Can Cost-effectiveness Analysis Inform Genotype-Guided Aspirin Use for Primary Colorectal Cancer Prevention?

This article is featured in Highlights of This Issue, p. 1033

Colorectal cancer is the third most common cancer worldwide, mostly affecting developed countries (1). Screening with colonoscopy reduces incidence and mortality from colorectal cancer (2). However, risk for adverse events, patient preference, and high costs can hamper adherence to screening (3, 4). Aspirin chemoprevention of colorectal cancer is supported by a large body of strong and consistent in vitro, in vivo, clinical, and epidemiologic evidence (5–8).

The leading mechanisms for aspirin chemoprevention stem from inhibiting COX-1 and COX-2 (9, 10). COX-1 and COX-2 produce proinflammatory prostaglandins which can promote tumorigenesis, whereas COX-2 increases angiogenesis and apoptosis (11–13). Direct COX-2 inhibition is believed to the main pathway for aspirin's effectiveness against colorectal cancer, especially in COX-2 overexpressing tumors (5, 10, 14–16).

Aspirin use, even at low doses, carries the risk of potential side effects. The US Preventive Services Task Force (USPSTF) recommends low-dose aspirin for primary prevention of cardiovascular disease (CVD) and colorectal cancer in adults ages 50 to 59 years, who have a 10% or greater CVD risk and are not at an increased risk of bleeding, that is, history of gastrointestinal (GI) ulcers, recent bleeding, or concurrent use of medications that increase the risk for bleeding (17). There remains uncertainty around broad adoption of aspirin chemoprevention (18). A better understanding of the underlying determinants of efficacy and safety is needed.

Pharmacogenetic determinants moderating the chemopreventive effect of aspirin may help identify individuals mostly likely to benefit (19, 20). Inherited genetic variability in drug metabolism or biological pathways of drug efficacy can affect response to chemopreventive agents. Genetic variability in prostaglandin synthesis modifies the association of NSAID, including aspirin, and risk for colorectal cancer (21). A meta-analyses of five prospective cohort studies identified pharmacogenetic patterns modifying the cancer-chemopreventive effect of aspirin (A.N. Holowatyj; submitted for publication). Polymorphisms in ALOX5AP, PTGS2, CRP, HPGD, IL23R, ALOX5, PTGS1, and PTGES identified individuals who were more or less likely to derive aspirin benefit.

An assessment of simultaneously occurring polymorphisms affecting efficacy and safety of NSAID chemoprevention may improve risk assessment and identification of individuals that are likely to benefit from chemoprevention, along with other known risk factors, without substantial side effects of low-dose aspirin use. An important public health impact may be achieved through genetically-guided aspirin use for targeted chemoprevention of colorectal cancer. Prior economic model studies have suggested a 22% reduction in all cancer mortality even in individuals without CVD risk factors (22, 23). For prevention of proximal colorectal cancer, combining low-dose aspirin with endoscopic screening may be cost-effective (24). To date, there have been no studies investigating the cost-effectiveness of aspirin for colorectal cancer chemoprevention considering a genetically-guided approach.

We performed a base-case cost-effectiveness analysis to compare the clinical and economic outcomes of genotype-guided aspirin use compared with no genetic testing, no aspirin in adults with average-risk for colorectal cancer. Our intent was to establish a framework which may be modified as evidence for genetically-guided aspirin use for targeted chemoprevention of colorectal cancer builds to determine the commercial application of testing in this population.

A health state transition decision analytic model comparing the clinical and economic outcomes of genotype-guided aspirin use versus no genetic testing, no aspirin in an average-risk population was developed. Average-risk population included adults at least 50 years with no previous diagnosis of colorectal cancer and an average colorectal cancer and CVD risk. Primary outcomes were quality-adjusted life-years (QALY), costs, and incremental cost-effectiveness ratio (ICER). Secondary outcomes included incidence of colorectal cancer, colorectal cancer-related death, polyp developed, polyps removed, myocardial infarction (MI), GI bleeding, and intracranial hemorrhage (ICH). The ICER was calculated as incremental cost of genotype-guided aspirin use offered per each additional QALY gained over no genetic testing, no aspirin. We used $100,000/QALY willingness-to-pay (WTP) threshold to define a cost-effective result. Therefore, an ICER below 100,000/QALY would indicate genotype-guided aspirin use is a cost-effective approach. A microsimulation analysis of 100,000 average-risk individuals was performed over their lifetime, allowing for health state transitions annually. A US third party payer perspective that considers direct medical costs only was utilized. Costs were reported in US Dollars. Future costs and QALYs were discounted to present day values (annual rate 3%). The model was constructed using TreeAge Pro 2020 (TreeAge Software) and followed good research practices of cost-effectiveness analyses outlined in the ISPOR Task Force report (25).

The model structure followed the natural history of colorectal cancer, inspired by the MISCAN-colon microsimulation model (26). Six major health states, which followed the natural pathologic development of colorectal cancer, are included: (i) no adenoma/no polyp, (ii) adenoma, (iii) preclinical colorectal cancer, (iv) clinical colorectal cancer, (v) post-colorectal cancer remission, and (vi) death (Fig. 1A). However, this model included fewer health states and did not account for polyp size when compared with the MISCAN model.

Cycle length is 1 year. In each cycle, individuals remain in their current state or transition to another state. An individual in any health state can directly transition to the “Death” state. We used a simplified Markov model that took into account the probability of staying in or transitioning to each health state and how that translates over the study timeframe.

This model simulated a solitary malignant adenoma that eventually developed into colorectal cancer. Once an adenoma develops, individuals' transition to the preclinical colorectal cancer or clinical colorectal cancer health state. Preclinical colorectal cancer was defined as adenoma and undiagnosed colorectal cancer. Individuals transition to early or advanced stage clinical colorectal cancer health state according to the clinical findings once cancer is diagnosed. Both early and advanced stage clinical colorectal cancer health state was represented as colorectal cancer health state in Fig. 1A. The developed cancer stage was captured and tracked at the simulated individual level. Tracking cancer stage affected few model parameters and assumptions, such as time to transition from preclinical colorectal cancer to clinical colorectal cancer and colorectal cancer survival parameters. Individuals surviving colorectal cancer can transition to a “post-colorectal cancer remission” state. Death from colorectal cancer and death due to other causes were captured separately.

On the basis of the natural history assumption for developing colorectal cancer, we modeled an average-risk population without previous colorectal cancer. Individuals enter the model at age 20 to capture the natural progression of polyps and colorectal cancer. It was assumed that no significant polyp formation occurs before age 20, thus everyone begins in the “no adenoma/no polyp” health state (27). The model starts capturing the clinical history of each individual and accumulating costs and QALYs at 50 years; because this age is considered the greatest risk factor for developing polyps (28) and it is the initial age at which estimates of benefits and harms of prophylactic aspirin are available (Fig. 1B; ref. 23). Individuals who transition to clinically diagnosed cancer at or before age 50 were excluded. A more detailed description of the model is provided in the Supplementary Materials and Methods.

The analysis compared genotype-guided aspirin use versus no genetic testing, no aspirin. In the genotype-guided aspirin comparator, it was assumed 71% individuals would adhere to testing based on a clinical trial report of blood testing to predict colorectal cancer risk (29). Those with testing result indicating a greater reduction in colorectal cancer risk with aspirin use were recommended 81 mg aspirin daily starting at age 50. For individuals who refused genetic testing or not offered genetic testing, aspirin was not recommended.

We assumed 70% of those with positive tests would adhere to recommended aspirin use, based on a pooled estimate of adherence levels (30). This pooled estimate was based on two studies, which followed patients for an average of 1 (31) and 5 (32) years, respectively. This model assumed that adherence level of aspirin did not change with time. Effectiveness of aspirin in preventing the development of colorectal cancer was derived from a meta-analysis of five prospective cohort studies (A.N. Holowatyj; submitted for publication), these five studies have been previously reported (33–37). Adverse effects aside, adherent individuals continued aspirin use until death. The model accounted for increased risk of GI bleeding and ICH associated with aspirin use. In the event of GI bleeding or ICH, aspirin was discontinued indefinitely and colonoscopy was used for screening. The model also captured the protective effect of aspirin against MI. Our model assumed only one GI bleeding, ICH, or MI event could occur per 1-year cycle.

For both model comparators in the base case model, we assumed 58% of individuals would adhere to colonoscopy recommendations (38). Adherent individuals would receive a colonoscopy at age 50 and every 10 years afterwards. Colonoscopy frequency was increased to every 5 years after polyp detection. With early stage colorectal cancer, colonoscopy is recommended annually for the first 2 years, then every 5 years thereafter. Perforation was assumed to be the only colonoscopy-associated complication, with a 0.05% probability (38).

In senario analyses, we assessed the impact of 100% adherence with colonscopy utlizlation and 100% adherence with aspirin in the study population.

The majority of probabilities, costs, and utilities were extracted from estimates in the published literature (Supplementary Tables S1–S5). Model parameters were obtained by age groups, cancer history, and cancer stage. Data were pooled from multiple sources to provide weighted estimates when feasible.

Individuals were stratified into low and increased colorectal cancer risk categories based on results from our previous meta-analysis assessing polymorphisms in aspirin pharmacodynamics or metabolism affecting efficacy (A.N. Holowatyj; submitted for publication). This analysis examined genotypes predicting colorectal cancer risk with and without regular aspirin use to create a genetic risk score. In the genotype-guided aspirin use arm in this study, aspirin was only recommended only for individuals in the increased risk group (genetic risk score of ≥1), whereas those in low-risk group did not receive aspirin (genetic risk score of 0). ORs and prevalence of risk of colorectal cancer risk categories are provided in Supplementary Table S1 and also derived from the meta-analysis (A.N. Holowatyj; submitted for publication).

Baseline probabilities of developing adenomas were estimated from age-specific colorectal cancer incidence rates from the NCI's SEER program, genetic risk of colorectal cancer without aspirin, and assumed a mean transition of 26.7 years to symptomatic colorectal cancer. To model heterogeneity of colorectal cancer risk, baseline probabilities for developing adenomas were modified by ORs (Supplementary Tables S1 and S2).

Colorectal cancer mortality was derived from age- and stage-specific SEER survival data (39). Background mortality was derived from National Vital Statistics data (40). Values, sources for other probabilities, and additional description of the methods are shown in Supplementary Tables S1 to S5.

A mean transition time of 20 years was assumed from the formation of an adenoma to preclinical colorectal cancer (26). In the absence of screening, the mean transition time from preclinical colorectal cancer to clinical colorectal cancer was assumed to be 6.7 years (27). These transition times varied by individual and were assumed to be exponentially distributed. SEER survival data is only available for 10 years after diagnosis of colorectal cancer. However, survival curves flatten 5 to 10 years after diagnosis indicating low likelihood of colorectal cancer-related mortality beyond this time period. Therefore, colorectal cancer-related mortality was modeled for 10 years after which individuals transitioned to a post-colorectal cancer state.

Costs of testing, screening strategies and their complications, health states, and average annual healthcare costs were included in the analysis (Supplementary Table S1). Cost data were derived from published literature and publicly available sources such as Healthcare Cost and Utilization Project (HCUP) and U.S. Centers for Medicare & Medicaid Services (CMS). The cost of the test, yet to be developed, was set at $1,000 reflecting a likely highest ceiling cost and reductions in testing cost would only improve the cost-effectiveness ratio in favor of genetic-guided aspirin use (41). Costs were inflated to 2019 dollars using the healthcare component of the personal consumption expenditure index (42).

Utility values associated with noncancer and cancer-related health states were modeled for one year (Supplementary Table S1). The utility values associated with GI bleeding accrued for 1 month, and 3 months for ICH and MI. The utility for the remainder of the year after GI bleeding, ICH, and MI was calculated on the basis of the previous health state. The disutility of colonoscopy and perforation from colonoscopy were subtracted from the utility of the year in which perforation occurred (Supplementary Table S1).

A microsimulation was performed of 100,000 individuals followed from 20 years of age until death. This technique allowed individual variation in risk for developing polyps or colorectal cancer; experiencing GI bleeding, ICH, or MI; and dying from cancer or noncancer related causes. Individual values were tracked and updated for every cycle based on age, primary prevention strategy, and cancer stage and status. Modifications to risk at baseline were made using relative risks (RR) and ORs obtained from literature review.

One-way sensitivity analyses (OSA) and probabilistic sensitivity analyses (PSA) were performed to examine uncertainty of selected parameters, such as probabilities and RRs of aspirin effects, costs of health states and procedures, and utilities associated with health states. Ranges used in the sensitivity analyses are reported in Supplementary Table S1. In the OSA, parameter values were varied across a plausible range, one parameter at a time, to determine the parameter's impact. The PSA evaluated joint parameter uncertainty by randomly sampling values for each parameter simultaneously from their respective distributions and running a microsimulation. The result of this microsimulation is a joint posterior distribution of our model outputs; costs and effectiveness. This process was repeated 1,000 times; each time, parameter values were resampled and mean costs and effectiveness outcomes were calculated. Beta distributions were used for probabilities and utilities, gamma distributions were used for costs, and log-normal distributions were used for RRs and ORs.

Compared with no genetic testing, no aspirin, the genotype-guided aspirin strategy was associated with 1,461 fewer polyps developed, 378 fewer polyps detected, 510 fewer diagnosed colorectal cancer cases, and 181 fewer colorectal cancer-related deaths in an average-risk population of 100,000 individuals (Fig. 2). There were 1,078 fewer MI cases. Genotype-guided aspirin strategy resulted in 1,430 more GI bleeding events and 323 more ICH cases. In the genotype-guided aspirin strategy, 71% of patients were tested and of those 70% had a positive result (49.7% of all patients in the group). Aspirin was used among 30% of the genotype-guided aspirin group.

The mean cost of using a genotype-guided aspirin strategy was $187,109 compared with $186,464 for no genetic testing, no aspirin (Table 1). The mean QALYs resulting from a genotype-guided aspirin strategy was 19.922 compared with 19.912 for no genetic testing, no aspirin. The ICER was $66,243 per each additional QALY.

Three scenarios were assessed to understand best case results if adherence with colonoscopy and aspirin use was considered 100% (Table 1). In scenario 1, 100% adherence with colonoscopy and 70% adherence with aspirin was utilized resulting in an ICER of $126,901 per each additional QALY gained using a genotype-guided aspirin strategy. In scenario 2, 58% adherence with colonoscopy and 100% adherence with aspirin was utilized resulting in an ICER of $50,649 per each additional QALY using a genotype-guided aspirin strategy. In scenario 3, 100% adherence with colonoscopy and 100% adherence with aspirin was utilized resulting in an ICER of $54,453 per each additional QALY gained using a genotype-guided aspirin strategy.

Cost-effectiveness results were most sensitive to probability of genetic testing uptake, cost of genetic testing, probability of adhering to colonoscopy screening, and RR of developing ICH from aspirin use. Lower ICER values resulted from more uptake of the genetic testing, lower genetic testing costs, lower adherence to colonoscopy screening, and lower risk of developing ICH from aspirin. However, our model results were less sensitive to variabilities in other model inputs, such as variability surrounding adherence to aspirin use.

PSA showed that the majority of the distribution that represents differences in cost (y-axis) and effectiveness (x-axis), where each model run was represented by a dot in Fig. 3A, were in the upper right quadrant of the cost-effectiveness plane. These results indicated that increased costs and increased effectiveness were associated with genotype-guided aspirin use over no genetic testing, no aspirin. Probability that genotype-guided aspirin use was considered cost effective was 30% at WTP of $50,000/QALY and 58% at WTP of $100,000/QALY (Fig. 3B).

To the best of our knowledge, this is the first study using a cost-effectiveness analysis model to incorporate aspirin pharmacogenomics for primary chemoprevention of colorectal cancer. Our model demonstrated that genotype-guided aspirin was associated with overall fewer polyps, and prevented more colorectal cancer cases, colorectal cancer-related deaths, and MI cases when compared with no genetic testing, no aspirin. There were higher rates of bleeding events with genotype-guided aspirin. Mean cost and QALYs associated with genotype-guided aspirin strategy were similar to those of no genetic testing, no aspirin. Genotype-guided aspirin yielded an ICER of $56,283 per QALY gained, and was cost effective in 70% of simulations at $100,000 WTP threshold.

Aspirin decreases the likelihood of developing colorectal cancer. It can also increase bleeding risk. Aspirin is recommended for primary prevention of CVD and colorectal cancer in adults ages 50 to 59 years, who have a 10% or greater CVD risk and are not at an increased risk of bleeding (17). Defining the cost–benefit balance of chemoprevention with aspirin is critical for the individual and health care system. Our model simulated the current aspirin recommendations and tested the hypothesis of using genotype-guided aspirin compared with not using aspirin or genetic testing in adults over 50 years of age. Results indicate genotype-guided aspirin may be a cost-effective alternative to no genetic testing. This model offers a framework to modify inputs to calculate the gained health benefit needed to justify the additional costs for genetic testing.

USPSTF does not currently recommend universal aspirin for primary colorectal cancer prevention citing uncertainties regarding the potential harms of long term aspirin, thus universal aspirin was not included as a comparator in this model. The model demonstrated that aspirin prevented polyp formation, and a reduction in incidence of diagnosed colorectal cancer cases, colorectal cancer-related death, and MIs in genotype-guided aspirin use population. These results support the clinical value of genetic testing. Genetically-guided aspirin use, alone or with colonoscopy, prevented colorectal cancer development from its earliest stages, surpassing the benefits of other colorectal cancer screening strategies. Benefits of using genetically-guided aspirin were offset by increased rates of GI bleeding and ICH. Thus, combining the benefits of these tests with other genetic tests that assess genetic variants predictive of bleeding (40, 43–45) may advance the selection of a target population with the greatest benefits-risk profile for using aspirin as a colorectal cancer-chemopreventive agent.

Genetic polymorphisms have also been suggested as potential biomarkers for aspirin-induced GI ulceration and bleeding (40, 44). For example, polymorphisms in CHST2 were associated with an increased risk of GI bleeding in low-dose aspirin users (44). Our initial exploration including rates of CHST2 polymorphisms (44) in the current model did not improve cost-effectiveness outcomes, most likely driven by the high prevalence and small effect size associated with variant alleles of CHST2 gene. Ideally, a less-prevalent gene with large effect size would provide a potential biomarker with more favorable outcomes. Future derivations of this model could also include proton-pump inhibitors to prevent aspirin-induced GI ulceration and bleeding (46).

This is the first study integrating the increasing ability and likelihood of genetic tests used with broad pharmacologic strategies to help tailor intervention. It is particularly relevant to chemoprevention, where the risk-benefit balance needs to be shifted as much as possible towards greater population benefit to make an intervention acceptable. We tested a real-life scenario relevant to health policy, building on established model parameters.

Model structure and inputs were reflective of pragmatic colorectal cancer-related practices and outcomes. Whenever feasible, we reported scenarios that took into account realistic levels of adherence to screening in the base case model, while providing best case levels of adherence as scenario analyses. In addition, we relied on SEER survival data as an indirect measure of colorectal cancer treatment effectiveness, rather than using colorectal cancer treatments efficacy reported in controlled treatment settings. However, assumptions were made when data reported in the literature was limited, as is reported in the Materials and Methods section.

Model inputs were derived from multiple sources, which may introduce heterogeneity. To address this, we included inputs from study populations similar to our population whenever feasible. We stratified model inputs by health state and age, whenever available, as our model followed simulated individuals over their lifetime. In addition, we conducted sensitivity analyses to account for uncertainty surrounding model inputs and to evaluate their impact on the results. Internal model validation assessing colorectal cancer incidence by age was performed and comparable to SEER estimates (Supplementary Table S6). External model validation is warranted as real-world or clinical data become available with genetically-guided aspirin use to further assess the validity of model outcomes and desired properties.

To compare our model results to previously published cost-effectiveness models, we assessed the output of no screening versus screening with colonoscopy as a common metric across models and the results are comparable. In three previous studies, the number of colorectal cancer cases prevented with colonoscopy versus no screening included 2,334 (47), 4,014 (24), and 4,428 (48) cases per 100,000 individuals screened with 50% (47) and 100% (24, 48) adherence to colonoscopy modeled. This represents a 25%, 68%, and 75% RR reduction in colorectal cancer incidence, with colonoscopy in these models, respectively. In our model, colonoscopy resulted in a reduction of 1,827 and 3,152 cases of colorectal cancer per 100,000 patients screened in scenarios of 58% and 100% adherence to colonoscopy, respectively. Representing a 28% and 49% RR reduction in colorectal cancer incidence in our model and aligned with the other published results. In two of these studies, colorectal cancer-related deaths were reported and colonoscopy versus no screening resulted in 652 (47) and 1,679 (24) fewer colorectal cancer-related deaths per 100,000 patients screened with colonoscopy versus no screening, representing a 23% and 68% RR reduction in colorectal cancer-related mortality with 50% and 100% adherence to colonoscopy, respectively. Our model reported 746 and 1,287 fewer colorectal cancer-related deaths per 100,000 patients screened with colonoscopy with 58% and 100% adherence to colonoscopy, respectively; representing a 28% and 48% risk reduction in colorectal cancer-related mortality with colonoscopy compared with no intervention. On the basis of this comparison, our overall model performance is generally aligned with the outputs of other models and variations can be expected on the basis of model design and other input variables such as incidence rates of colorectal cancer, efficacy of colonoscopy, and mortality rates of colorectal cancer.

We used polymorphism scores from a prior study as a hypothetical scenario to identify individuals who would benefit from genotype-guided aspirin use (A.N. Holowatyj; submitted for publication). These risk scores have not been validated and are utilized to represent the potential application of genetically-guided aspirin using the best data available; as future research evolves the model inputs can be updated. In addition, the overall effects of aspirin reported in study determining the genetic risk scores were smaller than reported in previous studies (A.N. Holowatyj; submitted for publication). For example, Cuzick and colleagues cites a reduction of 33% in colorectal cancer incidence among those that regularly used aspirin whereas the study used in this analysis reported a reduction of approximately 23% (22). A greater risk reduction would have resulted in a lower ICER. Therefore, our analysis may underestimate the true value of genetically-guided aspirin. The optimal duration of aspirin use that maximizes chemoprevention while minimizing adverse events is unknown. In our model aspirin use was continued until death or adverse event.

Long-term low-dose aspirin use plays an integral role in preventing CVD and can have a similar role in colorectal cancer in a target population with a favorable benefit-risk profile. Genotype-guided aspirin use in the future may help identify individuals most likely to benefit from aspirin use while minimizing aspirin-related side effects. This could lead to a potential decrease in uncertainty towards using aspirin as a colorectal cancer-chemopreventive strategy and increased patient adherence. The genetically-guided use of aspirin may be applied to other diseases and to medications that share the same drug metabolism or biological pathways as aspirin. The cost-effectiveness model presented here can provide a framework to assess the impact of further modifications towards targeting aspirin prevention in colorectal cancer.

In conclusion, genetically-guided aspirin for primary colorectal cancer chemoprevention may provide in the future a cost-effective alternative to no genetic testing, no aspirin and is associated with improved clinical outcomes on a population level. A genotype-guided aspirin strategy may prevent colorectal cancer, colorectal cancer-related deaths, and MIs. Caution for bleeding complications from long-term exposure to even low-dose aspirin is still warranted when considering the commercial viability of this testing approach.

E. Biltaji reports grants from NIH during the conduct of the study. C.M. Ulrich reports nonfinancial support from Bayer during the conduct of the study. No disclosures were reported by the other authors.

E. Biltaji: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. B. Walker: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. T.H. Au: Conceptualization, data curation, investigation, writing–review and editing. Z. Rivers: software, formal analysis, validation, investigation, visualization, writing–review and editing. J. Ose: Conceptualization, data curation, investigation, writing–review and editing. C.I. Li: Conceptualization, funding acquisition, project administration, writing–review and editing. D.I. Brixner: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing. D.D. Stenehjem: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing. C.M. Ulrich: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, project administration, writing–review and editing.

NCI of NIH, projects title “Discovery and Verification of Novel Biomarkers of Colorectal Cancer Recurrence” (R01CA189184). C.M. Ulrich and J. Ose received funding from the Huntsman Cancer Foundation, NCI of NIH, projects title “Metabolomic Strategies for Discovery and Validation of Biomarkers of Colorectal Cancer Recurrence” (R01 CA207371), and “Transdisciplinary Team Science in Colorectal Cancer” (U01 CA206110). Zachary Rivers received funding from the University of Minnesota Clinical and Translational Science Institute via the NIH's National Center for Advancing Translational Sciences, grants TL1R002493 and UTL1R002494.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.