prophylaxis management) have resulted in lower VTE rates.

2. Estimates linking both types of GCSs (in addition to conventional pharmacological management) are sparse and therefore highly uncertain. Results from the no interaction, random-effects NMA conducted and discussed in Chapter 3 reveal wide CrIs around the comparisons with no treatment. In addition, it would not be possible to link either type of GCSs to no treatment using the no interaction, lumped random-effect NMA.

The searches for systematic reviews of risk of DVT, and the searches for national and international guidelines addressing the risk of DVT identified several potentially relevant guidelines, risk models and risk algorithms; these have been presented in Chapter 3, Baseline risk of deep-vein thrombosis, and summarised in Appendix 4. Most of these approaches for calculating baseline risk of DVT, however, could not easily be applied within an economic model owing to their rather qualitative and descriptive nature.

The two guidelines from the ACCP for the prevention of VTE in orthopaedic surgery patients123 and the prevention of VTE in non-orthopaedic surgery patients124 were identified as the most comprehensive and rigorous guidelines and were considered to be the best source of VTE baseline risk estimates. The ACCP guidelines provide contemporary estimates for baseline risk of VTE events for patients on LMWH prophylaxis that can directly inform the economic model. These estimates were derived from large-scale RCTs that were considered to provide more precision and less bias than other alternatives considered. Table 25 summarises the baseline risk estimates from the ACCP guidelines for the orthopaedic123 and non-orthopaedic patients. 124

Venous thromboembolism

The results of the clinical effectiveness review, discussed in Chapter 3, were used to estimate the relative effectiveness of thigh-length versus knee-length GCSs, in addition to standard pharmacological prophylaxis, for postsurgical patients at risk of DVT. Results from all analyses have been reported in Chapter 3, Network meta-analysis results.

The relative effectiveness inputs that were subsequently used within the economic model are sourced from:

1. The base-case NMA based on the no interaction, random-effect analysis, using the predictive distribution output. The impact of this type of analysis on the uncertainty around the results is further discussed in Chapter 7.

2. The direct meta-analysis comparing thigh-length GCSs (with pharmacological prophylaxis) with knee-length GCSs (with pharmacological prophylaxis).

The justification for using the predictive distribution and its impact are discussed in Chapter 7, Expected value of perfect information results.

In the absence of differential estimates for each type of VTE event, the relative effectiveness estimates were applied to all VTE events included in the model (i.e. symptomatic DVT, asymptomatic DVT and symptomatic PE). The relative effectiveness inputs used in the economic model are summarised in Tables 37 and 38. These relative effectiveness rates were assumed to be common for all population subgroups evaluated.

It should be noted that, although the NMAs in Chapter 3, Network meta-analysis results, grouped LMWH, LDH and fondaparinux under the term ‘heparin’ as a comparator, the relative effectiveness results were subsequently applied to LMWH in the economic model. This is because of the strategies that are being compared in the analysis and the type of evidence that was used to inform the baseline risks (i.e. for patients on treatment with LMWH).

Bleeding

Bleeding events are a complication associated with pharmacological prophylaxis. Hence, we assumed that GCSs prophylaxis (thigh-length or knee-length stockings) in addition to standard LMWH prophylaxis is not expected to alter the incidence of bleeding events compared with LMWH prophylaxis alone. In the economic model, the bleeding event rates and their associated consequences are assumed to be the same across all three prophylaxis strategies (LMWH alone, LMWH plus thigh-length GCSs, LMWH plus knee-length GCSs).

Similarly to the baseline risks for VTE events, the baseline risks for bleeding events in the economic model have been sourced from the two guidelines from the ACCP for the prevention of VTE in orthopaedic surgery patients123 and the prevention of VTE in non-orthopaedic surgical patients. 124 These baseline risk rates for bleeding events refer to patients on treatment with LMWH and can be directly used in the economic model. Table 39 reports the baseline risk estimates for bleeding events used in the economic analysis.

Other complications

The only complication of prophylaxis included in the economic model is major bleeding events, a complication of pharmacological prophylaxis. There may be other important complications associated with VTE prophylaxis but they are more difficult to quantify. Mechanical prophylaxis is not without potential complications. Chapter 3, Results of studies included in the review of patient adherence and preference, presented evidence of patients not adhering to compression stockings usage, with the main reasons being related to discomfort, provision of stockings, removing stockings for bathing or no longer requiring them owing to ambulation. Incorrect GCSs use is also an issue and is related to wearing incorrectly sized stockings, or to the rolling down, binding or wrinkling of the stocking.

Incorrect GCSs use suggests that some disutility (i.e. reduced quality of life) is associated with stockings, but this disutility is difficult to quantify and might be negligible compared with the patient’s underlying condition. Perhaps of more concern is that the discomfort might cause patients not to adhere to GCS use or to wear the stockings incorrectly (especially thigh-length stockings); this might mean that the effectiveness estimated in trial conditions might not be replicated in practice.

To account, to some extent, for those concerns, the cost of nurse time for checking that stockings are fitted correctly has been included in the economic model (see Prophylaxis costs). In addition, the potential impact of differential levels of adherence for thigh-length and knee-length stockings that may arise in a real-world setting has been explored in scenario analyses.

To explore the effect of differential adherence to thigh-length versus knee-length stockings that may occur in clinical practice settings rather than controlled trial settings, scenarios were run in which the adherence of patients to thigh-length stockings was varied. The levels of adherence in these scenarios were informed by the studies reporting comparative adherence rates of thigh-length versus knee-length stockings in Chapter 3, Results of studies included in the review of patient adherence and preference.

Post-thrombotic syndrome, pulmonary hypertension and stroke

The risk of longer-term events was estimated to be conditional on the initial acute risk of VTE (and the type of VTE). In the absence of evidence, the risk of longer-term events was assumed to be the same for each of the population subgroups evaluated. To estimate these probabilities in the current model, we employed a similar approach and data sources to previously published cost-effectiveness studies.

In particular, the NICE CG924 estimates for PTS after an initial VTE event, PHT after a symptomatic PE event and the proportion of major bleeding episodes that lead to stroke were used as inputs in the economic analysis for this assessment. These inputs are summarised in Table 40. It should be noted that the 5-year PTS rates and 2-year PHT rates were annualised and converted to probabilities prior to being used as inputs within the economic model.

Prophylaxis costs

The unit costs of LMWH and GCSs, and the respective sources of this information are provided in Table 43. LMWH prices were taken from the British National Formulary130 for the recommended dose and GCSs published prices were sourced from the NHS Electronic Drug Tariff. 131

With respect to the cost of GCSs, however, differences were identified between the published prices (sourced from the NHS Electronic Drug Tariff131) and the estimates of the actual prices paid, as obtained from our clinical advisors for this assessment. Those different price levels for GCSs are reported in Table 43 and their impact on the cost-effectiveness results is explored in scenario analyses in Chapter 6.

The duration of prophylaxis in the NICE CG924 economic model reflected the average duration of prophylaxis in the RCTs in their clinical review and was assumed to be subgroup specific. The same duration of prophylaxis was considered in the current economic analysis and ranged between 7 and 10 days (Table 44).

To derive the total prophylaxis cost per strategy, the costs of nurse time and monitoring tests per strategy were added. These are reported in Table 45. For nursing time, the NICE CG924 assumed 5–10 minutes of nursing time per day for mechanical prophylaxis strategies (i.e. IPCDs, FPs and GCSs). In the current economic analysis, the lower end of this range was assumed (i.e. 5 minutes per day) owing to the fact that GCSs were assumed to be less complicated to fit and monitor than the other types of mechanical prophylaxis.

Table 46 summarises the total cost for each thromboprophylaxis strategy included in the economic model.

Venous thromboembolism treatment costs

To estimate the resource use and costs for diagnosing and treating symptomatic VTE episodes, we applied NHS reference costs for the treatment of symptomatic DVT and symptomatic PE events (Table 47). 132

In clinical practice, there would be no treatment cost associated with asymptomatic DVT. This has been a standard assumption in previous VTE prophylaxis models and was also assumed in the current analysis.

In the absence of more detailed information in the published literature, it is also assumed that the cost of treating a VTE episode does not vary by population subgroup (i.e. THR, TKR, GS).

Long-term costs

For PTS, PHT and stroke, the treatment pathways are varied and complex, and previous cost-effectiveness studies have used costs estimates from relevant cohort studies of patients from the literature. The NICE CG924 economic model identified these sources primarily through systematic reviews and rapid searches of the HEED and PubMed databases. No updated studies were identified to inform the current economic model; therefore, the NICE CG924 resource-use sources were updated and used to estimate the cost of long-term consequences of VTE and bleeding events (Table 48). All cost estimates have been inflated to 2012/13 prices, using the Hospital and Community Health Services Pay and Prices Index. 132

Bleeding events and stroke costs

The cost of treating major bleeding is assumed to vary primarily according to whether or not there is a decision to reoperate. Rates of reoperation after a bleeding episode differ per patient subgroup within the economic model and were sourced from the NICE CG924 economic model (see Table 48).

For patients with stroke, the NICE acute stroke guideline117 was used to inform the cost and resource use associated with stroke during the first year and for subsequent years (see Table 48).

Fatal events

Naturally for patients dying during the initial hospitalisation, their expected life-years in the model is zero. This refers to patients who experience (1) a fatal PE or (2) a fatal bleeding event.

Patients without long-term consequences

For patients surviving surgery, life expectancy has been estimated using a combination of general population data and subgroup-specific estimates.

• For TKR and GS patients, for the first 12 months standardised mortality ratios were applied to the relevant age- and sex-specific England and Wales mortality rate, so that for the first year after surgery disease-specific mortality was used. For these patients, from 12 months onwards, age- and sex-specific life expectancy for England and Wales was assumed using 2009–11 interim life tables for England and Wales137).

• For THR patients, the standardised mortality ratio was applied for 10 years.

Patients surviving with long-term consequences

In the absence of specific evidence, it was assumed that the life expectancy of patients with PTS would be the same as for other patients in their population subgroup who did not experience PTS.

For patients with PHT, a life expectancy of 5 years was assumed based on evidence from the literature, as used in the NICE CG92 economic model. 119,120

For patients with stroke, a life expectancy of 4.5 years was assumed, based on the NICE acute stroke guideline. 117

Patients with recurrent venous thromboembolism events

Recurrent VTE events (i.e. recurrent symptomatic DVT and recurrent symptomatic PE) were assumed not to have an impact on the life expectancy of patients within the economic model, apart from the proportion of patients for whom a recurrent symptomatic PE event would be fatal.

Scenario 1

In the base-case cost-effectiveness analysis, the estimates of relative effectiveness of thigh-length and knee-length GCSs as adjunctive treatments to LMWH, relative to LMWH alone, were obtained from the base-case NMA reported in Chapter 3 (random effect, no interaction). Chapter 3 concluded that the results of the NMA and the standard meta-analysis were similar regarding the relative effectiveness of thigh- versus knee-length GCSs and that the precision of the estimate of effect was not increased by the NMA. To ensure that this interpretation also applied to the cost-effectiveness results, a separate scenario was explored, which replaced the relative effectiveness estimate for thigh-length GCSs plus LMWH versus knee-length GCSs plus LMWH derived from the NMA, with the estimate derived from the standard meta-analysis.

Table 52 reports the probabilistic cost-effectiveness results for scenario 1. The results and interpretation of this scenario were consistent with those of the base-case and the interpretation of the clinical effectiveness results reported in Chapter 3.

Scenario 2

The national published prices of thigh- and knee-length GCSs were applied in the base-case. However, owing to local purchasing agreements, the national published prices may not appropriately reflect those actually being paid in practice. Hence, we undertook a separate scenario analysis which used local prices reported by our clinical advisors. Importantly, although the price of both stocking types in absolute terms was markedly lower in this scenario, the difference in price between the types of stockings (approximately £2) was exactly the same based on national and local prices. Therefore, it was not envisaged that this scenario would significantly alter the relative cost-effectiveness of thigh-length versus knee-length GCSs compared with the base case, but that the cost-effectiveness of both types of stockings compared with LMWH alone might be impacted. Hence, decision uncertainty surrounding all three strategies could also be significantly affected.

Table 53 presents the probabilistic results for scenario 2. In common with the base-case results, across all five of the populations, thigh-length GCSs plus LMWH appeared to dominate knee-length GCSs plus LMWH. However, in contrast to scenario 1, thigh-length GCSs plus LMWH either dominated or appeared cost-effective (i.e. ICER less than lower bound of the threshold) compared with LMWH alone in all five populations. Hence, applying lower stocking prices that may be more reflective of local prices improves the cost-effectiveness of both types of stockings compared with drug therapy alone (i.e. because the difference in acquisition costs between GCSs and drug-alone strategies reduces). Although there were two populations in the base-case for which thigh-length GCSs plus LMWH did not dominate LMWH alone (THR and low-risk GS), in this scenario thigh-length GCSs plus LMWH now also appeared to be the dominant strategy for the low-risk GS population. Although the lower VTE risks in the THR population meant that thigh-length GCSs plus LMWH did not dominate LMWH alone, the ICER was now more favourable and below conventional cost-effectiveness thresholds (£18,900 per QALY vs. £30,366 in the base case). Importantly, there was less decision uncertainty over whether or not thigh-length GCSs plus LMWH is the most cost-effective strategy compared with the base-case analysis. The error probability across the different populations ranged from between 0.21 and 0.47, compared with a range of between 0.21 and 0.70 in the base case.

Scenario 3

The review of patient adherence studies reported in Chapter 3, Patient adherence and preference, reported that the proportion of patients not wearing stockings or wearing stockings incorrectly appeared to be higher in patients receiving thigh-length GCSs than in those receiving knee-length GCSs. In addition, patient adherence was reported to be higher in the RCTs than the observational studies. Given that the base-case analysis is based on the intention-to-treat results reported in the trials, it is assumed that differences in adherence between the types of stocking are at least partially captured in the model. However, because the model is based on the intention-to-treat results from RCTs, the analysis may not capture differences in adherence that might happen in routine clinical practice outside a trial environment. Consequently, a separate scenario was undertaken to explore the robustness of the cost-effectiveness results to alternative assumptions regarding patient adherence. This scenario explored the impact of differences between the types of stocking (i.e. assuming that adherence to thigh-length GCSs might be lower in clinical practice compared with a trial setting but that adherence to knee-length GCSs would be unaffected). A separate adjustment was thus employed to the relative effectiveness applied to thigh-length GCSs by reducing this relative to LMWH alone by 10% and 25% (i.e. equivalent to assuming an adherence rate for thigh GCSs of between 75% and 90% of that reported in the trials; see Chapter 3, Patient adherence and preference).

The adjustment applied to the relative effectiveness estimates of thigh-length GCSs (10–25%) is based on differences in adherence between the stocking types based on the range of quantitative estimates reported in Chapter 3. Although this scenario provides a useful exploratory analysis of the robustness of the results to higher differences in adherence between the stocking types that may occur in clinical practice settings rather than controlled trial settings, this analysis should be considered as exploratory in nature. It is important to recognise that the results from this analysis are potentially conservative towards thigh-length GCSs compared with knee-length GCSs. First, the evidence on the effectiveness of GCSs and the focus of the model relates to the use of GCSs during the acute postsurgical period (i.e. between 10 and 14 days), rather than the long-term use of GCSs in the postacute period. Consequently, because a significant proportion of this time may be spent in a hospital setting, differences in adherence may be less evident than over a longer time period when a patient has subsequently been discharged. Second, it is difficult to ascertain the potential impact on VTE risks of any reported differences in adherence. The assumption made in this scenario is akin to assuming that those patients who would not be adherent to thigh-length GCSs (but, importantly, would have been compliant with knee stockings) receive no additional clinical benefit compared with drug therapy alone. However, this may be a conservative assumption, because an incorrectly fitted stocking and/or failure to wear the stocking for the recommended duration may result not in a complete loss of benefit but rather some reduction in magnitude of the expected benefit.

Tables 54 and 55 report the results assuming 90% adherence and 75% adherence, respectively, for thigh-length GCSs. In three of the populations considered (TKR, moderate-risk GS and high-risk GS), the ICER results were consistent with the base-case analysis; thigh-length GCSs plus LMWH still dominated the other strategies. However, decision uncertainty was higher in these scenarios, with error probabilities across the thresholds and scenarios of between 0.25 and 0.35, compared with 0.21–0.22 in the base case. Therefore, although assuming a lower adherence rate for thigh-length GCSs did not alter the cost-effectiveness conclusions based on the mean ICER, the uncertainty that thigh-length GCSs plus LMWH was the optimal strategy did increase.

In the GS low-risk population, thigh-length GCSs plus LMWH continued to dominate knee-length GCSs plus LMWH but remained more costly and more effective than LMWH alone. However, because the magnitude of the difference in QALYs decreased with lower adherence rates, the ICER estimates of thigh-length GCSs plus LMWH versus LMWH alone increased compared with the base-case analysis (£5774–13,807 across the two scenarios vs. £2632 in the base case). However, even assuming 75% adherence, these ICER estimates were still under the lower bound of the threshold (£20,000), indicating that thigh-length GCSs plus LMWH appeared cost-effective even when assuming lower adherence rates.

In the THR population, thigh-length GCSs plus LMWH continues to dominate knee-length GCSs plus LMWH, assuming either a 75% or 90% adherence rate. The ICER of interest from the fully incremental analysis was still estimated between thigh-length GCSs plus LMWH versus LWMH alone. Although in the base-case analysis this ICER was only marginally higher than the upper bound of the cost-effectiveness threshold, the ICER estimates in the scenarios increased to £34,297 per QALY (90% adherence) and £48,781 per QALY (75% adherence), such that there was significantly more decision uncertainty over whether or not thigh-length GCSs plus LMWH was the most cost-effective strategy.