High Outliers Research Articles

NSQIP-Based Quality Improvement Curriculum for Surgical Residents

General surgery training has historically lacked a standardized approach to resident quality improvement (QI) education aside from traditional morbidity and mortality conference. In 2013, the ACGME formalized QI as a component of residency training. Our residency chose the NSQIP Quality In-Training Initiative (QITI) as the foundation for our QI training. We hypothesized that a focused curriculum based on outcomes would produce change in culture and improve the quality of patient care. Quality improvement curriculum design and implementation were retrospectively reviewed. Institutional NSQIP data pre-, during, and post-curriculum implementation were reviewed for improvement. A QITI project committee designed a 2-year curriculum, with 3 parts: didactics, focused on methods of data collection, QI processes, and techniques; review of current institutional performance, practice, and complication rates; and QI breakout groups tasked with creating "best practice" guidelines addressing common complications in our NSQIP semi-annual reports. Educational presentations were given to the surgical department addressing reduction of cardiac complications, pneumonia, surgical site infections (SSIs), and urinary tract infections (UTIs). Twenty-four residents completed both years of the QITI curriculum. National NSQIP decile ranks improved in known high outlier areas: cardiac complications, ninth to fourth decile; pneumonia, eighth to first decile; SSIs, tenth to second decile; and UTIs, eighth to third decile. Pneumonia and SSI rates demonstrated statistical improvement after curriculum implementation (p < 0.003). Implementing a QITI curriculum with a full resident complement is feasible and can positively affect surgical morbidity and nationally benchmarked performance. Resident QI education is essential to future success in delivering high quality surgical care.

Robust and sparse estimators for linear regression models

Penalized regression estimators are popular tools for the analysis of sparse and high-dimensional models. However, penalized regression estimators defined using an unbounded loss function can be very sensitive to the presence of outlying observations, especially to high leverage outliers. The robust and asymptotic properties of ℓ1-penalized MM-estimators and MM-estimators with an adaptive ℓ1 penalty are studied. For the case of a fixed number of covariates, the asymptotic distribution of the estimators is derived and it is proven that for the case of an adaptive ℓ1 penalty, the resulting estimator can have the oracle property. The advantages of the proposed estimators are demonstrated through an extensive simulation study and the analysis of real data sets. The proofs of the theoretical results are available in the Supplementary material to this article (see Appendix A).

The Impact of Excluding Shock Patients on Hospital and Physician Risk-Adjusted Mortality Rates for Percutaneous Coronary Interventions: The Implications for Public Reporting

ObjectivesThe authors examined the impact of including shock patients in public reporting of percutaneous coronary intervention (PCI) risk-adjusted mortality. BackgroundThere is concern that an unintended consequence of statewide public reporting of medical outcomes is the avoidance of appropriate interventions for high-risk patients. MethodsNew York State’s PCI registry was used to compare hospital and physician risk-adjusted mortality rates and outliers from New York’s public report models with rates and outliers based on statistical models that include refractory shock patients and exclude both refractory shock and other shock patients. ResultsCorrelations between the public report model and each of the other 2 models were above 0.92 for hospital risk-adjusted rates and were 0.99 for all physician risk-adjusted rates (p < 0.0001). There were 11 physicians with lower than expected mortality rates (low outliers) and 41 physicians with higher than expected mortality rates (high outliers) across the 3 time periods in the public report, compared with 10 low outliers and 40 high outliers if all shock patients had been excluded. There was considerable overlap among outliers identified by the 3 models. Findings were similar for hospital outliers. ConclusionsRisk-adjusted hospital and physician mortality rates are highly correlated regardless of whether shock patients are included in public reporting. The numbers of outliers are similar, and outlier changes are minimal, although 10% to 15% of cardiologists who were outliers in either exclusion rule were not outliers in the other one. This information can form a basis for subsequent discussions regarding the exclusion of high-risk patients from public reporting.

Air quality data clustering using EPLS method

Nowadays air quality data can be easily accumulated by sensors around the world. Analysis on air quality data is very useful for society decision. Among five major air pollutants which are calculated for AQI (Air Quality Index), PM2.5 data is the most concerned by the people. PM2.5 data is also cross-impacted with the other factors in the air and which has properties of non-linear non-stationary including high noise level and outlier. Traditional methods cannot solve the problem of PM2.5 data clustering very well because of their inherent characteristics. In this paper, a novel model-based feature extraction method is proposed to address this issue. The EPLS model includes: (1) Mode Decomposition, in which EEMD algorithm is applied to the aggregation dataset; (2) Dimension Reduction, which is carried out for a more significant set of vectors; (3) Least Squares Projection, in which all testing data are projected to the obtained vectors. Synthetic dataset and air quality dataset are applied to different clustering methods and similarity measures. Experimental results demonstrate that EPLS is efficient in dealing with high noise level and outlier air quality clustering problems, and which can also be adapted to various clustering techniques and distance measures.

Hospital Variability in Use of Adjuvant Chemotherapy for Patients with Stage 2 and 3 Colon Cancer

Following oncologic resection, adjuvant chemotherapy is associated with decreased recurrence and improved survival in stage 3 colon cancer. However, there is controversy regarding its use in stage 2 colon cancer with high-risk features (tumor depth T4, poorly differentiated, positive margin, and/or inadequate lymph node retrieval). Consensus guidelines recommend no adjuvant chemotherapy in the absence of these high-risk features (low-risk stage 2). This study aimed to examine hospital characteristics associated with poor risk-adjusted, stage-specific guideline compliance for the use of adjuvant chemotherapy in stage 3 and low-risk stage 2 colon cancer. This was a retrospective study. Stepwise logistic regression was used to identify patient and hospital factors associated with administration of adjuvant chemotherapy. Hierarchical regression models were used to calculate risk- and reliability-adjusted rates of chemotherapy use and observed-to-expected ratios in each hospital's stage 2 low-risk and stage 3 patients. Data were retrieved from the National Cancer Database. Patients selected were adults treated with oncologic resection for stage 2 to 3 colon cancer between 2004 and 2010. The primary outcome measured was receipt of adjuvant chemotherapy. A total of 167,345 patients were identified at 1395 hospitals. The mean overall risk-adjusted adjuvant chemotherapy rate was 65.3% for stage 3 and 15.2% for low-risk stage 2. Analysis of low outlier hospitals for stage 3 colon cancer, where adjuvant chemotherapy was underutilized, demonstrated that 62.8% were low-volume centers and 51.4% were community centers. Of high outlier hospitals for stage 2 low-risk disease, where adjuvant chemotherapy was overutilized, 87.2% were low-volume hospitals and 67.2% were community centers. Selection bias and the inability to compare specific chemotherapy regimens were limitations of this study. Following oncologic resection, administration of adjuvant chemotherapy for low-risk stage 2 and stage 3 disease varies substantially among hospitals in the United States. Outlier hospitals were most likely to be low-volume community centers.

Estimation of flood frequencies from data sets with outliers using mixed distribution functions

ABSTRACTIn this paper the estimation of high return period quantiles of the flood peak and volume in the Kolubara River basin are carried out. Estimation of flood frequencies is carried out on a data set containing high outliers which are identified by the Rosner’s test. Simultaneously, low outliers are determined by the multiple Grubbs–Beck. The next step involved the usage of the mixed distribution functions applied to a data set from three populations: floods with low outliers, normal floods and floods with high outliers. The contribution of the data set with low outliers is neglected, since it should underestimate the flood quantiles with large return periods. Consequently, the best fitted mixed distribution from the applied types (EV1, GEV, P3 and LP3) was determined by using the minimum standard error of fit.

Multieigenvalue Communication

In the most general case, all three components—the discrete eigenvalues, the discrete spectral amplitudes, and the continuous spectrum—of the nonlinear Fourier transform of a signal can be independently modulated. This paper examines information transmission using only the discrete eigenvalues, and presents heuristic designs for multisoliton signal sets with spectral efficiencies greater than 3 b/s/Hz. The first design, called multieigenvalue position encoding, is based on an exhaustive search followed by pruning of the signal set to remove high pulsewidth or high bandwidth outliers. The second design, called trellis encoding, achieves comparable efficiencies to the fist method at much lower complexity. These multisoliton signals do not undergo any pulse broadening, but are significantly limited by bandwidth expansion if the system length is not much smaller than the dispersion length parameter. This limitation suggests that modulating the eigenvalues alone cannot address the problem of nonlinearity in commercial fiber transmission systems, and that our proposed methods are only meaningful when dispersion is very small and dominated by nonlinearity, e.g., close to the zero-dispersion wavelength at 1300 nm.

Colorectal surgical site infection reduction strategies

ObjectivePeaceHealth Sacred Heart Medical Center at RiverBend is a 379 bed hospital; 15,060 surgical procedures were performed in 2014 with 254 being colorectal surgical cases. Using the ACS NSQIP program the hospital was identified as a high outlier for surgical site infection (SSI) in all cases and in colorectal procedures in our July 2012 semiannual report (SAR). MethodsA best practice bundle to reduce SSI's and a colorectal enhanced recovery pathway were concurrently initiated ResultsAfter implementation of these best practice elements, our ACS NSQIP results showed a reduction in our colorectal SSI rate from 17.58% for 2011 to 5.11% (1st decile/low outlier) in the January 2015 SAR. Our SSI rate for all cases were reduced from 4.87% to 1.71% (1st decile/low outlier) during the same time period. A reduction in length of stay for our colorectal surgical patients was also noted. ConclusionsThe ACS NSQIP program was essential to identify colorectal SSI as an area of concern and for providing data to drive quality improvement measures.

Appending Limited Clinical Data to an Administrative Database for Acute Myocardial Infarction Patients: The Impact on the Assessment of Hospital Quality.

Hospitals' risk-standardized mortality rates and outlier status (significantly higher/lower rates) are reported by the Centers for Medicare and Medicaid Services (CMS) for acute myocardial infarction (AMI) patients using Medicare claims data. New York now has AMI claims data with blood pressure and heart rate added. The objective of this study was to see whether the appended database yields different hospital assessments than standard claims data. New York State clinically appended claims data for AMI were used to create 2 different risk models based on CMS methods: 1 with and 1 without the added clinical data. Model discrimination was compared, and differences between the models in hospital outlier status and tertile status were examined. Mean arterial pressure and heart rate were both significant predictors of mortality in the clinically appended model. The C statistic for the model with the clinical variables added was significantly higher (0.803 vs. 0.773, P<0.001). The model without clinical variables identified 10 low outliers and all of them were percutaneous coronary intervention hospitals. When clinical variables were included in the model, only 6 of those 10 hospitals were low outliers, but there were 2 new low outliers. The model without clinical variables had only 3 high outliers, and the model with clinical variables included identified 2 new high outliers. Appending even a small number of clinical data elements to administrative data resulted in a difference in the assessment of hospital mortality outliers for AMI. The strategy of adding limited but important clinical data elements to administrative datasets should be considered when evaluating hospital quality for procedures and other medical conditions.

Adaptive divergence despite strong genetic drift: genomic analysis of the evolutionary mechanisms causing genetic differentiation in the island fox (Urocyon littoralis).

The evolutionary mechanisms generating the tremendous biodiversity of islands have long fascinated evolutionary biologists. Genetic drift and divergent selection are predicted to be strong on islands and both could drive population divergence and speciation. Alternatively, strong genetic drift may preclude adaptation. We conducted a genomic analysis to test the roles of genetic drift and divergent selection in causing genetic differentiation among populations of the island fox (Urocyon littoralis). This species consists of six subspecies, each of which occupies a different California Channel Island. Analysis of 5293 SNP loci generated using Restriction-site Associated DNA (RAD) sequencing found support for genetic drift as the dominant evolutionary mechanism driving population divergence among island fox populations. In particular, populations had exceptionally low genetic variation, small Ne (range=2.1-89.7; median=19.4), and significant genetic signatures of bottlenecks. Moreover, islands with the lowest genetic variation (and, by inference, the strongest historical genetic drift) were most genetically differentiated from mainland grey foxes, and vice versa, indicating genetic drift drives genome-wide divergence. Nonetheless, outlier tests identified 3.6-6.6% of loci as high FST outliers, suggesting that despite strong genetic drift, divergent selection contributes to population divergence. Patterns of similarity among populations based on high FST outliers mirrored patterns based on morphology, providing additional evidence that outliers reflect adaptive divergence. Extremely low genetic variation and small Ne in some island fox populations, particularly on San Nicolas Island, suggest that they may be vulnerable to fixation of deleterious alleles, decreased fitness and reduced adaptive potential.

Multinomial interpoint distances

We explore the properties of the squared Euclidean interpoint distances (IDs) drawn from multinomial distributions. We consider the distances within one sample and across two samples and obtain their means, variances, covariances and distributions. We discuss applications of IDs for testing goodness of fit, the equality of high dimensional multinomial distributions, classification, and outliers detection. A simulation study compares the performance of the \(\chi ^2\) and the likelihood ratio statistics for testing equality of distributions, with methods based on the IDs.

Using NSQIP Data for Quality Improvement: The Cleveland Clinic SSI Experience

Prevention of surgical site infection (SSI) has become one of the primary objectives of national quality improvement programs in U.S. hospitals. This article would discuss the impact of participation in the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) on the outcomes and our experience for SSI reduction. Since 2005, our institution has been an active participant of the ACS-NSQIP, and our SSI rates have been compared with other NSQIP centers; in 2006, we were found to be a high outlier institution for SSI. ACS-NSQIP greatly facilitated our understanding of the reasons for high SSI rates and patient profile that is associated with high risks for SSIs and how we can make improvements. Thus, our department initiated an SSI reduction strategy program called the “CORS SSI Prevention Bundle Project” in 2014. This effort was resulted in a statistically significant decrease in overall SSI rates from 11.8% to 6.5% after colorectal surgery. In conclusion, participation in the ACS-NSQIP offers the opportunity to compare institutional SSI rates with those of other and recognize the facts for improvement. By using this data, a bundled approach improved clinical outcomes in our experience and this was attributed to multifaceted initiatives stemming from multiple team members and comprehensive data utilization.

Global High Dimension Outlier Algorithm for Efficient Clustering and Outlier Detection

Global High Dimension Outlier Algorithm for Efficient Clustering and Outlier Detection

Use of a quantile regression based echo state network ensemble for construction of prediction Intervals of gas flow in a blast furnace

The usual huge fluctuations in the blast furnace gas (BFG) generation make the scheduling of the gas system become a difficult problem. Considering that there are high level noises and outliers mixed in original industrial data, a quantile regression-based echo state network ensemble (QR-ESNE) is modeled to construct the prediction intervals (PIs) of the BFG generation. In the process of network training, a linear regression model of the output matrix is reported by the proposed quantile regression to improve the generalization ability. Then, in view of the practical demands on reliability and further improving the prediction accuracy, a bootstrap strategy based on QR-ESN is designed to construct the confidence intervals and the prediction ones via combining with the regression models of various quantiles. To verify the performance of the proposed method, the practical data coming from a steel plant are employed, and the results indicate that the proposed method exhibits high accuracy and reliability for the industrial data. Furthermore, an application software system based on the proposed method is developed and applied to the practice of this plant.

Advanced weight graph transformation matching algorithm

An efficient and accurate point matching algorithm named advanced weight graph transformation matching (AWGTM) is proposed in this study. Instead of relying only on the elimination of dubious matches, the method iteratively reserve correspondences which have a small angular distance between two nearest‐neighbour graphs. The proposed algorithm is compared against weight graph transformation matching (WGTM) and graph transformation matching (GTM). Experimental results demonstrate the superior performance in eliminating outliers and reserving inliers of AWGTM algorithm under various conditions for images, such as duplication of patterns and non‐rigid deformation of objects. An execution time comparison is also presented, where AWGTM shows the best results for high outlier rates.

Robust sampling-sourced numerical retrieval algorithm for optical energy loss function based on log–log mesh optimization and local monotonicity preserving Steffen spline

We introduce a new technique of interpolation of the energy-loss function (ELF) in solids sampled by empirical optical spectra. Finding appropriate interpolation methods for ELFs poses several challenges. The sampled ELFs are usually very heterogeneous, can originate from various sources thus so called “data gaps” can appear, and significant discontinuities and multiple high outliers can be present. As a result an interpolation based on those data may not perform well at predicting reasonable physical results. Reliable interpolation tools, suitable for ELF applications, should therefore satisfy several important demands: accuracy and predictive power, robustness and computational efficiency, and ease of use. We examined the effect on the fitting quality due to different interpolation schemes with emphasis on ELF mesh optimization procedures and we argue that the optimal fitting should be based on preliminary log–log scaling data transforms by which the non-uniformity of sampled data distribution may be considerably reduced. The transformed data are then interpolated by local monotonicity preserving Steffen spline. The result is a piece-wise smooth fitting curve with continuous first-order derivatives that passes through all data points without spurious oscillations. Local extrema can occur only at grid points where they are given by the data, but not in between two adjacent grid points. It is found that proposed technique gives the most accurate results and also that its computational time is short. Thus, it is feasible using this simple method to address practical problems associated with interaction between a bulk material and a moving electron. A compact C++ implementation of our algorithm is also presented.

Reduction of Venous Thromboembolism in Surgical Patients Using a Mandatory Risk-Scoring System: 5-Year Follow-Up of an American College of Surgeons National Surgical Quality Improvement Program.

Sheikh Khalifa Medical City's (SKMC) surgery institute was identified as a high outlier in the incidence of venous thromboembolism (VTE; deep vein thrombosis [DVT] and pulmonary embolism [PE]) based on the semiannual report of the American College of Surgeon's National Surgical Quality Improvement Program (ACS NSQIP) in June 2010. To report our rates of VTE at SKMC, the results, and 5-year follow-up after an ACS NSQIP quality improvement program. A multidisciplinary VTE task force was established in June 2010. We instituted a compulsory risk assessment for VTE and utilized the ACS NSQIP best practice guidelines to review cases of VTE. We prospectively evaluated the observed/expected (O/E) ratio for DVT/PE after implementing the action plan. The O/E ratio for PE/DVT in general and general/vascular (GV) surgery was 6.00 and 4.86 in June 2010. Our compliance with ordering antithrombotic prophylactic measures was as low and it improved to 100% and our O/E ratio decreased to 1.18 and 1.5 in July 2011 and stabilized for the next 4 years. Currently, our compliance with ordering antithrombotic prophylactic measures is 100%, and our last 2 O/E ratio for DVT/PE are 0.74 and 0.75 in GV surgery and 0.82 and 0.78 in the entire surgery institute, respectively, and we are considered an exemplary site of the ACS NSQIP in GV surgery. A compulsory risk assessment for VTE has led to an overall improvement in DVT/PE rates in the surgery institute and for GV surgery to become an exemplary site for the ACS NSQIP.

Exceedance Frequency Analysis of Contaminants in Streams Under Dry-Weather Conditions in Denton, Texas.

Percentages of dry-weather stream samples exceeding water quality criteria for ten parameters were compiled for mixed land use watersheds in north-central Texas. Most problematic were total suspended solids (TSS), total dissolved solids (TDS), ammonia, nitrate, phosphorus and copper. Nutrients had much higher exceedance frequency at a sampling station impacted by wastewater discharge. Whereas, TSS and TDS exceedance frequency was highest in predominantly agricultural and rangeland watersheds, and urbanized watersheds respectively. Total dissolved solids was most often exceeded in urbanized watersheds. For several parameters, especially TDS, TSS, ammonia and copper, median concentrations were below water quality thresholds in most watersheds, but exceedance frequency was high. For example, median TSS was less than its threshold in every watershed, but exceedance frequency was higher than 10 % in four of five watersheds - and nearly 43 % in one watershed. This pattern reflects the skewed nature of water quality data; often times, many observations cluster around the lowest values, causing the median to be relatively low, but several (high) outliers form the right-hand tail of the distribution. Results of this study indicate a need to examine exceedance frequency in addition to traditional descriptive measures to better understand dry-weather stream quality in watersheds.

Predictors of High Profit and High Deficit Outliers under SwissDRG of a Tertiary Care Center.

PrinciplesCase weights of Diagnosis Related Groups (DRGs) are determined by the average cost of cases from a previous billing period. However, a significant amount of cases are largely over- or underfunded. We therefore decided to analyze earning outliers of our hospital as to search for predictors enabling a better grouping under SwissDRG.Methods28,893 inpatient cases without additional private insurance discharged from our hospital in 2012 were included in our analysis. Outliers were defined by the interquartile range method. Predictors for deficit and profit outliers were determined with logistic regressions. Predictors were shortlisted with the LASSO regularized logistic regression method and compared to results of Random forest analysis. 10 of these parameters were selected for quantile regression analysis as to quantify their impact on earnings.ResultsPsychiatric diagnosis and admission as an emergency case were significant predictors for higher deficit with negative regression coefficients for all analyzed quantiles (p<0.001). Admission from an external health care provider was a significant predictor for a higher deficit in all but the 90% quantile (p<0.001 for Q10, Q20, Q50, Q80 and p = 0.0017 for Q90). Burns predicted higher earnings for cases which were favorably remunerated (p<0.001 for the 90% quantile). Osteoporosis predicted a higher deficit in the most underfunded cases, but did not predict differences in earnings for balanced or profitable cases (Q10 and Q20: p<0.00, Q50: p = 0.10, Q80: p = 0.88 and Q90: p = 0.52). ICU stay, mechanical and patient clinical complexity level score (PCCL) predicted higher losses at the 10% quantile but also higher profits at the 90% quantile (p<0.001).ConclusionWe suggest considering psychiatric diagnosis, admission as an emergencay case and admission from an external health care provider as DRG split criteria as they predict large, consistent and significant losses.

The relationship between processes and outcomes for injured older adults: a study of a statewide trauma system.

Age is a risk factor for death, adverse outcomes, and health care use following trauma. The American College of Surgeons' Trauma Quality Improvement Program (TQIP) has published "best practices" of geriatric trauma care; adoption of these guidelines is unknown. We sought to determine which evidence-based geriatric protocols, including TQIP guidelines, were correlated with decreased mortality in Pennsylvania's trauma centers. PA's level I and II trauma centers self-reported adoption of geriatric protocols. Survey data were merged with risk-adjusted mortality data for patients ≥65 from a statewide database, the Pennsylvania Trauma Systems Foundation (PTSF), to compare mortality outlier status and processes of care. Exposures of interest were center-specific processes of care; outcome of interest was PTSF mortality outlier status. 26 of 27 eligible trauma centers participated. There was wide variation in care processes. Four trauma centers were low outliers; three centers were high outliers for risk-adjusted mortality rates in adults ≥65. Results remained consistent when accounting for center volume. The only process associated with mortality outlier status was age-specific solid organ injury protocols (p=0.04). There was no cumulative effect of multiple evidence-based processes on mortality rate (p=0.50). We did not see a link between adoption of geriatric best-practices trauma guidelines and reduced mortality at PA trauma centers. The increased susceptibility of elderly to adverse consequences of injury, combined with the rapid growth rate of this demographic, emphasizes the importance of identifying interventions tailored to this population. III. Descriptive.