Problem A significant proportion of patients with diabetes mellitus do not get the benefit of treatment that would reduce their risk of progressive kidney disease and reach a nephrologist once significant loss of kidney function has already occurred.
Design Systematic disease management of patients with diabetes and kidney disease.
Setting Diverse population (approximately 800 000) in and around Birmingham, West Midlands, UK.
Key measures for improvement Number of outpatient appointments, estimated glomerular filtration rate (eGFR) at first contact with nephrologist, number of patients starting kidney replacement therapy (KRT) and mode of KRT at start.
Strategy for change Identification of patients with low or deteriorating trend in eGFR from weekly database review, specialist diabetes–kidney clinic, self-management of blood pressure and transfer to multidisciplinary clinic >12 months before end-stage kidney disease.
Effects of change New patients increased from 62 in 2003 to 132 in 2010; follow-ups fell from 251 to 174. Median eGFR at first clinic visit increased from 28.8 ml/min/1.73 m2 (range 6.1–67.0) in 2000/2001 to 35.0 (11.1–147.5) in 2010 (p<0.006). In 2010, the number of patients starting KRT fell 30% below the projected activity using 1993–2003 data as baseline (p<0.003). The proportion starting KRT with either a kidney transplant, peritoneal dialysis or haemodialysis via an arteriovenous fistula increased from 26% in 2000 to 55% in 2010.
Lessons learned Systematic disease management across a large population significantly improves patient outcomes, increases the productivity of a specialist service and could reduce healthcare costs compared with the current model of care.
- Chronic disease management
- quality improvement methodologies
- diabetes mellitus
- healthcare quality improvement
- statistical process control
Statistics from Altmetric.com
- Chronic disease management
- quality improvement methodologies
- diabetes mellitus
- healthcare quality improvement
- statistical process control
The National Health Service is the sole provider of kidney replacement therapy (KRT) for patients developing end-stage kidney disease (ESKD) in the West Midlands region of England. The Heart of England NHS Foundation Trust kidney service (HEKS) serves the population of east and north Birmingham (438 000), Solihull (220 000) and South Staffordshire (150 000). The population of east Birmingham has a 16.9% South Asian ethnicity and 4.1% African-American ethnicity (England averages=5.7% and 2.8%, respectively) and 60% of the population is in the most deprived national deprivation quintile.1 The incidence of diabetes is in the highest quartile2 and the incidence rate of KRT in east and north Birmingham was 167 per million population in 2008, 1.67 times the expected age- and sex-standardised rate. Solihull and South Staffordshire have a lower-than-average percentage of ethnic minorities and level of deprivation, and have an incidence of diabetes and KRT equal to the England average.3
Outline of the problem
ESKD has a major impact on survival and quality of life of patients with diabetes. Loss of kidney function due to diabetic nephropathy can be slowed or halted by blood pressure reduction, particularly through the use of ACE inhibitors (ACEIs) and angiotensin II receptor blockers (ARBs).4 A significant proportion of patients with diabetes do not receive effective treatment that would reduce the likelihood of ESKD. The 2008–2009 National Diabetes Audit reported that only 48.7% of patients with type 2 diabetes in England achieved the recommended maximum blood pressure (≤140/80 mm Hg for people with diabetes and ≤130/80 mm Hg for those with diabetes and kidney disease).5 East and north Birmingham, Solihull and South Staffordshire achieved 43.3%, 43.0% and 51.3%, respectively.5 As the early stages of chronic kidney disease (CKD) are asymptomatic, patients with declining kidney function are unaware of the risk to their health unless they are provided with this information in a way they can understand. Early referral to a nephrologist may prevent irreversible kidney damage;6 however, a recent systematic review could find little evidence about how to organise care of patients at high risk of ESKD.7
Key measures for improvement
We investigated whether a new model of care introduced in 2004:
reduced the need for patients to be followed up in hospital outpatient clinics;
enabled more patients to access a nephrologist at an earlier stage of kidney disease;
prevented kidney damage in the population and reduced the number of patients with diabetes reaching ESKD and starting KRT;
enabled a greater proportion of patients who did start KRT to do so via a long-term mode of therapy.
Process of gathering information
All patients starting long-term KRT within the HEKS were included. Diabetes status was identified from coded information and from haemoglobin A1c (HbA1c) and blood glucose values. Diabetes was not necessarily the cause of kidney failure. Patients with acute kidney injury requiring dialysis who recovered sufficient kidney function to stop dialysis or did not survive to leave hospital were excluded.
The effect on kidney function of attending the diabetes–kidney clinic was assessed by plotting a time-dependent graph of all available estimated glomerular filtration rate (eGFR) measurements for patients who first visited the diabetes renal clinic in 2005. The number of measurements varied as these were taken when clinically necessary. As the change in eGFR was often non-linear with short-term fluctuations, fitting a linear regression trendline did not appropriately describe the average long-term trend. Therefore, a straight line was drawn manually to achieve the best estimate of the average long-term slope before and after the first visit to the clinic.
As with many quality improvement projects, we chose to use statistical process control (SPC) methods.8 However, time series data measurements cannot be assumed to be independent. A strong trend over time before the intervention implies autocorrelation, which must be removed in order to assess the impact of the intervention accurately from an SPC chart. We converted such data to a deviation-from-aim chart using a linear forecast model calculated from the pre-intervention data and subtracting the post-intervention measurements from the forecasted values. This method represents an application of the Box–Jenkins time series analysis.9 SPC charts were created using Baseline software (SAASoft, http://www.saasoft.com). The limits of common cause variation over time were calculated from the average of the moving ranges, which are the absolute differences between consecutive measurements.
Analysis and interpretation
Data from the kidney service at the University Hospitals Coventry and Warwickshire (UHCWKS) were used to study the current model of care. UHCWKS serves a population of approximately 800 000 adjacent to the Heart of England service with a similar mix of deprived urban and more affluent suburban and rural communities.5 The proportion of patients of ethnic minority was lower in this population and a slightly greater percentage of patients with diabetes achieved the target blood pressure in 2008–2009 (table 1). The total number, mean age and sex distribution of patients with diabetes starting KRT were very similar in the two services.
The UHCWKS receives referrals from primary care and other hospital clinicians into general kidney clinics and a diabetes–kidney clinic for patients with proteinuria without a specific focus on those with declining eGFR. No change to the model of care was made at UHCW between 1990 and 2010.
As the purpose of the system is to increase the duration of KRT-free survival, the number of patients starting KRT (the ‘start KRT demand’) over time is a measure of the behaviour of the whole system. It is helpful to liken the number of patients reaching ESKD to the distance that the sea reaches up a beach. The distance reached is due to a combination of the tide, which is predictable from past experience, and the waves, which are not predictable. An SPC chart of the number of patients with diabetes starting KRT in the UHCWKS shows that the process is unstable with an apparently linear increase (figure 1A). The goodness of fit to a linear trend over time model was assessed by calculating the difference of each measurement from the trend and plotting a deviation-from-aim chart.9 The deviations are within the limits of common cause variation, so the hypothesis of a linear trend model is not disproved (figure 1B). In other words, the tide is steadily coming in between 1990 and 2010 and this can be subtracted to reveal the behaviour of the waves.
Strategy for change
The new model of care was operated by a nephrologist with an interest in diabetes (HCR). Routine reporting of eGFR10 by the Heart of England pathology laboratory was introduced in January 2004, supported by a CKD education programme for primary care clinicians. Urinary protein test reports included a recommendation for nephrology referral for protein:creatinine ratios >100 mg/mmol from 2009.11
From 1996, all patients attending diabetes clinics at Birmingham Heartlands and Solihull Hospitals were registered on the Proton database (http://www.ccl.com). Results of tests performed on samples taken in the community and hospital were automatically transferred from the laboratory into the database.
From August 2004, the database was interrogated weekly to identify patients not receiving KRT or attending the pre-dialysis specialist clinic who have an eGFR reported in the previous week ≤40 ml/min/1.73 m2 and those aged ≤65 years with an eGFR ≤50 ml/min/1.73 m2. Those with multiple results within the previous 6 days were not reported, as they were most probably hospital inpatients. The age threshold of 65 years restricted the list to a manageable size, focusing on patients at risk of progressing to ESKD.12
The mean number of patients identified from the database per week was 53 (range 26–74: eGFR <30 ml/min/1.73 m2, mean=20; eGFR 30–40 ml/min/1.73 m2, mean=27; aged <65 years, eGFR 41–50 ml/min/1.73 m2, mean=7). An average of 2756 eGFR graphs were reviewed per year. The mean number excluded per week due to having more than one result in the previous week was 9 (range 2–17).
For each patient, a graph of all eGFR values over the previous 10 years was generated from the database and viewed by a nephrologist (HCR). No numerical analysis was applied to the graph. Patients were selected if their eGFR graph suggested a declining trend or a sudden fall outside the usual variation for that patient and the source of the last blood sample was identified. A letter or email including the eGFR graph was sent to the general practitioner or hospital doctor to alert them to the change and advise appropriate action or referral to the diabetes–kidney clinic. The clinician was contacted by telephone if the drop in eGFR was sudden and severe. This surveillance activity took approximately 1 hour per week of the nephrologist's time. Referrals of patients by primary care and other hospital clinicians were also received in the conventional way. If referred patients did not attend, letters of advice were sent to the patient and the referring doctor.
During the diabetes–kidney clinic consultation, the following issues were addressed
The cause of kidney disease, excluding urinary obstruction and non-diabetic nephropathy such as interstitial nephritis and glomerulonephritis by renal biopsy where indicated.
The trend in kidney function and proteinuria, discussing with the patient the graph of eGFR over time.
Dietary salt intake and modification of medication to achieve blood pressure control, adding or increasing diuretic therapy if necessary.
Starting ACEI/ARB therapy where not already prescribed and possibly stopping ACEI/ARB or diuretic therapy to reveal any negative effect on eGFR.
Self-monitoring of blood pressure to improve medication concordance and prompt increases in medication to achieve target.15
Smoking cessation, referring to ‘stop-smoking’ service during the consultation if appropriate.
Following the consultation, a letter was sent to the patient detailing the discussion and care plan and including test results and an updated eGFR graph. Copies of these documents were also sent to the patient's general practitioner.
The patient was reviewed every two to four months in clinic until the home blood pressure was controlled and the eGFR decline halted or slowed when the patient was discharged to be monitored in the general diabetes clinic or primary care. The nephrologist then monitored the eGFR remotely via the weekly database extract.
Patients with declining eGFR were transferred to a multidisciplinary kidney team approximately 1 year before the time the eGFR was projected to reach 10 ml/min/1.73 m2. The team educated and prepared patients for dialysis, transplantation or conservative kidney management.16
Effects of change
Following the introduction of the system in 2004, patients with stable eGFR were discharged and the number of follow-up attendances declined (figure 2). With the addition of referral guidance on urinary protein reports in 2009, more patients were referred with protein:creatinine ratio >100 mg/mmol and normal eGFR. The median eGFR at the first clinic visit increased from 28.8 ml/min/1.73 m2 (range 6.1–67.0) in 2000/2001 to 35.0 ml/min/1.73 m2 (range 11.1–147.5) in 2010, p=0.0056 two-tailed Mann–Whitney U test.
In 66 patients first seen in the clinic during 2005, the mean estimated rate of decline in eGFR changed from −5.2 ml/min/1.73 m2/year prior to attending the clinic to +1.1 ml/min/1.73 m2/year afterwards until death, KRT or up to 4 years of follow-up, p<0.001 paired t test (figure 3).
The number of patients per year who transferred to the multidisciplinary team for preparation for KRT or conservative kidney management fell sharply in 2010 (figure 4).
The number of patients with diabetes starting KRT fell from 2005 (figure 5). A high proportion was from ethnic minorities (62% white European, 29% South Asian and 7% African-American) and the fall was in all ethnic groups (figure 6). The mean age of patients when KRT was started increased between 1990 and 2009 (figure 7A). In 2010, the mean age deviated below the limits of common cause variation of the projection from 1993 to 2003 (p<0.003) (figure 7B).
The main criteria for starting dialysis are the severity of a patient's symptoms and their metabolic abnormalities such as acidosis and uraemia. These become more severe as GFR falls. Change in the criteria for starting dialysis was quantified by plotting the mean eGFR at the start of dialysis over time. The mean eGFR increased until 1997 but was stable thereafter (figure 7A). There was no increase in the number of patients who did not receive dialysis and died due to advanced kidney failure; one patient died with an eGFR <10 ml/min/1.73 m2 without receiving dialysis in 2007, one in 2008, two in 2009 and none in 2010.
The 6-year delay between the start of the new model of care and the significant fall in numbers starting KRT is consistent with the reduced rate of deterioration in eGFR. Patients with an eGFR of 38 ml/min/1.73 m2 declining at 5 ml/min/1.73 m2/year would have started KRT after 6 years without intervention. Risk of death is related to age, so delaying the start of KRT in a cohort of patients exposes the older patients in that cohort to a greater risk of dying before starting KRT than the younger patients. The significant fall in mean age at the start of KRT in 2010 suggests that some older patients avoided the need for KRT prior to death.
A deviation from the aim chart shows a negative run for 6 years from 2005, and the number in 2010 was 14 (30%) lower than the projection of 47, below the lower limit of common cause variation (figure 8). The design of the SPC chart implies that the probability that this change was due to chance variation (type I error) is <3 in 1000 (p<0.003). Using the analogy of the tide and waves, subtraction of the incoming tide demonstrated by the UHCW data reveals that the number of patients in the HEKS receded further than would be expected from a reduction in the size of the waves. In other words, the tide reversed following implementation of the new model of care.
With the increase in median eGFR at first attendance and increased opportunity for timely preparation,17 the proportion of patients starting KRT with a transplant, peritoneal dialysis or haemodialysis via an arteriovenous fistula increased from 26% in 2000 to 55% in 2010 (figure 9).
By focusing on patients at greatest risk and replacing face-to-face consultations with remote reviewing of results, the system is cost effective in its use of specialists' time. Others have shown elements of the model of care to be effective in improving outcomes: a chronic disease programme for patients with CKD stages 4 and 5 based in UK primary care led to a reduction in the rate of decline in eGFR over a 9-month period,18 and a system of specialist care for members of Kaiser Permanente in Hawaii led to an increase in the proportion of patients commencing KRT as an outpatient.19 However, this is the first report of a reduction in the incidence of KRT following the implementation of a disease management system within existing resources.
A reduction in the incidence of KRT provides significant economic benefits. The cost of one patient year on haemodialysis is approximately £30 000 and peritoneal dialysis £15 000 without overheads.20 The reduction of 14 in the number of new patients commencing KRT in 2010, with a ratio of haemodialysis:peritoneal dialysis of 85:15, equates to an annual saving of £390 000. The increase in the proportion commencing KRT via a long-term mode of therapy as an outpatient will lead to lower costs of hospitalisation in the first months of KRT but will also increase length of survival and hence lifetime cost of dialysis.
Implementing the system requires a computerised disease register. Few nephrology centres have a database containing data on patients not under the care of a nephrologist. An alternative would be to analyse data held in laboratory databases.21 22 In the UK, these are located in hospitals and a small number of laboratories serve each local population. Patients with diabetes at high risk of ESKD could be identified from blood glucose, HbA1c, urinary protein and eGFR results.23 Time-related graphs of eGFR could routinely be provided to the requesting clinician and those showing a declining trend could be selected for review by a nephrologist.
It is difficult to predict the reduction in KRT that would be achieved by applying this model across the UK. The database did not include all patients with diabetes in the local population and so the impact may be underestimated. Conversely, the high incidence of KRT in Birmingham and its high ethnic minority population may mean that the reduction achieved was greater, although a preferential impact on one ethnic group was not obvious. Over the long term, the initial reduction in KRT incidence will be balanced against the continuing increase in the number of older people and patients with diabetes.
The authors wish to thank colleagues in the departments of kidney medicine, diabetes, vascular surgery and pathology, and local general practices for their vital contribution to the care of people with diabetes and Dr Shahrad Taheri for his helpful comments on the manuscript.
Funding Funded within current NHS resources.
Competing interest All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organisation for the submitted work, no financial relationships with any organisations that might have an interest in the submitted work in the previous three years and no other relationships or activities that could appear to have influenced the submitted work.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.