Geographic Variation in Postoperative Imaging for Low-Risk Breast Cancer.

Background: The objective of this study was to examine the presence and magnitude of US geographic variation in use rates of both recommended and high-cost imaging in young patients with early-stage breast cancer during the 18 month period after surgical treatment of their primary tumor. Methods: Using the Truven Health MarketScan Commercial Database, a descriptive analysis was conducted of geographic variation in annual rates of dedicated breast imaging and high-cost body imaging of 36,045 women aged 18 to 64 years treated with surgery for invasive unilateral breast cancer between 2010 and 2012. Multivariate hierarchical analysis examined the relationship between likelihood of imaging and patient characteristics, with metropolitan statistical area (MSA) serving as a random effect. Patient characteristics included age group, BRCA1/2 carrier status, family history of breast cancer, combination of breast surgery type and radiation therapy, drug therapy, and payer type. All MSAs in the United States were included, with areas outside MSAs within a given state aggregated into a single area for analytic purposes. Results: Descriptive analysis of rates of imaging use and intensity within MSA regions revealed wide geographic variation, irrespective of treatment cohort or age group. Increased probability of recommended postoperative dedicated breast imaging was primarily associated with age and treatment including both surgery and radiation therapy, followed by MSA region (odds ratio, 1.42). Increased probability of PET use-a high-cost imaging modality for which postoperative routine use is not recommended in the absence of specific clinical findings-was primarily associated with surgery type followed by MSA region (odds ratio, 1.82). Conclusions: In patients with breast cancer treated for low-risk disease, geography has effects on the rates of posttreatment imaging, suggesting that some patients are not receiving beneficial dedicated breast imaging, and high-cost nonbreast imaging may not be targeted to those groups most likely to benefit.

Validation of Prediction Models for Critical Care Outcomes Using Natural Language Processing of Electronic Health Record Data.

Importance: Accurate prediction of outcomes among patients in intensive care units (ICUs) is important for clinical research and monitoring care quality. Most existing prediction models do not take full advantage of the electronic health record, using only the single worst value of laboratory tests and vital signs and largely ignoring information present in free-text notes. Whether capturing more of the available data and applying machine learning and natural language processing (NLP) can improve and automate the prediction of outcomes among patients in the ICU remains unknown. Objectives: To evaluate the change in power for a mortality prediction model among patients in the ICU achieved by incorporating measures of clinical trajectory together with NLP of clinical text and to assess the generalizability of this approach. Design, Setting, and Participants: This retrospective cohort study included 101 196 patients with a first-time admission to the ICU and a length of stay of at least 4 hours. Twenty ICUs at 2 academic medical centers (University of California, San Francisco [UCSF], and Beth Israel Deaconess Medical Center [BIDMC], Boston, Massachusetts) and 1 community hospital (Mills-Peninsula Medical Center [MPMC], Burlingame, California) contributed data from January 1, 2001, through June 1, 2017. Data were analyzed from July 1, 2017, through August 1, 2018. Main Outcomes and Measures: In-hospital mortality and model discrimination as assessed by the area under the receiver operating characteristic curve (AUC) and model calibration as assessed by the modified Hosmer-Lemeshow statistic. Results: Among 101 196 patients included in the analysis, 51.3% (n = 51 899) were male, with a mean (SD) age of 61.3 (17.1) years; their in-hospital mortality rate was 10.4% (n = 10 505). A baseline model using only the highest and lowest observed values for each laboratory test result or vital sign achieved a cross-validated AUC of 0.831 (95% CI, 0.830-0.832). In contrast, that model augmented with measures of clinical trajectory achieved an AUC of 0.899 (95% CI, 0.896-0.902; P < .001 for AUC difference). Further augmenting this model with NLP-derived terms associated with mortality further increased the AUC to 0.922 (95% CI, 0.916-0.924; P < .001). These NLP-derived terms were associated with improved model performance even when applied across sites (AUC difference for UCSF: 0.077 to 0.021; AUC difference for MPMC: 0.071 to 0.051; AUC difference for BIDMC: 0.035 to 0.043; P < .001) when augmenting with NLP at each site. Conclusions and Relevance: Intensive care unit mortality prediction models incorporating measures of clinical trajectory and NLP-derived terms yielded excellent predictive performance and generalized well in this sample of hospitals. The role of these automated algorithms, particularly those using unstructured data from notes and other sources, in clinical research and quality improvement seems to merit additional investigation.

Multidisciplinary Critical Care and Intensivist Staffing: Results of a Statewide Survey and Association With Mortality.

PURPOSE: The role of multidisciplinary teams in improving the care of intensive care unit (ICU) patients is not well defined, and it is unknown whether the use of such teams helps to explain prior research suggesting improved mortality with intensivist staffing. We sought to investigate the association between multidisciplinary team care and survival of medical and surgical patients in nonspecialty ICUs. MATERIALS AND METHODS: We conducted a community-based, retrospective cohort study of data from 60 330 patients in 181 hospitals participating in a statewide public reporting initiative, the California Hospital Assessment and Reporting Taskforce (CHART). Patient-level data were linked with ICU organizational data collected from a survey of CHART hospital ICUs between December 2010 and June 2011. Clustered logistic regression was used to evaluate the independent effect of multidisciplinary care on the in-hospital mortality of medical and surgical ICU patients. Interactions between multidisciplinary care and intensity of physician staffing were examined to explore whether team care accounted for differences in patient outcomes. RESULTS: After adjustment for patient characteristics and interactions, there was no association between team care and mortality for medical patients. Among surgical patients, multidisciplinary care was associated with a survival benefit (odds ratio 0.79; 95% confidence interval (CI), 0.62-1.00; P = .05). When stratifying by intensity of physician staffing, although the lowest odds of death were observed for surgical patients cared for in ICUs with multidisciplinary teams and high-intensity staffing (odds ratio, 0.77; 95% CI, 0.55-1.09; P = .15), followed by ICUs with multidisciplinary teams and low-intensity staffing (odds ratio 0.84, 95% CI 0.65-1.09, p = 0.19), these differences were not statistically significant. CONCLUSIONS: Our results suggest that multidisciplinary team care may improve outcomes for critically ill surgical patients. However, no relationship was observed between intensity of physician staffing and mortality.

Revisit rates and associated costs after an emergency department encounter: a multistate analysis.

BACKGROUND: Return visits to the emergency department (ED) or hospital after an index ED visit strain the health system, but information about rates and determinants of revisits is limited. OBJECTIVE: To describe revisit rates, variation in revisit rates by diagnosis and state, and associated costs. DESIGN: Observational study using the Healthcare Cost and Utilization Project databases. SETTING: 6 U.S. states. PATIENTS: Adults with ED visits between 2006 and 2010. MEASUREMENTS: Revisit rates and costs. RESULTS: Within 3 days of an index ED visit, 8.2% of patients had a revisit; 32% of those revisits occurred at a different institution. Revisit rates varied by diagnosis, with skin infections having the highest rate (23.1% [95% CI, 22.3% to 23.9%]). Revisit rates also varied by state. For skin infections, Florida had higher risk-adjusted revisit rates (24.8% [CI, 23.5% to 26.2%]) than Nebraska (10.6% [CI, 9.2% to 12.1%]). In Florida, the only state with complete cost data, total revisit costs for the 19.8% of patients with a revisit within 30 days were 118% of total index ED visit costs for all patients (including those with and without a revisit). LIMITATION: Whether a revisit reflects inadequate access to primary care, a planned revisit, the patient's nonadherence to ED recommendations, or poor-quality care at the initial ED visit remains unknown. CONCLUSION: Revisits after an index ED encounter are more frequent than previously reported, in part because many occur outside the index institution. Among ED patients in Florida, more resources are spent on revisits than on index ED visits. PRIMARY FUNDING SOURCE: Agency for Healthcare Research and Quality.

Thirty-day hospital readmissions in systemic lupus erythematosus: predictors and hospital- and state-level variation.

OBJECTIVE: Systemic lupus erythematosus (SLE) has one of the highest hospital readmission rates among chronic conditions. This study was undertaken to identify patient-level, hospital-level, and geographic predictors of 30-day hospital readmissions associated with SLE. METHODS: Using hospital discharge databases from 5 geographically dispersed states, we studied all-cause readmission of SLE patients between 2008 and 2009. We evaluated each hospitalization as a possible index event leading up to a readmission, our primary outcome. We accounted for clustering of hospitalizations within patients and within hospitals and adjusted for hospital case mix. Using multilevel mixed-effects logistic regression, we examined factors associated with 30-day readmission and calculated risk-standardized hospital-level and state-level readmission rates. RESULTS: We examined 55,936 hospitalizations among 31,903 patients with SLE. Of these hospitalizations, 9,244 (16.5%) resulted in readmission within 30 days. In adjusted analyses, age was inversely related to risk of readmission. African American and Hispanic patients were more likely to be readmitted than white patients, as were those with Medicare or Medicaid insurance (versus private insurance). Several clinical characteristics of lupus, including nephritis, serositis, and thrombocytopenia, were associated with readmission. Readmission rates varied significantly between hospitals after accounting for patient-level clustering and hospital case mix. We also found geographic variation, with risk-adjusted readmission rates lower in New York and higher in Florida as compared to California. CONCLUSION: We found that ~1 in 6 hospitalized patients with SLE were readmitted within 30 days of discharge, with higher rates among historically underserved populations. Significant geographic and hospital-level variation in risk-adjusted readmission rates suggests potential for quality improvement.

N-gram support vector machines for scalable procedure and diagnosis classification, with applications to clinical free text data from the intensive care unit.

BACKGROUND: Existing risk adjustment models for intensive care unit (ICU) outcomes rely on manual abstraction of patient-level predictors from medical charts. Developing an automated method for abstracting these data from free text might reduce cost and data collection times. OBJECTIVE: To develop a support vector machine (SVM) classifier capable of identifying a range of procedures and diagnoses in ICU clinical notes for use in risk adjustment. MATERIALS AND METHODS: We selected notes from 2001-2008 for 4191 neonatal ICU (NICU) and 2198 adult ICU patients from the MIMIC-II database from the Beth Israel Deaconess Medical Center. Using these notes, we developed an implementation of the SVM classifier to identify procedures (mechanical ventilation and phototherapy in NICU notes) and diagnoses (jaundice in NICU and intracranial hemorrhage (ICH) in adult ICU). On the jaundice classification task, we also compared classifier performance using n-gram features to unigrams with application of a negation algorithm (NegEx). RESULTS: Our classifier accurately identified mechanical ventilation (accuracy=0.982, F1=0.954) and phototherapy use (accuracy=0.940, F1=0.912), as well as jaundice (accuracy=0.898, F1=0.884) and ICH diagnoses (accuracy=0.938, F1=0.943). Including bigram features improved performance on the jaundice (accuracy=0.898 vs 0.865) and ICH (0.938 vs 0.927) tasks, and outperformed NegEx-derived unigram features (accuracy=0.898 vs 0.863) on the jaundice task. DISCUSSION: Overall, a classifier using n-gram support vectors displayed excellent performance characteristics. The classifier generalizes to diverse patient populations, diagnoses, and procedures. CONCLUSIONS: SVM-based classifiers can accurately identify procedure status and diagnoses among ICU patients, and including n-gram features improves performance, compared to existing methods.

The effect of race and ethnicity on outcomes among patients in the intensive care unit: a comprehensive study involving socioeconomic status and resuscitation preferences.

OBJECTIVE: We sought to determine whether race or ethnicity is independently associated with mortality or intensive care unit length of stay among critically ill patients after accounting for patients' clinical and demographic characteristics including socioeconomic status and resuscitation preferences. DESIGN: Historical cohort study of patients hospitalized in intensive care units. SETTING: Adult intensive care units in 35 California hospitals during the years 2001-2004. PATIENTS: A total of 9,518 intensive care unit patients (6,334 white, 655 black, 1,917 Hispanic, and 612 Asian/Pacific Islander patients). MEASUREMENTS AND MAIN RESULTS: The primary outcome was risk-adjusted mortality and a secondary outcome was risk-adjusted intensive care unit length of stay. Crude hospital mortality was 15.9% among the entire cohort. Asian patients had the highest crude hospital mortality at 18.6% and black patients had the lowest at 15.0%. After adjusting for age and gender, Hispanic and Asian patients had a higher risk of death compared to white patients, but these differences were not significant after additional adjustment for severity of illness. Black patients had more acute physiologic derangements at intensive care unit admission and longer unadjusted intensive care unit lengths of stay. Intensive care unit length of stay was not significantly different among racial/ethnic groups after adjustment for demographic, clinical, and socioeconomic factors and do-not-resuscitate status. In an analysis restricted only to those who died, decedent black patients averaged 1.1 additional days in the intensive care unit (95% confidence interval, 0.26-2.6) compared to white patients who died, although this was not statistically significant. CONCLUSIONS: Hospital mortality and intensive care unit length of stay did not differ by race or ethnicity among this diverse cohort of critically ill patients after adjustment for severity of illness, resuscitation status, socioeconomic status, insurance status, and admission type. Black patients had more acute physiologic derangements at intensive care unit admission and were less likely to have a do-not-resuscitate order. These results suggest that among intensive care unit patients, there are no racial or ethnic differences in mortality within individual hospitals. If disparities in intensive care unit care exist, they may be explained by differences in the quality of care provided by hospitals that serve high proportions of minority patients.

Predictors of early postdischarge mortality in critically ill patients: a retrospective cohort study from the California Intensive Care Outcomes project.

PURPOSE: Existing intensive care unit (ICU) mortality measurement systems address in-hospital mortality only. However, early postdischarge mortality contributes significantly to overall 30-day mortality. Factors associated with early postdischarge mortality are unknown. METHODS: We performed a retrospective study of 8484 ICU patients. Our primary outcome was early postdischarge mortality: death after hospital discharge and 30 days or less from ICU admission. Cox regression models assessed the association between patient, hospital, and utilization factors and the primary outcome. RESULTS: In multivariate analyses, the hazard for early postdischarge mortality increased with rising severity of illness and decreased with full-code status (hazard ratio [HR], 0.33; 95% confidence interval [CI], 0.21-0.49). Compared with discharges home, early postdischarge mortality was highest for acute care transfers (HR, 3.18; 95% CI, 2.45-4.12). Finally, patients with very short ICU length of stay (<1 day) had greater early postdischarge mortality (HR, 1.86; 95% CI; 1.32-2.61) than those with longest stays (≥7 days). CONCLUSIONS: Early postdischarge mortality is associated with patient preferences (full-code status) and decisions regarding timing and location of discharge. These findings have important implications for anyone attempting to measure or improve ICU performance and who rely on in-hospital mortality measures to do so.

Relationship between discharge practices and intensive care unit in-hospital mortality performance: evidence of a discharge bias.

CONTEXT: Current intensive care unit performance measures include in-hospital mortality after intensive care unit admission. This measure does not account for deaths occurring after transfer to another hospital or soon after discharge and therefore, may be biased. OBJECTIVE: Determine how transfer rates to other acute care hospitals and early post-discharge mortality rates impact hospital performance assessments using an in-hospital mortality model. DESIGN, SETTING, AND PARTICIPANTS: Data were retrospectively collected on 10,502 eligible intensive care unit patients across 35 California hospitals between 2001 and 2004. MEASURES: We calculated the rates of acute care hospital transfers and early post-discharge mortality (30-day overall mortality-30-day in-hospital mortality) for each hospital. We assessed hospital performance with standardized mortality ratios (SMRs) using the Mortality Probability Model III. Using regression models, we explored the relationship between in-hospital SMRs and the rates of hospital transfers or early post-discharge mortality. We explored the same relationship using a 30-day SMR. RESULTS: In multivariable models, for each 1% increase in patients transferred to another acute care hospital, there was an in-hospital SMR reduction of -0.021 (-0.040-0.001). Additionally, a 1% increase in early post-discharge mortality was associated with an in-hospital SMR reduction of -0.049 (-0.142-0.045). Assessing hospital performance based upon 30-day mortality end point resulted in SMRs closer to 1.0 for hospitals at high and low ends of in-hospital mortality performance. CONCLUSIONS: Variations in transfer rates and potentially discharge timing appear to bias in-hospital SMR calculations. A 30-day mortality model is a potential alternative that may limit this bias.

Mortality probability model III and simplified acute physiology score II: assessing their value in predicting length of stay and comparison to APACHE IV.

BACKGROUND: To develop and compare ICU length-of-stay (LOS) risk-adjustment models using three commonly used mortality or LOS prediction models. METHODS: Between 2001 and 2004, we performed a retrospective, observational study of 11,295 ICU patients from 35 hospitals in the California Intensive Care Outcomes Project. We compared the accuracy of the following three LOS models: a recalibrated acute physiology and chronic health evaluation (APACHE) IV-LOS model; and models developed using risk factors in the mortality probability model III at zero hours (MPM(0)) and the simplified acute physiology score (SAPS) II mortality prediction model. We evaluated models by calculating the following: (1) grouped coefficients of determination; (2) differences between observed and predicted LOS across subgroups; and (3) intraclass correlations of observed/expected LOS ratios between models. RESULTS: The grouped coefficients of determination were APACHE IV with coefficients recalibrated to the LOS values of the study cohort (APACHE IVrecal) [R(2) = 0.422], mortality probability model III at zero hours (MPM(0) III) [R(2) = 0.279], and simplified acute physiology score (SAPS II) [R(2) = 0.008]. For each decile of predicted ICU LOS, the mean predicted LOS vs the observed LOS was significantly different (p

Variation in ICU risk-adjusted mortality: impact of methods of assessment and potential confounders.

BACKGROUND: Federal and state agencies are considering ICU performance assessment and public reporting; however, an accurate method for measuring performance must be selected. In this study, we determine whether a substantial variation in ICU mortality performance still exists in modern ICUs, and compare the predictive accuracy, reliability, and data burden of existing ICU risk-adjustment models. METHODS: A retrospective chart review of 11,300 ICU patients from 35 California hospitals from 2001 to 2004 was performed. We calculated standardized mortality ratios (SMRs) for each hospital using the mortality probability model III (MPM(0) III), the simplified acute physiology score (SAPS) II, and the acute physiology and chronic health evaluation (APACHE) IV risk-adjustment models. We compared discrimination, calibration, data reliability, and abstraction time for the models. RESULTS: Regardless of the model used, there was a large variation in SMRs among the ICUs studied. The discrimination and calibration were adequate for all risk-adjustment models. APACHE IV had the best discrimination (area under the receiver operating characteristic curve [AUC], 0.892) compared to MPM(0) III (AUC, 0.809), and SAPS II (AUC, 0.873; p < 0.001). The models differed substantially in data abstraction times, as follows: MPM(0)III, 11.1 min (95% confidence interval [CI], 8.7 to 13.4); SAPS II, 19.6 min (95% CI, 17.0 to 22.2); and APACHE IV, 37.3 min (95% CI, 28.0 to 46.6). CONCLUSIONS: We found substantial variation in the ICU risk-adjusted mortality rates that persisted regardless of the risk-adjustment model. With unlimited resources, the APACHE IV model offers the best predictive accuracy. If constrained by cost and manual data collection, the MPM(0) III model offers a viable alternative without a substantial loss in accuracy.

Race/ethnicity, socioeconomic status and the health of pregnant women.

We examined how traditional (income, education) and nontraditional (public assistance, material deprivation, subjective social standing) socioeconomic status (SES) indicators were associated with self-rated health, physical functioning, and depression in ethnically diverse pregnant women. Using multiple regression, we estimated the association of race/ethnicity (African American, Latino, Asian/Pacific Islander (PI) and white) and sets of SES measures on each health measure. Education, material deprivation, and subjective social standing were independently associated with all health measures. After adding all SES variables, race/ethnic disparities in depression remained for all minority groups; disparities in self-rated health remained for Asian/Pacific Islanders. Few race/ethnic differences were found in physical functioning. Our results contribute to a small literature on how SES might interact with race/ethnicity in explaining health.

Changes in the health status of women during and after pregnancy.

OBJECTIVE: To characterize the changes in health status experienced by a multi-ethnic cohort of women during and after pregnancy. DESIGN: Observational cohort. SETTING/PARTICIPANTS: Pregnant women from 1 of 6 sites in the San Francisco area (N=1,809). MEASUREMENTS AND MAIN RESULTS: Women who agreed to participate were asked to complete a series of telephone surveys that ascertained health status as well as demographic and medical factors. Substantial changes in health status occurred over the course of pregnancy. For example, physical function declined, from a mean score of 95.2 prior to pregnancy to 58.1 during the third trimester (0-100 scale, where 100 represents better health), and improved during the postpartum period (mean score, 90.7). The prevalence of depressive symptoms rose from 11.7% prior to pregnancy to 25.2% during the third trimester, and then declined to 14.2% during the postpartum period. Insufficient money for food or housing and lack of exercise were associated with poor health status before, during, and after pregnancy. CONCLUSIONS: Women experience substantial changes in health status during and after pregnancy. These data should guide the expectations of women, their health care providers, and public policy.

Prepregnancy health status and the risk of preterm delivery.

BACKGROUND: Despite extensive evaluation, our understanding of risk factors for premature delivery is incomplete. OBJECTIVE: To examine whether a woman's health status and risk factors before pregnancy are associated with a woman's risk of preterm delivery, independent of risk factors that occur during pregnancy. DESIGN, SETTING, AND PARTICIPANTS: Prospective cohort of pregnant women in the San Francisco Bay area who delivered a singleton infant (n = 1619). MAIN OUTCOME MEASURE: Preterm delivery (<37 weeks' gestational age). RESULTS: Sociodemographic characteristics alone explained 13.0% of the risk of preterm delivery, whereas risk factors that occurred before pregnancy explained 39.8% and risk factors that occurred during pregnancy explained 47.1%. After we adjusted for sociodemographic characteristics, prepregnancy risk factors, and pregnancy risk factors, women who reported poor physical function during the month before conception were nearly twice as likely to experience a preterm delivery (odds ratio, 1.97; 95% confidence interval, 1.18-3.30) as women with better physical function. CONCLUSION: A broader focus on the health of women prior to pregnancy may improve rates of preterm delivery.

Differences in mortality among patients with community-acquired pneumonia in California by ethnicity and hospital characteristics.

PURPOSE: To determine ethnic disparities in mortality for patients with community-acquired pneumonia, and the potential effects of hospital characteristics on disparities, we compared the risk-adjusted mortality of white, African American, Hispanic, and Asian American patients hospitalized for community-acquired pneumonia. METHODS: We studied patients discharged with community-acquired pneumonia in 1996 from an acute care hospital in California (n = 54,874). Logistic regression models were used to examine the association between ethnicity and hospital characteristics and 30-day mortality after adjusting for clinical characteristics. RESULTS: The overall 30-day mortality was 12.2%. After adjustment for demographic, clinical, and hospital characteristics, Hispanic (odds ratio [OR] = 0.81; 95% confidence interval [CI]: 0.73 to 0.90) and Asian American patients (OR = 0.88; 95% CI: 0.77 to 1.00) had lower mortality than did white patients, whereas African Americans had a similar mortality to whites (OR = 0.93; 95% CI: 0.83 to 1.06). There were no overall differences in mortality by hospital characteristics (i.e., teaching status, rural location, and public or district hospital). CONCLUSION: Hispanics and Asian Americans have a lower risk of death from community-acquired pneumonia than whites in California. No overall differences in mortality were observed by hospital characteristics.