CKD and Hospitalization in the Elderly : A Community-Based Cohort Study in the United Kingdom

Background: We previously have shown that chronic kidney disease (CKD) is associated with cardiovascular and all-cause mortality in community-dwelling people 75 years and older. The present study addresses the hypothesis that CKD is associated with a higher rate of hospital admission at an older age. Study Design: Cohort study. Setting & Participants: 15,336 participants from 53 UK general practices underwent comprehensive health assessment between 1994 and 1999. Predictor: Data for estimated glomerular filtration rate (eGFR, derived from creatinine levels using the CKD Epidemiology Collaboration [CKD-EPI] study equation) and dipstick proteinuria were available for 12,371 participants. Outcomes: Hospital admissions collected from hospital discharge letters for 2 years after assessment. Measurements: Age, sex, cardiovascular risk factors, possible biochemical and health consequences of kidney disease (hemoglobin, phosphate, and albumin levels; physical and mental health problems). Results: 2,310 (17%) participants had 1 hospital admission, and 981 (7%) had 2 or more. After adjusting for age, sex, and cardiovascular risk factors, HRs were 1.66 (95% CI, 1.21-2.27), 1.17 (95% CI, 0.95-1.43), 1.08 (95% CI, 0.90-1.30), and 1.11 (95% CI, 0.91-1.35) for eGFRs 30, 30-44, 45-59, and 75 mL/min/1.73 m, respectively, compared with eGFRs of 60-74 mL/min/1.73 m for hospitalizations during 6 months of follow-up. HRs were weaker for follow-up of 6-18 months. Dipstick-positive proteinuria was associated with an increased HR throughout follow-up (HR, 1.29 [95% CI, 1.11-1.49], adjusting for cardiovascular risk factors). Dipstick-positive proteinuria and eGFR 30 mL/min/1.73 m were independently associated with 2 or more hospital admissions during the 2-year follow-up. Adjustment for other health factors and laboratory measurements attenuated the effect of eGFR, but not the effect of proteinuria. Limitations: Follow-up limited to 2 years, selection bias due to nonparticipation in study, missing data for potential covariates, and single noncalibrated measurements from multiple laboratories. Conclusions: The study indicates that community-dwelling older people who have dipstick-positive proteinuria and/or eGFR 30 mL/min/1.73 m are at increased risk of hospitalization. Am J Kidney Dis. 57(5):664-672. © 2011 by the National Kidney Foundation, Inc.

2][3] There are few data about the overall effects of CKD on morbidity and overall health outcomes in the general population.A large study in the United States has shown that for those who are insured, individuals with CKD appear to be at higher risk of later hospitalization. 2However, such a study has a number of limitations.In particular, assessment of kidney function was not available for the entire population, but was measured selectively in people who had a clinical indication.This means that people included in the study potentially were unrepresentative of the population, making results difficult to generalize.In addition, such database studies have limited data for confounding and explanatory factors.
In the United Kingdom, there are limited data for hospital admissions for CKD.Using a large study of older people living in the community, we previously have shown that CKD is associated with a high burden of comorbid conditions 4 and higher risk of death that was independent of concurrent comorbid conditions and conventional cardiovascular risk factors. 1 In the present article, we report results for the association of CKD with rate of hospital admission.

Study Design
We used data from a cluster randomized trial of older people in the setting of general practice (the Medical Research Council [MRC] Study of Multidimensional Assessment of Older People).The study protocol and main results have been reported previously. 5,6In brief, this trial compared 2 methods of multidimensional assessment (universal vs targeted assessment) in people 75 years and older registered in 106 general practices from the MRC General Practice Research Framework in England, Wales, and Scotland selected to be representative of the UK general practice standardized mortality ratios and Jarman deprivation score (categorized into low, middle, and high scores). 7All patients 75 years or older registered with the practices were eligible and invited to participate unless they were resident in long-stay hospitals or nursing homes or had a terminal illness.The trial was approved by relevant local ethics committees.This report uses data from the 53 practices in the universal arm of the trial because in this arm, all patients were offered an in-depth health assessment, including a routine blood test.In the universal arm, 15,336 of 20,934 (73.2%) participants attended for the assessment; nonresponders were older and more likely to be women. 5

Data Collected at Assessment
Assessments were conducted in 1994-1999 and carried out by nurses trained in the study methods and assessments following a structured questionnaire and protocol. 5atients' height, weight, waist and hip circumferences, and blood pressure (average of 2 measurements each) were measured.A nonfasting blood sample was obtained for a biochemical screen that included serum creatinine, urea, potassium, albumin, calcium, phosphate, bilirubin, alkaline phosphatase, aspartate aminotransferase, and urate and a full blood cell count (hemoglobin, white blood cell count, and platelets).Urine dipstick for protein, glucose, and blood was performed, and if positive for protein, a midstream urine sample was obtained.Sociodemographic information, self-reported medical history, lifestyle, and medication data were obtained using nurse interview.Medication data were derived from participants bringing their medicines to the assessment, and drugs were coded into broad classes using the British National Formulary chapter headings.Diabetes was classified according to selfreport of a medical diagnosis, use of antidiabetic medication, or the presence of a high random blood glucose level.Participants were asked about alcohol consumption, smoking history, and perception of physical activity.Activities of daily living (ADLs) were categorized by the number of dependencies for 8 activities (washing, dressing, cutting toe nails, cooking, shopping, doing light housework, walking 50 yards, and going up and down stairs and steps).We defined full or partial dependency as being unable to perform 2 or more ADLs.A score Ͻ24 on the Mini-Mental State Examination 8 was considered to indicate cognitive impairment, and a score Ͼ5 on the Geriatric Depression Scale, 9 significant depression.Other variables included self-reported history of cancer, unexpected weight loss of more than half a stone (1/2 stone ϭ 3.175 kg), and history of falls in last 6 months (Ͻ2 vs Ն2).

Kidney Function
Of 45 local laboratories to which serum samples were sent, 37 used the modified Jaffé method and 7 used an enzymatic method for serum creatinine (in 1, the method was unknown).
The CKD Epidemiology Collaboration (CKD-EPI) Study equation 10,11 was used to calculate estimated glomerular filtration rate (eGFR) and categorized into eGFR groups 12 of Ն75, 60-74, 45-59, 30-44, Ͻ30 mL/min/1.73m 2 using nonstandardized serum creati-nine multiplied by 0.95 (this represents the difference between standardized and nonstandardized creatinine in the Modification of Diet in Renal Disease [MDRD] Study laboratory). 13Only those who had dipstick data and no evidence of urinary tract infection on the midstream urine sample were defined as having available urine dipstick data.Dipstick proteinuria was categorized as none/trace versus proteinuria (ϩ, ϩϩ, and ϩϩϩ).

Outcomes
Information for hospital admissions (defined as a stay of at least 1 night) for a 2-year period after the date of assessment was collected from the hospital discharge letter extracted from the practice records.Diagnostic codes for hospital admission were summarized using International Classification of Diseases, Tenth Revision chapter headings as circulatory (I00-I99), cancer or diseases of the blood system (C00-D89), of infectious origin (A00-B99, L00-L08, K65, M00-M03, and J00-J22), or other causes of hospital admissions.Multiple causes were allowed.Study participants were registered with the Office for National Statistics for mortality follow-up (date and cause of death).

Data Analysis
Data analyses were performed using Stata, version 11 (www.stata.com).Crude associations for eGFR and dipstick proteinuria with baseline criteria (using 2 tests and tests for trend as appropri- ate) and for hospitalization rates (with corresponding 95% confidence intervals [CIs]) were calculated.We censored participants at death (if it occurred outside the hospital) or the end of the 2-year follow-up after the health assessment.
A Cox proportional hazards model was used to model the outcome of time to first hospitalization after the baseline assessment.The proportional hazards assumption of the Cox hazard model did not hold for the entire follow-up of 2 years because there was a time-varying effect of eGFR, age, and sex, which changed rapidly within the first few months.In other words, there were strong selection effects over time that led to a changing hazard ratio (HR) dependent on the time of follow-up.Follow-up time therefore was divided into 2 periods (or time bands) for each patient: the first 6 months after the baseline assessment and the period from 6 months to 2 years (the end of follow-up).Cox proportional hazard models were fitted separately for the time from the start of the study up to 6 months and separately from 6 months to 24 months for those who were not hospitalized within 6 months and were alive (with baseline factors and measurements carried forward to the 6-month start).
All models were adjusted for the effect of age using 4 strata with cutoff points at ages 80, 85, and 90 years (model 1 in tables).We tested for interactions between eGFR and dipstick proteinuria and for eGFR and sex by fitting the respective interaction terms in the age-adjusted models and performing Wald test for exclusion of all interaction terms (in both the 6-month and 6-to-24 month follow-up cohorts).Model 2 adjusted for socioeconomic status, comorbid conditions, cardiovascular risk, or use of cardiovascular drugs.Variables were Jarman score, smoking status (current, ex-, or never smoker), alcohol use (never , ex-, and current drinker), self-reported physical activity (very active, fairly, not very, and not at all), waist-to-hip ratio (sex-specific quintiles), and comorbid conditions using self-reported history of cardiovascular disease (CVD; heart attack or stroke) and diabetes.Hypertension was modeled separately as physician-diagnosed hypertension, average blood pressure Ͼ140/90 mm Hg, or antihypertensive drug use.Cardiovascular drugs were statins or aspirin.We adjusted for angiotensin-converting enzyme-inhibitor and angiotensin II receptor blocker use separately from other antihypertensive agents.We ran models 1 and 2 separately for both eGFR and dipstick protein-uria and further ran models with both eGFR and proteinuria included.
We then added laboratory measurements (model 3) and other health measures (model 4) to investigate their role in the observed associations.Hemoglobin and phosphate levels were categorized as quintiles for each sex, and albumin, as quintiles for each sex and assay type.Other health measures were Mini-Mental State Examination scores (Յ23 vs Ͼ23), Geriatric Depression Scale score (Յ5 vs Ͼ5), overall health perception (poor vs not poor), and ADLs.
For those hospitalized, we derived separate dummy variables for the presence of infectious, cancer, cardiocirculatory, or other causes for hospitalization.Logistic regression analyses were carried out for the odds for specific causes of hospitalization (relative to not having this cause) as a function of measurements preceding that hospitalization (adjusted for age and sex).Secondary analysis was conducted for the association of CKD with total number of admissions (categorized as 0, 1, or Ն2) during follow-up using a multinomial logistic regression.In all models, robust standard errors were calculated to account for the study design of 53 practices from which participants were recruited.

Crude Associations With Subsequent Hospitalization
For 13,177 of 15,336 (86%) participants who completed the in-depth assessment, eGFR was calculated.Missing data included patient refusal of phlebotomy, poor veins, lost blood sample, or unknown.During the 2-year follow-up, 3,291 of 13,177 (25%) participants with eGFR data had at least 1 hospital admission; 2,310 (17%) had only 1 admission and 981 (7%) had 2 or more admissions.For those hospitalized at least once, the next admission occurred within a median of 98 (25th-75th percentile, 42-225) days.There were 12,371 participants who had both eGFR and dipstick proteinuria data.Six participants died on the day of admission to the hospital, and 2,279 died after being admitted to the hospital.There were 999 patients who died within 2 years of follow-up without entering the hospital; these were censored for the analysis at their death date.
Selected baseline characteristics and their associations with eGFR and proteinuria are listed in Table 1.Associations of baseline characteristics with subsequent hospital admissions are listed in Table 2. Hospitalization rates increased with increasing age, and men were more likely to be hospitalized than women.When investigating causes of admissions, 23.2% of all admissions were for circulatory reasons, 14.2% had infections as a contributing cause, and 11.6% had cancer or blood-related diseases as a contributing cause.
Subsequent analyses listed in Table 4 were based on people with complete information for confounding variables (n ϭ 10,977); results for analyses with all data with varying totals for each model are very similar (data not shown).Adjusting for age and sex of participants, we found a strong effect of eGFR Ͻ30 mL/min/1.73m 2 , which was stronger in the first 6 months of follow-up compared with the subsequent 18 months (model 1).In age-adjusted analysis, there was no evidence for an interaction of eGFR and sex in up to 6 months' follow-up (P ϭ 0.7) and during the subsequent 18 months of follow-up (P ϭ 0.8).The association of eGFR Ͻ30 mL/min/1.73m 2 with hospitalization attenuated, but remained significant, when adjusting further for all cardiovascular risk factors and underlying CVD, as well as Jarman score (model 2).HRs for eGFR Ͻ30 mL/min/1.73m 2 during the first 6-month period were confounded weakly by dipstick positivity.There was no evidence for an interaction of eGFR and dipstick positivity in up to 6 months of follow-up (P ϭ 0.7) and during the subsequent 18 months of follow-up (P ϭ 0.6).There was no evidence of time-dependent effects of dipstick positivity, and the age-and CVD risk-adjusted HR was 1.29 (95% CI, 1.11-1.49)for the total 2-year follow-up.

Associations of eGFR and Proteinuria With Number of Hospitalizations
Compared with eGFR of 60-74 mL/min/1.73m 2 , eGFR categories of 30-44 and Ͻ30 mL/min/1.73m 2 were associated with increased ORs for 2 or more hospitalizations during the 2-year follow-up.For those with eGFR Ͻ30 mL/min/1.73m 2 in particular, there was a more than doubled OR (Table 5).Adjustments for cardiovascular risk factors at baseline attenuated associations of eGFR with number of hospitalizations, with an increased OR remaining for only eGFR Ͻ30 mL/min/1.73m 2 and 2 or more admissions.Dipstick-positive proteinuria (not adjusted for eGFR) was associated with multiple hospitalizations during follow-up; adding potential cardiovascular confounding variables did not attenuate the association appreciably.In a model with both proteinuria and eGFR, we found that both dipstick-positive proteinuria and eGFR Ͻ30 mL/min/1.73m 2 were associated independently with the odds of multiple hospitalizations during the 2-year follow-up, even after adjustment for CVD (Table 5).

DISCUSSION
Our results show that dipstick-positive proteinuria is associated with an approximate 30% increased risk of single and multiple hospitalizations during the 2 years after measurement.We found a strong association of eGFR Ͻ30 mL/min/1.73m 2 with the shortterm incidence of hospitalization (Ͻ6 months) and a 50% increase in odds of more than 1 admission.This finding agrees with previous studies that examined only eGFR. 2,14Our results indicate the potential importance of dipstick testing and eGFR measurement in the early identification of older people who are at risk of hospitalization.Other studies in the United States using health insurance claims data have found broadly similar results of an association of eGFR with risk of subsequent hospitalization. 2,15However, many people with less severe degrees of CKD are not identifiable in US claims databases, limiting their utility. 16In the United Kingdom, to our knowledge, no community-based study has investigated whether CKD increases the risk of hospitalization.
It is unclear exactly what explains the associations found.Rate ratios for eGFR across the total follow-up a Trend in rate ratio.b Midstream urine positive.c Sex-specific.d Sex-and assay-specific.For albumin, 39 of 45 laboratories used bromocresol green, the rest used bromocresol purple, and quintiles were derived by assay and sex.Note: Values shown as rate ratio (95% confidence interval).Mantel-Haenszel tests for interaction between sex and eGFR and between eGFR and proteinuria were nonsignificant.Proteinuria presence was assessed using dipstick.
a In addition to age and sex, model 2 was adjusted for Jarman deprivation score (low, middle, and high), smoking status (current, ex-, or never smoker), alcohol (never, ex-, and current drinker), self-reported physical activity (very active, fairly, not very, and not at all), waist-to-hip ratio (sex-specific quintiles), and comorbid conditions using self-reported history of CVD (heart attack or stroke) and diabetes.Hypertension was modeled separately as physician-diagnosed hypertension, average blood pressure Ͼ140/90 mm Hg, or use of antihypertensive drugs.Cardiovascular drugs were statins or aspirin.We adjusted for angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers separately from other antihypertensive agents.
b eGFR adjusted for proteinuria.c Dipstick-positive proteinuria adjusted for eGFR.
were less marked for men than women (Table 3); however, these rate ratios conceal variations in the pattern of HRs during follow-up.Men, those who were older, and those with lower eGFRs had higher admission rates during the first few months of followup.For this reason, we split the time into a short-(up to 6 months) and long-term risk period (6-18 months).Adjustment for concurrent CVD risk factors attenuated the associations of eGFR Ͻ30 mL/min/1.73m 2 for the first 6 months of follow-up, suggesting that cardiovascular risk factors may explain some (but not all) associations.Adjustment for hemoglobin, phosphate, or albumin levels led to little attenuation of the association of eGFR with time to hospitalization; in contrast, there was more attenuation of the association of eGFR with number of hospitalizations.Further adjusting for health factors that indicate other aspects of physical and mental health almost completely attenuated the association of eGFR Ͻ30 mL/min/1.73m 2 with either rate or number of admissions.It therefore is possible that some of the association of eGFR Ͻ30 mL/min/1.73m 2 may be mediated through factors related to other health problems.No such attenuation was observed for effects of dipstick proteinuria.
Our study derives from a representative sample of the UK community-dwelling older population, with systematic testing of serum creatinine and dipstick proteinuria at baseline and systematic follow-up for hospitalization during 2 years after these measurements.Our findings are not applicable to people living in nursing homes.Competing risks may have led to underestimation of the true effect of eGFR and dipstick proteinuria on hospitalization.A quarter of participants died during the follow-up period; 7% died without being hospitalized, and 18%, with at least 1 admission.Because both low eGFR and proteinuria are associated with higher risk of death, the risk of hospital admission or multiple admissions in those who died is different from those with low eGFR who did not die (competing risks).We used Cox regression rather than Poisson regression because of the limitations of Poisson.A Poisson approach would have assumed: (1) a constant rate of hospitalization during a given observation period (and thus ignored the issue of removal of those at highest risk of the hospitalization or at risk of death during follow-up), and (2) independence of risk of subsequent hospitalization from having had previous hospitalizations.
We were able to collect hospital admission data in only the first 2 years of follow-up and therefore our study does not provide evidence for longer term risks of hospital admission.Measurement errors may have led to underestimation of associations of eGFR or dipstick proteinuria There may be some interlaboratory variation in the creatinine measurement method, 17 Note: N ϭ 10,977.Values shown as odds ratio (95% confidence interval).eGFRs are given in mL/min/1.73m 2 .Proteinuria assessed as dipstick positivity.
a In addition to age and sex, model 2 was adjusted for Jarman deprivation score (low, middle, and high), smoking status (current, ex, or never smoker), alcohol (never, ex-, and current drinker), selfreported physical activity (very active, fairly, not very, and not at all), waist-to-hip ratio (sex-specific quintiles), and comorbid conditions using self-reported history of CVD (heart attack or stroke) and diabetes.Hypertension was modeled separately as physician-diagnosed hypertension, average blood pressure Ͼ140/90 mm Hg, or use of antihypertensive drugs.Cardiovascular drugs were statins or aspirin.We adjusted for angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers separately from other antihypertensive agents.
b eGFR adjusted for proteinuria.c Adjusted for eGFR.
introducing random variation in serum creatinine and eGFR values.However, at high creatinine levels, there is less variation 17 and estimation formulas perform better, 18 which means that errors associated with risk estimates for eGFR Ͻ30 mL/min/1.73m 2 are likely to be minimal.Similarly, dipstick proteinuria readings are variable and less precise than urinary protein or albumin-creatinine ratios or 24-hour urine collections. 19,20Use of a midstream urine sample and exclusion of patients with culture-positive urine may have partially compensated for errors in dipstick proteinuria.Of 20,934 older people invited to join this study, 15,336 participated, of whom 12,371 had data for both dipstick protein and eGFR.Complete data for all potential confounders were available for only 10,799.Hence, there remains the possibility of selection bias in the fully adjusted analysis.However, analyses of the larger sets of data (12,371) gave virtually identical results.We cannot exclude the possibility of confounding caused by unmeasured factors, but this is a well-characterized data set that enabled us to adjust for an extensive range of confounding variables, including socioeconomic deprivation.
Our results show that both eGFR Ͻ30 mL/min/1.73m 2 and dipstick proteinuria are associated with increased risk of subsequent hospitalization.We found no association at higher eGFRs or for those with eGFR Ͼ75 mL/min/1.73m 2 .There is considerable interest in minimizing hospital admissions for older people. 21Our results show that decreased kidney function, particularly in the presence of proteinuria, identifies older people at high risk of subsequent hospital admission.

Table 1 .
Baseline Characteristics for 12,371 Participants With Both eGFR and Urine Data

Table 1 (
Cont'd).Baseline Characteristics for 12,371 Participants With Both eGFR and Urine Data

Table 2 .
Rates of Subsequent Hospital Admission According to Selected Participants' Baseline Characteristics

Table 2 (
Cont'd).Rates of Subsequent Hospital Admission According to Selected Participants' Baseline Characteristics

Table 3 .
Age-Adjusted Ratios of Hospital Admission Rates Across Categories of eGFR in Men and Women Stratified byPresence of Proteinuria

Table 4 .
Effects of Sequential Adjustments in Complete-Case Subset of Data on Associations of eGFR and Proteinuria WithSubsequent Hospitalization

Table 5 .
Effects of Sequential Adjustments in Complete-Case Subset of Data on Association of eGFR and Proteinuria (modeled separately) With Hospital Admissions During 2-Year Follow-up