The present study showed that patients who suffered an AKI episode caused by presumed ATN, who were followed by nephrologists during hospitalization and who were discharged free from dialysis remains with a high mortality even after hospital discharge. This mortality was much higher than those of the population of São Paulo City, and also higher than those reported in other studies 
. One explanation could be that we included all deaths occurred after hospital discharge (median time to death was 5 months) and some studies excluded the deaths up to 90 days 
Since half of non-survivors died in the first 6 months after discharge, and neoplasm was a much higher cause of death than in those who died after this period, it is reasonable to suppose that the high early mortality observed occurred in the most severely diseased patients.
The present study highlights the importance of previous comorbidities in long-term mortality after AKI. It showed that patients classified as high risk by the Khan index, those with chronic liver failure, those who were admitted to non-surgical services and those who had a second episode of AKI had a very high risk of death after hospital discharge.
The role of previous comorbidities on long-term mortality was more clearly shown by the finding that the non-survivors died from diseases already present before the AKI episode and that their causes of death were similar to the population of São Paulo City in the same age range.
The Khan index was first described as an indicator of the risk of mortality for patients with end-stage renal disease (ESRD) and was later transposed to AKI patients 
. Ali et al
showed that the index was associated with 6-month mortality after AKI 
. In the present study, the patients classified as high risk by the Khan index had a poorer survival rate than those classified as low risk. There was a trend (P
0.08) suggesting that the patients classified as medium risk also had a poorer prognosis than the low risk patients.
In addition to its presence on the Khan index, the presence of chronic liver disease was an independent factor associated with long-term mortality. It has been reported that the occurrence of AKI is frequent in patients with chronic liver failure and that it increases morbidity and mortality 
. Even for patients who had received a liver transplant, a reduced GFR before the transplant was an independent predictor of mortality 30 days and 2 years after the transplantation 
. Deshpande et al
propose that the combination of AKI and liver failure produces a “toxic milieu” that directly causes endothelial dysfunction affecting multiple organ systems, which contributes to the increased short- and long-term mortality of these patients 
. Our findings emphasize the catastrophic role of the interaction between chronic liver disease and AKI.
Admission to non-surgical wards was an independent factor associated with long-term mortality that could not be explained by the presence of any of the variables analyzed. Other studies have also found a poorer survival among non-surgical admissions, and this finding has not yet been explained 
The impact that a second episode of AKI has on in-hospital mortality has not been studied. In the present study, only 8% of the patients had a second episode of AKI during the same hospitalization; however, those who did have a second episode were at very high risk for long-term mortality. More studies are needed to evaluate the importance of a new AKI episode both on in-hospital mortality and on long-term mortality.
Data from animal studies have already shown that AKI can induce damage to other organs, such as the lungs 
and heart 
. Thus, each AKI episode may aggravate pre-existing dysfunctions of various organs, accelerating their decline and leading to death.
AKI may also aggravate previous kidney disease, accelerate the progression to ESRD and worsen survival rates. Experimental studies have shown that AKI induces permanent kidney damage 
. However, human studies are controversial, mainly because many studies on long-term mortality exclude patients with CKD 
. Some studies have shown that in-hospital mortality is lower in patients with AKI superimposed on CKD (AoCKD) when compared with patients with “pure” AKI 
. However, Ali et al
showed that the 6-month mortality of patients with AoCKD was higher than those with AKI (62.5% and 49.8%, respectively, P
. Long-term mortality was not associated with pre-existing CKD in the present study. However, we must note that our percentage of patients with presumed CKD (51%) was much higher than those reported by Ali et al
(15%). This difference may be because in the present study, all patients were seen by nephrologists, while in the Ali et al
study, the nephrologists attended only 29% of the patients 
Another controversial point is whether AKI severity is a good predictor of long-term mortality, as it has been shown to be for in-hospital mortality 
. Usually AKI severity can be evaluated by AKIN or RIFLE criteria. As there is no widely accepted AKI definition for retrospective research using databases, we defined AKI considering the adapted AKIN definition 
. This definition uses the ratio between the higher SCr measured during hospitalization and the previous SCr to define and classify the AKI episode and has been used by other authors 
. In contrast to these studies, which have shown that this classification can also predict long-term mortality 
, in the present study, AKI severity was not a risk factor for long-term mortality. One factor may explain this difference; as mentioned earlier, we only included patients followed by nephrologists. Ponce et al
showed that AKI patients evaluated by nephrologists were more seriously ill than those not evaluated 
. This finding may explain why 49% of our patients were classified as AKI severity III, while other authors, who also evaluated patients not attended by nephrologists, reported that 13% of patients were classified as AKI severity III 
. Moreover, patient with pre-renal AKI were not evaluated in the present study, as those with less than two days of follow-up were excluded.
Some studies found that the absence of renal recovery was a risk factor for long-term mortality 
, but in the present study, renal recovery was not associated with long-term mortality. Similar to our findings, Loef et al
did not find a difference in mortality 96 months after the AKI episode between those who recovered and those who did not show renal function recovery 
. Conversely, Mehta et al
showed that, 13 years after cardiac surgery, the survival curve of the patients who did not show renal function recovery was poorer than that of those who had recovered (P
. However, in the present study patients who were discharged on dialysis and also those with AKI not due to presumed ATN were excluded. These exclusions might explain this difference. It must be noted that the definition of renal recovery varies among studies, and it may be more appropriate to use a SCr measured after hospital discharge to define renal recovery 
The present study has some limitations. The lack of a control group, similar to the studied population but without AKI, does not allow us to establish a causal relationship between AKI and long-term mortality. Although we attempted to control for selection bias with statistical methods using multiple variables, our retrospective observational analysis is subject to bias from unmeasured factors. However, it is very difficult to find a matched control group. Wald et al
, who studied 4066 AKI survivors, had to exclude 297 more seriously ill patients because they could not match these patients to an adequate control 
. In the present study, we compared the long-term survival of AKI patients with the population of São Paulo City at the same age and found poorer survival among AKI patients. Although the general population is not an ideal control group, a similar approach was taken by Liaño et al
, who compared the mortality of AKI patients with the mortality of the population of Madrid and found poorer survival among AKI patients 
The use of death certificates to address the cause of death is subject to bias, as it depends on the person who completed the certificate. In 2005, the World Health Organization (WHO) examined the quality of the death certificates in 115 countries and classified the Brazilian death certificates to be of medium quality, based on the percentage of incomplete data and ill-defined codes 
. However, França et al
analyzed the data in various regions of Brazil and found that the quality of the data was high in the southeast (where São Paulo City is located) 
. Moreover, we observed that only 5 death certificates (3%) had an improperly defined code as the cause of death.
The present study only included patients followed by nephrologists, and thus, some patients with AKI may have not been evaluated (presumably those with less severe disease). Other important studies, primarily those based on the Program to Improve Care in Acute Renal Disease (PICARD), also included only patients attended by nephrologists 
. The results of the present study cannot be generalized. However, they point to the main importance of pre-existing comorbidities in patients with AKI due to presumed ATN who were discharged free from dialysis.
Our study also has several strengths. It was conducted in a developing country, where there is a lack of true epidemiologic studies on AKI 
. We used a globally accepted AKI definition that was based not on a calculated reference Scr but on a reported reference Scr 
. We lost only 40 patients (7%) after hospital discharge (). We were also able to clearly characterize the AKI episode, which allowed us to adjust the analysis of outcomes for many important factors. To the best of our knowledge it was the first time that the causes of long-term mortality were carefully studied.
In conclusion, long-term mortality after an AKI episode due to presumed ATN is high and is more closely related to previous comorbidities than to the characteristics of the AKI. In our view, all survivors of AKI should have a medical follow-up after hospital discharge and all efforts should be made to control their comorbidities.