Proenkephalin as a biomarker correlates with acute kidney injury: a systematic review with meta-analysis and trial sequential analysis

Two reviewers (LC Lin and HW Liao) independently searched the Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, EMBASE, MEDLINE, ClinicalTrials.gov and Cnki.net until Jun 26, 2023, using terms associated with AKI (“acute renal failure,” “acute kidney impairment,” “acute kidney insufficiency” and “AKI”) and PENK (“proenkephalin A,” “proenkephalin A 119-159,” “PenKid” and “PENK”). The search strategies are listed in the Additional file 1. We also manually checked the reference list of related review articles, editorials and identified studies to identify any further randomized controlled trials (RCTs). The full texts of potentially eligible RCTs and observational studies were retrieved and evaluated for inclusion. Additionally, we contacted the original authors to acquire additional information in cases where the data were incomplete.

This systematic review and meta-analysis was performed in accordance with the Preferred Reporting Items of Systematic Reviews and Meta-Analyses (PRISMA) recommendation and Cochrane methods. The study protocol was registered in PROSPERO [CRD42023424693].

Studies meeting the inclusion criteria were identified based on the following criteria: (1) evaluation of the diagnostic performance of PENK for AKI in adult patients and (2) provision of comprehensive information, including sample size, sensitivity and specificity at a designated cutoff value, thereby facilitating the pooling of data for accuracy analysis. Exclusion criteria encompassed duplicate publications, case reports, conference abstracts and non-original articles, such as reviews and commentaries. Language restrictions were not imposed.

Two reviewers (LC Lin and HW Liao) independently reviewed full-text articles, individual study protocols and the template for case report forms and evaluated the risk of bias in methodology. Discrepancies were resolved through discussions with a third investigator (VC Wu). The data extracted from the enrolled studies included the first author, publication year, study designs, sample/event sizes, clinical settings, patients' characteristics (age, sex, comorbidities and baseline renal function), timing of PENK measurement, AKI criteria, study endpoint and diagnostic accuracy assessment (specifically the cutoff value of PENK for the early diagnosis of AKI along with the corresponding sensitivity and specificity). The baseline characteristics of included studies are illustrated in Table 1.

The primary outcome of this study was the development of AKI, treated as a binary outcome.

The risk of bias and applicability of the individual study were evaluated using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool [17, 18]. Four crucial domains were assessed, namely patient selection, index test, reference standard and flow and timing. Each domain was categorized as having a low, unclear or high risk of bias. Any disagreements in the quality assessment were resolved by discussion and consensus [19]. The findings of the assessment were then visually depicted in a summarized graphical format.

We hypothesized that multiple factors, including baseline characteristics such as age, sex and pre-existing conditions such as hypertension, diabetes, CKD and cardiac events, could have a substantial impact on the observed patient outcomes in the reported studies. We also considered the use of mean values for grouping, and whether the studies included surgical patients exclusively or a combination of surgical and medical patients, along with patients suffering from sepsis. The AKI criteria employed—RIFLE (Risk, Injury, Failure, Loss, end-stage renal disease (ESRD)), AKIN (Acute Kidney Injury Network), KDIGO (Kidney Disease: Improving Global Outcomes), and the severity of AKI and the variation in follow-up durations (greater than 2 days or less than or equal to 2 days) were also examined. Additionally, to evaluate the potential influence of small-study effects on overestimation, we stratified our analysis based on study size [20].

The determination of true positives, true negatives, false positives and false negatives was carried out by utilizing the sample size, event rate, sensitivity and specificity information obtained from each respective study. In instances where the sensitivity and specificity values were not explicitly provided in the studies, we employed WebPlotDigitizer (version 4.6) to digitally extract the data from the receiver operating characteristic (ROC) curve [21]. The overall diagnostic performance of PENK was evaluated by utilizing a summary receiver operating characteristic (SROC) curve along with measures such as pooled sensitivity, specificity, positive likelihood ratio (LR) and negative LR [22]. The optimal cutoff point for PENK in association with the development of AKI was ascertained through the methodology introduced by Steinhauser et al. [23]. We adopted the logistic distribution assumption and employed the model that minimized the restricted maximum likelihood (REML). A weighting parameter of 0.5 was applied to ensure an equal emphasis on sensitivity and specificity. The optimal cutoff value was determined as the point that maximizes the Youden index [24]. When examining the diagnostic accuracy of AKI using PENK, we utilized the diagnostic odds ratio (DOR) for comparison. We compared the diagnostic capability of PENK with neutrophil gelatinase-associated lipocalin (NGAL), a kidney tubular damage marker known for its good diagnostic performance [10].

Fagan diagrams were used to examine the clinical applicability of PENK as an early indicator of AKI. Heterogeneity was quantified using the I² statistics, with substantial heterogeneity defined as I² > 50%. Subgroup analysis and meta-regression were conducted to investigate potential sources of heterogeneity observed between the included studies. Funnel plots were utilized to assess the presence of publication bias. Moreover, to account for type-I and type-II errors and predetermined number of patients was reached, trial sequential analysis (TSA) was performed. The TSA was set with a power level of 90% and a two-tailed α level of 0.05 [25‐28]. All statistical analyses were conducted using Stata software (version 16) with the midas package, R software (version 3.6.0) or TSA software (version 0.9.5.10 Beta).

A summary of the study selection process is provided in Additional file 1. The initial database search yielded 175 articles. Following the removal of duplicates, the titles and abstracts of 81 articles were assessed. Eventually, a total of 23 studies met the eligibility criteria for a full-text review. Among these, 11 observational studies comprising 3969 patients reported data on the occurrence of AKI with PENK and were included in the meta-analysis [15, 16, 29‐37]. We contacted the corresponding authors of nine studies via email for missing data clarification, and two provided additional information [32, 34]. The population characteristics and performance of plasma PENK in each individual study are summarized in Tables 1 and 2. The mean baseline SCr levels ranged from 0.88 to 1.95 mg/dL (77–173 µmol/l). The included studies encompassed a broad range of clinical settings, including sepsis (three studies, 1379 patients), acute heart failure (three studies, 2223 patients), cardiac surgery (two studies, 199 patients), liver transplant (one study, 57 patients) and contrast medium exposure (one study, 111 patients). Among the included studies, seven studies employed the KDIGO criteria for defining AKI, while two studies utilized the AKIN criteria. Additionally, two studies specifically focused on advanced stages of AKI, while the other studies encompassed any stage of AKI. The duration of follow-up varied across the studies.

The comprehensive evaluation of study quality was guided by the QUADAS-2 framework (Additional file 1: Fig. S1). Within the patient selection domain, it was determined that one study incurred a high risk of bias, attributed to the non-enrollment of consecutive patients. Furthermore, two studies were classified as having an unclear risk of bias due to insufficient information regarding patient selection. However, all studies under scrutiny demonstrated a low risk of bias in both the index test and reference standard domains. Regarding flow and timing, three studies warranted an unclear risk of bias as not all participants were included in the analysis [17, 18].

The incidence of AKI was determined from the complete set of included studies giving a total patient pool of 3,969 individuals. Of these, 929 developed AKI (23.4%). The diagnostic values, threshold levels and the sensitivity and specificity of PENK from each study are shown (Table 2). Among them, the optimal cutoff values of PENK, correlating with the development of AKI, were reported in eight studies, with a mean value of 91.5 pmol/L. The range of sensitivity for identifying AKI lies between 0.55 and 0.90, and the specificity extends from 0.52 to 1.00. The overall sensitivity of PENK for forecasting the occurrence of AKI, as shown in the forest plot, is 0.69 (95% CI 0.62–0.75), while the combined specificity is 0.76 (95% CI 0.68–0.82) (Fig. 1), indicating that PENK has a moderate ability to correctly identify AKI cases and a relatively good ability to accurately identify non-AKI cases. Significant heterogeneity was observed in terms of sensitivity (I² = 70.42%, p < 0.001) and specificity (I² = 93.94%, p < 0.001). In regard to the optimal threshold, we employed the different random intercepts and common random slope model to achieve the smallest REML criterion. The determined optimal cutoff value was 57.3 pmol/L.

The SROC curve, which illustrates the overall ability of PENK for the early diagnosis of AKI, shows an area under curve (AUC) of 0.77 (95% CI, 0.73–0.81). This suggests that PENK has a moderately accurate discriminatory ability in early detection of AKI (Fig. 2). The positive LR was calculated as 2.83 (95% CI 2.06–3.88), indicating that a positive PENK result increases the likelihood of AKI by approximately threefold. Conversely, the negative LR was found to be 0.41 (95% CI 0.33–0.52) (Fig. 3). Fagan nomograms were utilized to illustrate the effect of positive and negative results on the post-test probability of AKI development. By assuming a pre-test probability of AKI of 25%, based on the observed AKI incidence of 23.4% in this study, the Fagan nomogram demonstrates that when the PENK result is above the cutoff value, the post-test probability of AKI increases to 49%. Conversely, when the PENK result is below the cutoff value, the post-test probability of AKI decreases to 12% (Additional file 1: Fig. S2a). If the pre-test probability for AKI is set at 75%, the post-test probability of AKI increases to 89% when the PENK value is above the cutoff, whereas it decreases to 55% when the PENK value is below the cutoff (Additional file 1: Fig. S2b).

To explore the potential sources of the heterogeneity in PENK's diagnostic accuracy for AKI, we carried out subgroup analysis and meta-regression. Subgroup analysis was conducted encompassing various variables such as patient characteristics (age, gender, prevalence of HTN, DM and CKD), clinical settings (cardiac events versus non-cardiac events; sepsis versus non-sepsis; surgery versus medical/mixed), AKI severity, AKI definition (KDIGO criteria versus non-KDIGO criteria), follow-up duration and study size (Additional file 1: Table 1). The results derived from the thorough subgroup analysis demonstrate the robust performance of PENK across different patient groups, with the notable exception of those in the large size group.

Our findings revealed that PENK's overall diagnostic ability for AKI was higher in smaller studies (AUC: 0.81, 95% CI 0.78–0.85) compared to larger counterparts (AUC: 0.71, 95% CI 0.65–0.75). The outcomes stemming from the meta-regression analysis indicate that the coexistence of HTN and CHF appears to reduce in the specificity of PENK's diagnostic capacity concerning the onset of AKI (Fig. 4).

The cumulative Z-curve, as analyzed through trial sequential analysis (TSA), demonstrated that the required information size of 1723 patients was exceeded. Moreover, the penalized Z-curve exceeded the conventional threshold value of Z = 1.96, offering additional substantiation for the exclusion of AKI based on negative PENK levels. These findings strongly support the notion that PENK is an effective biomarker for ruling out the presence of AKI and that low levels of PENK provide robust evidence for excluding AKI (Fig. 5).

Three studies concurrently evaluated the diagnostic performance of NGAL for AKI [31, 35, 37].

Among these, two studies offered sufficient data to compute the diagnostic odds ratio (DOR) [31, 37]. As illustrated in Additional file 1: Fig. S3, the diagnostic accuracy of PENK for AKI was not inferior to NGAL (p = 0.62, test for between group differences (random-effects model)).

We created Deek’s funnel plot to assess the potential for publication bias, and these plots exhibited largely symmetrical patterns. This finding indicates that the likelihood of publication bias in this meta-analysis is non-significant (p = 0.28) (Additional file 1: Fig. S4).

According to the GRADE (Grading of Recommendations, Assessment, Development and Evaluation) framework, the strength of evidence regarding the diagnostic accuracy of PENK for AKI was assessed. The level of certainty surrounding the primary outcome was deemed to be low, primarily because the studies we included were of an observational kind. Despite minimal risk associated with bias, indirectness, imprecision and publication bias, the reliability of the evidence was reduced owing to inconsistencies found within the results of the studies we analyzed (see. Additional file 1).

Our analyses show that patients with lower PENK levels have a substantially reduced risk of developing AKI. Such negative predictive power provides invaluable insight to clinicians in their decision-making process. The 23rd Acute Disease Quality Initiative (ADQI) consensus group advocates incorporating biomarkers as complementary tools alongside traditional methods for AKI risk stratification, cause identification, severity assessment and prediction of recovery [9]. However, patients with heart failure and HTN can result in a higher false-positive rate when using PENK as an early indicator of AKI. The biological plausibility of this observation stems from the endogenous opioid system in the regulation of cardiovascular function and fluid homeostasis. Notably, elevated levels of PENK have been documented in patients with heart failure and are considered a protective mechanism to counter-regulate the sympathetic nervous system overdrive in the early stages of heart failure [42, 43]. Consequently, this adaptive response may have implications for the performance of PENK as an early indicator for AKI, leading to a higher rate of false-positive results in the presence of heart failure and HTN. Given that AKI is a common among hospitalized patients suffering from HTN or heart failure [40, 44‐47], it is warranted to search for more suitable biomarkers in these subgroups. The observed diminished diagnostic performance of PENK for AKI in larger studies merits attention, particularly as these studies included a higher proportion of patients with underlying heart failure. This factor complicates the task of determining whether the variance in diagnostic accuracy is primarily due to inherent small-study effects or is influenced by the elevated prevalence of heart failure in the larger study cohorts [20].

Incorporating PENK into patient care could potentially facilitate the identification of patients at high risk for AKI, who might benefit from more intensive surveillance and personalized prevention efforts. Such strategies may include optimizing fluid status, the judicious use of nephrotoxic agents and prophylactic intravenous hydration prior to contrast media exposure. Further prospective studies are essential to ascertain if PENK-guided interventions truly enhance patient outcomes. Additionally, a deeper understanding of the interactions between the endogenous opioid system, cardiovascular function and PENK's diagnostic performance in AKI is warranted to better inform clinical decision making.

While our study delivers promising outcomes, it is important to acknowledge several limitations. Firstly, the meta-analysis was hampered by moderate sample sizes across most studies, leading to significant heterogeneity. Although no noticeable publication bias was found, the limited number of studies precluded extensive subgroup analysis. Secondly, a variety of PENK analysis methods were employed in these studies, with nine studies using immunoluminometric assays, one study using enzyme-linked immunosorbent assay and one study lacking information on the specific assay used, and thus, determining an optimal PENK cutoff value may prove challenging. Thirdly, it is known that PENK levels are influenced by the glomerular filtration rate (GFR); however, only a few studies addressed this issue and had conflicting results. Two studies concluded that PENK at admission was an independent indicator of AKI, even when accounting for factors such as age, gender, medical history and estimated GFR [33, 37]. Nonetheless, Rosenqvist et al. observed a reduced diagnostic capacity of PENK for AKI when further considering estimated GFR [15]. This discrepancy calls for more research incorporating adjustments for baseline renal function to elucidate the true prognostic value of PENK. Fourthly, evidence from two studies suggested that changes in PENK levels over time may serve as more reliable indicator for AKI development compared to single baseline measurements [16, 29]. Due to the limited number of studies, we were unable to perform an analysis to determine the optimal timing and thresholds for these dynamic changes that would possible enhance the diagnostic accuracy. Furthermore, the small sample sizes in the majority of the included studies could potentially lead to an overestimation of effects and introduce bias due to sampling error [20, 48]. Finally, our meta-analysis exhibited heterogeneity in both pooled sensitivity and specificity, likely due to variations in study design, PENK measurement timing and the method of PENK analysis used. Despite these limitations, our study's conclusions are derived from a variety of studies with differing designs and clinical contexts. Future research should explore how the specific etiology of AKI and its severity affect PENK's diagnostic accuracy. These considerations could be integrated into upcoming randomized controlled trials, aiding in determining optimal cutoff values for various clinical settings, thereby improving the timely diagnosis and management of incident AKI. Additionally, further exploration of the underlying AKI mechanisms might enhance diagnostic performance and timely treatment, potentially reducing the high mortality rate among AKI patients.