scholarly journals Common sampling and modeling approaches to analyzing readmission risk that ignore clustering produce misleading results

2020 ◽  
Vol 20 (1) ◽  
Author(s):  
Huaqing Zhao ◽  
Samuel Tanner ◽  
Sherita H. Golden ◽  
Susan G. Fisher ◽  
Daniel J. Rubin
Keyword(s):  
Logistic Regression ◽  
Readmission Rate ◽  
Medical Center ◽  
Academic Medical Center ◽  
Model Performance ◽  
Readmission Rates ◽  
Modeling Approaches ◽  
Multiple Variables ◽  
Highly Correlated ◽  
Readmission Risk

Abstract Background There is little consensus on how to sample hospitalizations and analyze multiple variables to model readmission risk. The purpose of this study was to compare readmission rates and the accuracy of predictive models based on different sampling and multivariable modeling approaches. Methods We conducted a retrospective cohort study of 17,284 adult diabetes patients with 44,203 discharges from an urban academic medical center between 1/1/2004 and 12/31/2012. Models for all-cause 30-day readmission were developed by four strategies: logistic regression using the first discharge per patient (LR-first), logistic regression using all discharges (LR-all), generalized estimating equations (GEE) using all discharges, and cluster-weighted (CWGEE) using all discharges. Multiple sets of models were developed and internally validated across a range of sample sizes. Results The readmission rate was 10.2% among first discharges and 20.3% among all discharges, revealing that sampling only first discharges underestimates a population’s readmission rate. Number of discharges was highly correlated with number of readmissions (r = 0.87, P < 0.001). Accounting for clustering with GEE and CWGEE yielded more conservative estimates of model performance than LR-all. LR-first produced falsely optimistic Brier scores. Model performance was unstable below samples of 6000–8000 discharges and stable in larger samples. GEE and CWGEE performed better in larger samples than in smaller samples. Conclusions Hospital readmission risk models should be based on all discharges as opposed to just the first discharge per patient and utilize methods that account for clustered data.

scholarly journals Common sampling and modeling approaches to analyzing readmission risk that ignore clustering produce misleading results

2020 ◽  
Author(s):  
Huaqing Zhao ◽  
Samuel Tanner ◽  
Sherita H. Golden ◽  
Susan Fisher ◽  
Daniel J. Rubin
Keyword(s):  
Logistic Regression ◽  
Readmission Rate ◽  
Medical Center ◽  
Academic Medical Center ◽  
Model Performance ◽  
Readmission Rates ◽  
Modeling Approaches ◽  
Multiple Variables ◽  
Highly Correlated ◽  
Readmission Risk

Abstract Background There is little consensus on how to sample hospitalizations and analyze multiple variables to model readmission risk. The purpose of this study was to compare readmission rates and the accuracy of predictive models based on different sampling and multivariable modeling approaches. Methods We conducted a retrospective cohort study of 17 284 adult diabetes patients with 44 203 discharges from an urban academic medical center between 1/1/2004 and 12/31/2012. Models for all-cause 30-day readmission were developed by four strategies: logistic regression using the first discharge per patient (LR-first), logistic regression using all discharges (LR-all), generalized estimating equations (GEE) using all discharges, and cluster-weighted (CW) GEE using all discharges. Multiple sets of models were developed and internally validated across a range of sample sizes. Results The readmission rate was 10.2% among first discharges and 20.3% among all discharges, revealing that sampling only first discharges underestimates a population’s readmission rate. Number of discharges was highly correlated with number of readmissions (r = 0.87, P < 0.001). Accounting for clustering with GEE and CWGEE yielded more conservative estimates of model performance than LR-all. LR-first produced falsely optimistic Brier scores. Model performance was unstable below samples of 6 000–8 000 discharges and stable in larger samples. GEE and CWGEE performed better in larger samples than in smaller samples. Conclusions Hospital readmission risk models should be based on all discharges as opposed to just the first discharge per patient and utilize methods that account for clustered data.

scholarly journals Common sampling and modeling approaches to analyzing readmission risk that ignore clustering produce misleading results

2020 ◽  
Author(s):  
Huaqing Zhao ◽  
Samuel Tanner ◽  
Sherita H. Golden ◽  
Susan Fisher ◽  
Daniel J. Rubin
Keyword(s):  
Logistic Regression ◽  
Readmission Rate ◽  
Medical Center ◽  
Academic Medical Center ◽  
Model Performance ◽  
Readmission Rates ◽  
Modeling Approaches ◽  
Multiple Variables ◽  
Highly Correlated ◽  
Readmission Risk

Abstract Background: There is little consensus on how to sample hospitalizations and analyze multiple variables to model readmission risk. The purpose of this study was to compare readmission rates and the accuracy of predictive models based on different sampling and multivariable modeling approaches. Methods: We conducted a retrospective cohort study of 17 284 adult diabetes patients with 44 203 discharges from an urban academic medical center between 1/1/2004 and 12/31/2012. Models for all-cause 30-day readmission were developed by four strategies: logistic regression using the first discharge per patient (LR-first), logistic regression using all discharges (LR-all), generalized estimating equations (GEE) using all discharges, and cluster-weighted (CW) GEE using all discharges. Multiple sets of models were developed and internally validated across a range of sample sizes. Results: The readmission rate was 10.2% among first discharges and 20.3% among all discharges, revealing that sampling only first discharges underestimates a population’s readmission rate. Number of discharges was highly correlated with number of readmissions (r=0.87, P<0.001). Accounting for clustering with GEE and CWGEE yielded more conservative estimates of model performance than LR-all. LR-first produced falsely optimistic Brier scores. Model performance was unstable below samples of 6 000-8 000 discharges and stable in larger samples. GEE and CWGEE performed better in larger samples than in smaller samples. Conclusions: Hospital readmission risk models should be based on all discharges as opposed to just the first discharge per patient and utilize methods that account for clustered data.

Impacting Readmission Rates and Patient Satisfaction: Results of a Discharge Pharmacist Pilot Program

2011 ◽  
Vol 46 (11) ◽  
pp. 876-883 ◽  
Author(s):  
Samaneh Tavalali Wilkinson ◽  
Pal Aroop ◽  
J. Couldry Richard
Keyword(s):  
Patient Satisfaction ◽  
High Risk ◽  
Readmission Rate ◽  
Medical Center ◽  
Academic Medical Center ◽  
Control Group ◽  
Unplanned Readmission ◽  
Discharge Process ◽  
Readmission Rates ◽  
The Impact

Background Readmission to a hospital within 30 days of discharge has become a key quality outcome measure. With an observed 30-day readmission rate as high as 20% and attributed costs of almost $17.4 billion a year for Medicare patients, the potential implications for patients and the entire health care system are significant. Medication-related complications have been shown to increase the risk for unplanned readmission. Pharmacists have an opportunity to impact quality and cost by risk stratifying and identify patients at high risk for hospital readmission. Objective To study the impact of a pharmacist-driven discharge counseling program for high-risk patients identified by BOOST (Better Outcomes for Older adults through Safe Transitions) criteria on 30-day readmission rates. Method This was a prospective, cohort, nonrandomized trial at a single medical-surgical unit with telemetry capability at a single academic medical center including 669 patients who were older than 18 years. Primary outcome was 30-day readmission rate. Secondary outcomes were the number and type of pharmacist interventions, cost avoidance, and patient satisfaction results. Results The readmission rate for patients counseled by a pharmacist during the discharge process was 15.7% compared to 21.6% for patients not counseled by a pharmacist on discharge (relative risk [RR] 0.728; 95% confidence interval [CI], 0.514–1.032; P = .04). The readmission rate for adult medicine patients not counseled at discharge by a pharmacist in the study was comparable to the readmission rates of a similar patient control group at 3 months and 1 year prior to the initiation of the study (18.7% and 19.1% vs 19.6%). Conclusions Pharmacists' support in the discharge process facilitated increased communication on the multidisciplinary team and resulted in a lower unplanned readmission rate for patients.

scholarly journals Readmission Rates and Medication Selection for Patients with Heart Failure Preserved Ejection Fraction

2020 ◽  
Author(s):  
Troy Kramer ◽  
Carrie Vogler ◽  
Robert Robinson ◽  
Mukul Bhattarai
Keyword(s):  
Heart Failure ◽  
Logistic Regression ◽  
Ejection Fraction ◽  
Medical Center ◽  
Academic Medical Center ◽  
Univariate Analysis ◽  
Prescribing Patterns ◽  
Preserved Ejection Fraction ◽  
Multivariate Logistic Regression ◽  
Readmission Rates

Purpose Heart failure with preserved ejection fraction (HFpEF) has less guideline driven treatment options due to a lack of trials demonstrating medications with improved clinical outcomes for this patient population. The primary objective of this study is to determine which medications and dosages are related to high readmission rates for HFpEF patients. Methods A retrospective, single center, chart review was performed on patients with HFpEF at an academic medical center. Heart failure patients ages between 18-89 with an ejection fraction ≥45% from a transthoracic echocardiogram (TTE) were included. Primary outcomes include 30-day all cause readmission rates, prescribing patterns, and avoidance of potentially harmful medications. Descriptive statistics and multivariate logistic regression were used to assess potential risk factors. Results This study analyzed 455 patient admissions. Univariate analysis shows patients who were not readmitted were more likely to be on furosemide (54% vs 42%; p = 0.019). Conversely, readmitted patients were more likely to be taking bumetanide (4% vs 1%; p = 0.039). Lisinopril was the only angiotensin converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB) associated with lower readmission rates (p = 0.036). Multivariate logistic regression showed bumetanide on admission (OR 14.6, p = 0.001), discharged on rosuvastatin (OR 6.29, p = 0.003) and meloxicam therapy (OR 6.33, p = 0.003) to be independent predictors of hospital readmission. Conclusion Three independent pharmacologic predictors for 30-day readmissions for patients with HFpEF were therapy with bumetanide, meloxicam, or rosuvastatin. Further research is needed to clarify the significance of these results.

scholarly journals Missing the Mark: Inaccuracy of administrative data in identification of hospitalized patients with pneumonia and results of a systematic clinical reclassification process on readmission rates

2017 ◽  
Vol 6 (6) ◽  
pp. 35
Author(s):  
Marcus D. Ruopp ◽  
Joel C. Boggan ◽  
Thomas L. Holland ◽  
Mary Jane Stillwagon ◽  
Joseph A. Govert ◽  
...  
Keyword(s):  
Administrative Data ◽  
Readmission Rate ◽  
Medical Center ◽  
Academic Medical Center ◽  
Hospital Readmissions ◽  
Final Diagnosis ◽  
Healthcare Systems ◽  
Readmission Rates ◽  
Control And Prevention ◽  
One Year

Objective: Pneumonia readmissions carry financial ramifications under the Hospital Readmissions Reduction Program (HRRP). As readmission determination utilizes administrative data, healthcare systems should evaluate accuracy of pneumonia diagnoses. We sought to develop a systemic process for pneumonia classification review and determine potential effects on pneumonia readmissions in a tertiary academic medical center in the United States.Methods: We performed independent reviews of all pneumonia discharges within 48 hours of discharge over a one-year period. We reclassified all pneumonia discharges into four categories based on the Centers for Disease Control and Prevention reference standard. Secondary review of discordant classifications was performed by discharging providers to determine final diagnosis. The primary outcome was readmission rate within 30 days by pneumonia clinical classification category.Results: Two hundred seventy-eight discharges were reviewed, with overall readmission rate of 18.0%. Independent review confirmed 191 cases (68.7%) as definite or probable pneumonia, while 87 cases (31.3%) were classified as either probably not or not pneumonia. Readmission rates differed significantly between cases reviewed as pneumonia vs. those reviewed as unlikely to be pneumonia (14.1% vs. 26.4%, p < .02). Discharging attending physicians agreed with independent reviewers in 58/87 cases (66.6%), attenuating readmission differences (rate 16.8% for those finalized as pneumonia vs. 22.4% for another diagnosis, p = .32). Pneumonia readmissions were reduced by 1.2% using the classification standard.Conclusions: Complex conditions such as pneumonia may be inaccurately diagnosed in many patients, potentially affecting penalties associated with readmission rates. Therefore, it is imperative that healthcare systems adopt systematic review processes to standardize diagnoses and improve comparative administrative data.

Abstract WP338: The Impact of Social Determinants of Health on 30-Day Readmission After Ischemic Stroke

Stroke ◽  
2020 ◽  
Vol 51 (Suppl_1) ◽  
Author(s):  
Shivakrishna Kovi ◽  
Shumei Man ◽  
Ken Uchino ◽  
Anne S. Tang ◽  
Rocio Lopez
Keyword(s):  
Logistic Regression ◽  
Ischemic Stroke ◽  
Social Determinants ◽  
Household Income ◽  
Readmission Rate ◽  
Multivariate Logistic Regression ◽  
Readmission Rates ◽  
The Impact ◽  
Zip Code ◽  
Readmission Risk

Objective: Readmission after ischemic stroke presents immense social and financial burden on the patient, family and society. The impact of social determinants of health on post-stroke readmission have not been well studied. This study aimed to examine association of patient social determinants with 30-day readmission risk after ischemic stroke. Methods: We examined patients who were hospitalized for acute ischemic stroke (ICD-9 codes 433.1, 434, and 436) in the state of Florida and New York from February 2012 to November 2013, with a 30-day run-in time and 30-day follow up time. Data were obtained from the State Inpatient Database which was then linked with Census data on social determinants at zip code level. Multivariate logistic regression models were generated to study the association of patient factors with 30-day readmission, after adjusting for patient characteristics, in-hospital infection, hemiplegia, and 16 comorbidities. All statistical analyses were conducted using SAS Version 9.4 software (SAS Institute). Results: A total of 127,290 patients were included in the study. The overall 30-day readmission rate was 23%. The 30-day readmission rates differed by race, insurance but not age, sex or household income at zip code level. The 30-day readmission rate was higher amongst black (27%) and Hispanics (25%), and lower in Native American (17%) and White (22%). Comparable to older age groups, patients who suffered a stroke at young age (<50 years old) had 23% readmission rates at 30 days. In the multivariate logistic regression model, age was not a risk factor for readmission [adjusted odds ratio (OR) 1.0, 95% confidence interval (CI) 1.0-1.0]. Compared to white, black race had higher risk of 30-day readmission (OR1.2, 95% CI 1.1-1.2), as well as Hispanic race (OR1.1, 95% CI 1.1-1.2). Compared to Medicare insurance, patient on Medicaid had higher risk of 30-day readmission (OR 1.1, 95% CI 1.1-1.2). Self-pay patients had lower risk of readmission (OR 0.6, 95% CI 0.5-0.7). Conclusion: Race and medical insurance, not age, sex, or household income, has significant influence in30-day readmission risk after ischemic stroke. This will allow further targeted intervention for readmission reduction.

Abstract P389: External Validation of the 2CAN Score for Inpatient Stroke Detection

Stroke ◽  
2021 ◽  
Vol 52 (Suppl_1) ◽  
Author(s):  
oleg otlivanchik ◽  
Jenny Lu ◽  
Natalie Cheng ◽  
Daniel L Labovitz ◽  
charles esenwa ◽  
...  
Keyword(s):  
Logistic Regression ◽  
Negative Predictive Value ◽  
Predictive Value ◽  
Medical Center ◽  
Academic Medical Center ◽  
External Validation ◽  
Acute Stroke Care ◽  
Metabolic Encephalopathy ◽  
Cardiac Procedure ◽  
Academic Medical

Introduction: Up to 15% of all strokes occur in patients who are already hospitalized for other conditions. A validated clinical tool to help rapidly discriminate between mimics and stroke among inpatients could greatly improve acute stroke care. Recently, the 2CAN score was developed and validated at a single Midwest academic medical center to identify inpatient strokes; a score of ≥2 was highly sensitive and specific for stroke. We sought to externally validate the 2CAN score at our institution. Methods: We conducted a retrospective cohort study of consecutive inpatient stroke codes at a single Northeast academic medical center from 7/1/2018 to 11/1/2019. Pre-specified variables, including patient demographics, vascular risk factors, and clinical features (neurological examination, vital signs, laboratory values, and final diagnoses), were abstracted from the electronic medical record. We determined the sensitivity, specificity, positive and negative predictive value of a 2CAN score ≥2 for stroke (ischemic stroke, hemorrhagic stroke, or TIA) in our cohort. The 2CAN score consists of clinical deficit score (0-3 points), recent cardiac procedure (1 point), atrial fibrillation (1 point), and code called within 24 hours of admission (1 point). We used multivariate logistic regression to identify additional determinants of stroke. Results: We identified 111 inpatient stroke codes on 110 patients, mean age 67 ± 1 year, 46.8% women, and 73.8% Black or Hispanic. Final diagnosis was stroke for 54 codes (48.6%) and mimic for 57 codes (51.3%), most commonly toxic-metabolic encephalopathy. 2CAN score ≥2 had 96.3% sensitivity, 45.6% specificity, 62.7% positive predictive value, and 92.3% negative predictive value for stroke. In a multivariable logistic regression model, only recent cardiac procedure (OR: 5.5; 95% CI: 1.1-27.5) and high clinical deficit score (OR: 3.9; 95% CI: 1.9-6.1) predicted stroke. Conclusion: The 2CAN score is externally valid and helps distinguish stroke from mimic in inpatients; having a score of <2 makes stroke very unlikely.

scholarly journals Reducing Preventable Readmissions in a Neurosurgical Patient Population at an Academic Medical Center

Neurosurgery ◽  
2019 ◽  
Vol 66 (Supplement_1) ◽  
Author(s):  
Shayan Moosa ◽  
Lindsay Bowerman ◽  
Ellen Smith ◽  
Mindy Bryant ◽  
Natalie Krovetz ◽  
...  
Keyword(s):  
High Risk ◽  
Patient Population ◽  
Medical Center ◽  
Academic Medical Center ◽  
Readmission Rates ◽  
Academic Medical ◽  
Neurosurgical Patient ◽  
Neurosurgical Patients ◽  
Significant Difference

Abstract INTRODUCTION Hospital readmissions are extremely costly in terms of time and resources and negatively impact patient safety and satisfaction. In this study, we performed a Pareto analysis of 30-day readmissions in a neurosurgical patient population in order to identify patients at high-risk for readmission. Using this information, we implemented a new practice parameter with the goal of reducing preventable readmissions. METHODS Patient characteristics and causes for readmission were prospectively collected for all neurosurgical patients readmitted to an academic medical center within 30 d of discharge between July and October 2018. A program was then initiated where postoperative neurosurgical spine patients were contacted by phone at standardized intervals before their 2-wk follow-up appointment, with the purpose of more quickly addressing surgical concerns and/or coordinating care for general medical issues. Finally, 30-d readmission rates were compared between the initial 4-mo period and January 2019 through April 2019. RESULTS Prior to intervention, the largest group of readmitted patients included those who had undergone recent spinal surgery (16/47, 34%). Among spine readmissions during this time, 47% were readmitted before their two-week follow-up appointment, 67% lived over 50 miles from the medical center, and 40% were Medicare-insured. There was a statistically significant difference in the mean rate of spine readmissions per month in the periods before (7.0%) and after (3.0%) the program onset (P = .029, 57% decline). The total number of surgically and medically related spine readmissions decreased between the pre- and postintervention periods from 10 to 3 (70%) and 3 to 1 (67%), respectively. CONCLUSION Our data suggests that a large number of neurosurgical readmissions may be prevented by the simple process of early follow-up and consistent communication via telephone. Readmission rates may be further reduced by standardizing the coordination of postoperative general medical follow-up and providing thorough wound care teaching for high-risk patients.

scholarly journals Development, implementation, and prospective validation of a model to predict 60-day end-of-life in hospitalized adults upon admission at three sites

2020 ◽  
Vol 20 (1) ◽  
Author(s):  
Vincent J. Major ◽  
Yindalon Aphinyanaphongs
Keyword(s):  
Positive Predictive Value ◽  
Receiver Operating Characteristic ◽  
Predictive Value ◽  
Operating Characteristic ◽  
Medical Center ◽  
Academic Medical Center ◽  
Model Performance ◽  
Risk Of Death ◽  
Temporal Validation ◽  
Receiver Operating

Abstract Background Automated systems that use machine learning to estimate a patient’s risk of death are being developed to influence care. There remains sparse transparent reporting of model generalizability in different subpopulations especially for implemented systems. Methods A prognostic study included adult admissions at a multi-site, academic medical center between 2015 and 2017. A predictive model for all-cause mortality (including initiation of hospice care) within 60 days of admission was developed. Model generalizability is assessed in temporal validation in the context of potential demographic bias. A subsequent prospective cohort study was conducted at the same sites between October 2018 and June 2019. Model performance during prospective validation was quantified with areas under the receiver operating characteristic and precision recall curves stratified by site. Prospective results include timeliness, positive predictive value, and the number of actionable predictions. Results Three years of development data included 128,941 inpatient admissions (94,733 unique patients) across sites where patients are mostly white (61%) and female (60%) and 4.2% led to death within 60 days. A random forest model incorporating 9614 predictors produced areas under the receiver operating characteristic and precision recall curves of 87.2 (95% CI, 86.1–88.2) and 28.0 (95% CI, 25.0–31.0) in temporal validation. Performance marginally diverges within sites as the patient mix shifts from development to validation (patients of one site increases from 10 to 38%). Applied prospectively for nine months, 41,728 predictions were generated in real-time (median [IQR], 1.3 [0.9, 32] minutes). An operating criterion of 75% positive predictive value identified 104 predictions at very high risk (0.25%) where 65% (50 from 77 well-timed predictions) led to death within 60 days. Conclusion Temporal validation demonstrates good model discrimination for 60-day mortality. Slight performance variations are observed across demographic subpopulations. The model was implemented prospectively and successfully produced meaningful estimates of risk within minutes of admission.

Pharmacist Transition-of-Care Services Improve Patient Satisfaction and Decrease Hospital Readmissions

2020 ◽  
pp. 089719002095826
Author(s):  
Katherine L. March ◽  
Michael J. Peters ◽  
Christopher K. Finch ◽  
Lauchland A. Roberts ◽  
Katie M. McLean ◽  
...  
Keyword(s):  
Patient Satisfaction ◽  
Medical Center ◽  
Academic Medical Center ◽  
Hospital Readmissions ◽  
Transitions Of Care ◽  
Control Group ◽  
Readmission Rates ◽  
Improve Patient Satisfaction ◽  
Phone Calls ◽  
Improve Patient

Background: Pharmacists ability to directly impact patient satisfaction through increases in the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) surveys utilizing transitions-of-care (TOC) services is unclear. Methods: Retrospective analysis of TOC patients from 07/01/2018 to 03/31/2019 was conducted. Intervention (INV) patients received pharmacist medication reconciliation and education prior to discharge and post-discharge telephone follow-up. All other patients served as control group (CON). Primary outcome: Evaluate impact of TOC services on HCAHPS scores for “Communication about Medicines” and “Care Transitions.” Secondary outcomes: 30-day readmissions, quantification of prevented potential safety events, assessment of discharge prescriptions sent to the academic medical center outpatient pharmacy (MOP) for TOC patients. Results: Of 1,728 patients screened, 414 patients met inclusion criteria (INV = 414, CON = 1314). A significant improvement (14.7%; p = <0.0001) in overall medication-related HCAHPS results was seen when comparing pre- vs post-implementation of the TOC service. Statistically significant increases for individual questions “staff told you what the medicine was for” (14.2%; p = 0.018), “staff describe possible effects” (21.2%; p = 0.004), and “understood the purpose of taking medications” (11.4%; p = 0.035) were observed. A non-significant decrease in 30-day readmission rates for the groups was observed (CON 16.4%, INV 13.3%; p = 0.133); however, an unplanned subgroup analysis evaluating impact of discharge phone calls on 30-day readmission rates revealed a significant reduction of 17.3% to 12.4% (p = 0.007). One hundred forty-three medication safety event(s) were potentially prevented by the TOC pharmacist. Lastly, 562 prescriptions were captured at the MOP as a result of the TOC initiative. Conclusions: Pharmacy-based TOC models can improve patient satisfaction, prevent hospital readmissions, and generate revenue.

Sign in / Sign up

Export Citation Format

Share Document