The present study was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) and prospectively registered on PROSPERO (ID: CRD42023417848).

Inclusion criteria: (1)

Patients with confirmed deep venous thrombosis were included in the study (Figure 1);

Exclusion criteria: (7)

The research types were meta-analyses, reviews, guidelines, expert opinions, etc.;

We systematically searched PubMed, Embase, Cochrane and Web of Science databases until 7 April 2023. The search was conducted in the form of subject terms and free words, with no restrictions on regions. The detailed search strategies are shown in Supplementary Material S1.

In this systematic review, EndNote was first used to de-duplicate the retrieved literature, and then we initially screened eligible original studies by reading the titles and abstracts and downloaded their full text. Next, we further read the full text to screen the original studies that met the requirements of our review.

Before data extraction, a standardized spreadsheet was used to develop a data extraction plan. The extraction information included title, first author, publication year, author's country, study type, patient source, background disease, diagnostic criteria for post-thrombotic syndrome, follow-up duration, number of post-thrombotic syndrome cases, the total number of cases, number of post-thrombotic syndrome cases in the training set, the total number of cases in the training set, generation method of the validation set, overfitting method, number of post-thrombotic syndrome cases in the validation set, the total number of cases in the validation set, missing value processing method, variable screening, type of model used, modeling variables.

The above literature screening was carried out by two investigators (Tong Yu and Jialin Song). After the completion of cross-checking, the third researcher (Lingke Yu) was asked to assist in adjudicating if there was any dispute.

The original types of studies included in our systematic review were cohort or case-control studies, and thus we adopted the Newcastle-Ottawa Scale (NOS) to evaluate the quality of the included original studies (7). The variable applies to cohort studies and pathologically controlled studies, and the quality evaluation is mainly conducted from eight questions in three fields. Except for the full score of 2 points for comparability, the full score of the remaining seven questions is 1 point. A score of 7 to 9 was considered a high-quality study and a score of 4 to 7 was defined as a moderate-quality study.

For this study, two evaluators (Tong Yu and Jialin Song) performed a risk of bias assessment using NOS and cross-checked. If there was any discrepancy in the evaluation, the third evaluator (Wanlin Deng) was invited to participate in the decision.

In this systematic review, we pooled the incidence rates in cohort studies. The incidence of PTS was calculated as the percentage of subjects with PTS in each cohort study. The odds ratio (OR) of each predictor was derived from multivariate logistic/cox regression or ORs adjusted for other confounders. The c-index, sensitivity, and specificity of the prediction model constructed based on these predictors were used to reflect its overall accuracy. The primary outcomes encompassed the incidence of PTS and the c-index, sensitivity, and specificity of the prediction model. The secondary outcome was the ORs of predictors under multivariate analysis.

When performing independent predictor analyses, to avoid the influence of confounding factors, we extracted ORs from multivariate logistic/cox regressions from original studies or those adjusted for other confounding factors for meta-analysis. A random-effects model was used when the heterogeneity index I² > 50% in the meta-analysis of the OR of the predictors and subgroup analysis was performed according to the different levels of each predictor.

In addition, our outcome measures included indicators to evaluate the accuracy of the machine learning model (c-index, sensitivity, and specificity). When evaluating the c-index, we prioritized the use of a random-effects model for data processing due to differences in the inclusion of variables and inconsistent parameters across machine-learning models.

For the c-index with missing 95% confidence intervals and standard errors, we referred to the study of Debray TP et al. (8) and estimated the standard errors. Also, we performed a meta-analysis of sensitivity and specificity using a bivariate mixed-effects model. The meta-analysis of this study was conducted based on R4.2.0 (R development Core Team, Vienna, http://www.R-project.org).

A total of 2,375 papers were retrieved, 603 of which were duplicates, and 28 eligible studies were included according to the titles and abstracts. Twenty studies (9–28) were retained after full-text review. The literature screening process is shown in Figure 2.

The present study included 20 studies altogether, with a total number of 8,512 cases and 3,219 cases of morbidities. The period of publication is mainly from 2016 to 2023. The countries of publication included China (9, 11, 12, 18, 21), Norway (10, 28), Canada (13), the Netherlands (14, 16, 25), Japan (15), Indonesia (17), Australia (19), Switzerland (20), Israel (22, 23), Pakistan (24) and Poland (26, 27). For the background of morbidity, only one of the studies (10) reported a background of pregnancy-related venous thrombosis, while the remaining 19 had a background of deep vein thrombosis (DVT). As for the diagnostic criteria of PTS, two studies (14, 15) used the CEAP score (CEAP score ≥3) and the remaining 18 used the Villalta score (Villalta score ≥5). Among the 18 studies adopting the Villalta score, four used a new scale developed on the basis of the Villalta score to assess PTS, one (18) employed the Villalta score combined with the APTSD score, two (22, 23) used the Villalta score with SOX-PTS score, and one (16) used Villalta score and LET grading score. The follow-up period of the articles ranged from 3 months to 72 months. Seven articles (9, 11, 12, 18, 21, 25) constructed risk prediction tools with independent validation sets, with a total of 21 prediction models constructed, which were mainly based on logistic regression. Sixteen of the included articles were cohort studies. Table 1 shows the basic information.

Newcastle-Ottawa Scale (NOS) was used to evaluate the quality of the included articles. There were six 9-point articles, seven 8-point articles, six 7-point articles, and one 6-point article. In the cohort studies, scores were not obtained mainly in the three areas of representativeness of the exposure cohort, assessment of outcome events, and completeness of follow-up. In case-control studies, there were no scores for control selection and non-response rates. The specific assessment results are provided in Table 2.

Our study showed that the follow-up of PTS was mainly within 3 years, with a focus on 2-year incidence. The incidence of PTS in our study was 37.5% (95% CI: 27.8–47.7%). We also found huge differences in incidence rates among countries, which may be related to the level of medical treatment and local concern, and We should increase awareness of clinical prevention and control of PTS.

Independent predictors in our systematic review included old age, BMI, DVT location, history of venous thrombosis, smoking history, SOX-PTS score, varicose veins, and baseline Villalta score. The prediction models constructed based on these predictors had good accuracy, with a c-index of 0.77 (95% CI: 0.74–0.80), sensitivity of 0.75 (95% CI: 0.68–0.81), and specificity of 0.69 (95% CI: 0.60–0.77) in the training set. In the validation set, the c-index, sensitivity and specificity of the prediction models were 0.74 (95% CI: 0.69–0.79), 0.71 (95% CI: 0.64–0.78) and 0.72 (95% CI: 0.67–0.76) respectively.

Most reviews on PTS focus on summarizing relevant studies on PTS (e.g., long-term therapy (29), anticoagulant selection (30), catheter therapy (31, 32), etc.), describing and explaining the main results, findings and trends of the studies, and drawing conclusions from them. The PTS reviews describe the pathophysiology, risk factors, diagnosis, prevention, and treatment of PTS to improve our understanding of the disease and guide treatment.

As found in the review, two articles [(Kearon et al., 2020 (29) and Kruger et al., 2019 (30)] proposed male as an risk factor of PTS, which is contradictory with our findings. In this study, male OR = 1.189 (95%CI; 0.780–1.812) was not a risk factor of PTS. Meanwhile, Iding AFJ et al., 2023 (33) noted in their study that estrogen use in women reduces the incidence of PTS, so further research is needed to determine whether gender can be used as an independent predictor. Unprovoked DVT is DVT occurring in the absence of a transient risk factor. And the review of Kruger et al. (2019) described unprovoked DVT as an independent predictor, which differs from our findings, as in this paper unprovoked DVT OR = 1.291 (95% CI; 0.742–2.246) could not be regarded as an independent predictor. However, it has been suggested in one article (33) that Unprovoked DVT is closely related to the patient's cardiovascular disease. Therefore, it can be inferred that whether Unprovoked DVT can be an independent predictor is inextricably linked to the status of its study patients.The above may be due to the heterogeneity of the DVT patient population in the studies that included the above predictors and the diversity of clinical manifestations of PTS, leading to different results for independent predictors (34). Based on the present study and several previous studies, the main independent predictors of PTS were old age, obesity, proximal DVT, venous thromboembolism history, and baseline Villalta score. Therefore, the remaining controversial predictors require more research to explore the independent predictors of PTS.

Galanaud et al. 2018 (35), Kearon et al. 2020, INOVICH et al. 2017 (36), and Makedonov et al. 2020 (37) all noted in their studies that the prevalence of PTS was about 20%–50%, of which 5%–10% of patients would develop severe PTS. Kruger et al. 2019 (30) pointed out in a study that the incidence of PTS was 40%. In a meta-analysis (38) on the combined risk of PTS, mainly the risk of PTS caused by distal venous thrombosis, it was estimated that 20% of patients with venous thrombosis would develop PTS, among which 20% would develop severe PTS. The main background of this study is proximal venous thrombosis, the prevalence of which is 37.5% in this paper, probably due to differences in study countries, study populations, tools used to assess PTS, and the time interval between DVT and PTS assessment, making the prevalence of PTS reported in the studies highly variable.

Meanwhile, PTS is mentioned in several studies as significantly reducing the quality of life and being expensive, so early prediction and intervention for PTS is an important direction for treatment (30, 36, 37, 39). Development of risk predictors to predict the risk of PTS at the time of diagnosis of DVT is also ongoing to help guide the long-term treatment of patients with DVT (36). The purpose of this systematic meta-analysis was to comprehensively analyze data on incidence, predictors, and the accuracy and feasibility of predictive models from multiple studies so as to evaluate the predictive value of predictors and machine learning-based predictive models. This study allows us to further explore the causes of morbidity fluctuations, analyze the value of independent predictors and provide a reference for the development of more accurate and reliable machine prediction models, and the conclusions obtained can contribute to future relevant medical decision-making and disease prevention and treatment.

In this review, a consensus was reached by two independent reviewers at the literature selection stage to reduce the risk of selection bias. Further, this study is the first to focus on the incidence and predictors of PTS from the perspective of a systematic review. A comprehensive analysis of multiple independent predictors eliminates bias and noise in the study, thus estimating the overall effects and confidence intervals more accurately. Meanwhile, by examining the heterogeneity and consistency in different studies, the accuracy and feasibility of the early prediction models of PTS were further revealed, and the reliability of the conclusions was improved.

There are also some limitations in this paper. The number of cases in our included cohort studies was small, which may be due to the insufficient attention paid to venous thrombosis in clinical practice or some countries. Also, some of the predictors were included in a small number of literature reports and need to be interpreted with caution. Among the included literature, a small part of them have built risk prediction models, and thus more large-sample studies are needed for verification.