Looking to the future: predicting renal replacement outcomes in a large community cohort with chronic kidney disease
Nephrol Dial Transplant
Looking to the future: predicting renal replacement outcomes in a large community cohort with chronic kidney disease
Angharad Marks 0 1 2
Nicholas Fluck 0 1
Gordon J Prescott 0 2
Lynn Robertson 0 2
William G Simpson 0 1
William Cairns Smith 0 2
Corri Black 0 1 2
0 The Author 2015. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved
1 NHS Grampian , Aberdeen , UK
2 Aberdeen Applied Renal Research Collaboration, Division of Applied Health Sciences, University of Aberdeen , Aberdeen , UK
Correspondence and offprint requests to: Angharad Marks; E-mail: I N T R O D U C T I O N
chronic kidney disease; outcome; risk prediction
-
A B S T R AC T
Background. Chronic kidney disease (CKD) is common and
important due to poor outcomes. An ability to stratify CKD
care based on outcome risk should improve care for all. Our
objective was to develop and validate 5-year outcome prediction
tools in a large population-based CKD cohort. Model
performance was compared with the recently reported ‘kidney failure
risk equation’ (KFRE) models.
Methods. Those with CKD in the Grampian Laboratory
Outcomes Mortality and Morbidity Study-I (3396) and -II (18 687)
cohorts were used to develop and validate a renal replacement
therapy (RRT) prediction tool. The discrimination, calibration
and overall performance were assessed. The net reclassification
index compared performance of the developed model and the
3- and 4-variable KFRE model to predict RRT in the validation
cohort.
Results. The developed model (with measures of age, sex,
excretory renal function and proteinuria) performed well with a
Cstatistic of 0.938 (0.918–0.957) and Hosmer–Lemeshow (HL)
χ2 statistic 4.6. In the validation cohort (18 687), the developed
model falsely identified fewer as high risk (414 versus 3278
individuals) compared with the KFRE 3-variable model
(measures of age, sex and excretory renal function), but had more
false negatives (58 versus 21 individuals). The KFRE 4-variable
model could only be applied to 2274 individuals because of a
lack of baseline urinary albumin creatinine ratio data, thus
limiting its use in routine clinical practice.
Conclusions. CKD outcome prediction tools have been
developed by ourselves and others. These tools could be used to stratify
care, but identify both false positives and -negatives. Further
refinement should optimize the balance between identifying those
at increased risk with clinical utility for stratifying care.
In the UK, over 3.6 million adults are estimated to have chronic
kidney disease (CKD) [1]; 23 million in the USA [2, 3]. While
many remain undiagnosed, recognition is improving rapidly
and more are coming to medical attention [4]. People with
CKD are at increased risk of mortality, cardiovascular disease
and progressive kidney function decline [leading to renal
replacement therapy (RRT)] [5, 6]. Progression to poor outcomes
is highly variable and only a small proportion will require RRT
[4]. Important opportunities therefore exist for improving care,
maintaining function, reducing progression and minimizing
and managing complications. People with CKD often present
to primary care are often elderly and frequently have multiple
morbidities. An ability to identify which patients would benefit
most from interventions including referral to specialist services
is key. Stratification of patients by predicted risk of future
outcomes would potentially enable care pathways to be optimized
[7].
The literature regarding prognosis prediction in CKD has
been recently reviewed [8, 9] and the processes involved
summarized [10]. Of the studies identified in the reviews, 10
predicted progression of CKD or renal failure, three
cardiovascular events and five all-cause mortality. All but two
of the progression prediction models [11, 12] were developed
in patients referred to nephrology services. Thus, model utility
in other contexts, particularly the community, is not clear [13,
14]. Some models used variables not routinely available in
clinical practice, e.g. cystatin C. Very few models have been
externally validated. None have been applied in clinical practice.
Although Tangri et al. [15] developed models using a
population referred to nephrology services, these models contain
commonly available variables (including measures of age, sex and
excretory renal function), which were externally validated by
the authors in another referred population; and model
performance has since been reported in 595 referred individuals
[14]. Unlike many prediction model studies, model
performance metrics including discrimination, calibration and
reclassification [10] were reported. Thus, these ‘kidney failure risk
equation’ (KFRE) models have the best evidence for their use
to predict risk in CKD [15].
We aimed to report the development and validation of
models to predict first outcome (mortality or RRT initiation) by 5
years in a large community-based CKD cohort. We compared
the perform (...truncated)