Studies were eligible if they reported thromboembolic events including DVTs or PEs in the sub-acute phase of SCI (3–6 months post injury).

The three outcomes were incidence of PE in the sub-acute phase of SCI, incidence of DVT in the sub-acute phase of SCI and thromboprophylaxis used in the sub-acute phase.

The Cochrane Central Register of Controlled Trials (CENTRAL, March 2015), MEDLINE (HDAS, 1950 to March 2015) and EMBASE Classic and EMBASE (HDAS, 1980 to March 2015) databases were searched as summarized by the medical librarian at Stoke Mandeville Hospital. The search strategy was designed for maximum sensitivity using free text, thesaurus terms and exploded thesaurus terms. There were no language or time restrictions in the searches. Appendix 1 provides full details. Additionally, the references of relevant studies were screened for eligible papers.

Two review authors (S.A., A.H.) independently reviewed the titles and abstracts identified from the search. Full-text publications were obtained when necessary and eligibility was determined independently by the authors. The authors independently extracted data from the studies fulfilling the inclusion criteria and assessed quality of included studies. The data was extracted using an agreed data extraction form. Where the incidence of PE or DVT was not directly reported in the included studies, it was calculated from the results. We adopted the Cochrane risk of bias tool for non-randomized trials (ACROBAT-NRSI) to assess the quality of included studies [16]. The tool assesses domains of sequence generation, allocation concealment, confounding, outcome measurement and assessment and follow-up.

Risk of bias in the seven studies is summarized in Fig. 2. No randomized controlled or quasi randomized trials were found. None of the included studies pre-specified a list of potential VTE confounders in the sub-acute phase. None of the studies involved statistical modelling for confounding factors. We could not tell from the included reports if confounding factors, such as infection, surgical intervention or lack of thromboprophylaxis, contributed to VTE incidence. Diagnosis of PE was based on perfusion lung scan using ^99m TC macro-aggregated albumin particle in one study [17] and using ventilation-perfusion scans and pulmonary angiography in another [18]. El Masri and Silver [19] diagnosed PEs clinically using the criterion of sudden onset of pleuritic chest pain lasting more than 12 hours with hemoptysis, after the exclusion of chest infection and a cardiac event. CTPA and perfusion lung scan were used to diagnose PE in one study [20]. We could not tell how PE diagnosed by Chung et al. [21]. DVT was diagnosed in McKinley et al. [18] clinically or with 1–125 labelled fibrinogen uptake citation, impedance plethysmography, Doppler or venography. In Mackiewicz-Mile et al. [22] DVT was diagnosed by D-dimer and ultrasound Doppler. Lamb et al. [23] accepted any diagnostic method for VTE for inclusion; however we could not tell how patients were diagnosed with PE or DVT from the report. The study reported total VTE events without distinguishing between PE and DVT in the final results. Methods for collecting data were poorly reported and we could not tell if any of the studies used more than one outcome assessor to repeat the analysis to ensure reliability. However, as PE and DVT were diagnosed using objective imaging methods the risk of bias in outcome assessment was thought to be low. In Alabed et al. [20] data about CTPA and perfusion scans were collected from Centricity Radiology Information Systems. Images were reported by different consultant radiologists. The included papers did not report incomplete or missing data. We could not tell if this was because of lack or reporting or absence of missing data.

Five papers reported PEs in the sub-acute phase. We did not pool data in a meta-analysis because of differences in populations, study designs and outcome reporting in the included studies. We reported the incidence of PE as reported by the included papers. However, where the included papers did not report the incidence of PE, we calculated it, if possible, from the reported findings. Perkash et al. [17] is a retrospective case series that reviewed 50 spinal patients at the Veterans Administration Hospital, Palo Alto, California between 1976 and 1977. They reported a total of 4 PEs, of which 3 were in the sub-acute phase. Two PEs in the sub-acute phase were recurrent episodes of thromboembolism. All PEs were confirmed on perfusion lung scan. The calculated PE incidence is 6%. El Masri and Silver [19] reported on a retrospective case series of 102 consecutive patients admitted to the National Spinal Injuries Centre, Aylsebury, UK between 1976 and 1979. They reported 19 PEs, of which two were in the sub-acute phase. Both patients with PE had inadequate anticoagulation (less than 6 weeks) and one had multiple PEs prior to the sub-acute PE. The calculated sub-acute PE incidence was 2%. McKinley et al. [18] described a multicenter retrospective case series. The data of 6,772 spinal patients were reviewed from the National Spinal Cord Injury Statistical Centre, Birmingham, AL, from 1973 to 1998. The diagnosis of PE reported in the database was made clinically or confirmed with ventilation-perfusion scans or pulmonary angiography. The study reported a PE incidence of 0.5% within the first year of SCI. We could not tell the sub-acute PE incidence separately from the reported findings. Chung et al. [21] was a retrospective cohort study involving data retrieved from the National Health Insurance Research Database of Taiwan from 1998 to 2008. The study included 47,916 spinal patients and 191,664 matched non-SCI controls. It reported a PE incidence of 4.76 per 10,000-person-years in the sub-acute phase compared to 1.48 per 10,000-person-years in the non-spinal patient group. Alabed et al. [20] described a retrospective case series of 640 spinal patients at the aforementioned National Spinal Injuries Centre. The study reviewed every CTPA or perfusion lung scan for all new admissions between 2008 and 2013. The PE incidence in the sub-acute phase was 1.25%; 95% confidence interval 0.39–2.11.

Three studies reported on DVTs in the sub-acute phase. Lamb et al. [23] reported a retrospective case series that reviewed the medical charts of 287 spinal patients treated at the Spinal Cord Injury Centre, Zablocki Veteran's Affairs Medical Centre, Milwaukee, WI. VTE was counted if it was mentioned in the patient's chart regardless of diagnostic method used. The study reported total VTE incidence without specifying if the event was a PE or DVT. Eleven patients had VTE events in the sub-acute phase. We could not tell if these events were DVTs or PEs or both. The VTE incidence was 5.9% in the acute phase and 3.8% in the sub-acute phase. McKinley et al. [18] reported a DVT incidence rate of 2.1% in the first year post-injury. We could not tell the sub-acute DVT incidence separately from the reported findings. In the third study, Mackiewicz-Milewska et al. [22] presented an abstract of a prospective case series that included 63 spinal patients from the rehabilitation department in Bydgoszcz, Poland. The study reported five DVTs in the sub-acute phase diagnosed with Doppler ultrasound. The DVT incidence in the sub-acute phase was 7.9%.

The included studies poorly reported type, dose or duration of anticoagulation used. In the study of El Masri and Silver [19], 66 of 102 patients received prophylactic anticoagulation with phenindione 25 to 100 mg daily for up to 3 months post-injury, and the remaining 36 patients were either admitted after 3 months post injury or had contraindications to anticoagulation. No patient diagnosed with PE was adequately anti-coagulated at the time of PE due to contraindications or interventions. Lamb et al. [23] did not use prophylactic anticoagulation. Prophylactic compression stockings and leg exercise were used. Perkash et al. [17] reported the use of prophylactic compression stockings and leg exercise. We could not tell if prophylactic anticoagulation was used or not. In Alabed et al. [20], all patients received prophylactic anticoagulation with injections of low molecular weight heparin and use of compression stockings for up to 3 months post injury. We could not tell what type of thromboprophylaxis was used in McKinley et al. [18] or Chung et al. [21]. No study reported thromboprophylaxis beyond 90 days post SCI.

The risk of VTE in acute spinal cord injury is one of the highest among patients' groups. Current guidelines recommend 3 months thromboprophylaxis post injury.

There remains a significant risk of VTE in the sub-acute spinal injury period. However, a large prospective cohort studies is needed to establish the real risk of thromboembolism in this period. Double-blind randomized controlled trials are needed to establish the effect of extending thromboprophylaxis to cover the sub-acute injury period.