Results of the search

See Figure 1.

Figure 1

---

---

Study flow diagram.

Included studies

In the 2014 update, one additional study was added (Chin 2009), which resulted in a total of 19 RCTs that met the inclusion criteria. See the Characteristics of included studies table.

Excluded studies

For the 2014 update, a further 23 studies were excluded (Ayhan 2013; Cazaubon 2013; Gao 2012; GlaxoSmithKline 2009; Horner 2012; Ido 1995; Kahn 2012; Kim 2009; Lobastov 2013; Maxwell 2000; Necioglu 2008; Orken 2009; Rabe 2013; Rocca 2012; Sakon 2012; Serin 2010; Shilpa 2013; Sobieraj‐Teague 2012; Sultan 2011; Vignon 2013; Yang 2009; Yang 2010; Zhang 2011) making a total of 68 excluded studies. See the Characteristics of excluded studies table. Two additional reports have been removed from this update as they were not based on primary research (one book chapter and one review article).

The study design of 31 trials did not include the appropriate control and treatment groups for this review (Ayhan 2013; Chandhoke 1991; Fasting 1985; Gao 2012; GlaxoSmithKline 2009; Hansberry 1991; KANT study; Koopmann 1985; Lacut 2005; Lee 1989; Lobastov 2013; Maksimovic 1996; Marston 1995; Maxwell 2000; Moser 1980; Necioglu 2008; Nelson 1996; Norgren 1996; Nurmohamed 1996; Orken 2009; Porteous 1989; Rabe 2013; Ryan 2002; Sakon 2012; Serin 2010; Shilpa 2013; Silbersack 2004; Sobieraj‐Teague 2012; Vignon 2013; Yang 2009; Zhang 2011). Seven trials were not designed to assess the effectiveness of stockings in preventing DVT (Cazaubon 2013; Horner 2012; Ido 1995; Kahn 2012; Kim 2009; Rocca 2012; Yang 2010). One further trial was excluded as it used a retrospective control group (Caprini 1983).

Incidence of DVT was not assessed in five of the trials considered during the selection of studies for this review (Benko 2001a; Ibegbuna 1997; Lewis 1976; Manella 1981; Wilson 1994). One trial solely relied on a clinical diagnosis of DVT and did not use an objective method for confirmation (Wilkins 1952).

One study (Rasmussen 1988) was excluded as it relied solely upon the use of the Tc^99m plasmin test as, according to the authors, it is faster and less labour intensive compared to other tests. This test has high sensitivity (91% to 100%) but low specificity (33% to 67%) and consequently has not gained widespread use for diagnosing DVT (Bergqvist 1990). Therefore, it was likely that a number of positive results in this study may have been false positives, which may have accounted for the high reported incidence of DVT. This necessitated the confirmation of positive results using another method, which was not done in this study. Furthermore, the method of randomisation, which was not described, did not appear to be reliable as there was a substantial difference in the number of patients allocated to the GCS group (74 patients) and the GCS + heparin group (89 patients). Furthermore, these groups may not have been comparable due to considerable bias between the type of abdominal surgery that patients in each group underwent.

Similarly, one further trial was excluded as it was unclear whether the protocol for diagnosis of DVT was standardised throughout the study (Cohen 2007). It was also not clear whether all patients or only selected patients were scanned routinely, how symptomatic DVTs or PEs were 'objectively' diagnosed and whether this was standardised throughout the study. It seemed that some patients were scanned using ultrasound and some using venography. If this was the case, the split between the two methodologies was not clear. Furthermore, patients with asymptomatic DVT only seemed to have been assessed proximally; it was unclear whether they were also examined for more distal DVTs. It was not clear which veins were examined for proximal DVTs and whether this was standardised throughout the study. The authors were contacted to seek clarification, however these queries remained unanswered at the time of publication.

Seven trials were not adequately randomised and were thus excluded from this review (Borow 1983; Ibarra‐Perez 1988; Inada 1983; Ishak 1981; Liavag 1972; Moser 1976; Pitto 2008). One additional trial was excluded due to unclear randomisation, cause of dropouts and method of monitoring the occurrence of DVT in trial participants (Belcaro 1993). The method of randomisation was not made clear in one French trial and this was therefore excluded from this review (Marescaux 1981). One trial was not amenable to analysis as the figures were difficult to interpret (Mellbring 1986), and was therefore also excluded.

Four trials were excluded as they did not use the correct type of stockings (that is GCS), which was a pre‐specified criterion for inclusion. Of these, two trials used pneumatic compression (Ramos 1996; Westrich 1996), one trial used Tubigrip (Rosengarten 1970) and one trial used thick elastic compression stockings instead of thromboembolic deterrent stockings (Flanc 1969). The type of stocking used was not clear in one French trial (Patel 1988), which was thus excluded.

Stockings reduce the cross‐sectional area of the deep veins making the calf muscle pump more effectively and thereby improve blood flow. Authors of the CLOTS trial have suggested that severe leg weakness in patients with acute stroke may therefore account for the ineffectiveness of stockings in this patient group (CLOTS 2009). Two trials on patients with acute stroke (CLOTS 2009; Muir 2000) were thus excluded from this review but included in a separate Cochrane review focusing on patients with acute stroke (Naccarato 2010).

Four trials were only published as abstracts, making it difficult to accurately assess their methodology and to extract data (Bolton 1978; Brunkwall 1991; Perkins 1999; Sultan 2011). Further information provided by the authors of the Sultan 2011 trial showed that not all patients participating in this trial had been hospitalised, and this trial therefore did not meet our inclusion criteria.

The study design for two studies could not be adequately assessed as the reports were not published in English (Celebi 2001; Wille‐Jorgensen 1986). They currently await classification. Further details can be found in the table Characteristics of studies awaiting classification.

Allocation

Randomisation of participants to treatment and control groups was mentioned in nine trials and was done using:

• random number tables in six trials (Allan 1983; Bergqvist 1984; Tsapogas 1971; Turner 1984; Wille‐Jorgensen 1985; Wille‐Jorgensen 1991);

• coin toss in one trial (Scurr 1977);

• consecutively numbered boxes in one trial (Kalodiki 1996);

• date of birth in one trial (Torngern 1980), which was deemed to be an inadequate method of randomisation.

Of these trials, the method of randomisation seemed to be unclear in two (Hui 1996; Tsapogas 1971):

• in the Tsapogas trial (Tsapogas 1971) there was a discrepancy between the numbers of participants randomised to the treatment and control groups;

• in the Hui trial (Hui 1996), which looked at the effectiveness of thigh‐length versus knee‐length GCS, patients were randomised in a ratio of 1:1 in the thigh‐length GCS group and 1:4 in the knee‐length GCS group. The control group of the thigh‐length GCS group was also used as the control for the knee‐length GCS group.

The method of randomisation was not mentioned in the remaining 10 included trials (Barnes 1978; Chin 2009; Fredin 1989; Holford 1976; Hui 1996; Kierkegaard 1993; Ohlund 1983; Scurr 1987; Shirai 1985; Turpie 1989). Of note, Chin 2009 was the only included trial published after the publication of the CONSORT statement in 1996 (CONSORT 1996). Despite this, the Chin 2009 trial report did not include a power calculation and did not report the method of randomisation and use of allocation concealment, as advised in the CONSORT statement, suggesting risk of bias.

There was no mention of allocation concealment in 14 of the 19 RCTs. The remaining five studies (Barnes 1978; Holford 1976; Kalodiki 1996; Turpie 1989; Wille‐Jorgensen 1991) used sealed envelopes to conceal the allocation of patients to the treatment and control groups.

Blinding

It is inherently difficult to ensure adequate blinding for patients who wear stockings and those who do not. In seven trials (Allan 1983; Bergqvist 1984; Chin 2009; Kalodiki 1996; Turner 1984; Turpie 1989; Wille‐Jorgensen 1985) the radiologist reporting the scan results was unaware of whether the patient, or their leg, belonged to the treatment or control group. Also, in two trials (Fredin 1989; Kierkegaard 1993) the results of the studies were analysed without knowledge of the type of prophylaxis.

Incomplete outcome data

Results for all included patients were analysed in seven trials (Barnes 1978; Chin 2009; Kierkegaard 1993; Scurr 1987; Tsapogas 1971; Turner 1984; Turpie 1989). Patients lost to follow up were accounted for in the remaining trials, with some patients excluded post‐randomisation due to failure to comply with wearing the GCS because they found them uncomfortable.

Selective reporting

All included studies reported the incidence of DVTs, as stated in their aims. Visual inspection of the funnel plot showed that all included trials came within the expected confidence intervals, though there was a suggestion of minimal publication bias (Figure 4).

Figure 4

---

---

Funnel plot of comparison: Incidence of DVT with stockings and without stockings (all specialties).

Other potential sources of bias

None of the trials stratified patients according to DVT risk level. However, our own analysis of the papers indicated that all patients were in either moderate or high risk groups.

Six trials obtained funding or support from pharmaceutical companies or manufacturers of GCS. These companies included Kendall Co (Barnes 1978; Scurr 1987; Wille‐Jorgensen 1991), Beiersdorf AB (Bergqvist 1984), Brevet Hospital Products (Hui 1996) and Rhone‐Poulenc‐Rorer (Kalodiki 1996).

One trial (Barnes 1978) was terminated early as it was deemed unjustifiable to continue after revealing a major incidence of DVT amongst patients who did not wear stockings.

In one trial (Tsapogas 1971), patients in the treatment group were given an additional recommendation regarding exercise that was not given to the control group. This may have had an influence on the risk of thrombosis.

One trial was published in Japanese (Shirai 1985), which made it difficult to accurately appraise the study design.

Incidence of DVT

As part of the 2014 update, trials assessing the effectiveness of GCS as the sole method of prophylaxis were merged together with trials using a background method of thromboprophylaxis for all patients in addition to the use of GCS in the treatment group. This resulted in a total of 2745 analytic units (1681 individual patients and 1064 individual legs) in the meta‐analysis (Analysis 1.1). In the treatment group (GCS), 126 of the 1391 units developed DVT in comparison to 282 of the 1354 units in the control group (no GCS): Peto's odds ratio (OR) of 0.33 (95% Cl 0.26 to 0.41). This amounted to a 9% incidence of DVT in the treatment group in comparison to 21% in the control group.

The I² statistic for this analysis suggested 0% heterogeneity, with P > 0.1 (Analysis 1.1). This was supported by the corresponding forest plot (Figure 4), using Peto's ORs, which showed that results for all studies were within or on the 95% CI suggesting minimal publication bias.

Subgroup analysis by specialty

Subgroup analysis was performed based upon the specialty under which the participants were managed (Figure 5). There was no significant difference between specialty subgroups regarding the effectiveness of stockings in reducing the incidence of DVT (P = 0.08). The majority of participants were general surgical patients, who accounted for 1378 of 2745 participants (50%). Amongst this cohort the incidence of DVT was 44 of 687 (6%) in the treatment group and 140 of 691 (20%) in the control group, based on results from 9 of 19 trials (Allan 1983; Bergqvist 1984; Holford 1976; Scurr 1977; Scurr 1987; Torngern 1980; Tsapogas 1971; Wille‐Jorgensen 1985; Wille‐Jorgensen 1991) (Peto's OR 0.27, 95% CI 0.20 to 0.38).

Figure 5

---

---

Pie chart depicting the number of participants from each specialty included in the meta‐analysis.

Patients undergoing orthopaedic surgery accounted for 598 of 2745 participants (22%). Amongst patients undergoing orthopaedic surgery, 70 of 314 participants (22%) from the treatment group and 97 of 284 participants (34%) from the control group developed DVT, based on results from 6 of 19 trials (Barnes 1978; Chin 2009; Fredin 1989; Hui 1996; Kalodiki 1996; Ohlund 1983) (Peto's OR 0.47, 95% CI 0.32 to 0.68).

The four remaining trials on patients from other specialties provided small sample sizes for each of the specialities (Kierkegaard 1993; Shirai 1985; Turner 1984; Turpie 1989). However, their results favoured the use of stockings (Peto's OR 0.28, 95% CI, 0.16 to 0.48). Of note, only one trial considered medical patients (Kierkegaard 1993), making it difficult to confidently comment on the effect of stockings in these patients. Again, the results favoured the use of stockings in this study (Peto's OR 0.12, 95% CI, 0.03 to 0.51).

Incidence of proximal DVT

Proximal DVTs are felt to be of greatest clinical significance as they are more likely to embolise to the pulmonary veins and can thereby potentially result in fatal PE. The incidence of proximal DVT in the two experimental arms of the included trials was therefore assessed. Eight trials (Barnes 1978; Bergqvist 1984; Chin 2009; Fredin 1989; Kalodiki 1996; Kierkegaard 1993; Scurr 1987; Turpie 1989) provided data for the incidence of proximal DVT amongst 1035 included units (Analysis 2.1). The incidence of proximal DVT was 7 of 517 (1%) units in the treatment group and 28 of 518 (5%) units in the control group (Peto's OR 0.26, 95% CI 0.13 to 0.53). There was no significant difference between subgroup specialities regarding the effectiveness of stockings in reducing the incidence of proximal DVT (P = 0.79).

Incidence of PE

Five trials provided data for the incidence of PE amongst 569 included participants (Barnes 1978; Chin 2009; Holford 1976; Kalodiki 1996; Wille‐Jorgensen 1985). Routine screening for PE was only conducted in two of these studies (Holford 1976; Kalodiki 1996), using perfusion‐ventilation scintigraphy. This method was used to confirm clinically apparent PE in the remaining studies, except in one trial where PE was diagnosed at autopsy (Turpie 1989). The incidence of PE was 5 of 283 (2%) participants in the treatment group and 14 of 286 (5%) in the control group (Peto's OR 0.38, 95% CI 0.15 to 0.96) (Analysis 3.1). These results should be interpreted with caution in light of the aforementioned limitations in reporting of the incidence of PE in the included trials.

One further trial (Turpie 1989) reported that one patient was diagnosed with PE at autopsy but did not state which group this patient belonged to. The cause of death of this patient was, however, found to be cerebral oedema. Two further trials (Bergqvist 1984; Fredin 1989) reported a cumulative incidence of three cases of PE but did not specify which group these patients belonged to. One further trial (Torngern 1980) reported that none of the participants suffered fatal PE. There was no significant difference between subgroups regarding the effectiveness of stockings in reducing the incidence of PE (P = 0.76).

Adverse effects and complications from GCS

Seven of 19 trials mentioned the incidence of adverse effects but none of the trials stated which groups the patients belonged to (Bergqvist 1984; Chin 2009; Fredin 1989; Kalodiki 1996; Kierkegaard 1993; Torngern 1980; Wille‐Jorgensen 1991).

In one trial (Kierkegaard 1993) some patients experienced post‐phlebotic changes. Three trials mentioned the incidence of bleeding associated with the background antithrombotic measure used (Bergqvist 1984; Fredin 1989; Wille‐Jorgensen 1991). One trial (Kalodiki 1996) reported no difference in haemorrhagic complications between the treatment and control groups. One trial reported that none of the patients showed any signs of post‐operative haemorrhage or side effects (Torngern 1980). Similarly, one further trial reported no adverse events related to the use of GCS (Chin 2009).

Patients' complaints were reported in two further trials (Hui 1996; Turpie 1989). In one trial 23% of patients wearing above‐knee stockings and 16% of patients wearing below‐knee stockings found the stockings too uncomfortable and requested their removal (Hui 1996). Ambulant patients in another trial reported disturbance as the stockings fell down easily, which was likely to be due to improper fitting (Turpie 1989).