34
views
0
recommends
+1 Recommend
0 collections
    0
    shares
      • Record: found
      • Abstract: found
      • Article: found
      Is Open Access

      Feature Ranking in Predictive Models for Hospital-Acquired Acute Kidney Injury

      research-article

      Read this article at

      Bookmark
          There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

          Abstract

          Acute Kidney Injury (AKI) is a common complication encountered among hospitalized patients, imposing significantly increased cost, morbidity, and mortality. Early prediction of AKI has profound clinical implications because currently no treatment exists for AKI once it develops. Feature selection (FS) is an essential process for building accurate and interpretable prediction models, but to our best knowledge no study has investigated the robustness and applicability of such selection process for AKI. In this study, we compared eight widely-applied FS methods for AKI prediction using nine-years of electronic medical records (EMR) and examined heterogeneity in feature rankings produced by the methods. FS methods were compared in terms of stability with respect to data sampling variation, similarity between selection results, and AKI prediction performance. Prediction accuracy did not intrinsically guarantee the feature ranking stability. Across different FS methods, the prediction performance did not change significantly, while the importance rankings of features were quite different. A positive correlation was observed between the complexity of suitable FS method and sample size. This study provides several practical implications, including recognizing the importance of feature stability as it is desirable for model reproducibility, identifying important AKI risk factors for further investigation, and facilitating early prediction of AKI.

          Related collections

          Most cited references27

          • Record: found
          • Abstract: found
          • Article: found
          Is Open Access

          Gene selection and classification of microarray data using random forest

          Background Selection of relevant genes for sample classification is a common task in most gene expression studies, where researchers try to identify the smallest possible set of genes that can still achieve good predictive performance (for instance, for future use with diagnostic purposes in clinical practice). Many gene selection approaches use univariate (gene-by-gene) rankings of gene relevance and arbitrary thresholds to select the number of genes, can only be applied to two-class problems, and use gene selection ranking criteria unrelated to the classification algorithm. In contrast, random forest is a classification algorithm well suited for microarray data: it shows excellent performance even when most predictive variables are noise, can be used when the number of variables is much larger than the number of observations and in problems involving more than two classes, and returns measures of variable importance. Thus, it is important to understand the performance of random forest with microarray data and its possible use for gene selection. Results We investigate the use of random forest for classification of microarray data (including multi-class problems) and propose a new method of gene selection in classification problems based on random forest. Using simulated and nine microarray data sets we show that random forest has comparable performance to other classification methods, including DLDA, KNN, and SVM, and that the new gene selection procedure yields very small sets of genes (often smaller than alternative methods) while preserving predictive accuracy. Conclusion Because of its performance and features, random forest and gene selection using random forest should probably become part of the "standard tool-box" of methods for class prediction and gene selection with microarray data.
            Bookmark
            • Record: found
            • Abstract: found
            • Article: not found

            Serving the enterprise and beyond with informatics for integrating biology and the bedside (i2b2).

            Informatics for Integrating Biology and the Bedside (i2b2) is one of seven projects sponsored by the NIH Roadmap National Centers for Biomedical Computing (http://www.ncbcs.org). Its mission is to provide clinical investigators with the tools necessary to integrate medical record and clinical research data in the genomics age, a software suite to construct and integrate the modern clinical research chart. i2b2 software may be used by an enterprise's research community to find sets of interesting patients from electronic patient medical record data, while preserving patient privacy through a query tool interface. Project-specific mini-databases ("data marts") can be created from these sets to make highly detailed data available on these specific patients to the investigators on the i2b2 platform, as reviewed and restricted by the Institutional Review Board. The current version of this software has been released into the public domain and is available at the URL: http://www.i2b2.org/software.
              Bookmark
              • Record: found
              • Abstract: found
              • Article: not found

              A European Renal Best Practice (ERBP) position statement on the Kidney Disease Improving Global Outcomes (KDIGO) Clinical Practice Guidelines on Acute Kidney Injury: Part 1: definitions, conservative management and contrast-induced nephropathy†

              Introduction The broad clinical syndrome of acute kidney injury (AKI) encompasses various aetiologies, including specific kidney diseases (e.g. acute interstitial nephritis), non-specific conditions (e.g. renal ischaemia) as well as extrarenal pathology (e.g. post-renal obstruction). AKI is a serious condition that affects kidney structure and function acutely, but also in the long term. Recent epidemiological evidence supports the notion that even mild, reversible AKI conveys the risk of persistent tissue damage, and severe AKI can be accompanied by an irreversible decline of kidney function and progression to end-stage kidney failure [1–3]. The Kidney Disease Improving Global Outcomes (KDIGO) Clinical Practice Guidelines for AKI [4] were designed to systematically compile information on this topic by experts in the field. These guidelines are based on the systematic review of relevant trials published before February 2011. Nevertheless, for many sections of the guidelines, appropriate supporting evidence is lacking in the literature. As a consequence, variations in practice will inevitably occur when clinicians take into account the needs of individual patients, available resources and limitations unique to a region, an institution or type of practice. Therefore, in line with its philosophy [5], the European Renal Best Practice (ERBP) wanted to issue a position statement on these guidelines. A working group was established to produce guidance from the European nephrology perspective, based on the compiled evidence as presented, with an update of the literature up to March 2012, following the methodology as explained in the ERBP instructions to authors [6]. The present document will deal with the diagnosis and prevention of AKI, and contrast-induced nephropathy (CIN) (Sections 1–4 of the KDIGO document), and other chapters will be discussed in a separate position statement. As a general rule, we will only mention those guideline statements of the KDIGO document that we have amended, even when the change is small. If a KDIGO recommendation is not repeated, it can be considered as endorsed by ERBP as is, unless specifically stated otherwise. 1: AKI definition 1.1: Definition and classification of AKI 1.1.1 We recommend using a uniform definition of AKI, based on urinary output and on changes in serum creatinine (SCr) level. It is important that both criteria are taken into account. (1C) 1.1.2 We recommend diagnosing and indicating the severity of AKI according to the criteria in the table below: (ungraded statement) Stage 1: one of the following: Serum creatinine increased 1.5–1.9 times baseline Serum creatinine increase >0.3mg/dl (26.5 µmol/l) Urinary ouotput 3 times baseline Serum creatinine increases to >4.0mg/dl (353 µmol/l) Initiation of renal replacement therapy Urinary output 210 mg/dL) in the critically ill. However, this was a single-centre trial, and in a larger randomized multicentre trial of intensive versus conventional insulin therapy, the NICE-SUGAR trial [36], a blood glucose target of 81–108 mg/dL resulted in higher mortality than a target of <180 mg/dL, without any benefit in preventing or improving AKI. The same study also confirmed previous findings of increased incidence of hypoglycaemia, and the associated risk of death, when targeting low glycaemia levels. In two recent meta-analyses of trials on intensive versus conventional glycaemic control, pooled relative risk of death with intensive insulin therapy was only slightly lower, whereas relative risk of hypoglycaemia was much higher [37]. Lowering glycaemia could thus potentially be beneficial, but this small benefit is easily offset by the much higher risk of hypoglycaemia [38]. Overall, these data do not support the use of intensive insulin therapy aiming to control plasma glucose at 110 mg/dL or lower in critically ill patients as a general rule. On the other hand, it cannot be denied that insulin therapy for preventing severe hyperglycaemia is beneficial. Based on these considerations, ERBP suggests keeping glycaemia between 110 and 180 mg/dL. We strongly recommend regular control of glycaemia, with appropriate instructions on what action should be undertaken based on the result of a certain glycaemic value, when insulin therapy is initiated. In epidemiological studies, protein–calorie malnutrition is an important independent predictor of in-hospital mortality in patients with AKI, but very few systematic studies have assessed the impact of nutrition on clinical end points. Recommendations are therefore largely based on expert opinion. There is no evidence to support that giving proteins can invert the catabolic process in patients with AKI. According to ERBP, no meaningful guidance can be provided. As such, the ERBP group does not endorse the KDIGO statements relating to administration of proteins. As there is no proven benefit of administering high quantities of protein to patients with AKI, initiating high-volume CRRT with the sole aim to remove extra uraemic waste products resulting from high protein loading, cannot be recommended. Several RCT's have demonstrated the beneficial effect of providing enteral versus parenteral nutrition in different conditions as soon as possible in ICU patients [39, 40]. A recent large RCT indicated that early initiation of parenteral nutrition in patients not meeting the recommended caloric intakes by enteral feeding leads to higher mortality rates and longer ICU stay [41]. Although these studies have mostly not reported patients with AKI as a separate subgroup, there is no reason to believe that results would be different in this patient group. As parenteral feeding seems not to improve outcomes in a general ICU population, and as parenteral feeding can lead to accumulation of uraemic waste products and increased fluid loading, and thus ultrafiltration need, and, in AKI patients, it should only be used cautiously. 2.2.3 The use of diuretics in AKI. 2.2.3.1 We recommend diuretics should not be used to prevent AKI. (1B) 2.2.3.2 We suggest not using diuretics to increase urinary volume in established AKI, except for the management of volume overload. (2C) Rationale Since fluid retention is one of the major symptoms of impaired kidney function, diuretics are often used for patients with or developing AKI. Mostly, loop diuretics such as furosemide are administered to patients with AKI to convert oliguric to non-oliguric AKI, and to facilitate fluid management. However, some reports have indicated that the use of diuretics is associated with harmful effects maybe because circulating volume is reduced excessively, thereby worsening renal haemodynamics. The use of diuretics can also delay the recognition of AKI and nephrology consultation [42]. In meta-analyses, the use of furosemide was not associated with any significant clinical benefits in the prevention and treatment of AKI in adults, and high doses were associated with an increased risk of ototoxicity [43, 44]. The ERBP work group therefore endorses both recommendations on the use of diuretics in patients with AKI. 2.2.4 Pharmacological interventions. 2.2.4.1 We recommend low-dose dopamine should not be used to prevent or treat AKI. (1A) 2.2.4.2 We do not recommend using fenoldopam to prevent or treat AKI. (1C) 2.2.4.3 We do not recommend using atrial natriuretic peptide (ANP) to prevent (1C) or treat (1B) AKI. 2.2.4.4 We do not recommend using recombinant human (rh)IGF-1 to prevent or treat AKI. (1B) Rationale With multiple negative studies, including a randomized, double-blind, placebo-controlled trial of adequate size and power, ‘low-dose’ (1–3 mg/kg/min) dopamine has been abandoned for the prevention and treatment of AKI [32]. Smaller clinical studies have reported a potentially beneficial effect (prevention of need for RRT) of fenoldopam, a pure dopamine Type-1 receptor agonist, in patients with established AKI after cardiothoracic surgery [45], but larger trials are lacking. In contrast, results on the use of fenoldopam for the prevention of AKI were not positive. Taken together, no data from adequately powered multicentre trials with clinically significant end points and adequate safety are available to recommend fenoldopam to either prevent or treat AKI. In addition, concerns about a potentially harmful dose-dependent hypotensive action, and about the high cost remain. Also, the beneficial impact of norepinephrine on mortality and AKI is well established [30] in these conditions, and should remain as first-line therapy, also in the function of its low cost. As a consequence, ERBP does not recommend the use of fenoldopam. There are no trials to support the use of ANP, urodilatin and brain natriuretic peptide (BNP—nesiritide), for prevention or treatment of AKI. In view of the paucity of robust data from large intervention trials, and the fact that all substances may induce serious adverse effects such as hypotension and arrhythmias, the ERBP group considers that their use cannot be recommended. The list of substances tested in the setting of experimental and clinical AKI is long, and among them are recombinant human insulin-like growth factor-1 (IGF-1) and recombinant human erythropoietin. As with many other agents, clinical studies on IGF-1 were disappointing. Under these circumstances, the ERBP feels that their use cannot be recommended until proof of a beneficial effect is provided. 2.2.5 Prevention of aminoglycoside- and amphotericin-related AKI. 2.2.5.1 We suggest not using more than one shot of aminoglycosides for the treatment of infections unless no suitable, less nephrotoxic, therapeutic alternatives are available. (2A) 2.2.5.2 We recommend that, in patients with normal kidney function in steady state, aminoglycosides are administered as a single-dose daily rather than multiple-dose daily treatment regimens. (1B) An exception to this recommendation can be patients with endocarditis, where inconsistent evidence on non-inferiority of single versus multiple daily dosing is reported. (1D) 2.2.5.3 We recommend monitoring aminoglycoside drug levels when treatment with multiple daily dosing is used for more than 24h. (1A) 2.2.5.4 We suggest monitoring aminoglycoside drug levels when treatment with single-daily dosing is used for more than 48h. (2C) 2.2.5.5 We suggest using topical or local applications of aminoglycosides (e.g. respiratory aerosols, instilled antibiotic beads), rather than intravenous (i.v.) application, when feasible and suitable. (2B) 2.2.5.6 We recommend that patients receiving whatever formulation of amphotericin B should receive adequate sodium loading and potassium suppletion (1B). We suggest balancing the presumed lower nephrotoxicity of lipid formulations against their higher cost. (2D) 2.2.5.7 We suggest balancing the need for adequate antimycotic treatment against the potential risk of nephrotoxicity in selecting the most suitable antimycotic agent. (Ungraded statement) Rationale Aminoglycosides are highly potent, bactericidal antibiotics. They have many favourable attributes, including their remarkable stability, predictable pharmacokinetics, low incidence of immunologically mediated side effects and lack of haematologic or hepatic toxicity. Although nephrotoxicity, and ototoxicity, remain major concerns, these events appear to be due to cumulative exposure, and their occurrence after single shot administration is exceptional. On the other hand, due to their potent bactericidal activity, aminoglycosides can help to reverse sepsis-related haemodynamic instability, and thus risk for AKI. In the light of recent developments with progressive antimicrobial resistance to a number of other classes of agents, aminoglycosides remain useful antibiotics. In this perspective, ERBP does not object to the use of aminoglycosides as a single-shot administration in certain conditions. However, careful dosing and therapeutic drug monitoring should be applied to mitigate the risk of AKI with these antibiotics when more than one dose is administered. We recommend that they should be used for as short a period of time as possible. There are several approaches to avoid nephrotoxicity of amphotericin B in patients at risk. In the opinion of ERBP, the KDIGO guideline has focused too little attention to sodium loading as a potential nephroprotective strategy. Although there is no hard evidence to support the protective effect of sodium loading, the cost is low, and therefore ERBP recommends that it should be implemented in all patients receiving any formulation of amphotericin B. Numerous studies with lipid formulations of this drug have been published. However, a well-performed review on the topic pointed to the high risk of bias in these studies, making the conclusions rather weak [46]. The ERBP believes that there is insufficient evidence to recommend the use of the lipid formulations of amphotericin B as being clearly superior to the conventional formulation. Another approach to prevent amphotericin B nephrotoxicity is to use alternative agents, such as the azoles (voriconazole, fluconazole, itraconazole and posaconazole) and echinocandins (caspofungin, anidulafungin and micafungin). Although these agents have clearly a better record with regard to nephrotoxicity, there is the potential of hepatotoxicity, and there is uncertainty on the therapeutic equivalence. A Cochrane review [47] pointed to substantial biases in the RCT's dealing with this question. In this setting, ERBP believes that the recommendation as issued by KDIGO is too strong, ambivalent and not supported by the evidence. The ERBP workgroup judged that azoles and echinocandines can be used in low-grade infections, but that their role in life threatening infections is unclear, and that in these conditions, the risk of AKI should not outweigh the risk of death by uncontrolled infection. 3. Contrast-induced nephropathy Besides the KDIGO guidelines, many other bodies issued recommendations on the treatment and prevention of CIN. As early as 2007, a series of guidelines on the prevention of CIN in high-risk patients undergoing cardiovascular procedures were released [48], and in 2011, the European Society of Urogenital Radiology (ESUR) released their new guidelines on CIN [49]. 3.1 Definition, epidemiology and prognosis 3.1.1 We recommend that for CIN, the same definition and grading is used as for AKI (see 1.1). (Ungraded statement). 3.1.2 We recommend that before an intervention which encompasses a risk for CIN, a baseline serum creatinine should be determined. (Ungraded statement) 3.1.3 We suggest that in high-risk patients, a repeat serum creatinine is performed 12 and 72h after administration of contrast media. (2D) 3.1.4 We suggest not considering only CIN in individuals who develop changes in kidney function after administration of intravascular contrast media, but also other possible causes of AKI. (Not Graded) Rationale The ERBP work group is not aware of any pathophysiological or epidemiological reason why the definition and staging of CIN should be different from the general AKI definition. This definition is slightly different from the ESUR criteria [49] for contrast-induced nephropathy, which requires an increase in SCr by more than 25% or 44 µmol/L in the 3 days following intravascular administration of contrast medium (CM) in the absence of an alternative aetiology. Thus, many patients with an SCr increase ranging from 26.5 to 44 µmol/L following CM administration would be considered as presenting Stage 1 AKI but not as CI nephropathy. However, for the sake of clarity and uniformity, ERBP recommends to use the general AKI criteria. Remarkably, studies have also pointed out that in many hospitalized patients not receiving contrast, an increase in serum creatinine was observed [50]. As such, in patients who did receive contrast, one should be cautious to attribute AKI to the contrast, and other underlying causes for AKI should be explored. The moment when the repeat serum creatinine should be measured is a matter of debate. According to ESUR, it should be done in the 3 days following intravascular administration of CM. Some studies suggest that the peak of SCr could even occur later, especially in patients with diabetes and pre-existing CKD [51–56] which really underlines the need for an extended period of renal function survey. On the other hand, the percentage increase in serum creatinine from baseline after 12h showed a good prediction for later development of renal impairment [57]. The reliability of other renal function markers such as cystatin C should be further evaluated. On the other hand, the importance of urinary output for diagnosing CIN should be emphasized. 3.2.1 We recommend balancing the risk for CIN against the benefit of administering contrast. (Not Graded) 3.2.2 We recommend considering alternative imaging methods not requiring contrast administration in patients at increased risk for CIN, so long as these yield the same diagnostic accuracy. (Not Graded) Rationale Although these recommendations seem trivial, it is important to balance the potential risk of CIN against the potential gain of administering contrast in the clinical decision process. Risk for CIN increases with decreasing pre-existing GFR. A CIN Consensus Working Panel [58] agreed that CIN risk becomes clinically significant when the baseline SCr concentration is ≥1.3 mg/dL (≥115 mmol/L) in men and ≥1.0 mg/dL (≥88.4 mmol/L) in women, mostly equivalent to an eGFR <60 mL/min/1.73 m2. In light of more recent work [50], the ERBP work group agrees with KDIGO that this threshold could be lowered to 45 mL/min/1.73 m2. The risk of CIN also increased in the presence of diabetes, and dehydration. The risk may be lower when simple i.v. contrast is administered for imaging versus when contrast is used during an invasive intra-arterial procedure, where the risk of cholesterol embolization should also be taken into account [59]. It is unclear whether simple intra-arterial injection, e.g. digital subtraction angiography has a different risk from i.v. [60, 61]. The risk increases with the volume of contrast applied. There are no data available to know if the effect of repeated contrast administration is simply a consequence of the cumulative dosage of iodine, or whether repeated administrations are disproportionately more toxic than the administration of a certain volume of contrast in one shot. Another risk factor is the use of concurrent nephrotoxic medication: non-steroidal anti-inflammatory drugs, aminoglycosides, amphotericin B, high doses of loop diuretics and antiviral drugs like acyclovir and foscarnet, in particular. A special mention should be made on metformine, as accumulation of this drug in CIN can lead to dangerous situations. The ERBP group wants to point out that several drugs have a prolonged nephrotoxic action as a consequence of a long-lasting cellular accumulation in the kidney. In order to minimize the risk of kidney damage, these drugs would have to be stopped for days or even weeks, and not only hours, before contrast administration. The rationale for stopping loop diuretics is mainly based on their detrimental effect if used as pharmacological prevention against CIN [62]. Not only must loop diuretics be discontinued during and after contrast administration, but they should be stopped for as long as possible before the procedure in order to reduce the possibility of volume depletion. From this point of view, it is surprising to note that the possible detrimental effect of thiazide diuretics, which have a much longer action period, is almost never mentioned. It should be stressed that dehydration or any degree of volume depletion make medullary renal perfusion closely dependent of vasoactive hormones, and extremely sensitive to microvascular effects of intravascular contrast administration [63]. Apart from diuretics, clinical circumstances such as gastro-intestinal fluid losses may induce dehydration, and if possible it is wise to delay contrast administration until volume status has been corrected. To date, there is very little evidence on the detrimental effects of angiotensin-converting enzyme-inhibitor (ACE-I) concerning the renal risk of contrast administration. A randomized study showed a decreased incidence of CIN following the administration of captopril in diabetic patients undergoing coronary angiography [64], and more recently, it was observed that a captopril treatment stopped 36h before CM administration was neither associated with nor increased the risk of CIN in hydrated patients [65]. However, the risk associated with long-acting ACE-I and ARB is poorly defined and should be assessed through specific studies. Pharmacological prevention strategies of CIN 3.4.1 We recommend volume expansion with either isotonic sodium chloride or sodium bicarbonate solutions, rather than no volume expansion, in patients at increased risk for CIN. (1A) 3.4.2 We suggest using the oral route for hydration, on the premise that adequate intake of fluid and salt are assured. (2C) We suggest that, when oral intake of fluid and salt is deemed cumbersome in patients at increased risk of CIN, hydration should be performed by intravenous route. (2C) 3.4.3 We suggest using oral N-acetyl cysteine (NAC) only in patients who receive appropriate fluid and salt loading (2D). We recommend not using oral NAC as the only method for prevention of CIN. (1D) 3.4.4 We do not suggest using theophylline to prevent CIN. (2C) 3.4.5 We do not recommend using fenoldopam to prevent CIN. (1B) Rationale There is no doubt that before contrast media administration, adequate salt and fluid should be provided to prevent CIN The ERBP work group amended the statement on oral fluid loading by the KDIGO work group, as this was based on two small and relatively old studies, in which oral fluid intake did not confer the same degree of protection against CIN than i.v. fluid administration [66, 67]. However, a recent observational study showed a significant inverse correlation between the amount of oral fluid intake and the percentage changes in SCr as well as the absolute changes in eGFR in patients undergoing a coronary computed tomography angiography [68], and a prospective randomized trial comparing i.v. fluids with oral hydration with or without sodium bicarbonate found no differences in the incidence of CIN in patients with mild CKD [69]. It should be noted that the main difference between oral and i.v. fluid administration concerns not only the volume but the sodium content of the fluids as well [70]. In ambulatory patients, the i.v. route leads to a substantial increase in costs, and a risk for destruction of future vascular access. The ERBP work group accordingly does not recommend hospitalizing low-risk patients just for hydration. Most of the ambulatory patients have a relatively low risk for CIN, and in these patients, oral hydration should be recommended. When i.v. access is in place anyway, e.g. in hospitalized patients, the i.v. route can be used. NAC has a number of beneficial properties, including anti-oxidant functions and mediation of renal vasodilation, making it a suitable candidate to help prevent CIN. However, NAC has been the subject of a series of comprehensive reviews, and overall there appears to be insufficient evidence to support the universal use of NAC to prevent CIN despite its ease of administration [63]. It should be noted that in most trials reporting a benefit, NAC administration was associated with i.v. hydration using bicarbonate. Studies of NAC with bicarbonate administration have found a moderate benefit for this combination, compared with the combination of NAC–saline, and it is unclear in how far the benefit can be attributed to NAC per se. To date, 7 out of the 11 meta-analyses that have been published on this subject found a net benefit for NAC in the prevention of CIN [71]. NAC, however, has been reported to decrease SCr levels in normal volunteers with normal kidney function. This reduction in SCr was not accompanied by a change in serum cystatin C levels, suggesting an effect independent of a change in GFR, such as an increase in tubular secretion of creatinine or a decrease in creatinine production [72]. In conclusion, in view of its low costs and the high likelihood of absence of harm, there is no objection against oral NAC administration, but this should never replace adequate fluid loading. Effects of haemodialysis or haemofiltration 4.5.1: We do not recommend using prophylactic intermittent haemodialysis (IHD) or haemofiltration (HF) for the purpose of prevention of CIN only. (1C) Rationale The evidence collected by KDIGO demonstrates that IHD to prevent CIN in well pre-hydrated patients at risk is not effective, and that there is even a trend to more harm (more CIN, and more need for RRT) [73–75]. High-volume HF in this setting has been reported to be beneficial [76, 77]. The protocol used in these studies included HF at ICU, and with high volumes of bicarbonate fluid. It seems likely that under these conditions, the beneficial effects observed were due to volume expansion and loading with bicarbonate rather than to the removal of contrast media by the HF. In view of the high costs and logistical problems, the evidence seems too weak to recommend prophylactic HF at this moment.
                Bookmark

                Author and article information

                Contributors
                henryhu200211@163.com
                meiliu@kumc.edu
                Journal
                Sci Rep
                Sci Rep
                Scientific Reports
                Nature Publishing Group UK (London )
                2045-2322
                23 November 2018
                23 November 2018
                2018
                : 8
                : 17298
                Affiliations
                [1 ]ISNI 0000 0004 1790 3548, GRID grid.258164.c, Big Data Decision Institute (BDDI), , Jinan University, ; Guangzhou, 510632 China
                [2 ]Guangdong Engineering Technology Research Center for Big Data Precision Healthcare, Guangzhou, 510632 China
                [3 ]ISNI 0000 0001 2177 6375, GRID grid.412016.0, Division of Nephrology and Hypertension and the Kidney Institute, , University of Kansas Medical Center, ; Kansas City, 66160 USA
                [4 ]ISNI 0000 0004 1936 9000, GRID grid.21925.3d, Center for Critical Care Nephrology, Department of Critical Care Medicine, , University of Pittsburgh School of Medicine, ; Pittsburgh, 15260 USA
                [5 ]ISNI 0000 0001 2177 6375, GRID grid.412016.0, Department of Internal Medicine, Division of Medical Informatics, , University of Kansas Medical Center, ; Kansas City, 66160 USA
                Author information
                http://orcid.org/0000-0003-3752-0045
                http://orcid.org/0000-0002-1776-2533
                Article
                35487
                10.1038/s41598-018-35487-0
                6251919
                30470779
                57d901f4-8235-4246-8b4c-f1b576e9e044
                © The Author(s) 2018

                Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

                History
                : 5 July 2018
                : 2 November 2018
                Funding
                Funded by: FundRef https://doi.org/10.13039/501100001809, National Natural Science Foundation of China (National Science Foundation of China);
                Award ID: 91746204
                Award Recipient :
                Funded by: FundRef https://doi.org/10.13039/501100007162, Guangdong Science and Technology Department (Science and Technology Department, Guangdong Province);
                Award ID: 2017B030308008
                Award Recipient :
                Funded by: Guangdong Engineering Technology Research Center for Big Data Precision Healthcare (Grant No.603141789047) Fundamental Research Funds for the Central Universities (Grant No.21618315)
                Categories
                Article
                Custom metadata
                © The Author(s) 2018

                Uncategorized
                Uncategorized

                Comments

                Comment on this article