Length-of-Stay in the Emergency Department and In-Hospital Mortality: A Systematic Review and Meta-Analysis

The effect of emergency department (ED) length of stay (EDLOS) on in-hospital mortality (IHM) remains unclear. The aim of this systematic review and meta-analysis was to determine the association between EDLOS and IHM. We searched the PubMed, Medline, Embase, Web of Science, Cochrane Controlled Register of Trials, CINAHL, PsycInfo, and Scopus databases from their inception until 14–15 January 2022. We included studies reporting the association between EDLOS and IHM. A total of 11,337 references were identified, and 52 studies (total of 1,718,518 ED patients) were included in the systematic review and 33 in the meta-analysis. A statistically significant association between EDLOS and IHM was observed for EDLOS over 24 h in patients admitted to an intensive care unit (ICU) (OR = 1.396, 95% confidence interval [CI]: 1.147 to 1.701; p < 0.001, I2 = 0%) and for low EDLOS in non-ICU-admitted patients (OR = 0.583, 95% CI: 0.453 to 0.745; p < 0.001, I2 = 0%). No associations were detected for the other cut-offs. Our findings suggest that there is an association between IHM low EDLOS and EDLOS exceeding 24 h and IHM. Long stays in the ED should not be allowed and special attention should be given to patients admitted after a short stay in the ED.


Introduction
Prolonged length of stay (LOS) in the emergency department (ED), characterized by an inappropriately long period before final departure for an in-hospital bed, home, or another facility, is believed to adversely affect clinical outcomes. The time spent in the ED can be divided into distinct periods that are marked by time of arrival (triage registration), time of the start of care, time of the disposition decision (discharge or admission), time at the end of care, and time at ED departure ( Figure 1). EDLOS is defined as the time elapsed between the initial triage registration and physical departure from the ED. Boarding time (BT), defined as the time spent waiting for inpatient bed availability after the decision to admit the patient is made, is a significant contributor to the LOS. BT may also affect outcomes, as boarded patients require ongoing, often intensive care that several EDs are not well equipped to deliver [1][2][3]. The definition of prolonged EDLOS may vary. Prolonged ED visits have been defined as >4 h in the United Kingdom, >6 h in Canada and the U.S., and >8 h in Australia [4][5][6].
Prior studies have shown that ED boarding delays care, including the commencement of home medication, and increases the risk of adverse events, prolongs in-hospital LOS, and is associated with staff and patient dissatisfaction [7][8][9][10]. Prolonged ED BT also consumes already scarce ED resources, making them unavailable for the care of new patients and potentially affecting the outcomes of non-boarded patients [1,11].
Despite increased recognition of the adverse effects of prolonged EDLOS, its effect on patient mortality remains unclear. Several studies have found that ED crowding and increased BT are associated with higher mortality rates [11][12][13][14][15][16].
Crowding can increase both EDLOS and BT, since the rate of patient intake exceeds the capacity of the triage process. Throughput is also overwhelmed, because the number of patients requiring managing is high, and a lack of hospital beds throttles patient outflow [17]. Although there is a significant relationship between crowding, boarding time, and EDLOS, the relationship with in-hospital mortality (IHM) remains unclear.
Given the lack of evidence, additional research is needed to explore the association between EDLOS and IHM. This is important, considering recent evidence demonstrating the limited implementation and thus limited impact of hospital strategies to improve patient flow through the ED [1,2,17,18]. Prior studies have shown that ED boarding delays care, including the commencement of home medication, and increases the risk of adverse events, prolongs in-hospital LOS, and is associated with staff and patient dissatisfaction [7][8][9][10]. Prolonged ED BT also consumes already scarce ED resources, making them unavailable for the care of new patients and potentially affecting the outcomes of non-boarded patients [1,11].
Despite increased recognition of the adverse effects of prolonged EDLOS, its effect on patient mortality remains unclear. Several studies have found that ED crowding and increased BT are associated with higher mortality rates [11][12][13][14][15][16].
Crowding can increase both EDLOS and BT, since the rate of patient intake exceeds the capacity of the triage process. Throughput is also overwhelmed, because the number of patients requiring managing is high, and a lack of hospital beds throttles patient outflow [17]. Although there is a significant relationship between crowding, boarding time, and EDLOS, the relationship with in-hospital mortality (IHM) remains unclear.
Given the lack of evidence, additional research is needed to explore the association between EDLOS and IHM. This is important, considering recent evidence demonstrating the limited implementation and thus limited impact of hospital strategies to improve patient flow through the ED [1,2,17,18].
To address this knowledge gap, we performed a systematic review and meta-analysis (MA) which examined the association between EDLOS and IHM. We hypothesized that a longer EDLOS would predict greater IHM risk.

Materials and Methods
This systematic review and MA focused on studies analyzing the relationship between total EDLOS and IHM. Studies analyzing only the BT, which represents a time segment within the EDLOS (see Figure 1), and overcrowding studies that did not refer to the EDLOS were excluded.
The review follows the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines recommended by the Cochrane Handbook for Systematic Reviews of Interventions [19]. A PRISMA checklist is presented in Supplemental Table S1. The protocol for this review was registered in PROSPERO, CRD42016050422 (http://www.crd.york.ac.uk/PROSPERO, accessed on 29 November 2022). To address this knowledge gap, we performed a systematic review and meta-analysis (MA) which examined the association between EDLOS and IHM. We hypothesized that a longer EDLOS would predict greater IHM risk.

Materials and Methods
This systematic review and MA focused on studies analyzing the relationship between total EDLOS and IHM. Studies analyzing only the BT, which represents a time segment within the EDLOS (see Figure 1), and overcrowding studies that did not refer to the EDLOS were excluded.
The review follows the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines recommended by the Cochrane Handbook for Systematic Reviews of Interventions [19]. A PRISMA checklist is presented in Supplemental  Table S1. The protocol for this review was registered in PROSPERO, CRD42016050422 (http://www.crd.york.ac.uk/PROSPERO, accessed on 29 November 2022).

Data Sources and Searches
We defined EDLOS as the time elapsed between the initial triage patient registration and physical departure from the ED (Figure 1). Our primary endpoint was all-cause mortality.
A systematic search of the PubMed, Embase, Web of Science, Cochrane Controlled Register of Trials, CINAHL, PsycInfo, and Scopus databases was prepared by two medical librarians specializing in systematic reviews (L.Ö., J.C.), in close collaboration with D.L. and A.B. (emergency medicine expert physicians). All terms were searched in the fields for "Abstract" and "Article Title" (alternatively in the field for "Topic") and MeSH/Subject Headings/Thesaurus when available. The databases were first searched from their inception to January 2020 (L.Ö.). A search update was conducted in the same databases during manuscript preparation on 14-15 January 2022 (L.Ö.), to ensure the inclusion of recently published papers. No filters or limitations were applied to retrieve the best possible results. We screened all published studies related to ED boarding and crowding to identify those reporting data on EDLOS and IHM. Studies reporting EDLOS cut-off times were included in the MA. Studies not mentioning EDLOS or IHM were excluded. We also screened the reference lists of the selected studies manually. The reviewers also manually searched the gray literature (including congress and meeting abstracts) but excluded these sources when they were not subsequently followed by full-text articles published in scientific journals. Reproducible search strings, results, and technical notes for each database are presented in Supplemental Table S2.

Inclusion Criteria and Study Selection
All patients over 18 years old who visited an ED were included. Exposure was defined as the time spent in the ED from the arrival to the admission to inpatient bed. This time exposure was defined as a EDLOS cut-off chosen in selected studies. The outcome was IHM whatever the cause and the delay of death in the in-hospital bed was. We considered all studies based on a prospective or retrospective design, namely cohort studies, case-control studies, as well as randomized controlled trials.
Records identified in the literature search were uploaded to the Covidence (Veritas Health Innovation, 2021, https://www.covidence.org, accessed on 29 November 2022) systematic review software for blinded screening and automatic removal of duplicates. We extracted articles focused on the association between EDLOS and IHM in an adult ED setting. Studies analyzing the effects of boarding and ED crowding on mortality were also included when EDLOS was reported in their statistical analysis. Publications in English and other languages using translators when necessary were included.
Two emergency medicine specialists (D.L., A.B.) independently screened the titles and abstracts yielded by the literature searches. Any selection disagreements identified by Covidence were resolved by discussion to reach consensus or were adjudicated by a third independent reviewer (Z.B.). Full reports were obtained for all titles or abstracts that met the inclusion criteria. Both reviewers independently read all full-text articles, obtaining additional information from the study authors as needed to resolve questions about eligibility. An overview of the screening and selection process is presented in the PRISMA flow diagram ( Figure 2). Study data were extracted into a customized Microsoft Excel ® table, including the following study characteristics: design, setting, population, sample size, main objective, prognostic factors, and outcomes such as boarding, definition and values of EDLOS, crowding, type of mortality, results including precision and significance, and adjustment for confounding factors (e.g., age, comorbidities, diagnosis, triage severity code).

Data Extraction and Quality Assessment
The quality of each study was rated and recorded in a data collection form. Quality assessments were performed independently by two reviewers (A.B., Z.B.) using the Newcastle-Ottawa Quality Assessment Scale (NOS), a scale designed for non-randomized trials [20], and disagreements were resolved by discussion to reach consensus.
The NOS consists of four items on "study selection", one item on "comparability", and three items on "study outcome" [20]. Using this scale, reviewers can award one star for each of the four items on "selection", one star for each of the three items on "outcome", and a maximum of two stars for "comparability". Ratings were calculated independently by each reviewer, and the results were averaged. Studies of the highest quality were awarded nine stars.
The risk of bias was summarized for each study and incorporated into the overall findings and data synthesis.

Data Extraction and Quality Assessment
The quality of each study was rated and recorded in a data collection form. Quality assessments were performed independently by two reviewers (A.B., Z.B.) using the Newcastle-Ottawa Quality Assessment Scale (NOS), a scale designed for non-randomized trials [20], and disagreements were resolved by discussion to reach consensus.
The NOS consists of four items on "study selection", one item on "comparability", and three items on "study outcome" [20]. Using this scale, reviewers can award one star for each of the four items on "selection", one star for each of the three items on "outcome", and a maximum of two stars for "comparability". Ratings were calculated independently by each reviewer, and the results were averaged. Studies of the highest quality were awarded nine stars.
The risk of bias was summarized for each study and incorporated into the overall findings and data synthesis.
Odds ratios (ORs) were used to measure the potential association between EDLOS and IHM. For binary outcome variables, the measured effect was expressed as the logtransformed estimated OR. The weight of each study in the analysis was expressed as the inverse of the variance of the log-transformed estimated OR. The amount of between-study heterogeneity against the total variance was measured by I 2 and presented as 0-100%. Sensitivity analysis was performed by the leave-one-out method, in which one study at a time was removed iteratively to confirm that our findings were not dictated by any specific study. With this method, if the results are consistent, there is confidence that the overall MA results are robust.
To illustrate the foundations, we used forest plots to summarize and visualize the effect size of each study, including 95% confidence intervals (CIs), with respect to the study's weight. The location of the 95% CI for the OR in relation to 1, in the case of ORs, also indicated the significance of the effect size.
We used a DerSimonian-Laird random-effects model in our study. Because the weight of each study should be approximately the same, the weighted analysis for the randomeffects model was treated as an unweighted analysis.
To examine the influence of population characteristics on overall heterogeneity, we separated the studies into two subsets for each cut-off: intensive care unit (ICU) and non-ICU population subsets. Two additional meta-analyses were conducted for each subset.
Moreover, to improve the accuracy of our heterogeneity evaluation in the MA, we used the IVhet model in the Microsoft Excel ® MA package, designed particularly for use in meta-analyses with high heterogeneity (MetaXL, available at www.epigear.com, accessed on 29 November 2022) [21,22]. This method uses the quasi-likelihood estimator as an alternative to random-effects models with the problem of underestimation of the statistical error and overconfident estimates. The estimator retains a correct coverage probability and a lower observed variance than the random-effect model estimator, regardless of heterogeneity [23,24].
The symmetry of a funnel plot and Egger and Begg tests were used to qualitatively determine the presence of publication bias (MedCalc Software, version 19.6.1) [21,22].
To analyze the factors underlying heterogeneity, we performed a univariate metaregression analysis using the following factors: age, sex, country of study, ED population, and disease severity.
Detailed information is available in the Supplemental Text.

Quality of the Selected Studies
The methodological quality of the studies is presented in Supplemental Table S3. The evaluation was performed by two independent evaluators. The average quality score was 6.53 ± 1.23 (min.: 3; max.: 8), which can be considered intermediate.

Meta-analysis 3.2.1. Random-Effects Models
The DerSimonian-Laird random-effects model [73] showed no statistically significant association between EDLOS and IHM, regardless of the cut-off value used:  The ED populations included in these studies were divided into two categories: the patients admitted to the ICU (ICU-admitted population, representing the most critically The ED populations included in these studies were divided into two categories: the patients admitted to the ICU (ICU-admitted population, representing the most critically ill patients) and those not admitted to the ICU (non-ICU-admitted population; those admitted to lesser-acuity in-patient wards). Our meta-analysis identified an association between EDLOS and IHM for the 24 h cut-off only in ED ICU-admitted patients, with a significant OR of 1.396 (95% CI 1.147 to 1.701, I 2 = 0%; p < 0.001). Another association was found for a low EDLOS cut-off in the non-ICU-admitted ED patients' subgroup, with a significant OR of 0.581 (95% CI 0.453 to 0.745, I 2 = 0%; p < 0.001) (Supplemental Figures S1 and S2).
No significant association was found between EDLOS and IHM for any of the cutoff values when all studies, including both ICU and non-ICU populations, were tested together. After dividing the patients into the two population types to create a certain level of homogeneity in each subgroup, the effect of prolonged EDLOS on IHM could be identified. For all cut-off values, the overall effect size was close to 1, and was not statistically significant, but in the ICU subgroup, the effect size was above 1 (significant for 24 h cut-off), and in the non-ICU subgroup, the effect size was less than 1 (significant for a low EDLOS cut-off).

Cross-validation (Leave-one-out)
The results of the cross-validation performed by the leave-one-out method are given in Supplemental Tables S8-S10. This procedure was used in cases where insufficient data were available for partitioning between the training and test datasets. The sensitivity analysis confirmed the high heterogeneity among studies, which was not decreased by the exclusion of any single study (Supplemental Table S8).
We performed a sensitivity analysis in both ICU and non-ICU populations. The observed heterogeneity remained high in both subpopulations (Supplemental Tables S9 and S10). The exclusion of studies one by one, as suggested by Choi et al. [59], Intas et al. [45], Servia et al. [34], Soni et al. [42], Tilluckhdarry et al. [35], and Verma et al. [71], significantly reduced the heterogeneity in the ICU population for the 24 h cut-off value (Supplemental Table S9). Sensitivity analysis for the non-ICU population was possible only for a cut-off of 4 h, with the exclusion of the Paton et al. study (Supplemental Table S10) [44].
To summarize, for most cut-off values except for EDLOS <3 h and EDLOS >24 h, in the studies overall and in the ICU and non-ICU subgroups separately, no single study had a significant effect on the test results.

Inverse Variance Heterogeneity Model
Because of the high level of heterogeneity between studies, we decided to conduct a meta-analysis using the inverse variance heterogeneity (IVhet) model [23,24]. We did not find a significant difference in IHM between patients staying in the ED for any of the investigated cut-off periods (Supplemental Table S11). The use of the IVhet model allows reducing the underestimation of the statistical error and overconfident estimates. In all cases, even if the 95% CI for the effect size of the random-effects model revealed a significant result, the IVhet model provided a broader 95% CI for the same effect size, so that eventually, none of the effect sizes were statistically significant. This finding supported the main conclusion that there is no significant association between EDLOS and IHM.

Subgroup Meta-Analyses and Univariate Meta-Regression Analysis
We performed different meta-analyses to isolate subpopulations to explain the observed high heterogeneity. First, we excluded step-by-step studies because we observed that this exclusion decreased heterogeneity. The random-effects model confirmed the absence of an association (Supplemental Table S12). Next, we performed meta-analyses of studies that included the general ED population (Supplemental Table S12), specific disease populations, and patients with different severities of illness (ICU and non-ICU populations) (Supplemental Table S12). We found that the disease population and severity of illness were involved in the heterogeneity (Supplemental Table S12).
To explain the source of heterogeneity, a univariate meta-regression analysis was performed for each cut-off value separately. As expected, some of the factors had a significant effect on heterogeneity. For example, in meta-regression analysis for the 6 h cut-off, all the factors were significant at a 5% significance level.

Discussion
EDLOS and BT are used by hospital administrators as measures of the quality of care delivered in the ED. A prolonged EDLOS is a source of dissatisfaction for patients and family; however, this indicator in isolation is not sufficient to comprehensively evaluate the quality of care. Combining ED time and the occurrence of negative outcomes, such as adverse events and IHM, is comparatively more relevant, and could help to improve quality of care. We previously found that there was a trend that BT increases IHM [74]. This new systematic review and MA did not find a significant relationship between EDLOS and IHM for any of the studied cut-off time points. However, our research did uncover a new and relevant result for EDLOS >24 h in ED ICU-admitted patients and EDLOS <3 h in non-ICU-admitted ED patients. For these cut-offs and types of ED populations, we did not find heterogeneity (I 2 = 0). The absence of a statistically significant difference in IHM for the other cut-offs is likely multifactorial, including the heterogeneity among the studies and various other factors, including population characteristics (e.g., age, sex, triage severity score, type of disease, mode of arrival at the ED, ED daytime, time shift, etc.), variation in hospital organization, adherence to clinical guidelines, type of admission source, and other factors. We used IVhet, designed particularly for use in meta-analyses with high heterogeneity, to provide better validation for the same estimated effect size [23,24]. Regular random-effects models, such as inverse variance or DerSimonian-Laird [73,[75][76][77], emphasize the need for larger studies and indicate an underestimation of the statistical error. However, the IVhet model provides the correct coverage of the estimated effect size. The CI of the effect size obtained with this model was wider than that in other randomeffects models. All 95% CIs using the IVhet model included 1; thus, we can conclude that there was no significant association between EDLOS and IHM for cut-off values of 4-8 h, which represent the target times in some countries [4][5][6]. Cross-validation analysis did not reduce the heterogeneity (Supplemental Table S11). However, meta-regression analysis showed that factors, such as type of population, type of disease, and severity of illness, could explain the heterogeneity for EDLOS <3 h, 4 h, 5 h, and 6 h cut-off values (Supplemental Table S12). Most categorical variables (e.g., population type, severity score, and country) were found to be significant in at least some of the meta-regression models at different cut-offs. However, there was no consistent impact of one variable on all cut-offs.
In exploring this lack of association between EDLOS and IHM for some cut-offs, we recognize that processing time and patient care time are complex variables, combining many different factors that influence the EDLOS, quality of care, and patient safety in the ED [78,79]. Given the frenzied nature of the ED environment, crowding may prevent providers from giving critically ill patients the close and constant attention they need [80][81][82][83][84][85][86][87][88][89][90][91][92]. This could be expected to lead to worse outcomes for patients, including increased IHM, but the evidence that we found in this systematic review was mixed. While some studies suggested that EDLOS is an independent predictor of ICU mortality [3,25,[34][35][36]39,40,42,47,48,[51][52][53][54]58,67,71], others reported no adverse association [32,38,44,45,50,55,56,[59][60][61]63,69,70]. MA of the studies reporting IHM in patients admitted to the ICU showed an association with EDLOS over 24 h, with absence of heterogeneity. In most EDs, it is only acceptable to keep critical patients in the ED when there are no ICU beds. Many EDs are not designed to manage those patients optimally, due to a lack of trained emergency specialists in some countries, a lack of nurse resources, or the absence of a specific intensive care area where critical patients can be safely observed by a specific team. In ED patients who were admitted to non-ICU wards, some studies showed an association between EDLOS and IHM [26,29,33,37,43,62,64,66,68], while other studies did not [9,27,28,30,31,41,46,49,57,65,72]. Surprisingly, with cut-off values analyzed through different random-effects meta-analyses, we found a significant association between EDLOS < 3 h and IHM. Our data do not provide an explanation of this finding, and prospective studies analyzing all factors that contribute to the EDLOS are needed. It is typically rare to admit ED patients within 3 h, which is often below the threshold for obtaining all laboratory and imaging results, and for some patients' specialist consultations. Sicker patients and those with clear-cut diagnoses who receive certain specific treatments may account for early departures, which could explain this result.
Our recommendations for policy makers are that long stays in the ED must be discouraged, unless there is a specific track for these patients including a specific ED area with a dedicated team. Another lesson from our study is that ED teams must be careful when they admit patients after a short stay in the ED and should be sure that there is continued close monitoring to avoid the risk of clinical deterioration. This may be particularly true in older patients where clinical presentations are often atypical. Some critically ill patients will be treated in the ED [58]. In such cases, the outcome and EDLOS will be dependent on the rapidity to stabilize the patients and the decision to admit them to hospital [57].
Another recommendation is to fast-track the care of specific events, such as myocardial infarction or stroke, that will be directly addressed to the angiography laboratory and acute neurovascular unit, resulting in a markedly reduced EDLOS for these patients. The association between EDLOS at different cut-off time points and IHM at different hospital time points (24 h, 48 h, 72 h, 7 days, 28 days), including the occurrence of adverse events after the ED care, could be worth investigating.
In contrast to high priority/sicker patients, mortality in patients with lower triage could be positively associated with EDLOS. One explanation could be the effect of undertriaging, where patients with medical urgency remain undetected by the ED triaging system. Patients with non-specific symptoms and low clinical urgency often have increased hospitalization, increased EDLOS, increased mortality, and more often are frail and of advanced age [9,93].

Study Strengths
The strengths of this MA include an extensive comprehensive search strategy, strong eligibility criteria that enhance generalizability, rigorous use of the NOS approach for rating the quality of evidence, a robust step-by-step MA, and a large number of included studies. This is the first MA exploring the association between EDLOS and IHM.

Study Limitations
Our study had some limitations and potential presence of publication bias. Some studies included a univariate analysis, while others used multivariate analysis, making it difficult to compare the effect sizes. Therefore, we chose to use an overall univariate analysis using the crude data reported by the studies. We used various meta-analyses and used the IVhet method to confirm the absence of an association for the classical cut-offs observed in the EDs. Moreover, a dose-response meta-analysis model was inapplicable for the relationship of EDLOS and IHM, given the lack of sufficient EDLOS-specific comparisons within each included study (i.e., only two-time ranges in each study) [94]; a dose-response model would be useful to determine the golden time range of EDLOS for patients needing emergency healthcare, and meanwhile explain the heterogeneity of the results. In addition, our study explored the association between EDLOS and IHM, but did not address the causes of prolonged EDLOS. With 33 studies from 50 countries worldwide included in the systematic review, we believe the results are generalizable to larger, urban, academic EDs, which represent the vast majority of EDs contributing to this MA. Representation of smaller, rural, non-academic EDs is limited, and therefore, generalizability to these EDs is unclear. More studies are needed to evaluate the correlation between EDLOS and IHM in different countries and hospital types, with variable equipment and human resources to confirm the results for EDLOS <3 h and EDLOS >24 h. In addition, a better understanding of the role played by potential confounding factors can help to reduce heterogeneity for the other cut-offs.

Conclusions
This MA was designed to analyze the association between EDLOS and IHM; we did not find evidence supporting this hypothesis when all ED patients were included for each cut-off. However, we did find a new and relevant result confirming an association with EDLOS and IHM for patients exceeding 24 h in ED ICU-admitted patients and for low EDLOS below 3 h in non-ICU-admitted ED patients. Other factors involved in the negative outcomes after ED care should be carefully explored to determine the role of EDLOS in the occurrence of IHM.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/jcm12010032/s1, Supplemental Text. Detailed information for each selected study; Figure S1. Meta-analysis of studies including the ICU-admitted ED population with the same cut-off using the random-effects model (DerSimonian-Laird), Iˆ2 = I 2 ; Figure S2. Meta-analysis of studies including the non-ICU-admitted ED population with the same cut-off using the random-effects model (DerSimonian-Laird), Iˆ2 = I 2 ; Figure Table S3. Quality assessment of the studies. The Newcastle-Ottawa Quality Assessment Scale consists of 4 items on study selection, 1 item on comparability, and 3 items on study outcomes (see Reference 20 in main manuscript). According to this scale, studies can be awarded one star for each of the 4 items on selection and for each of the 3 items on outcomes, and a maximum of 2 stars for comparability. The highest quality studies are awarded up to nine stars. The evaluation was performed by two independent evaluators (E1 and E2). The average is 6.53 ± 1.23 (min.: 3; max.: 8); Table S4. Summary of the random-effects model results (DerSimonian-Laird); Table S5. Publication bias tests for all cut-offs; Table S6. Publication bias tests for all cut-offs in the ICU population; Table S7. Publication bias tests for all cut-offs in the non-ICU population; Table S8. Sensitivity analysis (leave-one-out) results of the overall data. Random-effects model (DerSimonian-Laird); Table S9. Sensitivity analysis (leave-one-out) results of the ICU population. Random-effects model (DerSimonian-Laird); Table  S10. Sensitivity analysis (leave-one-out) results of the non-ICU population. Random-effects model (DerSimonian-Laird); Table S11. Heterogeneity analysis: comparison between inverse variance (IV) and inverse variance heterogeneity (IVhet) methods. Estimated effect size and CI (95%); Table S12: Meta-regression analysis. Discrete (categorical) factors are population type, country, and number of patients included in the studies. The continuous factors are age and sex. Discrete (categorical) factors = population type, severity (ICU/not ICU), country, disease. Continuous factors = age and % male.
Author Contributions: All authors made a significant contribution to the scientific quality of this paper and fulfilled the ICMJE criteria. All authors have approved the final version of this manuscript and have agreed to be held accountable for all aspects of the submitted work, including ensuring that any questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. D.L., A.K., Z.B. and A.B. are considered co-first authors as they wrote the protocol, participated in the screening and selection of the studies, and drafted the manuscript. Institutional Review Board Statement: We utilized de-identified data pertaining to EDLOS cut-offs and mortality that did not require ethical committee approval. This is a systematic review and meta-analysis of peer reviewed studies; each study had its own ethics approval.

Informed Consent Statement: Not applicable.
Data Availability Statement: Template data collection forms, data extracted from included studies, data used for all analyses, analytic code, and any other materials used in the review are available from the corresponding author upon request. We also utilized de-identified data pertaining to EDLOS cut-offs and mortality that did not require ethical committee approval.

Acknowledgments:
We thank Detajin Junhasavasdikul, Jose Javier Trujillano Cabello, Georges Intas, and Kapil Dev Soni for sharing their databases with approval by their ethical committees. We would also like to thank Mariam Al Ahbabi at the UAEU National Medical Library for her help with ordering and uploading full-text papers to the screening module in Covidence.

Conflicts of Interest:
The authors declare no conflict of interest.