Validation of a United Kingdom Model to Predict Mortality in Incident Dialysis Patients in the Dialysis Outcomes and Practice Patterns Study Cohort: Introduction of a Clinical Risk Score.

To the Editor:

Patients with kidney failure represent a heterogeneous group, in which many factors, including age, the cause of kidney disease, comorbidities, and so forth, result in a wide variation of mortality risk. A number of predictive models are available to assess the patient’s individual risk of mortality at the time of dialysis initiation. However, few are applicable to patients treated with hemodialysis (HD) and peritoneal dialysis, many include nonroutinely available variables, and most importantly, few have been externally validated in independent cohorts, thus leaving their applicability and validity in clinical practice unanswered. Previously, we published a model to predict mortality with high accuracy in incident dialysis patients in the United Kingdom Renal Registry (UKRR) by employing routinely available variables (age, sex, race, and cause of kidney disease), comorbidities (diabetes, cardiovascular disease, and smoking), and laboratory measures (creatinine, hemoglobin, albumin, and calcium). Here, we briefly report the external validation of a United Kingdom (UK) model in the international cohort of the Dialysis Outcomes and Practice Patterns Study (DOPPS). We also translated the model into a clinical risk score.

The validation data set consisted of 3,612 patients participating in DOPPS phase 2 (enrollment 2002-2004) who received HD treatment 3 months after dialysis initiation, similar to the UK model., We restricted the UKRR data set to HD patients because peritoneal dialysis patients are not enrolled in DOPPS. The UK model was validated by exploring C-statistics (discrimination) and d’Agostino and Nam χ2 statistics (calibration) for 1- and 3-year mortality, as the data allowed. The original (fixed) coefficients were applied; yet, the baseline hazard function of DOPPS and its subsets (North America, Europe, Japan, and Australia/New Zealand) were considered (ie, recalibration). Finally, the UK model was transformed into a clinical score (see Item S1 for details in methodology and statistical analysis).

Patient characteristics and the outcomes of DOPPS, DOPPS by continent, and the HD cohort of UKRR are displayed in Table 1. A total of 675 (18.8%) patients from DOPPS and 1,193 (31.7%) patients from UKRR died within 3 years (1-year mortality, 355 [9.8%] and 468 [12.4%], respectively; Fig S1). The UK prediction model proved to have high accuracy in DOPPS (C-statistic, 0.74; χ2 statistic, 9.3) for 1-year mortality, while discrimination and calibration were adequate in patients from Europe (C-statistic, 0.74; χ2 statistic, 7.1), Japan (C-statistic, 0.82; χ2 statistic, 2.6), and Australia/New Zealand (C-statistic, 0.80; χ2 statistic, 2.9) but modest in North American patients (C-statistic, 0.69; χ2 statistic, 17.7). The model also indicated better performance in European patients for 3-year mortality (C-statistic, 0.71; χ2 statistic, 15.5) than in North American patients (C-statistic, 0.68, χ2 statistic, 8.79) (Table S1, Fig S2). We translated the UK prediction model into a clinical risk score (Fig 1), which indicated adequate performance in the original UKRR development (C-statistic, 0.74; χ2 statistic, 2.3) and validation (C-statistic, 0.72; χ2 statistic, 1.0) data sets, in HD as well as peritoneal dialysis patients (Table S2).

Table 1

Patient Characteristics of DOPPS Phase 2 and the HD Cohort of the UK Renal Registry

	DOPPS 2n = 3,612	DOPPS 2 Can/USn = 1,241	DOPPS 2 European = 1,776	DOPPS 2 Japann = 431	DOPPS 2 Aus/NZn = 164	P ValueAcross DOPPS 2 Continents	UK Renal Registry - HD n = 3,769	P Value DOPPS 2 vs UKRR-HD	P Value DOPPS 2 Europe vs UKRR-HD
Age	66 (54-75)	64 (53-75)	68 (55-75)	64 (55-73)	62 (48-72)	<0.001	66 (53-75)	0.37	0.002
Male sex	60.3%	56.1%	61.7%	66.8%	60.7%	<0.001	61.6%	0.28	0.94
BMI, kg/m2	24.5 (21.5-28.3)	26.1 (22.4-30.9)	24.6 (21.8-27.7)	21.2 (19.3-23.2)	25.4 (23.0-28.6)	<0.001	25.6 (22.3-30.0)	<0.001	<0.001
Race
White	74.2%	67.0%	96.5%	0%	82.3%	<0.001	72.9%	<0.001	<0.001
Black	8.8%	23.1%	1.7%	0%	0%		4.5%
Chinese/Japanese	13.5%	3.0%	0.8%	99.8%	3.1%		0.6%
Asian (Indian subcontinent)	0.3%	0.5%	0.1%	0 %	1.8%		8.3%
Other/unknown	3.2% / 0%	6.4%	0.8%	0.2%	12.8%		2.2% / 11.5%
Cause of kidney disease
Diabetes	29.9%	38.9%	21.5%	41.1%	22.6%	<0.001	20.1%	<0.001	<0.001
Glomerulonephritis	13.5%	6.9%	13.6%	32.0%	13.4%		10.0%
Polycystic kidney disease	4.5%	2.8%	5.9%	3.0%	6.7%		6.1%
Pyelonephritis	3.2%	1.6%	4.5%	2.3%	4.3%		8.8%
Renovascular disease	18.9%	25.1%	18.7%	3.3%	14.0%		16.8%
Other	12.7%	9.4%	16.2%	5.6%	18.3%		15.8%
Uncertain/missing	17.4%	15.2%	19.8%	12.8%	20.3%		21.9%
Modality changeb	2.3%	1.8%	2.6%	1.1%	5.8%	0.01	1.4%	0.06	0.003
Vascular accessc
Fistula	43.3%	17.2%	50.8%	82.4%	53.0%	<0.001	NA	--	--
Synth. graft	6.2%	11.4%	3.5%	2.6%	6.0%
Bov. graft	0.4%	0.9%	0.1%	0%	0%
Cuffed cath.	31.6%	54.5%	23.8%	0.2%	27.5%
Temp. cath.	18.1%	15.8%	21.5%	13.2%	13.4%
Other	0.5%	0.3%	0.4%	1.6%	0%
Comorbidities
Diabetesd	44.3%	58.4%	34.2%	47.6%	38.4%	<0.001	29.1%	<0.001	<0.001
CVDe	47.7%	55.2%	47.0%	30.8%	43.8%	<0.001	37.7%	<0.001	<0.001
Ischemic heart disease	32.0%	39.3%	31.1%	16.4%	28.8%	<0.001	na
Cerebrovascular disease	14.9%	16.3%	14.7%	12.6%	12.2%	0.20	na
Peripheral artery disease	25.0%	28.7%	26.3%	8.9%	25.6%	<0.001	na
Smokingf	18.6%	18.8%	16.9%	24.1%	20.1%	<0.001	16.5%	<0.001	<0.001
Laboratoryg
Hemoglobin, g/dL	10.8 ± 1.8	11.5 ± 1.7	10.7 ± 1.6	9.6 ± 1.5	10.6 ± 1.7	<0.001	11.0 ± 1.7	<0.001	<0.001
Albumin, g/L	3.6 (3.2-3.9)	3.6 (3.2-3.9)	3.6 (3.2-3.9)	3.7 (3.3-4.0)	3.5 (3.2-3.8)	<0.001	3.6 (3.2-3.9)	0.58	0.55
Calcium, mg/dL	8.94 (8.42-9.50)	8.94 (8.42-9.42)	9.10 (8.54-9.66)	8.42 (7.90-8.82)	9.22 (8.58-9.86)	<0.001	9.50 (9.06-10.06)	<0.001	<0.001
Creatinine, mg/dL	6.7 (5.2-8.7)	6.1 (4.6-8.1)	6.7 (5.3-8.5)	8.0 (6.5-9.9)	7.3 (6.0-9.6)	<0.001	7.2 (5.7-8.9)	<0.001	<0.001
Outcomes within 3 y
Death	18.8%	22.5%	20.0%	5.4%	12.3%	<0.001	31.7%	<0.001	<0.001
End of observation	61.1%	56.7%	59.8%	84.6%	46.0%		49.0%
Kidney transplantation	6.0%	4.9%	8.1%	0.7%	6.1%		9.9%
Recovery of renal function	1.0%	1.3%	1.1%	0.2%	0.6%		1.3%
Lost to follow-uph	10.4%	10.9%	9.3%	8.6%	22.7%		1.1%
Switch to PD	2.8%	3.7%	1.8%	0.2%	12.3%		7.1%

Note: Data are %, median (interquartile range) or mean ± standard deviation. P values of Χ2-test, Kruskal-Wallis-test, and ANOVA, as appropriate. Abbreviations: Aus, Australia; Can, Canada; HD, hemodialysis; NA, not available; NZ, New Zealand; PD, peritoneal dialysis; RRT, renal replacement therapy; US, United States.

Belgium, France, Germany, Italy, Spain, Sweden, United Kingdom.

Change from PD to HD within the first 90 days of RRT.

At enrollment DOPPS.

Including diabetes as cause of kidney disease.

Definitions of DOPPS (Cerebrovascular disease; Ischemic Heart Disease: angina, previous myocardial infarction, previous CABG or angioplasty; Peripheral Vascular Disease: PVD diagnosis, claudication, non-coronary angioplasty. vascular graft or aneurysm, amputation for PVD) and UKRR (any of angina, previous myocardial infarction, previous CABG or angioplasty, cerebrovascular disease, claudication, ischemic or neuropathic ulcers, non-coronary angioplasty, vascular graft or aneurysm, amputation for PVD).

Active smoker or stopped <1 year ago.

Measurements of treatment quarter 2, except creatinine.

Lost to follow-up, withdrawal of RRT, change to non DOPPS dialysis unit (DOPPS only).

Figure 1

(A) Clinical risk score and (B) estimated probability of death within 3 years by risk score values (green) and histogram or the number of observations (gray). The risk score points for the individual patient are to be summed up to result in a total risk score, which can then be compared with the probability of death in (B). Abbreviations: CVD, cardiovascular disease; GN, glomerulonephritis; HD, hemodialysis; PKD, polycystic kidney disease; PD, peritoneal dialysis; RVD, renal vascular disease; UK, United Kingdom.

Our analyses showed that basic patient characteristics and laboratory variables are sufficient to accurately predict mortality in incident dialysis patients in various international settings. The UK prediction model was also externally validated in the NECOSAD cohort, in which, however, the more recent AROii model, which was developed in European HD patients, indicated higher performance measures., Yet, conclusions drawn from the results of a prediction model should be applied to patients with caution because to our knowledge, none of these standardized models have ever been tested prospectively and in a randomized controlled trial to guide clinical decision making regarding whether to apply more or less therapy. However, the proposed UK clinical risk score can help researchers and clinicians in the field of HD and peritoneal dialysis to describe the underlying baseline mortality risk at the time of dialysis inception.