Introduction to Health Care Decision-Making 1 Course Introduction • This includes a set of core slides that can be incorporated into an existing Intro to Health Economics or Intro to Healthcare Decision-Making course. • The aims of this primer are to: • Introduce key concepts related to decision analysis in healthcare • Give an understanding of the design, conduct and analysis of economic evaluation in healthcare 2 Module 1: Decision Trees 3 Module 1: Decision Trees What is decision analysis? • Decision Analysis - A quantitative method for choosing from a set of alternatives under conditions of uncertainty • Process allows decision-makers to think clearly through elements of complex decisions: • Range of possible actions (or inaction) and their consequences • Impact of complex, unpredictable systems and processes (e.g., markets, health) • Actions of others (e.g., competitors, regulators, patients) • Incorporates what is known and also what is uncertain 4 Module 1: Decision Trees Decision Trees Decision Trees: • A decision tree is a decision support tool that uses a tree-like diagram or branching structure to represent the decision, competing strategies and their consequences. • Decision trees utilize various nodes to represent different elements, including decisions and uncontrollable events, resulting in a set of possible pathways. • Analyzing the decision tree considers all the pathways and allows you to choose your optimal strategy for each decision. • It is one way to display an algorithm of a decision problem. 5 Module 1: Decision Trees Decisions and Strategies Decisions and Strategies: • Decision trees will typically have one or more points where a choice needs to be made from a set of competing strategies. • For each strategy, there will be a set of consequences that will result from that choice. • The goal of the analysis is to determine the optimal strategy. What are the consequences of this choice? 6 Module 1: Decision Trees Decisions and Strategies • In this example, we have one decision with two strategies: • Decision: I want to buy a lottery ticket but there are two different lotteries I can enter. • Strategy 1: Should I buy a ticket for Lottery 1? OR • Strategy 2: Should I buy a ticket for Lottery 2? 7 Module 1: Decision Trees Pathways • For each of these strategies, there are two possible outcomes: • I could win OR • I could lose (not win) • In this example decision tree, there are 4 potential pathways, given the different outcomes in each scenario: Pathway 1 Pathway 2 Pathway 3 Pathway 4 8 Module 1: Decision Trees Pathways • Choosing a strategy does not guarantee a result. • Some outcomes or events will be outside our control. • In addition, the competing outcomes must be mutually exclusive. • For example if there is an event that may or may not occur, you need to have two outcomes: • One outcome where the event occurs AND • Another outcome where the event does not occur. • There may be a number of events/outcomes that could occur within a strategy – these constitute decision tree pathways. • Each model pathway is a unique path from the decision node to a final outcome where no further events are possible. • For a decision tree to be realistic, it must contain all possible outcomes or events, which could result in a large number of pathways. • Downstream outcomes could be dependent on earlier events occurring. 9 Module 1: Decision Trees Outcomes and probabilities • Each of these outcomes has a probability associated with how likely it is to occur. • Probability is the likelihood of a given outcome's occurrence, which is expressed as a number between 1 (absolute 100% certainty) and 0 (absolute impossibility – 0% certainty). • An outcome with a probability of 0.5 can be considered to have equal odds of occurring or not occurring: for example, the probability of a coin toss resulting in "heads" is 0.5, because the toss is equally as likely to result in "tails." 10 Module 1: Decision Trees Outcomes and probabilities • If I buy a ticket for Lottery 1, the chance of winning is 10%. Consequently the chance of losing is 90%. • If I buy a ticket for Lottery 2, the chance of winning is 40%. Consequently, the chance of losing is 60%. • There is no conditional probability in this example. 11 Module 1: Decision Trees Payoffs (Rewards) • Payoffs or Rewards are the values gained or lost as a result of each pathway. • These payoffs could be measured as different kinds of outcomes: • • • • Revenue Costs Life Expectancy Death • Typically in decision trees you enter payoffs at the endpoint for each pathway. 12 Module 1: Decision Trees Payoffs (Rewards) • In our example, there are two strategies, each strategy has 2 pathways, resulting in 4 total pathways or scenarios. • Payoff values need to be entered for each of those 4 pathway endpoints. • In this example: • • • • If I win Lottery 1 I win $1,000 If I lose Lottery 1 I win $0 If I win Lottery 2 I win $300 If I lose Lottery 2 I win $0 13 Module 1: Decision Trees Decision Tree • The model is now complete. • In Lottery 1 I have a 10% chance of winning $1,000. • In Lottery 2 I have a 40% chance of winning $300. • We can analyze the model to choose the optimal strategy. 14 Module 1: Decision Trees Analyzing the Model – Expected Values • Expected Value - a predicted value calculated as the sum of all possible values each multiplied by the probability of its occurrence. • Expected values are calculated for each strategy and outcome measure • In a decision tree, we can calculate the overall expected value for each strategy by multiplying the reward for each pathway by the probability of that pathway. • Note there may be multiple probabilities within each pathway and multiple outcome measured in the model, but not in this example. • Expected values help us determine the optimal strategy. 15 Module 1: Decision Trees Expected Values – Lottery Example • Strategy 1: Lottery ticket 1 offers us a 10% chance of winning 1,000. Therefore: • The expected value of playing Lottery 1 is: (0.10*1,000) + (0.90*0) = $100 • Lottery 1 gives us an expected value of $100. • Strategy 2: Lottery ticket 2 offers us a 40% chance of winning $300. Therefore: • The expected value of playing Lottery 2 is: (0.40*300) + (0.60*0) = $120 • Lottery 2 gives us an expected value of $120. 16 Module 1: Decision Trees Analyzing the model – Choosing the optimal strategy • Models will be configured to choose the method by which the optimal strategy is identified, based on the expected values. This could entail: • Maximizing expected value (e.g., revenue, gains in health, etc.) • Minimizing expected value (e.g., reducing deaths, costs, etc.) • In this example we are trying to maximize our expected value, by choosing the strategy that has the highest expected value. • In healthcare models, there may be a balance between multiple payoffs. • For example, minimize cost AND maximize life expectancy. 17 Module 1: Decision Trees Optimal Strategy – Lottery Example • Optimal strategy – the decision that chooses the strategy with the “best” expected value. • In this example, buying a ticket for Lottery 2 maximizes our expected reward, therefore it is the optimal strategy in this decision analysis. 18 Module 1: Decision Trees Decision Trees – Key points • Decisions and strategies • Pathways • Probabilities • Payoffs • Expected Value • Optimal strategy selection 19 Module 1: Decision Trees Homework assignment Die roll game: In this game, you have three choices of how to play. Each has its own goal and reward. 1. Roll a die once and win $100 only if you roll a “6”. 2. Roll a die twice and win $125 if both rolls are “5” or greater. 3. Roll a die three times and win $110 if all three rolls are “4” or greater. Build a model to identify the choice above that maximizes your reward. 20 Module 2: Decision Analysis in Healthcare 21 Module 2: Decision Analysis Healthcare Measurement and Valuation of Health – What is decision analysis in healthcare? • Health care systems are created by societies to maintain health and well-being. • Systems are expected to run efficiently, however we tend to find gaps in quality, safety, equity and access in our health care systems. • Efficiency relates to maximizing the quality of a comparable unit of health care delivered or unit of health benefit achieved for a given unit of health care resources used. • Health care cost increases continue to outpace the rise in wages, inflation, and economic growth. Resources are finite. Wants are infinite. • Decision analysis helps decision-makers reconcile spending with maximizing optimal health outcomes for individuals or populations, and serves an important role in healthcare decision-making. 22 Module 2: Decision Analysis Healthcare Model Perspective Model perspective - The point of view from which the costs and benefits are recorded and assessed. The choice of perspective must be derived logically from the research question. Perspective affects the measurements that get included in our decision analysis model. • Healthcare Provider • Patient • Society • Government • Employer 23 Module 2: Decision Analysis Healthcare Types of economic evaluation - Single outcome • Always looking to choose the optimal strategy • What is the evaluation method by which you will make your decision or choice? • If there a single primary outcome, you will choose the strategy with either the highest expected value or the lowest value. • Highest – Life Expectancy, QALYs, etc. • Lowest – Costs • Typically, however, we are trying to balance multiple competing outcomes: • For example, a new intervention improves health benefit or efficacy, but comes at an increased cost. • We want to minimize cost AND maximize life expectancy. 24 Module 2: Decision Analysis Healthcare Putting economic evaluation into consideration - Heart attacks • Imagine there is a $10 Million magic pill that prevents heart attacks in people forever. • Does it make sense to pay for this pill? • Given the benefit it provides, is it worth paying for this magic pill? • Consider that there are other causes of death. 25 Module 2: Decision Analysis Healthcare Types of economic evaluation – Cost-effectiveness analysis • Cost-effectiveness: In the context of health and medicine, cost-effectiveness analysis (CEA) is a method for evaluating tradeoffs between health benefits and costs resulting from alternative courses of action. • Cost-effectiveness analysis measures the incremental cost of achieving an incremental health benefit. • For example, a new treatment might cost an additional $10K to increase life expectancy by one year. Incremental Life expectancy Incremental Cost • Cost-effectiveness analysis requires measurements for cost and effectiveness. 26 Module 2: Decision Analysis Healthcare Outcome measures – Examples of Costs There may be many different types of costs that comprise the overall cost of a treatment strategy. Direct Costs – all consumption of resources (including direct medical and non-medical) resulting from a treatment or therapy • Treatment costs arise directly from the treatment (e.g. diagnosis, drug therapy, medical care, office visits, in-patient treatment, devices, surgery, etc.). • Other costs arise from the consequences of the disease or treatment (e.g. hospitalizations, ER visits, adverse events, etc.). Indirect Costs - represent changes in resources that occur not directly in relation to the treatment of the disease (e.g., productivity lost, caregiver time and resources, transportation, etc.) • Typically not included in healthcare models 27 Module 2: Decision Analysis Healthcare Outcome measures – Examples Clinical Effectiveness There are also different health outcomes that can be used to measure effectiveness. Quality Adjusted Life Year (QALY)* – The most common clinical/health outcome measure used in healthcare decision-making to assess the value for money of medical interventions. It is a generic measure of disease burden, including both the quality and the quantity of life lived. Utility* – (quality measure) the preference or value that an individual or society gives a particular health state. It is generally a number between 0 (representing death) and 1 (perfect health). Life Years (quantity measure) – measures quantity of years lived, but does not incorporate the quality of a person’s life A treatment will result in higher QALYs the longer you live and the healthier you are. 28 *As defined by the National Institute for Health and Care Excellence (NICE) Module 2: Decision Analysis Healthcare Outcome measures - QALYs The QALY can be calculated using the following formula: Years of Life x Utility Value = #QALYs For example: • If a person lives in imperfect health (utility is 0.8) over 5 years, that person will accumulate 4 QALYs. (5 Years of Life x 0.8 Utility Value = 4 QALYs) • Note that utilities can change over time, resulting in more complex calculations. 29 Module 2: Decision Analysis Healthcare Outcome measures – Other Clinical/Health Alternative health outcome measures: • Disability-Adjusted Life Years (DALYs) • Infections • Hospitalizations • ER visits • Successful transplants • Deaths 30 Module 2: Decision Analysis Healthcare Cost-effectiveness for comparing strategies Competing strategy is dominated Incr. Cost More costly What if an Intervention is less effective and more costly than existing treatments? There is no question that this is not something we should fund. Less costly Incr. Effectiveness Less Effective More Effective Reference strategy is at the center of the graph and represented by the blue circle 31 Module 2: Decision Analysis Healthcare Cost-effectiveness for comparing strategies Incr. Cost Cost increasing What if a new, competing intervention is more effective and less costly than existing treatments? There is no question that this is the best option for us to fund. Incr. Effectiveness Cost saving Competing strategy is dominant Less Effective More Effective Reference strategy is at the center of the graph and represented by the blue circle 32 Module 2: Decision Analysis Healthcare Cost-effectiveness for comparing strategies Incr. Cost Cost increasing The question we are most often dealing with is - What if an intervention is more expensive and more effective than current treatments? Is it worth it for us to pay for this new treatment? Does the additional effectiveness justify the additional cost? Optimal strategy is determined by ICER Cost saving Incr. Effectiveness Optimal strategy is determined by ICER Less Effective More Effective Reference strategy is at the center of the graph and represented by the 33 blue circle Module 2: Decision Analysis Healthcare Incremental cost-effectiveness plane • Cost-effectiveness analysis provides a mathematical basis to measure the relative value of the increased cost vs. improved health outcomes between competing strategies. Incr. Cost Cost increasing • In these cases, the estimation of value is based upon calculation of an Incremental CE Ratio (ICER). Cost saving Incr. Effectiveness Reference strategy is at the center of the graph and represented by the blue circle Less Effective More Effective 34 Module 2: Decision Analysis Healthcare Cost effectiveness - Incremental costs vs. incremental effectiveness We can only rationally justify higher cost strategies if they provide sufficient increases in health outcomes. • In cost-effectiveness analysis, this is the incremental cost-effectiveness ratio (ICER) and it is defined as the difference in cost (C) divided by the difference in effectiveness (E). CNew – CReference ICER = _____________ ENew – EReference • The ICER measures the additional cost per additional unit of effectiveness. *Note – incremental values are always calculated by comparing a more expensive strategy to a less expensive strategy. 35 Module 2: Decision Analysis Healthcare Willingness to Pay • The ICER can be used as a decision rule in resource allocation. • If a decision-maker is able to establish a willingness to pay value for the outcome of interest, it is possible to adopt this value as a threshold. • The Willingness to Pay (WTP) represents the additional amount a decision-maker is willing to pay per additional unit of effectiveness, thereby setting a limit on the ICER. 36 Module 2: Decision Analysis Healthcare Willingness to Pay • If for a given intervention the ICER is above this WTP threshold it will be deemed too expensive and thus should not be funded, whereas if the ICER lies below the WTP threshold the intervention can be judged as cost-effective. • On a CEA graph, the slope of the ICER and WTP lines provide this information. ICER from less expensive and more expensive strategy is < WTP, so we can fund the new intervention ICER from less expensive and more expensive strategy is > WTP, so we should not fund the new intervention 37 Module 2: Decision Analysis Healthcare Multiple strategies • There may be more than two strategies that need to be compared. • When there are multiple strategies, you will need to consider comparisons in pairs, starting with the least costly strategy. Drug 2 vs. Drug 1: ICER > WTP = Drug 2 not preferred Drug 1 vs. no treatment: ICER < WTP = Drug 1 preferred 38 Module 2: Decision Analysis Healthcare Multiple Strategies - Dominance • Dominated strategies are not on the cost-effectiveness frontier, and are therefore “dominated” by other, more cost-effective strategies. 39 Module 2: Decision Analysis Healthcare Other types of economic evaluation • Cost-effectiveness is the most commonly used economic analysis in healthcare, but there are other approaches including: • Cost-minimization - Cost-minimization analysis is a method of calculating treatment costs to project the least costly drug or therapeutic modality. • For example, budget impact models (BIMs) • Comparative effectiveness - evaluating and comparing health outcomes and the clinical effectiveness, risks and benefits of 2 or more medical treatments, services or items • Cost-utility - Cost-utility analysis (CUA), a subset of CEA, is used to determine cost relative to changes in utilities, especially quality of life. CUA is generally applied when multiple patientrelevant clinical outcome parameters are expressed in different units. • Cost-benefit - Cost-benefit analysis is used to value both incremental costs and outcomes in monetary terms and therefore allows a direct calculation of the net monetary cost of achieving a health outcome. 40 Module 2: Decision Analysis Healthcare Homework assignment • There are 2 strategies for treating a disease: • Standard of Care (SOC) – it costs $10K per year, which provides an average Life Expectancy of 10 years at a constant utility of 0.75 • Novel treatment – it costs $15K per year, which provides an average Life Expectancy of 11 years at a constant utility of 0.85 • Assignment: • Calculate the ICER between the two strategies (without a model). • Determine the optimal strategy based on a WTP threshold of $50K and explain why. 41 Module 3: Building a Healthcare Decision Tree Model (Medicine vs. Surgery) 42 Module 3: Building a Decision Tree Model Defining the healthcare problem Before you build any model, you need to have a firm understanding of the goals and measurable outcomes within the model. To establish this understanding, answer the following 3 questions: • What is the underlying problem? • What are our strategies/options? • How do we measure the outcomes associated with these strategies? 43 Module 3: Building a Decision Tree Model Defining the healthcare problem Establishing decision points and strategies: • What is the underlying problem? Patients with Disease X have historically only had one treatment option. A new treatment has become available. We want to identify the most cost-effective treatment for patients with Disease X. 44 Module 3: Building a Decision Tree Model Developing your model – defining strategies Establishing decision points and strategies: • What are our alternatives? The current SOC for treating this disease is surgery. There is a new medicine available that can also be used to treat patients for this disease. This medicine is more expensive than the SOC but has the potential to be more effective than the SOC. Thus, we have 2 options for treating patients that we want to consider in our decision analysis: 1. Medicine (New) 2. Surgery (SOC) 45 Module 3: Building a Decision Tree Model Create decisions and strategies • In this healthcare decision problem, we have two strategies to consider in treating patients with Disease X. • In a model, this is represented by a decision node with a branch for each strategy. 46 Module 3: Building a Decision Tree Model Developing your model – defining outcomes • How do we measure the outcomes associated with these strategies? • For this model, we will use a traditional cost-effectiveness analysis • Cost is measured in US$ • Effectiveness is measured in QALYs • The model must be configured for cost-effectiveness. 47 Module 3: Building a Decision Tree Model Developing your model – define patient pathways for each strategy Defining pathways: • Surgery: • It costs $50K one time for the procedure. • There is a 75% probability of success with surgery. • Medicine: • It costs $7K per year for the treatment over the course of a patient’s lifetime. • There is a 80% probability of success with Medicine. • We will know if a patient fails on medicine in Year 1, therefore in the model if a patient fails on Medicine, treatment is stopped after Year 1. • With either Surgery or Medicine, if treatment is successful, patients have an average Life Expectancy of 12 years at a constant utility of 0.85 • If either treatment fails, patients have an average Life Expectancy of 8 years at a constant utility of 0.70 48 Module 3: Building a Decision Tree Model Developing your model – create parameters to build the model Defining parameters: • Creating variables for your parameters provides transparency, consistency and flexibility. • Parameter variables should be defined once numerically at the beginning of your model. Parameter Variable Name Value Probability of success with Surgery pSuccessSurg 0.75 Probability of success with Medicine pSuccessMed 0.80 Cost of Surgery cSurgery 50K Cost of Medicine cMedAnnual 7K Life Expectancy Post Treatment Success lePostSuccess 12 Life Expectancy Post Treatment Failure lePostFailure 8 Utility Success Post Treatment uPostSuccess 0.85 Utility Failure Post Treatment uPostFailure 0.70 49 Module 3: Building a Decision Tree Model Developing your model – define patient pathways for each strategy • Each strategy consists of a chance node (represented by a ), with branches to the right to represent possible outcomes – in this example there are two outcomes for each strategy: • Success • Failure of treatment • Most models will have a more complex structure. 50 Module 3: Building a Decision Tree Model Developing your model – adding probabilities • For each outcome, we need to define the probability of that outcome occurring. • We know the probability of success with Medicine and the probability of success with Surgery. • The parameter variables ‘pSuccessMed’ and ‘pSuccessSurg’ represent the probability of success for each strategy. • For each chance node, probabilities must sum to 1. In TreeAge Pro, the complement to each known probability [1-pSuccessMed or 1-pSuccessSurg] is represented by a “#”. 51 Module 3: Building a Decision Tree Model Adding outcome measures • Each terminal node ( ) terminates and represents a full patient pathway and now we need to add outcome measures associated with each pathway. • Let’s consider the first pathway: Success from Medicine 52 Module 3: Building a Decision Tree Model Adding outcome measures Success from Medicine pathway Costs: • Medicine costs $7K per year over the course of the full life expectancy of 12 years. • cMedAnnual * lePostSuccess = $7K * 12 = $84K Effectiveness: • Utility with successful treatment is 0.85 over the course of the full life expectancy of 12 years. • uPostSuccess * lePostSuccess = 0.85 * 12 = 10.2 QALYs • In your model enter the formula referencing your parameters, NOT the calculated value. • Again, using variables as demonstrated above keeps the model flexible by allowing you to change your parameter values at any time. 53 Module 3: Building a Decision Tree Model Adding outcome measures • Let’s consider the next pathway: Failure from Medicine 54 Module 3: Building a Decision Tree Model Adding outcome measures Failure from Medicine pathway Costs: • Medicine costs $7K per year of treatment, but Medicine is stopped after 1 year in this scenario. • cMedAnnual * 1 Year = $7K * 1 = $7K Effectiveness: • Utility with treatment failure is 0.7 over the course of the full life expectancy of 8 years. • uPostFailure * lePostFailure = 0.7 * 8 = 5.6 QALYs 55 Module 3: Building a Decision Tree Model Adding outcome measures • The next two slides will walk through Success and Failure pathways associated with the Surgery strategy. 56 Module 3: Building a Decision Tree Model Adding outcome measures Success from Surgery pathway Costs: • Surgery costs $50K only one time over the course of the full life expectancy of 12 years. • cSurgery = $50K Effectiveness: • Utility with successful treatment is 0.85 over the course of the full life expectancy of 12 years. • uPostSuccess * lePostSuccess = 0.85 * 12 = 10.2 QALYs 57 Module 3: Building a Decision Tree Model Adding outcome measures Failure from Surgery pathway Costs: • Surgery costs $50K one time over the course of the full life expectancy of 8 years. • cSurgery = $50K Effectiveness: • Utility with treatment failure is 0.7 over the course of the full life expectancy of 8 years. • uPostFailure * lePostFailure = 0.7 * 8 = 5.6 QALYs 58 Module 3: Building a Decision Tree Model Completed Model • Our model is now complete and ready for analysis. 59 Module 4: Analyzing a Healthcare Decision Tree Model (Medicine vs. Surgery) 60 Module 4: Analyze the Tree Analyze the Tree • We want to compare Surgery against Medicine to make sure we are utilizing our healthcare dollars wisely. Ultimately, we want to know which is more cost-effective. • Evaluate the strategies independently and then compare them. • How do we evaluate each strategy? • How do we compare the strategies? 61 Module 4: Analyze the Tree Analyze the Tree – Evaluate each strategy • Let’s start with the Medicine strategy. • We have outcomes associated with our patient pathways: • Success = $84K and 10.2 QALYs • Failure = $7K and 5.6 QALYs • We need a weighted average of those pathways to determine an overall expected value for the Medicine strategy. This is where probabilities are utilized. • Cost: $84K * 0.8 + $7K * 0.2 = $68.6K • Effectiveness: 10.2 * 0.8 + 5.6 * 0.2 = 9.28 QALYs • Average values weighted by the probabilities ensure more likely pathways have a bigger impact on our overall strategy than less likely pathways. 62 Module 4: Analyze the Tree Analyze the Tree – Evaluate each strategy • Let us now evaluate the Surgery strategy. • We have outcomes associated with our patient pathways: • Success = $50K and 10.2 QALYs • Failure = $50K and 5.6 QALYs • We need a weighted average of those pathways to determine an overall expected value for the Surgery strategy. • Cost: $50K * 0.75 + $50K * 0.25 = $50K • Effectiveness: 10.2 * 0.75 + 5.6 * 0.25 = 9.05 QALYs • Each terminal node has a “P =“ expression which represents the likelihood of that 63 pathway relative to all pathways in the strategy (Product of all probabilities along path). Module 4: Analyze the Tree Analyze the Tree – Compare Strategies • The model is ready to run Cost-Effectiveness Analysis. • Let us do it first by hand, and then show how TreeAge Pro can do it for you. • As a reminder, cost-effectiveness analysis provides a mathematical basis to measure the relative value of the increased cost vs. improved health outcomes between competing strategies (i.e., increased effectiveness). • In cost-effectiveness analysis, this is the incremental cost-effectiveness ratio (ICER) and it is defined as the difference in cost (C) divided by the difference in effectiveness (E). CNew – Creference ICER = -----------------------ENew – Ereference = 68600 – 50000 ----------------------- 9.28 – 9.05 = 18600 ---------------- = 80870 0.230 64 Module 4: Analyze the Tree Analyze the Tree – Compare Strategies We can run the Rankings Report in TreeAge Pro to analyze the model and generate an ICER. CNew – Creference ICER = -----------------------ENew – Ereference = 68600 – 50000 ----------------------9.28 – 9.05 = 18600 ---------------- = 80870 0.230 65 Module 4: Analyze the Tree Analyze the Tree – Willingness to Pay (WTP) We can now compare the ICER to a willingness-to-pay (WTP) threshold: • ICER = the amount we have to pay for each additional unit of effectiveness • WTP = the amount we are willing to pay for each additional unit of effectiveness The analysis can then tell us is the ICER too high relative to the WTP? • ICER <= WTP, choose the more expensive/effective option as we can justify the higher cost • ICER > WTP, choose the less expensive/effective option, as we cannot justify the higher cost 66 Module 4: Analyze the Tree Analyze the Tree – Choose the Optimal Strategy • The ICER calculated is $80.87K. If our Willingness to Pay Threshold is $50K, then this ICER value is higher than our WTP, which means we cannot recommend funding the new treatment, Medicine. • Therefore, Surgery is the optimal strategy. • If there were dominated strategies, they would appear in the second grouping and not 67 rd th the first. You will never use the 3 and 4 groupings to determine the optimal strategy. Module 4: Analyze the Tree Analyze the Tree – Choose the Optimal Strategy • The TreeAgo Pro CEA graph function also compares strategies and identifies the optimal strategy. • A WTP line can be added to your graph as was done below. The WTP line will always pass through the optimal strategy. 68 Module 4: Analyze the Tree Analyze the Tree – Choose the Optimal Strategy via Net Benefits • Net Benefits Calculations combine cost, effectiveness and WTP into a single value. • The optimal strategy will have the highest Net Benefit value. • Net Monetary Benefits (NMB) = Effectiveness * WTP – Cost • Surgery: 9.05 * 50000 – 50000 = 402500 • Medicine: 9.28 * 50000 – 68600 = 395400 • Net Health Benefits (NHB) = Effectiveness – Cost / WTP • Both will identify the optimal strategy. • The NMB calculation in the Rankings Report also confirms Surgery as the optimal strategy. 69 Module 4: Analyze the Tree Homework Assignment • For Disease X, there exists surgery and generic medicine as treatment options. A New Medicine is entering the market and we need to determine which of the 3 treatments is the most cost-effective. • Surgery is a one-time cost of $50K. There is a 1% chance of dying from surgery. • Generic medicine costs $7K per year over the course of a lifetime. If a patient fails on generic medicine, treatment ends after one year. • New medicine costs $12K per year over the course of a lifetime. This treatment is also discontinued after one year if a patient fails on the New medicine. • Probability of success with surgery is 75%, success with Generic medicine is 70% and success with New medicine is 80% • Your WTP threshold is $50K. • Build a model to identify the optimal strategy and describe how you came to that conclusion. • Are there any dominated strategies? If so, which one(s) and how do you know? 70 Module 5: Considering Uncertainty 71 Module 5: Considering Uncertainty Completed Model Our model is now completed and has been analyzed using a single value for each parameter based on our best research and estimates. 72 Module 5: Considering Uncertainty Confidence in Parameters - Sensitivity Analysis But, how confident are we about those parameters? For example: • In the model, we assume the price of a new medicine is $7K/year. What if the drug is actually priced less at $6K, or more at $8K when it launches? • In the model, we assume the probability of success with surgery is 75%. What if that probability is closer to 70% in reality? 73 Module 5: Considering Uncertainty Confidence in our Conclusions - Sensitivity Analysis Given the uncertainty in our parameters, how confident are we of our conclusions? The application of Sensitivity Analysis to our model assesses the extent to which a model’s calculations and recommendations are affected by uncertainty. • Specific questions about the model sensitivity analysis can help answer are: • Is a model sensitive to a particular uncertainty? E.g., does varying a parameter’s value result in changes in optimal strategy? • How does uncertainty related to multiple parameters affect our overall confidence in the conclusions/recommendations? 74 Module 5: Considering Uncertainty One-Way Sensitivity Analysis One-Way Sensitivity Analysis: • Consider a parameter over a range of values around the base case value. • Re-evaluate the model multiple times across that range. • Is there a change in strategy within that uncertainty range? • Example: pSuccessSurg base case is 0.75 in the Medicine vs. Surgery model. Let us consider a range of values reflecting our uncertainty in this parameter by running 1-way sensitivity analysis over the range of 0.7 to 0.78. 75 Module 5: Considering Uncertainty One-Way Sensitivity Analysis • If we run our analysis using the range of pSuccessSurg values, we find that at the lower values within the uncertainty range (0.7, 0.71), Medicine becomes the optimal strategy, with ICERs below our WTP threshold of $50K. Medicine is the optimal strategy Surgery is the optimal strategy 76 Note: Not pasted here - Report includes outputs up to pSuccessSurg values of 0.78 Module 5: Considering Uncertainty One-Way Sensitivity Analysis • The Net Benefits (CE Thresholds) graph illustrates the pSuccessSurg threshold where the optimal strategy (highest NMB) shifts from Medicine to Surgery. 77 Module 5: Considering Uncertainty Tornado Diagrams • A Tornado Diagram is a set of one-way sensitivity analyses brought together in a single graph. It can include any number of parameters defined in the tree. • The horizontal bars depict how much the uncertainty of each parameter affects the ICER. • Red indicates an increase in the value from the base case, whereas blue indicates a decrease in the value from base case. ICER increases as parameter value increases CE threshold line ICER decreases as parameter value increases Base case ICER line 78 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) • Uncertainty always exists in multiple parameters in a model. • Studying parameters independently does not fully represent the overall uncertainty of the model. • Different parameter combinations could impact optimal strategy. It may take outliers of multiple parameters to cause a change in optimal strategy. • Probabilistic Sensitivity Analysis (PSA) enables us to study the combined uncertainty across multiple parameters. • PSA results estimate the total impact of uncertainty on the model, or the confidence that can be placed in the analysis results. 79 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) - Distributions • PSA requires a distribution for each parameter included in the analysis. • In Example model Medicine vs. Surgery_PSA, we have set up distributions for 4 parameters in the model. • This distribution represents the uncertainty around probability of success with surgery (Dist_pSuccessSurg). • Beta distribution type • Mean – 0.75 • Standard deviation – .03 • Sample per EV (once per model calc.) 80 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) - Distributions • Use the “Graph It” tool to sample each distribution and validate that it appropriately reflects your uncertainty. • The histogram below represents the range of samples generated for Dist_pSuccessSurg. • Note that the distribution mean is 0.75, so these values look reasonable. 81 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) - Distributions • Distributions must be referenced in the model (4 total in this model). • Variable pSuccessSurg is now set equal to the distribution. • When you run CEA, the variable uses the mean from the distribution. • When you run PSA, the variable uses the appropriate sample from the distribution. 82 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) – Running Analysis • PSA calculates the model many times with different sets of sampled parameter data from your distributions. • Sample distributions, calculate the model, and then repeat. • To run the analysis in TreeAge Pro: • Choose the Root Node of the model. • From the menu, choose Analysis > Monte Carlo Simulation > Sampling • You will be presented with aggregated output; however, our focus is on the PSA outputs. • We are going to highlight the Acceptability Curve and the ICE Scatterplot outputs. 83 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) – CE Acceptability Curve • The CE acceptability curve shows the percentage of model recalculations that favor each strategy across a range of WTP. • As WTP increases, more of the recalculations will favor the more effective strategy. • At a WTP of $50K, it shows Surgery is the optimal strategy approximately 70% of the time. 84 Module 5: Considering Uncertainty Probabilistic Sensitivity Analysis (PSA) – ICE Scatterplot • The ICE scatterplot also shows us that 70% favored the less expensive strategy, Surgery, as depicted by the dots above and to the left of the WTP line on the graph. • In almost all cases, Medicine was more expensive (IC > 0). • In a solid majority of cases, Medicine was more effective (IE > 0). • Given our PSA results, how confident are we that Surgery is the more cost-effective strategy? 85 Module 5: Considering Uncertainty Homework Assignment Run PSA on the Classroom Medicine vs. Surgery Model built in Modules 3 and 4. • Run PSA using the following three parameter distributions: o Add a Beta Distribution for pSuccessNewMed (Mean: 0.80; Std Dev .03) o Add a Gamma Distribution for cSurgery (Mean: 50000; Std Dev 5000) o Add a Gamma Distribution for cNewMedicine (Mean: 12000; Std Dev 3000) • Reference the distributions in the model. • Run the PSA and interpret the results. 86 Section 6: More Advanced Modelling Approaches and When to Apply Them 87 Module 6: Advanced Modelling Approaches Markov Models • Markov modelling is a technique that allows presentation and analysis of disease progression over time. • It is particularly suitable for diseases that are chronic and recurrent in nature. • Markov models take a long disease progression and break it down into shorter time cycles that repeat to cover the entire model time horizon. • Markov model structure consists of health states and events/transitions. • Markov cohort analysis sends a cohort of patients through disease progression pathways consisting of health states and events/transitions, cycle by cycle until the end of the total time horizon. • In each cycle, across the time horizon, the analysis will accumulate health utility and cost, which can be attached to any health state and/or event. 88 Module 6: Advanced Modelling Approaches Decision Trees Vs. Markov Models Decision Trees: • Used for “one off” decisions • Particularly suited to: • Acute care problems (“kill or cure”) • Once-only diseases/conditions • Short-term diagnostic/screening decisions Markov Models: • Represent disease processes which evolve over time • Suited to modelling the progression of chronic disease • Can handle recurrence • Estimate long term costs and life years gained/QALYs 89 Module 6: Advanced Modelling Approaches Markov Models – Example Events/Transitions Health States Terminal Nodes (return cohort to health states for next cycle) Markov Node (start of cycle) 90 Module 6: Advanced Modelling Approaches Markov Models – Cohort Analysis Cohort flow through events/transitions Accumulation of costs and effectiveness 91 Module 6: Advanced Modelling Approaches Individual vs. Cohort • All of the models we have considered thus far have been cohort models, which analyze data for populations. • Cohort analysis does not allow for patient-level data. • If we run individual patients through the model via microsimulation, we can then associate data with each patient which enhances our model capabilities in several ways: • Heterogeneity – each patient can have his/her own set of patient characteristics • Event tracking – events that occur to each patient can be recorded • Heterogeneity and event tracking allow patients to take their own individual paths based on their unique characteristics and prior events. • In many cases, this creates a more robust, realistic model. 92 Module 6: Advanced Modelling Approaches Discrete Event Simulation Discrete Event Simulation (DES) models are similar in structure to Markov models, but instead of patients moving from cycle to cycle in fixed time increments , patients move from event to event based on the sampled (variable) timing of those events happening. • Similar to Markov models, patients will accumulate cost and effectiveness at any time point in the patient pathway. • DES models are computationally more efficient than Markov models in that patients move from event to event vs. moving through cycles for events to occur. 93 Module 6: Advanced Modelling Approaches Discrete Event Simulation - Example • Time-to-event vs. probabilities • At a basic level: as a probability of an event increases, the time to event decreases • If we take the example below, the DES node (D) is similar to the Markov node as it starts with the health states to the right. • Transitions are calculated based on the time to that event occurring as opposed to probability of that event occurring. • Time to event is derived from distributions, and re-sampled after event, therefore DES models require microsimulation. 94 Other complex models Interactions among individuals in the model [parallel trials] • Patient pathways may be affected by other patients in the model • Infectious Disease models where the likelihood of infection may depend on the percentage of the cohort that is currently infected • Resource constraints/queues where patients may compete for a resource like a hospital bed Dynamic Cohort • Studying a patient population as a whole rather than looking for average values per patient • Could be used for budget impact 95