Our research protocol was registered with the International Prospective Register of systematic reviews (PROSPERO, CRD42021241699). Two authors independently searched Pubmed, Embase, and Cochrane to identify studies that compared the efficacy or safety of LMWHs with other treatment in the elderly (from inception to July 22, 2020). The following search terms were included: “low-molecular-weight,” “LMWH,” “nadroparin,” “enoxaparin,” “dalteparin,” and many other generic and trade names of LMWHs; “elderly,” “aged,” “elder people,” “old people,” “the old,” “old man,” and “aging.” The search details are provided in Supplementary Tables S1–S3. We also evaluated the bibliographies of published studies. The cutoff age (60 years old) for the elderly is defined by the World Health Organization (World Health Organization, 1989; Zhang et al., 2020). Studies were considered eligible if they fulfilled the following criteria: 1) studies that included patients aged ≥60 years; 2) interventions that were specific kinds of LMWHs, and the control group were unrestricted; 3) randomized controlled trials (RCTs); and 4) studies published in English. Exclusion criteria were 1) studies that did not report the interest endpoints; 2) the full-text version could not be acquired online; and 3) studies that were not relevant.

Efficacy endpoints included incidence of VTE or VTE-related death, deep venous thrombosis (DVT), and pulmonary embolism (PE). VTE or VTE-related death was defined as asymptomatic proximal DVT, symptomatic proximal DVT, symptomatic distal DVT, symptomatic nonfatal PE, or VTE-related death; for studies reporting only DVT or PE, the same data were also used for VTE or VTE-related death. DVT and PE were defined by the included studies (Fuentes et al., 2019). Safety endpoints included clinically relevant bleeding, major bleeding, minor bleeding, and all-cause death. Clinically relevant bleeding was defined as all bleeding events, and the data were used for clinically relevant bleeding when studies reported only major bleeding. Major and minor bleeding were defined according to the International Society on Thrombosis and Hemostasis criteria (Schulman and Kearon, 2005); All-cause death included VTE-related death and death for other reasons. The definition of outcomes in original RCTs is provided in Supplementary Table S5.

Two researchers (Yang and Zhou) screened the studies independently. Studies were preliminarily screened according to the title and abstract and finally included or excluded according to the full text. Studies were included only if they met the inclusion criteria (data were also collected from some subgroups of RCTs, meta-analysis, and pooled analyses). If there was any disagreement, another author also independently evaluated the studies, after which the decisions were finalized through a group discussion.

A unified data extraction form was used to extract the relevant data from engaged RCTs, including 1) characteristics of the studies: title, first author, year of publication, country, number of centers, treatment, study object, sample size, and so on; 2) patient characteristics: age, gender, weight, and definition of the elderly; 3) interest endpoints: VTE or VTE-related death, DVT, PE, clinically relevant bleeding, major bleeding, minor bleeding, and all-cause death. If the related events were reported as a percentage, the figure was rounded.

The study quality of publications included in this analysis was evaluated by two researchers using the Cochrane Collaboration’s risk of bias tool, which evaluated seven possible biases: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessments, incomplete outcome data, selective outcome reporting, and other potential bias. The potential for publication bias was assessed by Begg funnel plots.

All of our network analysis was performed using the STATA statistical software (version 13.0). Node-splitting analysis was used to calculate the inconsistencies of our models. Pooled results of direct and indirect comparisons were reported as relative ratios (RR) and 95% confidence intervals (CI). The I ² test was calculated as a quantitative measure of heterogeneity, with I ² ≥ 50% considered as substantial heterogeneity (Yatabe et al., 2018). The random effects model was used when I ² was >50%; otherwise, the fixed effect model was used. Ranking probabilities were calculated by the surface under the cumulative ranking analysis (SUCRA) for each outcome to increase the estimated precision of the effect sizes. The larger the SUCRA value, the better the LMWHs (Zhou et al., 2018). Considering the difference of patients involved, we further performed subgroup analysis based on patient characteristics (patients with DVT, patients with cardiac disease, medical patients, and patients >65 years old, > 70 years old, > 75 years old) and study characteristics (sample size >100, follow-up shorter than 60 days, follow-up longer than 60 days, single-center, multicenter, RCTs before 2010, and RCTs after 2010).

We finally included 27 RCTs in our network meta-analysis. The flow of our study selection is shown in Figure 1. A total of 30,441 patients were included, including 13,351 in the LMWHs group and 17,090 in the control group.

The characteristics of included RCTs are shown in Supplementary Table S4. The LMWHs included certoparin (n = 1), enoxaparin (n = 21), tinzaparin (n = 2), nadroparin (n = 1), and dalteparin (n = 2). The population included DVT (n = 4), cardiac disease (n = 6), medical (n = 7), and orthopedic surgery (n = 10) patients. All subjects were 60 years or older. The data were from five original RCTs (Dahan et al., 1986; Veiga et al., 2000; Lederle et al., 2006; Riess et al., 2010; Leizorovicz et al., 2011), 10 subgroup analyses of 10 main RCTs (Samama et al., 1999; Lopez-Beret et al., 2001; Simoons, 2001; Leizorovicz et al., 2004; Daskalopoulos et al., 2005; Antman et al., 2006; Hull et al., 2010; Montalescot et al., 2011; Cohen et al., 2016; Kim et al., 2016), one meta-analysis (Sardar et al., 2014), and four pooled analyses of RCTs (Sinnaeve et al., 2006; Friedman et al., 2010; Turpie et al., 2011; Pineo et al., 2013). The risk bias of included RCTs is shown in Figure 2.

There was no direct comparison among different LMWHs. When compared with other treatments, LMWHs were definitely more effective than placebo in preventing VTE or VTE-related death (RR 0.36, 95% CI 0.25–0.53) and safer than acenocoumarol regarding risk of bleeding events (RR 0.67, 95% CI 0.49–0.90). However, LMWHs were less effective than novel oral anticoagulant (NOACs) in preventing VTE or VTE-related death (RR 1.59, 95 %CI 1.33–1.91). LMWHs show efficacy and safety similar to that of UFHs (Supplementary Table S6).

The endpoint of VTE or VTE-related death was reported in 20 RCTs and involved four types of LMWH (certoparin, n = 1; enoxaparin, n = 16; nadroparin, n = 1; tinzaparin, n = 2) (Figure 3A). As shown in Figure 3B, the four kinds of LMWHs show no significant difference in preventing incidence of VTE or VTE-related death (Figure 4A). The rank probabilities of the four LMWHs are shown in Figure 5A, whereby tinzaparin provided the most benefit in VTE or VTE-related death (SUCRA 68.8%, cumulative probability 42.3%), whereas nadroparin provided the least benefit (SUCRA 34.3%, cumulative probability 21.2%).

The indirect analysis of DVT is shown in Figure 3B, which included nine RCTs and involved four types of LMWH (certoparin, n = 1; enoxaparin, n = 6; nadroparin, n = 1; tinzaparin, n = 1). There was no significant difference in the incidence of different LMWHs (Figure 4B). Tinzaparin performed best with regard to efficacy in VTE (SUCRA 73.0%, cumulative probability 57.5%), and enoxaparin had the worst efficacy (SUCRA 32.9%, cumulative probability 2.0%) (Figure 5B).

PE was reported in seven RCTs and included three LMWHs (certoparin, n = 1; tinzaparin, n = 1; enoxaparin, n = 5) (Figure 3C). The three LMWHs showed no significant difference regarding PE (Figure 4C). Certoparin was ranked first (SUCRA 82.9%, cumulative probability 66.0%) and tinzaparin was ranked last (SUCRA 19.4%, cumulative probability 10.6%) (Figure 5C).

Clinically relevant bleeding included major and minor bleeding. The results of indirect analysis are shown in Figure 3D, including 25 RCTs and five kinds of LMWHs (enoxaparin, n = 19; tinzaparin, n = 2; nadroparin, n = 1, dalteparin, n = 2; certoparin, n = 1). The five kinds of LMWH showed no significant difference in the incidence of clinically relevant bleeding (Figure 4D). The rank probabilities indicated that nadroparin had a 77.0% probability of being the best therapy (SUCRA: 84.8%) to prevent incidence of clinically relevant bleeding, whereas certoparin was the worst therapy (SUCRA: 22.8%, cumulative probability: 2.5%) (Figure 5D).

Incidence of major bleeding was reported in 15 RCTs (one was excluded because no incidence of major bleeding was observed) and included five LMWHs (Figure 3E). Eight RCTs reported incidence of minor bleeding with five LMWHs involved (Figure 3F). The indirect comparisons of the two endpoints are shown in Figures 4E,F. Different LMWHs showed no significant difference in both major and minor bleeding. Dalteparin showed the best efficacy for major bleeding (SUCRA 72.1%, cumulative probability 28.3%), and certoparin showed the worst efficacy (SUCRA 29.4%, cumulative probability 6.7%) (Figure 5E). Nadroparin had a 71.1% probability of being the best therapy to prevent the incidence of minor bleeding (SUCRA 84.1%), whereas enoxaparin was ranked last (SUCRA 20.7%, cumulative probability 0.7%) (Figure 5F).

A total of 14 RCTs reported the incidence of all-cause death after treatment with anticoagulants with five LMWHs involved (enoxaparin, n = 8; tinzaparin, n = 2; nadroparin, n = 1; dalteparin, n = 1; certoparin, n = 1; one was excluded because no incidence of major bleeding was observed) (Figure 3G). Indirect comparison indicated that no LMWHs showed significant superiority regarding the risk of all-cause death (Figure 4G). The rank probabilities indicated that tinzaparin delivered the greatest benefit for the risk of all-cause mortality (SUCRA 84.2%, cumulative probability 40.7%), and certoparin provided the least benefit (SUCRA 26.0%, cumulative probability 2.3%) (Figure 5G).

For comparisons involving 10 or more studies, Begg funnel plots for studies of VTE or VTE-related death, clinically relevant bleeding, major bleeding, and all-cause death were symmetrical (Supplementary Figure S1). Thus, there was no statistically significant evidence of publication bias in the studies reported herein.

To verify the stability of our results, we further performed subgroup analyses in populations with different characteristics: patients with DVT (Supplementary Table S7), internal medical patients (Supplementary Table S8), patients with cardiac disease (Supplementary Table S9) (orthopedic surgery patients were excluded because they used fewer than two types of LMWH), patients >65 years old (Supplementary Table S10), patients >70 years old (Supplementary Table S11), and patients >75 years old (Supplementary Table S12); and studies with different characteristics: sample size >100 (Supplementary Table S13), follow-up time <60 days (Supplementary Table S14), follow-up time >60 days (Supplementary Table S15), multicenter (Supplementary Table S16), single-center (Supplementary Table S17), RCTs before 2010 (Supplementary Table S18), and RCTs after 2010 (Supplementary Table S19). Different LMWHs showed similar efficacy and safety with regard to the incidence of VTE or VTE-related death, clinically relevant bleeding, and all-cause mortality. In addition, tinzaparin showed the greatest overall benefit in VTE and VTE-related death and all-cause mortality in our ranking analysis, whereas nadroparin was ranked first in reducing the incidence of bleeding events.

Considering that enoxaparin constituted the majority of LMWHs we included, we conducted a subgroup analysis that excluded the RCTs of enoxaparin. After these 21 RCTs were discarded, dalteparin was associated with a decreased risk of clinically relevant bleeding (RR 1.76, 95% CI 1.10–2.82) in comparison with cetoparin (Supplementary Table S20).