The prevalence of CKD is increasing worldwide due to the aging of the general population and the increasing prevalence of diabetes, hypertension, and CV disease . Among the 18,412 patients hospitalized during 2007, the overall prevalence of CKD was high (about 25%). In particular, nearly half of patients older than 70 years of age had CKD. The prevalence of CKD in patients with CV disease, diabetes and hypertension was only slightly lower than the prevalence in older individuals (44%, 41%, and 37%, respectively). CKD has been shown to be an independent risk factor for adverse cardiovascular events and mortality. In fact, for each CKD patient who lives long enough to develop end-stage renal disease requiring dialysis, several patients have already had fatal or non-fatal CV events [1, 32]. Therefore, identifying patients with CKD has become a major health issue. However, identification of kidney disease does not often happen early in the course of the disease, not only because it progresses asymptomatically until kidney function is severely compromised, but also because the reported "normal" range for serum creatinine values does not take into consideration patients' age and anthropomorphic characteristics. To solve this problem, a number of standardized formulas for estimating patients' creatinine clearance  or glomerular filtration rate [7, 8] have been developed.
The health information systems that are currently in place in various health organizations  use the MDRD formula to calculate an eGFR value for each patient who has serum creatinine levels tested in order to overcome problems associated with the variability of "normal" creatinine values and their interpretation by physicians.
The lack of attention paid to kidney disease by physicians is well known from the literature. This study is the first one to quantify this phenomenon in a highly specialized hospital where patient eGFRs are calculated in a standardized fashion each time the serum creatinine is evaluated. Thus, all physicians who request serum creatinine values are always informed about that patient's renal function, and it is up to the physician to place the proper emphasis on it.
At least theoretically, being aware of this value and correctly interpreting it should help with the diagnosis of CKD , as timely identification is extremely important for slowing progression and reducing the associated increase in cardiovascular risk [36, 37]. In sharp contrast with this hypothesis, the present study highlights that in 80% of hospitalizations in which there was documented evidence of renal failure by eGFR, CKD was not listed as a final ICD-9 code diagnosis, thus underscoring the pitfalls associated with this approach [38–40]. Notably, we found that while only 19% of patients with stage 3 or higher CKD (eGFR < 60 ml/min) received a correct ICD-9 diagnosis of kidney failure, this percentage rose to 56% in patients with stage 4 or 5 CKD (eGFR < 30 ml/min). Although diagnosing half of patients with advanced kidney disease based on available laboratory data should not be considered a "good" result, it is evident that ICD-9 performance improved considerably in this subgroup of patients. There may be several explanations for this. Perhaps physicians do actually take into account lower values of reported eGFR (i.e., values <30 ml/min) when assigning ICD-9 codes at discharge, but fail to recognize moderate CKD (stage 3) as a clinical problem. Alternatively, it is possible that they use creatinine values when assigning an ICD-9 diagnosis of CKD, because creatinine values are usually only elevated above the "normal" range when eGFR falls below 30 ml/min; for higher values of eGFR, creatinine values remain "normal", and therefore an ICD-9 diagnosis of CKD may remain overlooked. Whatever the cause, our study highlights the presence of a "gray area", corresponding to stage 3 CKD, in which most of the diagnoses are missed. In particular, when considering the first admissions of 18,412 patients admitted during 2007, 4,096 had an eGFR between 30 and 59 ml/min, but only 537 (13.1%) correctly received a diagnosis of chronic kidney disease. Thus, only one in eight patients with stage 3 CKD received an ICD-9 diagnosis of kidney failure. This is of great concern when considering the importance of early diagnosis not only in reducing the burden of kidney disease, but also in preventing the adverse effects of inadequate drug dosing or inappropriate exposure to nephrotoxic agents.
Another important point is the very low detection rate of CKD in the elderly that we observed in this study. Although the estimated prevalence ratio for CKD was 3.41 (3.20-3.63, p < 0.001) in patients ≥ 70 years of age as compared to patients <70 years of age, the ICD-9 sensitivity for detection of CKD was similar between the two groups (19% vs. 19.2%, p = 0.68), indicating that physicians failed to pay the necessary attention to renal function when evaluating higher-risk elderly patients.
On the other hand, detection of CKD through ICD-9 codes was significantly better in patients with diabetes mellitus, hypertension, or CV disease as compared to patients without diabetes, hypertension, or CV disease (sensitivity values: 26.8%, 22.2%, and 23.7%, vs. 17.1%, 16.9, and 16.1%, respectively; p < 0.001), although these values are still low when considering the widely described relationship between such comorbidities and CKD.
A statistically significant difference in the detection of CKD was also evident in patients admitted in medical departments as compared to surgical departments (17.4% vs. 13.0%, p = 0.001). It is possible that medical specialists are slightly more careful about evaluating eGFR at the time of discharge than surgical specialists. However, this phenomenon likely only has a small amount of clinical relevance given the magnitude of the problem.
The performance of ICD-9 codes in detecting CKD varied significantly between females and males (sensitivity: 12% vs. 28.4%, p < 0.001). We are unable to find any obvious explanation for this performance variation at the present time and are planning further analyses of our data to further explore the spectrum effect of ICD-9 performance.
Our study has several limitations. First, the diagnosis of chronic kidney disease was based on a single determination of serum creatinine level and eGFR [20, 41] because our database was created using administrative data about a large number of hospitalized patients, and because nearly every hospitalization lasted less than three months. Consequently, we may have overestimated the number of patients with chronic kidney disease. However, the risk of overestimating CKD is counterbalanced by the use of a single centralized laboratory and by the availability of serum creatinine calibration[42, 43], allowing the reduction of misclassification bias. Secondly, although eGFR has been widely used to diagnose CKD in hospitalized patients  with a wide spectrum of pathologies, including acute coronary syndromes , congestive heart failure  and stroke , equations for estimated GFR perform better for healthy, stable patients than for acutely ill, hospitalized patients . To reduce the impact of acute renal failure on the study, the last eGFR for hospitalization was analyzed, intensive care and nephrology unit admissions were excluded, and patients whose discharge forms included ICD-9-CM code diagnoses, renal replacement therapy, acute renal failure, or contrast administration were excluded from the analysis.