Introduction

As is well known, venous thromboembolism (VTE), including deep vein thrombosis (DVT) and pulmonary embolism (PE), significantly affect patient health [1]. It is notably linked to malignant tumors, with cancer-associated VTE posing a common life-threatening complication for adult cancer patients. Such patients have a 4-6.5 times higher risk of developing VTE compared to non-cancer individuals [2, 3]. Furthermore, established cancer-associated VTE significantly contributes to morbidity and is the second leading cause of death among cancer patients [4]. National and international guidelines recommend routine assessment for individuals at high risk of VTE [5,6,7,8]; however, the recommended assessment tools are not specific to cancer patients. In 2017, Gerotziafas et al. developed the COMPASS-CAT risk assessment model based on cancer patients, incorporating tumor-related factors, laboratory characteristics, and clinical features [9]. Several domestic and international studies have applied the COMPASS-CAT assessment scale for solid tumor prediction. Although these studies suggest that the model can comprehensively assess VTE risk in cancer patients, variations in study subjects, regions, scenarios, sample sizes, and threshold values have resulted in diverse outcomes. Therefore, this study evaluates the predictive performance of the COMPASS-CAT thrombosis risk assessment scale for VTE in cancer patients through meta-analysis, aiming to provide a reference for clinical practitioners in selecting appropriate venous thrombosis assessment tools.

Materials and methods

Inclusion and exclusion criteria

Inclusion Criteria:

  1. 1.

    Study Type: Diagnostic studies.

  2. 2.

    Study Subjects: Individuals aged ≥ 18 years, diagnosed with cancer through histopathological or cytological examination.

  3. 3.

    Assessment Tool: The COMPASS-CAT thrombosis risk assessment scale.

  4. 4.

    Diagnostic Criteria: DVT is defined as a positive result in venous color ultrasound, venography, or magnetic resonance venography (MRV) (including central venous catheter (CVC)-associated VTE). For PE, a positive result in pulmonary arteriography, nuclear lung ventilation/perfusion scanning, or Magnetic Resonance Pulmonary Angiography (MRPA) [10, 11].

  5. 5.

    Outcome Measures: Sensitivity (SEN), specificity (SPE), positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and the area under the summary receiver operating characteristic curve (SROC AUC).

Exclusion Criteria:

  1. 1.

    Cross-sectional studies, animal experiments, literature reviews, meta-analyses, and systematic reviews that are non-diagnostic in nature.

  2. 2.

    Studies with incorrect methodology, incomplete data, or those where 2 × 2 table data cannot be extracted.

  3. 3.

    Literature with duplicate publications or data.

  4. 4.

    Non-Chinese and non-English literature.

Search strategy

A computerized search was conducted in PubMed, Web of Science, The Cochrane Library, Embase, CINAHL, OVID, Chinese Biomedical Literature Database (CBM), China National Knowledge Infrastructure (CNKI), WanFang Data, and VIP database from their inception up to April 19, 2023. The search terms in Chinese databases included thrombosis, venous thrombosis, venous thromboembolism, venous thromboembolism, deep vein thrombosis, deep vein thrombosis, pulmonary thromboembolism, pulmonary embolism, pulmonary thromboembolism; cancer, cancer, tumor, tumor, malignancy, leukemia, malignant diseases; COMPASS-CAT, COMPASS-CAT risk assessment, COMPASS-CAT thrombotic risk assessment scale, COMPASS-CAT thrombotic risk assessment model, COMPASS-CAT score, risk assessment. In English databases, the terms included “venous thrombosis, thrombosis, pulmonary embolism, venous thromboembolism, deep venous thrombosis, deep vein thrombosis, pulmonary thromboembolism, phlebothrombosis; neoplasms, cancer, tumor, carcinoma, malignancy, oncology, malignant neoplasm, benign neoplasm; COMPASS-CAT, COMPASS-CAT risk assessment model, COMPASS-CAT risk assessment scale, COMPASS-CAT risk score, risk assessment”. The search strategy was designed by combining MeSH terms and free text terms and was adjusted according to each database. Additionally, manual searches were conducted to trace the references of the included studies to avoid omissions. An example of the search strategy in PubMed is as follows:

#1 venous thrombosis OR thrombosis OR pulmonary embolism OR venous thromboembolism OR deep venous thrombosis OR deep vein thrombosis OR pulmonary thromboembolism OR phlebothrombosis.

#2 neoplasms OR cancer OR tumor OR carcinoma OR malignancy OR oncology OR malignant neoplasm OR benign neoplasm.

#3 COMPASS-CAT OR COMPASS-CAT risk assessment model OR COMPASS-CAT risk assessment scale OR COMPASS-CAT risk score.

#4 #1 AND #2 AND #3.

Literature screening and data extraction

Based on the objectives of this study and the inclusion and exclusion criteria, two researchers independently conducted a preliminary screening based on the titles and abstracts of the literature. In cases of disagreement, a decision was made in consultation with a third researcher. After excluding duplicate and irrelevant literature, the full texts were further reviewed, and those that could not be accessed or did not meet the inclusion and exclusion criteria were excluded. Data extracted included the first author, publication year, country, sample size, cancer type, gold standard, study type, cutoff values, true positives, false positives, false negatives, and true negatives. If data were incomplete, attempts were made to contact the original authors via email to obtain the relevant data. If relevant data were ultimately unobtainable, the literature was excluded.

Quality assessment

The quality of the included studies was independently assessed by two researchers according to the Cochrane-recommended diagnostic test accuracy study bias assessment tool (QUADAS-2), comprising 13 items. Each literature was evaluated against these items with three answers: “Yes,” “No,” and “Unclear.” A study was classified as high quality (Grade A) if it fully met the criteria; medium quality (Grade B) if it partially met the criteria; and low quality (Grade C) if it did not meet the criteria. In cases of disagreement, discussions were held, or a decision was made by a third researcher.

Statistical methods

In accordance with the QUADAS-2 criteria, RevMan 5.3 was used to fill in the quality of the included studies and to create diagrams. Meta-analysis of diagnostic studies was conducted using Stata MP 17.0, while the pooled SEN, SPE, PLR, NLR, DOR, and 95% confidence interval (CI) were calculated. Summary Receiver Operating Characteristic (SROC) curves were drawn, and the area under the ROC curve (AUC) was estimated. Heterogeneity analysis was performed using the I² test, with I² > 50% and P < 0.05 indicating significant heterogeneity, prompting further investigation into sources of heterogeneity. If I² < 50% and P > 0.05, heterogeneity among the included studies was considered low. Sensitivity analysis was conducted to evaluate the stability of the study results by assessing the impact of excluding any one study on the overall effect size. Deeks’ funnel plot asymmetry test was used to assess publication bias, with P > 0.05 indicating no publication bias.

Results

Literature search results

A systematic search yielded a total of 4,948 related articles. After deduplication using EndNote X9.1 software, 4,007 articles remained. Upon reading titles and abstracts, 3,987 articles were excluded, including 3,792 articles irrelevant to the topic, 3 conference abstracts, and 9 reviews. Full-text review further excluded 7 studies for incomplete or duplicate data. Ultimately, 13 studies [9, 12,13,14,15,16,17,18,19,20,21,22,23] were included. The literature screening process and results are shown in Fig. 1.

Fig. 1
figure 1

Literature screening process and results

Full size image

Basic characteristics of included literature

The 13 included articles comprised a total of 8,665 cases, among which VTE occurred in 1,089 cases. The basic characteristics of the included literature are presented in Table 1.

Table 1 Basic Characteristics of Included Literature
Full size table

Quality assessment of included literature

The quality assessment results of the included literature are shown in Fig. 2.

Fig. 2
figure 2

Risk of bias assessment results

Full size image

Meta-analysis results

The 13 studies were quantitatively meta-analyzed. The pooled SEN was 0.76 [95%CI (0.61, 0.86)], with high heterogeneity among the studies (I² = 96.23%). The pooled SPE was 0.67 [95%CI (0.52, 0.79)], with high heterogeneity (I² = 99.69%). Pooled PLR was 2.3 [95%CI (1.7, 3.1)]; Pooled NLR was 0.36 [95%CI (0.25, 0.52)]; DOR was 6 [95%CI (4, 10)]; and the AUC was 0.77 [95%CI (0.74, 0.81)]. The forest plots for sensitivity and specificity are shown in Fig. 3, and the SROC fitting curve is in Fig. 4.

Fig. 3
figure 3

Sensitivity and specificity forest plots

Full size image
Fig. 4
figure 4

SROC fitting curve

Full size image

Sensitivity analysis

Sensitivity analysis of the included studies revealed anomalies in two papers [14, 21], as seen in Fig. 5. After excluding these two anomalous studies, the quantitative synthesis showed that pooled SEN was 0.77 [95%CI (0.60, 0.88)], with I² = 97.48%; Pooled SPE was 0.66 [95%CI (0.48, 0.80)], with I² = 99.69%. Pooled PLR, NLR, DOR, and AUC were 2.2 [95%CI (1.6, 3.2)], 0.36 [95%CI (0.23, 0.54)], 6 [95%CI (4, 10)], and 0.78 [95%CI (0.74, 0.81)], respectively. The minimal changes in values after exclusion indicate good stability of the meta-analysis results.

Fig. 5
figure 5

Sensitivity analysis

Full size image

Publication bias

The Deeks’ test result showed no significant publication bias in the included studies (P = 0.48). The publication bias result is presented in Fig. 6.

Fig. 6
figure 6

Publication bias test

Full size image

Meta-regression and subgroup analysis

Meta-regression was used to identify potential sources of heterogeneity among the included literature, considering factors like country and diagnostic setting, cancer type, study type, and sample size. Results showed that country and diagnostic setting significantly influenced the heterogeneity of the combined sensitivity results (P < 0.001), suggesting that they were potential sources of heterogeneity. Subgroup analysis based on country and diagnostic setting, cancer type, study type, and sample size revealed that the sensitivity of the COMPASS-CAT scale in non-Chinese outpatient settings was 0.94, with an AUC of 0.85, whereas diagnostic accuracy of the COMPASS-CAT thrombosis risk assessment scale was higher in foreign outpatient settings., The specificity of this scale in assessing venous thrombosis risk was higher in lung cancer patients (specificity: 0.75). In prospective studies, sensitivity, specificity, DOR, and AUC were all higher than those in retrospective studies. The results of the subgroup analysis are presented in Table 2.

Table 2 Subgroup Analysis Results
Full size table

Discussion

The incidence, recurrence rate, and mortality of VTE in patients with active cancer are high, making it a common and life-threatening complication and the second leading cause of death. Therefore, preventing VTE is crucial [24]. Early identification of high-risk patients for VTE and targeted prevention and treatment require reliable assessment tools [25]. Guidelines recommend the Caprini and Padua thrombosis risk assessment scales [5, 10], but both scales are based on non-cancer patients and do not consider the impact of cancer-related factors on thrombosis risk. The Khorana thrombosis risk assessment scale, designed for outpatient chemotherapy patients, predicts VTE risk based on cancer site and pre-chemotherapy blood cell counts. However, multiple reports indicate poor predictive ability of this scale across various cancer types [26,27,28]. PROTECHT and CONKO scores refine the Khorana scale for assessing thrombosis risk. PROTECHT incorporates chemotherapy agents like gemcitabine and platinum-based drugs [29], while CONKO, focusing on advanced pancreatic cancer patients treated with low molecular weight heparin, opts for performance status (PS) over BMI. Although PS scores have been shown to be associated with an increased risk of venous thromboembolism in hospitalized cancer patients [30], the CONKO score is almost identical to the Khorana score. A study by Van Es [31] on 876 cancer patients found PROTECHT seems to be potentially more effective in distinguishing between low and high-risk groups. The COMPASS-CAT thrombosis risk assessment scale, tailored for cancer patients, has been used in many domestic and international studies to predict VTE risk in solid tumor patients. Yan AR et al. [32] focused on outpatient lung cancer patients. Their results showed that the COMPASS-CAT score showed a significantly better discriminatory capability, but the pooled data was obtained from only three studies, with high heterogeneity. Furthermore, sensitivity and specificity analyzes were not performed in their study. Conversely, our study conducted a meta-analysis of eligible 13 studies, involing 8665 patients with lung cancer, breast cancer, colorectal cancer, or ovarian cancer showed that the COMPASS-CAT assessment scale had a pooled sensitivity of 0.76 and a pooled specificity of 0.67 for VTE prediction, with a false negative rate of 0.24% and a false positive rate of 0.33%, indicating that its clinical use could easily identify at-risk patients, but also lead to false positives for VTE risk. The AUC was 0.77, which was between 0.7 and 0.9, indicating a certain level of accuracy for this assessment.

Subgroup analysis in our study showed that the pooled sensitivity and AUC of the COMPASS-CAT assessment scale in foreign outpatient settings was 0.94 and 0.86, respectively, which were significantly higher than those in domestic inpatients. This may be due to the scale being originally developed based on foreign outpatient data, making it better at identifying VTE risk in these patients, and thus having higher diagnostic value. The AUC of 0.78 in inpatients suggests that the COMPASS-CAT assessment scale also has a certain level of accuracy in domestic inpatient settings.

This study has several limitations. First, the included studies are highly heterogeneous. Although factors such as country and diagnostic setting, cancer type, study type, and sample size have explained the sources of heterogeneity to some degree, heterogeneity remains high. This may be related to differences in case selection, gold standards, and tumor treatment protocols among the studies. Not all studies used consistent thresholds. Despite only two employing different ones and the ‘shoulder-arm’ shape of the sROC curve is not prominent, threshold effects on outcomes can’t be disregarded. Therefore, sensitivity analysis was conducted to address this issue, and the results demonstrated good stability. Future research should aim to define optimal risk stratification thresholds. Second, the study focuses on lung cancer, limiting broader applicability. Third, most of the included literature is retrospective, with fewer prospective studies. This may be because patients require regular ultrasound and imaging examinations, increasing the workload of medical staff. Furthermore, patients need to visit hospitals repeatedly, leading to reduced compliance, and they pay less attention to VTE, making it difficult to conduct prospective studies. Last, only Chinese and English literature was included, potentially introducing selection bias.

In conclusion, the COMPASS-CAT thrombosis risk assessment scale is a relatively objective tool for predicting VTE risk in cancer patients. It has good sensitivity and specificity, and it holds a diagnostic value in identifying the risk of VTE in cancer patients, aiding healthcare professionals in early identification of patients at high risk for VTE. However, the number of studies included in this research is limited and the heterogeneity among them is significant. Therefore, large-scale, multicenter, high-quality prospective studies are still needed to validate these findings.