Diagnostic and prognostic value of fibroblast growth factor 23 in acute kidney injury: systematic review and meta-analysis Authors: Xiaohang Pei, Hang Gao, Lina Zhang Published: 2026 ABSTRACT AbstractBackground: Acute kidney injury (AKI) is associated with high mortality and adverse outcomes. Fibroblast growth factor 23 (FGF23) has emerged as a potential biomarker for AKI; however, its diagnostic and prognostic utility remains inconsistent.Methods: We conducted a systematic review and meta-analysis of studies evaluating circulating intact FGF23 (iFGF23) or C-terminal FGF23 (cFGF23) (PROSPERO: CRD42022302659). PubMed, EMBASE, CNKI, and Wanfang databases were searched through June 9, 2026. QUADAS-2 was used for quality assessment. A random-effects bivariate model pooled sensitivity, specificity, positive/negative likelihood ratio (PLR/NLR), diagnostic odds ratio (DOR), and area under the summary receiver operating characteristic curve (SROC AUC).Results: Twenty-three studies were included: 17 diagnostic, 6 prognostic (one addressing both). For AKI diagnosis, the pooled sensitivity was 0.79 (95% CI 0.73–0.86), specificity 0.82 (95% CI 0.75–0.89), PLR 4.40 (95% CI 2.59–6.21), NLR 0.25 (95% CI 0.16–0.34), DOR 17.49 (95% CI 8.67–35.16), and SROC AUC 0.87 (95% CI 0.81–0.92). Substantial heterogeneity was observed (I2 = 67%), with iFGF23 demonstrating higher accuracy than cFGF23 (AUC 0.91 vs 0.81). For AKI mortality, pooled sensitivity was 0.77 (95% CI 0.69–0.84), specificity 0.76 (95% CI 0.70–0.82), DOR 10.89 (95% CI 6.86–17.30), and SROC AUC 0.77 (95% CI 0.70–0.83). Significant heterogeneity was noted (I2 = 86.2% for sensitivity, 80.4% for specificity). No significant publication bias was detected.Conclusions: Circulating FGF23 exhibits moderate-to-high diagnostic and moderate prognostic performance in AKI, though interpretation is limited by substantial heterogeneity. It may serve as a complementary biomarker for risk stratification, pending further validation with standardized protocols. Introduction Acute kidney injury (AKI) is a clinical syndrome characterized by a rapid decline in renal function, which is associated with high mortality and poor prognosis. The incidence of AKI among hospitalized patients has been reported to exceed 10% [1]. In the intensive care units (ICUs), the incidence of AKI can reach as high as 50%, with a mortality rate exceeding 60% [2,3]. Moreover, AKI increases the risk of chronic kidney disease (CKD), end-stage renal disease (ESRD), and other long-term adverse outcomes [4]. Early identification of high-risk patients, timely diagnosis, and the implementation of effective interventions are critical to improving outcomes. However, current AKI diagnosis relies predominantly on changes in serum creatinine and urine output, which are delayed indicators and may fail to detect early or subclinical tubular injury. Therefore, there is an urgent need to identify novel biomarkers that can detect AKI earlier and predict clinical outcomes more accurately. Fibroblast growth factor 23 (FGF23), a hormone predominantly secreted by osteocytes and osteoblasts, plays a critical role in regulating phosphate homeostasis and vitamin D metabolism [5,6]. In the setting of AKI, circulating FGF23 levels rise rapidly due to decreased renal clearance, upregulated bone synthesis in response to tubular injury, and inflammation-mediated induction [7,8]. This rapid elevation positions FGF23 as a potentially attractive early biomarker of AKI. Experimental and clinical studies have consistently demonstrated that elevated FGF23 correlates with AKI severity, impaired renal recovery, and increased mortality [9,10]. Despite this, the magnitude and consistency of its diagnostic and prognostic value across studies remain uncertain, largely owing to differences in study design, patient populations, clinical settings, and assay methodology [8,11]. Two commercially available assay types introduce particular complexity: intact FGF23 (iFGF23) detects only the full-length, biologically active hormone, whereas C-terminal FGF23 (cFGF23) recognizes both the intact molecule and biologically inactive C-terminal fragments [12,13]. In AKI, impaired renal clearance of FGF23 fragments and possible alterations in proteolytic cleavage may lead to some degree of discordance between iFGF23 and cFGF23 measurements, which could partly contribute to the inconsistent findings reported across studies. To date, no comprehensive meta-analysis has systematically compared the diagnostic and prognostic accuracy of FGF23 in AKI while specifically accounting for assay type (iFGF23 vs cFGF23), age group (adult vs pediatric), and clinical setting. Therefore, this study aimed to summarize the diagnostic and prognostic value of circulating FGF23 in AKI and explore sources of heterogeneity across different subgroups. Methods This systematic review and meta-analysis was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [14]. The review protocol was developed a priori and prospectively registered in PROSPERO (CRD42022302659), where it is publicly accessible. No important amendments were made to the prespecified eligibility criteria, outcomes, or statistical methods relative to the registered protocol. Minor changes were made to the search strategy and author list during the conduct of the review, but these did not affect the core design or analysis plan. Inclusion and exclusion criteria Studies were included if they met all of the following criteria: (1) enrolled patients with AKI or at risk of AKI, based on established criteria such as KDIGO (Kidney Disease: Improving Global Outcomes), RIFLE (risk, injury, failure, loss, end-stage renal disease), or AKIN (acute kidney injury network); (2) measured serum or plasma FGF23 levels (either iFGF23 or cFGF23); (3) reported sufficient data to evaluate diagnostic or prognostic performance (allowing construction of a 2 × 2 contingency table); and (4) were original clinical research articles. Studies were excluded if they were: case reports, reviews, editorials, conference abstracts without full-text data, or animal experiments; duplicate publications; or studies that lacked sufficient data for extraction despite attempts to contact the corresponding authors. Literature search strategy PubMed, EMBASE, China National Knowledge Infrastructure (CNKI), and Wanfang databases were searched from inception to June 9, 2026. The search strategy combined Medical Subject Headings (MeSH terms) and free-text keywords, including ‘fibroblast growth factor 23,’ ‘FGF-23,’ ‘FGF23,’ ‘acute kidney injury,’ ‘acute renal failure,’ ‘AKI,’ ‘ARF,’ and related terms. The full electronic search strategies for each database (PubMed, Embase, CNKI, and Wanfang), including all keywords, controlled vocabulary terms, filters and limits used, are provided in Supplementary Appendix: Full search strategies. In addition, reference lists of all included studies and relevant review articles were manually screened to identify any potentially eligible studies not captured by the database searches. Literature screening and data extraction Two investigators (X.P. and H.G.) independently screened all records in two stages. First, titles and abstracts of the retrieved records were screened against the predefined inclusion and exclusion criteria to identify potentially eligible studies. Second, the full text of all potentially eligible articles was obtained and independently assessed by the same two reviewers to confirm eligibility. Discrepancies at any stage were resolved through discussion with a third researcher (L.Z.). Reference management software was used only for deduplication and citation management; no automation tools or machine-learning algorithms were used to perform or assist with study selection. Two reviewers (X.P. and H.G.) independently extracted data from each included study using a standardized, piloted data extraction form. For each study, we collected information on first author, publication year, country, study design, population characteristics, AKI definition, FGF23 assay type (intact or C-terminal), sample size, and 2 × 2 diagnostic data (true positives, false positives, false negatives and true negatives) or prognostic outcomes, as appropriate. Any discrepancies in extracted data between the two reviewers were resolved by discussion and, when necessary, by consultation with a third investigator (L.Z.). For the diagnostic performance analysis, we included all studies that met the eligibility criteria and reported sufficient data for circulating FGF23 (iFGF23 or cFGF23, measured in serum or plasma) in distinguishing patients with and without AKI. For the prognostic accuracy analysis, we broadly screened studies evaluating adverse AKI outcomes, including all-cause mortality, persistent kidney dysfunction, renal recovery, or composite endpoints. However, for the quantitative meta-analysis, only studies with all-cause mortality as the endpoint were pooled. Studies assessing non-mortality endpoints were identified but excluded from the pooled prognostic synthesis and are discussed qualitatively in the Results. For each eligible study and assay type, we extracted or reconstructed 2 × 2 contingency tables based on the authors’ diagnostic thresholds; when multiple cutoffs were reported, we used the threshold identified as optimal by the original study (typically from ROC analysis). Quality assessment of included studies The methodological quality of both diagnostic and prognostic studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies–2 (QUADAS‑2) tool. For prognostic studies, we adopted a diagnostic accuracy framework (FGF23 as index test, adverse outcome as reference standard) to enable 2 × 2 table extraction and direct comparability with the diagnostic component. QUADAS-2 evaluates four domains (patient selection, index test, reference standard, and flow and timing) and concerns regarding applicability in the first three domains. Each domain was judged as having low, high or unclear risk of bias. Two investigators (X.P. and H.G.) performed the assessments independently, and disagreements were resolved by discussion with a third investigator (L.Z.). We did not perform a formal GRADE or other structured certainty‑of‑evidence assessment for the diagnostic or prognostic outcomes; instead, we interpreted the overall strength of evidence qualitatively in light of the observed risk of bias, inconsistency, imprecision and indirectness as described in the Results and Discussion. Statistical analysis Statistical analyses were conducted using Stata 17.0 and Review Manager (RevMan) 5.4 software. We calculated or reconstructed 2 × 2 contingency tables and derived sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR). Characteristics of the included studies were summarized in descriptive tables, and pooled estimates of diagnostic and prognostic accuracy were displayed in forest plots and summary receiver operating characteristic (SROC) curves, with 95% confidence and prediction regions where appropriate. Meta-analyses were primarily performed using the midas package (version 3.02) in Stata, with midas mle for maximum likelihood estimation of the bivariate random-effects model. Metandi with restricted maximum likelihood was used as the alternative when MLE failed to converge, which occurred in prognostic analyses and small diagnostic subgroups. When zero cells occurred in the 2 × 2 tables, Haldane-Anscombe correction (adding 0.5 to all cells) was applied to enable stable logit transformation and model convergence. Spearman correlation coefficients between the logit-transformed sensitivity and logit (1 − specificity) were calculated to assess the potential for threshold effects causing heterogeneity. Cochrane Q and I2 tests were performed to evaluate non-threshold heterogeneity, with p < 0.1 or I2 > 50% indicating significant heterogeneity. Prespecified subgroup analyses were conducted by: (i) FGF23 assay type (iFGF23 vs. cFGF23); (ii) study design (prospective vs. retrospective); (iii) sample size (<100 vs. ≥100 participants); (iv) age group (adults vs. children); and (v) country (China vs. non-China). Sensitivity analyses were performed to explore result stability. Publication bias was evaluated using Deeks’ funnel plot asymmetry test, with p < 0.10 indicating significant asymmetry. Fagan nomograms were constructed to illustrate the potential clinical utility of circulating FGF23 testing. Results Literature screening results A total of 482 articles were retrieved through database searches, including 98 from PubMed, 313 from EMBASE, 42 from CNKI, and 29 from Wanfang. After removing 127 duplicate records using reference management software, 355 records underwent title and abstract screening, of which 276 were excluded for not meeting the inclusion criteria. The remaining 79 articles were assessed by full-text review; 56 were subsequently excluded for the following reasons: not investigating the diagnostic or prognostic value of circulating FGF23 (n = 31), conference abstracts without full-text data (n = 12), insufficient data for extraction (n = 8), repeat studies (n = 4), and review article (n = 1). Ultimately, 23 studies [15–37] were included in the final meta-analysis (13 English-language [15,19,22,23,25–31,35,36] and 10 Chinese-language [16–18,20,21,24,32–34,37] articles). For the prognostic synthesis, one study [37] assessing renal recovery as the endpoint was not pooled in the quantitative analysis for mortality prognosis due to distinct pathophysiological processes, leaving 6 studies [31–36] for the mortality prognostic meta-analysis. The literature screening process is depicted in Figure 1. Flowchart of study identification, inclusion, and exclusion. Basic characteristics of included studies Twenty-three studies were included in the meta-analysis, 17 reported the diagnostic value of circulating FGF23 in AKI, and 7 analyzed its prognostic value, among them one study [31] addressing both aspects (thus contributing to both analyses). The included studies were conducted in China (n = 13), the USA (n = 3), Egypt (n = 3), Germany (n = 1), Thailand (n = 1), India (n = 1), and Poland (n = 1), reflecting a predominance of single center cohorts from Asia. All studies were single-center clinical studies with sample sizes ranging from 19 to 859. In terms of study design, there were 15 prospective cohort studies, 7 retrospective cohort studies, and 1 nested case-control study. Twenty studies enrolled adult populations and 3 enrolled pediatric patients [27,28,30]. Clinical settings included cardiac surgery with cardiopulmonary bypass (n = 5), general ICU (n = 6), sepsis-related AKI (n = 3), contrast-induced nephropathy (n = 1), acute respiratory distress syndrome (ARDS; n = 1), acute decompensated heart failure (ADHF; n = 1), decompensated cirrhosis undergoing therapeutic paracentesis (n = 1), acute cerebral infarction (n = 1), elective intra-abdominal non-vascular surgery (n = 1), traumatic hemorrhagic shock following severe fractures (n = 1), and severe burn injury (n = 1). Regarding FGF23 assay type, 16 studies measured intact FGF23 (iFGF23), 6 measured C-terminal FGF23 (cFGF23), and 1 measured both [19]; the latter study contributed two independent datasets to the diagnostic subgroup analyses. Most studies defined AKI according to KDIGO guidelines, while 3 used AKIN criteria and 1 employed RIFLE classification. Detailed characteristics of the included studies are presented in Table 1. Table 1.Characteristics of included studies.StudyAimCountryDesignPatientSample size (n)Age (years)Male (%)AKI criteriaFGF23 measurementFayed 2026 [15]DiagnosisEgyptPCdecompensated cirrhosis having therapeutic paracentesis8054.3 ± 10.980.0%KDIGOiFGF23Liu 2026 [16]DiagnosisChinaRCelderly ACI patients9069.9 ± 5.556.7%KDIGOiFGF23Zang 2026 [32]PrognosisChinaPCCKD complicated with AKI9567.9 ± 5.152.6%KDIGOiFGF23Gao 2025 [17]DiagnosisChinaPCsevere degree burn14344.3 ± 11.154.6%KDIGOiFGF23Qi 2025 [18]DiagnosisChinaPCtraumatic hemorrhagic shock after severe fractures10747.0 ± 9.786.0%KDIGOiFGF23Zhang 2025 [19]DiagnosisChinaPCICU9658.7 ± 18.053.1%KDIGOiFGF23& cFGF23Zhang D 2025 [20]DiagnosisChinaPCelderly elective intra-abdominal non-vascular surgery100NA57.0%KDIGOiFGF23Gao 2024 [21]DiagnosisChinaRCICU30860.0 (45.0–70.0)56.5%KDIGOiFGF23Wang 2024 [33]PrognosisChinaRCSepsis15763.00 ± 9.663.7%KDIGOiFGF23Zhang 2024 [34]PrognosisChinaRCSepsis20267.1 ± 6.056.9%KDIGOiFGF23Pei 2022 [22]DiagnosisChinaPCSepsis16272.0 (58.8)59.9%KDIGOiFGF23Pramong 2021 [23]DiagnosisThailandPCADHF6270.4 ± 11.841.9%KDIGOcFGF23Xu 2021 [24]DiagnosisChinaRCARDS21056.6 ± 3.162.9%KDIGOiFGF23Gupta 2020 [35]PrognosisIndiaPCICU3739.7 ± 15.756.3%AKIN stage 3iFGF23Fayed 2019 [36]PrognosisEgyptPCICU3060.7 ± 19.056.7%AKINiFGF23Xu 2019 [37]PrognosisChinaRCICU10447.0 ± 17.042.3%KDIGOiFGF23Li 2018 [25]DiagnosisChinaRCCIN20260.0 ± 10.681.7%SCr >0.5 mg/dL or a minimum of 25% increase within 72 h after contrast administrationiFGF23Rygasiewicz 2018 [31]Diagnosis & PrognosisPolandPCICU7961.0 (47.0–75.0)65.0%KDIGOcFGF23Shaker 2018 [26]DiagnosisEgyptPCCPB8047.7 ± 12.867.5%SCr >0.3 mg/dL or >50% within 48 hours after surgeryiFGF23Volovelsky 2018 [27]DiagnosisUSAPCCPB760.7 (0.3–4.7)55.5%KDIGO stages 2–3cFGF23Hanudel 2016 [28]DiagnosisUSAPCCPB323.0 (1.0–7.0)56.0%SCr >50% within 48 hours after surgerycFGF23Speer 2015 [29]DiagnosisGermanyPCCPB85963.7 ± 1.568.6%AKIN score ≥1cFGF23Ali 2013 [30]DiagnosisUSAnested case-controlCPB194.0 (2.0–7.0)47.0%Pediatric-modified RIFLE class score using eGFRcFGF23AKI, acute kidney injury; FGF23, fibroblast growth factor 23; iFGF23, intact FGF23; cFGF23, C-terminal FGF23; PC, prospective cohort; RC, retrospective cohort; ACI, acute cerebral infarction; ADHF, acute decompensated heart failure; CIN, contrast-induced nephropathy; CPB, cardiopulmonary bypass; ICU, intensive care unit; ARDS, acute respiratory distress syndrome. n, number; SCr, serum creatinine; eGFR, estimated glomerular filtration rate; RIFLE, risk, injury, failure, loss of kidney function, and end-stage renal failure; AKIN, acute kidney injury network. Characteristics of included studies. AKI, acute kidney injury; FGF23, fibroblast growth factor 23; iFGF23, intact FGF23; cFGF23, C-terminal FGF23; PC, prospective cohort; RC, retrospective cohort; ACI, acute cerebral infarction; ADHF, acute decompensated heart failure; CIN, contrast-induced nephropathy; CPB, cardiopulmonary bypass; ICU, intensive care unit; ARDS, acute respiratory distress syndrome. n, number; SCr, serum creatinine; eGFR, estimated glomerular filtration rate; RIFLE, risk, injury, failure, loss of kidney function, and end-stage renal failure; AKIN, acute kidney injury network. Quality assessment results Quality assessment results of included studies are shown in Figure 2. With respect to the patient selection domain, the majority of studies demonstrated low risk of bias; the exception was Ali et al. [30], which used a case-control design and was rated at high risk for selection bias. For the index test domain, most studies did not clearly report whether FGF23 measurements were interpreted independently of the AKI reference standard, and none prespecified diagnostic thresholds, resulting in a high or unclear risk of bias for this domain in the majority of studies. The reference standard domain was generally rated at low risk, as most studies applied established AKI criteria. Concerns regarding flow and timing were limited, with most studies rating at low risk for this domain. Overall, most studies were judged to be at low or unclear risk of bias across key domains, but the frequent ‘unclear’ ratings for the index test domain and the small number and size of prognostic studies warrant cautious interpretation of the pooled estimates. Risk of bias and applicability concerns of included studies according to the QUADAS-2 tool. Diagnostic value of circulating FGF23 for AKI In total, 17 studies [15–31] evaluated the diagnostic value of circulating FGF23 in AKI and provided extractable data. One of these studies reported ROC data for both iFGF23 and cFGF23 [19]; so a total of 18 diagnostic datasets were included in the pooled analysis, with the iFGF23 and cFGF23 results extracted and analyzed separately. According to the QUADAS‑2 assessment, risk of bias was generally low in the reference standard and flow and timing domains, whereas the index test domain was frequently rated as high because no study prespecified diagnostic thresholds; most derived cutoffs post hoc via ROC analysis (Supplementary Table 1). The Spearman correlation coefficient between logit-transformed sensitivity and logit (1 - specificity) was −0.318 (p = 0.197), suggesting the absence of a significant threshold effect. Non-threshold heterogeneity was significant (Cochran Q test p < 0.001; I2 = 67%), necessitating the use of a random-effects bivariate model. The pooled diagnostic performance of circulating FGF23 for AKI was as follows: sensitivity 0.79 (95% CI 0.73–0.86), specificity 0.82 (95% CI 0.75–0.89) (Figure 3), PLR 4.40 (95% CI 2.59–6.21), NLR 0.25 (95% CI 0.16–0.34), DOR 17.49 (95% CI 8.67–35.16), and a combined AUC of 0.87 (95% CI 0.81–0.92) (Figure 4). These results indicate that circulating FGF23 has moderate-to-high overall diagnostic performance for AKI. Forest plots of sensitivity and specificity for the diagnostic performance of serum FGF23 in acute kidney injury. Summary receiver operating characteristic (SROC) curve for the diagnostic performance of serum FGF23 in acute kidney injury. To assess the robustness of the pooled estimates, we conducted sensitivity analyses by excluding studies with extreme values (Shaker 2018 [26], sensitivity 0.98), high weights (Speer 2015 [29], n = 860), or nested case–control study with small sample size (Ali 2013 [30], n = 19). Sensitivity analyses demonstrated that pooled sensitivity (0.77–0.80), specificity (0.80–0.84), and AUC (0.85–0.88) were highly robust to exclusion of individual studies, with minimal fluctuation across scenarios. Pooled DOR ranged from 13.77 to 20.19, indicating moderate stability with consistent direction of effect. Notably, exclusion of Shaker 2018 reduced the pooled DOR from 17.49 to 13.77 and markedly decreased sensitivity heterogeneity (I2 from 56.76% to 45.53%), while specificity heterogeneity remained substantial (I2 from 80.75% to 80.07%). Exclusion of Speer 2015 (the largest study, n = 859) modestly increased DOR (17.49–19.23) with minimal impact on heterogeneity, suggesting that this study primarily contributed statistical weight rather than explaining observed between-study variability. Exclusion of Ali 2013 [30] (nested case-control design) had negligible impact on all estimates, confirming that this methodologically distinct study did not distort the pooled analysis (Supplementary Table 2). These findings suggest that Shaker 2018 was an important source of heterogeneity, but its exclusion did not materially alter the overall diagnostic performance. To explore potential sources of heterogeneity, prespecified subgroup analyses were further conducted by FGF23 assay type, sample size, study design, age of population, and country (Table 2). The results showed that iFGF23 demonstrated higher diagnostic performance than cFGF23 across all metrics (AUC 0.91 vs 0.81; sensitivity 0.81 vs 0.80; specificity 0.88 vs 0.70; DOR 29.59 vs 8.94). Subgroup analyses by sample size (≥100 vs <100 patients) and study design (prospective vs retrospective) revealed numerically higher AUC estimates in smaller studies (0.88 vs 0.85) and prospective studies (0.88 vs 0.86); however, the 95% confidence intervals of these subgroup estimates overlapped substantially, suggesting that these factors were not major contributors to the observed heterogeneity. The adult subgroup showed good diagnostic performance (AUC 0.88, 95% CI 0.81–0.93; DOR 19.97, 95% CI 8.79–45.35). For the pediatric subgroup, individual study sensitivities ranged from 0.80 to 0.92 and specificities from 0.67 to 0.70; pooled AUC could not be estimated due to limited sample size. The China subgroup showed an AUC of 0.84 (95% CI 0.77–0.90), while the non-China subgroup showed an approximate AUC of 0.79 (95% CI 0.73–0.84). These findings suggest that assay type may be a meaningful contributor to heterogeneity. Table 2.Subgroup analyses of the diagnostic performance of serum FGF23 for acute kidney injury.SettingNo. of studiesSensitivity (95% CI)Specificity (95% CI)PLR (95% CI)NLR (95% CI)DOR (95% CI)AUC (95% CI)Main analysis180.79 (0.73–0.86)0.82 (0.75–0.89)4.40 (2.59–6.21)0.25 (0.16–0.34)17.49 (8.67–35.16)0.87 (0.81–0.92)cFGF2370.80 (0.70–0.89)0.70 (0.59–0.80)2.61 (1.42–3.80)0.29 (0.12–0.47)8.94 (3.15–25.37)0.81 (0.68–0.90)iFGF23110.81 (0.71–0.90)0.88 (0.80–0.95)6.52 (2.32–10.71)0.22 (0.10–0.34)29.59 (10.02–87.40)0.91 (0.82–0.96)Prospective120.82 (0.74–0.90)0.80 (0.71–0.90)4.16 (1.88–6.44)0.22 (0.10–0.35)18.69 (6.53–53.47)0.88 (0.78–0.94)Retrospective60.76 (0.67–0.85)0.85 (0.76–0.94)5.06 (2.21–7.92)0.28 (0.17–0.39)18.12 (8.36–39.28)0.86 (0.79–0.91)Sample size <100100.82 (0.73–0.92)0.81 (0.71–0.90)4.23 (1.91–6.54)0.22 (0.09–0.35)19.23 (6.42–57.61)0.88 (0.78–0.94)Sample size ≥10080.75 (0.67–0.84)0.83 (0.74–0.93)4.52 (1.82–7.22)0.29 (0.19–0.40)15.30 (6.46–36.22)0.85 (0.76–0.90)China100.74 (0.66–0.81)0.83 (0.74–0.89)4.38 (2.77–6.92)0.31 (0.22–0.43)14.12 (7.04–28.43)0.84 (0.77–0.90)Other country80.86 (0.75–0.92)0.80 (0.65–0.90)4.36 (2.12–8.98)0.18 (0.09–0.37)24.64 (6.06–100.20)0.79 (0.73–0.84)Adult150.79 (0.72–0.86)0.84 (0.77–0.91)4.95 (2.51–7.40)0.25 (0.15–0.34)19.97 (8.79–45.35)0.88 (0.81–0.93)Pediatric30.80-0.920.67-0.702.40-3.060.12-0.308.00-25.67NAFGF23: fibroblast growth factor 23; iFGF23: intact FGF23; cFGF23: C-terminal FGF23; CI: confidence interval; PLR: positive likelihood ratio; NLR: negative likelihood ratio; DOR: diagnostic odds ratio; AUC: area under the curve. Subgroup analyses of the diagnostic performance of serum FGF23 for acute kidney injury. FGF23: fibroblast growth factor 23; iFGF23: intact FGF23; cFGF23: C-terminal FGF23; CI: confidence interval; PLR: positive likelihood ratio; NLR: negative likelihood ratio; DOR: diagnostic odds ratio; AUC: area under the curve. Deeks’ funnel plot asymmetry test indicated no significant publication bias (p = 0.16). Given the modest number of predominantly single-center studies and the presence of between-study heterogeneity, the overall risk of bias due to missing results arising from reporting biases for this diagnostic synthesis was judged as unclear rather than definitively low. To assess the clinical utility of circulating FGF23, a Fagan nomogram was constructed. As shown in Figure 5, when the pretest probability was set at 20%, the post-test probability of a positive result indicating AKI was 52%, while the post-test probability of a negative result missing AKI was 6%, demonstrating clinically meaningful probability shifts. Fagan nomogram illustrating the clinical impact of serum FGF23 testing for acute kidney injury. Prognostic value of circulating FGF23 in AKI Six studies [31–36] analyzed the prognostic value of circulating FGF23 for in-hospital or short-term mortality in AKI patients. One additional study (n = 104) [37] assessed renal recovery as the endpoint and reported a sensitivity of 0.429 and specificity of 0.822 for predicting renal recovery; however, this study was not pooled in the quantitative synthesis, as mortality and renal recovery represent distinct pathophysiological processes and clinical trajectories that are not methodologically appropriate for combined diagnostic performance analysis. The small sample size and distinct endpoint precluded meaningful quantitative synthesis with mortality studies. This study is discussed qualitatively in the Discussion section. Overall risk of bias was acceptable in the patient selection and reference standard domains. The Spearman correlation coefficient between logit-transformed sensitivity and logit (1 − specificity) was −0.928 (p = 0.008), indicating a significant threshold effect. Additionally, substantial non-threshold heterogeneity was observed in both the sensitivity (Cochran Q test, p < 0.001; I2 = 86.2%) and specificity (I2 = 80.4%, p < 0.001) directions, likely reflecting differences in FGF23 assay types (iFGF23 vs. cFGF23), patient populations, clinical settings, and timing of FGF23 measurement across studies. The pooled prognostic performance for mortality prediction was: sensitivity 0.77 (95% CI 0.69–0.84), specificity 0.76 (95% CI 0.70–0.82) (Figure 6), PLR 3.26 (95% CI 2.58–4.11), NLR 0.30 (95% CI 0.22–0.42), DOR 10.89 (95% CI 6.86–17.30), and SROC AUC 0.77 (95% CI 0.70–0.83) (Figure 7), suggesting that circulating FGF23 has moderate overall prognostic accuracy. Forest plots of sensitivity and specificity for the prognostic performance of serum FGF23 in acute kidney injury. Summary receiver operating characteristic (SROC) curve for the prognostic performance of serum FGF23 in acute kidney injury. Sensitivity analysis identified Zang 2026 [32] and Rygasiewicz 2018 [31] as potential outliers. Exclusion of Zang 2026 [32] resulted in model non-convergence, indicating that this study was structurally critical to the model. Exclusion of Rygasiewicz 2018 [31] (fn = 1) yielded stable convergence, with minimal changes in sensitivity (0.77–0.74), specificity (0.76–0.77), and AUC (0.77 to 0.75), all changes were below the clinically significant threshold and 95% CIs overlapped substantially with the primary analysis. Notably, exclusion of Rygasiewicz 2018 [31] reduced sensitivity variance by 72.4%, confirming this study as the primary source of sensitivity heterogeneity (Supplementary Table 3). The persistently negative covariance across all convergent scenarios confirmed a genuine threshold effect as an inherent operating characteristic of this biomarker. Given the small number of included studies (n = 6), subgroup analyses were not performed; the sources of heterogeneity are further discussed in the Discussion. Deeks’ funnel plot asymmetry test indicated no significant publication bias (p = 0.20). However, given the very small number of predominantly single-center studies and the substantial between-study heterogeneity, the overall risk of bias due to missing results arising from reporting biases for this prognostic synthesis was judged as unclear rather than definitively low. Discussion This meta-analysis of 17 diagnostic studies and 6 prognostic studies (mortality endpoint) demonstrates that circulating FGF23 has moderate-to-high diagnostic performance for AKI (SROC AUC 0.87) and moderate prognostic performance for mortality (SROC AUC 0.77), supporting its potential as a clinically useful biomarker across both domains. These findings are consistent with prior single-center studies [10,38–40] and earlier meta-analyses reporting elevated FGF23 in AKI [41,42], and extend the evidence base by providing the first systematic comparison of iFGF23 versus cFGF23 assay types across adult and pediatric populations, together with a pooled prognostic evaluation. The diagnostic performance observed in this study, with a DOR of 17.49, indicates strong discriminative ability, as a DOR exceeding 10 is generally considered reflective of a diagnostically useful test. The PLR of 4.40 suggests that a positive FGF23 result provides moderate confirmatory value, whereas the NLR of 0.25 indicates only modest rule-out ability and underscores that FGF23 alone cannot safely exclude AKI. Taken together, these results confirm that circulating FGF23 is not a replacement for current diagnostic criteria but may serve as a complementary indicator, particularly when used alongside other biomarkers. Two prior meta-analyses have examined the diagnostic performance of circulating FGF23 in AKI. Sun et al. [41] pooled 11 studies (n = 1,946) in critically ill patients and reported an overall SROC AUC of 0.86, sensitivity of 82%, and specificity of 77%, consistent with the estimates obtained in the present study. Tao et al. [42] restricted their analysis to seven prospective studies in adults and reported an AUC of 0.83. The present study, incorporating 23 studies with a search updated through June 2026, yielded comparable overall estimates (AUC 0.87, sensitivity 0.79, specificity 0.82) and additionally provides the first adequately powered subgroup comparison of assay types and the first pooled prognostic analysis. Notably, our subgroup findings regarding assay type diverge from those of Sun et al. [41], who reported that cFGF23 had more stable diagnostic performance (I2 = 26.1%) while the iFGF23 subgroup, based on only two studies, generated an unstable estimate with extreme heterogeneity (I2 = 94.3%), leading those authors to tentatively favor cFGF23. In contrast, with a larger evidence base including 11 iFGF23 datasets and 7 cFGF23 datasets, we found that iFGF23 demonstrated higher overall AUC (0.91 vs 0.81) and specificity (0.88 vs 0.70). This discrepancy likely reflects the instability of earlier iFGF23 estimate due to its extremely limited sample. These divergent results emphasize the importance of adequate subgroup sample sizes and consistent reporting metrics in diagnostic meta-analyses, and highlight the need for further prospective studies directly comparing iFGF23 and cFGF23 performance in the same patient cohorts. Substantial heterogeneity was observed among the diagnostic studies (I2 = 67%). Spearman correlation analysis did not identify a significant threshold effect, suggesting that differences across studies were not primarily driven by variation in diagnostic cutoff values. In subgroup analyses, iFGF23 showed higher diagnostic performance than cFGF23 (AUC 0.91 vs 0.81; specificity 0.88 vs 0.70). These patterns are consistent with the known biological and analytical features of FGF23 [37,38]: in AKI, reduced renal clearance and altered processing of FGF23 fragments may disproportionately affect C-terminal fragments detected by cFGF23 assays, leading to lower specificity, whereas iFGF23 assays quantify only the biologically active full-length hormone and may better reflect pathophysiologically relevant changes. Numerically higher AUCs were also observed in smaller studies and retrospective designs compared with larger and prospective studies, a pattern compatible with small-study effects and spectrum bias, whereby selected populations and overfitted, study-specific cutoff values yield more optimistic test performance. However, the wide and overlapping confidence intervals across these subgroups suggest that sample size and study design are unlikely to be the predominant drivers of heterogeneity. Overall, assay type appears to be a more plausible contributor to between-study variability, while other factors such as AKI etiology, timing of FGF23 measurement, and baseline renal function could not be formally investigated because of limited reporting and remain important topics for future studies. The pediatric subgroup comprised only three studies. This limitation precludes definitive conclusions about FGF23 diagnostic performance in children, where developmental changes in bone mineral metabolism, growth hormone axis, and baseline FGF23 levels may substantially alter the biomarker’s operating characteristics. The absence of age-standardized reference ranges and pediatric-specific assay validation represents a critical evidence gap, particularly given the distinct AKI pathophysiology in infants and children compared with adults. The lower approximate AUC in non-Chinese studies compared with Chinese studies (0.79 vs 0.84) should be interpreted cautiously. This difference may stem from methodological differences in assay standardization, timing of measurement, or clinical practice patterns. These findings underscore the need for multicenter studies spanning diverse geographic and ethnic populations to establish generalizable diagnostic thresholds. The Fagan nomogram analysis provides additional insight into the clinical utility of circulating FGF23. Using a pretest probability of 20% (reflecting published estimates of AKI incidence in hospitalized patients), a positive FGF23 result increases the post-test probability to 52%, while a negative result reduces it to 6%. The low post-test probability associated with a negative result suggests that FGF23 may be particularly useful for ruling out AKI in low-to-moderate risk patients, whereas a positive result alone is insufficient for definitive diagnosis and should prompt further evaluation. With respect to study quality, QUADAS-2 assessment indicated that most studies demonstrated low risk of bias in the patient selection, reference standard, and flow and timing domains. However, the index test domain was rated ‘unclear’ in the majority of studies, as most did not report whether FGF23 measurements were interpreted independently of the AKI reference standard or whether diagnostic thresholds were prespecified. This gap in reporting represents a methodological limitation that could potentially inflate the apparent diagnostic performance of FGF23 across the included studies. This meta-analysis also evaluated the prognostic value of circulating FGF23 across six studies, yielding a pooled AUC of 0.77. However, the extremely high heterogeneity substantially limits confidence in this estimate. A key contributor of heterogeneity may arise from differences in AKI etiology, disease severity at the time of FGF23 measurement, and clinical setting (ICU vs non-ICU). The small number of prognostic studies (n = 6) precluded formal subgroup analysis or meta-regression to disentangle these sources. Sensitivity analysis confirmed that Rygasiewicz 2018 [31], with only one false-negative case, was the primary source of sensitivity heterogeneity; its exclusion reduced sensitivity variance by 72.4% while minimally affecting pooled estimates, supporting the robustness of the overall findings. The persistently negative covariance across convergent scenarios confirmed a genuine threshold effect as an inherent operating characteristic of this biomarker. Consequently, the pooled prognostic AUC should be interpreted as a rough summary measure of FGF23 performance for mortality prediction in AKI rather than as a precise estimate for any single endpoint. Future studies with larger sample sizes, standardized outcome definitions, and prespecified FGF23 measurement timepoints are needed to clarify the prognostic role of FGF23 in specific AKI subpopulations. One identified study (Xu 2019) evaluated renal recovery as an alternative endpoint, however, this endpoint was not pooled with mortality studies due to distinct pathophysiological processes. The prognostic value of FGF23 for renal recovery warrants separate investigation in future research. The divergent threshold effect patterns between diagnostic and prognostic analyses warrant explanation. The absence of a significant threshold effect in diagnostic studies suggests that FGF23 discriminates AKI from non-AKI relatively consistently across different cutoff values when referenced against established creatinine-based criteria. In contrast, the strong threshold effect in prognostic studies indicates that the predictive value of FGF23 for adverse outcomes is critically dependent on the chosen cutoff and the timing of measurement relative to AKI onset. Consequently, prognostic performance is inherently more sensitive to cutoff selection and measurement timing, explaining the observed threshold effect. Taken together with the diagnostic findings, these considerations indicate that the certainty of the evidence supporting the diagnostic use of circulating FGF23 for AKI is low to moderate, owing to concerns about risk of bias in the index test domain, moderate heterogeneity across studies, and imprecision related to relatively small samples with wide confidence intervals. In contrast, the certainty of the evidence for the prognostic use of FGF23 in AKI is very low because of very serious inconsistency from heterogeneous endpoints and effect estimates, serious imprecision related to the small number and size of studies, and potential reporting biases, such that the true prognostic performance of FGF23 may differ substantially from the summary estimates reported here. The body of evidence included in this review has several important limitations. First, most included studies used optimal cutoff values derived post hoc from ROC curve analyses rather than prespecified thresholds, and no internationally validated diagnostic cutoff for circulating FGF23 currently exists; this precludes direct clinical implementation and limits comparability across studies, underscoring the need for prospective threshold validation studies. Second, there was considerable clinical and methodological heterogeneity in patient populations, AKI etiologies, diagnostic criteria, and FGF23 assay types and measurement timepoints. Although prespecified subgroup analyses were used to explore some of these factors, complete resolution of heterogeneity was not achievable, leaving residual uncertainty around the pooled estimates. The prognostic analysis was framed as diagnostic performance for outcome prediction, which facilitated direct comparison with the diagnostic component but may not capture all biases relevant to traditional prognostic factor studies (e.g., time-varying effects, competing risks). Future studies should standardize etiology classification and sampling timepoints, and employ complementary time-to-event approaches for prognostic evaluation. Besides, geographic concentration of included studies (13 of 23 from China) reflects active research interest in FGF23 biomarkers within Asian medical centers, but raises important questions about external validity. We could not perform subgroup analyses by continent or race/ethnicity due to insufficient representation from other regions and inconsistent reporting of racial data. Future multicenter studies spanning diverse populations are needed to establish generalizable thresholds and assess potential population-specific performance differences. Third, the prognostic analysis was based on a small number of studies with heterogeneous clinical settings and FGF23 assay types, which substantially constrains the precision and generalizability of the prognostic conclusions. Finally, although no significant publication bias was detected by Deeks’ funnel plot, the overall number and size of available studies remain modest, and the presence of small-study effects or selective reporting cannot be entirely excluded. Conclusion This meta-analysis demonstrates that circulating FGF23 has moderate-to-high diagnostic performance for AKI and moderate prognostic performance for mortality, supporting its potential as a clinically useful complementary biomarker. However, the prognostic analysis was limited by a small number of studies and substantial heterogeneity, and conclusions in this domain should be interpreted with caution. Future large-scale, multicenter prospective studies using standardized iFGF23 assays, prespecified diagnostic thresholds, and clearly defined outcome endpoints are needed to validate these findings and establish the clinical utility of FGF23 in specific AKI subpopulations. Supplementary Material Supplement material.docx