American Journal of Kidney Diseases
Volume 55, Issue 3 , Pages 419-422, March 2010

The Map Is Not the Territory—Mapping Out the Course and Cost of CKD

  • Kevin C. Abbott, MD, MPH

      Affiliations

    • Corresponding Author InformationAddress correspondence to Kevin C. Abbott, MD, MPH, Dialysis and Nephrology Service, Walter Reed Army Medical Center, 6900 Georgia Ave, Washington, DC 20307
    • ,
  • Cristina M. Yuan, MD

Walter Reed Army Medical Center, Washington, DC

Article Outline

 

Related Articles, pp. 452 and 463

Two important characteristics of maps should be noticed. A map is not the territory it represents, but, if correct, it has a similar structure to the territory, which accounts for its usefulness. If the map shows a different structure from the territory represented—for instance, shows the cities in a wrong order…then the map is worse than useless, as it misinforms and leads astray.

Alfred Korzybski, American scientist and philosopher

Patients rely on nephrology providers to guide them through the landscape of chronic kidney disease (CKD). However, if we are to be honest, the current map of CKD bears an uncanny resemblance to early maps found in museums, sometimes decorated with pictures of mythical and dangerous beasts. For example, the map of North America in 1803 was skewed, as regarded detail, toward the East Coast (Fig 1). North of Mexico and Louisiana were “Parts Unknown”.1 By 1806, thanks to the Lewis and Clark “Voyage of Discovery” funded by the US government, the geography of this terra incognita was known in much greater detail. The expedition returned with its own tales of spectacular and dangerous beasts—grizzly bears.2

Our knowledge of the causes, progression, costs, and optimal therapy of CKD is similar to the geographic knowledge of North America existing in 1803. We have some of the details, but there are still wide swaths of relatively unexplored territory. While exploring, we should take advantage of existing knowledge to predict what we might find, and to optimally prepare for the journey. The 2 articles by Hoerger et al, published in this issue of the American Journal of Kidney Diseases, take advantage of the detailed epidemiologic and outcomes data that we do possess to piece together a mosaic map, namely, a validated cost-effectiveness model of the incidence, progression, and treatment of CKD.3, 4 Not unexpectedly, this model is much more detailed and accurate in some areas than in others, just as the map of North America in 1803 was. However, when constructed and used properly, such a “map” is a powerful tool.

The optimal management of CKD, with the goals of preventing progression to renal replacement therapy and reducing associated morbidity and mortality, is of compelling interest to clinical nephrologists, primary care providers, and to those who pay for CKD management (patients, insurers, and the federal government). While several detailed practice guidelines exist for CKD, many of the recommendations are based on weak evidence and/or professional opinion.5, 6 Moreover, little information is available regarding the cost-effectiveness of screening and therapy for CKD in a very large population of potential patients. Hoerger et al have developed a model in which the cost-effectiveness of screening and therapeutic interventions may be assessed based on a detailed model of the natural history, therapy, and progression of CKD. They then take the model for a “test-drive,” analyzing screening for microalbuminuria in CKD.

The model used is a microsimulation model, developed by large contract research organizations under contract with the US Centers for Disease Control and Prevention. Microsimulation models attempt to predict the behavior of a population over time by modeling individuals or individual units making up the population. The model may be designed to allow either constant (static) or changing (dynamic) patterns of behavior. Transition probabilities (sets of rules that allow for simulated change, based on population-derived probabilities) are applied to govern the simulated progress of the individual through time. If the model is a good one, it should closely simulate the true behavior of the population and subgroups of individuals in it over time, and it can be used to predict the economic and demographic impact of individual interventions.7 This technique, most frequently used for developing economic and traffic models, recently has been applied to medical situations. One example is the Cancer Intervention and Surveillance Modeling Network (CISNET),8 which focuses on developing models of screening, treatment, and risk factors on cancer incidence and outcomes. Another example, the Coronary Heart Disease Policy Model, first introduced in 1987 and updated repeatedly,9 continues to inform policy makers and clinicians about the cost-effectiveness of existing and new interventions for the prevention and treatment of coronary heart disease.10 Such models may assist in the development of public health policy, guiding decisions so that both clinical outcomes and resource use are optimized. They can also be used to estimate the expected value of perfect information, a measure of the value to society of minimizing model uncertainty. While it might seem paradoxical to estimate dollar amounts for perfection, the process is analogous to estimating values as “n approaches infinity,” a procedure most readers should remember from their calculus days. We will never have perfect information, but we can estimate its value (if it could be achieved). Therefore, we should certainly never pay more for imperfect information. Estimates of this expected value of perfect information can define this cost “ceiling,” or “willingness to pay.”11

The technical supplement to the first article3 details the modeling method, which was programmed in TreeAge Pro 2008 using the Markov Monte Carlo microsimulation function. Data sources include 2003 US census estimates, National Health and Nutrition Examination Survey (NHANES) data, Framingham risk profiles, and the US Renal Data System (USRDS). The model, which accounts for the epidemiology of CKD and its natural history impacted by treatment, was validated by comparing the prevalence estimates provided by the model with those from a national survey (NHANES) and previous studies. It simulates incidence, progression, and treatment in a cohort of individuals from age 30 until censored at death or age 90. All 5 stages of CKD are included, as are common risk factors (which change with progressing age). These include diabetes status, systolic blood pressure and hypertension, left ventricular hypertrophy, total cholesterol, high-density lipoprotein cholesterol, and smoking status. Incidence of myocardial infarction, stroke, and angina are also included. The primary summary outcome measures include discounted medical costs, quality-adjusted life-years (QALYs), and lifetime incidence of end-stage renal disease. Progression was defined by albuminuria status and estimated glomerular filtration rate. Overall, the model showed good agreement with NHANES data for CKD prevalence, although its predicted 3.7% incidence of end-stage renal disease over a lifetime is somewhat higher than that of the USRDS initiation rate, which is 3%. As pointed out by the authors, this overestimation would tend to bias toward interventions, as the benefit of reduced progression might be overestimated. The authors rigorously assessed their model over a wide range of different conditions (sensitivity analysis), and used different estimation methods for estimated glomerular filtration rate, including the recently published Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation.12

In the second article,4 the authors modeled the cost-effectiveness of universal micro- and macroalbuminuria screening one time at age 50, and at 1-, 2-, 5-, and 10-year intervals beginning at age 50, followed by treatment with either an angiotensin-converting enzyme inhibitor and/or an angiotensin receptor blocker, designed to slow progression of early CKD. Targeted screening was also simulated for individuals with diabetes, individuals with hypertension but not diabetes, and individuals with neither diabetes nor hypertension. Analysis included a “no screening” and a “usual care” scenario. Cost-effectiveness ratios were generated against both no screening and usual care. A threshold of $50,000/QALY versus usual care was used to identify a cost-effective screening program, with lesser ratios representing greater cost-effectiveness.

Whether the $50,000/QALY threshold should still be the standard is currently controversial,13 as the authors discuss. The simulation demonstrated that screening for and treatment of microalbuminuria in patients with diabetes or hypertension is cost-effective versus usual care (depending on screening frequency) and that screening was not cost-effective in patients without current diabetes or hypertension. It is intriguing that such a model might allow for truly individualized partnerships between patients and providers in the future. For example, figure 2 from the Hoerger et al article4 shows that the optimal age for screening differs between CKD patients with diabetes, those with hypertension, and those with neither diabetes nor hypertension. Figure 3 from the paper shows that the model is most sensitive to differences in microalbuminuria incidence and treatment adherence.4 If such models become more fully defined, especially if reliable biomarkers are available, it might be possible for providers to develop a fully personalized prognosis and treatment plan for individual patients.

Enthusiasm must, of course, be tempered with realism. Cost-effectiveness calculations are only as good as the assumptions that are made, and cannot be applied to patient populations that are not included in the assumptions. As the authors point out, the base model will require continuous updating as new data emerge on the natural history of CKD progression, which can be expected to change as medical treatment and national demographics change. Furthermore, this is a model of CKD in the United States, and can only be applied in that country. Additionally, the impact of kidney transplants was not included in the model. Lastly, because of the statistical and mathematical complexity of the model, clinicians and medical policy makers may be tempted to invest it with more power than it has. A key limitation in this regard, which the authors acknowledge, is the occasional need to extrapolate cross-sectional data for estimates of disease progression. While there is precedent for this, it can lead to well-described biases,14 and additional longitudinal studies of CKD may be necessary to strengthen this model. Longitudinal studies have not been performed for every circumstance or demographic group, and estimates are extrapolated from “islands” of greater certainty.15 The model will therefore need—and was designed—to be updated and revalidated, and its cost-effectiveness predictions will change as new data emerge from prospective screening and therapy research in living individuals. We do not want to use this model blindly, like a driver relying on an outdated global positioning system. This is a model of what we know about the natural history of CKD, and it only approximates the true natural history. The model and the 1803 North American map have much in common; they are early guides, but we need regular updates!

Back to Article Outline

Acknowledgements 

The views expressed in this article are those of the authors and do not necessarily reflect the official policy or position of the Department of the Army, Navy, Department of Defense, nor the US Government.

Financial Disclosure: The authors declare that they have no relevant financial interests.

Back to Article Outline

References 

  1. Luffman J. A Map of North America (Library of Congress Map Collections). http://memory.loc.gov/cgi-bin/map_item.pl?data=/home/www/data/gmd/gmd3/g3300/g3300/ct000174.jp2&style=gmd&itemLink=D?gmd:2:./temp/~ammem_Srkg::&title=A%20map%20of%20North%20America%20;%20Outline%20of%20North%20America,%20in%20correspond%20to%20the%20mapAccessed January 7, 2009
  2. Lewis M, Clark W. In:  Bergon Frank editors. The Journals of Lewis and Clark. New York, NY: Penguin Books; 1989;p. 110–111(2003 edition)
  3. Hoerger TJ, Wittenborn JS, Segel JE, et al. A health policy model of CKD: 1 (Model construction, assumptions, and validation of health consequences). Am J Kid Dis. 2009;55(3):452–462
  4. Hoerger TJ, Wittenborn JS, Segel JE, et al. A health policy model of CKD: 2 (The cost-effectiveness of microalbuminuria screening). Am J Kid Dis. 2009;55(3):463–473
  5. National Kidney Foundation Outcomes Quality Initiative (NKF-KDOQI). http://www.kidney.org/professionals/KDOQIAccessed January 7, 2009
  6. Kidney Disease: Improving Global Outcomes (KDIGO). http://www.kdigo.orgAccessed January 7, 2009
  7. International Microsimulation Association. www.microsimulation.org/index.htmAccessed January 7, 2009
  8. Cancer Intervention and Surveillance Modeling Network. http://cisnet.cancer.govAccessed January 7, 2009
  9. Weinstein MC, Coxson PG, Pass TM, Stason WB, Goldman L. Forecasting coronary heart disease incidence, mortality, and cost: the Coronary Heart Disease Policy Model. Am J Public Health. 1987;77(11):1417–1426
  10. Pletcher MJ, Lazar L, Bibbins-Domingo K, et al. Comparing impact and cost-effectiveness of primary prevention strategies for lipid-lowering. Ann Intern Med. 2009;150(4):243–254
  11. Briggs A, Claxton K, Sculpher M. Decision modelling for health economic evaluation. In: Oxford, UK: Oxford University Press; 2006;p. 170–195
  12. Levey AS, Stevens LA, Schmid CH, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150(9):604–612
  13. Kilner JF. Selecting patients when resources are limited: a study of US medical directors of kidney dialysis and transplantation facilities. Am J Public Health. 1988;78(2):144–147
  14. Szklo M, Nieto FJ. Analysis of Age, Birth Cohort, and Period Effects. In: Epidemiology: Beyond the Basics. Gaithersburg, MD: Aspen Publishers; 2000;p. 4–16
  15. In:  Egger M,  Smith GD,  Altman D editor. Systematic Reviews in Health Care: Meta-analysis in context. 2nd ed.. London, UK: BMJ Publishing Group; 2001;p. 3–17

PII: S0272-6386(10)00031-4

doi:10.1053/j.ajkd.2010.01.003

Refers to article:

  • A Health Policy Model of CKD: 2. The Cost-Effectiveness of Microalbuminuria Screening , 01 February 2010

    Thomas J. Hoerger, John S. Wittenborn, Joel E. Segel, Nilka R. Burrows, Kumiko Imai, Paul Eggers, Meda E. Pavkov, Regina Jordan, Susan M. Hailpern, Anton C. Schoolwerth, Desmond E. Williams, Centers for Disease Control and Prevention CKD Initiative
    American Journal of Kidney Diseases March 2010 (Vol. 55, Issue 3, Pages 463-473)

  • A Health Policy Model of CKD: 1. Model Construction, Assumptions, and Validation of Health Consequences , 01 February 2010

    Thomas J. Hoerger, John S. Wittenborn, Joel E. Segel, Nilka R. Burrows, Kumiko Imai, Paul Eggers, Meda E. Pavkov, Regina Jordan, Susan M. Hailpern, Anton C. Schoolwerth, Desmond E. Williams, Centers for Disease Control and Prevention CKD Initiative
    American Journal of Kidney Diseases March 2010 (Vol. 55, Issue 3, Pages 452-462)

American Journal of Kidney Diseases
Volume 55, Issue 3 , Pages 419-422, March 2010

Article Tools

* Image Usage & Resolution