Danube validation

From Prosit
Jump to navigationJump to search

Vlc.png This page contains development contents which may not yet represent final PROSIT Content.



External Validation

Comparison of the model structure with other models

  • Comparison of the structures of published diabetes models with Danube A (incl. purpose of a model)


CDC Diabetes Cost- Effectiveness Group, 2002
071217 structure CDC Diabetes Cost Effectiveness Group 2002.jpg

Reference: CDC Diabetes Cost-effectiveness Group Cost-effectiveness of intensive glycemic control, intensified hypertension control, and serum cholesterol level reduction for type 2 diabetes.JAMA. 2002 May 15;287(19):2542-51.

The model by the CDC Diabetes Cost- Effectiveness Group shows the following differences:

  • The model by CDC doesn't include Macroalbuminuria
  • The CDC model doesn't offer the possibility for regression from Microalbuminuria to normal and from Macroalbiminuria to Microalbuminuria, as the Danube A model does.
  • The second state of the CDC model "Clinical Nephropaty" is not taken into account in the Danube A model.
Palmer et al, 2000
080327 structure Palmer et al Switzerland.jpg

Reference: The cost-effectiveness of different management strategies for type I diabetes: a Swiss perspective.Diabetologia. 2000 Jan;43(1):13-26.

The model by Palmer et al. includes many states the Danube A model doesn't. The following shows the comparable differences between Danube A and the model by Palmer et al.:

  • The model by Palmer et al. distinguishes two types of Dialysis: Haemodialysis and Peritoneal Dialysis.
  • It also distinguishes the states of non-specific mortality and ESRD- specific mortality, whereas the Danube A model combines those two states.
  • The Palmer model takes Kidneytransplantation into account.
  • Another included state in the Palmer model ist "Graft failure".
  • The Palmer model doesn't allow for regression from "Macroalbuminuria" to "Microalbuminuria" and from "Microalbuminuria" to "No Nephropathy".
Palmer et al, 2003
071217 structure Palmer et al 2003.jpg

Reference: Palmer AJ, Annemans L, Roze S, Lamotte M, Rodby RA, Cordonnier DJ.An economic evaluation ofirbesartan in the treatment of patients with type 2 diabetes, hypertension and nephropathy:cost-effectiveness of Irbesartan in Diabetic Nephropathy Trial (IDNT) in the Belgian and French settings.Nephrol Dial Transplant. 2003 Oct;18(10):2059-66

  • The structure of this model is a little different from the Danube A model. The following points are distinguishable:
    • This model regards ESRD with transplantation and dialysis whereas the Danube A model takes only dialysis into acccount.
    • Also compared to the Danube A model the state "doubling of serum creatinine" is taken into account.
    • Regression from ESRD with Transplant to ESRD with dialysis is possible
    • Direct transitions from the normal state, which means "no Nephropathy" to any other state are possible.
Palmer et al, 2005
071219 structure Palmer et al 2005.jpg

Reference: Palmer AJ, Chen R, Valentine WJ, Roze S, Bregman B, Mehin N, Gabriel S.Cost-consequence analysis in a French setting of screening and optimal treatment of nephropathy in hypertensive patients with type 2 diabetes.Diabetes Metab. 2006 Feb;32(1):69-76

The structure of the model by Palmer at al. shows the following differences to the structure of the Danube A model:

  • The model by Palmer includes serveral states, which are not supported by the Danube A model.
    • "early overt Nephropathy"
    • "advanced overt Nephropathy"
    • "doubling of serum creatinine"
    • "ESRD treated with Transplant"
  • This model doesn't allow regression, whereas the Danube A model allows this.


Zhou et al, Diabetes Care, 2005
071206 structure Zhou et al.jpg

Reference: Zhou H, Isaman DJ, Messinger S, Brown MB, Klein R, Brandle M, Herman WH. A computer simulation model of diabetes progression, quality of life, and cost.Diabetes Care. 2005 Dec;28(12):2856-63.

The model by Zhou et al. has a very similar structure compared to the Danube A model. Several differences are obvious:

  • The Zhou model regards transplantation and dialysis whereas the Danube A model takes only dialysis into acccount.
  • In the Zhou model the progression from "normal", i.e. diabetes without nephropathy, to ESRD is a one-way course whereas the Danube A modell allows for regression in the states "macroalbuminuria" and "microalbuminuria".
  • The Danube A model takes Macroalbuminuria into account, whereas in the Zhou model it is called Proteinuria
  • In the Zhou model a direct progression from "normal" to "Proteinuria" is possible. The Danube model requires a progression to the state "Microalbuminuria" first.


Hoerger et al, 2004
071217 structure Hoerger et al, 2004.jpg

Reference: Hoerger TJ, Harris R, Hicks KA, Donahue K, Sorensen S, Engelgau M.Screening for type 2 diabetes mellitus: a cost-effectiveness analysis.Ann Intern Med. 2004 May 4;140(9):689-99.

  • The Hoerger model has a combined state "Low Microalbuminuria or high Albuminuria".
  • In the Hoerger model no regression from any state is possible, whereas the Danube A model allows for a regression from "Microalbuminuria" and "Macroalbuminuria"


CORE- Model, 2004

Reference: Palmer AJ, Roze S, Valentine WJ, Minshall ME, Foos V, Lurati FM, Lammert M, Spinas GA The CORE Diabetes Model: Projecting long-term clinical outcomes, costs and cost-effectiveness of interventions in diabetes mellitus (types 1 and 2) to support clinical and reimbursement decision-making.Curr Med Res Opin. 2004 Aug;20 Suppl 1:S5-26.

  • The CORE model regards different states of End- stage renal disease:
    • Haemodialysis
    • Peritoneal Dialysis
    • Kidney transplant

The Danube A model only takes Dialysis into account.


Eddy and Schlessinger, 2003

Reference: Eddy DM, Schlessinger L.Archimedes: a trial-validated model of diabetes.Diabetes Care. 2003 Nov;26(11):3093-101.

  • The structure of the Archimedes model is not comparable to the stucture of toe Danube A model:
    • It doesn't define states, but rather represents progression of the disease oin terms of underlying biological variables
    • Transition probabilities depend on each person and the different biological and behavioral factors

Comparison of transition probabilities with other models

  • Literature comparison with methods and transition probabilities of other published models


Modell p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12 p13
Danube A n.a Krolewski, AS et al. (1995)
Forsblom, CM et al. (1998)
Ruggenenti et al. (2004)
Newman, DJ et al. (2005)
Gaede, P et al. (2004)
Strippoli, GFM et al. (2004)
n.a Deferrari, G et al. (1998)
Strippoli, GFM et al. (2004)
Newman, DJ et al. (2005) Hovind, P et al. (2001) n.a Valmadrid, CT et al. (2000) Adler, AI et al. (2002 n.a van Dijk, PCW et al. (2005) Roper, NA et al. (2002)
Statistisches Bundesamt (2005)
Roper, NA et al. (2002)
CORE n.a DCCT 1995 n.a n.a DCCT 1995 n.a n.a n.a n.a Ritz E. et al 1996 n.a Wolfe et al. 1999 & USRDS 2001 & US Renal Registery 2002 n.a
Zhou et al n.a Gall et al. (1997) n.a n.a Ravid et al. (1993) n.a n.a n.a n.a US Renal Data System (2000) n.a US Renal Data System (2000) n.a
Hoerger n.a UKPDS 38 n.a n.a UKPDS 38 n.a n.a n.a n.a Eastman et al. (1997) & Humphrey et al. (1994) n.a n.a n.a
CDC Group * * * * * * * * * * * * *
* No specified data about the transition probabilities. The model is based on the following studies: 
  • UKPDS 33
  • WSCPS
  • Eastman et al. (1997)
  • Dong et al (1997)

Cross-Testing

  • Cross-testing of diabetes disease models

For the crosstesting the GDM, IMIB Model and the Danube Model are compared to eachother. Therefore the following input parameters were used:

    • Sex: male
    • Age: 45 and 75
    • HbA1c: 7,5% and 9,5%
    • Systolic blood pressure: 160 and 130
    • Smoker: no
                                      Survival rate        |     Macroalbuminuria
Cohort Age Blood pressure sys. HbA1c GDM[1] IMIB[2] Danube A[3] GDM IMIB Danube A
1 45 160 9.5 77.9% 51.9% 57.4% 19.0% 30.8% 17.2%
2 45 160 7.5 79.3% N.A. 61.9% 2.2% 16.5% 9.6%
3 45 130 9.5 83.0% 56.7% 57.5% 3.3% 31.3% 17.4%
4 45 130 7.5 81.9% 59.6% 62.0% 2.8% 17.4% 9.7%
5 75 160 9.5 11.8% 2.9% 1.1% 9.4% 6.7% 4.0%
6 75 160 7.5 11.8% 3.1% 1.5% 1.0% 3.4% 2.2%
7 75 130 9.5 13.3% 3.4% 1.1% 1.3% 7.2% 4.0%
8 75 130 7.5 13.6% 3.7% 1.5% 0.1% 3.7% 2.2%

Differences between the IMIB model and the Danube A model can be explained by the possibility of remission in the Danube model. The IMIB model doesn't allow for remission. Other than that can be said, that the IMIB model uses the life table of the USA, whereas the Danube A model uses the life table of Germany.

==> If we eliminate the following visible differences between the IMIB model and the Danube Model the results of the crosstesting become more similar:

  • the regressions from Macroalbuminuria to Microalbuminuria
  • the regression from Microalbuminuria to No Nephropathy
  • use the life table of the USA instead of the German


                                      Survival rate        |     Macroalbuminuria
Cohort Age Blood pressure sys. HbA1c GDM[1] IMIB[2] Danube A[3] GDM IMIB Danube A
1 45 160 9.5 77.9% 51.9% 47.34% 19.0% 30.8% 26.96%
2 45 160 7.5 79.3% N.A. 53.85% 2.2% 16.5% 15.84%
3 45 130 9.5 83.0% 56.7% 47.43 3.3% 31.3% 26.96%
4 45 130 7.5 81.9% 59.6% 53.85% 2.8% 17.4% 15.84%
5 75 160 9.5 11.8% 2.9% 0.77% 9.4% 6.7% 6.34%
6 75 160 7.5 11.8% 3.1% 1.22% 1.0% 3.4% 3.57%
7 75 130 9.5 13.3% 3,4% 0.77% 1.3% 7.2% 6.34%
8 75 130 7.5 13.6% 3.7% 1.22% 0.1% 3.7% 3.57%


[1] GDM( Global Diabetes Model) :Brown JB, Russell A, Chan W, Pedula K, Aickin M.The global diabetes model: user friendly version 3.0.Diabetes Res Clin Pract. 2000 Nov;50 Suppl 3:S15-46.
[2]IMIB
[3] Danube A

Model outcomes

  • Target: Compare the results of the Danube A model to those of clinical long-term studies
  1. Comparison of the development of the amount of dialysis patients, between the Danube A model and the Retrolective Study Self- Monitoring of Blood Glucose and Outcome in Patients with Type 2 Diabetes (ROSSO 4 [1]).

In this study the further development of diabetes was monitored retrospectively from diagnosis of type 2 diabetes. Data for more than 3000 patient files of 192 GP practices in Germany were collected.

    • Patient characteristics:
      • Cohort:3142 persons
      • Sex: 51% female
      • Initial age: 62.45 years
      • Blood pressure: 149.20 mmHg (systolic) / 86.96 mmHg(diastolic)
      • HbA1c: 7.65%
      • Initial amount of dialysis patients: 0.4%
      • No Nephropathy: 74.6%
      • Microalbuminuria: 20%
      • Macroalbuminuria: 5%
Comparison of ROSSO study results with Danube A
Year 0 1 2 3 4 5 6 7
Rosso 0.4% 0.6% 0.7% 0.8% 0.8% 0.9% 1.0% 1.1%
Danube A 0.4% 0.4% 0.5% 0.6% 0.7% 0.8% 0.9% 1.1%



  • Conclusion: As the table shows, the model forecast confirms the progression of the disease at comparable incidences. As at diagnosis of the disease only a minority of end-stage renal disease can be related to diabetes, one would expect lower incidence rates as it was observed in the patient files. The absolute deviation between observed and modelled data is never more than an absolute 0.2%.

Internal Validation

Input

Medical evidence:

medicine

The Danube A model includes the costs for the following drugs

  • Anti- diabetics
    • Insulin
    • Oral Antidiabetics
  • Blood pressure- lowering drugs
    • ACE inhibitor
    • AT1-inhibitor
    • Aldosterone inhibitor
    • Diuretics
    • Calcium antagonists
  • ASS

cohort characteristics

  1. The model offers the following cohort characteristics input parameters:
    • initial age (in years)
    • distribution between the sexes (in %)
    • duration of diabetes (in years)
    • smoker/ nonsmoker
    • HbA1c level (in %)
    • blood pressure (in mmgH)
    • BMI (in kg/sqm)
    • Anti- hypertensive treatment (in %)
    • ACE inhibitors (in %)
  2. The Danube A model has an age restriction. It works only for ages between 45 and 75. This is because of the age range of most Diabetes studies.

Economic evidence: cost data set, discount factors, inflation rates

  • The model allows the user to enter the desired discounting rate for the costs per year and a discounting rate for the outcome per year. Default values are 3,5%

Model

  • Correct arithmetics: plausibility of calculated results, correct implementation of formulas, algorithms, independence of variables, internal links etc.:
    • The blood pressure is only regarded in state p7( Macroalbuminuria -> Microalbuminuria) of the Danube Model.
    • The states p9 (Macroalbuminuria -> Death) and p6 (Microalbuminuria -> Death) both take the excess mortality into account, even though they both include p13 (No Nephropathy -> Death), which already includes the excess mortality. ==> Excess mortality is used twice.
  • Method: techniques
    • mathematical method
    • medical interventions
    • economical perspectives
  • Documentation: transparency, proof, development („the documentation is the model“)

Output

  • Correct reporting of model results: output tables, diagrams

Face Validity

Behaviour of the finished model with regard to model outcomes:

  • Extrem value analysis

Sensitivity Analysis

For the Sensitivity Analysis the input parameters were varied for +10% and -10%. The Tornado diagrams below show, which parameters have an extraordinary impact on the results on the development of Diabetes modelled with Danube A .

SenAnalyse LE.JPG

SenAnalyse ESRD.JPG

SenAnalyse Kosten.JPG


  • Interfaces & model handling: * Workflow logic, independence of user, reproducibility of results

Additional Validation Issues

Direct comparison of the two Danube models (2006 beta version vs. 2008 release candidate)

  • The structure of the Danube model from 2006 is identical with the structure of the Danube model 2008. They both regard the five same states of 'no nephropathy', 'Microalbuminuria', 'Macroalbuminuria', 'ESRD' and 'Death'. In both models a regression from 'Microalbuminuria' to 'no nephropathy' and from 'Macroalbuminuria' to 'Microalbuminuria' is possible.
  • The time horizon in the model 2006 is 40 years whereas the model 2008 has a time horizon of 60 years.
  • Cycle length of the Markov- Cycle is one year in both models.
  • The three main assumptions, such as the assumption that "Risks in Type 1 Diabetes and Type 2 Diabetes are similar", or the assumption that "Progression from micro- to macroalbuminuria does not depent on the duration of microalbuminuria", are the same.
  • Both models have the same amount of transition probabilities (p1-p13), which are calculated the same way.
  • Input-Parameters are the same
  • Costs:
    • ----------Cost for Drugs includes AT1-inhibitors in the model 2008------??
    • Costs for monitoring still included in the model 2006 the monitoring of the HbA- level.
    • The average German base-rate in the Danube A model of 2006 was, based on data of 2005, 2721,12€ whereas the model 2008 has, based on the data of 2006,a base-rate of 2756,62€
    • The calculation of the costs for drugs in the model of 2008 includes the manufacturer and denotation