Continuous and early prediction of future moderate and severe Acute Kidney Injury in critically ill patients: Development and multi-centric, multi-national external validation of a machine-learning model.

BACKGROUND: Acute Kidney Injury (AKI) is a major complication in patients admitted to Intensive Care Units (ICU), causing both clinical and economic burden on the healthcare system. This study develops a novel machine-learning (ML) model to predict, with several hours in advance, the AKI episodes of stage 2 and 3 (according to KDIGO definition) acquired in ICU. METHODS: A total of 16'760 ICU adult patients from 145 different ICU centers and 3 different countries (US, Netherland, Italy) are retrospectively enrolled for the study. Every hour the model continuously analyzes the routinely-collected clinical data to generate a new probability of developing AKI stage 2 and 3, according to KDIGO definition, during the ICU stay. RESULTS: The predictive model obtains an auROC of 0.884 for AKI (stage 2/3 KDIGO) prediction, when evaluated on the internal test set composed by 1'749 ICU stays from US and EU centers. When externally tested on a multi-centric US dataset of 6'985 ICU stays and multi-centric Italian dataset of 1'025 ICU stays, the model achieves an auROC of 0.877 and of 0.911, respectively. In all datasets, the time between model prediction and AKI (stage 2/3 KDIGO) onset is at least of 14 hours after the first day of ICU hospitalization. CONCLUSIONS: In this study, a novel ML model for continuous and early AKI (stage 2/3 KDIGO) prediction is successfully developed, leveraging only routinely-available data. It continuously predicts AKI episodes during ICU stay, at least 14 hours in advance when the AKI episode happens after the first 24 hours of ICU admission. Its performances are validated in an extensive, multi-national and multi-centric cohort of ICU adult patients. This ML model overcomes the main limitations of currently available predictive models. The benefits of its real-world implementation enable an early proactive clinical management and the prevention of AKI episodes in ICU patients. Furthermore, the software could be directly integrated with IT system of the ICU.

External validation of a deep-learning model to predict severe acute kidney injury based on urine output changes in critically ill patients.

OBJECTIVES: The purpose of this study was to externally validate algorithms (previously developed and trained in two United States populations) aimed at early detection of severe oliguric AKI (stage 2/3 KDIGO) in intensive care units patients. METHODS: The independent cohort was composed of 10'596 patients from the university hospital ICU of Amsterdam (the "AmsterdamUMC database") admitted to their intensive care units. In this cohort, we analysed the accuracy of algorithms based on logistic regression and deep learning methods. The accuracy of investigated algorithms had previously been tested with electronic intensive care unit (eICU) and MIMIC-III patients. RESULTS: The deep learning model had an area under the ROC curve (AUC) of 0,907 (± 0,007SE) with a sensitivity and specificity of 80% and 89%, respectively, for identifying oliguric AKI episodes. Logistic regression models had an AUC of 0,877 (± 0,005SE) with a sensitivity and specificity of 80% and 81%, respectively. These results were comparable to those obtained in the two US populations upon which the algorithms were previously developed and trained. CONCLUSION: External validation on the European sample confirmed the accuracy of the algorithms, previously investigated in the US population. The models show high accuracy in both the European and the American databases even though the two cohorts differ in a range of demographic and clinical characteristics, further underlining the validity and the generalizability of the two analytical approaches.

A deep-learning model to continuously predict severe acute kidney injury based on urine output changes in critically ill patients.

BACKGROUND: Acute Kidney Injury (AKI), a frequent complication of pateints in the Intensive Care Unit (ICU), is associated with a high mortality rate. Early prediction of AKI is essential in order to trigger the use of preventive care actions. METHODS: The aim of this study was to ascertain the accuracy of two mathematical analysis models in obtaining a predictive score for AKI development. A deep learning model based on a urine output trends was compared with a logistic regression analysis for AKI prediction in stages 2 and 3 (defined as the simultaneous increase of serum creatinine and decrease of urine output, according to  the Acute Kidney Injury Network (AKIN) guidelines). Two retrospective datasets including 35,573 ICU patients were analyzed. Urine output data were used to train and test the logistic regression and the deep learning model. RESULTS: The deep learning model defined an area under the curve (AUC) of 0.89 (± 0.01), sensitivity = 0.8 and specificity = 0.84, which was higher than the logistic regression analysis. The deep learning model was able to predict 88% of AKI cases more than 12 h before their onset: for every 6 patients identified as being at risk of AKI by the deep learning model, 5 experienced the event. On the contrary, for every 12 patients not considered to be at risk by the model, 2 developed AKI. CONCLUSION: In conclusion, by using urine output trends, deep learning analysis was able to predict AKI episodes more than 12 h in advance, and with a higher accuracy than the classical urine output thresholds. We suggest that this algorithm could be integrated in the ICU setting to better manage, and potentially prevent, AKI episodes.

[Acute Kidney Injury: the nephrology plus value and competence and a good organization can ameliorate the prognosis]. / Insufficienza Renale Acuta: il plus valore e la competenza del nefrologo, unitamente ad una buona organizzazione, possono migliorare la prognosi.

Epidemiology of Acute Kidney Injury (AKI) has changed radically in the past 15 years: we have observed an exponential increase of cases with high mortality and residual disability, particularly in those patients who need dialysis treatment. Those who survive AKI have an increased risk of requiring dialysis after hospital discharge over the short term as well as long term. They have an increased risk of deteriorating residual kidney function and cardiovascular events as well as a shorter life expectancy. Given the severe prognosis, difficulties of treatment, high level of resources needed, increased workload and consequently costs, several aspects of AKI have not been sufficiently investigated. Any national register of AKI has not been developed and its absence has an impact on provisional strategies. Specific training should be planned beginning with University, which should include practical training in Intensive Care Units. A definition of the organizational characteristics and requirements for the care of AKI is needed. Treatment of AKI is not based exclusively on dialysis efficiency or technology, but also on professional skills, volume of activity, clinical experience, model of healthcare organizations, continuity of processes and medical activities to guarantee such as a closed-staff system. Progress in knowledge and technology has only partially modified the outcome and prognosis of AKI patients; consequently, new strategies based on increased awareness, on the implementation of professional skills, and on revision, definition and updating of resources for the organization of AKI management are needed and expected over the short term.

Outcome in noncritically ill patients with acute kidney injury requiring dialysis: Effects of differing medical staffs and organizations.

Acute kidney injury requiring dialysis (AKI-D) treatment has significantly increased in incidence over the years, with more than 400 new cases per million population/y, 2/3 of which concern noncritically ill patients. In these patients, there are little data on mortality or on information of care organization and its impact on outcome. Specialty training and integrated teams, as well as a high volume of activity, seem to be linked to better hospital outcome. The study investigates mortality of patients admitted to and in-care of nephrology (NEPHROpts), a closed-staff organization, and to other medical wards (MEDpts), representing a model of open-staff organization.This is a single center, case-control cohort study derived from a prospective epidemiology investigation on patients with AKI-D admitted to or in-care of the Hospital of Perugia during the period 2007 to 2014. Noncritically ill AKI-D patients were analyzed: inclusion and exclusion criteria were defined to avoid possible bias on the cause of hospital admittance and comorbidities, and a propensity score (PS) matching was performed.Six hundred fifty-four noncritically ill patients were observed and 296 fulfilled inclusion/exclusion criteria. PS matching resulted in 2 groups: 100 NEPHROpts and 100 MEDpts. Characteristics, comorbidities, acute kidney injury causes, risk-injury-failure acute kidney injury criteria, and simplified acute physiology score (SAPS 2) were similar. Mortality was 36%, and a difference was reported between NEPHROpts and MEDpts (20% vs 52%, χ = 23.2, P < 0.001). Patients who died differed in age, serum creatinine, blood urea nitrogen/s.Creatinine ratio, dialysis urea reduction rate (URR), SAPS 2 and Charlson score; they presented a higher rate of heart disease, and a larger proportion required noradrenaline/dopamine for shock. After correction for mortality risk factors, multivariate Cox analysis revealed that site of treatment (medical vs nephrology wards) represents an independent risk factor of mortality (relative risk = 2.13, 95% confidence interval = 1.25, 3.63; P < 0.01). Other independent risk factors were age, URR, s.Creatinine at hemodialysis beginning, and SAPS 2 score.In our context, we have documented that noncritically ill AKI-D patients, who represented 2/3 of the population, had high in-hospital mortality (36%), and that a closed-staff specialty medical organization, such as a Nephrology team, seems to guarantee a better outcome than general medical organizations. The significance in healthcare system organization and resource allocation could be important.

Six-year single-center survey on AKI requiring renal replacement therapy: epidemiology and health care organization aspects.

Evidence regarding hospital-based acute kidney injury (AKI) reveals a continuous increase in incidence over the years, at least in intensive care units (ICU). Fewer reports are available for non critically-ill patients admitted to general or specialist wards other than ICU (non-ICU). The consequence of greater incidence is an increase in therapies such as dialysis; but how the health care organization deals with this problem is not clearly known. Here we quantified the incidence of dialysis-requiring AKI (AKI-D) among patients admitted to a University Hospital which serves a population of 354,000 inhabitants. Between 2007 and 2012, the incidence of AKI-D increased from 209 to 410 per million population (pmp)/year; age of patients and cardiovascular comorbid pathologies also increased. AKI-D was more frequent in non-ICU and 32% of patients were admitted to ICU. Considering the site of treatment of non-ICU patients, in 2007 the ratio of patients admitted to non-ICU wards apart from Nephrology to those admitted to Nephrology was 1:1, but in 2012 the ratio increased to 2.4:1 (p < 0.05). The complexity of acute disease, measured with the New Simplified Acute Physiology Score (SAPS II), did not reveal differences over the years. The number of dialysis treatments/year increased by 82%, and the total hours/year increased by 86%. Low-efficiency daily dialysis was performed in 52.4% of patients admitted to ICU, but dialysis sessions longer than 8 h were performed in only 40% of cases. Overall, 6-year mortality was 48.8%, without significant differences over the years. Mortality in ICU was 65.6%, and in non-ICU 41.2% (p < 0.001). Dialysis treatments needed to be continued after hospital discharge in 21% of patients. We conclude that dialysis-requiring AKI is becoming more common, and that two-thirds of patients are admitted as non-ICU: in these patients, during the last year of the study, the treatment site was more frequently in non-ICUs other than Nephrology. Over the 6-year period, the local healthcare organization had to dispense 80% more dialysis treatments/year in terms of total number and hours of treatment. One-fifth of surviving patients needed to continue dialysis after hospital discharge. Our data highlight the public health importance of AKI and the need for adequate resources for Nephrology.

Changes in the perceived epidemiology of primary hyperaldosteronism.

Primary aldosteronism has been considered a rare disease in the past years, affecting 1% of the hypertensive population. Subsequently, growing evidence of its higher prevalence is present in literature, although the estimates of disease range from 5 up to 20%, as in type 2 diabetes and resistant hypertension. The main reasons for these variations are associated with the selection of patients and diagnostic procedures. If we consider that hypertension is present in about 20% of the adult population, primary aldosteronism can no longer be considered a rare disease. Patients with primary aldosteronism have a high incidence of cardiovascular, cerebrovascular and kidney complications. The identification of these patients has therefore a practical value on therapy, and to control morbidities derived from vascular damage. The ability to identify the prevalence of a disease depends on the number of subjects studied and the methods of investigation. Epidemiological studies are affected by these two problems: there is not consensus on patients who need to be investigated, although testing is recommended in subjects with resistant hypertension and diabetes. The question of how to determine aldosterone and renin levels is open, particularly if pharmacological wash-out is difficult to perform because of inadequate blood pressure control.

Blood pressure assessment in haemodialysis patients: comparison between pre-dialysis blood pressure and ambulatory blood pressure measurement.

AIM: Hypertension is common in haemodialysis (HD) patients. Determining the most appropriate method of blood pressure (BP) measurement, representative of target organ damage, is still an issue. BP variations between pre- and post-HD treatment, or between on-dialysis day and off-dialysis day, are common. The aim of this study was to examine the possible differences between pre-HD office BP (OBP) levels, inter-HD (iHD) or HDday 24 h ambulatory BP measurement (ABPM) with 48 h ABPM, where the latter was considered the gold standard. METHODS: 163 HD patients were studied. BP was monitored consecutively for 48 h with a Takeda TM2421 device, then sub-analysed into two periods of 24 h: HD and iHD day. An average of 12 sessions pre-HD OBP measurements was determined. RESULTS: OBP significantly overestimates systolic (SBP) and diastolic BP (DBP) when compared with 48 hABPM. SBP and DBP are significantly higher on iHD day than on HD day: 141.2 1 20.8 versus 137.9 1 20.9, and 77.1 1 11.1 versus 76.1 1 10.9 (P < 0.01). No differences of SBP night/day ratio were reported between 48 hABPM and iHD 24 h ABPM or HD 24 h ABPM. The highest correlations were reported between 48 h SBP/DBP with iHD or HD 24 h ABPM (r 2 = 0.95, P < 0.001), while the lowest between 48 h SBP/DBP and OBP (r 2 = 0.40,P < 0.01, r 2 = 0.12, P < 0.01). The narrowest limits of agreement using the Bland and Altman test were reported between 48 h SBP or DBP and 24 h iHD or HD day ABPM. Considering 48 h ABPM, 80.5% of patients had BP higher than the norm, compared with 61.7% of patients in the case of OBP (c2 = 13.28, P < 0.001). The sensibility for detecting hypertension for iHD day 24 h ABPM was 98.4%, with specificity of 90%. The sensibility of 24 h HDday ABPM was 90.3%, with specificity 96.6%. In the case of OBP, sensibility and specificity were considerably lower, that is, 72.6% and 83.3% respectively. CONCLUSION: Significant differences are shown between OBP and 48 h ABPM in the recognition of a hypertensive state. OBP measurement has a lower sensibility and specificity than 24 h ABPM, which remains a valid alternative approach to 48 h ABPM in HD patients. Errors of OBP estimation should be taken into account, with possible negative impact on treatment strategies and epidemiology studies

Prognostic value of 24-hour ambulatory blood pressure monitoring and of night/day ratio in nondiabetic, cardiovascular events-free hemodialysis patients.

BACKGROUND: The use of 24-hour ambulatory blood pressure monitoring is increasing in end-stage renal disease (ESRD) patients but the prediction power for cardiovascular complications of time-averaged ambulatory blood pressure components has been little investigated in these patients. METHODS: We analyzed the prognostic power of 24-hour ambulatory blood pressure monitoring for all-cause and cardiovascular mortality in 168 nondiabetic, events-free hemodialysis patients selected from a total dialysis population of about 450 patients. RESULTS: During the follow-up period (38 +/- 22 months), 48 patients died, 29 of them of cardiovascular causes. On univariate Cox regression analyses, the night/day systolic ratio resulted to be the sole blood pressure indicator to be associated with all-cause and cardiovascular mortality while left ventricular hypertrophy (LVH) was a strong predictor of these outcomes. In multivariable Cox models not including LVH, the night/day systolic ratio maintained an independent prognostic value for incident outcomes. However, when both risk factors, LVH and night/day systolic ratio, were introduced into Cox models, LVH was no longer a significant predictor while the night/day systolic ratio became a predictor of marginal statistical significance. CONCLUSION: The night/day ratio emerges as the sole ambulatory blood pressure monitoring-derived indicator providing significant prognostic information in patients with ESRD. However, this indicator as well as LVH loses substantial prediction power in statistical models including both risk factors. The results suggest that the night/day systolic ratio and LVH provide overlapping prognostic information, a phenomenon in keeping with the hypothesis that they represent a common pathway leading to adverse outcomes in ESRD.

Association between brain natriuretic peptide and extracellular water in hemodialysis patients.

BACKGROUND: Brain natriuretic peptide (BNP) is a hormone released by the left ventricle (LV) as a consequence of pressure or volume load. BNP increases in left ventricle hypertrophy (LVH), LV dysfunction, and it can also predict cardiovascular mortality in the general population as well as those undergoing hemodialysis (HD). We investigated the association between BNP and volume load in HD patients. METHODS: We studied 32 HD patients (60 +/- 17.1 years) treated thrice-weekly for at least 6 months. Exclusion criteria were: LV dysfunction, atrial fibrillation, malnutrition. Blood chemistries and BNP were determined on mid-week HD day. Blood pressure (BP) and cardiac diameters were determined on mid-week inter-HD day by using 24-hour ambulatory blood pressure monitoring and echocardiography. Bioimpedance was performed after HD and extracellular water (ECW%), calculated as a percentage of total body water, was considered as the index of volume load. RESULTS: Patients were divided into quartiles of 8 patients depending on the BNP value: 1st qtl BNP < or =45.5 pg/ml (28.4 +/- 10.9 pg/ml), 2nd qtl BNP > 45.5 pg/ml and < or =99.1 pg/ml (60.9 +/- 15.8 pg/ml), 3rd qtl BNP > 99.1 pg/ml and < or =231.8 pg/ml (160.5 +/- 51.8 pg/ml), 4th qtl BNP > 231.8 pg/ml (664.8 +/- 576.6 pg/ml). No inter-quartile differences were reported in age, HD age, body mass index spKt/V, or blood chemistries. As expected patients in the 4th BNP quartile showed the highest values of 24-hour pulse pressure (PP) and LV mass index (LVMi). The study of body composition revealed significant differences in ECW%, which was higher in the 4th quartile when compared to the others (4th q: 50 +/- 9.6%, vs 1st q. 40.1 +/- 2.4%, 2nd q. 41.9 +/- 5%, 3rd q. 42.8 +/- 6.9%). Using multiple stepwise linear regression where BNP was the dependent variable, and PP and ECW% the independent variables, only ECW% maintained statistical significance as a predictor of BNP levels (PP: Beta = 0.86, p = 0.58; ECW%: Beta = 0.64, p < 0.001 p < 0.001). CONCLUSIONS: Few studies have investigated the relationship between plasma BNP and volume load, and direct evidence is lacking. We used bioimpedance and the determination of ECW% to assess volume state in HD patients finding an association between BNP and ECW. The increased synthesis and release of BNP from the LV in HD patients appear to be mainly related to volume stress rather than to pressure load.

Association between extracellular water, left ventricular mass and hypertension in haemodialysis patients.

BACKGROUND: Hypertension and left ventricular hypertrophy (LVH) are present in the majority of patients undergoing haemodialysis (HD). These two pathologies persist after dialysis onset, and pharmacological therapy is often required for adequate control of blood pressure (BP). Although fluid overload is a determinant of hypertension, clinical assessment of this parameter remains difficult and unsatisfactory. Bioimpedance analysis (BIA) spectroscopy and the relative determination of extracellular water (ECW%) may provide a simple and inexpensive tool for investigating fluid overload. We studied 110 patients on thrice-weekly HD to determine whether ECW body content correlates with hypertension and LVH in this patient population. METHODS: Hypertension was determined according to the WHO criteria (office BP >/= 140/90 and/or the use of antihypertensive therapy). Twenty-four hour BP monitoring and echocardiography were performed on midweek inter-HD days. Blood chemistries, dialysis dose (spKt/V) and bioimpedance were analysed on midweek HD days. RESULTS: Hypertension was present in 74.5% of patients. There were no differences for age, spKt/V, haemoglobin, serum creatinine and residual renal function between normotensive and hypertensive patients. Twenty-four hour systolic BP (SBP), 24 h diastolic BP and 24 h pulse pressure were higher in hypertensive patients, in spite of antihypertensive therapy. LVH was present in 61.8% of patients. BIA revealed that ECW% was increased in LVH+ patients (LVH+ = 47.5 +/- 7.9%, LVH- = 42.4 +/- 6.2%, P = 0.01) and in hypertensive patients compared with normotensives (46.5 +/- 7.7% vs 43 +/- 7.2%, P = 0.02). Dry body weights and inter-HD body weight increases did not differ between hypertensive and normotensive patients nor between patients with or without LVH. ECW was correlated with SBP (r = 0.35, P < 0.01) and with left ventricular mass index (LVMi(g/sqm)) (r = 0.49, P < 0.001). A stepwise multiple linear regression model revealed that LVMi(g/sqm) was significantly correlated with ECW%, SBP and male gender (r = 0.65, P < 0.001). CONCLUSIONS: LVH and hypertension are present in a majority of HD patients and they are closely correlated with one another. We found associations between fluid load, measured by BIA and expressed as ECW, and BP and LVM.

Behavior of non-protein-bound and protein-bound uremic solutes during daily hemodialysis.

BACKGROUND: In the last few years, renewed interest in daily short hemodialysis (DHD; six 2-hour sessions per week) has become apparent as a consequence of the better clinical outcome of patients treated by this schedule. Uremic syndrome is characterized by the retention of a large number of toxins with different molecular masses and chemical properties. Some toxins are water soluble and non-protein bound, whereas others are partially lipophilic and protein bound. There is increased evidence that protein-bound toxins are responsible for the biochemical and functional alterations present in uremic syndrome, and the kinetics of urea is not applicable to these substances for their removal. The aim of this study is to investigate whether DHD is accompanied by increased removal of non-protein-bound and protein-bound toxins and a decrease in their prehemodialysis (pre-HD) serum levels. PATIENTS AND METHODS: We studied 14 patients with end-stage renal disease treated by standard HD (SHD; three 4-hour sessions per week) for at least 6 months and randomly assigned them to a two-period crossover study (SHD to DHD and DHD to SHD). Patients maintained the same dialyzer, dialysate, and Kt/V during the entire study. At the end of 6 months of SHD and 6 months of DHD, we evaluated hemoglobin levels, hematocrits, recombinant human erythropoietin doses, and pre-HD and post-HD concentrations of serum urea, creatinine, uric acid, and the following protein-bound toxins: 3-carboxy-4-methyl-5-propyl-2-furanpropionic acid, p-cresol, indole-3-acetic acid, indoxyl sulfate, and hippuric acid. RESULTS: Values for hemoglobin, hematocrit, and recombinant human erythropoietin dose did not change during the two study periods. Pre-HD concentrations of creatinine, urea, and uric acid decreased on DHD (creatinine, from 8.7 +/- 1.9 to 7.8 +/- 1.6 mg/dL; P < 0.05; urea, from 149.4 +/- 28.8 to 132.7 +/- 40 mg/dL; P = 0.05; uric acid, from 9.14 +/- 1.49 to 8.16 +/- 1.98 mg/dL; P = 0.06). Concerning protein-bound toxins, lower pre-HD levels during DHD were reported for indole-3-acetic acid (SHD, 0.16 +/- 0.04 mg/dL; DHD, 0.13 +/- 0.03 mg/dL; P = 0.01), indoxyl sulfate (SHD, 3.35 +/- 1.68 mg/dL; DHD, 2.85 +/- 1.08 mg/dL; P = 0.02), and p-cresol at the borderline of significance (SHD, 0.96 +/- 0.59 mg/dL; DHD, 0.78 +/- 0.33 mg/dL; P = 0.07). CONCLUSION: Such non-protein-bound compounds as uric acid, creatinine, and urea were removed significantly better by DHD, and pre-HD serum levels were reduced. Furthermore, pre-HD concentrations of some protein-bound solutes, such as indole-3-acetic acid, indoxyl sulfate, and p-cresol, also were lower during DHD.

Daily haemodialysis improves indices of protein glycation.

BACKGROUND: Advanced glycation end-products (AGEs) accumulate in uraemia, regardless of hyperglycaemic conditions, and may contribute to the onset of some long-term complications, such as atherosclerosis, amyloidosis, and neurodegenerative processes. In this study, we compare a daily with a standard 3 times/week dialysis rhythm (DHD and SHD, respectively) in correcting some protein glycation indices in end-stage renal disease (ESRD) patients. METHODS: Twenty-one normoglycaemic and 11 diabetic patients on chronic haemodialysis (HD) with low-flux dialysers were studied in a prospective protocol to compare two different dialysis schedules, namely: 4 h, 3 times/week (SHD) and 2 h, 6 times/week (DHD). The patients were studied before and after 6 months of DHD. To further check the effect of DHD on glycation parameters, 4 normoglycaemic HD patients were studied in a third step in which they returned for 3 months to the SHD rhythm. Also, 11 chronic renal failure (CRF) patients not yet on HD and 11 age- and sex-matched healthy controls were studied. A new HPLC method was used to measure the following glycation indexes on plasma: the early product furosine and the advanced products protein-bound and free pentosidine, and two heterogeneous classes of low molecular mass (LMM) AGE peptides. RESULTS: All the parameters studied showed an accumulation that worsened with the progression of renal failure (controls