Can Pay-for Performance Incentive Levels be Determined using a Cost-Effectiveness Framework?

Abstract

Background:

Healthcare payers in the U.S. are increasingly tying provider payments to quality and value using pay-for-performance policies. Cost-effectiveness analysis quantifies value in healthcare, but is not currently used to design or prioritize pay-for-performance strategies or metrics. Acute ischemic stroke (AIS) care provides a useful application to demonstrate how simulation modeling can be used to determine cost-effective levels of financial incentives used in pay-for-performance policies, and associated challenges with this approach.

Methods and Results:

Our framework requires a simulation model that can estimate quality-adjusted life years (QALYs) and costs resulting from improvements in a quality metric. A monetary level of incentives can then be back-calculated using the lifetime discounted QALY (which includes effectiveness of quality improvement) and cost (which includes incentive payments and cost offsets from quality improvements) outputs from the model. We applied this framework to an AIS micro-simulation model to calculate the difference in population-level net monetary benefit (NMB, willingness-to-pay of $50,000–150,000/QALY) accrued under current Medicare policy (stroke payment not adjusted for performance) compared to various hypothetical pay-for-performance policies. Performance measurement was based on time-to-thrombolytic treatment with tissue-type plasminogen activator (tPA). Compared to current payment, equivalent population-level NMB was achieved in pay-for-performance policies with 10-minute door-to-needle time reductions (5,057 more AIS cases/year in the 0–3 hour window) incentivized by increasing tPA payment by as much as 18–44% depending on willingness-to-pay for health.

Conclusions:

Cost-effectiveness modeling can be used to determine the upper bound of financial incentives used in pay-for-performance policies, although currently this approach is limited due to data requirements and modeling assumptions. For tPA payments in AIS, our model-based results suggest financial incentives leading to a 10-minute decrease in door-to-needle time should be implemented but not exceed 18–44% of current tPA payment. In general, the optimal level of financial incentives will depend on willingness-to-pay for health and other modeling assumptions around parameter uncertainty and the relationship between quality improvements and long-run quality-adjusted life expectancy and costs.

Introduction

Healthcare payers in the U.S. and other developed countries are increasingly tying provider payments to quality and value of care using pay-for-performance policies. Under these policies, physicians and/or hospitals are paid more for meeting evidence-based quality targets. Determining the level of incentives for high-quality care is a key challenge faced by policy makers, who currently use ad-hoc approaches, such as 1–2% of hospital payments or $4,000 in semiannual provider payments, to approximate the incentives necessary for individuals or organizations to change their behavior.^1–3

Cost-effectiveness analysis provides a formal method to quantify value in healthcare, but is not currently being used to inform pay-for-performance policies. In contrast to ad-hoc approaches for determining the level of these incentives, cost-effectiveness analysis can quantitatively weigh health benefits produced from incentivizing higher quality care against the incremental costs needed to achieve these health improvements. From a healthcare payer perspective, these incremental costs are incentive payments to providers, in addition to the net costs of higher utilization of incentivized care minus the cost savings from avoided adverse health events. In 2014, the American College of Cardiology/American Heart Association joint policy stated that they would start using cost-effectiveness information to define value assessments in practice guidelines and performance measures, further specifying that cost-per-quality-adjusted life year (QALY) results below $50,000/QALY constitutes “high value,” between $50,000/QALY and $150,000/QALY indicates “intermediate value”, and greater than $150,000/QALY indicates “low value care.”⁴ These benchmarks have yet to be implemented in pay-for-performance policies.

Acute stroke care provides a useful application for demonstrating this new approach of using cost-effectiveness analysis to determine the maximum that the health care system would be willing to pay in terms of financial incentives to improve the quality of care. The American Stroke Association has endorsed 15 quality measures for acute stroke care in 2014, including time-to-intravenous thrombolytic therapy.⁵ Previously published studies have modeled the cost-effectiveness of intravenous thrombolytic therapy with tissue-type plasminogen activator (tPA) for different time-to-treatment windows, but none of these studies connected cost-effectiveness results to pay-for-performance targets.^6–9 Therefore, the objective of our study was to illustrate how pay-for-performance incentives can be quantitatively bounded using cost-effectiveness modeling, through the application of reimbursement to hospitals for faster time-to-tPA for acute ischemic stroke.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Cost-effectiveness-based framework

Our framework requires a simulation model that can estimate QALY gains and incremental costs resulting from improvements in a recognized quality metric. Then, using the lifetime discounted QALY and cost outputs from the model, a monetary level of incentives can be back-calculated using the following equations:

1. Incremental net monetary benefit from quality improvement (iNMBQI) =∑_j(λ*ΔQALYs - Δcosts), where j is the j^th individual being simulated in the model, λ is the willingness-to-pay for health set at $100,000/QALY (a commonly-used cost-effectiveness threshold for the United States) or some other value, and the Δ symbols refer to the difference with and without the quality improvements.^{10, 11}

   For a specified level of quality improvement, the downstream effects in terms of improved health outcomes (i.e., ΔQALYs) and costs (Δcosts, which includes incentive payments) are predicted by the simulation model. A monetary value can be assigned to the incremental health benefits, given λ. Under an assumed value of λ, the maximal incentive level (a component of Δcosts) should be no greater than that which enables iNMBQI to equal zero.

2. When iNMBQI = $0, the decision-maker is indifferent to scenarios with and without the quality improvements. The upper bound for a monetary incentive for the quality improvement, which is part of Δcosts, can therefore be solved for. If iNMBQI is >$0, it would imply that a health care payer would be willing to pay more in incentive payments. If iNMBQI is <$0, it would imply that the financial incentive cost is too expensive.

Overview of acute stroke application

We used a previously published computer-based acute ischemic stroke micro-simulation model comparing a tPA versus no tPA treatment to calculate population-level incremental net monetary benefit (iNMB) accrued under current Medicare policy (stroke payment not adjusted for performance) compared to various hypothetical pay-for-performance policies.^{8, 12} Compared to current Medicare tPA payment policy (status quo), pay-for-performance policies would be expected to result in higher costs (from incentive payments) and improved health outcomes (from improved quality of acute stroke care). For any improvement in acute stroke care quality (operationalized as door-to-needle time reductions), we calculated the corresponding level of tPA payment increase (i.e., the level of financial incentives) to achieve equivalent population-level iNMB compared to the status quo policy. Per economic evaluation theory, those calculated payment increases correspond to the upper bounds payers should be willing to pay for the corresponding improvements in acute stroke care.¹³

Our acute stroke model was programmed in Tree-Age Pro 2019 (TreeAge Software, Williamstown, Massachusetts) and results were also analyzed using Microsoft Excel with Visual Basic for Applications.

Acute stroke model structure and inputs

The key biological relationship that we modeled was time from stroke onset to thrombolytic treatment with tissue-type plasminogen activator (tPA) and acute stroke outcome. Specifically, shorter times from stroke onset to tPA administration are associated with better acute (90-day) stroke outcomes as characterized by the Modified Rankin Scale (mRS), which ranges from 0 (no stroke symptoms) to 6 (death).^{14, 15} Short-term (i.e., 90-day) mRS outcomes influence long-term mortality, morbidity, and costs, which are the components needed¹⁶ to conduct a cost-effectiveness analysis. Figure 1 shows the schematic of the model with the tradeoffs between the health gains from shorter time-to-treatment against the added costs from financial incentives required to generate these improvements in acute stroke care quality. Table 1 shows the main variables and data sources used to estimate base-case values for the model-based analyses. Additional details on the acute stroke simulation model are shown in the Appendix.

Figure 1.

Schematic of model-based analyses to estimate cost-effective levels of financial incentives used in acute stroke pay-for-performance policies. We used a computer-based simulation model to quantify the tradeoffs in acute stroke care (specifically, shorter door-to-needle times) versus the financial incentives needed to achieve the health gains, which were quantified using quality-adjusted life years (QALYs). We quantified these tradeoffs using incremental net monetary benefit), which monetizes health gains and subtracts net costs to create a single metric for overall value.

Table 1.

Key model variables with base-case values and data sources

Variable	Base-case value	Description of data source	Reference(s)
General parameters
Age (years, mean and distribution range)	71 (40–90)	Meta-analysis of individual-level data from tPA RCTs	Emberson 201415
Male (%)	55
Discount rate (%)	3.0	Expert panel recommendation	Sanders 201617
Distribution of times from stroke onset to hospital arrival (% of all ischemic strokes)*
0–0.5 hours	6.9	Observed data from Get with The Guidelines (GWTG) stroke program hospitals	Tong 201218
0.5–2.0 hours	13.7		Tong 201218
2.0–3.0 hours	4.5		Tong 201218
3.0–3.5 hours	1.8		Tong 201218
3.5–4.5 hours	2.8		Tong 201218
Time from hospital arrival to tPA treatment start (door-to-needle time)
Average time (minutes)	67 minutes	Observational data from the GWTG Target Stroke initiative	Fonarow 201419
Time reduction (base-case, sensitivity analysis range)	10 minutes (0–30)		Fonarow 201419
tPA effectiveness (odds ratio of good stroke outcome [90-day mRS 0–1]) by time from stroke onset to treatment start†
1.0 hour	1.88	Meta-analysis of individual-level data from tPA RCTs	Emberson 201415
3.0 hours	1.48		Emberson 201415
4.5 hours	1.23		Emberson 201415
tPA cost (difference in Medicare reimbursement for acute stroke with tpA – acute stroke without tPA)
Base-case (pay-for-performance increases)	$7,997 (0–50%)	Inpatient sample of community hospitals in the U.S	AHRQ/HCUP 201120

^*note these times are added to door-to-needle times to calculate onset-to-treatment times

^†odds ratio was modeled as a function of time from stroke onset to treatment start (in minutes, represented by “t” in the following equation) based on data from Emberson et al. 2014: OR (mRS 0–1) = 2.115*e^(−0.002*t)

Note: “RCT” stands for randomized controlled trial, “tPA” stands for tissue-type plasminogen activator

We modeled door-to-needle time reductions by shifting the distribution of time-to-treatment windows towards earlier times, resulting in more patients in earlier treatment windows (0–3.0 hours) and fewer patients in the time window that just missed tPA eligibility (i.e., the 4.5–5.5 hour window). The effectiveness of acute stroke treatment was quantified from mRS-specific outcome distributions for patients who did and did not receive tPA based on pooled data from multiple tPA randomized controlled trials.^{14, 15} We estimated the incremental net costs of tPA (the cost difference of acute stroke costs with tPA and acute stroke costs without tPA) based on cost data from commercial plan and Medicare claims, accounting for age and stroke disability (90-day mRS state, shown in Appendix Table A-1).²¹ For hypothetical pay-for-performance policies, we modeled tPA payment increases up to 50%, assuming these increased payments would incentivize providers to improve door-to-needle times. We reported the upper bounds of tPA payment increases that would be justified for door-to-needle time reductions of 5, 10, and 20 minutes.

Post-stroke simulation model

Lifetime quality-adjusted life years (QALYs) and long-term stroke-related healthcare costs were simulated using a state-transition model with mRS-based health states (Appendix Figure A-1). Patients entered the post-stroke model with the mRS outcome projected from the acute stroke model. In yearly cycles patients were exposed to risks of dying from non-stroke causes²² or having recurrent stroke (annual probability of 5.1%), 19% of which were fatal.⁷ We estimated age- and sex-specific risks of dying from non-stroke causes by subtracting the proportions of deaths related to stroke²² from all-cause life tables²³ for each year of age, by sex. In the model, each mRS state has a corresponding utility and long-term cost (Appendix Table A-1), which were used to estimate the lifetime discounted QALYs and costs. Base-case long-term cost values were based on a 2017 study by Shireman et al., which accounted for Medicare inpatient, outpatient, and nursing facility costs (annual costs ranging across mRS states from $11,345 to $69,048).²⁴

Cost-effectiveness analysis and incremental net monetary benefit

We calculated incremental cost-effectiveness ratios (ICERs) based on QALY and cost projections for tPA treatment versus no tPA treatment for two patient groups (in the 0–3.0 and 3.0–4.5 hour treatment windows). We assumed patients in the 4.5–5.5 hour window were not treated per current clinical recommendations.²⁵ For pay-for-performance policies, the total cost of the tPA strategies depended on the level of tPA incentive payment increase. The analysis was conducted from a healthcare payer perspective over a lifetime horizon and future healthcare costs and QALYs were discounted at 3% annually.¹⁷ All costs are reported in 2019 dollars.

To estimate population-level iNMB, we first calculated the population size for annual acute stroke patients in the 0–5.5 window without other non-time-related contraindications to tPA (such as uncontrolled hypertension or other risk factors for bleeding adverse events) using the following formula: 795,000 annual new or recurrent strokes in the United States²⁶ * 87% ischemic strokes²⁶ * 29.4% of these strokes presenting within the 0–5.5 treatment window¹⁸ * 77.8% eligible for tPA (for reasons aside from time from stroke onset)²⁷ = 158,059 acute stroke patients. Unlike the patient group-specific ICERs, which were calculated on a per-patient basis conditional on being in a specific time-to-treatment window, population-level iNMB accounted for both the effectiveness and costs of tPA in addition to how many patients in the population were treated with tPA. For pay-for-performance policies, reductions in door-to-needle time would affect population-level iNMB, but not patient group-specific ICERs. Our base-case analysis aimed to determine the increase in tPA payment for a door-to-needle time reduction that would result in the same population-level iNMB as the status quo (i.e., iNMBQI = $0). Per cost-effectiveness standards, incentives should not exceed levels where the resulting improvements in quality (and health) lead to negative iNMBQI.

Sensitivity analyses

In addition to our base-case analysis, we sought to find equivalent iNMB for different payment increases for tPA when administered in the 0–3.0 and 3.0–4.5 hour windows, to model the possibility that differential payment increases could result in maximized door-to-needle time reductions. We also performed a two-way sensitivity analysis on different combinations of the levels of door-to-needle time reductions (0–30 minutes) and tPA payment increases (0–50%) to show combinations of these inputs that resulted in positive, negative, or neutral changes in iNMB for the pay-for-performance policy compared to status quo care (defined by door-to-needle times and tPA payment levels). We performed this two-way sensitivity analyses using a willingness-to-pay value of $100,000/QALY for the base-case analysis and separately for $50,000/QALY and $150,000/QALY in sensitivity analyses. We also performed a sensitivity analysis assuming that door-to-needle time reductions would not add patients to the 0–4.5 hour treatment window (because many stroke centers already rush to treat patients close to the 4.5 hour treatment threshold). In this sensitivity analysis, door-to-needle time reductions would only improve treatment times for the 142,490 annual ischemic strokes already in the 0–4.5 hour treatment window.

Results

Assuming status quo door-to-needle time, treating acute stroke patients with tPA in the 0–4.5 treatment window was modeled to result in 15,638 more annual good stroke outcomes (mRS 0–1) compared to no tPA. Reducing the door-to-needle time for the stroke population by 10 minutes moved 5,057 more annual strokes to the earlier (0–3 hour) treatment window compared to status quo projections. Table 2 shows the time distributions in the 0–3, 3–4.5, and 4.5–5.5 hour windows under status quo and pay-for-performance policies that resulted in 5- and 10-minute door-to-needle time reductions. Door-to-needle time reductions of 5, 10, and 20 minutes would result in 253, 518, and 1,062 additional good stroke outcomes (mRS 0–1) annually, respectively, compared to tPA treatment assuming status quo door-to-needle times.

Table 2:

Model-based results for acute stroke treatment with tPA under various payment and quality (door-to-needle) time scenarios

Time category (patient group)	% of patients*	tPA payment	Per-person incremental costs†	Per-person incremental QALYs†	ICER‡	Population-level iNMB#
No tPA payment adjustment and status quo door-to-needle time (72 minutes)
0–3.0 hours	66.5	$7,997	−$15,413	0.451	tPA dominant	$6.357 billion
3.0–4.5 hours	23.6	$7,997	−$11,235	0.223	tPA dominant	$1.255 billion
4.5–5.5 hours	9.9	[not treated]	[not treated]	[not treated]	[not treated]	$0
Sum iNMB:						$7.612 billion
Increased tPA payment by 14.2% for 0–4.5 hour window and 5 minute reduction in door-to-needle time
0–3.0 hours	66.6	$9,129**	−$15,521	0.457	tPA dominant	$6.506 billion
3.0–4.5 hours	21.9	$9,129**	−$11,160	0.219	tPA dominant	$1.106 billion
4.5–5.5 hours	9.5	[not treated]	[not treated]	[not treated]	[not treated]	$0
Sum iNMB:						$7.612 billion
Increased tPA payment by 30.9% for 0–4.5 hour window and 10 minute reduction in door-to-needle time
0–3.0 hours	70.1	$10,473**	−$15,627	0.462	tPA dominant	$6.604 billion
3.0–4.5 hours	20.9	$10,473**	−$11,140	0.218	tPA dominant	$1.008 billion
4.5–5.5 hours	9.0	[not treated]	[not treated]	[not treated]	[not treated]	$0
Sum iNMB:						$7.612 billion

^*among patients who would receive tPA 0–5.5. hours from stroke onset

^†lifetime discounted costs and QALYs; incremental difference for tPA treatment compared to no tPA treatment

^‡incremental cost-effectiveness ratio (ICER) of tPA compared to no tPA treatment in same time category (i.e., conditioning on time-to-treatment); “tPA dominant” indicates tPA strategy had more QALYs and lower costs than no tPA treatment; “QALY” stands for quality-adjusted life year; lifetime QALY and cost results used to calculate ICERs were discounted at 3%

^#population-level incremental net monetary benefit (iNMB) for tPA treatment vs. no tPA treatment, which is calculated using both: 1) percent of patients in a given time category; 2) and cost-effectiveness tPA treatment within the given time category. iNMB scaled up to population-level assuming annual 158,059 incident ischemic strokes in the 0–5.5. hour time window. iNMB calculated as: (incremental QALYs* willingness-to-pay for health – incremental costs)

^**these tPA payment amounts were solved for such that population iNMBQI = $0 for each quality improvement scenario

Table 2 shows the cost-effectiveness results of tPA by treatment window for various potential pay-for-performance polices (defined by varying door-to-needle time reductions and tPA incentive payment levels). Treatment with tPA resulted in more QALYs and lower costs compared to no tPA treatment (i.e., treatment with tPA was dominant) in both the 0–3.0 and 3.0–4.5 hour treatment windows and remained dominant compared to no treatment when tPA payments increased. A door-to-needle time reduction of 5 minutes coupled with a 14.2% tPA payment increase resulted in the same population-level iNMB ($7.612 billion) compared to status quo door-to-needle times and tPA payment; in other words, the threshold tPA payment increase for a 5-minute door-to-needle time reduction was $9,129, which represents a 14.2% increase from the status quo tPA cost of $7,997. The same population-level iNMB result was achieved with a 30.9% payment increase for tPA in the 0–4.5 hour treatment windows coupled with a door-to-needle time reduction of 10 minutes.

Figure 2 shows the combined effects of door-to-needle time reductions (ranging from 0–30 minutes) and tPA payment increases (ranging from 0–100%) on population-level iNMB. These results were sensitive to changes in the willingness-to-pay for health (Appendix Figure A-2 and Figure A-3). A door-to-needle time reduction of 10 minutes had equivalent population-level iNMB compared to the status quo with tPA payment increases of 18.3% and 43.6% using willingness-to-pay values of $50,000/QALY and $150,000/QALY, respectively. In the sensitivity analysis assuming no additional stroke patients were treated as a result of door-to-needle time reductions, we found a lower threshold payment increase of (28.3% for a 10-minute reduction) compared to our base-case result (30.9% for the same payment increase). This result was expected because the only difference between these scenarios were the 15,569 acute stroke patients who gained (or did not gain) treatment eligibility (i.e., crossed the 4.5-hour threshold) due to door-to-needle time improvements.

Figure 2.

Two-way sensitivity analysis showing the difference in population-level incremental net monetary benefit (pay-for-performance scenarios compared to the status quo) for different combinations of the levels of door-to-needle time reductions and tPA payment increases. The green regions show combinations of values that resulted in positive iNMB compared to status quo acute stroke treatment (population-level ICER<$50,000/QALY); yellow indicates similar iNMB (population-level ICER around $100,000/QALY); and red indicates negative iNMB (population-level ICER>$150,000/QALY).

Discussion

In this paper, we provide a cost-effectiveness-based framework for how pay-for-performance incentive levels can be determined. Our suggested framework converts the projected costs and effects from quality improvements to lifetime discounted QALYs and costs, then back-calculates the upper value of a financial incentive (a component of incremental costs) that can be justified given a specified willingness-to-pay for health gained (cost-per-QALY) threshold. This framework provides a quantitative justification for quality incentive levels used in pay-for-performance policies. However, there are technical challenges related to data requirements and modeling assumptions associated with fully implementing this approach, as highlighted in our application to acute stroke care.

We applied our framework to illustrate how the upper bound of financial incentives for faster times-to-tPA, a quality measure endorsed by the American Stroke Association⁵, could be determined using cost-effectiveness analysis. Assuming a willingness-to-pay for health of $100,000/QALY, and that pay-for-performance on this quality measure would be effective in reducing door to needle time, for which there is no direct evidence for or against to date, our findings suggest that tPA payments can be increased up to 14% or 31% if incentives lead to door-to-needle time reductions of 5 or 10 minutes, respectively. A door-to-needle time reduction of 10 minutes, which was the average difference observed in the large GWTG Stroke Quality Initiative, would result in 5,057 more annual strokes presenting in the earlier (0–3 hour) treatment window and 518 more annual good stroke outcomes (mRS 0–1) compared to status quo door-to-needle time (national average of 67 minutes). Our modeling results show that these gains in health from improved acute stroke outcomes would be cost-effective with a corresponding increase of tPA payment from $7,997 to $10,437 (30.9% increase).

These results vary directly according to willingness-to-pay health thresholds; for example, a 10-minute reduction in door-to-needle time could justify corresponding tPA payment increases of 18% and 44% using willingness-to-pay values of $50,000 and $150,000/QALY, respectively. To transform such model-based results into policy, decision-makers would first have to decide whether they want health gains from pay-for-performance policies to reach ACC/AHA benchmarks for “high value” (cost-per-QALY gained below $50,000/QALY), “intermediate value” (below $150,000/QALY), or another standard, such as the $100,000/QALY standard commonly-used in cost-effectiveness analyses performed in the US.^{4, 11} The policy maker would also have to choose whether to base incentive payments on reaching an absolute level of performance (such as percent of eligible stroke patients with door-to-needle times less than 60 minutes, or average door-to-needle time) versus improvement compared to a relevant baseline (such as a 10-minute door-to-needle time reduction compared to the previous year). Our analyses assumed the latter, but it is an open question as to whether payers should prefer one approach over the other, or perhaps a combination of achievement, improvement, and consistency, as used by Medicare’s hospital value-based purchasing program.²⁸ Cost-effectiveness modeling could accommodate any of these alternatives, but the analyses would require more assumptions and modeling complexity for payment formulas that combine multiple measures of performance.

Previous studies have evaluated the cost-effectiveness of tPA treatment versus no treatment for relevant patient groups (in the 0–3.0 and 3.0–4.5 hour treatment windows) and found that tPA was cost-effective (ICERs less than $50,000/QALY) in both groups.^{6, 7} Our base-case analysis showed that tPA dominated no treatment in these treatment windows using newer post-stroke chronic cost estimates (that include nursing facility costs) that were not used in previous cost-effectiveness analyses. In addition to using more recent sources for model inputs, our study adds to the literature by connecting these findings to a pay-for-performance policy for an endorsed American Stroke Association quality measure. To be clear, our analysis is not a threshold analysis for the highest price of tPA that would still meet cost-effectiveness standards (a greater than seven-fold tPA price increase to $61,000 would still have an ICER less than $100,000/QALY compared to no treatment in the 0–3.0 hour treatment window, based on our model); instead, we evaluated the cost-effectiveness of hypothetical pay-for-performance policies (with improved door-to-needle times but increased costs due to added financial incentives) compared to status quo acute stroke care.

Developers of quality measurement have long indicated that the development of a useful quality measure “culminates in an analysis of the measure’s cost-effectiveness”, but payers in the U.S. have not yet explicitly included cost-effectiveness analysis in value-based payment models.^{29, 30} Our approach is an innovative application of model-based cost-effectiveness analysis, because we evaluated the use of a quality-enhancing policy intervention (incentivizing door-to-needle time improvements) that involves an acute stroke treatment (tPA) as opposed to solely evaluating the treatment.^{13, 31} Systematic reviews have shown that pay-for-performance policies are often not effective, which would be a reason against adopting these polices;^{32, 33} our approach extends this criteria beyond effectiveness to cost effectiveness. Payers could choose not to implement a pay-for-performance program that may be effective, but not effective enough, or too costly to justify the upper bound of incentive payment determined using cost-effectiveness modeling. In other words, the calculated upper bound of incentive payment could be viewed as a maximum payment used during negotiations between payers and providers before pay-for-performance policies are implemented.

Our framework also allows for decision-makers to consider additional aspects around uncertainty compared to usual methods for designing pay-for-performance policies. Following usual practice (i.e., before any cost-effectiveness modeling is performed), decision-makers consider the uncertainty underlying the effectiveness evidence for a quality measure, such as the relationship between door-to-needle time and good stroke outcomes, before deciding to attach financial incentives to performance on that measure. With our cost-effectiveness analysis-based approach, probabilistic sensitivity analysis could also allow for uncertainty from multiple model inputs to be incorporated in the results. These probabilistic analyses require additional information beyond base-case analyses, however. In particular, there must be sufficient information to inform a probability distribution for each model input variable included or individual-level data for bootstrapping. In our analysis, we lacked sufficient information to describe the uncertainty around the relationship between door-to-needle time and stroke outcomes, in addition to the uncertainty around stroke cost inputs, and therefore we did not perform a probabilistic sensitivity analysis. While incorporating uncertainty into pay-for-performance policy design using our full cost-effectiveness-based approach presents an additional data challenge, a restricted version of our framework using cost-effectiveness-based point estimates for payment increases would still have advantages over current, ad-hoc approaches used to determine pay-for-performance incentive levels.

Our study has several limitations beyond issues related to uncertainty described above. First, we do not have evidence that added financial incentives on acute stroke providers will lead to door-to-needle time reductions, because randomized controlled trials for financial incentive policies are rare.¹ Pay-for-performance policy developers must determine the value of these incentives nonetheless, and we believe our cost-effectiveness modeling-based approach is a more quantitative, structured, and transparent approach compared to subjectively approximating amounts deemed sufficient to motivate quality improvement, such as increasing hospital payments by 1 to 2 percent.² Second, our framework applies to process measures (such as door-to-needle time) as opposed to more commonly-used utilization-based outcome-measures, which can be hard to convert to QALYs (such as reducing 30-day readmissions).^{29, 34} However, we believe our approach can be complementary to existing value-based policies, such as bonuses and penalties related to evidence-based care.³⁵ Third, private payers may have financial reasons to focus on shorter time horizons (such as 1- or 3-year horizons) compared to recommended cost-effectiveness time horizons (such as lifetime horizons, which we used in our analysis), which should capture all relevant cost and health outcomes affected by an intervention.¹⁶ Private payers may also face different prices for services or prefer to base decisions on budgetary impact (as opposed to cost effectiveness), which may limit how generalizable our approach could be. Therefore, decision-makers with broader accountability, such as state and federal policy makers, should promote (or continue to promote, in the case of the ACC/AHA⁴) tying financial incentives to cost effectiveness as a way to bridge the gap between societal goals (which are reflected by cost-effectiveness results) and the bottom lines of payers, physicians, and hospitals (all of whom should respond to financial incentives on quality). Financial incentives can influence practice in acute stroke; previous research has shown that hospitals used more tPA when DRG 559 was created, which increased payment for ischemic strokes treated with thrombolysis.³⁶

In conclusion, we showed how value-based acute ischemic stroke payment adjustments can be systematically determined using a model-based cost-effectiveness framework that could potentially be generalized to other quality measures across health conditions. For tPA payments in acute stroke care, cost-effective financial incentives leading to a 10-minute decrease in door-to-needle time could be as high as 18–44% (depending on willingness-to-pay for health) of current tPA payments based on our model results. In general, the optimal level of financial incentives will depend on willingness-to-pay for health and other modeling assumptions around parameter uncertainty and the relationship between quality improvements and long-run quality-adjusted life expectancy and costs.

What is known

• The levels of monetary incentives used in pay-for-performance policies are currently set using ad-hoc or subjective method (e.g., 1–2% of total payments).

• Cost-effectiveness analysis provides a quantitative metric for measuring value in health care, but is not currently used to determine the levels of incentives used in pay-for-performance policies.

What the Study Adds

• The upper bounds for financial incentives used in pay-for-performance policies can be set using information from cost-effectiveness modeling.

• For tissue-type plasminogen activator (tPA) payments in acute stroke care, cost-effective financial incentives leading to a 10-minute decrease in door-to-needle time could be as high as 18–44% of current tPA payment based on simulation model results.

• In general, the optimal level of financial incentives will depend on willingness-to-pay for health and other modeling assumptions around parameter uncertainty and the relationship between quality improvements and long-run quality-adjusted life expectancy and costs.