Use Of Risk Models Research Articles

Predicting electricity infrastructure induced wildfire risk in California

This paper examines the use of risk models to predict the timing and location of wildfires caused by electricity infrastructure. Our data include historical ignition and wire-down points triggered by grid infrastructure collected between 2015–2019 in Pacific Gas & Electricity territory along with various weather, vegetation, and very high resolution data on grid infrastructure including location, age, and materials. With these data we explore a range of machine learning methods and strategies to manage training data imbalance. The best area under the receiver operating characteristic we obtain is 0.776 for distribution feeder ignitions and 0.824 for transmission line wire-down events, both using the histogram-based gradient boosting tree algorithm with under-sampling. We then use these models to identify which information provides the most predictive value. After line length, we find that weather and vegetation features dominate the list of top important features for ignition or wire-down risk. Distribution ignition models show more dependence on slow-varying vegetation variables such as burn index, energy release content, and tree height, whereas transmission wire-down models rely more on primary weather variables such as wind speed and precipitation. These results point to the importance of improved vegetation modeling for feeder ignition risk models, and improved weather forecasting for transmission wire-down models. We observe that infrastructure features make small but meaningful improvements to risk model predictive power.

Improved credit default prediction using machine learning and its impact on risk-weighted assets of banks

The use of risk models, especially credit risk models, has been a standard for banks for many years. Banks use models not only for business decision purposes but also for regulatory purposes when comparing their risk with the available regulatory capital. Furthermore, banks need to efficiently allocate their capital in the current competitive and regulatory environments. As part of this process, they develop models to predict the probability of default (PD), which are further used to calculate risk-weighted assets (RWA). This paper gives an overview of how banks calculate RWA for credit risk. We compare the performance of traditional PD models based on logistic regression with a machine learning (ML) algorithm based on gradient boosting. This shows that an improvement in PD model performance by using ML algorithms can also lead to a decrease of RWA, therefore releasing additional capital for the banks. We developed and calibrated PD models based on logistic regression and light gradient boosting machine (GBM) approaches and compared them in terms of discriminatory power and the impact on RWA to prove this statement. This paper shows that the use of ML leads in our case study to an improvement in the model’s discriminatory power of 5 per cent in terms of Gini and releasing RWA of approximately 6.5 per cent.

MA05.10 Performance of Draft 2020 USPSTF Lung-Cancer Screening Guidelines and Potential for use of Risk Models to Reduce Racial/Ethnic Disparities

MA05.10 Performance of Draft 2020 USPSTF Lung-Cancer Screening Guidelines and Potential for use of Risk Models to Reduce Racial/Ethnic Disparities

Validation of Risk Prediction Models to Inform Clinical Decisions After Acute Kidney Injury

Rationale & ObjectiveThere is limited evidence to guide follow-up after acute kidney injury (AKI). Knowledge gaps include which patients to prioritize, at what time point, and for mitigation of which outcomes. In this study, we sought to compare the net benefit of risk model–based clinical decisions following AKI.Study DesignExternal validation of 2 risk models of AKI outcomes: the Grampian -Aberdeen (United Kingdom) AKI readmissions model and the Alberta (Canada) kidney disease risk model of chronic kidney disease (CKD) glomerular (G) filtration rate categories 4 and 5 (CKD G4 and G5). Process mining to delineate existing care pathways.Setting & ParticipantsValidation was based on data from adult hospital survivors of AKI from Grampian, 2011-2013.PredictorsKDIGO-based measures of AKI severity and comorbidities specified in the original models.OutcomesDeath or readmission within 90 days for all hospital survivors. Progression to new CKD G4-G5 for patients surviving at least 90 days after AKI.Analytical ApproachDecision curve analysis to assess the “net benefit” of use of risk models to guide clinical care compared to alternative approaches (eg, prioritizing all AKI, severe AKI, or only those without kidney recovery).Results26,575 of 105,461 hospital survivors in Grampian (mean age, 60.9 ± 19.8 [SD] years) were included for validation of the death or readmission model, and 9,382 patients (mean age, 60.9 ± 19.8 years) for the CKD G4-G5 model. Both models discriminated well (area under the curve [AUC], 0.77 and 0.86, respectively). Decision curve analysis showed greater net benefit for follow up of all AKI than only severe AKI in most cases. Both original and refitted models provided net benefit superior to any other decision strategy. In process mining of all hospital discharges, 41% of readmissions and deaths occurred among people recovering after AKI. 1,464 of 3,776 people (39%) readmitted after AKI had received no intervening monitoring.LimitationsBoth original models overstated risks, indicating a need for regular updating.ConclusionsFollow up after AKI has potential net benefit for preempting readmissions, death, and subsequent CKD progression. Decisions could be improved by using risk models and by focusing on AKI across a full spectrum of severity. The current lack of monitoring among many with poor outcomes indicates possible opportunities for implementation of decision support.

Risk assessment and genetic counseling for hereditary breast and ovarian cancer syndromes-Practice resource of the National Society of Genetic Counselors.

Cancer risk assessment and genetic counseling for hereditary breast and ovarian cancer (HBOC) are a communication process to inform and prepare patients for genetic test results and the related medical management. An increasing number of healthcare providers are active in the delivery of cancer risk assessment and testing, which can have enormous benefits for enhanced patient care. However, genetics professionals remain key in the multidisciplinary care of at-risk patients and their families, given their training in facilitating patients' understanding of the role of genetics in cancer development, the potential psychological, social, and medical implications associated with cancer risk assessment and genetic testing. A collaborative partnership of non-genetics and genetics experts is the ideal approach to address the growing number of patients at risk for hereditary breast and ovarian cancer. The goal of this practice resource is to provide allied health professionals an understanding of the key components of risk assessment for HBOC as well as the use of risk models and published guidelines for medical management. We also highlight what patient types are appropriate for genetic testing, what are the most appropriate test(s) to consider, and when to refer individuals to a genetics professional. This practice resource is intended to serve as a resource for allied health professionals in determining their approach to delivering comprehensive care for families and individuals facing HBOC. The cancer risk and prevalence figures in this document are based on cisgender women and men; the risks for transgender or non-binary individuals have not been studied and therefore remain poorly understood.

Abstract A110: Could use of individualized risk models mitigate health disparities in eligibility for lung cancer screening?

Abstract Background: Current US Preventive Services Task Force (USPSTF) lung cancer screening guidelines recommend screening ever-smokers aged 55-80 years with ≥30 pack-years who currently smoke or have quit in the last 15 years. However, USPSTF guidelines could exacerbate health disparities, since they do not account for race/ethnicity, gender or socioeconomic status. In particular, African-Americans have a higher risk of developing lung cancer despite smoking less than their white counterparts, suggesting that current guidelines may eliminate many African Americans from screening despite having equivalent risk to whites. Validated lung cancer risk models, some of which include race/ethnicity, have been proposed for determining screening eligibility, although it is unknown whether these would mitigate any potential disparities. Methods: Using data from the US-representative 2015 National Health Interview Survey, we calculated those eligible for screening (overall and by subpopulations) under USPSTF guidelines, and three prominent validated lung cancer risk models (Bach, PLCOM2012 and LCRAT) at a range of risk thresholds. The PLCOM2012 and LCRAT models include race/ethnicity, and the Bach and LCRAT models include gender. We also calculated the distributions of pack-years, cigarettes per day (cpd) and years smoked by race/ethnicity. Results: Using a 2.3% 6-year lung cancer risk threshold for each model, a higher proportion of those selected by risk models are African-American (PLCOM2012: 10%, LCRAT: 12%, Bach: 9%) compared to USPSTF guidelines (7%). In contrast, a lower proportion of those selected by risk models are Asian-American (PLCOM2012: 1.6%, LCRAT: 1.4%, Bach: 2.1%) compared to USPSTF guidelines (2.7%). The models disagree on the proportion selected who are Hispanic (PLCOM2012: 1.5%, Bach: 5.1%, LCRAT: 2.6%), although the proportion selected by USPSTF guidelines (3.9%) is within their range. Some of the differences between models may occur as a result of how each model assigns risk to smokers with <30 pack-years, which occurs more frequently in racial/ethnic minorities: 82% of African-American ever-smokers aged 50-80, 84% of Hispanic and 80% of Asians had smoked <30 pack-years, compared to 67% of whites. Conclusions: Substantially higher proportions of those selected for lung screening by risk models were African-American than by USPSTF guidelines. However, risk models might select smaller proportions of Asian-Americans than USPSTF guidelines and disagree on the proportion selected who are Hispanic. Given historical lack of equal access to healthcare among minorities, the use of risk models to determine screening eligibility requires careful thought. Citation Format: Hormuzd Katki, Martin Skarzynski, Li Cheung, Christine Berg, Anil Chaturvedi, Rebecca Landy, Corey Young. Could use of individualized risk models mitigate health disparities in eligibility for lung cancer screening? [abstract]. In: Proceedings of the Twelfth AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2019 Sep 20-23; San Francisco, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(6 Suppl_2):Abstract nr A110.

Basic principles of risk score formulation in medicine.

Risk models play a vital role in monitoring health care performance. Despite extensive research and the widespread use of risk models in medicine, there are methodologic problems. We reviewed the methodology used for risk models in medicine. The findings suggest that many risk models are developed in an ad hoc manner. Important aspects such as the selection of risk factors, handling of missing values, and size of the data sample used for model development are not dealt with adequately. Methodologic details presented in publications are often sparse and unclear. Model development and validation processes are not always linked to the clinical aim of the model, which may affect their clinical validity. We make some suggestions in this review for improving methodology and reporting.

Multiple predictively equivalent risk models for handling missing data at time of prediction: With an application in severe hypoglycemia risk prediction for type 2 diabetes.

The presence of missing data at the time of prediction limits the application of risk models in clinical and research settings. Common ways of handling missing data at the time of prediction include measuring the missing value and employing statistical methods. Measuring missing value incurs additional cost, whereas previously reported statistical methods results in reduced performance compared to when all variables are measured. To tackle these challenges, we introduce a new strategy, the MMTOP algorithm (Multiple models for Missing values at Time Of Prediction), which does not require measuring additional data elements or data imputation. Specifically, at model construction time, the MMTOP constructs multiple predictively equivalent risk models utilizing different risk factor sets. The collection of models are stored and to be queried at prediction time. To predict an individual's risk in the presence of incomplete data, the MMTOP selects the risk model based on measurement availability for that individual from the collection of predictively equivalent models and makes the risk prediction with the selected model. We illustrate the MMTOP with severe hypoglycemia (SH) risk prediction based on data from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) study. We identified 77 predictively equivalent models for SH with cross-validated c-index of 0.77±0.03. These models are based on 77 distinct risk factor sets containing 12-17 risk factors. In terms of handling missing data at the time of prediction, the MMTOP outperforms all four tested competitor methods and maintains consistent performance as the number of missing variables increase.

Digital interface device for field soil hydraulic conductivity measurement

Accurate and reliable measurements of field-saturated hydraulic conductivity (Kfs) are essential for accurate determinations of soil and solution water fluxes. Kfs is a crucial parameter to design drainage, irrigation systems and many other environmental, agricultural and industrial installations Reynolds et al. (1983). There is a high demand for Kfs data. Its evaluation is preferable to be done in situ. A popular device to measure the Kfs is the head well permeameter such as the Guelph Permeameter. However, it is an analog equipment based on a Mariotte’s bottle that requires full attention and training to be manually operated, where the reading of the flux values (Q, in cm3/min) is time-consuming and costly. In this paper, we describe a device, with a digital readout, to measure the steady state rate of water level change Rs, in order to obtain Qs and Kfs. The method is based on the monitoring of the vacuum variation of the Mariotte’s bottle. Our results show no statistical differences between the measurements values of manual and automatic recording using paired t-test. In addition, our device identifies measurement artifacts due to the variation of the ambient temperature during measurements. The necessary accuracy for the digitalization method becomes feasible due to the availability of accurate piezoelectric transducers (Micro-Electro-Mechanical System – MEMS). Automating and digitizing these processes will reduce time, costs and increase the accuracy and availability of data. The device can be directly adapted to commercially available head well permeameters as shown in our experimental results. The more widespread use and understanding of this parameter can improve the rational use of water for irrigation and use in risk models for forecasting available soil water and therefore crop productivity.

Abstract IA08: Individualized risk based lung cancer screening: The way forward

Abstract The United States Preventive Services Task Force (USPSTF) recommends annual low-dose computed tomography (CT) lung-cancer screening, as does the Center for Medicare and Medicaid Services (CMS) and many major medical organizations. The USPSTF criteria, similar to those used for entry into the definitive National Lung Screening Trial, are for persons aged 55-80 years, who currently smoke or quit within the past 15 years, and who have accumulated at least 30 pack-years of cigarette smoking. Risk calculators as compared to micro-simulation modeling as used for the USPSTF can more specifically account for demographic, clinical, and smoking characteristics and personalize risk. The USPSTF criteria may miss smokers at high risk for lung cancer who would have been selected for CT screening by individual risk calculators. Recently, the National Comprehensive Cancer Network guidelines for lung cancer screening added consideration of use of risk models to select ever-smokers for screening. Online risk tools are available to assist individuals and health care providers as to whether lung cancer screening may be helpful. Several lung cancer risk models are available, and comparisons across them have been done. In this presentation, a comparison we performed of the statistical predictive properties of 8 published risk models in 2 large prospective U.S. cohorts will be presented. Identifying a consensus cost-effective risk threshold to define screening eligibility is crucial. Another issue is how best to account for competing morbidities that might make screening less useful. Perhaps a life-years gained approach could help to adjust for this. Additionally, risk models require accurate risk factor data, which are rarely available in electronic health records and may require special collection. It also poses challenges in the health care delivery setting given its complexity. To better capture high-risk smokers and prevent premature deaths from lung cancer, eligibility for lung cancer screening should be based on exceeding a cost-effective risk threshold that balances CT screening benefits and harms, using a lung cancer risk tool validated in the U.S. population. Citation Format: Christine D. Berg. Individualized risk based lung cancer screening: The way forward [abstract]. In: Proceedings of the Fifth AACR-IASLC International Joint Conference: Lung Cancer Translational Science from the Bench to the Clinic; Jan 8-11, 2018; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2018;24(17_Suppl):Abstract nr IA08.

Surgeons' assessment versus risk models for predicting complications of hepato-pancreato-biliary surgery (HPB-RISC): a multicenter prospective cohort study

BackgroundSeveral studies advise the use of risk models when counseling patients for hepato-pancreato-biliary (HPB) surgery, but studies comparing these models to the surgeons' assessment are lacking. The aim of this study was to assess whether risk prediction models outperform surgeons' assessment for the risk of complications in HPB surgery. MethodsThis prospective study included adult patients scheduled for HPB surgery in three centers in the UK and the Netherlands. Primary outcome was the rate of postoperative major complications. Surgeons assessed the risk prior to surgery while blinded for the formal risk scores. Risk prediction models were retrieved via a systematic review and risk scores were calculated. For each model, discrimination and calibration were evaluated. ResultsOverall, 349 patients were included. The rate of major complications was 27% and in-hospital mortality 3%. Surgeons' assessment resulted in an AUC of 0.64; 0.71 for liver and 0.56 for pancreas surgery (P = 0.020). The AUCs for nine existing risk prediction models ranged between 0.57 and 0.73 for liver surgery and between 0.51 and 0.57 for pancreas surgery. ConclusionIn HPB surgery, existing risk prediction models do not outperform surgeons' assessment. Surgeons' assessment outperforms most risk prediction models for liver surgery although both have a poor predictive performance for pancreas surgery. Registration informationREC reference number (13/SC/0135); IRAS ID (119370). Trialregister.nlNTR4649.

Use of Risk Model for Assessment of Residents' Perception of Complexity of Surgical Steps: Example of Modular Component Steps of Lumbar Spinal Fusion Surgery.

Quality improvement projects increasingly emphasize standardization of surgical work flow to optimize operative room efficiency. Removing special cause variability resulting from nonsurgical waste is an obvious target; however, resident surgical education must be maintained, even in the setting of process improvement. To describe the impact of resident-identified "risky" or "uncomfortable" procedural steps on operative time during transforaminal lumbar interbody fusion (TLIF). TLIF procedure steps were defined. An 8 2-part questions survey regarding comfort level and perceived risk assessment at each step was developed and completed by junior (17) and senior residents (10), and by faculty (6) from orthopedic, and neurological surgery. A risk matrix was constructed defining 2 zones: a "danger zone"; responses were high risk (3-5) and low comfort (1-3), and a "safe zone"; responses were low risk (1-2) and high comfort (4-5). One-tailed Chi-square with Yates correction was performed. Risk matrix analysis showed a statistical difference among "danger zone" respondents between junior resident and faculty groups for exposure, pedicle screw placement, neural decompression, interbody placement, posterolateral fusion, and hemostasis. A radar graph identifies percent of respondents who fall within the "danger zone". Resident perception of surgical complexity can be evaluated for procedural steps using a risk matrix survey. For TLIF, residents may assign more risk and may be less comfortable performing steps in a training-level-dependent manner. Identification of particular high-risk or uncomfortable steps should prompt strict faculty oversight to improve patient safety, monitor resident education, and reduce operative time.

A Continuous-Class Queueing Model With Proportional Hazards-Based Routing

Motivated by jail overcrowding and the U.S. correctional system's widespread use of risk models to aid in inmate release decisions both prior to trial (i.e., pretrial release) and near the end of their jail sentence (i.e., split sentencing), we formulate and analyze a queueing network model with two novel features: there is a continuum of customer classes corresponding to an inmate's continuous risk level p, and routing in the network is dictated by a Cox proportional hazards model, where the hazard rate associated with recidivism (i.e., committing another crime) during release is proportional to eγp for some parameter γ. We perform an exact analysis of a continuous-class M/M/c/c (i.e., Erlang B) model where preemptive priority is awarded according to the risk level p, and use it to develop approximate performance measures for a queueing network of a jail with a two-threshold policy, which dictates who is granted pretrial release and who receives a split sentence. For a slightly simplified version of the model, we derive sufficient conditions under which no inmates are offered pretrial release unless all inmates are given a split sentence.

Understanding physician-level barriers to the use of individualized risk estimates in percutaneous coronary intervention

The foundation of precision medicine is the ability to tailor therapy based upon the expected risks and benefits of treatment for each individual patient. In a prior study, we implemented a software platform, ePRISM, to execute validated risk-stratification models for patients undergoing percutaneous coronary intervention and found substantial variability in the use of the personalized estimates to tailor care. A better understanding of physicians' perspectives about the use of individualized risk-estimates is needed to overcome barriers to their adoption. In a qualitative research study, we conducted interviews, in-person or by telephone, with 27 physicians at 8 centers that used ePRISM until thematic saturation occurred. Data were coded using descriptive content analyses. Three major themes emerged among physicians who did not use ePRISM to support decision making: (1) "Experience versus Evidence," physicians' preference to rely upon personal experience and subjective assessments rather than objective risk estimates; (2) "Omission of Therapy," the perception that the use of risk models leads to unacceptable omission of potentially beneficial therapy; and (3) "Unnecessary Information," the opinion that information derived from risk models is not needed because physicians' decision making is already sound and they already know the information. Barriers to the use of risk models in clinical practice include physicians' perceptions that their experience is sufficient, that models may lead to omission of therapy in patients that may benefit from therapy, and that they already provide good care. Anticipating and overcoming these barriers may improve the adoption of precision medicine.

Risk Everywhere: Modeling and Managing Volatility

Based on a unique high-frequency dataset for more than fifty commodities, currencies, equity indices, and fixed income instruments spanning more than two decades, we document strong similarities in realized volatilities patterns across assets and asset classes. Exploiting these similarities within and across asset classes in panel-based estimation of new realized volatility models results in superior out-of-sample R2's compared to forecasts from existing models and more conventional procedures that do not incorporate the information in the high-frequency intraday data and/or the commonalities in the volatilities. We present a framework to evaluate the utility gains from the use of risk models, highlighting the interplay between transaction costs, the speed of different risk models, and their practical implementation.

149 Use of Risk Model for Assessment of Residents Perception of Complexity of Surgical Steps

Abstract INTRODUCTION: Quality improvement projects have begun to standardize surgical work flow as a component to optimize operative room (OR) efficiency. Removing special cause variability resulting from nonsurgical waste is an obvious target; however, surgical resident education must be maintained even in the setting of process improvement. There are no published data describing the impact on operative time of resident-identified risky or uncomfortable procedural steps during posterior instrumented fusion (PIF). Self-identification of risk or discomfort in surgical steps may allow for shorter OR time and reduced cost, without sacrificing resident education. METHODS: PIF procedure steps were defined. An 8 two-part question survey regarding surgeon comfort level and perceived risk assessment at each step was developed, and completed by junior (17) and senior residents (10), and faculty (6) from orthopedic and neurological surgeons. A risk matrix was constructed defining 2 zones, a “danger zone” where responses were both high risk (3–5) and low comfort (1–3) and a “safe zone” where responses were low risk (1–2) and high comfort (4–5). One-tailed χ2 with Yates correction was performed. RESULTS: Risk matrix analysis showed a statistical difference among danger zone respondents between junior resident and faculty groups for exposure, pedicle screw placement, neural decompression, interbody placement, posterolateral fusion, and hemostasis (Table 2). Radar graph identifies percentage of respondents who fall within the danger zone (Figure 1). CONCLUSION: Resident perception of surgical complexity can be evaluated for procedural steps using a risk matrix survey. For PIF, residents assign more risk and are less comfortable performing steps in a training-dependent manner. Identification of particular high-risk steps, which are uncomfortable, should prompt strict faculty oversight to improve patient safety, monitor resident education, and reduce operative time.

All else being equal, men and women are still not the same: using risk models to understand gender disparities in care.

Despite growing awareness of the burden of cardiovascular disease (CVD) in women over the past 2 decades, reports of disparities in the delivery of health care for these diseases have persisted over time.1 Under-treatment of women as compared with men has been described for primary prevention, stable coronary artery disease (CAD) and suspected or diagnosed acute coronary syndromes in both observational studies and controlled experiments.2–6 However, because men and women are nonidentical in many ways that might be relevant for treatment decisions, a critical question for policy makers and clinicians is whether these gender-based variations in treatment correspond to lower quality care. A demonstration of clinical inequity (unequal treatment despite equal clinical need) rather than strict inequality (unequal treatment regardless of need or condition) has been proposed as a framework to identify inappropriate variations in healthcare utilization.7 Following this model, it is important to consider whether observed gender differences correspond to inequities or may reflect appropriate variations explained by differences in treatment indications and contraindications or patient preferences.7 In a well-known study published over a decade ago, investigators devised a clever experiment to address just this question of identifying inappropriate healthcare variations.8 Using actors posing as patients with chest pain, primary care physicians were less likely to refer women for cardiac catheterization than men. Because these patients were carefully designed to be identical in age, cardiac risk factors, and symptoms, followed a script standardizing patient communication, and only differed by gender (and race), it seemed obvious that biased decision-making must account for treatment differences. In this commentary, we illustrate why analyses that attempt to control for all known cardiac risk factors that affect treatment decisions might still yield misleading findings about the presence, magnitude, and causes of gender disparities. This residual bias occurs …

Risk-Based Approaches for the Remediation of a UST Site: A Case Study

Risk-based corrective action (RBCA) is a scientifically-accepted approach to remediate polluted sites. Under the RBCA approach, the risks to human health associated with polluted sites are assessed and appropriate remedial measures are taken to reduce risks. RBCA evaluations involve the use of risk models to assess health risks to different receptors. In Taiwan, the Soil and Groundwater Remediation Act was enforced in 2000. The government of Taiwan also use risk assessment protocols to obtain acceptable remediation goals for polluted sites with low risks. In this study, the application of risk assessment to derive remediation goals and develop remedial strategies at a fuel-oil spill site were performed. After the risk evaluation, the soil and groundwater remediation goals for total petroleum hydrocarbon (TPH), benzene, toluene, ethylbenzene, and xylenes (BTEX) were determined. The remediation levels meet with the requirements for minimum target risk levels (cancer risk = 1×10-6 and hazard quotient = 1). The developed risk-based cleanup goals are calculated based on actual land use and exposure pathways rather than on assumed maximum exposure. After the completion of RBCA process, the remediation cost can be significantly reduced. Based on the risk assessment results, source zone remediation, natural attenuation, and long term monitoring were recommended as the remediation strategies of the studied site to reduce risks to human health. Results from this study provide a streamlined process for future risk assessment work at petroleum-hydrocarbon polluted sites in Taiwan. RBCA is a sound and defensible basis for site closure, and it offers a more logical framework for making site closure decisions.

Rational Blinders: Strategic Selection of Risk Models and Bank Capital Regulation

The regulatory use of banks' internal models aims at making capital requirements more accurate and reducing regulatory arbitrage, but may also give banks incentives to choose their risk models strategically. Current policy answers to this problem include the use of risk-weight floors and leverage ratios. I show that banks for which those are binding reduce their credit supply, which drives interest rates up, invites other banks to adopt optimistic models and possibly increases aggregate risk in the banking sector. Instead, the strategic use of risk models can be avoided by imposing penalties on banks with low risk-weights when they suffer abnormal losses or bailing out defaulting banks that truthfully reported high risk measures. If such selective bail-outs are not desirable, second-best capital requirements still rely on internal models, but less than in the first-best.

Selecting Antiplatelet Therapy at the Time of Percutaneous Intervention for an Acute Coronary Syndrome

On average, acute coronary syndrome patients treated with prasugrel experience fewer ischemic complications, but more bleeding, than those receiving clopidogrel. However, heterogeneity in treatment effects can alter the likelihood of benefits and risks of an individual patient. We developed predictive models of the benefits (reduced ischemic events) and risks (increased bleeding) to support targeting prasugrel to those who benefit most from treatment. Using 12 579 patients from Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With Prasugrel-Thrombolysis in Myocardial Infarction 38 (TRITON-TIMI 38), we fit risk models for ischemic events (cardiovascular death, spontaneous myocardial infarction, stroke) and bleeding (TIMI major/minor) over a 14.8-month follow-up and then calculated each patient's predicted risk for major ischemia and bleeding with both prasugrel and clopidogrel. We found substantial heterogeneity of the treatment effect of prasugrel (mean absolute reduction in the ischemia risk with prasugrel=1.5±3.0%, ranging from an 8.4% increased risk to a 31.2% reduction in risk for ischemia compared with clopidogrel). The mean absolute increase in the bleeding risk with prasugrel versus clopidogrel was 1.3±1.4% and ranged from a 7.9% lower risk to an 11.2% higher risk with prasugrel. The ratio of the difference in predicted ischemia risk/difference in predicted bleeding risk between prasugrel and clopidogrel was calculated for each patient to identify the proportion likely to benefit from prasugrel. Considering both ischemia and bleeding risk, a large proportion of TRITON participants (42%) were predicted to experience net benefit with prasugrel, a rate that increased if patients more strongly preferred avoiding ischemic events than bleeding. The expected benefits and risks of prasugrel versus clopidogrel depend highly on patient characteristics. The use of risk models could support individualized thienopyridine selection to maximize the benefits and safety of these drugs.