Impact of a machine learning algorithm on time to palliative care in a primary care population: protocol for a stepped-wedge pragmatic randomized trial.

BACKGROUND: As primary care populations age, timely identification of palliative care need is becoming increasingly relevant. Previous studies have targeted particular patient populations with life-limiting disease, but few have focused on patients in a primary care setting. Toward this end, we propose a stepped-wedge pragmatic randomized trial whereby a machine learning algorithm identifies patients empaneled to primary care units at Mayo Clinic (Rochester, Minnesota, United States) with high likelihood of palliative care need. METHODS: 42 care team units in 9 clusters were randomized to 7 wedges, each lasting 42 days. For care teams in treatment wedges, palliative care specialists review identified patients, making recommendations to primary care providers when appropriate. Care teams in control wedges receive palliative care under the standard of care. DISCUSSION: This pragmatic trial therefore integrates machine learning into clinical decision making, instead of simply reporting theoretical predictive performance. Such integration has the possibility to decrease time to palliative care, improving patient quality of life and symptom burden. TRIAL REGISTRATION: Clinicaltrials.gov NCT04604457 , restrospectively registered 10/26/2020. PROTOCOL: v0.5, dated 9/23/2020.

Real-World Incidence of Breakthrough Coronavirus Disease 2019 Hospitalization After Vaccination vs Natural Infection in a Large, Local, Empaneled Primary Care Population Using Time-to-Event Analysis.

We followed 106 349 primary care patients for 22 385 3099 person-days across 21 calendar months and documented 69 breakthrough coronavirus disease 2019 (COVID-19) hospitalizations: 65/102,613 (0.06%) among those fully vaccinated, 3/11 047 (0.03%) among those previously infected, and 1/7,313 (0.01%) among those with both statuses. These data give providers real-world context regarding breakthrough COVID-19 hospitalization risk.

Real-time risk prediction of colorectal surgery-related post-surgical complications using GRU-D model.

BACKGROUND: Post-surgical complications (PSCs) have been an increasing concern for hospitals in light of Medicare penalties for 30-day readmissions. PSCs have become a target for quality improvement and benchmark for the healthcare system. Over half (60 %) of the deep or organ space surgical site infections are discovered after discharge, leading to a readmission. There has thus been a push to develop risk prediction models for targeted preventive interventions for PSCs. OBJECTIVE: We experiment several Gated Recurrent Unit with Decay (GRU-D) based deep learning architectures with various feature sampling schemes in predicting the risk of colorectal PSCs and compare with atemporal logistic regression models (logit). METHOD: We used electronic health record (EHR) data of 3,535 colorectal surgical patients involved in the national surgical quality improvement program (NSQIP) between 2006 and 2018. Single layer, stacked layer, and multimodal GRU-D models with sigmoid activation were used to develop risk prediction models. Area Under the Receiver Operating Characteristic curve (AUROC) was calculated by comparing predicted probability of developing complications versus truly observed PSCs (NSQIP adjudicated) within 30 days after surgery. We set up cross-validation and an independent held-out dataset for testing model performance consistency. RESULTS AND CONCLUSION: The primary contribution of our study is the formulation of a novel real-time PSC risk prediction task using GRU-D with demonstrated clinical utility. GRU-D outperforms the logit model in predicting wound and organ space infection and shows improved performance as additional data points become available. Logit model outperforms GRU-D before surgery for superficial infection and bleeding. For the same sampling scheme, there is no obvious advantage of complex architectures (stacked, multimodal) over single layer GRU-D. Obtaining more data points closer to the occurrence of PSCs is more important than using a more frequent sampling scheme in training GRU-D models. The fourth predicted risk quartile by single layer GRU-D contains 63 %, 59 %, and 66 % organ space infection cases, at 4 h before, 72 h after, and 168 h after surgery, respectively, suggesting its potential application as a bedside risk assessment tool.

Anemia in Critically Ill Patients With Acute Respiratory Distress Syndrome and Posthospitalization Physical Outcomes.

OBJECTIVE: Anemia is common during critical illness and often persists after hospital discharge; however, its potential association with physical outcomes after critical illness is unclear. Our objective was to assess the associations between hemoglobin at intensive care unit (ICU) and hospital discharge with physical status at 3-month follow-up in acute respiratory distress syndrome (ARDS) survivors. METHODS: This is a secondary analysis of a multisite prospective cohort study of 195 mechanically ventilated ARDS survivors from 13 ICUs at 4 teaching hospitals in Baltimore, Maryland. Multivariable regression was utilized to assess the relationships between ICU and hospital discharge hemoglobin concentrations with measures of physical status at 3 months, including muscle strength (Medical Research Council sumscore), exercise capacity (6-minute walk distance [6MWD]), and self-reported physical functioning (36-Item Short-Form Health Survey [SF-36v2] Physical Function score and Activities of Daily Living [ADL] dependencies). RESULTS: Median (interquartile range) hemoglobin concentrations at ICU and hospital discharge were 9.5 (8.5-10.7) and 10.0 (9.0-11.2) g/dL, respectively. In multivariable regression analyses, higher ICU discharge hemoglobin concentrations (per 1 g/dL) were associated with greater 3-month 6MWD mean percent of predicted (3.7% [95% confidence interval 0.8%-6.5%]; P = .01) and fewer ADL dependencies (-0.2 [-0.4 to -0.1]; P = .02), but not with percentage of maximal muscle strength (0.7% [-0.9 to 2.3]; P = .37) or SF-36v2 normalized Physical Function scores (0.8 [-0.3 to 1.9]; P = .15). The associations of physical outcomes and hospital discharge hemoglobin concentrations were qualitatively similar, but none were statistically significant. CONCLUSIONS: In ARDS survivors, higher hemoglobin concentrations at ICU discharge, but not hospital discharge, were significantly associated with improved exercise capacity and fewer ADL dependencies. Future studies are warranted to further assess these relationships.

Associations of Skilled Nursing Facility Quality Ratings With 30-Day Rehospitalizations and Emergency Department Visits.

Skilled nursing facilities (SNFs) increasingly provide care to patients after hospitalization. The Centers for Medicare & Medicaid Services reports ratings for SNFs for overall quality, staffing, health inspections, and clinical quality measures. However, the relationship between these ratings and patient outcomes remains unclear. In this retrospective cohort study, we reviewed the electronic health records of 3,923 adult patients discharged from the hospital and admitted to 9 SNFs served by a health care delivery system. We used Cox proportional hazards models to examine associations between the overall quality and individual ratings and our primary outcomes of 30-day rehospitalizations and 30-day emergency department visits. Patients in higher-rated facilities had a 13% lower risk of 30-day rehospitalization than patients in lower-rated facilities (hazard ratio, 0.87; 95% CI, 0.76-0.99). The risk of emergency department visits was also lower for patients in facilities with a higher overall quality rating and a higher quality measures rating. Staffing and health inspection ratings were not associated with our primary outcomes. These findings may help inform providers and nursing home policy makers.

Prediction of individual outcomes for asthma sufferers.

We consider the problem of individual-specific medication level recommendation (initiation, removal, increase, or decrease) for asthma sufferers. Asthma is one of the most common chronic diseases in both adults and children, affecting 8% of the US population and costing $37-63 billion/year in the United States of America. Asthma is a complex disease, whose symptoms may wax and wane, making it difficult for clinicians to predict outcomes and prognosis. Improved ability to predict prognosis can inform decision making and may promote conversations between clinician and provider around optimizing medication therapy. Data from the US Medical Expenditure Panel Survey (MEPS) years 2000-2010 were used to fit a longitudinal model for a multivariate response of adverse events (Emergency Department or in-patient visits, excessive rescue inhaler use, and oral steroid use). To reduce bias in the estimation of medication effects, medication level was treated as a latent process which was restricted to be consistent with prescription refill data. This approach is demonstrated to be effective in the MEPS cohort via predictions on a validation hold out set and a synthetic data simulation study. This framework can be easily generalized to medication decisions for other conditions as well.

Assessing the Safety of Overlapping Surgery at a Children's Hospital.

IMPORTANCE: Media reports have questioned the safety of overlapping surgical procedures, and national scrutiny has underscored the necessity of single-center evaluations of its safety; however, sample sizes are likely small. We compared the safety profiles of overlapping and nonoverlapping pediatric procedures at a single children's hospital and discussed methodological considerations of the evaluation. DATA AND DESIGN: Retrospective analysis of inpatient pediatric surgical procedures (January 2013 to September 2015) at a single pediatric referral center. Overlapping and nonoverlapping procedures were matched in an unbalanced manner (m:n) by procedure. Mixed models adjusting for Vizient-predicted risk, case-mix, and surgeon compared inpatient mortality and length of stay (LOS). RESULTS: Among 315 overlapping procedures, 256 (81.3%) were matched to 645 nonoverlapping procedures. There were 6 deaths in all. The adjusted odds ratio for mortality did not differ significantly between nonoverlapping and overlapping procedures (adjusted odds ratio = 0.94 vs overlapping; 95% CI, 0.02-48.5; P = 0.98). Wide confidence intervals were minimally improved with Bayesian methods (95% CI, 0.07-12.5). Adjusted LOS estimates were not clinically different by overlapping status (0.6% longer for nonoverlapping; 95% CI, 9.7% shorter to 12.2% longer; P = 0.91). Among the 87 overlapping procedures with the greatest overlap (≥60 min or ≥50% of operative duration), there were no deaths. CONCLUSIONS: The safety of overlapping and nonoverlapping surgical procedures did not differ at this children's center. These findings may not extrapolate to other centers. LOS or intraoperative measures may be more appropriate than mortality for safety evaluations due to low event rates for mortality.

Safety of Overlapping Surgery at a High-volume Referral Center.

OBJECTIVE: To compare safety profiles of overlapping and nonoverlapping surgical procedures at a large tertiary-referral center where overlapping surgery is performed. BACKGROUND: Surgical procedures are frequently performed as overlapping, wherein one surgeon is responsible for 2 procedures occurring at the same time, but critical portions are not coincident. The safety of this practice has not been characterized. METHODS: Primary analyses included elective, adult, inpatient surgical procedures from January 2013 to September 2015 available through University HealthSystem Consortium. Overlapping and nonoverlapping procedures were matched in an unbalanced manner (m:n) by procedure type. Confirmatory analyses from the American College of Surgeons-National Surgical Quality Improvement Program investigated elective surgical procedures from January 2011 to December 2014. We compared outcomes mortality and length of stay after adjustment for registry-predicted risk, case-mix, and surgeon using mixed models. RESULTS: The University HealthSystem Consortium sample included 10,765 overlapping cases, of which 10,614 (98.6%) were matched to 16,111 nonoverlapping procedures. Adjusted odds ratio for inpatient mortality was greater for nonoverlapping procedures (adjusted odds ratio, OR = 2.14 vs overlapping procedures; 95% confidence interval, CI 1.23-3.73; P = 0.007) and length of stay was no different (+1% for nonoverlapping cases; 95% CI, -1% to +2%; P = 0.50). In confirmatory analyses, 93.7% (3712/3961) of overlapping procedures matched to 5,637 nonoverlapping procedures. The 30-day mortality (adjusted OR = 0.69 nonoverlapping vs overlapping procedures; 95% CI, 0.13-3.57; P = 0.65), morbidity (adjusted OR = 1.11; 95% CI, 0.92-1.35; P = 0.27) and length of stay (-4% for nonoverlapping; 95% CI, -4% to -3%; P < 0.001) were not clinically different. CONCLUSIONS: These findings from administrative and clinical registries support the safety of overlapping surgical procedures at this center but may not extrapolate to other centers.

Incorporation of Treatment Response, Tumor Grade and Receptor Status Improves Staging Quality in Breast Cancer Patients Treated with Neoadjuvant Chemotherapy.

BACKGROUND: Improved staging systems that better predict survival for breast cancer patients who receive neoadjuvant chemotherapy (NAC) by accounting for clinical pathological stage plus estrogen receptor (ER) and grade (CPS+EG) and ERBB2 status (Neo-Bioscore) have been proposed. We sought to evaluate the generalizability and performance of these staging systems in a national cohort. METHODS: The National Cancer Database (2006-2012) was reviewed for patients with breast cancer who received NAC and survived ≥90 days after surgery. Four systems were evaluated: clinical/pathologic American Joint Committee on Cancer (AJCC) 7th edition, CPS+EG, and Neo-Bioscore. Unadjusted Kaplan-Meier analysis and adjusted Cox proportional hazards models quantified overall survival (OS). Systems were compared using area under the curve (AUC) and integrated discrimination improvement (IDI). RESULTS: Overall, 43,320 patients (5-year OS 76.0, 95% confidence interval [CI] 75.4-76.5%) were included, 12,002 of whom had evaluable Neo-Bioscore. AUC at 5 years for CPS+EG (0.720, 95% CI 0.714-0.726) and Neo-Bioscore (0.729, 95% CI 0.716-0.742) were improved relative to AJCC clinical (0.650, 95% CI 0.643-0.656) and pathologic (0.683, 95% CI 0.676-0.689) staging. Both CPS+EG (IDI 7.2, 95% CI 6.6-7.7%) and Neo-Bioscore (IDI 9.8, 95% CI 8.0-11.6%) demonstrated superior discrimination when compared with AJCC clinical staging at 5 years. Comparison of CPS+EG with Neo-Bioscore yielded an IDI of 2.6% (95% CI 0.9-4.5%), indicating that Neo-Bioscore is the best staging system. CONCLUSIONS: In a heterogenous national cohort of breast cancer patients treated with NAC and surgery, the incorporation of chemotherapy response, tumor grade, ER status, and ERBB2 status into the staging system substantially improved on the AJCC TNM staging system in discrimination of OS. Neo-Bioscore provided the best staging discrimination.

Implications of CA19-9 elevation for survival, staging, and treatment sequencing in intrahepatic cholangiocarcinoma: A national cohort analysis.

BACKGROUND: Optimal management of patients with intrahepatic cholangiocarcinoma (ICCA) and elevated CA19-9 remains undefined. We hypothesized CA19-9 elevation above normal indicates aggressive biology and that inclusion of CA19-9 would improve staging discrimination. METHODS: The National Cancer Data Base (NCDB-2010-2012) was reviewed for patients with ICCA and reported CA19-9. Patients were stratified by CA19-9 above/below normal reference range. Unadjusted Kaplan-Meier and adjusted Cox-proportional-hazards analysis of overall survival (OS) were performed. RESULTS: A total of 2,816 patients were included: 938 (33.3%) normal; 1,878 (66.7%) elevated CA19-9 levels. Demographic/pathologic and chemotherapy/radiation were similar between groups, but patients with elevated CA19-9 had more nodal metastases and less likely to undergo resection. Among elevated-CA19-9 patients, stage-specific survival was decreased in all stages. Resected patients with CA19-9 elevation had similar peri-operative outcomes but decreased long-term survival. In adjusted analysis, CA19-9 elevation independently predicted increased mortality with impact similar to node-positivity, positive-margin resection, and non-receipt of chemotherapy. Proposed staging system including CA19-9 improved survival discrimination over AJCC 7th edition. CONCLUSION: Elevated CA19-9 is an independent risk factor for mortality in ICCA similar in impact to nodal metastases and positive resection margins. Inclusion of CA19-9 in a proposed staging system increases discrimination. Multi-disciplinary therapy should be considered in patients with ICCA and CA19-9 elevation. J. Surg. Oncol. 2016;114:475-482. © 2016 Wiley Periodicals, Inc.

Revolutionizing Postoperative Ileus Monitoring: Exploring GRU-D's Real-Time Capabilities and Cross-Hospital Transferability.

Background: Postoperative ileus (POI) after colorectal surgery leads to increased morbidity, costs, and hospital stays. Identifying POI risk for early intervention is important for improving surgical outcomes especially given the increasing trend towards early discharge after surgery. While existing studies have assessed POI risk with regression models, the role of deep learning's remains unexplored. Methods: We assessed the performance and transferability (brutal force/instance/parameter transfer) of Gated Recurrent Unit with Decay (GRU-D), a longitudinal deep learning architecture, for real-time risk assessment of POI among 7,349 colorectal surgeries performed across three hospital sites operated by Mayo Clinic with two electronic health records (EHR) systems. The results were compared with atemporal models on a panel of benchmark metrics. Results: GRU-D exhibits robust transferability across different EHR systems and hospital sites, showing enhanced performance by integrating new measurements, even amid the extreme sparsity of real-world longitudinal data. On average, for labs, vitals, and assisted living status, 72.2%, 26.9%, and 49.3% respectively lack measurements within 24 hours after surgery. Over the follow-up period with 4-hour intervals, 98.7%, 84%, and 95.8% of data points are missing, respectively. A maximum of 5% decrease in AUROC was observed in brutal-force transfer between different EHR systems with non-overlapping surgery date frames. Multi-source instance transfer witnessed the best performance, with a maximum of 2.6% improvement in AUROC over local learning. The significant benefit, however, lies in the reduction of variance (a maximum of 86% decrease). The GRU-D model's performance mainly depends on the prediction task's difficulty, especially the case prevalence rate. Whereas the impact of training data and transfer strategy is less crucial, underscoring the challenge of effectively leveraging transfer learning for rare outcomes. While atemporal Logit models show notably superior performance at certain pre-surgical points, their performance fluctuate significantly and generally underperform GRU-D in post-surgical hours. Conclusion: GRU-D demonstrated robust transferability across EHR systems and hospital sites with highly sparse real-world EHR data. Further research on built-in explainability for meaningful intervention would be highly valuable for its integration into clinical practice.

Effect of an Artificial Intelligence Decision Support Tool on Palliative Care Referral in Hospitalized Patients: A Randomized Clinical Trial.

CONTEXT: Palliative care services are commonly provided to hospitalized patients, but accurately predicting who needs them remains a challenge. OBJECTIVES: To assess the effectiveness on clinical outcomes of an artificial intelligence (AI)/machine learning (ML) decision support tool for predicting patient need for palliative care services in the hospital. METHODS: The study design was a pragmatic, cluster-randomized, stepped-wedge clinical trial in 12 nursing units at two hospitals over a 15-month period between August 19, 2019, and November 17, 2020. Eligible patients were randomly assigned to either a medical service consultation recommendation triggered by an AI/ML tool predicting the need for palliative care services or usual care. The primary outcome was palliative care consultation note. Secondary outcomes included: hospital readmissions, length of stay, transfer to intensive care and palliative care consultation note by unit. RESULTS: A total of 3183 patient hospitalizations were enrolled. Of eligible patients, A total of 2544 patients were randomized to the decision support tool (1212; 48%) and usual care (1332; 52%). Of these, 1717 patients (67%) were retained for analyses. Patients randomized to the intervention had a statistically significant higher incidence rate of palliative care consultation compared to the control group (IRR, 1.44 [95% CI, 1.11-1.92]). Exploratory evidence suggested that the decision support tool group reduced 60-day and 90-day hospital readmissions (OR, 0.75 [95% CI, 0.57, 0.97]) and (OR, 0.72 [95% CI, 0.55-0.93]) respectively. CONCLUSION: A decision support tool integrated into palliative care practice and leveraging AI/ML demonstrated an increased palliative care consultation rate among hospitalized patients and reductions in hospitalizations.

Risk Prediction Model for 6-Month Mortality for Patients Discharged to Skilled Nursing Facilities.

OBJECTIVE: Hospitalized patients discharged to skilled nursing facilities (SNFs) for post-acute care are at high risk for adverse outcomes. Yet, absence of effective prognostic tools hinders optimal care planning and decision making. Our objective was to develop and validate a risk prediction model for 6-month all-cause death among hospitalized patients discharged to SNFs. DESIGN: Retrospective cohort study. SETTING AND PARTICIPANTS: Patients discharged from 1 of 2 hospitals to 1 of 10 SNFs for post-acute care in an integrated health care delivery system between January 1, 2009, and December 31, 2016. METHODS: Gradient-boosting machine modeling was used to predict all-cause death within 180 days of hospital discharge with use of patient demographic characteristics, comorbidities, pattern of prior health care use, and clinical parameters from the index hospitalization. Area under the receiver operating characteristic curve (AUC) was assessed for out-of-sample observations under 10-fold cross-validation. RESULTS: We identified 9803 unique patients with 11,647 hospital-to-SNF discharges [mean (SD) age, 80.72 (9.71) years; female sex, 61.4%]. These discharges involved 9803 patients alive at 180 days and 1844 patients who died between day 1 and day 180 of discharge. Age, comorbid burden, health care use in prior 6 months, abnormal laboratory parameters, and mobility status during hospital stay were the most important predictors of 6-month death (model AUC, 0.82). CONCLUSION AND IMPLICATIONS: We derived a robust prediction model with parameters available at discharge to SNFs to calculate risk of death within 6 months. This work may be useful to guide other clinicians wishing to develop mortality prediction instruments specific to their post-acute SNF populations.

Surface Estimation, Variable Selection, and the Nonparametric Oracle Property.

Variable selection for multivariate nonparametric regression is an important, yet challenging, problem due, in part, to the infinite dimensionality of the function space. An ideal selection procedure should be automatic, stable, easy to use, and have desirable asymptotic properties. In particular, we define a selection procedure to be nonparametric oracle (np-oracle) if it consistently selects the correct subset of predictors and at the same time estimates the smooth surface at the optimal nonparametric rate, as the sample size goes to infinity. In this paper, we propose a model selection procedure for nonparametric models, and explore the conditions under which the new method enjoys the aforementioned properties. Developed in the framework of smoothing spline ANOVA, our estimator is obtained via solving a regularization problem with a novel adaptive penalty on the sum of functional component norms. Theoretical properties of the new estimator are established. Additionally, numerous simulated and real examples further demonstrate that the new approach substantially outperforms other existing methods in the finite sample setting.

Improving time to palliative care review with predictive modeling in an inpatient adult population: study protocol for a stepped-wedge, pragmatic randomized controlled trial.

BACKGROUND: Palliative care is a medical specialty centered on improving the quality of life (QOL) of patients with complex or life-threatening illnesses. The need for palliative care is increasing and with that the rigorous testing of triage tools that can be used quickly and reliably to identify patients that may benefit from palliative care. METHODS: To that aim, we will conduct a two-armed stepped-wedge cluster randomized trial rolled out to two inpatient hospitals to evaluate whether a machine learning algorithm accurately identifies patients who may benefit from a comprehensive review by a palliative care specialist and decreases time to receiving a palliative care consult in hospital. This is a single-center study which will be conducted from August 2019 to November 2020 at Saint Mary's Hospital & Methodist Hospital both within Mayo Clinic Rochester in Minnesota. Clusters will be nursing units which will be chosen to be a mix of complex patients from Cardiology, Critical Care, and Oncology and had previously established relationships with palliative medicine. The stepped wedge design will have 12 units allocated to a design matrix of 5 treatment wedges. Each wedge will last 75 days resulting in a study period of 12 months of recruitment unless otherwise specified. Data will be analyzed with Bayesian hierarchical models with credible intervals denoting statistical significance. DISCUSSION: This intervention offers a pragmatic approach to delivering specialty palliative care to hospital patients in need using machine learning, thereby leading to high value care and improved outcomes. It is not enough for AI to be utilized by simply publishing research showing predictive performance; clinical trials demonstrating better outcomes are critically needed. Furthermore, the deployment of an AI algorithm is a complex process that requires multiple teams with varying skill sets. To evaluate a deployed AI, a pragmatic clinical trial can accommodate the difficulties of clinical practice while retaining scientific rigor. TRIAL REGISTRATION: ClinicalTrials.gov NCT03976297 . Registered on 6 June 2019, prior to trial start.

Use of a Computerized Algorithm to Evaluate the Proportion and Causes of Potentially Preventable Readmissions Among Patients Discharged to Skilled Nursing Facilities.

OBJECTIVES: Older patients discharged to skilled nursing facilities (SNFs) for post-acute care are at high risk for hospital readmission. Yet, as in the community setting, some readmissions may be preventable with optimal transitional care. This study examined the proportion of 30-day hospital readmissions from SNFs that could be considered potentially preventable readmissions (PPRs) and evaluated the reasons for these readmissions. DESIGN: Retrospective cohort study. SETTING AND PARTICIPANTS: Post-acute practice of an integrated health care delivery system serving 11 SNFs in the US Midwest. Patients discharged from the hospital to an SNF and subsequently readmitted to the hospital within 30 days from January 1, 2009, through November 31, 2016. METHODS: A computerized algorithm evaluated the relationship between initial and repeat hospitalizations to determine whether the repeat hospitalization was a PPR. We assessed for changes in PPR rates across the system over the study period and evaluated the readmission categories to identify the most prevalent PPR categories. RESULTS: Of 11,976 discharges to SNFs for post-acute care among 8041 patients over the study period, 16.6% resulted in rehospitalization within 30 days, and 64.8% of these rehospitalizations were considered PPRs. Annual proportion of PPRs ranged from 58.2% to 66.4% [mean (standard deviation) 0.65 (0.03); 95% confidence interval CI 0.63-0.67; P = .36], with no discernable trend. Nearly one-half (46.2%) of all 30-day readmissions were classified as potentially preventable medical readmissions related to recurrence or continuation of the reason for initial admission or to complications from the initial hospitalization. CONCLUSIONS AND IMPLICATIONS: For this cohort of patients discharged to SNFs, a computerized algorithm categorized a large proportion of 30-day hospital readmissions as potentially preventable, with nearly one-half of those linked to the reason for the initial hospitalization. These findings indicate the importance of improvement in postdischarge transitional care for patients discharged to SNFs.

Challenges of Modeling Outcomes for Surgical Infections: A Word of Caution.

Background: We developed a novel analytic tool for colorectal deep organ/space surgical site infections (C-OSI) prediction utilizing both institutional and extra-institutional American College of Surgeons-National Surgical Quality Improvement Program (ACS-NSQIP) data. Methods: Elective colorectal resections (2006-2014) were included. The primary end point was C-OSI rate. A Bayesian-Probit regression model with multiple imputation (BPMI) via Dirichlet process handled missing data. The baseline model for comparison was a multivariable logistic regression model (generalized linear model; GLM) with indicator parameters for missing data and stepwise variable selection. Out-of-sample performance was evaluated with receiver operating characteristic (ROC) analysis of 10-fold cross-validated samples. Results: Among 2,376 resections, C-OSI rate was 4.6% (n = 108). The BPMI model identified (n = 57; 56% sensitivity) of these patients, when set at a threshold leading to 80% specificity (approximately a 20% false alarm rate). The BPMI model produced an area under the curve (AUC) = 0.78 via 10-fold cross- validation demonstrating high predictive accuracy. In contrast, the traditional GLM approach produced an AUC = 0.71 and a corresponding sensitivity of 0.47 at 80% specificity, both of which were statstically significant differences. In addition, when the model was built utilizing extra-institutional data via inclusion of all (non-Mayo Clinic) patients in ACS-NSQIP, C-OSI prediction was less accurate with AUC = 0.74 and sensitivity of 0.47 (i.e., a 19% relative performance decrease) when applied to patients at our institution. Conclusions: Although the statistical methodology associated with the BPMI model provides advantages over conventional handling of missing data, the tool should be built with data specific to the individual institution to optimize performance.

Improving the delivery of palliative care through predictive modeling and healthcare informatics.

OBJECTIVE: Access to palliative care (PC) is important for many patients with uncontrolled symptom burden from serious or complex illness. However, many patients who could benefit from PC do not receive it early enough or at all. We sought to address this problem by building a predictive model into a comprehensive clinical framework with the aims to (i) identify in-hospital patients likely to benefit from a PC consult, and (ii) intervene on such patients by contacting their care team. MATERIALS AND METHODS: Electronic health record data for 68 349 inpatient encounters in 2017 at a large hospital were used to train a model to predict the need for PC consult. This model was published as a web service, connected to institutional data pipelines, and consumed by a downstream display application monitored by the PC team. For those patients that the PC team deems appropriate, a team member then contacts the patient's corresponding care team. RESULTS: Training performance AUC based on a 20% holdout validation set was 0.90. The most influential variables were previous palliative care, hospital unit, Albumin, Troponin, and metastatic cancer. The model has been successfully integrated into the clinical workflow making real-time predictions on hundreds of patients per day. The model had an "in-production" AUC of 0.91. A clinical trial is currently underway to assess the effect on clinical outcomes. CONCLUSIONS: A machine learning model can effectively predict the need for an inpatient PC consult and has been successfully integrated into practice to refer new patients to PC.