A GPT-based EHR modeling system for unsupervised novel disease detection.

OBJECTIVE: To develop an Artificial Intelligence (AI)-based anomaly detection model as a complement of an "astute physician" in detecting novel disease cases in a hospital and preventing emerging outbreaks. METHODS: Data included hospitalized patients (n = 120,714) at a safety-net hospital in Massachusetts. A novel Generative Pre-trained Transformer (GPT)-based clinical anomaly detection system was designed and further trained using Empirical Risk Minimization (ERM), which can model a hospitalized patient's Electronic Health Records (EHR) and detect atypical patients. Methods and performance metrics, similar to the ones behind the recent Large Language Models (LLMs), were leveraged to capture the dynamic evolution of the patient's clinical variables and compute an Out-Of-Distribution (OOD) anomaly score. RESULTS: In a completely unsupervised setting, hospitalizations for Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection could have been predicted by our GPT model at the beginning of the COVID-19 pandemic, with an Area Under the Receiver Operating Characteristic Curve (AUC) of 92.2 %, using 31 extracted clinical variables and a 3-day detection window. Our GPT achieves individual patient-level anomaly detection and mortality prediction AUC of 78.3 % and 94.7 %, outperforming traditional linear models by 6.6 % and 9 %, respectively. Different types of clinical trajectories of a SARS-CoV-2 infection are captured by our model to make interpretable detections, while a trend of over-pessimistic outcome prediction yields a more effective detection pathway. Furthermore, our comprehensive GPT model can potentially assist clinicians with forecasting patient clinical variables and developing personalized treatment plans. CONCLUSION: This study demonstrates that an emerging outbreak can be accurately detected within a hospital, by using a GPT to model patient EHR time sequences and labeling them as anomalous when actual outcomes are not supported by the model. Such a GPT is also a comprehensive model with the functionality of generating future patient clinical variables, which can potentially assist clinicians in developing personalized treatment plans.

Mandatory Child Protective Services Reporting for Substance-Exposed Newborns and Peripartum Outcomes: A Difference-in-Differences Analysis.

This cross-sectional study examines data across 17 birthing hospitals before and after a policy change at Boston Medical Center in how reporting decisions are made in cases of prenatal substance exposure.

A Novel Machine Learning Algorithm for Creating Risk-Adjusted Payment Formulas.

Importance: Models predicting health care spending and other outcomes from administrative records are widely used to manage and pay for health care, despite well-documented deficiencies. New methods are needed that can incorporate more than 70 000 diagnoses without creating undesirable coding incentives. Objective: To develop a machine learning (ML) algorithm, building on Diagnostic Item (DXI) categories and Diagnostic Cost Group (DCG) methods, that automates development of clinically credible and transparent predictive models for policymakers and clinicians. Design, Setting, and Participants: DXIs were organized into disease hierarchies and assigned an Appropriateness to Include (ATI) score to reflect vagueness and gameability concerns. A novel automated DCG algorithm iteratively assigned DXIs in 1 or more disease hierarchies to DCGs, identifying sets of DXIs with the largest regression coefficient as dominant; presence of a previously identified dominating DXI removed lower-ranked ones before the next iteration. The Merative MarketScan Commercial Claims and Encounters Database for commercial health insurance enrollees 64 years and younger was used. Data from January 2016 through December 2018 were randomly split 90% to 10% for model development and validation, respectively. Deidentified claims and enrollment data were delivered by Merative the following November in each calendar year and analyzed from November 2020 to January 2024. Main Outcome and Measures: Concurrent top-coded total health care cost. Model performance was assessed using validation sample weighted least-squares regression, mean absolute errors, and mean errors for rare and common diagnoses. Results: This study included 35 245 586 commercial health insurance enrollees 64 years and younger (65 901 460 person-years) and relied on 19 clinicians who provided reviews in the base model. The algorithm implemented 218 clinician-specified hierarchies compared with the US Department of Health and Human Services (HHS) hierarchical condition category (HCC) model's 64 hierarchies. The base model that dropped vague and gameable DXIs reduced the number of parameters by 80% (1624 of 3150), achieved an R2 of 0.535, and kept mean predicted spending within 12% ($3843 of $31 313) of actual spending for the 3% of people with rare diseases. In contrast, the HHS HCC model had an R2 of 0.428 and underpaid this group by 33% ($10 354 of $31 313). Conclusions and Relevance: In this study, by automating DXI clustering within clinically specified hierarchies, this algorithm built clinically interpretable risk models in large datasets while addressing diagnostic vagueness and gameability concerns.

Metabolic Disease and The Risk of Post-COVID Conditions: A Retrospective Cohort Study.

Objective: To examine the influence of having a baseline metabolic disorder (diabetes, hypertension, and/or obesity) on the risk of developing new clinical sequelae potentially related to SARS-CoV-2 in a large sample of commercially insured adults in the US. Design setting and participants: Deidentified data were collected from the IBM/Watson MarketScan Commercial Claims and Encounters (CCAE) Databases and Medicare Supplemental and Coordination of Benefits (MDCR) Databases from 2019 to 2021. A total of 839,344 adults aged 18 and above with continuous enrollment in the health plan were included in the analyses. Participants were grouped into four categories based on their COVID-19 diagnosis and whether they had at least one of the three common metabolic disorders at baseline (diabetes, obesity, or hypertension). Measures and methods: ICD-10-CM codes were used to determine new symptoms and conditions after the acute phase of SARS-CoV-2 infection, defined as ending 21 days after initial diagnosis date, or index period for those who did not have a COVID-19 diagnosis. Propensity score matching (PSM) was used to create comparable reference groups. Cox proportional hazard models were conducted to estimate hazard ratios and 95% confidence intervals. Results: Among the 772,377 individuals included in the analyses, 36,742 (4.8%) without and 20,912 (2.7%) with a baseline metabolic disorder were diagnosed with COVID-19. On average, COVID-19 patients with baseline metabolic disorders had more 2.4 more baseline comorbidities compared to those without baseline metabolic disorders. Compared to adults with no baseline metabolic condition, the risks of developing new clinical sequelae were highest among COVID-19 patients with a baseline metabolic condition (HRs ranging from 1.51 to 3.33), followed by those who had a baseline metabolic condition but with no COVID-19 infection (HRs ranging from 1.33 to 2.35), and those who had COVID-19 but no baseline metabolic condition (HRs ranging from 1.34 to 2.85). Conclusions: In a large national cohort of commercially insured adults, COVID-19 patients with a baseline metabolic condition had the highest risk of developing new clinical sequelae post-acute infection phase, followed by those who had baseline metabolic condition but no COVID-19 infection and those who had COVID-19 but no baseline metabolic disorder.

Target trial emulation for comparative effectiveness research with observational data: Promise and challenges for studying medications for opioid use disorder.

Medications for opioid use disorder (MOUD) increase retention in care and decrease mortality during active treatment; however, information about the comparative effectiveness of different forms of MOUD is sparse. Observational comparative effectiveness studies are subject to many types of bias; a robust framework to minimize bias would improve the quality of comparative effectiveness evidence. This paper discusses the use of target trial emulation as a framework to conduct comparative effectiveness studies of MOUD with administrative data. Using examples from our planned research project comparing buprenorphine-naloxone and extended-release naltrexone with respect to the rates of MOUD discontinuation, we provide a primer on the challenges and approaches to employing target trial emulation in the study of MOUD.

Food Pantry Referral and Utilization in a Pediatric Primary Care Clinic.

INTRODUCTION: This study aimed to characterize progression from screening for food insecurity risk to on-site food pantry referral to food pantry utilization in pediatric primary care. METHODS: This retrospective study included 14,280 patients aged 0-21 years with ≥1 pediatric primary care visit from March 2018 to February 2020. Analyses were conducted in 2020-2022 using multivariable regression to examine patient-level demographic, clinical, and socioeconomic characteristics and systems-related factors associated with progression from screening positive for food insecurity risk to food pantry referral to completing ≥1 food pantry visit. RESULTS: Of patients screened for food insecurity risk, 31.9% screened positive; 18.5% of food-insecure patients received an on-site food pantry referral. Among patients referred, 28.9% visited the food pantry. In multivariable models, higher odds of referral were found for patients living near the clinic (AOR=1.28; 95% CI=1.03, 1.59), for each additional health-related social need reported (AOR=1.23; 95% CI=1.16, 1.29), and when the index clinic encounter occurred during food pantry open hours (AOR=1.62; 95% CI=1.30, 2.02). Higher odds of food pantry visitation were found for patients with a preferred language of Haitian Creole (AOR=2.16; 95% CI=1.37, 3.39), for patients of Hispanic race/ethnicity (AOR=3.67; 95% CI=1.14, 11.78), when the index encounter occurred during food pantry open hours (AOR=1.96; 95% CI=1.25, 3.07), for patients with a clinician letter referral (AOR=6.74; 95% CI=3.94, 11.54), or for patients with a referral due to a screening-identified food emergency (AOR=2.27; 95% CI=1.30, 3.96). CONCLUSIONS: There was substantial attrition along the pathway from screening positive for food insecurity risk to food pantry referral and utilization as well as patient-level characteristics and systems-related factors associated with successful referrals and utilization.

Value-Based Purchasing Design And Effect: A Systematic Review And Analysis.

During the past two decades in the United States, all major payer types-commercial, Medicare, Medicaid, and multipayer coalitions-have introduced value-based purchasing (VBP) contracts to reward providers for improving health care quality while reducing spending. This systematic review qualitatively characterized the financial and nonfinancial features of VBP programs and examined how such features combine to create a level of program intensity that relates to desired quality and spending outcomes. Higher-intensity VBP programs are more frequently associated with desired quality processes, utilization measures, and spending reductions than lower-intensity programs. Thus, although there may be reasons for payers and providers to opt for lower-intensity programs (for example, to increase voluntary participation), these choices apparently have consequences for spending and quality outcomes.

Structural Racism in Behavioral Health Presentation and Management.

Nia is a first-grade student with a history of trauma who was brought in by ambulance to the pediatric emergency department for "out of control behavior" at school. This is the first of multiple presentations to the emergency department for psychiatric evaluation, stabilization, and management throughout her elementary and middle school years. Several of the visits resulted in admission to the inpatient pediatric service, where she "boarded" while awaiting transfer to an inpatient psychiatric facility. At times, clinical teams used involuntary emergency medications and physical restraints, as well as hospital security presence at the bedside, to control Nia's behavior. Nia is Black and her story is a case study of how structural racism manifests for an individual child. Her story highlights the impact of adultification bias and the propensity to mislabel Black youth with diagnoses characterized by fixed patterns of negative behaviors, as opposed to recognizing normative reactions to trauma or other adverse childhood experiences-in Nia's case, poverty, domestic violence, and Child Protective Services involvement. In telling Nia's story, we (1) define racism and discuss the interplay of structural, institutional, and interpersonal racism in the health care, education, and judicial systems; (2) highlight the impact of adultification bias on Black youth; (3) delineate racial disparities in behavioral health diagnosis and management, school discipline and exclusion, and health care's contributions to the school-to-prison pipeline; and finally (4) propose action steps to mitigate the impact of racism on pediatric mental health and health care.

Pediatric surgical site infections in 287 hospitals in the United States, 2015-2018.

Among 287 US hospitals reporting data between 2015 and 2018, annual pediatric surgical site infection (SSI) rates ranged from 0% for gallbladder to 10.4% for colon surgeries. Colon, spinal fusion, and small-bowel SSI rates did not decrease with greater surgical volumes in contrast to appendix and ventricular-shunt SSI rates.

Development and Assessment of a New Framework for Disease Surveillance, Prediction, and Risk Adjustment: The Diagnostic Items Classification System.

Importance: Current disease risk-adjustment formulas in the US rely on diagnostic classification frameworks that predate the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). Objective: To develop an ICD-10-CM-based classification framework for predicting diverse health care payment, quality, and performance outcomes. Design Setting and Participants: Physician teams mapped all ICD-10-CM diagnoses into 3 types of diagnostic items (DXIs): main effect DXIs that specify diseases; modifiers, such as laterality, timing, and acuity; and scaled variables, such as body mass index, gestational age, and birth weight. Every diagnosis was mapped to at least 1 DXI. Stepwise and weighted least-squares estimation predicted cost and utilization outcomes, and their performance was compared with models built on (1) the Agency for Healthcare Research and Quality Clinical Classifications Software Refined (CCSR) categories, and (2) the Health and Human Services Hierarchical Condition Categories (HHS-HCC) used in the Affordable Care Act Marketplace. Each model's performance was validated using R 2, mean absolute error, the Cumming prediction measure, and comparisons of actual to predicted outcomes by spending percentiles and by diagnostic frequency. The IBM MarketScan Commercial Claims and Encounters Database, 2016 to 2018, was used, which included privately insured, full- or partial-year eligible enrollees aged 0 to 64 years in plans with medical, drug, and mental health/substance use coverage. Main Outcomes and Measures: Fourteen concurrent outcomes were predicted: overall and plan-paid health care spending (top-coded and not top-coded); enrollee out-of-pocket spending; hospital days and admissions; emergency department visits; and spending for 6 types of services. The primary outcome was annual health care spending top-coded at $250 000. Results: A total of 65 901 460 person-years were split into 90% estimation/10% validation samples (n = 6 604 259). In all, 3223 DXIs were created: 2435 main effects, 772 modifiers, and 16 scaled items. Stepwise regressions predicting annual health care spending (mean [SD], $5821 [$17 653]) selected 76% of the main effect DXIs with no evidence of overfitting. Validated R 2 was 0.589 in the DXI model, 0.539 for CCSR, and 0.428 for HHS-HCC. Use of DXIs reduced underpayment for enrollees with rare (1-in-a-million) diagnoses by 83% relative to HHS-HCCs. Conclusions: In this diagnostic modeling study, the new DXI classification system showed improved predictions over existing diagnostic classification systems for all spending and utilization outcomes considered.

Risk Factors for Admission Within a Hospital-Based COVID-19 Home Monitoring Program.

Background: Despite increasing vaccination rates, coronavirus disease 2019 (COVID-19) continues to overwhelm heath systems worldwide. Few studies follow outpatients diagnosed with COVID-19 to understand risks for subsequent admissions. We sought to identify hospital admission risk factors in individuals with COVID-19 to guide outpatient follow-up and prioritization for novel therapeutics. Methods: We prospectively designed data collection templates and remotely monitored patients after a COVID-19 diagnosis, then retrospectively analyzed data to identify risk factors for 30-day admission for those initially managed outpatient and for 30-day re-admissions for those monitored after an initial COVID-19 admission. We included all patients followed by our COVID-19 follow-up monitoring program from April 2020 to February 2021. Results: Among 4070 individuals followed by the program, older age (adjusted odds ratio [aOR], 1.05; 95% CI, 1.03-1.06), multiple comorbidities (1-2: aOR, 5.88; 95% CI, 2.07-16.72; ≥3: aOR, 20.40; 95% CI, 7.23-57.54), presence of fever (aOR, 2.70; 95% CI, 1.65-4.42), respiratory symptoms (aOR, 2.46; 95% CI, 1.53-3.94), and gastrointestinal symptoms (aOR, 2.19; 95% CI, 1.53-3.94) at initial contact were associated with increased risk of COVID-19-related 30-day admission among those initially managed outpatient. Loss of taste/smell was associated with decreased admission risk (aOR, 0.46; 95% CI, 0.25-0.85). For postdischarge patients, older age was also associated with increased re-admission risk (aOR, 1.04; 95% CI, 1.01-1.06). Conclusions: This study reveals that in addition to older age and specific comorbidities, the number of high-risk conditions, fever, respiratory symptoms, and gastrointestinal symptoms at diagnosis all increased odds of COVID-19-related admission. These data could enhance patient prioritization for early treatment interventions and ongoing surveillance.

Development and validation of predictive models for COVID-19 outcomes in a safety-net hospital population.

OBJECTIVE: To develop predictive models of coronavirus disease 2019 (COVID-19) outcomes, elucidate the influence of socioeconomic factors, and assess algorithmic racial fairness using a racially diverse patient population with high social needs. MATERIALS AND METHODS: Data included 7,102 patients with positive (RT-PCR) severe acute respiratory syndrome coronavirus 2 test at a safety-net system in Massachusetts. Linear and nonlinear classification methods were applied. A score based on a recurrent neural network and a transformer architecture was developed to capture the dynamic evolution of vital signs. Combined with patient characteristics, clinical variables, and hospital occupancy measures, this dynamic vital score was used to train predictive models. RESULTS: Hospitalizations can be predicted with an area under the receiver-operating characteristic curve (AUC) of 92% using symptoms, hospital occupancy, and patient characteristics, including social determinants of health. Parsimonious models to predict intensive care, mechanical ventilation, and mortality that used the most recent labs and vitals exhibited AUCs of 92.7%, 91.2%, and 94%, respectively. Early predictive models, using labs and vital signs closer to admission had AUCs of 81.1%, 84.9%, and 92%, respectively. DISCUSSION: The most accurate models exhibit racial bias, being more likely to falsely predict that Black patients will be hospitalized. Models that are only based on the dynamic vital score exhibited accuracies close to the best parsimonious models, although the latter also used laboratories. CONCLUSIONS: This large study demonstrates that COVID-19 severity may accurately be predicted using a score that accounts for the dynamic evolution of vital signs. Further, race, social determinants of health, and hospital occupancy play an important role.

Projected Long-Term Impact of the Coronavirus Disease 2019 (COVID-19) Pandemic on Hepatitis C Outcomes in the United States: A Modeling Study.

BACKGROUND: The coronavirus disease 2019 (COVID-19) pandemic disrupted access to and uptake of hepatitis C virus (HCV) care services in the United States. It is unknown how substantially the pandemic will impact long-term HCV-related outcomes. METHODS: We used a microsimulation to estimate the 10-year impact of COVID-19 disruptions in healthcare delivery on HCV outcomes including identified infections, linkage to care, treatment initiation and completion, cirrhosis, and liver-related death. We modeled hypothetical scenarios consisting of an 18-month pandemic-related disruption in HCV care starting in March 2020 followed by varying returns to pre-pandemic rates of screening, linkage, and treatment through March 2030 and compared them to a counterfactual scenario in which there was no COVID-19 pandemic or disruptions in care. We also performed alternate scenario analyses in which the pandemic disruption lasted for 12 and 24 months. RESULTS: Compared to the "no pandemic" scenario, in the scenario in which there is no return to pre-pandemic levels of HCV care delivery, we estimate 1060 fewer identified cases, 21 additional cases of cirrhosis, and 16 additional liver-related deaths per 100 000 people. Only 3% of identified cases initiate treatment and <1% achieve sustained virologic response (SVR). Compared to "no pandemic," the best-case scenario in which an 18-month care disruption is followed by a return to pre-pandemic levels, we estimated a smaller proportion of infections identified and achieving SVR. CONCLUSIONS: A recommitment to the HCV epidemic in the United States that involves additional resources coupled with aggressive efforts to screen, link, and treat people with HCV is needed to overcome the COVID-19-related disruptions.

Assessment of Federal Value-Based Incentive Programs and In-Hospital Clostridioides difficile Infection Rates.

Importance: Health care facility-onset Clostridioides difficile infection (HO-CDI) rates reported to the US Centers for Disease Control and Prevention's National Healthcare Safety Network (NHSN) became a target quality metric for 2 Centers for Medicare & Medicaid Services (CMS) value-based incentive programs (VBIPs) in October 2016. The association of VBIPs with HO-CDI rates is unknown. Objective: To examine the association between VBIP implementation and HO-CDI rates. Design, Setting, and Participants: This interrupted time series study evaluated HO-CDI rates among adults hospitalized from January 2013 to March 2019 at 265 acute-care hospitals. Interventions: Implementation of VBIPs in October 2016. Main Outcomes and Measures: Quarterly rates of HO-CDI per 10 000 patient-days, as reported to NHSN by participating hospitals, were evaluated. Generalized estimating equations were used to fit negative binomial regression models to estimate immediate program effect size (ie, level change) and changes in the slope of HO-CDI rates, controlling for each hospital's predominant method of CDI testing (ie, nucleic acid amplification test [NAAT], enzyme immunoassay [EIA] for toxin, or other testing methods). Results: The study cohort included 24 332 938 admissions, 109 371 136 patient-days, and 74 681 HO-CDI events at 265 hospitals (145 [55%] with 100-399 beds; 205 [77%] not-for-profit hospitals; 185 [70%] teaching hospitals; 229 [86%] in metropolitan areas). Compared with EIA, rates of HO-CDI were higher when detected by NAAT (adjusted incidence rate ratio [aIRR], 1.55; 95% CI, 1.40-1.70; P < .001) and other testing methods (aIRR, 1.47; 95% CI, 1.26-1.71; P < .001). There were no significant changes in testing methods used by hospitals immediately after VBIP implementation. Controlling for CDI testing method, VBIP implementation was associated with a 6% level decline in HO-CDI rates in the immediate postpolicy quarter (aIRR, 0.94; 95% CI, 0.89-0.99; P = .01) and a 4% decline in slope per quarter (aIRR, 0.96; 95% CI, 0.95-0.97; P < .001). Results were similar in a sensitivity analysis using a 1-year roll-in period accounting for the period after the announcement of the HO-CDI VBIP policy and prior to its implementation. Conclusions and Relevance: In this study, VBIP implementation was associated with improvements in HO-CDI rates, independent of CDI testing method. Given that CMS payment policies have not previously been associated with improvements in other targeted health care-associated infection rates, future research should focus on elucidating the specific processes that contributed to improvement in HO-CDI rates to inform the design of future VBIP interventions.

Time to SARS-CoV-2 PCR Clearance in Immunocompromising Conditions: Is Test-Based Removal From Isolation Necessary in Severely Immunocompromised Individuals?

To determine the association between immunosuppression and time to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) polymerase chain reaction (PCR) clearance, we studied 3758 adults retested following initial SARS-CoV-2 infection. Cox proportional hazards models demonstrated delayed PCR clearance with older age, multiple comorbidities, and solid organ transplant but not by degree of immunocompromise. These findings challenge current retesting practices.