Systematic review and meta-analysis of dysregulated miRNAs in patients with severe pneumonia

Literature retrieval

Literature was sourced from PubMed (https://pubmed.ncbi.nlm.nih.gov/), EMBASE (https://www.embase.com/), the Cochrane Library (https://www.cochranelibrary.com/), and Web of Science databases (https://www.webofscience.com), updated through October 2024. The keywords utilized for study selection included miRNA or microRNA, alongside diagnosis, predictive, diagnostic, biomarker, or indicator, in conjunction with severe pneumonia, pneumonia, lung inflammation, lung infection, or acute respiratory distress syndrome. Manually retrieved the list of relevant references to select potential articles.

Literature selection

The inclusion criteria for this study were as follows: (1) Studies that specifically investigated the role of miRNAs in the clinical diagnosis of pneumonia; (2) studies were case-control design featuring a case group comprising patients diagnosed with severe pneumonia alongside a control group of healthy individuals or patients with non-pneumonia respiratory conditions; (3) studies that established and utilized clear diagnostic criteria for severe pneumonia; (4) studies providing precise and quantitative measurements of miRNA levels, including the methodology used for detection; and (5) studies containing comprehensive data necessary for calculating true positives (TP), false positives (FP), false negatives (FN), and true negatives (TN).

Conversely, the exclusion criteria were as follows: (1) duplicate studies; (2) review articles, meta-analyses, case reports, conference proceedings, and lecture summaries; (3) articles that do not pertain to miRNAs or the diagnosis of pneumonia; and (4) articles lacking sufficient data to calculate the necessary diagnostic parameters (TP, FP, FN, TN).

Data extraction and quality assessment

Two independent researchers, Jinmei Xu and Qiaoke Li, conducted a thorough extraction of data following a comprehensive review of the full texts. Disagreements are resolved through discussion with other authors when they arise. The principal extracted variables encompassed the first author, country of publication, year of publication, sample size, mean age, and gender distribution of both patients and controls, as well as details regarding the miRNA sample source and detection methodology. According to the sensitivity, specificity, patient number, and control number in the articles, the corresponding values of TN (=specificity×control number), TP (=sensitivity×patient number), FP (=control number – TN), and FN (=patient number – TP) were calculated.

All studies underwent meticulous review by two independent evaluators, Jinmei Xu and Qiaoke Li. Any disagreements were addressed through discussions involving the other authors. In cases of uncertainty regarding the inclusion of documents, a consensus was reached through collaborative dialogue. The quality assessment was conducted utilizing the Quality Assessment for Diagnostic Accuracy Studies version 2 (QUADAS-2) evaluation system [15]. The diagram was generated using Review Manager 5.0 [16]. Key components such as patient selection, index test, reference standard, and flow and timing were incorporated within the QUADAS-2 framework.

Statistical analysis

The analysis of the extracted data was conducted using STATA 16.0. The extracted values for TP, FP, TN, and FN data were included into forest plots to evaluate sensitivity and specificity. A summary receiver operating characteristic curve (SROC) was generated to derive the pooled area under the curve (AUC), along with the corresponding sensitivity and specificity metrics. Subsequently, we calculated the diagnostic likelihood ratios (DLR) for both positive and negative results, as well as the diagnostic score and diagnostic odds ratio (DOR), to facilitate a comprehensive assessment of clinical applicability.

Inter-study heterogeneity was evaluated using Cochran’s Q and the Higgins I² test [17]. A p-value from Cochran’s Q of less than 0.1 indicates the presence of heterogeneity, while an I² value exceeding 50 % suggests a significant degree of heterogeneity within the study. In instances where heterogeneity was detected, a random effects model was employed for data analysis, whereas a fixed effects model was applied when heterogeneity was absent.

Basic characteristics of studies

An extensive online search across various databases, including PubMed, EMBASE, Cochrane Library, and Web of Science, yielded a total of 1,580 publications. After the removal of duplicates, 146 articles remained. Subsequently, a review of titles led to the retention of 322 documents, with 216 studies excluded based on abstract evaluations, resulting in a final selection of 106 articles. Ultimately, 12 articles [13], 14], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], encompassing 17 tests, were chosen for the final analysis after a thorough full-text evaluation eliminated 94 additional studies. The systematic literature search and study selection process are illustrated in Figure 1.

Figure 1:

Flow chart for document screening.

A total of 12 articles were incorporated into the meta-analysis, encompassing 620 controls and 789 patients. Of these studies, 11 were conducted in China, while one took place in Egypt. Notably, five articles comprised two distinct tests. The specimens utilized primarily included serum and extracellular vesicles (EVs) derived from bronchoalveolar lavage fluid (BALF). Across all studies, a total of 16 miRNAs were identified. The specific details captured within the literature are succinctly summarized and presented in Table 1.

Quality assessment and sensitivity analysis

The QUADAS-2 tool evaluated the quality of the literature across four distinct domains (including patient selection, index test, reference standard, and flow and timing) to discern the risk of bias and relevance to clinical applicability. Figure 2 illustrates that within the context of bias risk, no instances of high risk were identified across all four domains. However, regarding applicability concerns, a significant risk was observed in patient selection and reference standards. Notably, only one study was categorized as having high risk in both patient selection and reference standards.

Figure 2:

Quality assessment risk map and summary chart.

Diagnostic accuracy

The information pertaining to TP, FP, FN, TN, sensitivity, specificity, and AUC derived from the literature is presented in Table 2. The mixed-model correlation was measured at −0.68, with a heterogeneity proportion attributed to threshold effects calculated at 0.46. This indicates the presence of additional factors contributing to variations beyond mere threshold effects.

Diagnostic results from 17 tests were meticulously compiled, summarizing the sensitivity and specificity of various diagnostic assessments. The I² for pooled sensitivity was measured at 75.36, while the I² value for specificity reached 76.96, indicating considerable heterogeneity (all p<0.01, Figure 3A). The findings from the SROC analysis, illustrated in Figure 3B, revealed a specificity of 0.88 (0.81–0.93) and a sensitivity of 0.79 (0.73–0.84), with an AUC of 0.89 (0.86–0.92). These results underscore the robust diagnostic accuracy of miRNAs in distinguishing severe pneumonia. Furthermore, the overall diagnostic likelihood ratio (DLR) for positive results was 6.82 (4.25–10.95), while the DLR for negative results was 0.24 (0.19–0.31) (Figure 4A). The combined diagnostic score was calculated to be 3.34 (2.82–3.86), and the DOR stood at 28.28 (16.80–47.58) (Figure 4B).

Figure 3:

Diagnostic value of miRNAs for severe pneumonia. A, sensitivity and specificity of individual studies and their aggregated data. B, SROC diagram illustrating the performance of miRNAs in diagnosing severe pneumonia.

Figure 4:

Diagnostic accuracy analysis. A, forest plot of DLR positive and DLR negative. B, forest plot of diagnostic score and odds ratio.

Evaluation of clinical application

The Fagan’s nomogram revealed a positive post-test probability of 87 % and a negative post-test probability of 19 %, given a prior-test probability of 50 %. This indicates an elevation in the likelihood of severe pneumonia from 50 to 87 %, while the probability of no severe pneumonia diminishes from 50 to 19 % (Figure 5A). The scatter matrix illustrates that the distribution is predominantly characterized by a positive likelihood ratio (PLR) of less than 10 and a negative likelihood ratio (NLR) exceeding 0.1 (Figure 5B). The absence of studies located in the lower left quadrant suggests that the sensitivities reported in these articles are robust, significantly enhancing the probability of accurately diagnosing severe pneumonia (Figure 5B).

Figure 5:

Evaluation of clinical application and publication bias analysis. A, fagan plots of post-detection of miRNA. B, the likelihood matrix of the population distribution of included studies. C, univariable meta-regression accompanied by subgroup analysis. D, publication bias analysis.

Subgroup analysis

To address the inherent heterogeneity among the studies included in the meta-analysis and to explore potential sources of variability in the diagnostic efficacy of miRNAs for severe pneumonia, we conducted subgroup analyses. The choice of subgroups was guided by several factors (such as specimen source, age, diagnostic criteria, pneumonia type, control type, and expression trend) that are known to influence the expression and diagnostic potential of miRNAs. These factors were selected based on their biological relevance, clinical significance, and the availability of data across the studies.

Subgroup analyses were conducted based on specimen source, age, diagnostic criteria, pneumonia type, control type, and expression trend to investigate the origins of heterogeneity. The univariable meta-regression analysis revealed significant differences in sensitivity and specificity based on the specimen subgroup (BALF vs. serum); however, no significant variations were observed among age subgroups (adults vs. children), diagnostic criteria (shown vs. not shown), control types (low stage vs. healthy), or expression trends (up vs. down) (Figure 5C, Table 3). This suggests that the high heterogeneity may primarily be attributed to the specimen type (P_h<0.01).

Publication bias and sensitivity analysis

Deeks’ funnel plot asymmetry test revealed no significant evidence of publication bias in the original analyses (Figure 5D, p=0.59).

Sensitivity analysis revealed that the normal quantile (Figure 6A) and Chi-square quantile (Figure 6B) analyses indicated that the distributions of the studies conform to the assumptions of normality and homoscedasticity. Cook’s distance chart (Figure 6C) demonstrated that the studies conducted by Sun, Y. (for miR-193a-5p) and Haroun (for miR-155) exhibited Cook’s distances exceeding 4, suggesting that these two studies may exert a considerable influence on the robustness of the current findings. Outlier detection plot (Figure 6D) illustrated that the majority of scatter points were nearly horizontal, signifying a linear correlation between the independent and dependent variables. Collectively, these results suggest a commendable level of robustness within the present analysis.

Figure 6:

Sensitivity analysis. A, normal quantile plot. B, chi-square quantile figure. C, Cook’s distance chart. D, outlier detection plot.