Guidelines for Data and Safety Monitoring in Pragmatic Randomized Clinical Trials Using Case Studies.

Pragmatic randomized clinical trials (pRCTs) have a unique set of considerations for data and safety monitoring. Because of their unconventional trial designs coupled with collection of multilevel data and implementation outcomes in real-world settings, thoughtful consideration is needed on the presentation of the trial design and accruing data to facilitate review and decision-making by the trial's data and safety monitoring board (DSMB). To our knowledge, there is limited information available in practical guidelines for generalists and medical general practitioners on what to monitor and to report to the DSMB during the conduct of pRCTs and what the DSMB should focus on in its review of reports. This article discusses these matters in the context of 3 case studies focusing on a set of critical data and safety monitoring questions that would be of interest to the generalist conducting pRCTs. In considering these questions, we provide tabular and graphical illustrations of how data can be presented to the DSMB while drawing attention to those areas that the DSMB should focus on in its review of the trial. The strategies and viewpoints discussed herein provide practical guidelines and can serve as a resource for the generalist conducting pRCTs.

Patient Satisfaction With a Multisite, Multiregional Remote Patient Monitoring Program for Acute and Chronic Condition Management: Survey-Based Analysis.

BACKGROUND: Remote patient monitoring (RPM) is an option for continuously managing the care of patients in the comfort of their homes or locations outside hospitals and clinics. Patient engagement with RPM programs is essential for achieving successful outcomes and high quality of care. When relying on technology to facilitate monitoring and shifting disease management to the home environment, it is important to understand the patients' experiences to enable quality improvement. OBJECTIVE: This study aimed to describe patients' experiences and overall satisfaction with an RPM program for acute and chronic conditions in a multisite, multiregional health care system. METHODS: Between January 1, 2021, and August 31, 2022, a patient experience survey was delivered via email to all patients enrolled in the RPM program. The survey encompassed 19 questions across 4 categories regarding comfort, equipment, communication, and overall experience, as well as 2 open-ended questions. Descriptive analysis of the survey response data was performed using frequency distribution and percentages. RESULTS: Surveys were sent to 8535 patients. The survey response rate was 37.16% (3172/8535) and the completion rate was 95.23% (3172/3331). Survey results indicated that 88.97% (2783/3128) of participants agreed or strongly agreed that the program helped them feel comfortable managing their health from home. Furthermore, 93.58% (2873/3070) were satisfied with the RPM program and ready to graduate when meeting the program goals. In addition, patient confidence in this model of care was confirmed by 92.76% (2846/3068) of the participants who would recommend RPM to people with similar conditions. There were no differences in ease of technology use according to age. Those with high school or less education were more likely to agree that the equipment and educational materials helped them feel more informed about their care plans than those with higher education levels. CONCLUSIONS: This multisite, multiregional RPM program has become a reliable health care delivery model for the management of acute and chronic conditions outside hospitals and clinics. Program participants reported an excellent overall experience and a high level of satisfaction in managing their health from the comfort of their home environment.

Identifying treatment heterogeneity in atrial fibrillation using a novel causal machine learning method.

BACKGROUND: Lifelong oral anticoagulation is recommended in patients with atrial fibrillation (AF) to prevent stroke. Over the last decade, multiple new oral anticoagulants (OACs) have expanded the number of treatment options for these patients. While population-level effectiveness of OACs has been compared, it is unclear if there is variability in benefit and risk across patient subgroups. METHODS: We analyzed claims and medical data for 34,569 patients who initiated a nonvitamin K antagonist oral anticoagulant (non-vitamin K antagonist oral anticoagulant (NOAC); apixaban, dabigatran, and rivaroxaban) or warfarin for nonvalvular AF between 08/01/2010 and 11/29/2017 from the OptumLabs Data Warehouse. A machine learning (ML) method was applied to match different OAC groups on several baseline variables including, age, sex, race, renal function, and CHA2DS2 -VASC score. A causal ML method was then used to discover patient subgroups characterizing the head-to-head treatment effects of the OACs on a primary composite outcome of ischemic stroke, intracranial hemorrhage, and all-cause mortality. RESULTS: The mean age, number of females and white race in the entire cohort of 34,569 patients were 71.2 (SD, 10.7) years, 14,916 (43.1%), and 25,051 (72.5%) respectively. During a mean follow-up of 8.3 (SD, 9.0) months, 2,110 (6.1%) of patients experienced the composite outcome, of whom 1,675 (4.8%) died. The causal ML method identified 5 subgroups with variables favoring apixaban over dabigatran; 2 subgroups favoring apixaban over rivaroxaban; 1 subgroup favoring dabigatran over rivaroxaban; and 1 subgroup favoring rivaroxaban over dabigatran in terms of risk reduction of the primary endpoint. No subgroup favored warfarin and most dabigatran vs warfarin users favored neither drug. The variables that most influenced favoring one subgroup over another included Age, history of ischemic stroke, thromboembolism, estimated glomerular filtration rate, Race, and myocardial infarction. CONCLUSIONS: Among patients with AF treated with a NOAC or warfarin, a causal ML method identified patient subgroups with differences in outcomes associated with OAC use. The findings suggest that the effects of OACs are heterogeneous across subgroups of AF patients, which could help personalize the choice of OAC. Future prospective studies are needed to better understand the clinical impact of the subgroups with respect to OAC selection.

A prediction model for asthma exacerbations after stopping asthma biologics.

BACKGROUND: Little is known regarding the prediction of the risks of asthma exacerbation after stopping asthma biologics. OBJECTIVE: To develop and validate a predictive model for the risk of asthma exacerbations after stopping asthma biologics using machine learning models. METHODS: We identified 3057 people with asthma who stopped asthma biologics in the OptumLabs Database Warehouse and considered a wide range of demographic and clinical risk factors to predict subsequent outcomes. The primary outcome used to assess success after stopping was having no exacerbations in the 6 months after stopping the biologic. Elastic-net logistic regression (GLMnet), random forest, and gradient boosting machine models were used with 10-fold cross-validation within a development (80%) cohort and validation cohort (20%). RESULTS: The mean age of the total cohort was 47.1 (SD, 17.1) years, 1859 (60.8%) were women, 2261 (74.0%) were White, and 1475 (48.3%) were in the Southern region of the United States. The elastic-net logistic regression model yielded an area under the curve (AUC) of 0.75 (95% confidence interval [CI], 0.71-0.78) in the development and an AUC of 0.72 in the validation cohort. The random forest model yielded an AUC of 0.75 (95% CI, 0.68-0.79) in the development cohort and an AUC of 0.72 in the validation cohort. The gradient boosting machine model yielded an AUC of 0.76 (95% CI, 0.72-0.80) in the development cohort and an AUC of 0.74 in the validation cohort. CONCLUSION: Outcomes after stopping asthma biologics can be predicted with moderate accuracy using machine learning methods.

Clinician Adoption of an Artificial Intelligence Algorithm to Detect Left Ventricular Systolic Dysfunction in Primary Care.

OBJECTIVE: To compare the clinicians' characteristics of "high adopters" and "low adopters" of an artificial intelligence (AI)-enabled electrocardiogram (ECG) algorithm that alerted for possible low left ventricular ejection fraction (EF) and the subsequent effectiveness of detecting patients with low EF. METHODS: Clinicians in 48 practice sites of a US Midwest health system were cluster-randomized by the care team to usual care or to receive a notification that suggested ordering an echocardiogram in patients flagged as potentially having low EF based on an AI-ECG algorithm. Enrollment was between June 26, 2019, and July 30, 2019; participation concluded on March 31, 2020. This report is focused on those clinicians randomized to receive the notification of the AI-ECG algorithm. At the patient level, data were analyzed for the proportion of patients with positive AI-ECG results. Adoption was defined as the clinician order of an echocardiogram after prompted by the alert. RESULTS: A total of 165 clinicians and 11,573 patients were included in this analysis. Among patients with positive AI-ECG, high adopters (n=41) were twice as likely to diagnose patients with low EF (33.9%) vs low adopters, n=124, (16.9%); odds ratio, 1.62; 95% CI, 1.21 to 2.17). High adopters were more often advanced practice providers (eg, nurse practitioners and physician assistants) vs physicians, Family Medicine vs Internal Medicine specialty, and tended to have less complex patients. CONCLUSION: Clinicians who most frequently followed the recommendations of an AI tool were twice as likely to diagnose low EF. Those clinicians with less complex patients were more likely to be high adopters. TRIAL REGISTRATION: Clinicaltrials.gov Identifier: NCT04000087.

Time to first treatment is an independent prognostic factor for Multiple Myeloma (MM).

INTRODUCTION: Multiple myeloma (MM) is an incurable plasma cell neoplasm. In this study, we aimed to analyze the impact of time to initiation of systemic therapy for MM on overall survival (OS). METHODS: We identified cases diagnosed with MM from the National Cancer Database from 2004 to 2013. RESULTS: A total of 38,178 MM patients were included in the analysis. The median time to systemic therapy in our cohort was 17 days (range 0-120). The median OS for patients who initiated therapy > 30-days after diagnosis was longer than those who received it ≤ 7 days (46 vs. 27-month, p < 0.001). On multivariable analysis, patients who received treatment ≤ 7 days from diagnosis had worse mortality compared with those receiving treatment > 30 days (HR 1.5; 95% CI 1.4-1.6). CONCLUSIONS: In our study, time to initiation of systemic therapy was an independent prognostic factor in MM. Similar to other lymphoid malignancies, this metric may be a surrogate for high-risk disease in MM, and future trials may need to investigate time-to-treatment as a factor to allow enrollment of potentially sick patients.

Pilot implementation to assess the feasibility and care team impact of an app-based interactive care plan to remotely monitor breast cancer survivors.

PURPOSE: To assess the feasibility of an app-based, electronic health record (EHR)-integrated, interactive care plan (ICP) for breast cancer (BC) survivors. METHODS: A single-arm pilot study was conducted with female BC survivors. ICP tasks included quarterly quality of life (QOL) questionnaire; monthly assessments of fatigue, insomnia, sexual dysfunction, hot flashes, and recurrence symptoms; and daily activity reminders. Embedded decision trees escalated recurrence symptoms to providers. On-demand education was available for self-management of treatment-related toxicities. The primary objective was to assess patients' engagement with ICP tasks against feasibility thresholds of 75% completion rate. Secondary objectives were evaluation of the system's functionality to track and escalate symptoms appropriately, and care team impact measured by volume of escalation messages generated. We report preliminary results 6 months after the last patient enrolled. RESULTS: Twenty-three patients enrolled August to November 2020. Mean age was 50.1 years. All patients engaged with at least one ICP task. The monthly average task completion rates were 62% for the QOL questionnaire, 59% for symptom assessments, and 37% for activity reminders. Task completion rate decreased over time. Eleven of 253 symptoms and QOL questionnaires (4.3%) generated messages for care escalation. CONCLUSION: Implementation of an app-based, EHR-integrated ICP in BC survivors was feasible and created minimal provider burden; however, patient engagement was below the feasibility threshold suggesting that changes may enhance broad implementation and adoption. IMPLICATIONS FOR CANCER SURVIVORS: An ICP may facilitate remote monitoring, symptom control, and recurrence surveillance for cancer survivors as strategies to enhance patient engagement are applied.

Association of Area-Level Socioeconomic Deprivation With Hypoglycemic and Hyperglycemic Crises in US Adults With Diabetes.

Importance: Social determinants of health play a role in diabetes management and outcomes, including potentially life-threatening complications of severe hypoglycemia and diabetic ketoacidosis (DKA) or hyperglycemic hyperosmolar state (HHS). Although several person-level socioeconomic factors have been associated with these complications, the implications of area-level socioeconomic deprivation are unknown. Objective: To examine the association between area-level deprivation and the risks of experiencing emergency department visits or hospitalizations for hypoglycemic and hyperglycemic crises (ie, DKA or HHS). Design, Setting, and Participants: This cohort study used deidentified administrative claims data for privately insured individuals and Medicare Advantage beneficiaries across the US. The analysis included adults with diabetes who met the claims criteria for diabetes between January 1, 2016, and December 31, 2017. Data analyses were performed from November 17, 2020, to November 11, 2021. Exposures: Area deprivation index (ADI) was derived for each county for 2016 and 2017 using 17 county-level indicators from the American Community Survey. ADI values were applied to patients who were living in each county based on their index dates and were categorized according to county-level ADI quintile (with quintile 1 having the least deprivation and quintile 5 having the most deprivation). Main Outcomes and Measures: The numbers of emergency department visits or hospitalizations related to the primary diagnoses of hypoglycemia and DKA or HHS (ascertained using validated diagnosis codes in the first or primary position of emergency department or hospital claims) between 2016 and 2019 were calculated for each ADI quintile using negative binomial regression models and adjusted for patient age, sex, health plan type, comorbidities, glucose-lowering medication type, and percentage of White residents in the county. Results: The study population included 1 116 361 individuals (563 943 women [50.5%]), with a mean (SD) age of 64.9 (13.2) years. Of these patients, 343 726 (30.8%) resided in counties with the least deprivation (quintile 1) and 121 810 (10.9%) lived in counties with the most deprivation (quintile 5). Adjusted rates of severe hypoglycemia increased from 13.54 (95% CI, 12.91-14.17) per 1000 person-years in quintile 1 counties to 19.13 (95% CI, 17.62-20.63) per 1000 person-years in quintile 5 counties, corresponding to an incidence rate ratio of 1.41 (95% CI, 1.29-1.54; P < .001). Adjusted rates of DKA or HHS increased from 7.49 (95% CI, 6.96-8.02) per 1000 person-years in quintile 1 counties to 8.37 (95% CI, 7.50-9.23) per 1000 person-years in quintile 5 counties, corresponding to an incidence rate ratio of 1.12 (95% CI, 1.00-1.25; P = .049). Conclusions and Relevance: This study found that living in counties with a high area-level deprivation was associated with an increased risk of severe hypoglycemia and DKA or HHS. The concentration of these preventable events in areas of high deprivation signals the need for interventions that target the structural barriers to optimal diabetes management and health.

Making sense of diabetes medication decisions: a mixed methods cluster randomized trial using a conversation aid intervention.

PURPOSE: To determine the effectiveness of a shared decision-making (SDM) tool versus guideline-informed usual care in translating evidence into primary care, and to explore how use of the tool changed patient perspectives about diabetes medication decision making. METHODS: In this mixed methods multicenter cluster randomized trial, we included patients with type 2 diabetes mellitus and their primary care clinicians. We compared usual care with or without a within-encounter SDM conversation aid. We assessed participant-reported decisions made and quality of SDM (knowledge, satisfaction, and decisional conflict), clinical outcomes, adherence, and observer-based patient involvement in decision-making (OPTION12-scale). We used semi-structured interviews with patients to understand their perspectives. RESULTS: We enrolled 350 patients and 99 clinicians from 20 practices and interviewed 26 patients. Use of the conversation aid increased post-encounter patient knowledge (correct answers, 52% vs. 45%, p = 0.02) and clinician involvement of patients (Mean between-arm difference in OPTION12, 7.3 (95% CI 3, 12); p = 0.003). There were no between-arm differences in treatment choice, patient or clinician satisfaction, encounter length, medication adherence, or glycemic control. Qualitative analyses highlighted differences in how clinicians involved patients in decision making, with intervention patients noting how clinicians guided them through conversations using factors important to them. CONCLUSIONS: Using an SDM conversation aid improved patient knowledge and involvement in SDM without impacting treatment choice, encounter length, medication adherence or improved diabetes control in patients with type 2 diabetes. Future interventions may need to focus specifically on patients with signs of poor treatment fit. CLINICAL TRIAL REGISTRATION: ClinicalTrial.gov: NCT01502891.

Asthma biologic trial eligibility and real-world outcomes in the United States.

OBJECTIVE: To compare the outcomes of real-world patients who would have been eligible for asthma biologics to those who would not have been eligible. METHODS: We used data from the OptumLabs Data Warehouse (OLDW) to categorize patients into eligible and ineligible groups based on clinical trials (n = 19 trials) used for Food and Drug Administration (FDA) approval. We then compared the change in the number of asthma exacerbations before and after biological initiation between the two groups. RESULTS: The percentage of people who would have been eligible for asthma biologic clinical trials ranged from 0-10.2%. The eligible group had a greater reduction in number of asthma exacerbations compared to the ineligible group based on eligibility criteria from 1 omalizumab trial (1.52, 95% CI 1.25, 1.8 in eligible vs. 0.47, 95% CI 0.43, 0.52 in ineligible) and from 1 dupilumab trial (1.6, 95% CI 0.92, 2.28 in eligible vs. 0.52, 95% CI 0.38, 0.65 ineligible). Notably, 15 of the 19 trials had fewer than 11 eligible people, limiting additional comparisons. CONCLUSIONS: Fewer than 1 in 10 people in the United States treated with asthma biologics would have been eligible to participate in the trial for the biologic they used. Where comparisons could be made, trial eligible people have a greater reduction in exacerbations.Supplemental data for this article is available online at https://doi.org/10.1080/02770903.2021.2010749 .

Socioeconomic disadvantage and human papillomavirus (HPV) vaccination uptake.

IMPORTANCE: Despite availability of safe and effective human papillomavirus (HPV) vaccines, vaccination uptake remains low in the U.S. Research examining the impact of neighborhood socioeconomic status on HPV vaccination may help target interventions. OBJECTIVE: To examine the association between area deprivation and HPV vaccine initiation and completion. DESIGN, SETTING, PARTICIPANTS: Retrospective cohort study of individuals aged 11-18 years residing in the upper Midwest region. Receipt of HPV vaccination was examined over a three-year follow-up period (01/01/2016-12/31/2018). MAIN OUTCOMES AND MEASURES: Outcomes of interest were initiation and completion of HPV vaccination. Demographic data were collected from the Rochester Epidemiology Project (REP). Area-level socioeconomic disadvantage was measured by calculating an Area Deprivation Index (ADI) score for each person, a measure of socioeconomic disadvantage derived from American Community Survey data. Multivariable mixed effect Cox proportional hazards models were used to examine the association of ADI quartiles (Q1-Q4) with HPV vaccine series initiation and completion, given initiation. RESULTS: Individuals residing in census block groups with higher deprivation had significantly lower likelihood of HPV vaccine initiation (Q2: HR = 0.91, 0.84-0.99 Q3: HR = 0.83, 0.76-0.90; Q4: HR = 0.84, 0.74-0.96) relative to those in the least-deprived block groups (Q1). Similarly, those living in block groups with higher deprivation had significantly lower likelihood of completion (Q2: HR = 0.91, 0.86-0.97; Q3: HR = 0.87, 0.81-0.94; Q4: HR = 0.82, 0.74-0.92) compared to individuals in the least-deprived block groups (Q1). CONCLUSIONS AND RELEVANCE: Lower probability of both HPV vaccine-series initiation and completion were observed in areas with greater deprivation. Our results can inform allocation of resources to increase HPV vaccination rates in our primary care practice and provide an example of leveraging public data to inform similar efforts across diverse health systems.

Provider Perspectives on Artificial Intelligence-Guided Screening for Low Ejection Fraction in Primary Care: Qualitative Study.

BACKGROUND: The promise of artificial intelligence (AI) to transform health care is threatened by a tangle of challenges that emerge as new AI tools are introduced into clinical practice. AI tools with high accuracy, especially those that detect asymptomatic cases, may be hindered by barriers to adoption. Understanding provider needs and concerns is critical to inform implementation strategies that improve provider buy-in and adoption of AI tools in medicine. OBJECTIVE: This study aimed to describe provider perspectives on the adoption of an AI-enabled screening tool in primary care to inform effective integration and sustained use. METHODS: A qualitative study was conducted between December 2019 and February 2020 as part of a pragmatic randomized controlled trial at a large academic medical center in the United States. In all, 29 primary care providers were purposively sampled using a positive deviance approach for participation in semistructured focus groups after their use of the AI tool in the randomized controlled trial was complete. Focus group data were analyzed using a grounded theory approach; iterative analysis was conducted to identify codes and themes, which were synthesized into findings. RESULTS: Our findings revealed that providers understood the purpose and functionality of the AI tool and saw potential value for more accurate and faster diagnoses. However, successful adoption into routine patient care requires the smooth integration of the tool with clinical decision-making and existing workflow to address provider needs and preferences during implementation. To fulfill the AI tool's promise of clinical value, providers identified areas for improvement including integration with clinical decision-making, cost-effectiveness and resource allocation, provider training, workflow integration, care pathway coordination, and provider-patient communication. CONCLUSIONS: The implementation of AI-enabled tools in medicine can benefit from sensitivity to the nuanced context of care and provider needs to enable the useful adoption of AI tools at the point of care. TRIAL REGISTRATION: ClinicalTrials.gov NCT04000087; https://clinicaltrials.gov/ct2/show/NCT04000087.

Association Between Area-Level Socioeconomic Deprivation and Diabetes Care Quality in US Primary Care Practices.

Importance: Diabetes management operates under a complex interrelationship between behavioral, social, and economic factors that affect a patient's ability to self-manage and access care. Objective: To examine the association between 2 complementary area-based metrics, area deprivation index (ADI) score and rurality, and optimal diabetes care. Design, Setting, and Participants: This cross-sectional study analyzed the electronic health records of patients who were receiving care at any of the 75 Mayo Clinic or Mayo Clinic Health System primary care practices in Minnesota, Iowa, and Wisconsin in 2019. Participants were adults with diabetes aged 18 to 75 years. All data were abstracted and analyzed between June 1 and November 30, 2020. Main Outcomes and Measures: The primary outcome was the attainment of all 5 components of the D5 metric of optimal diabetes care: glycemic control (hemoglobin A1c <8.0%), blood pressure (BP) control (systolic BP <140 mm Hg and diastolic BP <90 mm Hg), lipid control (use of statin therapy according to recommended guidelines), aspirin use (for patients with ischemic vascular disease), and no tobacco use. The proportion of patients receiving optimal diabetes care was calculated as a function of block group-level ADI score (a composite measure of 17 US Census indicators) and zip code-level rurality (calculated using Rural-Urban Commuting Area codes). Odds of achieving the D5 metric and its components were assessed using logistic regression that was adjusted for demographic characteristics, coronary artery disease history, and primary care team specialty. Results: Among the 31 934 patients included in the study (mean [SD] age, 59 [11.7] years; 17 645 men [55.3%]), 13 138 (41.1%) achieved the D5 metric of optimal diabetes care. Overall, 4090 patients (12.8%) resided in the least deprived quintile (quintile 1) of block groups and 1614 (5.1%) lived in the most deprived quintile (quintile 5), while 9193 patients (28.8%) lived in rural areas and 2299 (7.2%) in highly rural areas. The odds of meeting the D5 metric were lower for individuals residing in quintile 5 vs quintile 1 block groups (odds ratio [OR], 0.72; 95% CI, 0.67-0.78). Patients residing in rural (OR, 0.84; 95% CI, 0.73-0.97) and highly rural (OR, 0.81; 95% CI, 0.72-0.91) zip codes were also less likely to attain the D5 metric compared with those in urban areas. Conclusions and Relevance: This cross-sectional study found that patients living in more deprived and rural areas were significantly less likely to attain high-quality diabetes care compared with those living in less deprived and urban areas. The results call for geographically targeted population health management efforts by health systems, public health agencies, and payers.

Drug labeling changes and pediatric hematology/oncology prescribing: Measuring the impact of U.S. legislation.

BACKGROUND/AIMS: The Pediatric Research Equity Act and Best Pharmaceuticals for Children Act are intended to promote the conduct of clinical trials that generate pediatric-specific evidence about drug safety and efficacy. This study assesses the quality of evidence generated through Pediatric Research Equity Act-mandated and Best Pharmaceuticals for Children Act-incentivized clinical trials of hematology/oncology drugs and characterizes subsequent changes in pediatric drug utilization rates. METHODS: Trial characteristics (blinding, randomization, and comparator group) were determined for clinical trials that supported pediatric label changes. Using data from OptumLabs® Data Warehouse, a de-identified administrative claims database, we calculated pediatric utilization rates for each drug. We calculated monthly utilization rates from January 2003 (or from the first month in which data were available) to December 2018. RESULTS: We identified 11 hematology/oncology drugs that underwent pediatric label changes under the Pediatric Research Equity Act Pediatric Research Equity Act and/or Best Pharmaceuticals for Children Act, and we identified 15 trials supporting these changes. Of these trials, 36% (5/14) were randomized, 31% (4/13) were blinded, and 36% (5/14) used a comparator group. A median of 49 children (interquartile range 29.5) received the drug under investigation across these trials. Pediatric label changes were not associated with subsequent changes in pediatric drug utilization. Although some drugs saw increased pediatric use after gaining new pediatric indications, this pattern was not consistently observed. In addition, there was no evidence to suggest that drugs were utilized less frequently after they failed to receive pediatric indications. CONCLUSIONS: Clinical trials of hematology/oncology drugs conducted under the Pediatric Research Equity Act Pediatric Research Equity Act and Best Pharmaceuticals for Children Act generally have low methodological rigor, and the resulting label changes are not consistently associated with changes in pediatric utilization. Alternative regulatory strategies and study designs may be necessary to maximize the impact of newly generated knowledge on drug utilization.

Persistence of asthma biologic use in a US claims database.

BACKGROUND: Little is known on the persistence of asthma biologic use in clinical practice. OBJECTIVE: To evaluate the persistence of asthma biologic use and time to clinical response in clinical practice. METHODS: A cohort of people with asthma who used at least 1 asthma biologic was constructed using data from 2003 to 2019 in the OptumLabs Data Warehouse. Treatment persistence was defined by the length of time that a person continuously used an asthma biologic, allowing for a lapse in use up to 4 months before confirming that a person stopped. Clinical response to treatment (defined as a decline in asthma exacerbations of at least 50% compared with the 6 months before starting an asthma biologic) was described over time and in relation to biologic persistence. RESULTS: There were 9575 people who had at least 1 episode of asthma biologic use. There were 5319 people (64%, 95% confidence interval, 63%-65%) who completed 6 months or more on an asthma biologic and 3284 (45%, 95% confidence interval, 44%-46%) who completed 12 months or more. Of people with 1 or more asthma exacerbation 6 months before index biologic use, 63%, 76%, 80%, and 81% realized a 50% or more reduction in postindex asthma exacerbations in the first 6 months, 6 to 12 months, 12 to 18 months, and 18 to 24 months, respectively. CONCLUSION: Between 48% and 64% of people remained on an asthma biologic for 6 months or more after first use. Most people who achieved a reduction in asthma exacerbations did so in the first 6 months of treatment.

Artificial intelligence-enabled electrocardiograms for identification of patients with low ejection fraction: a pragmatic, randomized clinical trial.

We have conducted a pragmatic clinical trial aimed to assess whether an electrocardiogram (ECG)-based, artificial intelligence (AI)-powered clinical decision support tool enables early diagnosis of low ejection fraction (EF), a condition that is underdiagnosed but treatable. In this trial ( NCT04000087 ), 120 primary care teams from 45 clinics or hospitals were cluster-randomized to either the intervention arm (access to AI results; 181 clinicians) or the control arm (usual care; 177 clinicians). ECGs were obtained as part of routine care from a total of 22,641 adults (N = 11,573 intervention; N = 11,068 control) without prior heart failure. The primary outcome was a new diagnosis of low EF (≤50%) within 90 days of the ECG. The trial met the prespecified primary endpoint, demonstrating that the intervention increased the diagnosis of low EF in the overall cohort (1.6% in the control arm versus 2.1% in the intervention arm, odds ratio (OR) 1.32 (1.01-1.61), P = 0.007) and among those who were identified as having a high likelihood of low EF (that is, positive AI-ECG, 6% of the overall cohort) (14.5% in the control arm versus 19.5% in the intervention arm, OR 1.43 (1.08-1.91), P = 0.01). In the overall cohort, echocardiogram utilization was similar between the two arms (18.2% control versus 19.2% intervention, P = 0.17); for patients with positive AI-ECGs, more echocardiograms were obtained in the intervention compared to the control arm (38.1% control versus 49.6% intervention, P < 0.001). These results indicate that use of an AI algorithm based on ECGs can enable the early diagnosis of low EF in patients in the setting of routine primary care.

Comparative Effectiveness of Machine Learning Approaches for Predicting Gastrointestinal Bleeds in Patients Receiving Antithrombotic Treatment.

Importance: Anticipating the risk of gastrointestinal bleeding (GIB) when initiating antithrombotic treatment (oral antiplatelets or anticoagulants) is limited by existing risk prediction models. Machine learning algorithms may result in superior predictive models to aid in clinical decision-making. Objective: To compare the performance of 3 machine learning approaches with the commonly used HAS-BLED (hypertension, abnormal kidney and liver function, stroke, bleeding, labile international normalized ratio, older age, and drug or alcohol use) risk score in predicting antithrombotic-related GIB. Design, Setting, and Participants: This retrospective cross-sectional study used data from the OptumLabs Data Warehouse, which contains medical and pharmacy claims on privately insured patients and Medicare Advantage enrollees in the US. The study cohort included patients 18 years or older with a history of atrial fibrillation, ischemic heart disease, or venous thromboembolism who were prescribed oral anticoagulant and/or thienopyridine antiplatelet agents between January 1, 2016, and December 31, 2019. Exposures: A cohort of patients prescribed oral anticoagulant and thienopyridine antiplatelet agents was divided into development and validation cohorts based on date of index prescription. The development cohort was used to train 3 machine learning models to predict GIB at 6 and 12 months: regularized Cox proportional hazards regression (RegCox), random survival forests (RSF), and extreme gradient boosting (XGBoost). Main Outcomes and Measures: The performance of the models for predicting GIB in the validation cohort, evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value, and prediction density plots. Relative importance scores were used to identify the variables that were most influential in the top-performing machine learning model. Results: In the entire study cohort of 306 463 patients, 166 177 (54.2%) were male, 193 648 (63.2%) were White, the mean (SD) age was 69.0 (12.6) years, and 12 322 (4.0%) had experienced a GIB. In the validation data set, the HAS-BLED model had an AUC of 0.60 for predicting GIB at 6 months and 0.59 at 12 months. The RegCox model performed the best in the validation set, with an AUC of 0.67 at 6 months and 0.66 at 12 months. XGBoost was similar, with AUCs of 0.67 at 6 months and 0.66 at 12 months, whereas for RSF, AUCs were 0.62 at 6 months and 0.60 at 12 months. The variables with the highest importance scores in the RegCox model were prior GI bleed (importance score, 0.72); atrial fibrillation, ischemic heart disease, and venous thromboembolism combined (importance score, 0.38); and use of gastroprotective agents (importance score, 0.32). Conclusions and Relevance: In this cross-sectional study, the machine learning models examined showed similar performance in identifying patients at high risk for GIB after being prescribed antithrombotic agents. Two models (RegCox and XGBoost) performed modestly better than the HAS-BLED score. A prospective evaluation of the RegCox model compared with HAS-BLED may provide a better understanding of the clinical impact of improved performance.