OUP user menu

Incidence of renal dysfunction over 6 months in patients with chronic heart failure due to left ventricular systolic dysfunction: contributing factors and relationship to prognosis

Ramesh de Silva, Nikolay P. Nikitin, Klaus K.A. Witte, Alan S. Rigby, Kevin Goode, Sunil Bhandari, Andrew L. Clark, John G.F. Cleland
DOI: http://dx.doi.org/10.1093/eurheartj/ehi696 569-581 First published online: 19 December 2005


Aims To determine the prevalence and incidence of renal dysfunction (RD) in patients with chronic heart failure (CHF), to identify contributory factors and predictors of worsening renal function (WRF), and to explore the relationship between RD and mortality.

Methods and results Prospective data on 1216 patients with CHF were analysed. The glomerular filtration rate (GFR) was used to determine renal function, and WRF was defined as an increase in serum creatinine of >26.5 µmol/L (>0.3 mg/dL). The prevalence of RD defined as a GFR of <60 mL/min was 57%. During 6 months, WRF occurred in 161 (13.0%) patients. Predictors of WRF were vascular disease, the use of thiazide diuretics, and a baseline urea >9 mmol/L. Two hundred and sixty-three (21.6%) patients died, and baseline RD and WRF both predicted a higher mortality (P<0.001), whereas an improvement in renal function over the first 6 months predicted a lower mortality (hazard ratio 0.8, 95% confidence interval 0.6–1.0).

Conclusion In ambulatory patients with CHF, RD is common, commonly deteriorates over a relatively short period of time, is unlikely to recover substantially, and augurs a poor prognosis.

  • Chronic heart failure
  • Renal dysfunction


Chronic heart failure (CHF) is common, is an important cause of hospitalizations, and is associated with significant morbidity and mortality.1 Despite some recent successes, the prognosis of heart failure remains poor.24 Reports from randomized controlled trials suggest that renal dysfunction (RD) is common in patients with heart failure and is associated with an adverse prognosis.5,6 However, thesestudies usually enrolled younger patients with fewer co-morbidities than those managed in clinical practice. Moreover, patients with significant RD were often excluded from the trials.712 Information on the prevalence and natural history of RD in patients hospitalized due to worsening heart failure has been published,1315 but few data from epidemiologically representative cohorts of ambulatory outpatients with CHF exist.

RD in patients with heart failure is multi-factorial. Renal function may act as a barometer of cardiac function. As heart failure progresses, preferential renal vasoconstriction occurs diverting blood away from the kidney to maintain blood flow to the heart, brain, and, during exercise, skeletal muscle. A decline in arterial pressure combined with an increase in venous pressure leads to a fall in the trans-renal pressure gradient, and despite efferent arteriolar constriction, glomerular filtration declines. However, disease of the kidneys and cardiovascular system shares many common aetiologies,16,17 and therefore renal parenchymal disease, renovascular disease, and urinary obstruction may also be important causes of RD in patients with heart failure.

RD may confer an adverse prognosis because it is a marker of more severe vascular disease, greater age, worse cardiac function, or greater likelihood of life-saving treatment being withheld. Alternatively, or in addition, RD may cause salt and water retention due to impaired nephron function and reduced diuretic efficacy. RD may also lead to failure to excrete toxic substances such as oxidized catecholamines, uric acid, and other uraemic factors.18 However, the effectiveness of interventions to improve renal function in heart failure remains largely anecdotal and little advice is given by guidelines for heart failure.19,20

Our aims were to determine the prevalence and incidence ofRD in a large community-based management programme for patients with CHF due to left ventricular systolic dysfunction, to identify possible contributory factors and predictors of worsening renal function (WRF), and to explore the relationship between RD and mortality.


Ethical approval

All participants provided written informed consent, and the study was carried out in accordance with the Helsinki Declaration II and the European Standards for Good Clinical Practice. Ethical approval was granted by the Local Research Ethics Committee.

Study design

This was a single centre, observational, prospective study. Cross-sectional data were used to determine the prevalence of RD in patients with CHF and left ventricular systolic dysfunction. Longitudinal data were analysed to investigate the incidence of RD, predictors of WRF, and their inter-relationship with prognosis.


Patients were identified from a community-based heart failure programme that accepts all patients with a suspected diagnosis of heart failure or major cardiac dysfunction but retains only those in whom the left ventricular ejection fraction is <45% on echocardiography. Heart failure is defined as current symptoms of heart failure, or a history of symptoms controlled by ongoing therapy, due to cardiac dysfunction and in the absence of any more likely cause. The only exclusion criteria for this study were the inability to provide written consent, pregnancy, and renal replacement therapy with dialysis or transplantation.

Clinical assessment

Clinical information obtained included past medical history and drug and smoking history. Clinical examination included assessment of height, weight, heart rate, rhythm, and blood pressure. Diabetes was defined as a previous diagnosis of this condition and sub-categorizedas requiring insulin, oral hypoglycaemic agent, or diet-controlled. Hypertension and chronic obstructive pulmonary disease were defined as a prior history of receiving drug treatment for these conditions. Patients with previous myocardial infarction, coronary artery bypass surgery, positive tests for ischaemia, or coronary artery disease on angiography were considered to have ischaemic heart disease (IHD). Vascular disease was defined as a previous clinical diagnosis with investigations that identified one or more of the following: peripheral vascular disease, renal artery stenosis, cerebrovascular disease, or abdominal aortic aneurysm. Cardiomyopathy was defined as the presence of left ventricular systolic dysfunction in the absence of any known cause. Daily doses of diuretics were expressed in furosemide equivalents (bumetanide 1 mg=furosemide 40 mg). Doses of agents blocking the renin–angiotensin–aldosterone system [both angiotensin-converting enzyme (ACE)-inhibitors and angiotensin receptor blockers] were expressed as the percentage of the maximum recommended daily dose used in CHF patients.19,20 All medications were those that the patients were taking at the time of referral to the heart failure clinic. Blood tests included urea, creatinine, sodium, and haemoglobin.


Following the baseline visit, all patients with a diagnosis of heart failure were reviewed at approximately 3 and 6 months. Patients not on optimal ACE-inhibition and/or beta-blocker therapy at baseline were seen at 2, 4, and 6 weeks for up-titration of their medications and followed up as above. Data collection was performed between March 1999 and November 2003, but only patients who survived more than 6 months were included in the analysis.

Assessment of renal function

We considered two different methods of assessing renal function. These were the serum creatinine (SCr) and the glomerular filtrationrate (GFR) using the simplified modification of diet in renal disease (MDRD) prediction equation [186×SCr−1.154 (mg/dL) × Age−0.203 (years)×0.742 if female, ×1.212 if black].21,22

Definitions of RD and changes in renal function

The primary definition of RD was a GFR <60 mL/min.6,23,24 We also considered a SCr >130 µmol/L (1.5 mg/dL) as a marker of RD.25 For the presentation of data, we divided patients into three groups based upon the baseline SCr [normal ≤106 µmol/L (≤1.2 mg/dL), minor increase 106–177 µmol/L (>1.2–2.0 mg/dL), marked increase >177 µmol/L (>2.0 mg/dL)]26 and four groups based upon the baseline GFR (normal ≥90 mL/min, mild impairment 60–89 mL/min, moderate impairment 30–59 mL/min, and severe renal impairment ≤29 mL/min) to compare the clinical characteristics of patients with different severities of RD. These values were defined by the Kidney Disease Outcome Quality Initiative guidelines.27

Changes in renal function were defined using two methods. The primary method used was an increase or decrease in SCr of >26.5 µmol/L (>0.3 mg/dL), a convention previously used by other groups.1315,28 In addition, we also used a change in the GFR category between normal, mildly impaired, moderately impaired, and severely impaired, as mentioned above27 to assess changes in renal function over a period of 6 months.

Statistical analysis

Continuous variables are presented as mean±standard deviation. Categorical data are presented as percentages. Analysis of variance was used for continuous data, whereas tests for trend were applied across the three SCr categories and the four GFR categories, respectively. The evaluation for linear trend in percentages was carried out using the Cochrane–Armitage test.29

Multivariable logistic regression models for WRF were developed using all baseline variables. Data from the logistic regressions are presented as odds ratios with 95% confidence intervals (CIs). The odds ratio is an approximation to the relative risk.30,31 Models were adjusted for length of follow-up because not all patients had the same length of follow-up. We did not consider the duration of heart failure as a potential variable due to the uncertainty of defining the onset of the syndrome. All the continuous variables were assessed for linearity by including a squared term. For all but one baseline variable (GFR), the linearity assumption was satisfied. However, our preference was to analyse all the continuous variables by categories essential for presentation purposes. We checked for co-linearity by calculating Pearson's correlation coefficients. Model building was based on backwards elimination (P-value for entry=0.05; P-value for removal=0.1). Backwards elimination is preferable to forward selection.32 Models were validated using re-sampling based on 10-fold cross-validation.33 The data were divided into 10 subsets of approximately equal size while maintaining the frequency of WRF within each of the subsets. Hence, each subset contained approximately 16 cases of WRF. For each subset, we generated a model for RD, leaving out one subset at a time. The omitted subset was used to calculate the misclassification rate per model. We are aware of the problems of model building using stepwise methods and these issues are discussed later.

The matched pairs odds ratio was used to assess changes in treatment from baseline to 6 months. The 95% CIs were calculated according to McNemar's test.34 Kaplan–Meier survival curves35 are presented for mortality data using the guidance of Pocock et al.36 The log-rank test was used to assess the equality of survivor function across groups. The Cox regression model was used to calculate hazard ratios with 95% CI. The Cox regression model is semi-parametricin the sense that no assumption concerning event-free survival times is necessary. The proportional hazards model is based on the assumption that the effect of a risk factor, expressed as a hazard ratio, is constant over time. The assumption of proportionality was tested for each variable, using the method of Grambsch and Therneau.37 The output of this test is a P-value, which for all our variables tested was not significant. Hence, we did not include time-dependent variables in our Cox models. We did not validate these models using cross-validation because this method is less well developed than the logistic model. A difficulty is understanding how to deal with censoring.38 Instead, we produced a subset of variables on univariate Cox analysis that were significantly associated with follow-up time. We included all of these variables in the final Cox model whether significant or not. We further adjusted the final model for age, sex, and New York Heart Association (NYHA) functional class. Variables used in the equations to calculate renal function were not excluded in any analysis.

SPSS (version 11) and GLIM4 statistical computer packages were used to analyse the data.39 An arbitrary level of 5% statistical significance was used throughout (two-tailed). One of the statistical issues to address is the problem of multiple testing when many variables are present and the possible inflation of Type I error. However, there is no consensus on what procedure to adopt to allow for multiple comparisons.40 Hence, in order to account for the inflation of experimentwise Type I error due to multiple testing, we have followed the recommendations of Perneger41 and not adjusted for this. There were two outcome measures: all-cause mortality and WRF.


Study sample

A total of 2621 patients with suspected heart failure were assessed. Of these, 905 patients werereferredfromgeneral outpatient clinics, 1087 patients were referred by primary care physicians, and 629 patients were identified while in hospital with suspected heart failure and subsequently seen following stabilization and discharge. In 1406 patients, the diagnosis of heart failure due to left ventricular systolic dysfunction was confirmed following clinical assessment and investigations as described above and were entered into thelong-term management programme. At 6 months, 91 patients had died, 21 patients were lost to follow-up, and76 patients had missing data points. The remaining 1216 patients had complete data sets at both baseline (Table 1) and 6-months follow-up and constituted our study sample.

We compared the baseline data between the 190 patients with missing 6 months data and our study sample of 1216 patients. Four variables showed significant differences: mean heart rate [71 (SD 15) vs. 76 (SD 17) b.p.m.], patients with hypertension [96 (51%) vs. 502 (41%)], patients with dilated cardiomyopathy [23 (12%) vs. 94 (8%)] and the number of patients taking diuretics [20 (63%) vs. 893 (73%), P<0.05 for all]. Other baseline data did not differ significantly.


The mean age at baseline was 71 years (range 22–95 years) and 837 (69%) were male. One-fifth (n=257) were aged ≥80. Mean ejection fraction was 34% (SD 10) and CHF was due to IHD in 798 (66%) patients. The mean values for SCr and GFR were 122 µmol/L (SD 51) [1.4 mg/dL (SD 0.6)] and 57 mL/min (SD 21), respectively. Figure 1 shows the distributions for the SCr and GFR values. The prevalence of RD defined as using an SCr of >130 µmol/L (1.5 mg/dL) and a GFR of <60 mL/min was 32 and 57%, respectively.

Contributing factors and associations with renal function at baseline

Five hundred and forty-eight patients (45%) had a normal SCr, 526 (43%) had a minor increase, and 142 (12%) had a marked increase. GFR was normal (≥90 mL/min) in only 82 patients (7%), mildly reduced (60–89 mL/min) in 437 patients (36%), moderately reduced (30–59 ml/min) in 577 patients (47%), and markedly reduced (≤29 ml/min) in 120 patients (10%).

Patients with severely impaired renal function were older, more often female and had lower left ventricular ejection fractions, diastolic blood pressure, and haemoglobin levels than patients with normal or mildly impaired renal function (Table 1). As the severity of renal impairmentincreased, so did the prevalence of IHD, hypertension, vascular disease, and other common co-morbidities. The proportion of patients taking diuretics, and the dose administered, increased as renal function declined. The proportion of patients taking ACE-inhibitors or beta-blockers however was not different between the groups. As renal function worsened, the proportion of patients taking aspirin, statins, and calcium channel blockers declined.

Independent predictors of a low GFR at baseline were increasing age, low haemoglobin, poorer NYHA functional class, presence of IHD, vascular disease and hypertension, and the use of spironolactone and loop diuretics (data not shown).

Changes in renal function assessed by SCr

During the 6-month follow-up, mean SCr rose by 4 µmol/L (0.05 mg/dL) (95% CI=2–4). WRF, defined as a rise in SCr of >26.5 µmol/L (0.3 mg/dL), occurred in 161 (13.0%) patients. One hundred and twenty-one (9.7%) patients showed an improvement by this amount. Figure 2A shows these changes in SCr as frequency histograms. Figure 3A shows the number of patients in whom renal function worsened, improved, or remained the same over 6 months period, stratified by the presence or absence of baseline renal impairment, defined by the SCr values mentioned earlier. Table 2 indicates the risk factors for WRF. Factors associated with WRF were a history of vascular disease, low systolic and diastolic blood pressures, low ejection fraction, high baseline blood urea levels, and thiazide and diuretic use. The relationships with all other variables were not statistically significant. The model-building exercise produced a wide variety of models but with some commonalties (Table 3). A total of 12 variables appeared in 10 different predictor models for RD but no model was repeated more than once. Misclassification rates varied greatly (Table 4). From these models, the presence of vascular disease, use of thiazide diuretics, and blood urea levels were strongly associated with WRF.

Changes in renal function assessed by the GFR

During the 6-month follow-up period, mean GFR fell by 2 mL/min (95% CI=1–3). Figure 2B shows these changes in GFR as frequency histograms. Figure 3B shows the number of patients in each category of calculated GFR at baseline and at follow-up and changes between groups. GFR deteriorated by at least one category in 229 patients (19%) and improved by at least one category in 145 (12%).

We looked at whether drug treatment at baseline and 6months could explain the change in renal function as assessed by GFR. Table 5 shows the changes in treatment that occurred between baseline and 6 months in the 374 patients who had a change in their GFR category. Beta-blocker use increased regardless of change in GFR, probably reflecting continued efforts to optimize therapy. Patients were more likely to be taking aspirin at baseline than follow-up if their GFR worsened, suggesting that worseningGFR may provoke aspirin withdrawal. There was no significant difference in the percentage of the maximum recommended daily dose of ACE-inhibitors used at baseline (50±48%) or follow-up (54±47%) either in patients whose GFR improved [mean difference 3.0, 95% CI −4.6–10.6, P=0.44] or in patients whose GFR worsened [mean difference 4.6, 95% CI −1.1–10.3, P=0.11]. The daily dose of diuretics used was not significantly different in patients whose GFR improved [mean difference 5.6 mg, 95% CI −2.6–13.8, P=0.41]. However, in patients whose GFR worsened, the diuretic dose was higher at 6 months when compared to baseline [mean difference in furosemide-equivalent dose of 11.7 mg, 95% CI 4.8–18.6, P=0.001].

Relationships to prognosis

There were 263 deaths at follow-up representing a crude death rate of 21.6%. The median time to follow-up was 16.5 months (interquartile range 9.6–26.3 months). Kaplan–Meier survival curves for baseline SCr and GFR, along with the numbers at risk, are shown in Figure 4. Patients with worse renal function at baseline had a poorer prognosis (P<0.001). The relationship between death and renal disease was investigated by Cox regression analysis from which hazard ratios and 95% CI were generated. Patients with WRF had a poorer prognosis, although the severity of RD rather than its change appeared the most important determinant of outcome. A total of nine variables were significantly associated with follow-up time on univariate analysis. All these variables were included in the final Cox model (Table 6). GFR was eliminated from the model by SCr. When considering the joint effects of baseline SCr and WRF on death, a synergistic relationship was noted (Figure 5). Recovery in renal function (a fall in the SCr of >26.5 µmol/L (0.3 mg/dL)) was associated with a better prognosis (hazard ratio 0.8, 95% CI 0.6–1.0) adjusted for age, sex, and NYHA class).


This study provides unique information on the prevalence and natural history of RD in a large epidemiologically representative population of outpatients with relatively stable CHF due to left ventricular systolic dysfunction. It shows that RD is common, commonly deteriorates over a relatively short period of time, is unlikely to recover substantially, and augurs a poor prognosis.


In patients with a recent hospitalization for worsening heart failure, the prevalence of RD has been reported to be between 24 and 75% depending on the definition used.14 One-third of our patients had evidence of renal impairment using an SCr threshold of 130 µmol/L (1.5 mg/dL) but more than half had a calculated GFR of <60 mL/min. Among patients in the SOLVD Prevention and Treatment trials, the prevalence of RD defined as a GFR of <60 mL/min was 21 and 36%, respectively, but in these studies, patients with SCr 220 µmol/L (2.5 mg/dL) and those aged >79 were excluded.6 The absence of such exclusion criteria, combined with higher rates of co-morbidity, probably accounts for the greater prevalence of RD in our population and is similar to the findings reported recently by others.42

Incidence of WRF

The natural progression of RD in outpatients with CHF is oneof the deterioration. Although the incidence of deterioration is much less than in patients admitted with an exacerbation of heart failure using the same definition for WRF,13,15 the gross rate of 13% over 6 months is substantial. Even if the rate of ‘recovery’ is subtracted, the net rate of 3% over 6 months is still of concern.

Causes of RD and reasons for WRF

Haemodynamic and renal factors

Patients with co-morbid vascular disease, higher levels of baseline blood urea, and lower ejection fractions were at greater risk of developing WRF. Two previous reports analysing WRF in hospitalized patients with CHF describe systemic hypertension, diabetes, history of CHF, and high baseline SCr values as independent risk factors for the development of WRF (using the same definition as us).13,15 These data are consistent with the concept that WRF in patients with heart failure is multi-factorial. First, there is a potential ‘haemodynamic’ element related to declining cardiac function, renal blood flow, and perfusion pressure leading to a fall in GFR. Secondly, there is an intrinsic renal element. Many patients, especially those with peripheral vascular disease, have renovascular disease and these patients are at greater risk of developing WRF either due to progression of arterial disease or in response to haemodynamic deterioration.9,43


RD could also be a side effect of treatments for heart failure. Loop and thiazide diuretics and spironolactone were all associated with WRF, although in an observational study such as this cause and effect cannot usually be distinguished. There must have been a reason to alter diuretic therapy and that will usually have been an attempt to improve symptoms. Also, as GFR declines, less diuretic will be filtered and intra-tubular delivery will decline with a loss of diuretic effect. Higher doses may be required in the presence of RD to achieve the same diuretic effect.

We showed no difference in the frequency or dose of ACE-inhibitors or angiotensin receptor blocker use in patients with varying degrees of RD, which contrasts with previous reports.44,45 Although ACE-inhibitors may reduce renal perfusion pressure, efferent arteriolar tone, and GFR in patients with heart failure in the short-term, there is evidence that they retard long-term deterioration of renal function in other clinical settings.46,47 The net long-term effect of ACE-inhibition on renal function in most patients with heart failure appears fairly neutral. Similarly, beta-blockers did not exert marked effects on the incidence of RD.

Statins have been reported to reduce the risk of WRF in patients with renal artery disease.48 In our study, patients with moderate or severe RD were less likely to be taking a statin, suggesting that chronic use of these agents might retard progression of atherosclerotic renal disease, although other mechanisms of benefit should not be discounted. During 6 months of follow-up, we were unable to confirm a protective effect of statins on WRF but statins were associated with a markedly lower mortality. The hypothesis that the observed association between better prognosis and statin treatment is cause and effect is being tested in two large randomized controlled trials.49

Relationships to prognosis

The relationship between RD and survival seen in our cohort of patients has been observed in several other multi-centre studies of CHF and, more recently, in clinical practice.42 In the SOLVD treatment trial,6 patients with CHF who had a GFR <60 mL/min were more likely to die, especially from worsening CHF. Similar findings were reported by Mahon et al.50 These studies show that even minor reductions in GFR, which do not necessarily increase SCr above the normal range, are associated with a worse outcome. The seven-fold increase in the risk of death in the presence of a markedly raised baseline SCr and WRF when compared with a normal baseline SCr and no WRF observed in our cohort implies a synergistic association between these two variables (Figure 5). Interestingly, it did not appear that GFR was a better predictor of prognosis than SCr and therefore it seems appropriate when using markers of RD to assess prognosis to use the simplest available measure. In our study, the severity of RD appeared more prognostically important than the change.

The reasons for the increased mortality in patients with RD and CHF are multi-factorial and complex. RD may limit the use of life-saving interventions such as ACE-inhibitors, angiotensin receptor blockers, and beta-blockers, although this was not obvious in the current study. Patients with RD have more peripheral vascular disease and may be at higher risk of vascular events. RD may lead to diuretic resistance and sodium and water retention leading to an increase in cardiac filling pressures, progressive ventricular dilation, and hyponatraemia.51,52 Electrolyte disturbances may increase the risk of arrhythmias,53 and it has been reported that an increase in mortality is observed in certain groups of patients with CHF who are prescribed spironolactone and ACE-inhibitors due to the adverse effects of hyperkalaemia.54 Other reasons for a higher mortality include abnormal calcium metabolism,55 hyperparathyroidism,56 increased coaguability, hyperhomocysteinaemia,57 and ureamic cardiomyopathy.

The improvement in prognosis seen with recovery of renal function has not been reported before in a stable, community-based population with heart failure. The reasons are multi-factorial and probably due to a reverse of the mechanisms discussed above. Reports of resolution of ventricular dysfunction after renal transplantation in patients with renal failure58,59 or after the correction of bilateral renal artery stenosis60 further suggest that RD may cause or exacerbate CHF.

Model selection procedures

In epidemiological studies, statistical models are often determined by data-driven selection methods. Such methods usually have only a heuristic basis and their sampling properties are largely unknown. Forward selection and backward elimination are two of the most widely used (and abused) selection methods. Indeed, many computer packages have such automated stepwise methods built into their software. Statistical objections to automated selection methods have long been known. For example, results lead to standard errors that may be too low.6163 If important variables are omitted from the final model, the regression coefficients are known to be biased.64 Finally, the selection method affects the properties of the tests of the final model itself.65 Many ‘solutions’ to these objections have been proposed with most focusing on resampling. Perhaps the best known resampling methods are cross-validation, the jack-knife, and bootstrapping. It is not our aim here to discuss which of these resampling methods is best but an excellent overview is given Sauerbrei.32

In the past, it may have been true that a single model was typically fitted to a given data set.66 However, modern day computing resources mean that it is possible to explore many models simultaneously so that the ‘one-fit’ model is no longer appropriate. This fits in with the notion of having a portfolio of plausible models.67 Finally, Copas68 stated that ‘a good predictor may include variables which are not significant, exclude others which are, and may involve coefficients which are systematically biased’, sentiments with which we agree. In conclusion, stepwise methods are not designed to select ‘best’ models or to indicate their relative importance but rather designed to select subsets from data sets ‘padded with extraneous variables, for example, those that contain everything we could measure’.69

Study limitations

The definition of WRF we used has arisen by convention rather than been justified scientifically.1315 An increase in SCr of as little as 17.7 µmol/L (0.2 mg/dL) is associated with an adverse outcome.14 Other investigators have used a rise in SCr above a threshold to define renal insufficiency [e.g. SCr >221 µmol/L (>2.5 mg/dL) or a percentage increase from baseline (e.g. >25% increase)].70,71 Each definition of WRF has its merits and problems. For instance, the final SCr may still be in the normal range even if there is a substantial increase. This may not carry the same adverse prognosis as a similar rise in SCr that results in an increased final value. Likewise the definition used for an improvement in renal function is not evidence based. The study analysed data at two defined points during patient follow-up, but does not take into account the factors in the intervening periods that could effect the renal function such as hospitalizations with significant haemorrhage, dehydration, or exposure to intravenous contrast. We also do not have temporal information as to when the doses of diuretics were changed in relationship to changes in renal function. Finally, because of the inherent nature of the study, patients who died or had incomplete data sets at follow-up were not included in the analysis. This could have the potential to introduce a selection bias.


RD is common in CHF and is associated with a poor prognosis, which is only partly explained by its association with poorer ventricular function. Renovascular disease, perhaps both of extra- and intra-renal vessels, may be an important under-recognized risk factor for WRF in patients with heart failure. Whether routine investigation for and treatment of renal artery stenosis in this setting is beneficial is uncertain, but is being tested in a subset of patients with heart failure in a randomized controlled trial (ASTRAL, Angioplasty and Stent for Renal Artery Lesions).72

Conflict of interest: none declared.

Figure 1 (A) Distribution of SCr in 1216 patients with heart failure. Mean 122 (+51) µmol/L. Sixteen patients with values >300 µmol/L are not shown. (B)Distribution of the GFR calculated by the MDRD prediction equation in 1216 patients with heart failure. Mean 57.3 (+21.4) mL/min.

Figure 2 (A) Distribution of the change in SCr over a period of 6 months (15patients with an improvement in their SCr by >100 µmol/L and 16 patients with a worsening of their SCr by >100 µmol/L are not shown). (B) Distribution of the change in the GFR (in mL/min) over a period of 6months (five patients with an improvement in their GFR by >50 mL/min and eight patients with a worsening of their GFR by >50 mL/min are not shown).

Figure 3 (A) The number of patients in whom renal function worsened, improved, or remained the same over the 6-month period. Definitions use the SCr and are as follows: RD=SCr >130 µmol/L, WRF=an increase in SCr by 26.5 µmol/L, improving renal function = a decrease in SCr by 26.5 µmol/L, no change in renal function=change in SCr of less than ±26.5 µmol/L. Numbers outside the boxes represent the number of patients. (B) Change in the GFR in millilitres per minute over 6 months for 1216 patients. Figures are mean (standard deviation) for GFR. Numbers outside the boxes represent the number of patients.

Figure 4 (A) SCr and relationship to prognosis. (B) Calculated GFR and relationship to prognosis.

Figure 5 Relationship between baseline SCr, WRF, and death. WRF defined as an increase in SCr of >26.5 µmol/L (>0.3 mg/dL), SCr low≤106 µmol/L (≤1.2 mg/dL), medium=106–177 µmol/L (1.2–2.0 mg/dL), and high ≥ 177 µmol/L (>2.0 mg/dL).

View this table:
Table 1

Baseline characteristics for 1216 patients, as classified by calculated GFR

VariableTotal (n=1216)Normal
>≥90 mL/min (n=82)
>60 to <90 mL/min (n=437)
>30 to <60 mL/min (n=577)
><30 mL/min (n=120)
P-value for trend
Age (years)71 (10.8)59.6 (11.9)68.1 (11.4)73.8 (8.9)75.9 (8.9)<0.0001
Male sex837 (69%)66 (81.6%)321 (70.4%)387 (65.2%)63 (52.2%)<0.0001
Weight (kg)77.6 (17)82.7 (17.1)79.4 (17.3)75.9 (16.3)76.0 (18.0)<0.0001
BMI27.8 (5.6)28.1 (5.1)27.6 (5.9)27.4(5.5)27.9(5.5)0.64
NYHA classes
 Class I160 (12.8%)14 (17.1%)68 (14.9%)65 (11.0%)7 (5.9%)0.0009a
 Class II772 (61.7%)53 (64.6%)280 (61.4%)346 (58.3%)68 (57.0%)
 Class III304 (24.3%)15 (18.3%)86 (18.9%)159 (26.8%)40 (33.1%)
 Class IV16 (1.3%)02 (0.4%)9 (1.5%)5 (4.1%)
Heart rate (b.p.m.)75.6 (16.8)77.1 (19.3)76.2 (17.0)75.4 (16.6)74.6 (16.2)0.21
Systolic BP (mmHg)134.9 (25.8)135.1 (25.1)135.6 (25.0)134.6 (26.1)135.3 (28.4)0.87
Diastolic BP (mmHg)76.8 (25.1)78.7 (14.6)77.0 (36.2)75.9 (14.6)71.4 (13.6)<0.0001
Pulse pressure (mmHg)59.2 (21.3)56.1 (20.7)58.1 (20.2)58.6 (21.3)63.9 (22.9)<0.0001
Ejection fraction34.2 (10.2)36.9 (9.9)35.4 (10.5)33.1 (9.8)33.2 (10.2)<0.0001
Baseline characteristics
 IHD798 (65.6%)41 (50.0%)265 (58.1%)400 (67.4%)92 (76.0%)<0.0001
 Cardiomyopathy94 (7.7%)8 (9.8%)50 (11.0%)25 (4.2%)11 (9.1%)0.012
 Past h/o hypertension502 (41.3%)33 (41.3%)150 (32.9%)256 (43.2%)63 (52.1%)0.0001
 Current smoker114 (9.4%)17 (20.7%)40 (9.0%)46 (8.0%)11 (9.1%)0.024
 Past smoker627 (51.7%)35 (43.7%)218 (47.8%)309 (52.0%)65 (54.4%)0.11
 Diabetes257 (21.1%)17 (20.7%)87 (21.2%)123 (20.5%)30 (24.8%)0.27
 AF307 (25.2%)19 (24.2%)91 (20.0%)170 (28.8%)27 (23.0%)0.18
 Vascular diseaseb204 (16.8%)5 (6.1%)56 (12.3%)111 (18.9%)32 (26.4%)<0.0001
 Chronic obstructive pulmonary disease109 (9.0%)10 (12.2%)34 (7.5%)59 (10.0%)6 (5.0%)0.58
 RAAS blockade918 (76.0%)61 (74.4%)330 (72.6%)453 (76.4%)74 (62.0%)0.19
 RAAS blocker dose/day50.3 (47.5)48.4 (44.4)51.4 (49.0)51.3 (46.7)40.0 (45.9)0.21
 Beta-blocker625 (51.4%)44 (53.7%)237 (51.9%)287 (48.4%)57 (47.8%)0.12
 Diuretic893 (73.4%)45 (55.9%)272 (59.4%)469 (79.0%)107 (89.2%)<0.0001
 Diuretic (dose/day)49.9 (48.1)32.5 (40.7)36.3 (40.9)57.3 (50.0)71.9 (53.1)<0.0001
 Thiazide60 (5%)3 (3.7%)22 (4.8%)31 (5.3%)4 (3.3%)0.74
 Spironolactone244 (20.1%)9 (11.0%)66 (14.5%)134 (22.6)35 (28.9%)<0.0001
 Aspirin591 (48.6%)38 (47.4%)226 (49.6%)281 (47.3%)46 (38.0%)0.074
 Statin406 (33.4%)32 (39.0%)172 (37.7%)169 (28.5%)33 (27.3%)0.0008
 Calcium channel blocker133 (10.9%)16 (19.5%)59 (13.5%)48 (8.3%)10 (8.3%)0.006
 NSAID59 (4.9%)4 (4.9%)19 (4.3%)29 (5.0%)7 (5.8%)0.56
Blood tests
 Haemoglobin (g/dL)13.1 (1.6)13.8 (1.5)13.8 (1.6)13.1 (1.8)11.7 (1.5)<0.001
 SCr (µmol/L)122.5 (50.7)71.1 (9.7)90.8 (13.3)130.7 (26.0)231.5 (66.8)<0.0001
 Serum sodium (mmol/L)138.3 (3.8)137.6 (3.5)138.6 (3.3)138.5 (3.8)136.9 (5.4)0.078
 Urea (mmol/L)9.0 (5.5)5.4 (2.4)6.3 (2.0)9.6 (4.3)18.1 (9.0)<0.0001

Continuous variables are presented as mean (standard deviation), whereas categorical variables are expressed as numbers (percentage). P-values are for differences between GFR groups (columns 3, 4, 5, and 6). See text for detailed definitions. BMI, body mass index; BP, blood pressure; h/o, history of; PMH, past medical history; AF, atrial fibrillation; RAAS, renin–angiotensin–aldosterone system; NSAID, non-steroidal anti-inflammatory drugs.

aNYHA class I vs. II–IV.

bVascular disease is a composite of stroke, transient ischaemic attacks, peripheral vascular disease, renal artery stenosis, or abdominal aortic aneurysms.

View this table:
Table 2

Risk factors for WRF [defined as a rise in SCr of >26.5 µmol/L (0.3 mg/dL)] in 1216 patients with heart failure

VariableBaseline variableNumber without WRFNumber with WRFOR95% CI
Age (years)≤65292401
Ejection fraction (%)40+334351
BMIUnderweight <21.7112181.0
Ideal 21.7–27.1447751.00.6–1.8
Overweight 27.1–31252671.71.0–3.0
Obese >31244370.90.5–1.7
NYHA gradeI137171
Heart rate (b.p.m.)≤64287451
Systolic blood pressure (mmHg)150+305421
Diastolic blood pressure (mmHg)80+466551
Pulse pressure (mmHg)<50365571
Baseline haemoglobin (g/dL)>14.5242301
Baseline sodium (mmol/L)≥140437621
Baseline SCr (µmol/L)≤106500821.0
Urea (mmol/L)≤6342361
Baseline GFR (mL/min)≥9066161.0
60 to <90395610.60.3–1.1
30 to <60498950.80.4–1.4
 Past history of hypertensionNo627871
 Atrial fibrillationNo7881211
 Current smokerNo9611411
 Past smokerNo509801
 Chronic obstructive pulmonary diseaseNo9621451
 Vascular diseaseNo8901221
 RAAS blockadeNo256421
 Calcium channel blockerNo9331501

OR, odds ratio; BMI, body mass index; RAAS, renin–angiotensin–aldosterone system; NSAID, non-steroidal anti-inflammatory drugs. ORs calculated adjusting for length of follow-up (months). Example interpretation: patients taking thiazides are 1.9 times as likely to have a WRF when compared with those not taking them.

View this table:
Table 3

Predictor models for WRF

Baseline variableSubset (excluding)Number of times included
Vascular disease+++++++++9
Ejection fraction++++4
Pulse pressure++++4
Current smoking++2

+ means that the variable has been included in the model. Subset1 means all subsets excluding subset 1, and so on. SBP, systolic blood pressure; BMI, body mass index; DBP, diastolic blood pressure.

View this table:
Table 4

Misclassification rates

Omitted subsetPer cent misclassified
View this table:
Table 5

Treatment changes in patients whose GFR category (see text and Figure 3B) either improved (n==145) or deteriorated (n=229) over a period of 6 months

Treatment at baselineTreatment at 6 months
GFR improvedGFR worsened
NoYesORmp(95% CI)NoYesORmp(95% CI)
RAAS blockade
 No26 14 28 28
 Yes 15 900.9(0.4–2.0) 29 1441(0.6–1.8)
 No105 7 156 22
 Yes 13 200.5(0.2–1.4) 16 351.4(0.7–3.3)
 No49 19 62 59
 Yes 9 682.1(0.9–5.0) 17 913.5(2.0–5.0)
 No28 6 53 18
 Yes 13 980.5(0.1–1.2) 17 1411.1(0.5–2.0)
 No138 4 206 6
 Yes 2 12.0(0.3–20.0) 6 111(0.3–3.7)
 No91 13 127 15
 Yes 4 373.3(0.9–14.3) 25 620.6(0.3–1.1)
 No68 10 104 5
 Yes 14 530.7(0.3–1.7) 39 810.1(0.04–0.3)
 No131 0 204 0 0
 Yes 9 50 15 10
 No138 0 221 2
 Yes 4 30 3 30.7(0.07–5.3)

Figures highlighted in bold represent changes in therapy. RAAS, renin–angiotensin–aldosterone system. ORmp, matched pairs odds ratio; CCB, calcium channel blockers; NSAID, non-steroidal anti-inflammatory drugs. Data are presented as a series of 2×2 tables. ORmp with corresponding 95% CIs was calculated for all 2×2 combinations. For example, of the 229 patients whose GFR deteriorated, 81 were taking aspirin both at baseline and at 6 months, 104 were not taking aspirin at either baseline or 6 months, 39 were taking aspirin at baseline but not at 6 months, and five were taking aspirin at 6 months but not at baseline. Hence, for aspirin, the matched pairs odds ratio is 5/39=0.1, calculated as the ratio of the discordant (for treatment) pairs. An odds ratio>1 means that the drug use is more likely at 6 months of follow-up; an odds ratio<1 means that the drug use is less likely at 6 months of follow-up; an odds ratio=1 means that drug use did not change from baseline to 6 months of follow-up.

View this table:
Table 6

A Cox-regression model predicting mortality (excluding patients who died in the first 6 months of follow-up or who did not have repeat SCr measurement)

VariableLevelHR (95% CI)P-value
Vascular diseaseNo1.0
Yes1.3 (0.8–2.1)0.35
Chronic obstructiveNo1.0
 pulmonary diseaseYes1.8 (1.0–3.1)0.05
Loop diureticsNo1.0
Yes1.4 (0.8–2.4)0.35
Yes1.7 (1.1–2.6)0.017
Ejection fraction (%)>401.0
30–391.9 (1.1–3.3)
<301.6 (1.1–2.4)0.02
Baseline urea (mmol/L)≥61.0
6.1–9.00.8 (0.4–1.5)
>91.5 (0.8–2.9)0.06
Baseline SCr (µmol/L)<1061.0
106–1771.1 (0.6–1.8)
>1771.5 (0.8–2.9)0.81
Baseline sodium (mmol/L)≥1401.0
137–1390.9 (0.6–1.5)
≤1361.2 (0.8–2.0)0.75
Haemoglobin (g/dL)>141.0
13.3–14.51.1 (0.6–2.0)
12.0–13.21.6 (0.9–2.9)
≤11.90.9 (0.5–1.8)0.18

HR, hazard ratio. Adjusted for age, sex, and NYHA class. All variables included were significant on univariate analysis.


View Abstract