Prediction models for the mortality risk in chronic dialysis patients: a systematic review and independent external validation study

Figures & data

Figure 1 Flowchart of study selection for systematic review and external validation.

Abbreviations: ESRD, end-stage renal disease; NECOSAD, Netherlands Cooperative Study on the Adequacy of Dialysis.

Table 1 General characteristics of models to predict risk of all-cause mortality in dialysis patients

Reference	Events:sample size	Study population	Patient inclusion time span	Outcome, prediction horizon	Statistical model	Validation	Predictors, n	Discrimination, C-statistic (CI) (of derivation)	Discrimination, C-statistic (CI) (of validation)	Calibration plot/table, slope	Presentation of model
Hemke et al,^Citation43 The Netherlands, “easy” model	1,225	Incident dialysis patients	1997–2007	3-, 5-, 10-year mortality	Cox	Internal split-sample validation (610 patients)	10	–	0.79, 0.79, 0.78a	Yes	Full prediction formula provided
Cao et al,^Citation40 China	371:5,405	Incident CAPD patients	2012	1-year mortality	Cox	External validation (1,764 patients)	6	0.71 (0.6–0.83)	0.72 (0.63–0.81)	–	β-value per predictor given
Doi et al,^Citation44 Japan	62:688	Incident HD patients	2006–2011	1-year mortality	Logistic	Internal bootstrap validation	6	0.83	0.83 (0.79–0.89)	Yes	β-value per predictor given and score chart
Floege et al,^Citation7 Germany*	1,060, 1,714:9,722b	Incident and prevalent HD patients	2007–2009	1- and 2-year mortality	Cox	External validation (10,615 patients)	12, 13b	–	0.73, 0.72b	Yes	Full prediction formulae provided
Chua et al,^Citation37 Singapore	169:983	Incident dialysis patients	2005–2010	1-year mortality	Logistic and linear	–	7	0.74, 0.74, 0.75 (logistic, linear, score)	–	–	Logistic, linear model, and score chart presented
Zhao et al,^Citation11 China, model 3	175:903	Incident CAPD patients	2006–2011	2-year mortality	Cox	Internal split-sample validation (451 patients)	10	0.79	0.76 (0.68–0.84)	Yes	Full prediction formula provided
Hemke et al,^Citation42 The Netherlands	–, 3502,** 4549**:6934 c	Incident RRT patients	1995–2005	3-, 5-, 10-year mortality	Cox	Internal split-sample validation (6,934 patients)	4	–	0.71 (0.7–0.72), 0.72 (0.7–0.72), 0.72 (0.71–0.73)c	Yes	Full prediction formula provided
Shiotsu et al,^Citation39 Japan	50:550	Prevalent HD patients	2009	2-year mortality	Cox	Bootstrapping and external validation (303 patients)	4	0.73 (0.66–0.8)	0.72 (0.63–0.82)	Yes	β-value per predictor given and score chart
Holme et al,^Citation9 Norway*	880:1,868	Prevalent HD patients	2003–2004	3-year mortality	Cox	Internal split-sample validation (905 patients)	5	0.73 (0.71–0.75)	0.72 (0.69–0.76)	Yes	Full prediction formula provided
Quinn et al,^Citation10 Canada	1,326:8,103	Incident dialysis patients	1998–2005	1-year mortality	Logistic	Internal split-sample validation, twice (4,052 and 4,051 patients) and bootstrapping	15	0.77	0.76 and 0.76	Yes	Full prediction formula provided and score chart
Wagner et al,^Citation8 UK, model 3*	1,078:3,631	Incident dialysis patients	2002–2004	3-year mortality	Cox	Internal split-sample validation (1,816 patients)	11	0.75 (0.73–0.77)	0.73 (0.71–0.76)	Yes	HR per predictor given***
Cohen et al,^Citation38 USA	43:512	Prevalent HD patients	2006–2007	6-month mortality	Cox	External validation (514 patients)	5	0.87 (0.82–0.92)	0.8 (0.73–0.88)	–	HR per predictor given
Mauri et al,^Citation45 Spain*	558:3,455	Incident HD patients	1997–2003	1-year mortality	Logistic	Internal split-sample validation (2,283 patients)	10	0.78	0.78	Yes	Full prediction formula provided
Geddes et al,^Citation3 UK, MVR model*	351,** 837**:1,139d	Incident RRT patients	1997	1- and 5-year mortality	Cox	Internal split-sample validation (1,171 patients)	17	–	–	Yes	Full prediction formula provided****
Foley,^Citation41 Canada	73:325	Incident dialysis patients	1980–1991	6-month mortality (logistic)	Logistic and Cox model	–	14	–	–	Yes	Full prediction formula (logistic), HRs for Cox and score chart
Hutchinson et al,^Citation55 Canada*	101**:220	Incident HD patients	1970–1975	5-year mortality	Cox	–	3	–	–	Yes	Full prediction formula and score chart

Notes:

* Externally validated in the current independent study;

** deducted/calculated based on information available in article;

*** baseline hazard provided upon request;

**** shrinkage factor provided upon request.

a 3, 5, 10-year mortality, respectively;

b 1 and 2-year mortality events respectively;

c 3, 5 and 10 year mortality events, respectively;

d 1, and 5-year mortality events, respectively.

“–” not provided.

Abbreviations: CAPD, continuous ambulatory peritoneal dialysis; HD, hemodialysis; HR, hazard ratio; MVR, multivariate regression; RRT, renal replacement therapy.

Table 2 Baseline characteristics of external validation cohort (NECOSAD) at 3 months after dialysis initiation

Characteristics	Missing	Total (n=1,943)	HD patients (n=1,266)	PD patients (n=677)
Age, years	0.2%	62.8 (49.8–72.2)	66.7 (55.6–74.4)	54.3 (43.0–65.1)
Sex (% male)	0.2%	62.3%	59.4%	67.8%
Current smokers	10.2%	22.5%	21.2%	24.8%
Ethnicity (% Caucasian)	0.2%	91.9%	92.0%	91.6%
Primary kidney disease	0.2%
Diabetes mellitus		14.4%	14.1%	15.1%
Glomerulonephritis		12.4%	9%	18.8%
Renal vascular disease		16.9%	19.5%	11.8%
Other		56.3%	57.4%	54.3%
Comorbidities
Cardiovascular disease	9.8%	35.3%	40.9%	25.3%
Diabetes mellitus	0.2%	5.6%	6.8%	3.3%
Cancer	10.8%	9.6%	11.9%	5.5%
Medication
ACE inhibitors	0.6%	24.5%	22.4%	28.6%
Calcium antagonists	0.6%	34.1%	32.4%	37.3%
β-blockers	0.6%	31.7%	30.3%	34.2%
Diuretics	0.6%	27.3%	27.0%	28.0%
Erythropoietin	1.5%	85.3%	91.7%	73.2%
Laboratory
Albumin, g/L	5.8%	36.0 (5.3)	35.8 (5.3)	36.2 (5.3)
mGFR, mL/min/1.73 m^2	21.0%	3.4 (1.8–5.4)	3.1 (1.6–4.9)	4.1 (2.2–6.3)
Hemoglobin, mmol/L	1.6%	6.9 (1.0)	6.7 (0.9)	7.4 (1.0)
HS-CRP mg/L	56.6%	5.0 (3.0—14.0)	6.0 (3.0–16.0)	4.0 (3.0–9.0)
BMI (kg/m^2)	10.1%	24.7 (4.2)	24.7 (4.4)	24.8 (3.9)
Vascular access mode	4.6%	NA		NA
Catheter		NA	21.0%	NA
Graft/fistula		NA	79.0%	NA

Notes:

Continuous variables presented as mean values with standard deviations for normally distributed variables. Continuous variables presented as median values with interquartile ranges for non-normal distributed variables. Categorical variables presented as valid percentages.

Abbreviations: NECOSAD, Netherlands Cooperative Study on the Adequacy of Dialysis; HD, hemodialysis; PD, peritoneal dialysis; mGFR, measured glomerular filtration rate (combined creatinine and urea clearance in 24-hour urine samples at 3 months); BMI, body-mass index; NA, not applicable; HS-CROP, high sensitivity C-reactive protein.

Figure 2 C-statistics reported in the original study compared to those found through external validation (for the original time frame and original dialysis modality).

Note: The validation in the original study of Floege et al⁷ was an external validation also.

Figure 3 Kaplan–Meier survival curves by quartile of the prognostic index.

Table 3 Discrimination results of external validation for HD and PD patients

Reference	Time frame	Original population	Discrimination: C-statistic,* HD and PD
Floege et al^Citation7	1 year	HD	0.740 (0.738–0.742)
Floege et al^Citation7	2 years	HD	0.740 (0.737–0.742)
Holme et al^Citation9	3 years	HD	0.734 (0.730–0.737)
Wagner et al^Citation8	1 year	HD and PD	0.752 (0.750–0.753)
	2 years		0.738 (0.737–0.738)
	3 years		0.730 (0.729–0.731)
Mauri et al^Citation45	1 year	HD	0.728 (0.724–0.730)
Geddes et al^Citation3	1 year	HD and PD	0.721 (0.717–0.723)
	2 years		0.706 (0.704–0.708)
	3 years		0.705 (0.702–0.707)
	5 years		0.698 (0.695–0.700)
Hutchinson et al^Citation55	1 year	HD	0.710 (0.708–0.711)
	2 years		0.702 (0.701–0.706)
	3 years		0.700 (0.699–0.702)
	5 years		0.697 (0.696–0.699)

Note:

* C-statistics presented as medians and interquartile ranges from pooled imputation results.

Abbreviations: HD, hemodialysis; PD, peritoneal dialysis.

Table 4 Discrimination results of external validation stratified for dialysis modality

Reference	Time frame	Original population	Discrimination: C-statistic*
			HD	PD
Floege et al^Citation7	1 year	HD	0.701 (0.699–0.709)	0.778 (0.768–0.780)
Floege et al^Citation7	2 years	HD	0.711 (0.705–0.718)	0.775 (0.767–0.777)
Holme et al^Citation9	3 years	HD	0.705 (0.700–0.708)	0.767 (0.765–0.768)
Wagner et al^Citation8	1 year	HD and PD	0.712 (0.711–0.715)	0.709 (0.789–0.793)
	2 years		0.712 (0.709–0.713)	0.770 (0.770–0.775)
	3 years		0.703 (0.702–0.706)	0.763 (0.762–0.767)
Mauri et al^Citation45	1 year	HD	0.670 (0.668–0.675)	0.800 (0.791–0.808)
Geddes et al^Citation3	1 year	HD and PD	0.676 (0.671–0.677)	0.772 (0.769–0.774)
	2 years		0.671 (0.668–0.674)	0.748 (0.746–0.749)
	3 years		0.667 (0.664–0.669)	0.750 (0.748–0.753)
	5 years		0.663 (0.660–0.665)	0.737 (0.735–0.740)
Hutchinson et al^Citation55	1 year	HD	0.667 (0.666–0.671)	0.767 (0.763–0.771)
	2 years		0.667 (0.666–0.669)	0.752 (0.748–0.754)
	3 years		0.663 (0.661–0.665)	0.749 (0.748–0.753)
	5 years		0.663 (0.662–0.665)	0.742 (0.741–0.745)

Note:

* C-statistics presented as medians and interquartile ranges from pooled imputation results.

Abbreviations: HD, hemodialysis; PD, peritoneal dialysis.

Figure 4 Calibration plots per model.

Abbreviations: HD, hemodialysis; PD, peritoneal dialysis.

Table 5 Calibration results of external validation for HD and PD patients

Reference	Time frame	Original population	Calibration: HD and PD
			Calibration slope (SE)	Calibration-in-the-large*
Floege et al^Citation7	1 year	HD	0.71 (0.04)	14.0% vs 13.5%
Floege et al^Citation7	2 years	HD	0.83 (0.04)	22.1% vs 26.9%
Holme et al^Citation9	3 years	HD	0.96 (0.05)	47.6% vs 37.2%
Wagner et al^Citation8	1 year	HD and PD	0.82 (0.04)	4.0% vs 13.5%
	2 years		0.82 (0.04)	8.9% vs 26.9%
	3 years		0.82 (0.04)	14.2% vs 37.2%
Mauri et al^Citation45	1 year	HD	0.42 (0.04)	12.0% vs 12.6%
Geddes et al^Citation3	1 year	HD and PD	0.82 (0.05)	41.7% vs 13.5%
	2 years		0.82 (0.05)	59.3% vs 26.9%
	3 years		0.82 (0.05)	69.7% vs 37.2%
	5 years		0.82 (0.05)	80.9% vs 58.2%
Hutchinson et al^Citation55	1 year	HD	0.50 (0.03)	45.3% vs 13.5%
	2 years		0.50 (0.03)	51.4% vs 26.9%
	3 years		0.50 (0.03)	60.3% vs 37.2%
	5 years		0.50 (0.03)	70.8% vs 58.2%

Note:

* Predicted vs observed risk of death.

Abbreviations: HD, hemodialysis; PD, peritoneal dialysis; SE, standard error.

Table 6 Calibration results of external validation stratified for dialysis modality

Reference	Time frame	Original population	Calibration: HD		Calibration: PD
			Calibration slope (SE)	Calibration-in-the-large*	Calibration slope (SE)	Calibration-in-the-large*
Floege et al^Citation7	1 year	HD	0.67 (0.04)	16.7% vs 16.9%	0.76 (0.08)	9.1% vs 7.1%
Floege et al^Citation7	2 years	HD	0.79 (0.05)	26.2% vs 30.9%	0.91 (0.08)	14.4% vs 19.5%
Holme et al^Citation9	3 years	HD	0.92 (0.06)	52.8% vs 42.2%	1.04 (0.09)	37.8% vs 27.3%
Wagner et al^Citation8	1 year	HD and PD	0.85 (0.05)	5.0% vs 16.9%	0.87 (0.08)	2.1% vs 7.1%
	2 years		0.85 (0.05)	11.1% vs 30.9%	0.87 (0.08)	4.7% vs 19.5%
	3 years		0.85 (0.05)	17.6% vs 42.2%	0.87 (0.08)	7.7% vs 27.3%
Mauri et al^Citation45	1 year	HD	0.34 (0.05)	16.9% vs 15.7%	0.78 (0.15)	2.9% vs 6.7%
Geddes et al^Citation3	1 year	HD and PD	0.75 (0.06)	47.6% vs 16.9%	1.02 (0.10)	30.7% vs 7.1%
	2 years		0.75 (0.06)	66.0% vs 30.9%	1.02 (0.10)	46.8% vs 19.5%
	3 years		0.75 (0.06)	76.1% vs 42.2%	1.02 (0.10)	57.7% vs 27.3%
	5 years		0.75 (0.06)	86.1% vs 64.2%	1.02 (0.10)	71.2% vs 44.3%
Hutchinson et al^Citation55	1 year	HD	0.46 (0.03)	51.6% vs 16.9%	0.56 (0.05)	33.6% vs 7.1%
	2 years		0.46 (0.03)	58.1% vs 30.9%	0.56 (0.05)	38.8% vs 19.5%
	3 years		0.46 (0.03)	67.3% vs 42.2%	0.56 (0.05)	47.2% vs 27.3%
	5 years		0.46 (0.03)	77.5% vs 64.2%	0.56 (0.05)	58.4% vs 44.3%

Note:

* Predicted vs observed risk of death.

Abbreviations: HD, hemodialysis; PD, peritoneal dialysis; SE, standard error.

HemkeACHeemskerkMBvan DiepenMDekkerFWHoitsmaAJImproved mortality prediction in dialysis patients using specific clinical and laboratory dataAm J Nephrol201542215816726406283

 PubMed Web of Science ®Google Scholar

CaoXYZhouJHCaiGYPredicting one-year mortality in peritoneal dialysis patients: an analysis of the China peritoneal dialysis registryInt J Med Sci201512435436126019685

 PubMed Web of Science ®Google Scholar

DoiTYamamotoSMorinagaTSadaKEKuritaNOnishiYRisk score to predict 1-year mortality after haemodialysis initiation in patients with stage 5 chronic kidney disease under predialysis nephrology carePloS One2015106e012918026057129

 PubMed Web of Science ®Google Scholar

FloegeJGillespieIAKronenbergFDevelopment and validation of a predictive mortality risk score from a European hemodialysis cohortKidney Int2015875996100825651366

 PubMed Web of Science ®Google Scholar

ChuaHRLauTLuoNPredicting first-year mortality in incident dialysis patients with end-stage renal disease: the UREA5 studyBlood Purif2014372859224589505

 PubMed Web of Science ®Google Scholar

ZhaoCLuoQXiaXRisk score to predict mortality in continuous ambulatory peritoneal dialysis patientsEur J Clin Invest201444111095110325263820

 PubMed Web of Science ®Google Scholar

HemkeACHeemskerkMBvan DiepenMWeimarWDekkerFWHoitsmaAJSurvival prognosis after the start of a renal replacement therapy in the Netherlands: a retrospective cohort studyBMC Nephrol20131425824256551

 PubMed Web of Science ®Google Scholar

ShiotsuYMoriYNishimuraMPrognostic utility of plasma S100A12 levels to establish a novel scoring system for predicting mortality in maintenance hemodialysis patients: a two-year prospective observational study in JapanBMC Nephrol2013141623324110

 PubMed Web of Science ®Google Scholar

HolmeIFellströmBCJardinAGSchmiederREZannadFHoldaasHPrognostic model for total mortality in patients with haemodialysis from the Assessments of Survival and Cardiovascular Events (AURORA) studyJ Intern Med2012271546347121812843

 PubMed Web of Science ®Google Scholar

QuinnRRLaupacisAHuxJEOliverMJAustinPCPredicting the risk of 1-year mortality in incident dialysis patients: accounting for case-mix severity in studies using administrative dataMedical care201149325726621301370

 PubMed Web of Science ®Google Scholar

WagnerMAnsellDKentDMPredicting mortality in incident dialysis patients: an analysis of the United Kingdom Renal RegistryAm J Kidney Dis201157689490221489668

 PubMed Web of Science ®Google Scholar

CohenLMRuthazerRMossAHGermainMJPredicting six-month mortality for patients who are on maintenance hemodialysisClin J Am Soc Nephrol201051727919965531

 PubMed Web of Science ®Google Scholar

MauriJMCleriesMVelaEDesign and validation of a model to predict early mortality in haemodialysis patientsNephrol Dial Transplant20082351690169618272779

 PubMed Web of Science ®Google Scholar

GeddesCCvan DijkPCMcArthurSThe ERA-EDTA cohort study: comparison of methods to predict survival on renal replacement therapyNephrol Dial Transplant200621494595616339161

 PubMed Web of Science ®Google Scholar

FoleyRNParfreyPSHeffertonDSinghISimmsABarrettBJAdvance prediction of early death in patients starting maintenance dialysisAm J Kidney Dis19942368368458203366

 PubMed Web of Science ®Google Scholar

HutchinsonTAThomasDCMacGibbonBPredicting survival in adults with end-stage renal disease: an age equivalence indexAnn Intern Med19829644174237039443

 PubMed Web of Science ®Google Scholar