An estimated 425 million people globally have diabetes, accounting for 12% of the world's health expenditures, and yet 1 in 2 persons remain undiagnosed and untreated. Applications of artificial intelligence (AI) and cognitive computing offer promise in diabetes care. The purpose of this article is to better understand what AI advances may be relevant today to persons with diabetes (PWDs), their clinicians, family, and caregivers. The authors conducted a predefined, online PubMed search of publicly available sources of information from 2009 onward using the search terms “diabetes” and “artificial intelligence.” The study included clinically-relevant, high-impact articles, and excluded articles whose purpose was technical in nature. A total of 450 published diabetes and AI articles met the inclusion criteria. The studies represent a diverse and complex set of innovative approaches that aim to transform diabetes care in 4 main areas: automated retinal screening, clinical decision support, predictive population risk stratification, and patient self-management tools. Many of these new AI-powered retinal imaging systems, predictive modeling programs, glucose sensors, insulin pumps, smartphone applications, and other decision-support aids are on the market today with more on the way. AI applications have the potential to transform diabetes care and help millions of PWDs to achieve better blood glucose control, reduce hypoglycemic episodes, and reduce diabetes comorbidities and complications. AI applications offer greater accuracy, efficiency, ease of use, and satisfaction for PWDs, their clinicians, family, and caregivers.

Increased scrutiny of artificial intelligence (AI) applications in healthcare highlights the need for real-world evaluations for effectiveness and unintended consequences. The complexity of healthcare, compounded by the user- and context-dependent nature of AI applications, calls for a multifaceted approach beyond traditional in silico evaluation of AI. We propose an interdisciplinary, phased research framework for evaluation of AI implementations in healthcare. We draw analogies to and highlight differences from the clinical trial phases for drugs and medical devices, and we present study design and methodological guidance for each stage.

Compared with women who discontinued use of an atypical antipsychotic medication before the start of pregnancy, women who continued treatment with olanzapine or quetiapine had an increased risk of gestational diabetes that may be explained by the metabolic effects associated with these two drugs.

IMPORTANCEThe lack of standards in methods to reduce bias for clinical algorithms presents various challenges in providing reliable predictions and in addressing health disparities. OBJECTIVE To evaluate approaches for reducing bias in machine learning models using a real-world clinical scenario. DESIGN, SETTING, AND PARTICIPANTSHealth data for this cohort study were obtained from the IBM MarketScan Medicaid Database. Eligibility criteria were as follows: (1) Female individuals aged 12 to 55 years with a live birth record identified by delivery-related codes from January 1, 2014, through December 31, 2018; (2) greater than 80% enrollment through pregnancy to 60 days post partum; and (3) evidence of coverage for depression screening and mental health services. Statistical analysis was performed in 2020. EXPOSURES Binarized race (Black individuals and White individuals). MAIN OUTCOMES AND MEASURES Machine learning models (logistic regression [LR], randomforest, and extreme gradient boosting) were trained for 2 binary outcomes: postpartum depression (PPD) and postpartum mental health service utilization. Risk-adjusted generalized linear models were used for each outcome to assess potential disparity in the cohort associated with binarized race (Black or White). Methods for reducing bias, including reweighing, Prejudice Remover, and removing race from the models, were examined by analyzing changes in fairness metrics compared with the base models. Baseline characteristics of female individuals at the top-predicted risk decile were compared for systematic differences. Fairness metrics of disparate impact (DI, 1 indicates fairness) and equal opportunity difference (EOD, 0 indicates fairness). RESULTS Among 573 634 female individuals initially examined for this study, 314 903 were White (54.9%), 217 899 were Black (38.0%), and the mean (SD) age was 26.1 (5.5) years. The risk-adjusted odds ratio comparing White participants with Black participants was 2.06 (95% CI, 2.02-2.10) for clinically recognized PPD and 1.37 (95% CI, 1.33-1.40) for postpartum mental health service utilization. Taking the LR model for PPD prediction as an example, reweighing reduced bias as measured by improved DI and EOD metrics from 0.31 and −0.19 to 0.79 and 0.02, respectively.

Objective To evaluate whether psychostimulants used to treat attention-deficit hyperactivity disorder (ADHD) are associated with risk of adverse placental-associated pregnancy outcomes including preeclampsia, placental abruption, growth restriction, and preterm birth. Methods We designed a population-based cohort study where we examined a cohort of pregnant women and their liveborn infants enrolled in Medicaid from 2000 to 2010. Women who received amphetamine–dextroamphetamine or methylphenidate monotherapy in the first half of pregnancy were compared to unexposed women. We considered atomoxetine, a non-stimulant ADHD medication, as a negative control exposure. To assess whether the risk period extended to the latter half of pregnancy, women who continued stimulant monotherapy after 20 weeks were compared to those who discontinued. Risk ratios (RRs) and 95% confidence intervals (CIs) were estimated with propensity score stratification to control for confounders. Results Pregnancies exposed to amphetamine/dextroamphetamine (n=3331), methylphenidate (n=1515), and atomoxetine (n=453) monotherapy in early pregnancy were compared to 1,461,493 unexposed pregnancies. Among unexposed women, the risks of the outcomes were 3.7% for preeclampsia, 1.4% for placental abruption, 2.9% for small for gestational age, and 11.2% for preterm birth. The adjusted RR for stimulant use was 1.29 for preeclampsia (95% CI 1.11–1.49), 1.13 for placental abruption (0.88–1.44), 0.91 for small for gestational age (SGA; 0.77–1.07) and 1.06 for preterm birth (0.97–1.16). Compared to discontinuation (n=3527), the adjusted RR for continuation of stimulant use in the latter half of pregnancy (n=1319) was 1.26 for preeclampsia (0.94–1.67), 1.08 for placental abruption (0.67–1.74), 1.37 for SGA (0.97–1.93), and 1.30 for preterm birth (1.10–1.55). Atomoxetine was not associated with the outcomes studied. Conclusion Psychostimulant use during pregnancy was associated with a small increased relative risk of preeclampsia and preterm birth. The absolute increases in risks are small and thus, women with significant ADHD should not be counseled to suspend their ADHD treatment based on these findings.

Objective Given the increasing use and broadening of indications for antipsychotic medications in the general population, as well as the paucity of information on the safety of this drug class during pregnancy, the study aim was to document patterns of antipsychotic medication use in pregnant women. Method Medicaid Analytic eXtract data (2001–2010) from pregnant women who delivered live-born infants was used. Antipsychotic use at both the class and individual drug level was defined based on dispensed outpatient prescriptions. Users’ characteristics, including mental disorder diagnoses, were described. Temporal trends in use, as well as discontinuation patterns and polytherapy with other psychotropic medications during pregnancy were evaluated. Results Among 1,522,247 pregnancies, the prevalence of atypical antipsychotic use at any time during pregnancy increased three-fold, from .4% to 1.3%, over the 10-year period while the use of typical antipsychotics remained stable around .1%. The increase in atypical use was largely driven by more frequent use in patients with bipolar disorder. Quetiapine and aripiprazole were the most frequently dispensed drugs, and polytherapy with antidepressants (65.2%), benzodiazepines (24.9%), and/or other mood stabilizers (22.0%) was common among women using antipsychotics during pregnancy. More than 50% of women receiving an antipsychotic in the 3 months prior to pregnancy discontinued during pregnancy. Conclusions A growing number of pregnant women in Medicaid are exposed to atypical antipsychotics, frequently in combination with other psychotropic medications. This study highlights the importance of documenting the use and safety of these drugs during pregnancy to inform therapeutic decision making for pregnant women with psychiatric disorders.

ObjectiveTo compare the risk of in-hospital mortality associated with haloperidol compared with atypical antipsychotics in patients admitted to hospital with acute myocardial infarction.DesignCohort study using a healthcare database.SettingNationwide sample of patient data from more than 700 hospitals across the United States.Participants6578 medical patients aged more than 18 years who initiated oral haloperidol or oral atypical antipsychotics (olanzapine, quetiapine, risperidone) during a hospital admission with a primary diagnosis of acute myocardial infarction between 2003 and 2014.Main outcome measureIn-hospital mortality during seven days of follow-up from treatment initiation.ResultsAmong 6578 patients (mean age 75.2 years) treated with an oral antipsychotic drug, 1668 (25.4%) initiated haloperidol and 4910 (74.6%) initiated atypical antipsychotics. The mean time from admission to start of treatment (5.3 v 5.6 days) and length of stay (12.5 v 13.6 days) were similar, but the mean treatment duration was shorter in patients using haloperidol compared with those using atypical antipsychotics (2.4 v 3.9 days). 1:1 propensity score matching was used to adjust for confounding. In intention to treat analyses with the matched cohort, the absolute rate of death per 100 person days was 1.7 for haloperidol (129 deaths) and 1.1 for atypical antipsychotics (92 deaths) during seven days of follow-up from treatment initiation. The survival probability was 0.93 in patients using haloperidol and 0.94 in those using atypical antipsychotics at day 7, accounting for the loss of follow-up due to hospital discharge. The unadjusted and adjusted hazard ratios of death were 1.51 (95% confidence interval 1.22 to 1.85) and 1.50 (1.14 to 1.96), respectively. The association was strongest during the first four days of follow-up and decreased over time. By day 5, the increased risk was no longer evident (1.12, 0.79 to 1.59). In the as-treated analyses, the unadjusted and adjusted hazard ratios were 1.90 (1.43 to 2.53) and 1.93 (1.34 to 2.76), respectively.ConclusionThe results suggest a small increased risk of death within seven days of initiating haloperidol compared with initiating an atypical antipsychotic in patients with acute myocardial infarction. Although residual confounding cannot be excluded, this finding deserves consideration when haloperidol is used for patients admitted to hospital with cardiac morbidity.

ObjectiveTo compare the risk of serious infections associated with use of systemic steroids, non-biologic agents, or tumor necrosis factor α (TNF) inhibitors in pregnancy.DesignObservational cohort study.SettingPublic (Medicaid, 2001-10) or private (Optum Clinformatics, 2004-15) health insurance programs in the US.Participants4961 pregnant women treated with immunosuppressive drugs for rheumatoid arthritis, systemic lupus erythematosus, ankylosing spondylitis, psoriatic arthritis, or inflammatory bowel disease.Exposure for observational studiesExposure was classified into steroid, non-biologic, or TNF inhibitors on first filled prescription during pregnancy. Because TNF inhibitors are not used to treat systemic lupus erythematosus, patients with this condition were excluded from comparisons involving TNF inhibitors.Main outcome measureThe main outcome was occurrence of serious infections during pregnancy, defined by hospital admission for bacterial or opportunistic infections. Hazard ratios were derived using Cox proportional hazard regression models after adjustment for confounding with propensity score fine stratification. A logistic regression model was used to conduct a dose-response analysis among women filling at least one steroid prescription.Results71 out of 4961 pregnant women (0.2%) treated with immunosuppressive agents experienced serious infections. The crude incidence rates of serious infections per 100 person years among 2598 steroid users, 1587 non-biologic users, and 776 TNF inhibitors users included in this study were 3.4 (95% confidence interval 2.5 to 4.7), 2.3 (1.5 to 3.5), and 1.5 (0.7 to 3.0), respectively. No statistically significant differences in the risk of serious infections during pregnancy were observed among users of the three immunosuppressive drug classes: non-biologics v steroids, hazard ratio 0.81 (95% confidence interval 0.48 to 1.37), TNF inhibitors v steroids 0.91 (0.36 to 2.26), and TNF inhibitors v non-biologics 1.36 (0.47 to 3.93). In the dose-response analysis, higher steroid dose was associated with an increased risk of serious infections during pregnancy (coefficient for each unit increase in average prednisone equivalent mg daily dose=0.019, P=0.02).ConclusionsRisk of serious infections is similar among pregnant women with systemic inflammatory conditions using steroids, non-biologics, and TNF inhibitors. However, high dose steroid use is an independent risk factor of serious infections in pregnancy.