Fast construction of interpretable whole-brain decoders.

Researchers often seek to decode mental states from brain activity measured with functional MRI. Rigorous decoding requires the use of formal neural prediction models, which are likely to be the most accurate if they use the whole brain. However, the computational burden and lack of interpretability of off-the-shelf statistical methods can make whole-brain decoding challenging. Here, we propose a method to build whole-brain neural decoders that are both interpretable and computationally efficient. We extend the partial least squares algorithm to build a regularized model with variable selection that offers a unique "fit once, tune later" approach: users need to fit the model only once and can choose the best tuning parameters post hoc. We show in real data that our method scales well with increasing data size and yields interpretable predictors. The algorithm is publicly available in multiple languages in the hope that interpretable whole-brain predictors can be implemented more widely in neuroimaging research.

Flexible Utility Function Approximation via Cubic Bezier Splines.

In intertemporal and risky choice decisions, parametric utility models are widely used for predicting choice and measuring individuals' impulsivity and risk aversion. However, parametric utility models cannot describe data deviating from their assumed functional form. We propose a novel method using cubic Bezier splines (CBS) to flexibly model smooth and monotonic utility functions that can be fit to any dataset. CBS shows higher descriptive and predictive accuracy over extant parametric models and can identify common yet novel patterns of behavior that are inconsistent with extant parametric models. Furthermore, CBS provides measures of impulsivity and risk aversion that do not depend on parametric model assumptions.

From Karen to Katie: using baby names to understand cultural evolution.

How do psychological processes shape how culture evolves? We investigated how a cultural item's popularity is shaped by the recent popularity of other items with features in common. Specifically, using more than 100 years of first-names data, we examined how a name's popularity is influenced by the popularity of that name's component phonemes in other names in the previous year. Building on mere-exposure research, we found that names are more likely to become popular when similar names have been popular recently. These effects are nonlinear, however, and overpopularity hurts adoption. In addition, these effects vary with phoneme position. We demonstrate the causal impact of similarity on cultural success in a natural experiment using hurricane names. An exogenous shock to a phoneme's frequency, due to the presence of the phoneme in the names of prominent hurricanes, boosts the popularity of names that share that phoneme. Taken together, our results suggest how the similarity between cultural items affects how popular they become and how culture evolves more broadly.

Using Testlet Response Theory to analyze data from a survey of attitude change among breast cancer survivors.

In this paper we examine alternative measurement models for fitting data from health surveys. We show why a testlet-based latent trait model that includes covariate information, embedded within a fully Bayesian framework, can allow multiple simultaneous inferences and aid interpretation. We illustrate our approach with a survey of breast cancer survivors that reveals how the attitudes of those patients change after diagnosis toward a focus on appreciating the here-and-now, and away from consideration of longer-term goals. Using the covariate information, we also show the extent to which individual-level variables such as race, age and Tamoxifen treatment are related to a patient's change in attitude.The major contribution of this research is to demonstrate the use of a hierarchical Bayesian IRT model with covariates in this application area; hence a novel case study, and one that is certainly closely aligned with but distinct from the educational testing applications that have made IRT the dominant test scoring model.

Public reporting on hospital process improvements is linked to better patient outcomes.

The Centers for Medicare and Medicaid Services publicly reports so-called process performance at all U.S. hospitals, such as whether certain recommended treatments are given to specific types of patients. We examined whether hospital performance on key process indicators improved during the three years since this reporting began. We also studied whether or not these changes improved patient outcomes or yielded other quality improvements, such as reduced hospital readmission rates. We found that, from 2004 to 2006, hospital process performance improved and was associated with better patient and quality outcomes. Most notably, for acute myocardial infarction, performance improvements were associated with declines in mortality rates, lengths-of-stay, and readmission rates. Although we cannot conclude that public reporting caused the improvement in processes or outcomes, these results are encouraging, since improving process performance may improve quality more broadly.

Analyzing risk response dynamics on the web: the case of Hurricane Katrina.

Using a unique data set that documented the hourly web-surfing behavior of over 140,000 Internet users in five southeastern states in August 2005, we explore the dynamics of information gathering as Hurricane Katrina developed and then hit South Florida and the Northern Gulf Coast. Using both elementary statistical methods and advanced techniques from functional data analysis,((1)) we examine both how storm events (such as the posting of warnings) affected traffic to weather-related websites, and how this traffic varied across locations and by characteristics of the web user. A general finding is that spatial-temporal variation in weather-site web traffic generally tracked the timing and scale of the storm threat experienced by a given area. There was, however, considerable variation in this responsiveness. Residents in Florida counties that had been most directly affected by Hurricane Dennis just a month earlier, for example, displayed more active visitation rates than those who had been less affected. We also find evidence of a gender effect where male users displayed a disproportionately larger rate of visitation to weather sites given the onset of storm warnings than females. The implications of this work for the broader study of behavioral risk response dynamics during hazards are explored.

Hospital performance measures and quality of care.

Increasingly, quality improvement initiatives emphasize public reporting of hospital performance measures, to encourage providers to improve, to help consumers pick providers, and to determine provider payments. Although these measures are based on compliance with well-established processes of care, it is unknown whether quality measured in this way is correlated with, or predictive of, clinical outcomes. This Issue Brief summarizes studies that examine and quantify the relationship between frequently used measures of hospital performance and hospital mortality.

Does hospital performance on process measures directly measure high quality care or is it a marker of unmeasured care?

OBJECTIVE: Quality measures may be associated with improved outcomes for two reasons. First, measured activities may directly improve care. Second, success on these measures may be a marker for other unmeasured aspects of high quality care. Our objective is to test the contribution of both possible effects. DATA SOURCES: 2004 Medicare data on hospital performance from Hospital Compare and risk-adjusted mortality rates from Medicare Part A claims. STUDY DESIGN: We studied 3,657 acute care U.S. hospitals and compared observed differences in condition-specific hospital mortality rates based on hospital performance with expected differences in mortality from the clinical studies underlying the measures. PRINCIPAL FINDINGS: Differences in observed mortality rates across U.S. hospitals are larger than what would be expected if these differences were due only to the direct effects of delivering measured care. CONCLUSIONS: Performance measures reflect care processes that both improve care directly and are also markers of elements of health care quality that are otherwise unmeasured. This finding suggests that process measures capture important information about care that is not directly measured, and that these unmeasured effects are in general larger than the measured effects.

Assessment of medical devices: how to conduct comparative technology evaluations of product performance.

OBJECTIVES: U.S. expenditures on medical devices (US dollars 70 billion in 2003) are one of the fastest growing components of hospital costs. Physicians' selection of medical devices lacks an evidence base on the comparative clinical effectiveness of these products. Comparative studies (e.g., vendor 1 versus vendor 2, technology A versus technology B) are increasingly promoted in the public sector as a means of cost containment, value-based purchasing, and quality improvement. This study illustrates how hospitals and physicians can conduct comparative technology assessments of product performance. METHODS: Surgeons evaluated comparable medical devices manufactured by eight different vendors in standardized surgical procedures. Devices included sutures and endomechanical products, which account for US dollars 2.5 billion of total device spending. Evaluations covered multiple performance dimensions, including ergonomics, functionality, clinical acceptability, and vendor preference. RESULTS: One vendor's products garnered consistently high ratings from surgeons, while two other vendors garnered consistently low ratings. Differences in ratings were statistically significant and persist when controlling for physician background characteristics and prior experience. Study results were used by a large hospital group purchasing organization to select which vendors to contract with for these products. CONCLUSIONS: Comparative technology evaluations assist physicians and hospitals in making cost-effective purchases of devices. These evaluations provide robust information on the performance of products routinely used by clinicians. Such evaluations can be carefully designed to have scientific rigor and clinical credibility.

Surgeon evaluation of suture and endo-mechanical products.

BACKGROUND: Surgeons select medical instruments without comparative performance data. This analysis seeks to determine if suture and endo-mechanical products made by different vendors have equivalent performance profiles or are clearly distinguished by physicians on different dimensions. MATERIALS AND METHODS: A sample of 45 surgeons evaluated eight vendors of five categories of suture and endo-mechanical products: clip appliers, staplers, trocars, needles and sutures, and endoscopic specimen retrieval devices. Surgeons rated each vendor's products in each category on multiple performance dimensions at six animal laboratories at academic medical centers around the U.S. between April and September 2005. Performance dimensions included the product's clinical acceptability, ergonomics, functionality, overall performance, and relative rank-order preference. RESULTS: Physician evaluations of vendor performance vary widely. Vendors rated as clinically equivalent on a given product received different performance ratings by physicians. Ethicon's products (Somerville, NJ) were rated consistently high by physicians across product categories. This suggests the presence of some superior brand performance. Nevertheless, within some categories, there were alternative vendors (U.S. Surgical [Mansfield, MA], Applied Medical [Rancho Margarita, CA]) whose products are rated similar to the brand leader. This suggests there are often multiple vendors from which to choose. There was also evidence of idiosyncratic physician preference, especially due to the physician's gender, height, and glove size. CONCLUSIONS: Suture and endo-mechanical products made by different vendors do not have equivalent performance profiles. Specific brand seems to be the most important determinant of physician evaluations of the different vendors' products. These results suggest the value and importance of conducting head-to-head comparisons of multiple vendors of the same product.

Relationship between Medicare's hospital compare performance measures and mortality rates.

CONTEXT: In response to concerns about the quality of care in US hospitals, the Centers for Medicare & Medicaid Services began measuring hospital performance and reporting this performance on their Web site, Hospital Compare. It is unknown whether these process performance measures are related to hospital-level outcomes. OBJECTIVE: To determine whether quality measured with the process measures used in Hospital Compare are correlated with and predictive of hospitals' risk-adjusted mortality rates. DESIGN, SETTING, AND PARTICIPANTS: Cross-sectional study of hospital care between January 1 and December 31, 2004, for acute myocardial infarction, heart failure, and pneumonia at acute care hospitals in the United States included on the Hospital Compare Web site. Ten process performance measures included in Hospital Compare were compared with hospital risk-adjusted mortality rates, which were measured using Medicare Part A claims data. MAIN OUTCOME MEASURES: Condition-specific inpatient, 30-day, and 1-year risk-adjusted mortality rates. RESULTS: A total of 3657 acute care hospitals were included in the study based on their performance as reported in Hospital Compare. Across all acute myocardial infarction performance measures, the absolute reduction in risk-adjusted mortality rates between hospitals performing in the 25th percentile vs those performing in the 75th percentile was 0.005 for inpatient mortality, 0.006 for 30-day mortality, and 0.012 for 1-year mortality (P<.001 for each comparison). For the heart failure performance measures, the absolute mortality reduction was smaller, ranging from 0.001 for inpatient mortality (P = .03) to 0.002 for 1-year mortality (P = .08). For the pneumonia performance measures, the absolute reduction in mortality ranged from 0.001 for 30-day mortality (P = .05) to 0.005 for inpatient mortality (P<.001). Differences in mortality rates for hospitals performing in the 75th percentile on all measures within a condition vs those performing lower than the 25th percentile on all reported measures for acute myocardial infarction ranged between 0.008 (P = .06) and 0.018 (P = .008). For pneumonia, the effects ranged between 0.003 (P = .09) and 0.014 (P<.001); for heart failure, the effects ranged between -0.013 (P = .06) and -0.038 (P = .45). CONCLUSIONS: Hospital performance measures predict small differences in hospital risk-adjusted mortality rates. Efforts should be made to develop performance measures that are tightly linked to patient outcomes.

Length of stay, conditional length of stay, and prolonged stay in pediatric asthma.

OBJECTIVE: To understand differences in length of stay for asthma patients between New York State and Pennsylvania across children's and general hospitals in order to better guide policy. DATA SOURCES/STUDY SETTING: All pediatric admissions for asthma in the states of Pennsylvania and New York using claims data obtained from each state for the years 1996-1998, n = 38,310. STUDY DESIGN: A retrospective cohort design to model length of stay (LOS), the probability of prolonged stay, conditional length of stay (CLOS or the LOS after stay is prolonged), and the probability of readmission, controlling for patient factors, state, location and hospital type. ANALYTIC METHODS: Logit models were used to estimate the probability of prolonged stay and readmission. The LOS and the CLOS were estimated with Cox regression. Model variables included comorbidities, income, race, distance from hospital, and insurance type. Prolonged stay was based on a Hollander-Proschan "New-Worse-Than-Used" test, corresponding to a three-day stay. PRINCIPAL FINDINGS: The LOS was longer in New York than Pennsylvania, and the probabilities of prolonged stay and readmission were much higher in New York than Pennsylvania. However, once an admission was prolonged, there were no differences in CLOS between states (when readmissions were not added to the LOS calculation). In both states, children's hospitals and general hospitals had similar adjusted LOS. CONCLUSIONS: Management of asthma appears more efficient in Pennsylvania than New York: Less severe patients are discharged faster in Pennsylvania than New York; once discharged, patients are less likely to be readmitted in Pennsylvania than New York. However, once a stay is prolonged, there is little difference between New York and Pennsylvania, suggesting medical care for severely ill patients is similar across states. Differences between children's and general hospitals were small as compared to differences between states. We conclude that policy initiatives in New York, and other states, should focus their efforts on improving the care provided to less severe patients in order to help reduce overall length of stay.