USING MODELING TO IMPROVE RESOURCE ALLOCATION
* Finalists for the Lee B. Lusted Student Prize
Methods: We evaluated patient-level data from women undergoing IVF at Washington University between 2001 and 2008 to determine the cumulative probability of having a live birth stratified by IVF insurance coverage. We collected demographic data (age, patient IVF insurance coverage status, ZIP code of residence, average family income in the patient’s ZIP code of residence, antral follicle count associated with each oocyte retrieval, race, infertility diagnosis, and obstetrical history), and IVF cycle data(fresh versus frozen-embryo transfer, IVF cycle number, year of IVF cycle, total gonadotropin used, peak estradiol, use of intracytoplasmic sperm injection (ICSI), and number of embryos transferred) to control for potential confounders. Our primary outcome was the cumulative chance of having a live birth from IVF after five cycles according to IVF insurance coverage status, which was determined by incorporating the probability of obtaining a live birth each cycle and the patient’s decision to return for additional treatment cycles.
Results: Women with IVF insurance coverage were more likely to return for an additional IVF cycle after an unsuccessful cycle. The cumulative probability of having a live birth after five cycles for those with IVF insurance coverage was 0.635 (95% CI = [0.5592, 0.7132]) and 0.5553 (95% CI = [0.4738, 0.6393]) for those without coverage. The difference in the cumulative live birth rate between the insured and the uninsured was significant at 0.0793 with 95% CI = [0.0237, 0.1337]. This indicated that the insured patients had a significantly higher cumulative live birth rate after five cycles of IVF when we explicitly took into account the probability of returning for an additional cycle of IVF after failure.
Conclusions: By increasing the likelihood that individuals will return for additional IVF treatment after a failed IVF cycle, IVF insurance coverage is associated with an increased cumulative chance of having a live birth.
Purpose: Thrombolytic treatment (tissue-type plasminogen activator [tPA]) is only recommended for acute ischemic stroke patients with stroke onset time <4.5 hours due to higher bleeding risks with onset time >4.5 hours. tPA is not recommended when stroke onset time is unknown (14-28% of ischemic strokes). Magnetic resonance imaging (MRI) of the brain can be used to estimate stroke onset time with considerable accuracy. We projected health benefits, risks, and costs of image-based treatment decisions versus the current recommendation (no treatment) for acute stroke patients with unknown stroke onset time.
Methods: We developed a micro-simulation model that assigned patients a true stroke onset time from a beta distribution (average value 6.5 hours with uniform probability between 2.5-10.5 hours). Brain MRI used to estimate stroke onset time had estimated sensitivity and specificity of 0.62 and 0.78 respectively from the literature, cost $488, and delayed treatment by 30 minutes. True stroke onset time affected the impact of tPA on the probability of a favorable acute stroke outcome (modified Rankin score of 0-1 [mRS0-1]; odds ratio range 1.0-2.6 depending on true onset time) and the risk of major bleeding events (substantial intracerebral hemorhrage [sICH]; odds ratio range 3.5-8.5). Cost, utility, and disease progression parameters were estimated from published sources and depended on treatment status (tPA cost $16,740), mRS (utility values range 0.2-0.8), and sICH (54% fatal) outcomes. Discounted lifetime costs and health benefits (quality-adjusted life years [QALYs]) were projected for each strategy. In a sensitivity analysis, true stroke time was assumed to be left-skewed (average 4.8 hours with skewed beta distribution).
Results: With no treatment, 45.1% and 1.0% patients with unknown stroke time experienced mRS0-1 and sICH outcomes, respectively, with 5.125 lifetime QALYs and $86,949 lifetime costs; the image-based strategy resulted in 46.4% mRS0-1, 3.0% sICH, 5.150 QALYs, and $92,356 costs. The incremental cost-effectiveness ratio (ICER) for image-based treatment versus no treatment was $220,000/QALY. The ICER using a left-skewed beta distribution was $91,000/QALY. Results were sensitive to MRI sensitivity and specificity (Figure).
Conclusions: The cost-effectiveness of image-based treatment decisions versus the current recommendation of no tPA treatment for acute stroke patients with unknown onset time is sensitive to assumptions about true stroke onset time. Cost-effective image-based treatment (<$100,000/QALY) for these patients could be achieved with improved MRI diagnostic performance.
Purpose: Proper evaluation and diagnosis of coronary artery disease (CAD) is an essential part of public health strategies. In Brazil, the Unified National Health System (SUS) currently reimburses exercise electrocardiogram (Ex-ECG), stress echocardiogram (ECHO), and single-photon emission computed tomography (SPECT), but not coronary computed tomography coronary angiogram (CT) or stress cardiac magnetic resonance (MRI). We sought to compare the cost effectiveness of different testing strategies, measured as cost per correct diagnosis.
Methods: We built a decision-analytic model, comparing eleven strategies combining sequential tests (figure 1) for evaluating patients with possible stable angina, from the public health system's perspective. We used available data from published meta-analyses of test performance and 2013 SUS reimbursement rates as source of costs for diagnostic tests. The costs of CT and MRI were estimated based on costs in the private sector. All results are expressed in International Dollars (I$). The main parameter test's performance and costs are: Ex-ECG (sensitivity 65%, specificity 67% and cost I$ 16.1), ECHO (sensitivity 85%, specificity 77% and cost I$ 88.7), SPECT (sensitivity 87%, specificity 64% and cost I$ 425.6), CT (sensitivity 88%, specificity 87% and cost I$ 102.7), MRI (sensitivity 89%, specificity 80% and cost I$ 203.8) and CA (sensitivity 100%, specificity 100% and cost I$ 330.5).
Results: Figure 1 illustrates the cost-effectiveness results for each pretest probability. Strategies using Ex-ECG as initial test were the least costly alternatives, but showed the disadvantage of generating a larger number of false-positive initial tests and false-negative final diagnosis. Strategies 6 and 9, based on CT and ECHO as initial test, result in almost superposable cost-effectiveness results. Strategy 8, based on C-MRI, was highly effective for diagnosing stable CAD, but its relatively high cost resulted in unfavorable ICERs in moderate- and high-risk scenarios. Noninvasive strategies based on SPECT (3 and 5) generated consistently unfavorable results, due to the high cost of SPECT when compared to other noninvasive tests, and have been dominated in all scenarios.
Conclusions: Incorporation of coronary computed tomography into SUS would add a cost-effective option for CAD diagnosis. Stress cardiac magnetic resonance yielded acceptable ICER only at low pretest probability. Stress echocardiography is an alternative option among currently available tests.
The prevalence of hepatitis C virus (HCV) in United Stated prisons is between 16-41%; however, no standard protocols exist for HCV screening. The objective of our study was to evaluate the cost-effectiveness HCV screening in prisons and HCV prevention in society by interventions in prisons.
We developed an agent-based simulation model that simulated the transmission and progression of HCV disease in United States population in prisons and society. Chronic stages of HCV were modeled as Markov states. We used Bureau of Justice Statistics data to simulate movement of people between prisons and society by incorporating data on arrest, recidivism and duration of sentence. We evaluated three screening scenarios: risk-based screening only (RISK), 1-time screening of all existing inmates followed by screening of any incoming inmate for 2 years (2YR), and 1-time screening of all existing inmates followed by screening of any incoming inmate for 5 years (5YR). Inmates with length of sentence more than 12 months were eligible for treatment with recently approved direct-acting antivirals. We assigned treatment based on patient’s HCV genotype, treatment-history and tolerance to interferon, as recommended by the new clinical guidelines. We projected the total cost, quality-adjusted life years (QALYs), cumulative incidence of cirrhosis, decompensated cirrhosis (DC), hepatocellular carcinoma (HCC), and liver-related deaths (LRD). We also projected the number of new HCV infections in society because of HCV-infected people released from prisons.
The total cost under RISK, 2YR, and 5YR were $11.08 million, $11.68 million and $11.75 million per 10,000 people, respectively. The corresponding total QALYs were 189 417, 189 439 and 189 440, respectively. The incremental cost-effectiveness ratio of screening for 2YR and 5YR were $28,000 and $60,000 per QALY, respectively. Compared with RISK, 2YR and 5YR could avoid 56,700 and 63,300 new HCV infections in the next 30 years, where 53% of these infections can be attributed to HCV-infected persons released from correctional facilities.
HCV screening followed by treatment with recently approved therapies in prisons is highly cost-effective. Our study quantifies reduction in HCV transmission and disease burden in society at large as a result of screening in prisons. Resources spent in prisons can substantially reduce the burden of HCV in both prisons and the society at large.
Method: In collaboration with an expert panel we developed a one-year decision tree model that captured successful treatment as well as various complications including deep vein thrombosis and pulmonary embolism. We undertook a review of the literature to determine the clinical effectiveness and the risk of complications for the five methods. Data on resource use were collected from surgical departments while utility weights for varicose vein disease and treatment complication were captured from scientific literature. Uncertainty in utility and probability parameters was expressed with beta distributions while gamma distributions were used for costs.
Result: In the societal perspective, the laser ablation strategy compared to no therapy entailed one year costs of $2,193 and 0.180 QALY ($11,888 per QALY), and had a 42% probability of being cost-effective using the national willingness-to-pay threshold of $85,000. The incremental cost per QALY of replacing laser ablation with steam vein sclerosis was $55,244 while the three other alternatives were dominated. In a health care perspective, however, the steam vein sclerosis strategy had the lowest incremental cost-effectiveness ratio ($5,731 per QALY) while all other treatments were dominated, but still steam vein sclerosis had only a 50% probability of being cost-effective.
Conclusion: In a societal perspective, laser ablation is a cost-effective intervention with the current willingness to pay for a QALY, while other strategies were either cost-ineffective or were dominated. In a health care perspective, steam vein sclerosis was the treatment of choice.