key: cord-0926719-bx4g2mj5
authors: Li, Jiawen; Zhou, Kaiyu; Duan, Hongyu; Yue, Peng; Zheng, Xiaolan; Liu, Lei; Liao, Hongyu; Wu, Jinlin; Li, Jinhui; Hua, Yimin; Li, Yifei
title: Value of D-dimer in predicting various clinical outcomes following community-acquired pneumonia: A network meta-analysis
date: 2022-02-23
journal: PLoS One
DOI: 10.1371/journal.pone.0263215
sha: 3f199a56d377a3c50589eaf57654f729db14996a
doc_id: 926719
cord_uid: bx4g2mj5

BACKGROUND: Whether high D-dimer level before treatment has any impact on poor outcomes in patients with community-associated pneumonia (CAP) remains unclear. Therefore, we conducted the first meta-analysis focusing specifically on prognostic value of high D-dimer level before treatment in CAP patients. METHODS: Pubmed, Embase, the Cochrane Central Register of Controlled Trials and World Health Organization clinical trials registry center were searched up to the end of March 2021. Randomized clinical trials (RCT) and observational studies were included to demonstrate the association between the level of D-dimer and clinical outcomes. Data were extracted using an adaptation of the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS-PF). When feasible, meta-analysis using random-effects models was performed. Risk of bias and level of evidence were assessed with the Quality in Prognosis Studies tool and an adaptation of Grading of Recommendations Assessment, Development, and Evaluation. Data were analyzed using STATA 14.0 to complete meta and network analysis. MAIN OUTCOMES AND MEASURES: Besides d-dimer levels in CAP patients with poor outcomes, we also analyzed proportion of patients with or without poor outcomes correctly classified by the d-dimer levels as being at high or low risk. The poor outcome includes severe CAP, death, pulmonary embolism (PE) and invasive mechanical ventilators. RESULTS: 32 studies with a total of 9,593 patients were eventually included. Pooled effect size (ES) suggested that d-dimer level was significantly higher in severe CAP patients than non-severe CAP patients with great heterogeneity (SMD = 1.21 95%CI 0.87–1.56, I(2) = 86.8% p = 0.000). D-dimer level was significantly elevated in non-survivors compared to survivors with CAP (SMD = 1.22 95%CI 0.67–1.77, I(2) = 85.1% p = 0.000). Prognostic value of d-dimer for pulmonary embolism (PE) was proved by hierarchical summary receiver operating characteristic curve (HSROC) with good summary sensitivity (0.74, 95%CI, 0.50–0.89) and summary specificity (0.82, 95%CI, 0.41–0.97). Network meta-analysis suggested that there was a significant elevation of d-dimer levels in CAP patients with poor outcome than general CAP patients but d-dimer levels weren’t significantly different among poor outcomes. CONCLUSION: The prognostic ability of d-dimer among patients with CAP appeared to be good at correctly identifying high-risk populations of poor outcomes, suggesting potential for clinical utility in patients with CAP.

Introduction Community-acquired pneumonia (CAP) is defined as pneumonia acquired outside the hospital and has led to life-years lost globally [1] . Approximately 6.6% to 16.7% of hospitalized patients with CAP would enter the severe stage. Unfortunately, mortality rate is supposed to reach up to 28.8% among patients with severe CAP [2] . A considerable proportion of patients with CAP in the emergency department can be treated as outpatients. However, unpredictable disease course and uncertain outcomes are challenges for clinicians, hindering the early identification of patients at high risk. Several risk scores, such as the pneumonia severity index (PSI) and CURB-65 (confusion, urea nitrogen, respiratory rate, blood pressure, age �65 years), can be used to assess the severity of pneumonia and predict prognosis [3] [4] [5] . For CAP, low risk was defined as PSI score classes I to III and CURB-65 score class 1. High risk was defined by PSI score classes IV-V and CURB-65 score classes 2-5. However, they were more suitable for research than clinical decision and their performance is still controversial [6] . C-reactive protein (CRP) [7, 8] and procalcitonin (PCT) [9, 10] had been reported be a prognostic marker of outcome during severe CAP and ventilator-associated pneumonia (VAP). D-dimer is the fibrinolytic degradation products of crosslinked fibrin and is applied as a useful marker for the diagnosis of pulmonary embolism. It mirrors the severity of infection and has emerged as the extensively studied and promising blood biomarker for the risk stratification of patients with CAP [11] [12] [13] .

However, whether d-dimer level is an ideal index to predict the prognosis of communityacquired pneumonia remains debatable [14, 15] , because no relevant studies focusing on ddimer levels before treatment in patients with CAP were available to conduct meta-analyses previously. For this reason, a meta-analysis was performed to systematically and quantitatively evaluate the prognostic accuracy of the d-dimer level before treatment in CAP. To our knowledge, this is the first meta-analysis specifically focusing on d-dimer levels in patients with CAP. In consideration of different disease process and different therapeutical strategies of COVID-19 due to its distinct biology and pathogen, we haven't included publications of COVID-19 to prevent unsolvable heterogeneity.

Titles and abstracts of search results were screened independently (Jiawen Li, Hongyu Duan). The full texts of the remaining results were assessed independently by another 2 of us (Yifei Li, Yimin Hua) for inclusion based on predetermined criteria. Any discrepancies should be resolved through discussion, potentially with a third reviewer. We manually searched the reference lists of included studies and existing systematic reviews as well as all articles citing the included studies on Google Scholar. Potentially relevant reports were then retrieved as complete manuscripts and assessed for compliance to inclusion and exclusion criteria.

In accordance to the objectives of our meta-analysis, we developed a 'Population, Index prognostic factor, Comparator prognostic factor, Outcome, Timing, Settings' (PICOTS) framework adapted from the guideline proposed by Riley et al [16] . Our study inclusion criteria were as follows according to PICOTS framework: (1)Population: CAP patients with a welldefined diagnostic reference standard for pneumonia; (2)Index prognostic factor: before-treatment d-dimer levels measured by enzyme-linked immunosorbent assay (ELISA), quantitative latex assay, immunoturbidimetric or other convinced assay machine; (3)Outcome: severe CAP patients identified by PSI, CURB-65 or other definitive scale, pulmonary embolism (PE), death or invasive ventilation. (6) If studies were based on overlapping patients, the most completed one was chosen. We used the following criteria for study exclusion: (1) patients acquired pneumonia in clinical settings (e.g. VAP or hospital-acquired pneumonia (HAP)); (2) studies were published in other language; (3) conference abstracts, reviews, case reports, and experiment studies.

Two reviewers (Jiawen Li, Hongyu Duan) independently extracted study data and assessed risk of bias, with discrepancies resolved by a third investigator in a blinded fashion. Quality of evidence was assessed by the modified Grading of Recommendations Assessment, Development, and Evaluation system (GRADE) by consensus among the authors [19, 20] .

The essential data was extracted according to the modification of CHARMS (checklist for critical appraisal and data extraction for systematic reviews of prediction modelling studies) for prognostic factors (CHARMS-PF) [16] . When an included study reported different cut-off values, we chose one which made both sensitivity and specificity more than 50% as possible. When an included literature reported the same outcome at different follow-up timepoint (e.g. 7-days mortality and 30-days mortality), we chose the earliest one. We extracted data of PSI if included studies reported severity of CAP by both PSI and CURB-65. Mean and standard deviation were estimated based on sample size, median and quartile if included studies did not reported mean and standard deviation [21, 22] . All the baseline characteristics of included studies were shown on Table 1 .

The included studies were further investigated for risk of bias using an adapted version of the Quality in Prognosis Studies (QUIPS) tool [23] , which assessed the study-specific risk of bias across to six bias domains: study participation, study attrition, prognostic factor measurement, outcome, measurement, study confounding, statistical analysis and reporting. A study that satisfied the criteria of low risk of bias in all 6 domains was designated as having low overall risk of bias. A study with a high risk of bias in 1 or more domains was designated as having high overall risk of bias. Details on each signaling question of the QUIPS tool are elaborated on Table 2 . We did not exclude any publication with high risk of bias according to QUIPS.

Analyses were performed for both adjusted and unadjusted estimates. To combine comparative continuous data with dichotomous data, we transformed logarithm odds ratios to effect size, assuming a normal underlying distribution [24] Quantitative synthesis was first conducted by comparing the d-dimer levels of CAP patients with various outcome. The betweenstudy heterogeneity was evaluated by the χ2-based Q statistics and I 2 test, and a significant heterogeneity was as P<0. 1 [25] or I 2 >50%. When significant heterogeneity was observed, we would apply the random effects models for analysis. Otherwise, we would apply the fixed effects models. A sensitivity analysis was also conducted by sequential removal of each study. Here, we applied funnel plots as well as Egger's test [26] to assess publication bias. A two-sided P value of 0.05 was deemed as statistical significance.

The proportion patients with poor outcome correctly classified by the d-dimer levels as high risk was defined by dividing true-positive results by the sum of true-positive and falsenegative results. The proportion of patients without poor outcome correctly classified by the d-dimer levels as low risk was defined by dividing true-negative results by the sum of true-negative and false-positive results. It was similar in concept to sensitivity and specificity although sensitivity and specificity are more appropriately reported at a particular time point in prognostic studies. Dose-response meta-analysis (DRMA) was conducted only for adjusted outcomes with more than 3 categories of exposures. When pooled effects had significant heterogeneity, and included more than 9 studies, subgroup analyses were carried out based on methodologies of d-dimer measurements, study design, location, sample size, risk of bias, number of categories and effect size type.

For network meta-analysis, we evaluated global inconsistency by fitting consistency and inconsistency model [27] , and evaluated local inconsistency between direct and indirect estimates by using a node-splitting procedure [28] . In order to further quantify the d-dimer level of various outcome, we calculated the frequentist analogue of the surface under the cumulative ranking curve (SUCRA) for each outcome [29] .

Data was analyzed using STATA Version 14.0 [30]. The network was evaluated using frequentist multivariate meta-analysis (commands network meta and mvmeta) in Stata 14.0. Besides, publication bias and sensitivity analysis were also conducted by STATA version 14.0. 

Nine-hundred and twenty-five articles were retrieved from databases, of which 32 studies with a total of 9,593 patients were eventually included (Fig 1) . No additional relevant articles were identified in the bibliographies of the original articles. The characteristics of the included studies are listed in Table 1 . 

QUIPS tool had been used to assess the quality of included studies ( were evaluated as low risk of bias [43, 53] and the remaining 10 studies had medium risk of bias. The number of individual domains rated as high risk ranged from 0 to 3.

Severity. We excluded one study [59] for quantitative synthesis because it simply defined severe CAP as pneumonia with pleural effusion. Twelve studies [11, 31, 32, 34, 35, 37-43] including 1,394 participants reported the plasma d-dimer level of both severe CAP and nonsevere CAP patients. Pooled effect size (ES) suggested d-dimer level was significantly higher in severe CAP patients than non-severe CAP patients with great heterogeneity (SMD = 1.21 95% CI 0.87-1.56, I 2 = 86.8% p = 0.000) (Fig 2) and pooled result with adjusted OR [36, 43] (OR = 1.07 95%CI 1.01-1.13, I 2 = 60.2% p = 0.113) was consistent with continuous data (S1 Fig) . By subgroup analysis, conversion of median and quartiles, combination of mean and SD, 

methodologies of d-dimer measurements, high risk of bias and tools of d-dimer measurement did not contribute to heterogeneity significantly. Sensitivity analysis of pooled ES of continuous data suggested that no study contribute much to the pooled estimate as our findings remained consistent (S2 Fig). Funnel plots were used to assess publication bias. We detected asymmetry in the funnel plot which was further ascertained by Egger's test (P = 0.031 ; Fig 3) , suggesting the presence of publication bias.

Mortality. Mortality associated with CAP was an important clinical outcome. 12 articles had been included to make quantitative analysis. Among them, 10 articles [11, 38, 39, 41-43, 50, 52-54] including 4,117 participants reported the plasma D-dimer level of survivors and non-survivors with CAP, which confirmed significant elevation of D-dimer level in non- The presence of asymmetric distribution of funnel plots suggested that there might be publication bias in pooled result of continuous data (S5 Fig). Paradoxically, qualitative analysis by Egger's test did not indicate publication bias (p = 0.168).

Pulmonary embolism. Pooled effect based on continuous data from three studies suggest a significant elevation in CAP patients with PE than those without PE (SMD = 0.75 95%CI 0.11-1.38, I 2 = 84.5% p = 0.002) (S6 Fig). As there were only three enrolled studies, the publication bias and subgroup analysis were unavailable. However, data from a series of studies could be converted to the form of fourfold table of diagnostic test to demonstrate the prognostic value Network analysis. We pooled effects of different outcomes by network meta-analysis of frequentist statistics (S10 Fig) and loop inconsistency test suggested significant heterogeneity between direct and indirect comparisons (S11 Fig). The results provide evidence that there 

D-dimer in pneumonia was significant elevation of d-dimer levels in CAP patients with poor outcomes than general CAP patients. But wide 95% prediction interval cross null value (0) reminded us potential heterogeneity of included studies (Fig 6) . D-dimer levels weren't significantly different among poor outcomes although the SUCRA statistic showed that d-dimer level in CAP patients requiring mechanical ventilators ranked first, followed by non-survivors, severe patients and patients with PE. (S12 Fig). Funnel plots suggested potential publication bias based on its slight asymmetry (S13 Fig). Quality of evidence. Most of included studies are retrospective observational studies and had high or moderate risk of bias. Great heterogeneity existed in pooled effects and contributed to inconsistency of evidence. Based on GRADE system and the above considerations, the quality of evidence of our study should be low or very low. 

Our study is the first meta-analysis ever published about prognostic value of D-dimer in patients with CAP. It shows that elevated d-dimer level is significantly associated with CAP severity, mortality and PE occurrence in general analysis regardless their age, gender, race and region and the type of assay of d-dimer measurement. With the aid of method of diagnostic meta-analysis, we confirmed prognostic value of d-dimer for high-risk population of PE in patients with CAP. Although we were unable to establish a firm evidence on the independent prognostic value of d-dimer levels, our results on d-dimer levels were consistent and robust. Network meta-analysis further confirmed the evidence of conventional meta-analysis and suggested d-dimer levels were not significantly different in various poor outcomes. This biomarker may be helpful in the early identification of patients with high risk of poor outcomes to make special therapeutic strategy as soon as possible.

Previous studies reported that patients with elevated D-dimer levels were more likely to suffer from thromboembolism [62] , digestive cancer [63] , traumatic brain injury [64] , and aortic dissection [65] , which are associated with coagulation disorders. However, the pathophysiology of D-dimer elevation in CAP is only partially understood. Indeed, D-dimer elevation has also been observed in children and adults without any symptoms of pneumonia. Elevation of D-dimer had been reported to be correlated with several inflammatory and coagulation factors, including C-reaction protein, procalcitonin, IL-6, prothrombin time (PT), activated partial thromboplastin time (APTT) and thrombin time (TT) [7, 9, 66, 67] . Besides, several studies showed some patients with high d-dimer levels who died from severe CAP did not present obvious disorders of coagulation function [31] .

As an unconventional detection index, the value of d-dimer varied widely across different studies, resulting in great unmanageable heterogeneity in pooled effects. This is partly because of diverse measuring apparatus and assays. The primary outcome is problematic for time-toevent analyses, particularly if studies have short follow-up and significant censoring. On the other hand, it is difficult to make subgroup analysis because of limited number of included studies and insufficiently reported data to make stratification. D-dimer is only reported as a confounding factor in most of included studies so the data from them is unable to support to make dose-response meta-analysis. Moreover, given the heterogeneity in study designs and data reporting, as well as the lack of availability of individual patient data, meta-analysis of hazard ratios was not feasible. There is uncertain statistical bias in combine means and SDs into one group and estimating the sample mean and SD from the sample size, median, range and/ or interquartile range.

This study found that the prognostic ability of d-dimer to predict multiple poor outcomes among patients with CAP. But it is difficult to distinguish high-risk populations of different poor outcomes according to d-dimer levels. Additional, more rigorously structured research appears to be needed to better quantify the association of d-dimer levels with poor outcomes in patients with CAP and to demonstrate clinical utility. 

Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study

Incidence, direct costs and duration of hospitalization of patients hospitalized with community acquired pneumonia: A nationwide retrospective claims database analysis

Management-based risk prediction in community-acquired pneumonia by scores and biomarkers. The European respiratory journal

Validity of severity scores in hospitalized patients with nursing home-acquired pneumonia

Guidelines for the management of adult lower respiratory tract infections-full version. Clinical microbiology and infection: the official publication of the European Society of Clinical Microbiology and Infectious Diseases

Prognostic value of procalcitonin in pneumonia: A systematic review and meta-analysis. Respirology (Carlton, Vic)

Plasma ddimer levels correlate with outcomes in patients with community-acquired pneumonia

A rapid quantitative D-dimer assay at admission correlates with the severity of community acquired pneumonia

D-dimer levels in assessing severity and clinical outcome in patients with community-acquired pneumonia. A secondary analysis of a randomised clinical trial. European journal of internal medicine

What is the most reliable marker in the differential diagnosis of pulmonary embolism and community-acquired pneumonia? Blood coagulation & fibrinolysis: an international journal in haemostasis and thrombosis

Diagnostic value of D dimer in pulmonary embolism and pneumonia. Respiration; international review of thoracic diseases

Can the presence of significant coagulopathy be useful to exclude symptomatic acute pulmonary embolism? Anaesthesia and intensive care

D-dimer threshold increase with pretest probability unlikely for pulmonary embolism to decrease unnecessary computerized tomographic pulmonary angiography

Prevalence of venous thromboembolic events and diagnostic performance of the wells score and revised geneva scores for pulmonary embolism in patients with interstitial lung disease: a prospective study

Pulmonary Embolism in Pneumonia: Still a Diagnostic Challenge? Results of a Case-Control Study in 100 Patients

Prognostic value of albumin-red cell distribution width score in patients with severe community-acquired pneumonia. Annals of palliative medicine

The mortality risk factor of community acquired pneumonia patients with chronic obstructive pulmonary disease: a retrospective cohort study

Predictors for mortality from respiratory failure in a general population

Early prediction of treatment failure in severe community-acquired pneumonia: The PRoFeSs score

The impact of coagulation parameters on the outcomes of patients with severe community-acquired pneumonia requiring intensive care unit admission

PubMed Central PMCID: PMC3154260 clinical data collection, and Diagnostic Products Corporation for the cytokine assays

Prognostic value of plasma Ddimer level in adults with community-acquired pneumonia: a prospective study. Revista portuguesa de pneumologia

Prevalence and significance of coagulation abnormalities in community-acquired pneumonia

Risk factors for pulmonary embolism in patients preliminarily diagnosed with community-acquired pneumonia: a prospective cohort study

Procoagulant activity in children with community acquired pneumonia, pleural effusion and empyema

Lung vessels thrombosis in hospitalized patients with community-acquired pneumonia: role of endothelial function, hemostasis, fibrinolysis and inflammation on different phases of treatment

Evaluation of variation in coagulation among children with Mycoplasma pneumoniae pneumonia: a case-control study

Diagnostic accuracy of biomarker D-dimer in patients after stroke suspected from venous thromboembolism: A diagnostic meta-analysis. Clinical biochemistry

Clinical significance of plasma D-dimer and fibrinogen in digestive cancer: A systematic review and meta-analysis

Prognostic role of D-dimer level upon admission in patients with traumatic brain injury

Improved rule-out diagnostic gain with a combined aortic dissection detection risk score and D-dimer Bayesian decision support scheme

D-dimer antigen: current concepts and future prospects

Biomarkers for predicting short-term mortality in community-acquired pneumonia: A systematic review and meta-analysis