Журнал «Медицина неотложных состояний» 2 (65) 2015

The aim of study: to assess the impact on the outcome and treatment of acute kidney injury (AKI) some factors that could potentially have an impact on disease outcomes.

Material and Methods. The results of treatment were examined in 106 patients. 14 of them had moderate AKI, 16 — severe and 76 patients recorded extremely severe AKI (oliguric stage more than three weeks). Was tasked with identifying factor variables associated with the results of treatment in patients with AKI; assess the extent and direction of the effect of these factor variables on the results obtained. For analysis methods were used constructing multivariate classification models. Quality constructed models to evaluate their sensitivity and specificity, calculated 95 % confidence interval (95 % CI) indicators. To identify the factors that are most associated with the results of treatment method was used "genetic algorithm" (GA) selection. To assess the adequacy of mathematical models and multivariate tests predicting the effectiveness of treatment used performance area under ROC-curve (Area Under Curve — AUC) of 95 % CI. To assess the possibility of using models built on the new data used method of checking their prognostic characteristics in the confirmation set (used to select a random number generator). To evaluate the degree of influence on the resulting factorial signs a method of constructing logistic regression models was used. There were calculated odds ratio (OR) and their 95 % CI.

As the factor variables in constructing mathematical models of 18 indicators were analyzed: age, sex, AKI form ( prerenal — 1, renal-2 subrenal-3), duration of illness before admission to the ICU, duration of oliguria in days before admission to ICU, the presence of multiple organ disorders, character of multiple organ disorders (primary, secondary), presence of AKI risk factors (no = 0, there is a = 1), nephroprotection before ICU admission (0-not carried out, 1 — carried out); nephroprotection in ICU (1-early 0 — late), the need for dialysis (without HD = 0, HD = 1), the amount of hemodialysis, the need for mechanical ventilation, the level of urea on admission to the ICU, serum creatinine at admission to the ICU, The highest level of urea, the highest level of creatinine. The results of treatment based on the outcome: survived, died; renal function recovered, loss. Also assessed the duration of oliguric stage (indicates the severity of AKI), the need for renal replacement therapy (hemodialysis) and the number of the procedures.

Results. One of the main goals of successful treatment AKI patient is to reduce the duration of oliguric stage. The faster restoration of diuresis leads the less risk of severe multiple organ disorders and formation of irreversible changes in the kidneys. Long oliguria determines the severity of AKI . There was assessed the impact of various factors on the duration of oliguric stage. When analyzing the risk of prolonged oligoanuria (considered duration for more than 5 days) in the first stage of the study was to construct a linear neural network forecasting model on the same factor 16 signs. Sensitivity model training set made up 90.0 % (95 % CI 78.5 % — 97.4 %), specificity — 91.7 % (95 % CI 80.1 % — 98.5 %). In the confirmation set sensitivity of the model accounted for 71.4 % (95 % CI 43.3 % — 92.6 %), specificity — 93.8 % (95 % CI 75.4 % — 100 %). To identify the factors that are most associated with the risk of a long oligoanuria was conducted selection of significant attributes using the GA. As a result of analysis were selected 3 factorial trait: AKI form (1 prerenal, renal 2 subrenal 3) (X1), nephroprotection before ICU admission (0-not carried out, 1 — conducted) (X2), nephroprotection in ICU (1-early, 0 — late) (X3). On a dedicated set of factor variables was constructed linear neural network model predicting the risk of a long oligoanuria. The sensitivity of this model in the training set made ​​up 85.0 % (95 % CI 72.0 % — 94.5 %), a specificity of — 77.8 % (95 % CI 62.4 % — 90.0 %). In the confirmation set the sensitivity of the model was 64.3 % (95 % CI 35.8 % — 100 %), specificity — 81.3 % (95 % CI 57.1 % — 96.7 %). To assess the significance of 3 selected from a total of 16-factor variables and evaluate the adequacy of the constructed models predicting the risk of a long oligoanuria during the treatment of patients with AKI, we used the method of comparison ROC-curve models. In the analysis, found that the area under the ROC-curve for the linear neural network model based on all 16 signs factor was AUC1 = 0,94 ± 0,02, for a linear neural network model based on a 3-factorial sign isolated — AUC2 = 0 , 84 ± 0,04. Thus, reducing the number of factor variables from 16 and up to 3 does not lead to a significant change, indicating the high importance of the selected factor variables (AKI variant 1- prerenal, 2-renal ,3- subrenal ) (X1), nephroprotection before ICU admission (0-not carried out, 1 — conducted) (X2), nephroprotection in the ICU (1-early 0 — late) (X3) to predict the risk of a long oligoanuria. From the analysis of the coefficients of the logistic regression model implies that for renal variant AKI (2) the risk of prolonged oligoanuria (p = 0,001) increased, OR = 8.7 (95 % CI 2.3 — 32.9), compared with prerenal (1). Found that the risk of a long oligoanuria statistically significant (p = 0,032) reduced in case of nephroprotection before admission to the ICU, OR = 0.3 (95 % CI 0.1 — 0.9). Also showed a reduction (p = 0,014) risk of prolonged oligoanuria if early nephroprotection in ICU was done, OR = 0.06 (95 % CI 0.01 — 0.56).

Analysis of the risk of death was the same as in risk of prolonged oliguria. The main factors, associated with risk of death were: the presence and character of multiple organ disorders (MOD- primary — 1, secondary — 2, No-0) (X1), nephroprotection in the ICU (1-early; 0 — late) (X2). Sensitivity of this model was 83.3 % (95 % CI 34.3 % — 100 %), specificity — 100 % (95 % CI 92.3 % — 100 %). Found, that the risk of death was statistically significantly (p = 0,001) reduced, OR = 0.009 (95 % CI 0.001 — 0.050) during early nephroprotection in ICU.

Conclusion. Individual nephroprotection (based on variant, ethiology of AKI, renal blood flow changes ) is one of the decisive factors in the implementation of effective intensive care and improving outcomes of AKI. Nephroprotection should be carried out at all stages of AKI, as the severity and prevalence of renal tubular damage is difficult to estimate reliably.