In a health care economy with limited resources, the provision of service that yields the greatest health benefit at the lowest cost is a priority. Informed decision making in value-based care requires information on benefits and costs. Organizing, prioritizing, and analyzing the many variables that must be considered in making common health care decisions is difficult for even the most experienced clinician. Rather than performing a rigorous formal analysis of the options and their possible beneficial and detrimental outcomes, most caregivers rely on informal heuristics to make the many decisions they are required to make each day. Understanding health care economics analyses and methods may empower the clinician to participate in value-based health care decision making and guide choices toward the provision of optimal health care.
A formal, logical process that incorporates complex data and models the clinical decision-making process can help the clinician to understand the potential impact of different factors in the likely outcomes of an intervention. Economic studies, including cost-benefit analysis (CBA), cost-effectiveness analysis (CEA), and cost-utility analysis (CUA) organize the available clinical and economic data and can help understand possible answers to the increasingly relevant question: “what intervention gives the greatest benefit proportional to its cost?”.1
The purpose of this article is to introduce the topic of economic studies in health care. It is intended to introduce a reader to some of the key elements of economic studies and to spur further in-depth reading about the subject. It is hoped that after reading this article, one can understand the methodology of health care economic analysis and approach economic studies with a critical and discerning eye that will guide optimal care.
MEDICAL DECISION ANALYSIS
Medical decision analysis is a formal method of organizing and connecting probabilities of events and their outcomes in a way that allows an assessment of the possible impacts of a decision.2 The decision model is based upon an index health state and the probability of change with a specific intervention. The resultant analysis enables the clinician, administrator, or policymaker to evaluate a probability distribution of expected outcomes and to understand the cost of the index decision on the total potential cost of care. The model is based upon clinical, epidemiological, patient-centered, and economic data.
Decision analyses, regardless of how apparently simple they are, involve many assumptions and choices, some of which may not be obvious. The most important of these is the perspective of the analysis. With some decisions, only 1 perspective may be important or makes sense. On the contrary, many problems, particularly those in public health, may have many possible perspectives, such as the individual, community, insurance company, hospital, or state or national government. For example, in the surgical management of adolescent idiopathic scoliosis, the hospital perspective may be limited to the episode of care, or the inpatient stay. The insurance perspective may be limited to the duration of time that adolescents are covered by the insurance of their parents. The patient and the physician share a much longer perspective, and a time frame that extends for a lifetime. The decision analysis model must be for a specific time frame. This is important both for estimating costs and outcomes and for determining the type of model that is appropriate. The latter is an advanced topic that will not be discussed further here but is covered in several of the references.
Medical decision making is complex and may encompass countless considerations for the patient, the surgeon, the hospital, the insurance company, and for public health and social welfare. Medical modeling involves choosing a finite number of options and data elements and estimating the distribution of outcomes based upon the best available data. Branching decision trees are the most common model used to organize and visualize the possible outcomes of a decision and their probabilities. A simple decision tree for the treatment of a benign painful condition is shown in Figure 1. The decision on whether or not to have the procedure is represented by the square “node.” The circular nodes represent subsequent possible events that result from the decision. Events are assumed to flow from left to right through the tree.
Each branch exiting a chance node is assigned a probability. The branches are assumed to represent all possible outcomes at the point represented by the node. Consistent with probability theory, the sum of all probabilities at each node must be 1, representing all possible outcomes. Although these are modeled as chance events, in reality one can model subsequent decisions, also. For example, if recurrent radiculopathy is modeled as a possible outcome after lumbar microdiscectomy a decision tree could include a branch for reoperation. The decision would be modeled as a simple probability based on data regarding the proportion of patients who undergo reoperation under similar circumstances.
The probabilities used in a decision analysis are ideally derived from high-quality data sources. Peer-reviewed literature is the usual source for most probabilities used in decision analysis of health care problems. In some cases, data are not available in a form that is directly translatable to probabilities in a decision tree. In other cases, such as the incidence of complications after a procedure or adverse reactions to a medication, published incidence data may be directly used in the decision tree.
Finally, the decision tree is populated with outcomes and cost data. The selection of the types of outcomes used in the tree and the methods for determining costs are discussed briefly in the following text. For an economic analysis, each path through the tree is associated with a cost and with an outcome. A process of “rolling back” probabilities and outcomes or costs is used to analyze the tree. This is shown in Figures 2 and 3. Starting at the final branches, the outcome value (or cost) of each branch is multiplied by the probability associated with that outcome. All of these partial valuations are added together for each node. This gives the expected value for a particular node. Then, working backward, a similar process is performed at each node until the decision node, the first node, is reached. Then, the expected values for each decision branch can be compared, and the best option, based on the greatest expected outcome or lowest expected cost, may be selected.
Outcomes of care are difficult to measure, and we do not have a clear consensus on the outcome measure that best reflects the patient's health care experience. Populating the decision tree with estimates of the value of the various outcomes is a fundamental component of the value equation, and determines the end result of the value calculation. Each branch that represents a distinct outcome is given a value. The differences between CEA, CBA, and CUA are mainly differences in the way that outcomes are valued (Table 2).
A CBA values both costs and outcomes in monetary units. This allows for costs and outcomes to be combined, and the optimal choice is the one with the greatest net gain (or smallest loss).3 Although this simplifies the analysis of the decision tree in some ways, this method has drawbacks. In health care analyses, it may be difficult or objectionable to value health benefits or lives in monetary terms. Although CBA is the standard mode of analysis in some fields, it is rare to see it in the field of health economics. An example of a CBA in spine surgery is the study on cost savings of vancomycin powder in instrumented and noninstrumented spine surgical procedures. Emohare et al4 demonstrated that the expenditure of $1152 on intraoperative vancomycin may save more than $500,000 in revision surgery for infection.
CEA is an alternative that avoids the potential problems of valuing outcomes in money that uses “natural units” to assess outcome.5 For example, if one were assessing different immunization protocols, a natural way to assess outcomes would be the number of cases. Other possibilities of natural outcomes would include deaths, pain score, or serum cholesterol levels. This method has at least 2 obvious advantages. First, readily available outcomes can be used. The outcomes can be selected to take advantage of data that are available during the course of treatment. Second, avoiding the need to convert outcomes from 1 form to another eliminates 1 potentially large source of uncertainty. The disadvantages of CEA are mainly limitations of generalizability. Although results given in lives saved have an understandable impact, other disease-specific such as cases averted or pain score changes may lack context, particularly if the disease or outcome metric is unfamiliar. An example of a cost-effectiveness study is the Swedish Lumbar Spine Study on fusion compared with nonoperative care for chronic low back pain.6 The authors concluded that the incremental cost per Oswestry Disability Index unit gained by surgery compared with nonoperative care averaged 5200 Swedish Krone. The limitation of cost-effectiveness as an outcome is that the value of an Oswestry Disability Index unit is not well defined, or translatable to other comparative health conditions.
Finally, CUA uses “health-state utilities” based on von Neumann-Morganstern expected utility theory, to value health outcomes.7,8 Utility scores are preference-based assessments of various health states that conventionally range from 0 (death) to 1 (perfect health). The most common and familiar utility in health economic analyses is the quality-adjusted life year (QALY).9 Quality adjusted life years is the unit of measure derived from the area under the curve of health status preference (utility score) over time. By incorporating both length and quality of life, measuring outcome in QALYs can avoid many of the pitfalls of using either monetary or natural units to value outcome. QALYs can be obtained for and compared among disparate health conditions. They can be estimated from several off-the-shelf questionnaires including the EuroQol Group 5-Dimension Self-Report Questionnaire and the Short Form-12. Cost-utility outcomes are most useful for procedures in which the intended outcome of care is an improvement of quality of life. Most discretionary spine procedures in adults are amenable to evaluation using the cost per unit of improvement health status, or utility. CUA is less useful in procedures in which the primary goal of care is to avoid the consequences of disease progression, or to prolong survival. Specifically, CUA in adolescent idiopathic scoliosis may underestimate the value of care in the patient with limited preoperative health compromise. The value of care in adolescent idiopathic scoliosis is measured by the value of avoiding the expected consequences of deformity progression. Disability-adjusted life year estimation may be most useful for preventative interventions. Similarly, an en bloc excision of an asymptomatic chordoma of the sacrum may improve survival, but would not be expected to improve patient preference for health status as a result of care. Therefore, CUA studies may be most appropriate for conditions in which the goal of care is improvement of patient reported health status.
As with cost estimates, health states that occur in the future should be discounted. Discounting future costs without discounting future outcomes will result in erroneous results. Conventionally, outcomes (and costs) that occur more than 1 year in the future are discounted. The same discount rate should be used for costs and outcomes.
Costs are an important component of any cost-benefit, cost-effectiveness, or cost-utility model. Estimating costs for the model is complex, and may vary significantly depending on the perspective of the stakeholder. The investigator must first determine what costs should be included in the analysis. A cost, in economic terms, is any use of a resource. For analysis, all costs must be in the same units, usually currency. Time and materials are both types of costs and need to be appropriately valued and incorporated into the decision model. Costs include direct costs, which are quantifiable and measurable, and indirect costs including the costs of keeping lights on and full staffing in a hospital. Some of the most significant costs of illness and care include lost productivity and wages, burden on families, care provider costs, and nonmedical costs. Therefore, direct costs alone underestimate the total cost of an illness. Costs in economic analyses are distinct from charges. Retail prices are not the same as economic costs. In health care analyses, this is frequently relevant when determining hospital costs associated with care delivery. Hospital charges include markups over what was paid for materials or services and, in general should not be used in economic analyses because they are not directly translatable to direct costs. Instead, the investigator should determine or estimate the true cost of the service, material, or procedure. Direct costs are not transparent and hospitals and manufacturers are limited in complete disclosure of cost information. In some models, the reimbursement from Medicare or other payers may be used as an estimate of cost, but this methodology is limited because the contribution margin (revenue-direct cost) may be highly variable between procedures and payers. Published literature may also be a useful resource for estimating cost data.
Costs for similar services or materials can vary significantly between institutions and geographical regions. Bederman et al10 demonstrated 8-fold variation in the costs of implants and in total costs of care for common spine procedures. The expected audience for the analysis may also be an important consideration in deciding whether the variation is important. Sensitivity analysis can be used to determine the effect of changes in cost estimates on the overall results of the study. A final consideration that is relevant to perspective is the necessity of discounting future costs (and benefits) that occur in the medium term to long term. Generally, it is appropriate to discount costs that occur more than 1 year in the future; in most situations, the present value of a dollar is greater than the future value of that dollar. In the base case analysis, a discount rate close to the average inflation rate is used. Sensitivity analysis may be used to test the robustness of the results over a range of discount rates.
The perspective of cost analysis will determine, to a significant degree, what costs should be included in a model.11,12 A cost from one perspective, for example, a hospital, may not be from another, for example the patient or the payer. Specifically, the hospital total costs include direct costs for materials and services, and indirect costs for fixed capital expenses. Adding a fixed percentage to direct costs to determine indirect costs is problematic in procedures that have a significant component of cost attributable to medical devices because the direct costs of devices are not proportional to the cost of maintaining fixed costs of the hospital infrastructure. The hospital perspective of cost is limited to the episode of care, and readmission to another hospital, prolonged recovery, and return to work are not relevant to the hospital perspective of costs. The payer has a similarly limited perspective on costs. The predicted probability of an insured patient losing private insurance within 12 months is more than 20%.13 A private payer has little financial incentive to be concerned about the 5-year follow-up of a 63-year female undergoing surgery for degenerative spondylolisthesis because that patient is unlikely to retain private insurance during that time period. The perspective of cost for the employer or dependent family may focus upon return to work, productivity, and quality of life after a spine procedure, and these components of cost are not in the purview of the hospital or payer. The patient and the physician share a long-term perspective with consideration of costs during a lifetime, and with prioritization of a durable improvement of quality of life. Costs remain the most complex component of the value equation, and the perspective of the stakeholder determines what components of cost to model. The most complete model will include direct costs, indirect costs, and social costs.
ANALYSIS AND INTERPRETATION
After the decision analysis model has been developed and populated with the estimated probabilities, costs, and outcomes, the overall analysis can be performed. The “gold standard” for this is the incremental cost-effectiveness ratio (ICER). This is a measure of the cost associated with the “incremental” benefit of 1 treatment compared with the next most effective option. This is an important distinction as the result is usually different, often significantly so, than that obtained from simply calculated the ratio of cost to outcome. An example analysis is shown in Table 1. Note that although the cost-effectiveness ratio and the ICER are both measured in dollars per QALY, their values are vastly different. Although the expected cost of surgery is $5100 and its expected outcome is 0.96 QALYs, its incremental benefit over the no surgery option is 0.1 QALY with a $4600 greater cost. The ICER, a ratio of the difference in cost and the difference in outcome, is the measure of the true benefit of 1 option compared with another. In the example, the ICER is $46,000 per QALY.
Within a model, it is generally fairly straightforward to determine which option is best, or whether there is a difference between options at all. Furthermore, for an individual or an organization with a specific budget or other constraints the implications of the analysis may be clear. For analyses conducted from a societal perspective, however, the significance of the results is less certain. Whether or not an intervention or program is “cost-effective” is a judgment based on many factors.14 Although an upper threshold of $60,000 to $100,000 per QALY is often discussed as a limit for cost-effectiveness, this is a societal, not a medical, judgment. Because of these difficulties, further analyses may be performed to determine the probability of an intervention being cost-effective at different thresholds.15
EXAMPLE: SPINE PATIENT OUTCOMES RESEARCH TRIAL HERNIATED LUMBAR DISC STUDY
Tosteson et al16 published a CUA based on the intervertebral (lumbar) disc herniation arm of the Spine Patient Outcomes Research Trial in 2008. Using primary outcomes data from both the randomized and observational arms of that study and cost data estimated from Medicare charges and payments, as well as self-reported out-of-pocket resource use and missed work, the authors calculated a base-case ICER of $69,403 per QALY (95% confidence interval, $49,523–$94,999 per QALY) for surgical management compared with usual nonoperative care. A subgroup analysis showed that the cost per QALY was lower for the Medicare population than for the non-Medicare population. Sensitivity analyses were performed for some of the cost assumptions but were not reported for other variables, such as outcomes or probabilities. Combined with similar results from previous studies, the findings reported by Tosteson et al16 indicate that, during a period of 2 years, lumbar discectomy for the treatment of intervertebral disc herniation provides a clinical benefit over usual nonoperative care at a cost below society's willingness-to-pay threshold.
UNCERTAINTY IN AND LIMITATIONS OF ECONOMIC ANALYSES
Medical decision analysis involves modeling under conditions of uncertainty. The results of economic studies of health interventions are best understood as estimates based on a specific model and data that should, similar to other economic analyses, be tested and updated as new information becomes available.17 The robustness of the results of a CUA or other economic analysis should be thoroughly tested. This sensitivity analysis gives the investigator and the reader an understanding of the likelihood that the base case results accurately represent the underlying reality. Figures 4 and 2A, B demonstrate examples of probability distributions of outcomes for operative and nonoperative care. Using these probability distributions, the investigator can model the likelihood of specific outcomes, and the cost of reaching that outcome. If, for example, altering cost, outcomes, and probability estimates for important variables within plausible ranges does not significantly change the results, one can be reasonably assured that the model will apply to a fairly wide set of circumstances. On the contrary, if a relatively small change in one estimate dramatically alters the results the investigator or reader will likely want to ensure that estimates of that variable are as accurate as possible. At the same time, neither the robustness of the model nor its absence indicates the accuracy of the model itself, which should be tested thoroughly against confirmatory data from other sources.
Health economic analyses, similar to randomized clinical trials, are most useful at a population level. An individual patient may have significantly different risks, costs, or values than the average values used in an overall analysis. Eliciting these differences may be difficult and is not generally part of a clinical encounter. A strongly robust model may, however, reassure the patient and clinician that the results are likely to apply in a particular circumstance.
Economic analyses of health interventions are an important tool for many audiences, including patients, physicians, hospitals, and health care systems, insurance companies, and governmental policy makers. A single economic study should not be interpreted as a definitive demonstration of the cost-effectiveness of an intervention, but CEA, CBA, and CUA that are designed well and conducted can serve as a helpful decision aid for its intended audience. Physicians and other providers, who have a thorough understanding of the clinical conditions and interventions, are critical to the proper design, conduct, and interpretation of economic analyses.
In a health care economy with limited resources, value is the priority of health care decisions.18 Informed choice in a value-based health care economy is characteristically made with incomplete information. Decision analysis is the process of making calculated decisions under circumstances of uncertainty. Medical decision modeling uses the best available data to guide informed choice. The literature on cost analysis in spine interventions remains limited, and the introduction and adoption of new techniques and technologies outpace the studies of value and cost analysis. Reimbursement, coverage levels, and access to care for patients with spinal disorders are dependent upon demonstration of value of specific interventions. Future research using the techniques reviewed in this article is a priority for guiding an evidence-based approach to spine care in a value-based health care economy.
- Health economic analyses integrate cost data with clinical outcomes data to estimate the economic impact associated with any gain in outcomes comparing one treatment option with another.
- Cost-benefit, cost-effectiveness, and cost-utility analyses measure outcomes in units of money, health outcomes, and QALYs, respectively.
- Understanding concepts such as decision analysis, discounting, and sensitivity analysis will help clinicians interpret the results of economic studies.
1. Drummond MF, Sculpher MJ, Torrance GW, et al. Methods for the Economic Evaluation of Health Care Programmes. 3rd ed. Oxford, England: Oxford University Press; 2005.
2. Soto J. Health economic evaluations using decision analytic modeling. Int J Technol Assess Health Care 2002;18:94–111.
3. Clemmer B, Haddix AC. Cost-benefit analysis. In: Haddix AC, Teutsch SM, Shaffer PA, et al., eds. Prevention Effectiveness: A Guide to Decision Analysis and Economic Evaluation. New York, NY: Oxford University Press; 1996:85–102.
4. Emohare O, Ledonio CG, Hill BW, et al.Cost savings analysis of intrawound vancomycin powder in posterior spinal surgery. Spine J 2014:S1529–9430.
5. Haddix AC, Shaffer PA. Cost-effectiveness analysis. In: Haddix AC, Teutsch SM, Shaffer PA, et al., eds. Prevention Effectiveness: A Guide to Decision Analysis and Economic Evaluation. New York, NY: Oxford University Press; 1996:103–29.
6. Fritzell P, Hägg O, Jonsson D, et al.Swedish Lumbar Spine Study Group. Cost-effectiveness of lumbar fusion and nonsurgical treatment for chronic low back pain in the Swedish Lumbar Spine Study: a multicenter, randomized, controlled trial from the Swedish Lumbar Spine Study Group. Spine (Phila Pa 1976) 2004;29:421–34.
7. von Neumann J, Morgenstern O. Theory of Games and Economic Behavior. Princeton, NJ: Princeton University Press; 1944.
8. Dasbach E, Teutsch SM. Cost-utility analysis. In: Haddix AC, Teutsch SM, Shaffer PA, et al., eds. Prevention Effectiveness: A Guide to Decision Analysis and Economic Evaluation. New York, NY: Oxford University Press; 1996:130–42.
9. Torrance GW. Measurement of health state utilities for economic appraisal. A review. J Health Econ 1986;5:1–30.
10. Bederman S, Pahlavan S, Berven S. Variations in Costs of Spinal Implants [abstract]. In: American Academy of Orthopaedic Surgeons 2013 Annual Meeting; 2013 19–23; Chicago, Illinois.
11. Weinstein MC, Stason WB. Foundations of cost-effectiveness analysis for health and medical practices. N Engl J Med 1977;296:716–21.
12. Torrance GW, Siegel JE, Luce BR. Framing and designing the cost-effectiveness analysis. In: Gold MR, Siegel JE, Russell LB, et al., eds. Cost-effectiveness in Health and Medicine. New York, NY: Oxford University Press; 1996:54–81.
13. Cutler DM, Gelber AM. Changes in the incidence and duration of periods without insurance. N Engl J Med; 2009;360:1740–8. doi:10.1056/NEJMsa0804668.
14. Doubilet P, Weinstein MC, McNeil BJ. Use and misuse of the term “cost-effective” in medicine. N Engl J Med 1986;314:253–6.
15. Fenwick E, Marshall DA, Levy AR, et al. Using and interpreting cost-effectiveness acceptability curves: an example using data from a trial of management strategies for atrial fibrillation. BMC Health Serv Res 2006;6:52.
16. Tosteson ANA, Skinner JS, Tosteson TD, et al. The cost-effectiveness of surgical versus
nonoperative treatment for lumbar disc herniation over two years. Evidence from the Spine Patient Outcomes Research Trial (SPORT). Spine 2008;33:2108–15.
17. Weinstein MC, O'Brein B, Hornberger J, et al. Principles of good practice for decision analytic modeling in health-care evaluation: report of the ISPOR Task Force on Good Research Practices—Modeling Studies. Value Health 2003;6:9–17.
18. Porter M, Teisberg E Redefining Healthcare: Creating Value-Based Competition on Results. Boston, MA: Harvard Business School Press. 2006.