Deep vein thrombosis (DVT) and pulmonary embolism, in combination often referred to as venous thromboembolism (VTE), is a common postoperative complication that can be fatal. The aim of perioperative thromboprophylaxis, which has become a standard of care in patients undergoing major surgery, is to prevent these (fatal) thromboembolic events. The currently recommended pharmacological agents as well as non-pharmacological modalities used for the prevention of these complications were all accepted based on reduced incidences of (asymptomatic) venous thrombosis using contrast venography or other surrogate markers to detect this disease [1,2]. Moreover, in recent prevention studies with novel anticoagulants, venography was used as the main component of the primary efficacy outcome [3,4].
Initially, in the 1970s, there was reluctance to adopt thromboprophylaxis in the surgical setting because of doubts about the translation of a reduced asymptomatic thrombosis risk into clinical benefit . However, based on systematic reviews as well as consensus conferences, prophylaxis is now widely accepted and applied. Recently, again concerns were voiced that differences in the frequency of venographically detected asymptomatic postoperative DVT seen with the use of various prophylactic regimens may not reflect the true differences in the clinical effectiveness of these regimens. In other words, it is questioned whether venographically detected asymptomatic postoperative DVT is indeed a valid substitute for the clinical entity of VTE and hence whether venography is a valid surrogate outcome measure [6–8].
Based on the existing literature, a set of criteria was defined to assess the validity of a surrogate outcome [9,10]. These generally applicable criteria are that: (i) there is a biological plausible link between the surrogate measurement and the clinical entity of interest, (ii) the test used to measure the surrogate outcome is accurate with objective criteria for normal and abnormal test results and has a high inter-observer agreement, and (iii) the surrogate outcome fully captures the net effects of the intervention.
At least three conditions should be met before the presence of a biological plausible link between the surrogate outcome measure and the clinical entity of interest can be accepted. These conditions are: (i) Is there a similar pathophysiological substrate for what the surrogate outcome measures as for the real disease?; (ii) Does the surrogate outcome precede the clinical disease and can it be detected earlier? and (iii) Is the association between the surrogate outcome and the disease quantifiable?
For venography, the first requirement has been addressed in postmortem and surgical thrombectomy studies, which revealed that intra-luminal defects seen on contrast venography (and pulmonary angiography) indeed reflect a venous thromboembolic mass consisting of fibrin strands and platelets. In addition, studies employing radioactive fibrinogen have indirectly shown that fibrin is formed at sites of venographically detected thrombi .
Contrast venography satisfies the second condition although it depends on the timing of the examination. Venous thrombi tend to start in the calf veins and gradually extend proximally. They usually become symptomatic when they are larger and obstruct the venous outflow. For example, when venography is performed approximately 1 week after a major surgery, the thrombi mostly have a small to moderate size and hence are usually asymptomatic. However, if venography is performed several weeks after surgery, a substantial proportion of patients with postoperative thrombi will have developed symptomatic disease .
The third condition concerns the natural history of the abnormality identified by the surrogate measurement. For obvious reasons, there are only limited data on what happens to venographically detected asymptomatic thrombosis which is left untreated. In fact there is only one study evaluating the natural history . In this study, 40 postoperative patients with fibrinogen legscan detected and venographically confirmed thrombosis were followed up. In about one-third of the patients, the thrombus lysed spontaneously, whereas 10% of the patients developed symptomatic pulmonary embolism. These figures are supported by data from a retrospective analysis of untreated thrombosis after total hip arthroplasty .
Additional information to assess the quantifiable association between the surrogate outcome and the disease can be derived from large follow-up studies in cohorts of patients undergoing certain operative procedures who received in-hospital prophylaxis. Among 19 586 patients who underwent elective hip surgery, White et al.  found that the cumulative incidence of symptomatic VTE was 2.8% during a 3-month postoperative follow-up period. Numerous studies in this patient category have shown that about 20% of these patients had thrombi as detected by contrast venography, at hospital discharge . Hence, these studies suggest that approximately one of seven to 10 patients with venographically detected thrombi develop symptomatic disease.
The second major criterion concerns the performance and interpretation of contrast venography. Rabinov and Paulin  clearly described how this examination should be performed to obtain an adequate result. They also defined a set of criteria for determination of a normal or abnormal test result. Contrast venography is widely considered to be an accurate test for both proximal and deep calf vein thrombosis, albeit that formal assessment of sensitivity and specificity is not feasible as there is no other gold standard, in particular, in the setting of prophylaxis.
Inter-observer agreement of contrast venography findings has been shown to be good  (kappa 0.65) and can be further improved by long leg images and a strict adherence to the binary outcome definition of normal or abnormal.
To satisfy the criterion that the effect of the intervention on the surrogate outcome predicts the effect on the clinical outcome, clinical follow-up studies should reveal that the relative reduction in the rate of venographically detected venous thrombosis induced by preventive measures is proportionally related to the reduction in clinical outcomes. At least three lines of evidence are available to address this criterion. The first comes from the introduction of unfractionated heparin (UFH) as thromboprophylaxis in the 1960s to 1970s, in surgical patients. The aim was to reduce the risk of postoperative pulmonary embolism. A meta-analysis of the studies comparing UFH with placebo (or no treatment) in various surgical interventions revealed a reduction of 70% in the surrogate outcome of scintigraphically or venographically detected venous thrombosis . This observation is consistent with the international multicentre trial in more than 4000 surgical patients, which revealed a 70% reduction in clinically confirmed VTE (75% reduction in fatal and non-fatal pulmonary embolism) in favor of prevention by UFH . The second concerns the comparison of low-molecular-weight heparin (LMWH) with placebo or no treatment in patients undergoing general surgery . A systematic review of all randomized trials confirmed that the significant reduction observed in asymptomatic deep vein thrombosis obtained with LMWH [relative risk (RR) 0.28; 95% confidence interval (CI) 0.14–0.54] was associated with a similar reduction in clinical pulmonary embolism (RR 0.25; 95% CI 0.08–0.79) as well as in clinically manifest VTE (RR 0.29; 95% CI 0.11–0.73).
The third line of evidence stems from studies on prolonged prophylaxis after hospital discharge in patients undergoing major orthopedic surgery. These studies aimed at assessing the relative efficacy of 4 weeks of thromboprophylaxis vs. only 1 week during hospital stay. Patients were followed up for 4 weeks during which, contrast venography was performed unless clinically apparent VTE occurred earlier. The reduction of all and proximal DVT (as assessed by contrast venography) was identical to that observed for symptomatic thromboembolism (all approximately 65%) . A similar extent in the reduction of the surrogate measurement and clinical disease was recently observed in a study of prolonged prophylaxis in patients undergoing surgery for fractured hip .
The application of surrogate outcomes in medical research is based on practical, financial and ethical aspects [10,21]. But, ever since their introduction, the use of surrogate outcomes to support conclusions about the effectiveness of therapeutic interventions has been met by clinicians and regulatory agencies with both enthusiasm and skepticism. Indeed, several initially widely applied surrogate outcomes such as ventricular premature beats and exercise tolerance for the evaluation of new antiarrhythmic and inotropic agents, respectively, proved to be unreliable or even misleading predictors for the true clinical outcome and have thereby fuelled the concerns about the reliability of surrogate outcomes in general . Although caution should be exercised, one needs to be careful not to discredit the valuable surrogate outcomes.
The present review demonstrates convincingly that contrast venography as a measure of subclinical venous thrombotic disease is a valid surrogate outcome in the setting of thromboprophylaxis for surgical and medical patients. This conclusion is based on at least two major elements generally accepted as critical for a proper surrogate outcome. First, a quantifiable association between the findings on venography and the clinical outcome is present. This association supports the causal link between the surrogate and the clinical outcome, albeit the data from observational studies on untreated asymptomatic thrombosis are scarce because of the possible lethal outcome of extending thrombosis. Nevertheless, the relationship is consistent across the available observational studies.
Secondly, improvement in the surrogate outcome induced by therapeutic interventions is consistently and proportionally associated with an improvement in the target clinical outcome. This is based on data from numerous randomized trials in different clinical settings. Although, in certain subgroups, slight differences are observed in the proportion of the reduction, e.g knee vs. hip surgery , this variation does not invalidate the overall conclusions.
It should be realized that for the optimal use of venography as a surrogate outcome several important prerequisites need to be met. These include the standard execution and documentation of the test, a blinded evaluation by experienced readers who use the internationally accepted criteria for normal, abnormal and inadequate test results.
In conclusion, the critical application of the predefined criteria for the validity of contrast venography as a surrogate outcome reveals that this test can be used with confidence in the evaluation of new thromboprophylactic regimens.