Early referral strategies for management of people with markers of renal disease: a systematic review of the evidence of clinical effectiveness, costeffectiveness and economic analysis

Angiotensin converting enzyme inhibitors

In people with non-diabetic CKD, ACE Is have little benefit over placebo or other antihypertensive treatments in reducing all-cause mortality (ACM). 43 A statistically significant reduction in the risk of developing ESRD alone, or doubling of serum creatinine and ESRD [relative risk (RR) reduction of 40% and 30% respectively] was observed from a meta-analysis of 11 trials and after adjustment for difference in study baseline characteristics. 43 The risk reduction was modified by, but independent of, change in BP and proteinuria. A benefit on renal outcomes from ACE Is was observed for those with proteinuria above 0.5 g/day. Below 0.5 g/day, a benefit could not be excluded but the findings were less robust.

Even in diabetics with CKD, ACE Is had no benefit on ACM44 unless patients were treated at maximum tolerable dose (RR 0.78, 95% CI 0.61 to 0.98),44 ACE Is reduced the risk of ESRD by 31% in diabetics when compared with placebo. 44 ACE Is can prevent the progression of micro- to macroalbuminuria (reported RR reduction of 55–65% versus placebo) in diabetic renal patients. 44,45 Regression from micro- to normoalbuminuria was also increased (RR 3.06, 95% CI 1.76 to 5.35). 44 When analysis was restricted to those with proteinuria, the findings were similar, with no evidence of benefit from ACE Is (as compared with placebo) on mortality, but a statistically significant effect on reduction of ESRD or doubling of serum creatinine (RR 0.60, 95% CI 0.49 to 0.73). 45

Angiotensin receptor blockers

Similar to ACE Is, ARBs have no significant impact on ACM in diabetics with CKD when compared with placebo or standard antihypertensive agents,44,46 or CVD morbidity and mortality. 46 However, there was evidence of a reduction in progression of CKD,44 with statistically significant risk reductions for ESRD (22%) and doubling of serum creatinine (21%). ARBs reduced progression from micro- to macroalbuminuria by more than 50% as well as significantly increasing the numbers returning from micro- to normoalbuminuria. Regarding adverse effects, the only significant increase observed was in hyperkalaemia.

Strippoli and colleagues44 identified three studies that compared ACE Is with ARBs. They found no evidence of a difference in effect on mortality or renal outcomes.

A systematic review by Casas and colleagues47 reported on effects of ACE Is or ARBs on renal outcomes, compared with placebo or other active interventions. Evidence of statistically significant risk reductions for ESRD was observed (RR 0.87, 95% CI 0.75 to 0.99). Small reductions in creatinine concentration (mean difference –7.07 µmol/l, 95% CI –13.26 to –0.88) and urinary albumin excretion (mean difference –15.7 mg/day, 95% CI –24.73 to –6.74) were reported, but significant heterogeneity among studies was noted. No significant difference was observed in change of GFR or on the composite end point of doubling of creatinine and ESRD. ACE Is/ARBs had no significant effects on GFR, ESRD or doubling of creatinine in diabetics. However, small reduction in urinary albumin excretion was observed in those with diabetic kidney disease (mean difference –12.21 mg/day, 95% CI –21.68 to –2.74). Casas and colleagues47 noted that larger studies were more likely to report smaller benefits, suggesting publication or other source of small study bias.

In people with CKD and diabetes with microalbuminuria, NICE and SIGN recommend treatment with ACE Is or ARBs regardless of BP. In non-diabetics with CKD, hypertension should be controlled using the range of available antihypertensives. However, both guidelines recommend that in the presence of proteinuria and CKD, hypertension should be managed first line with ACE Is/ARBs. 23,42

Data sources and search strategy

Sensitive electronic searches were undertaken to identify studies comparing early referral to other care options for people with CKD. Initial searching was undertaken between January and April 2008, with the main search updated in February 2009. Electronic searches were restricted to reports published in the English language since 1990. We searched for meeting abstracts from January 2006 only. In addition, reference lists of all included studies were scanned to identify additional potentially relevant studies. The search strategy is summarised in Appendix 1 and included clinical and cost effectiveness studies. Search terms did not restrict based on timing of referral; studies of early or late referral were identified.

The following databases were searched:

• Ovid MEDLINE, 1950 to 4 February 2009

• Ovid MEDLINE In-Process and Non-Indexed Citations

• EMBASE, 1988 to week 5 2008

• Science Citation Index, 4 February 2008

• ISI Proceedings, 4 February 2008

• British Nursing Index, 1994 to January 2008

• British Nursing Index Archive, 1985–96

• Health Management Information Consortium, January 2008

• Cumulative Index to Nursing & Allied Health Literature, 1982 to week 1 December 2007

• Social Science Citation Index, 4 February 2008

• Cochrane Central Register of Controlled Trials: Cochrane Library Issue 1 2008

• Cochrane Database of Systematic Reviews: Cochrane Library Issue 1 2008

• National Research Register Archive (up to October 2007), 19 March 2008

• The UK Clinical Research Network, 19 March 2008

• The European Dialysis and Transplant Nurses Association/European Renal Care Association 35th International Conference 2006 and 36th International Conference 2007 Final Programmes.

Additional searching was performed in NHS Economic Evaluation Database to support the cost-effectiveness literature review.

Study selection

Initial searching indicated that we would find very few randomised controlled trials (RCTs) of early referral strategies versus standard care. We therefore considered evidence from any study design that compared a strategy for early referral with a relevant comparator group. We included prospective and retrospective study designs. The research focused on adults with markers for early renal disease in either primary or secondary care. We included any intervention that aimed to achieve the early referral of people with markers of renal disease to specialist nephrology care. For retrospective studies, the definition of early referral was taken as referral at least 12 months before RRT. For prospective study designs, referral prior to reaching stage 5 CKD was required but other definitions of early referral were accepted (e.g. based on proteinuria thresholds in diabetic nephropathy). In the absence of a ‘gold standard’ for care, acceptable comparators included usual care, later referral (defined as less than 12 months prior to RRT for retrospective studies) or primary care. Outcomes of interest included:

• renal function

• onset of RRT

• quality of life

• ACM

• CVD mortality

• hospitalisations

• emergency dialysis

• survival on dialysis.

Details of the care package were also sought.

Titles, abstracts and keywords were reviewed by two systematic reviewers, independently, to identify studies that met the inclusion criteria outlined above. Full papers were then considered for inclusion. Discrepancies were resolved by discussion, with involvement of a third reviewer if necessary.

The search identified that there was a substantial evidence base on the effect of late referral (defined as less than 12 months prior to RRT) and, where studies included mortality as an outcome, this has been included in Appendix 5. A supplementary search for literature about potential barriers to early referral was conducted using MEDLINE (1996 to week 2 April 2008) and EMBASE (1996 to week 15 2008).

Quality assessment of included studies

Two systematic reviewers independently assessed the methodological quality of the included studies. The quality of included studies was assessed using the criteria outlined in the Centre for Reviews and Dissemination Report 4. 58 Checklists were adapted from those developed by NICE. 59

Data extraction

Data extraction from the included studies was carried out independently by two systematic reviewers. Study characteristics, outcome results and aspects of study quality were collected using a standardised form (see Appendix 2). Any discrepancies were resolved by discussion and, where necessary, by involvement of a third reviewer (two occasions relating to terminology).

Data synthesis

Study characteristics and results were tabulated; analysis was qualitative. It was not possible to pool results for quantitative analysis because of the heterogeneity of study characteristics and the diversity in reporting of outcomes.

Study selection

Our inclusion criteria for formal appraisal were studies assessing the costs and consequences (long- or short-term) of early referral strategies for individuals with markers of renal disease. Early referral was defined as referral to a specialist prior to stage 4 CKD (i.e. eGFR higher than 30 ml/min/1.73 m²). We were not explicitly interested in extracting data from studies assessing the cost-effectiveness of population screening for proteinuria, but studies that looked at screening in non-diabetic populations were obtained to help inform the development of the economic model. Finally, studies assessing the cost-effectiveness of single interventions in individuals with CKD, such as the use of ACE Is to slow progression, were excluded from formal appraisal.

Quality assessment of included studies

Studies meeting our inclusion criteria were appraised using the British Medical Journal guidelines for reviewers of economic evaluations. 61 The quality of included decision models was also assessed against a published checklist for good practice in decision analytic modelling in health technology assessment. 62

Data extraction

Plans were made to formally extract data from studies directly assessing the cost-effectiveness of nephrology referral for individuals with stage 3 disease compared with referral at stage 4 or 5 (current practice).

Natural history

A systematic literature review was conducted to identify relevant literature on the natural history of CKD. MEDLINE (1950 to March week 2 2008) and EMBASE (1996 to March week 4 2008) were searched. An internet search (Google scholar; http://scholar.google.co.uk/) was performed and potentially relevant studies were identified from clinical experts. Searches were restricted to the English language and were from 1998. Bibliographic searching of any included study was also undertaken (see Appendix 1 for the search strategy).

Systematic reviews, meta-analyses, observational cohorts and follow-up studies (prospective or retrospective) of adult populations with at least 2 years’ of follow-up were considered. RCTs were excluded owing to the highly selected nature of their participants. Those studies where the main study populations were defined as having CKD were included. However, studies with broader participant inclusion criteria (diabetes, CVD or hypertension) were considered for inclusion when participants with CKD accounted for a substantial number of the total study population. Studies with less than 100 subjects were excluded. While a definition of CKD based on the KDOQI classification was considered to be optimal, the definition of CKD used by the authors was accepted.

Included studies were quality assessed. Systematic reviews and meta-analyses were evaluated based on a methodology checklist outlined by NICE. 63 Other study designs were evaluated based on a quality assessment tool adapted from various methodological criteria. 39,64–67

Care model

Database searching was carried out in MEDLINE (1950 to present) and EMBASE (1996 to 2008 week 25) combining keywords and medical subject heading (see Appendix 1). Internet and bibliographic searches of included studies were performed. No relevant studies were identified from the Cochrane review database and Health Technology Assessment (HTA) journal database of published reports. Any full papers or abstracts published in English (evaluation, audit, description of care) that reported a model of care for the management of CKD patients were included. Care models of management of RRT patients were excluded.

ACM risk in those with CKD compared with no CKD

Eriksen and Ingebretsen96 reported a hazard ratio (HR) of 1.25 (95% CI 1.14 to 1.37) for each 10-ml/min/1.73 m² decrease in eGFR. Five primary studies reported a small increased risk of mortality (risk ratios ranged from 1.12 to 1.78) for people with CKD as compared with those without CKD (Figure 4). 74,76,81,85,92 These studies were based on general populations undergoing health screening76,81,85 or clinical record database review,92 and all but one92 adjusted for comorbidities, age and sex. Herzog and colleagues92 adjusted for comorbidities only. A further two population health screening studies, comparing ACM in CKD cohorts with the population as standardised mortality rates (SMRs), noted higher risks of mortality with SMR 2.2 (95% CI 2.1 to 2.4)96 and SMR 8.3 (95% CI 7.5 to 9.2),98 but did not adjust for comorbidity (Figure 4). Evans and colleagues98 included a cohort of participants with marked renal impairment with a serum creatinine level of at least 250 µmol/l. John and colleagues104 reported age and sex-adjusted SMRs for a CKD cohort compared with the general population, based on laboratory data (SMR 1.53, 95% CI 1.44 to 1.62).

FIGURE 4.

Risk of all-cause mortality. Comparing those with CKD with those without CKD. ^aThree studies reported by Tonelli 200668 as ‘general populations’ with adjusted analysis; Sihvonen 2004115 included only people with rheumatoid arthritis so recoded as ‘clinical population’ in this review. Clin, clinical; CVD, cardiovascular diseases; DM, diabetes mellitus; F, female; FHS, Framingham Heart Study; Gen, general; HBP, high blood pressure; HR, hazard ratio; M, male; NHANES I, National Health and Nutritional Survey I; Pop, population; RR, relative risk; SMR, standardised mortality rate.

Tonelli and colleagues68 reported an unadjusted increased risk of death among those with CKD compared with no CKD in 93% of the 37 studies included in their meta-analysis (RR range 0.94–5.00); however, significant heterogeneity was noted. From a series of meta-regression analyses, authors reported that a greater risk associated with CKD was observed in younger patients, women and studies in ‘general populations’. In nine studies classed as including a ‘general population’, the risk for mortality in the CKD group was threefold higher (RR 3.0, 95% CI 2.18 to 4.11), but again substantial heterogeneity was observed. Studies based on population health survey data (NHANES I114 and II,70 FHS73) reported lower HR/RR of 1.38–2.26. One population-based heart risk factor study in North America (ARIC71) reported higher risk estimates (3.54); three studies included populations known to be at high risk of mortality and CKD (people over 65 years, aboriginal Australians, clinical records based cohort). Adjusted analysis for general population studies (possible for three studies) was also presented by Tonelli and colleagues68 and this reduced heterogeneity as well as reducing the risk estimate to close to 1 for all but one of the studies. Further details are given in Appendix 5.

Two studies reported adjusted risk of ACM by CKD stage (Table 5). 75,85 Go and colleagues,75 studying a clinical population, reported an increasing risk with stage from HR 1.2 for stage 3a to 5.9 for stage 5 (compared with no CKD). Astor and colleagues,85 studying a general population, also reported an increase in risk with stage.

ACM rate in populations with CKD

Four studies reported ACM per 10,000 person-years. 74–76,85 All reported death rates among those with CKD that were greater than for those with no CKD (definitions varied) (Table 6). Death rate varied substantially between studies; Wiener and colleagues74 reported the highest rate of 3080 deaths per 10,000 person-years, considerably more than any of the other studies. This study also reported a high death rate (980 per 10,000 person-years) in the population with no CKD. There was evidence that death rate increased with the degree of proteinuria85 and stage,75 and was higher in men than in women. 76

Seven primary studies presented cumulative mortality for a specified time period. 81,83,86,92,96,98,105 At 1 year, mortality varied from only 3%98 to 15%. 105 At 5 years, 31–39% died. 96,98 For studies reporting 9.0–12.5 years’ follow-up, cumulative mortality ranged from 19.7% to 52.0%. 81,83,86,96 Five other studies presented the proportion of deaths in a given mean follow-up time and thus were not directly comparable. 68,76,84,101,104 Drey and colleagues95 reported an exceptionally high proportion of deaths (69%) in mean 5.5 years. The results of these are summarised in Table 6.

Four primary studies reported the proportion of deaths for different stages of CKD. 75,83,106,111 The proportion of people dying was observed to increase with declining eGFR.

CVD risk in those with CKD compared with no CKD

As for ACM, most studies observed a higher risk of CVD deaths or events in those with CKD as compared with people without CKD. As shown in Figure 5, the adjusted risk estimates for general population studies were reasonably consistent (RR 1.14–1.81) with the exception of studies by Astor and colleagues85 (RR 2.12, 95% CI 1.65 to 2.73), Weiner and colleagues74 (HR 1.09, 95% CI 0.91 to 1.29) (pooled analysis of four population-based health studies in the USA), and Tonelli and colleagues. 68 Across the 14 studies included in a review by Tonelli and colleagues68 that reported on CVD mortality and morbidity, the unadjusted RRs ranged from 1.43 to 3.73. The risk for CVD mortality was 2.47 (95% CI 1.42 to 4.30) in the CKD group as compared with the non-CKD population for three general population studies, but substantial heterogeneity was observed. 68 The authors described adjusted analysis for all 14 studies being particularly sensitive to age, with a greater risk associated with CKD in younger people (no details in paper). Di and colleagues69 also reported an increased risk of CVD morbidity and mortality (RR 1.41, 95% CI 1.19 to 1.68) from the meta-analysis of seven general population studies, with statistically significant heterogeneity observed (p = 0.045).

FIGURE 5.

Risk of cardiovascular morbidity and mortality. Comparing those with CKD with those without CKD. CVD, cardiovascular diseases; DM, diabetes mellitus; F, female; Gen, general; HBP, high blood pressure; HR, hazard ratio; M, male; Pop, population; RR, relative risk; SMR, standardised mortality rate.

Two studies, reporting the risk of CVD events associated with different stages of CKD, reported an increase with decreasing eGFR (Table 7).

CVD rate in populations with CKD

The rate of CVD deaths or events was reported in five studies and varied between studies by population, age, sex and CKD stage. 32,74–76,85 Rates of CVD events and deaths more than doubled from stage 3a to stage 3b.

Cardiovascular disease morbidity and mortality was reported for follow-up periods that ranged from a median of 23 months84 to 12.5 years. 76 Frequency of CVD death was low in the study by Levin and colleagues,84 where study participants were selected because they were thought to be clinically healthier. Proportions experiencing CVD death or a CVD event increased with CKD stage and were higher in men than in women. 32,75 Summarised results of the CVD rates and proportions experiencing CVD morbidity and mortality are presented in Table 8.

CKD with diabetes

The results from cohorts that included only people with diabetes mellitus are summarised in Appendix 6. In studies with at least 6–10 years’ follow-up, mortality was reported as ranging from 19.7%79 to 35.0%88 with cardiovascular deaths accounting for a substantial proportion of this. GFR declined at between 3.8 and 4.5 ml/min/1.73 m²/year and, after 6.5 years, 7% of the study population was reported to have ESRD. 88 Three studies presented data that compared diabetics with non-diabetics. 101,105,111 Jones and colleagues105 observed that diabetic nephropathy was a predictor for having ‘fast progressive’ disease (eGFR decline of > 5 ml/min/year). Hemmelgarn and colleagues101 showed that CKD progression increased as the stages increased, with higher progression in late stages (CKD stages 3 and 4). Relative to the non-diabetic population, Orlando and colleagues111 found a higher risk of progression among those with diabetes in stage 1–2 (RR versus no diabetes mellitus 1.32, 95% CI 1.13 to 1.55), but no difference in progression in higher stages.

CKD with cardiovascular disease and hypertension (HBP)

While many studies included people with CVD and hypertension (HBP), few presented data on outcomes stratified by these subgroups. One study compared the risk of CKD progression for different CKD stages for those with CVD at baseline with those without CVD. 111 Compared with no CVD at baseline, people with CVD in CKD stage 1–3 were at a 35–52% increased risk of progression; in stage 4–5 there was no difference in the risk of progression. 111 Others presented the number accepted onto RRT as a proportion in a given follow-up time (see Appendix 6). Tonelli and colleagues68 analysed ACM and CVD mortality for studies recruiting only patients with CVD or patients with HBP at baseline, comparing CKD with no CKD. For those with CVD and CKD at baseline, the RR of ACM compared with no CKD was 1.71 (95% CI 1.49 to 1.96), substantially lower than that reported in the studies in general populations summarised in the section All-cause mortality: general CKD population (RR 3.0), and possibly reflecting the nature of the ‘general population’ studies that included known high risk groups. The RR for CVD mortality was 1.8 (95% CI 1.45 to 2.24), again lower than the studies of general populations. For studies of populations with hypertension, Tonelli and colleagues68 reported an ACM RR of 2.15 (95% CI 1.17 to 2.61) and a CVD mortality RR of 2.35 (95% CI 1.52 to 3.64). See Appendix 6 for detailed results.

Gender, race and age in CKD

The age standardised risk of mortality was reported by Evans and colleagues98 to be higher in females with CKD (SMR 12.3, 95% CI 10.3 to 14.5) than in males (SMR 7.2, 95% CI 6.3 to 8.1), as compared with the general population. Males had higher cumulative mortality and cumulative incidence of ESRD after 10 years than females. 96 Drey and colleagues95 found a similar proportion of deaths in males (69.34%) as in females (67.76%) during 5.5 years of mean follow-up. Meisinger and colleagues76 and Irie and colleagues81 reported a higher risk of CVD mortality in females than in males (see Table 44, Appendix 6), while risk of ACM was similar in both genders. The rate of decline in GFR was found to be higher in males than in females. 82,101 See Appendix 6 for detailed results.

Chronic kidney disease seemed to have a greater impact on ACM risk (HR 1.83, 95% CI 1.33 to 2.52) for African-Americans than for white people (1.31, 95% CI 1.16 to 1.49) (ACM in those with CKD versus no CKD). 74 Similarly, risk of cardiovascular morbidity increased more in African-Americans with CKD (2.09, 95% CI 1.33 to 2.52) than in white people with CKD (1.01, 95% CI 0.84 to 1.21). 74 Additionally, Hsu and colleagues103 reported the 5-year incidence rate of ESRD cases was higher among African-Americans (5.4%) than among white people (1.1%).

The mortality and incidence rate ratios for ESRD were significantly higher in younger groups (less than 69 years old) than in older patients when comparing CKD with no CKD. 96 Hsu and colleagues103 also presented incident ESRD rates that were greater among younger patients (20–60 years old) than among older ones (61–74 years old). The mortality rate ranged from 0.4% per year for 18- to 44-year-old patients with an eGFR ≥ 60 ml/min/1.73 m² to 36.0% per year for the most elderly patients (85–100 years old) with an eGFR ≤ 15 ml/min/1.73 m². 110 The RR of mortality increased with decreasing eGFR, but the relative impact was less as age increased. 110 For example, risk of ACM for those with an eGFR of 40–49 ml/min/1.73 m² decreased from HR 1.90 (95% CI 1.35 to 2.67) in 18–44 year olds to HR 1.35 (95% CI 1.32 to 1.39) in 65–74 year olds. Similarly, for those with eGFR of 30–39 ml/min/1.73 m², the HR decreased from 3.58 (95% CI 2.54 to 5.05) in 18–44 year olds to HR 1.81 (95% CI 1.75 to 1.87) in 65–74 year olds. 110 Imai and colleagues82 reported that the mean rate of decline of GFR from an initial GFR of 30–39 ml/min/1.73 m² was higher (3.28, SD 0.72 ml/min/year) in younger males (40–49 years old) than in those in older age groups, with mean rate of decline 0.91 (SD 3.28) in 50–59 year olds, 0.98 (SD 0.18) in 60–69 year olds and 1.24 (SD 0.25) in 70–79 year olds. No difference in progression rate across age groups in male participants with a starting GFR of 60–69 ml/min/1.73 m² was observed, although the progression rate was much lower (0.31–0.36, SD 0.01–0.03 ml/min/year). 82 Among female participants, the mean rate of decline of GFR from an initial GFR of 60–69 ml/min/1.73 m² was higher for ages 40–49 years (0.45, SD 0.01) ml/min/year than for 50–79 year olds (range 0.24–0.29, SD 0.01–0.02 ml/min/year). See Appendix 6 for detailed results.

Renal function: progression and onset of RRT

Five studies reported creatinine clearance or eGFR at the start of RRT but did not report renal function at the time of referral, at first diagnosis or during follow-up. 56,119,121–123 Only two studies reported renal function over time as an outcome. 111,120

Renal function at the onset of RRT was not substantially or consistently altered by referral to the specialist team at least 12 months prior to requiring RRT. 56,119,121–123

Orlando and colleagues111 retrospectively compared those receiving care by primary care physicians alone with those receiving specialist nephrology input to assess progression (defined by change of stage or death). More progressed to a more advanced stage of CKD in the nephrology referral group at each stage of disease than the primary care only group, but fewer died. There was little difference in time spent in follow-up between groups in stages 1 and 2. For stage 3, 316 more days were spent in follow-up in the nephrology group than the primary care only group. The difference in deaths was particularly marked for those with stage 3 or worse CKD. Cox HRs for the risk of progression, adjusting for age, race, medication use, comorbidities and risk factors showed no statistically significant difference between nephrology referral and primary care only for those with stage 1 or 2 disease. For those with stage 3 or worse, nephrology care reduced progression or death (after adjustment) with an HR of 0.8 (95% CI 0.61 to 0.90) for stage 3 disease and 0.75 (95% CI 0.45 to 0.89) for stage 4. Of the 1553 participants, 26 progressed to ESRD during the 5-year follow-up with six (0.4%) in the primary care only group and 20 (1.29%) in the specialist care group. The higher progression in the referred group reflects, at least in part, that fewer people died during stages 3–5 and thus survived to require dialysis. 111

Martínez-Ramírez and colleagues120 reported that during the 1-year prospective follow-up of diabetic patients with evidence of early or overt nephropathy, only one patient deteriorated to requiring RRT (in the unreferred group). Among the group where no referral to a specialist service was available, mean (SD) eGFR deteriorated from 78.6 (28.1) ml/min/1.73 m² to 66.6 (29.9) ml/min/1.73 m², a statistically significant fall of 12.0 ml/min/1.73 m² (p < 0.05). In the cohort referred early to specialist nephrology care, mean (SD) eGFR was 83.8 (26.1) ml/min/1.73 m² at baseline and was not statistically significantly altered after 12 months [80.4 (35.5) ml/min/1.73 m²]. A similar pattern was observed regardless of whether the patient had early or overt diabetic nephropathy (defined by the presence of microalbuminuria with eGFR > 60 ml/min/1.73 m² versus proteinuria or eGFR < 60 ml/min/1.73 m²). 120

Table 12 summarises the data from these studies.

Mortality

All seven studies reported mortality as an outcome. 56,111,119–123 Roderick and colleagues,56 however, did not report it by the referral groups of interest here. Martínez-Ramírez and colleagues120 reported three deaths during the 1-year follow-up, all in the group who were not referred to specialist care, and all cardiovascular deaths. Orlando and colleagues111 reported higher numbers of deaths among those cared for in primary care than among those referred to a specialist. This pattern held at all stages of CKD.

In all of the retrospective studies, starting their cohorts from the initiation of dialysis, better dialysis survival was associated with early referral. This improvement was observed from 90 days119 to 5 years121 after initiation of dialysis. Jungers and colleagues121 reported survival was highest in the group referred more than 72 months prior to dialysis. The difference in survival could be observed from 3 months after starting dialysis and was marked by 5 years (77.3% in those referred > 72 months prior to dialysis versus 57.8% in those referred less than 6 months prior to dialysis; p < 0.001). CVD accounted for more than 50% of the deaths. The earlier referral groups (> 36 months and > 72 months) experienced fewer CVD deaths than those referred later. Adjusting for age, sex and comorbidities (diabetes mellitus, HBP, CVD), the RR of death for those with referral > 72 months prior to dialysis, as compared with < 6 months, was 0.24 (95% CI 0.10 to 0.59) at 1 year and 0.53 (95% CI 0.35 to 0.79) at 5 years. Kessler and colleagues119 reported an association between timing of referral and survival in the first 90 days after initiating dialysis, with an almost threefold increase in risk (HR 2.7, 95% CI 1.2 to 6.3) once referral was less than 12 months prior to dialysis, and as great as five times higher (HR 5.2, 95% CI 2.2 to 12.3) if referral was less than 1 month prior to dialysis after adjustment for age, sex and systolic BP. For survival over 3 months, the only independently associated factor relating to referral timing was for those referred between 1 and 4 months prior to dialysis, where a HR of 2.2 (95% CI 1.4 to 3.5) was observed compared with referral < 1 month. Khan and colleagues121 reported lower 1-year mortality on dialysis in those referred during the 24 months prior to dialysis (25–35%) as compared with those with no pre-dialysis referrals (51%) [HR 1.5 (95% CI 1.44 to 1.55) after adjustment for age, sex, race, erythropoietin injections, non-nephrology care and comorbidities]. There was evidence of a ‘dose-response’ gradient, with those receiving the most nephrology visits having the lowest mortality. Kinchen and colleagues123 reported lower mortality on dialysis in those referred more than 12 months, as compared to those referred less than 4 months, prior to dialysis. The differential in survival was observed to at least 3 years. An HR of 1.6 (95% CI 1.11 to 2.24) for later referral, as compared to referral > 12 months prior to dialysis, was observed after adjustment for type of dialysis, demographic characteristics, socioeconomic status, years of smoking, exercise status and Index of Coexistent Disease score. Table 13 provides a more detailed summary of the mortality data reported.

Hospitalisations

Three studies reported hospitalisations as an outcome. 56,111,121

Orlando and colleagues111 compared hospitalisations in the group referred to nephrology specialists with those with renal disease managed only in primary care. The number of hospitalisation days differed little between the two groups (mean 2.8 versus 2.5 days respectively; p = 0.03).

One study reported on the impact of early referral on hospitalisation in the first 6 months after dialysis was initiated. 56 Roderick and colleagues56 observed that those in the group with referral < 1 month prior to dialysis had more hospitalisation episodes within the first 6 months of dialysis than all others (1–4 months, 4–12 months and > 12 months) (mean 2.6 versus 1.7, p = 0.001). The median length of stay was shorter in those referred more than 1 month prior to dialysis (10 versus 18 days in the < 1-month group). Lengthening referral time to greater than 1 month prior to dialysis did not have any substantial effect on this outcome.

Jungers and colleagues121 reported the impact of referral on the duration of initial hospitalisation at the time of starting dialysis. They found a statistically significantly lower duration of initial hospitalisation in those referred at least 6 months prior to dialysis [mean 23.8 (SD 17.1) days] versus those referred 6 to 35 months prior to dialysis [mean of 7.5 (SD 8.9) days; p < 0.001] and some evidence of a dose response with improvements out to referral > 72 months prior to dialysis.

Emergency dialysis

Kessler and colleagues119 reported statistically significantly higher emergency dialysis among those referred late than those referred more than 12 months prior to dialysis (83.3% versus 29.1%; p < 0.001).

Quality of life

None of the included studies reported quality of life as an outcome.

Blood pressure control

In the two prospective studies of early referral for CKD, improvements in BP control were observed. Martínez-Ramírez and colleagues120 reported statistically significant improvements in systolic BP in the referred cohort [140 (SD 30) mmHg to 130 (SD 21) mmHg] while BP control deteriorated in the unreferred group [140 (SD 19) mmHg to 145 (SD 23) mmHg] (p < 0.05). Orlando and colleagues111 reported that a higher proportion of patients in the nephrology care group had good BP control (41% versus 36%; p = 0.06).

At the initiation of dialysis, Jungers and colleagues121 reported BP was lower in all the group referred more than 6 months prior to dialysis and lowest in the group referred more than 72 months prior to dialysis [systolic BP mean 171 (SD 23) mmHg in the < 6-month group versus 148 (SD 17) mmHg in the 6- to 35-month group and 141 (SD 12) mmHg in the ≥ 72 months group; p < 0.001]. However, Kessler and colleagues119 and Kinchen and colleagues123 did not observe statistically significant differences in BP control at the start of dialysis. Khan and colleagues122 and Roderick and colleagues56 did not report BP control.

Other clinical markers

Four studies56,111,121,123 reported serum albumin was statistically significantly higher if referred early, but none presented information about the underlying diagnoses. Those with nephrotic syndrome may have presented earlier, but have relatively preserved renal function despite their apparently poor clinical markers, e.g. low albumin as a result of heavy proteinuria. Kinchen and colleagues123 described statistically significant differences in serum albumin < 36 g/l at the time of initiation of dialysis (60.5% versus 77.9%; p < 0.001), when comparing early and late referral. Orlando and colleagues111 reported albumin levels < 40 g/l in 59% versus 49% (p < 0.001) (early versus late respectively). Roderick and colleagues56 noted albumin was lower in those referred less than 1 month prior to dialysis [mean 32 g/l (SD 0.83)] than for those with referral > 12 months [mean 37 g/l (SD 0.56)] (p < 0.001). Khan and colleagues122 found no statistically significant difference in albumin. Kinchen and colleagues123 also described statistically significant differences in haematocrit < 0.3% (56% versus 68.1%; p < 0.001). Khan and colleagues,122 Roderick and colleagues56 and Orlando and colleagues111 did not observe substantial differences in haematocrit between early and late referral groups.

Treatments

Martínez-Ramírez and colleagues120 reported statistically significant improvements in systolic BP in the referred cohort. In terms of treatments, the use of ACE Is increased more in the referred group and there was substantially higher use of ARBs and statins. Non-steroidal anti-inflammatory drug use fell in the referred group, but increased in the unreferred group (Table 14). Orlando and colleagues,111 however, reported no statistically significant differences between the groups for lipid lowering agent or ACE I use.

Roderick and colleagues56 reported that vitamin D supplementation (late 20% versus early 40%), phosphate binders (late 29% versus early 44%), sodium bicarbonate (late 10% versus early 28%), lipid lowering agents (late 6% versus early 11%) and erythropoietin (late 5% versus early 23%) were all prescribed more frequently in those referred more than 12 months before starting dialysis.

Kinchen and colleagues123 described statistically significant differences in the following factors when comparing early and late referral: exercise one or more times per week (26.3% versus 14.9% respectively; p < 0.001); and erythropoietin treatment (25.3% versus 12.7%; p < 0.001).

Evidence of clinical effectiveness of early referral

Despite the difficulties around the definition of CKD and referral, we sought to report what evidence was available about the effect of early referral on clinical outcomes.

Progression

One of the key aims of referral to specialist nephrology services is to initiate interventions to stop or slow progression towards ESRD. Many studies have demonstrated the effectiveness of ACE Is and ARBs in reducing the progressive decline in eGFR in trials. 42–46 BP control is also important. 42 The comparison of nephrology referral for all those with diabetic nephropathy versus those with no access to nephrologists in Mexico120 demonstrated better preservation in function in those referred to a specialist. The comparator group, with no access to specialist services, experienced a mean decline in renal function of 12 ml/min/1.73 m², which is approximately three times higher than we found described in the natural history review. 88 This pattern was observed regardless of degree of baseline microalbuminuria. There was a high proportion of stage 1 and 2 CKD patients in this study and, as described by Djamali and colleagues,83 the absolute rate of decline may be at its fastest during these stages. Orlando and colleagues111 reported a higher risk (RR 1.32, 95% CI 1.13 to 1.55) of progression in stage 1–2, but not stage 3–5 for people with diabetes than those with no diabetes. The high rate of decline in function in the primary care only group may, therefore, be generalisable to a wider clinical setting.

Similarly, Orlando and colleagues111 reported better outcomes for a composite end point of death or progression in those referred to specialists with stage 3 disease or worse. However, looking at renal progression alone, the crude data suggest more people in specialist care experienced progression. This may be as a result of differences in case-mix, survival and referral bias (where the sick or low risks are not referred by GPs) rather than due to differences in care. Predialysis survival was noted to be better in the group referred to specialists. 111 This finding may reflect a selection bias where people considered to have other serious comorbidities were less likely to be referred. However, a composite end point of death and renal progression was found to be significantly reduced by referral to a specialist for people with stage 3–5 CKD, even after adjustment for comorbidities and age.

Mortality

If patients progress to ESRD and survive to dialysis then there was evidence that post-dialysis survival was improved by early referral, even for referral more than 72 months prior to dialysis. Improved survival was sustained beyond that relating to initial establishment on RRT. This suggests that the survival advantage was not simply explained by technical preparation for dialysis and nutritional status. The differential effect of early referral to a specialist (> 72 months) on survival post dialysis lasted for at least 5 years.

The clinical difference at initiation of dialysis may reflect differences in care that are markers for overall care, or may in part reflect an improvement in clinical status of the patient at the time he or she starts dialysis that is causally linked to survival. Some hypothesise that they are markers for improved cardiovascular risk modification that then manifest in improved survival long after the early dialysis phase. 121 However, in the types of studies presented, the differences in clinical markers may also reflect selection bias, where clinical staffs are selecting the healthiest and fittest to refer for dialysis. 123,124

Kessler and colleagues119 noted those referred late had low contact with primary care compared with those referred early. If CKD was diagnosed only within 1 month of requiring dialysis, then 43.4% of patients received no regular primary care input.

There was little evidence that hospitalisations were affected in the pre- or post-dialysis phase by early specialist referral beyond 1 month prior to starting dialysis.

Other measures of effectiveness

Given the chronic nature of this condition and the important potential to effect quality of life it was perhaps surprising that no studies reported aspects of quality of life.

The nature of what early referral meant in terms of clinical care was essentially not described. The limited detail available has been summarised above. Looking at the impact of different care on aspects of management (of risk factors) and prescribing identified inconsistencies between studies. However, BP control and/or ACE I/ARB prescribing were the most consistently identified differences between referral and standard non-specialist care. The use of CVD risk modifiers such as statins was also reported to be higher in specialist care by some authors. 111,120

Multidisciplinary care

Multidisciplinary care was the most commonly described approach to the management of CKD populations. Table 19 details the services and settings. MDC models were described in primary care150 and hospital-based clinics. 142,143,145,146,151,152,154 The MDC disease management programmes based in primary care involved community-based teams of nurses, dieticians, social workers and GPs with a wide scope including a named nurse available to each patient. 150 Hospital care-based MDC nephrology clinics comprised nurses, nurse-educators, social workers, dieticians, nephrologists and academics. Joint diabetics and renal clinics included MDC teams with specialists from clinical specialties. 152 Some described a case management approach with a named team for individual patients. 145

Structured care by an individual specialty

Apart from nephrologists, CKD management has been influenced by the involvement of other health-care professionals. Pharmacy programmes included structured clinics solely led by pharmacists using protocols to manage the medication of patients with microalbuminuria and nephropathy. 148 Pharmacists also operated within structured disease management programmes implemented by pharmacists and diabetes specialists managing diabetic nephropathy patients together. 149 Joy and colleagues156 described the support of pharmacists as part of a multidisciplinary team for the management of CKD patients. Some care programmes were run by renal specialist nurses, or renal specialists working closely with general nurses. 157,158 Two studies compared the care given in primary care with that of specialists (nephrology and diabetology) in Italy. 137,138 Care by GPs was also compared with virtual clinic management (Table 20). 144

Educational intervention

Educational interventions for care in CKD were described by some of the studies. Generally, two types of educational programme were observed: education of general staff and patient education. 141,142,146,150,153–158 The education programmes varied in content and delivery style and are detailed further in Table 21.

Patients surveillance

Patient surveillance, including regular clinical and laboratory examinations (e.g. BP, weight, haemoglobin, calcium, phosphate, other metabolic parameters) along with medical review, was the most commonly reported element of care within all models of care. Patient clinic visits or follow-up varied from monthly to annually depending on the level of kidney function and clinical need. 142,146,147,151,152 Appointments lasted for 30–35 minutes143,148 and some included discussion of a patient in a multidisciplinary team meeting. Harnett and colleagues144 reported so-called ‘virtual methods’ to conduct this surveillance with a patient attending a practice nurse for examinations and the results being logged for a remote specialist team to review them in the context of that patient’s clinical record. 147 None of the studies described in any detail how patients were selected for inclusion in the CKD care model. Most appeared to rely on measures of eGFR (or creatinine thresholds), but there was rarely a description of why patients were originally tested (i.e. clinical need versus screening) or whether other factors were used to influence selection for further management.

Patient education

Patient education included reviewing pre-specified educational topics with patients,142 discussion about CKD, dietary restriction, monitoring BP, medicine management, lifestyle modification,146,150 education on dialysis154 and discussion about clinical parameters. 153 Some CKD management programmes developed a guideline, self-care manual or performance-based programme for patient education with education classes held prior to clinics. 153 A patient-centred education and management approach was reported where renal nurses were working along with other specialists and primary care teams. 157,158

Health-care professional education

Only three papers reported education provided to health-care professionals. 141,154,156 In two studies, primary family physicians were provided with educational sessions. 141,154 The educational intervention to physicians included lectures based on interactive theory–practice models and discussions (one-to-one or small group discussion of real cases led by investigators). Lectures included a theory course covering various aspects such as basic anatomy and physiology of kidney, epidemiology and clinical measures and advanced topics such as associated comorbidities, prevention and management.

Medicine management

With the aim of optimisation of treatment, most of the CKD programmes considered medicine management; in particular the use of ACE Is and ARBs. 145,149 Pharmacist-led structured care models included medicine management and optimisation of treatment which were based on prescribing recommendations and clinical standards. 148 Medicine management included a full drug history and ascertaining patients’ knowledge of their medicine treatments. 148

Managing/preventing complications

Some of the MDC clinics focused on the management and prevention of complications associated with CKD. 146,149,150,152,154 Education on lifestyle and cardiovascular risk factors was provided to reduce CVD risk and to delay CKD progression. 146,149,150 To prevent the development of diabetic nephropathy, some care focused on screening for microalbuminuria and clinical proteinuria. 152 Ghossein and colleagues154 mentioned that a comprehensive renal care programme included screening for, and treating, the complications of CKD such as anaemia and renal osteodystrophy. Additionally, some MDCs endeavoured to run structured screening programmes (screening for comorbid conditions, microalbuminuria or proteinuria, or to identify those to prepare for dialysis and transplant). 152,154 The separation of ‘new referral’ assessment from follow-up was reported to enable more focused clinics and to identify patients where no follow-up was required. 143

Nutritional advice

Nutritional advice was provided as a part of education programmes146,153 or as a separate element of more comprehensive care models. 145,150,154

Social care

A few models of care included a component related to social care services. 142,145,146,151 Social care workers were involved to decrease barriers to care by interviewing patients and by direct intervention if needed. 145 Others stated that social workers were also involved in providing education to patients. 146

Clinical markers

Patients under the care of a MDC were consistently reported to have improved clinical surrogates such as BP, glycosylated haemoglobin and cholesterol when compared with before the start of treatment in the MDC (Table 23). Patients under the care of the SIMON programme had a significant reduction in mean BP after referral. 147 Likewise BP was improved when patients attended multidisciplinary nephrology clinics and redesigned services. 143,151 It was observed that in the redesigned service for diabetic nephropathy patients, the mean fall in BP was –11/7 mmHg (148/80 mmHg at baseline to 137/73 mmHg at follow-up, p < 0.001) and 33% achieved a BP target of 125/75 mmHg. 143 Moreover, patients’ glycosylated haemoglobin and low density lipoprotein cholesterol improved and there was minimal progression of microalbuminuria to diabetic nephropathy, and the proportion of patients with proteinuria fell. Similarly, in a multidisciplinary nephrology clinic, mean BP fell from 151 ± 27/80 ± 13 mmHg at the initial visit to 135 ± 18/75 ± 11 mmHg at follow-up of 4 years. 151

Structured care programmes run by pharmacists resulted in clinically and statistically significant improvements in mean BP (from 150.5 ± 19.3/79.7 ± 10.2 mmHg to 132.6 ± 15.2/67.8 ± 10.5 mmHg, p < 0.001). Similar trends of significant results were observed for cholesterol and glycosylated haemoglobin in CKD patients. 148 Likewise, CKD management by collaborative approach between pharmacists and diabetes specialists proved to be beneficial with improvements in clinical surrogates (BP and low density lipoprotein cholesterol) and risk reduction of end points compared with those receiving usual care. 149 Those using educative intervention among family physicians have shown clinical improvements in BP and kidney function. 141

Where comparator groups of patients were available, the findings were less consistent.

Curtis and colleagues142 reported patients exposed to MDC had clinical benefits including higher levels of haemoglobin, albumin and calcium at dialysis start than those receiving standard care. Minutolo and colleagues137 reported that hypertensive patients with CKD under nephrology care seemed to have improved BP (21.5% patients reached BP target of 130/80 mmHg, 95% CI 15.6 to 27.4, p < 0.0001) compared with primary care (5.8%, 95% CI 2.9 to 8.6). Patients were 2.6 times less likely to have reached their target BP in primary care versus nephrology care (after adjustment for age, diabetes and eGFR). However, in patients with diabetic renal disease, there was little difference in the proportions achieving target BP (10–14%), with similarly low levels in nephrology, diabetology and primary care. 137

Harris and colleagues,145 reporting the only RCT, showed no difference in weight, BP or creatinine clearance when they compared people managed in an intensive MDC management programme with those receiving standard care.

Prescription changes

Statistically significantly higher prescriptions of renin–angiotensin aldosterone system (RAAS) inhibitor agents were noted. 143,147 Prescription rates were found to be significantly higher in specialist care than in primary care, and the choice of medicines differed. 137,138 Nephrologists were more likely to use dual blockade of RAAS using ACE Is and ARBs and were more likely to prescribe loop diuretics than either primary care physicians or diabeteologists. 138 These differences existed after accounting for clear differences in case-mix. Use of antihypertensives (including RAAS inhibitor agents) and cholesterol reducing drugs (statins) appeared to increase with the involvement of pharmacists and diabetes specialists in CKD care. 148,149 Prescription of ACE I, in those without contradictions, increased from 66% to 100%, while that of ARBs (single or in combination) increased by 29%. 148 Prescriptions for antihypertensive agents by physicians who underwent an educative intervention were higher than those by physicians without educative intervention. 141 Harris and colleagues,145 however, reported little difference in prescribing behaviour between those cared for in intensive multidisciplinary case management and those receiving standard care.

Long-term outcomes

Long-term outcomes have been reported only for MDC models. See Table 24 for outcomes reflecting mortality and CKD progression.

Attitudes of patients/health-care providers

The service delivered by joint diabetes and renal clinics for the patients with established diabetic kidney disease at Eastbourne, UK, has been appreciated by patients and was considered to be a level of care expected by GPs. 152 Joint clinics have also been supported by NICE. 173

The perception of Canadian nephrologists towards MDC-based CKD clinics has been presented in one survey. 159 More than 90% of nephrologists reported that MDC-based CKD clinics were easily accessible to them. Regarding decision-making on referral to MDC clinics, most (more than 80%) of the nephrologists found calculated creatinine clearance the most useful method rather than depending on estimated months before ESRD. Fifty-seven per cent of nephrologists reported that they referred patients with a creatinine clearance of 20–29 ml/min, while around 30% reported earlier referral with creatinine clearance between 30 and 59 ml/min as the best time for referral. Others referred at lower creatinine clearance (10–19 ml/min). The survey also presented information about perceptions of the availability of multidisciplinary staff. More than 90% of nephrologists reported that they had excellent access to nurses, dieticians and social workers. Around 64% believed that pharmacists were comparatively less commonly available. According to nephrologists’ opinion, nurses provided more information about dialysis to patients than about palliative care, prognosis or pre-emptive transplant.

Overall, nephrologists (94.5%) believed that MDC-based clinics were superior to conventional clinics. Most of them (83%) also thought that this kind of clinic could be managed by nurses using algorithms, thus helping to reduce the burden on nephrologists. They were supportive of the value of MDC clinics for those with stable and slowly progressing CKD but reported that the clinics were not sufficiently funded.

Optimal management

Four papers proposed similar ranges of strategies of care for the management of early CKD patients that they hoped would lead to improved outcomes. 160–163 The emphasis was on early detection of CKD and associated comorbid conditions. The proposals focused mainly on the need for delaying CKD progression, and preventing or treating complications by using timely intervention. Importantly, these papers identified the need for precise measures to stratify CKD patients into groups that may help to predict future risks. Patient education, to increase awareness of CKD and risk of comorbidities in CKD, was recommended as a part of optimal management. Moreover, two papers highlighted the importance of having co-ordinated care for the management of early CKD. 162,164

Authors proposed that implementing the strategy of optimal management and early detection of the disease may help reduce hospitalisation, morbidity and mortality, and eventually help to reduce cost. For example, identifying comorbid conditions like anaemia and CVD at earlier states, and treating them, may reduce the risk of future morbidity and mortality. Controlling BP, proteinuria and other metabolic parameters may reduce the risk of CKD-related comorbidities. Authors were proponents of the use of ACE Is and ARBs, identifying CKD at earlier stages and referring patients to the nephrologists to slow CKD progression. The use of existing data from patient database systems to develop an intervention programme to optimally managed CKD patients was suggested. 160

Integrated approach for management

Seven papers proposed the advantages of collaborative approaches in CKD management. 164,165,167–172 Involvement of multidisciplinary staff (GPs with specialist interest, specialist nurses, pharmacists, nutritionists, etc.) and co-operation between these staff may aid early referral and management. The key role of nurses was highlighted as: patient educators, care managers communicating between care providers and support staff, and following up patients to monitor their progression. 166–168 Four of the papers suggested that patients should be managed and evaluated in primary care by skilled support staff, with only complex or severe cases referred to specialists or nephrologists in secondary care. 166,168,171,172 Furthermore, they noted the importance of care plans being developed based on evidence-based practice guidelines. Recognition that CKD was a chronic and complex condition was found to be useful in supporting collaboration and education needs. 164 The proponents of an integrated approach for management also hoped that it would contribute to reducing mortality and hospitalisation and delaying progression. Moreover, this co-operative care potentially provided a mechanism to deliver optimal management of CKD at early stages and thus increase quality of care and improve quality of life of patients while reducing cost.

Similar to the above proposals, A vision for the future of renal services, 2002126 also focused on prioritising early diagnosis and treatment of those with CKD and those who were at risk of developing CKD. The authors recommended the need for clinical networks as a model of future services provision. Details of the network focused on preparation for end-stage disease and those with progressive renal disease, but did note the need to integrate services across disciplines and to deliver a flexible and appropriate care for the individual patient’s needs. They also noted the need to tackle inequalities, giving attention to the elderly, children, socially disadvantaged, ethnic minorities and those with comorbidity. 126

Key themes from aspirational proposals

Thematic review of the aspiration studies identified the following key elements to care:

CVD risk estimation

Annual cardiovascular event risks for the different disease states were built up through a staged approach. Annual base risks were first estimated for each cohort stratum using the demographic/risk profiles in Table 27 and the QRISK2 (ClinRisk Ltd, University of Nottingham, UK) web-based risk calculator. 199 This is based on a CVD risk algorithm which has been developed and validated for the general population of England and Wales. 200 Adjustments were made for gender and the proportion of people with hypertension within each stratum. These risk estimates reflect the average annual probabilities of cardiovascular events in cohorts with these risk profiles, without CKD or pre-existing CVD. Included in the definition of CVD events are coronary heart disease (angina and myocardial infarction), stroke and transient ischaemic attacks. Although there is likely to be a tendency for those with an ACR ≥ 30 mg/g or with CVD to be older than those with CKD alone, we assumed a constant age of 72 years across the strata when calculating base risks for incorporation in the model. This is because Markov models have to be analysed for cohorts of a single age. In addition, the effect of increasing age on cardiovascular events was incorporated in the model using an HR for increasing age (see below).

Following estimation of the base risks, we estimated annual probabilities of CVD events for each disease state by multiplying the base risks by RRs and/or HRs for CVD events associated with reduced eGFR, ACR 30–299 mg/g, ACR ≥ 300 mg/g and pre-existing CVD (Table 28). The RRs associated with reduced eGFR, ACR 30–299 mg/g and ACR ≥ 300 mg/g were obtained from a study based on data from NHANES III. 84 This study reported RRs for cardiovascular mortality by eGFR category (15–59 ml/min/1.73 m², 60–89 ml/min/1.73 m² and ≥ 90 ml/min/1.73 m²) and ACR levels (< 30 mg/g, 30–299 mg/g and ≥ 300 mg/g). We assume that the RRs for any CVD event (fatal and non-fatal) would be similar. The RRs were adjusted for all the factors in Table 27 used to generate the base risks.

To estimate the increased risk of cardiovascular events in people with pre-existing CVD, we used an adjusted HR associated with prevalent CVD based on data reported by Parikh and colleagues. 194 We estimated the HR inferred by existing CVD in people with CKD by dividing the HR associated with having CKD and CVD by the HR associated with having CKD without CVD.

In order to incorporate the increased probability of cardiovascular events with increasing age, we incorporated an HR for CVD events associated with a 10-year increase in age. 196 The HR was adjusted for pre-existing CVD and CKD and other classical risk factors.

When using HRs to estimate annual event probabilities, base risks were converted to average rates before being multiplied by HRs, and then converted back into risks (probabilities). The resultant annual CVD event risks for each disease state in the model are reported in Table 29.

The risk estimates reported in Table 29 represent approximations of the annual risk of experiencing any CVD event (fatal and non-fatal myocardial infarction, angina, fatal and non-fatal stroke, and transient ischaemic attacks). We assume that fixed proportions of people experiencing CVD events would die as a result within the year of the event. Proportions of 25% and 50% were selected for CVD events occurring in those with CKD stage 3a, and CKD stages 3b and 4 respectively in order that the model predicted cumulative cardiovascular mortality in line with the different CVD mortality rates observed for these subgroups in similar cohorts32,194,196 (see Model validation). Given the great uncertainty surrounding these assumptions, we subjected them to extensive sensitivity analysis. For purposes of costing, 33.5% of non-fatal events were assumed to be major (myocardial infarction or stroke) and 66.5% of non-fatal events were assumed to be non-major (angina, transient ischaemic attacks) – based loosely on numbers of non-fatal major and other CVD events reported in the Heart Protection Study. 201 Again, these assumptions were subjected to sensitivity analysis.

Chronic kidney disease progression

In order to generate annual transition probabilities for CKD stage progression, we used a data set of all individuals with a recorded eGFR (calculated from recorded serum creatinine results) extracted from the Primary Care Clinical Informatics Unit (PCCIU) research database for another study. 39 The PCCIU research database comprises patient data from 320 Scottish primary care practices (∼1.8 million patients). This represents 38% of Scottish practices and has been shown to be representative of the Scottish population. We simulated CKD progression over a 5-year period using reported rates of eGFR decline by level of urine albumin/protein at baseline, and calculated the average proportion of individuals that would be expected to transit annually from each CKD state to the next state. In the first instance we applied annual rates of eGFR decline of –2.5 ml/min/year for people with a normal ACR, –3.5 ml/min/year for people with microalbuminuria and –5.5 ml/min/year for people with proteinuria as reported in the MDRD study. 202 However, application of the resultant CKD transition probabilities resulted in estimates of cumulative ESRD incidence significantly higher than the rates reported for general CKD cohorts in the literature. We therefore proportionally reduced these rates until they gave estimates of ESRD close to the cumulative 10-year incidence of ESRD observed in the study by Eriksen and Ingebretsen. 96 This study consisted mainly of individuals with early stage 3 disease (eGFR 45–59 ml/min/1.73 m²) and so probably underestimated the incidence of CKD in our cohort. The resultant transition probabilities are displayed in Table 30. We assumed that individuals in the cohort had established CKD and so, by definition, could not regress to less severe CKD states. It was also assumed that individuals could progress a maximum of one CKD stage per year.

Albumin–creatinine ratio

Individuals in the cohort were allowed to develop ACR 30–299 mg/g or ACR ≥ 300 mg/g over the course of the simulation. As we could find no evidence on the rate at which people with a low eGFR develop these complications, we used the rates reported for people with diabetes in the base-case analysis; 2% (95% CI 1.9 to 2.2) per year for development of ACR 30–299 mg/g, and 2.8% (95% CI 2.5 to 3.2) per year for development of ACR ≥ 300 mg/g from ACR 30–299 mg/g. 203

Mortality from other cause

In each cycle of the model, individuals experience an age- and sex-adjusted probability of dying from causes other than CVD or ESRD. These probabilities were estimated by removing deaths due to CVD and renal disease from the data used to derive the interim life-tables published by the UK Office for National Statistics. 204,205

Standard practice

In order to come up with estimates of annual resource use and costs of referral by stage of CKD and level of comorbidity (CVD and ACR > 30 mg/g), we used a combination of published data and assumptions based on expert opinion.

One of the main sources of current resource use data for unreferred CKD was the study by Stevens and colleagues,192 which reported hypertension rates and antihypertensive medication use by stage of CKD for a retrospective sample of 11,731 patients with an eGFR below 60 ml/min/1.73 m². The cohort was identified, using computerised records, from patients across 17 primary practices in England, who had a valid serum creatinine test between 1998 and 2003. The cohort was largely unknown to nephrology services.

Primary care resource use and medication costs were estimated by CKD stage for the four cohort strata described in Table 27. We based our costing, as far as possible, on resource use data reported by Stevens and colleagues. 192 However, as this study relied on data collected between 1998 and 2003, we also attempted to update the initial costing to reflect possible increases in the use of resources for these individuals in recent years. The initial estimates were used in the base-case analysis.

Early referral strategy

In estimating the costs of a formal referral strategy for people with reduced eGFR on GP registers, we followed the description of the SIMON programme, as reported by Jones and colleagues. 147

In this shared primary and secondary care nephrology scheme, people with reduced eGFR were first referred to nephrologist for assessment. Those considered to be stable and uncomplicated were then managed under an SCS where they were monitored by nephrologists through clinical and biochemical reviews recorded in primary care. Those considered to be unstable and/or complicated were managed under hospital outpatient care, but could be referred to shared care at a later date should their condition stabilise.

The scheme required an administrative system where patient records and reviews were stored on a central database. For those under shared care, administrators sent out blood and urine test forms every 6–12 months, depending on clinical condition, and asked that the patient attend their primary care practice to have the tests performed. The patient’s BP, weight, medications and urinalysis (ACR) results were recorded by practice nurses and then sent back to the administrative staff. Results were entered into the database and nephrologists reviewed the results via an electronic system, which used specialist software to chart trends in eGFR and other blood tests. If a patient’s renal function deteriorates significantly, or there is concern about BP control, the patient can be recalled for hospital care. Through this system, nephrologists can also advise on medication use.

We used the above description along with available unit cost data and several assumptions to estimate the average annual cost of implementing this system for cohorts of patients with varying levels of renal insufficiency and comorbidity.

Additional hospital costs associated with CVD events

Costs associated with CVD events were taken from a published cost-effectiveness study which assessed patient level hospitalisation costs associated with major cardiovascular events (myocardial infarction, stroke), other CVD events (angina, transient ischaemic attacks, etc.) and CVD deaths. 201 This study also reported hospitalisation costs in the years following CVD events, when no other CVD events occurred (see Table 7). These costs were applied in the model as transition costs. Given the high CVD mortality risk in people with ESRD (∼0.125 per year), we assumed that 50% of these people would experience a CVD event each year; with a CVD event fatality rate of 25%, this results in an annual risk of CVD mortality of 0.125. Those experiencing non-fatal CVD events incur the associated hospitalisation costs.

Sensitivity analysis

Following the base-case analysis we explored the impact of varying uncertain model parameters and assumptions. The model was rerun for the following scenarios, most of which are biased against early referral:

• Annual rates of eGFR decline were doubled.

• The risk of ACR > 30 mg/g development was set to zero.

• CVD event risks were halved.

• The effect of nephrology referral on CVD events was set to zero.

• The effect of nephrology referral on eGFR decline (CKD progression) was set to zero.

• The effect of nephrology referral on CVD events and eGFR decline was simultaneously halved.

• The effects of nephrology referral on CVD events and eGFR decline were halved and constrained to last 5 years (the median length of follow-up in the study by Orlando and colleagues). 111

• Costs for standard care were set at their upper limit.

• Costs of care under early nephrology referral were doubled.

• Costs associated with fatal and non-fatal CVD events were varied within the 95% confidence limits of their assigned distributions (see Probabilistic sensitivity analysis).

The cost of caring for people with ESRD was also subjected to sensitivity analysis. For the lower limit we applied the lowest dialysis cost estimates reported by Baboolal and colleagues207 and also assumed that only 70% of patients with ESRD would end up on dialysis, i.e. 30% would be managed conservatively. For the upper limit we applied high dialysis cost estimates207 and assumed that 60% of patients would commence dialysis within a year of transiting to ESRD, and that the proportion on dialysis would subsequently increase to 90%. In addition, we explored the impact of allowing for increasing cardiovascular event rates by CKD stage, and assessed the potential impact of allowing for non-linear rates of eGFR decline (by applying higher rates of eGFR decline for people with stage 4 CKD). Finally, we considered a scenario where individuals with an ACR ≥ 30 mg/g were considered separately from the main cohort.

Probabilistic sensitivity analysis

The impact of joint uncertainty across all model parameters was assessed using probabilistic sensitivity analysis. Monte Carlo simulation was employed whereby values were simultaneously selected for each parameter from an assigned distribution and the results recorded. The process was repeated 1000 times to give an estimate of the sampling distribution of cost and effect differences between the referral strategies. These results were then used to generate cost-effectiveness acceptability curves.

Gamma distributions were fitted to all cost parameters. As we had no information on the statistical precision of our CKD cost estimates (under standard practice), base-case costs were treated as means for these parameters and variances were selected so that our feasible high cost estimates fell below the 97.5th percentiles of the resultant distributions. For CKD costs under nephrology referral, distributions were centred on our base-case estimates and assigned variances so that feasible high estimates fell within the 95% CIs of resultant distributions. The same principals were followed when assigning distributions to costs associated with ESRD. For costs associated with cardiovascular events, we applied gamma distributions using means and reported standard errors. 201

For probabilities of CKD stage progression we assigned beta distributions. We centred these distributions on the base-case estimates and selected variances so that high estimates fell within the 97.5th percentiles of the resultant distributions. For probabilities of developing microalbuminuria and proteinuria, we assigned beta distributions using point estimates and CIs reported for individuals with diabetes. 203 For RRs of CVD events conveyed by CKD, ACR > 30 mg/g, and ACR > 300 mg/g, we assigned log normal distributions using reported point estimates and CIs. 84 The same approach was used to assign a log normal distribution to the HR associated with nephrology referral. 111 For CVD event fatality rates, and the ratio of major to minor CVD events (used in the cost calculation), we reduced and increased point estimates by 50% and assigned uniform distributions.

Given the uncertainty underlying our effectiveness estimate, we also ran a probabilistic sensitivity analysis where the distribution for this parameter was centred on a 10% risk reduction (as opposed to a 20% risk reduction).