A Novel Explainable AI Method to Assess Associations between Temporal Patterns in Patient Trajectories and Adverse Outcome Risks: Analyzing Fitness as a Risk Factor of ADRD.

We present a novel explainable artificial intelligence (XAI) method to assess the associations between the temporal patterns in the patient trajectories recorded in longitudinal clinical data and the adverse outcome risks, through explanations for a type of deep neural network model called Hybrid Value-Aware Transformer (HVAT) model. The HVAT models can learn jointly from longitudinal and non-longitudinal clinical data, and in particular can leverage the time-varying numerical values associated with the clinical codes or concepts within the longitudinal data for outcome prediction. The key component of the XAI method is the definitions of two derived variables, the temporal mean and the temporal slope, which are defined for the clinical concepts with associated time-varying numerical values. The two variables represent the overall level and the rate of change over time, respectively, in the trajectory formed by the values associated with the clinical concept. Two operations on the original values are designed for changing the values of the two derived variables separately. The effects of the two variables on the outcome risks learned by the HVAT model are calculated in terms of impact scores and impacts. Interpretations of the impact scores and impacts as being similar to those of odds ratios are also provided. We applied the XAI method to the study of cardiorespiratory fitness (CRF) as a risk factor of Alzheimer's disease and related dementias (ADRD). Using a retrospective case-control study design, we found that each one-unit increase in the overall CRF level is associated with a 5% reduction in ADRD risk, while each one-unit increase in the changing rate of CRF over time is associated with a 1% reduction. A closer investigation revealed that the association between the changing rate of CRF level and the ADRD risk is nonlinear, or more specifically, approximately piecewise linear along the axis of the changing rate on two pieces: the piece of negative changing rates and the piece of positive changing rates.

Hybrid Value-Aware Transformer Architecture for Joint Learning from Longitudinal and Non-Longitudinal Clinical Data.

Transformer is the latest deep neural network (DNN) architecture for sequence data learning, which has revolutionized the field of natural language processing. This success has motivated researchers to explore its application in the healthcare domain. Despite the similarities between longitudinal clinical data and natural language data, clinical data presents unique complexities that make adapting Transformer to this domain challenging. To address this issue, we have designed a new Transformer-based DNN architecture, referred to as Hybrid Value-Aware Transformer (HVAT), which can jointly learn from longitudinal and non-longitudinal clinical data. HVAT is unique in the ability to learn from the numerical values associated with clinical codes/concepts such as labs, and in the use of a flexible longitudinal data representation called clinical tokens. We have also trained a prototype HVAT model on a case-control dataset, achieving high performance in predicting Alzheimer's disease and related dementias as the patient outcome. The results demonstrate the potential of HVAT for broader clinical data-learning tasks.

Hybrid Value-Aware Transformer Architecture for Joint Learning from Longitudinal and Non-Longitudinal Clinical Data.

Transformer is the latest deep neural network (DNN) architecture for sequence data learning that has revolutionized the field of natural language processing. This success has motivated researchers to explore its application in the healthcare domain. Despite the similarities between longitudinal clinical data and natural language data, clinical data presents unique complexities that make adapting Transformer to this domain challenging. To address this issue, we have designed a new Transformer-based DNN architecture, referred to as Hybrid Value-Aware Transformer (HVAT), which can jointly learn from longitudinal and non-longitudinal clinical data. HVAT is unique in the ability to learn from the numerical values associated with clinical codes/concepts such as labs, and also the use of a flexible longitudinal data representation called clinical tokens. We trained a prototype HVAT model on a case-control dataset, achieving high performance in predicting Alzheimer’s disease and related dementias as the patient outcome. The result demonstrates the potential of HVAT for broader clinical data learning tasks.

Recent trends in tranexamic acid use during postpartum hemorrhage in the United States.

INTRODUCTION: Postpartum hemorrhage (PPH) was the second leading cause of maternal death, accounting for approximately 14% of all pregnancy-related deaths between 2017 and 2019 in the United States. Several large multi-center studies have demonstrated decreased PPH rates with the use of tranexamic acid (TXA). Little data exists regarding the prevalence of TXA use in obstetric patients. METHODS: We identified over 1.2 million US pregnancies between January 1, 2015 and June 30, 2021, with and without PPH by International Statistical Classification of Disease and Related Health Problems, Tenth Revision codes using Cerner Real-World Database™. TXA use and patient characteristics were abstracted from the electronic medical record. RESULTS: During delivery, TXA was used approximately 1% of the time (12,394 / 1,262,574). Pregnant patients who did and did not receive TXA during delivery had similar demographic characteristics. Pregnant patients who underwent cesarean delivery (4,356 / 12,394), had a term delivery (10,199 / 12,394), and had comorbid conditions were more likely to receive TXA during hospitalization for delivery. The majority of TXA was use was concentrated in Arizona, Colorado, Idaho, New Mexico, Nevada, Utah, and Wyoming. During the study period the use of TXA increased in both patients with PPH and those without. CONCLUSION: The data illustrate a rapid increase in the use of TXA after 2017 while the total number of pregnancies remained relatively constant. The observed increase in TXA use may reflect changing practicing patterns as the support for use of TXA in the setting of PPH prophylaxis increases.

Identification of colorectal cancer using structured and free text clinical data.

Colorectal cancer incidence has continually fallen among those 50 years old and over. However, the incidence has increased in those under 50. Even with the recent screening guidelines recommending that screening begins at age 45, nearly half of all early-onset colorectal cancer will be missed. Methods are needed to identify high-risk individuals in this age group for targeted screening. Colorectal cancer studies, as with other clinical studies, have required labor intensive chart review for the identification of those affected and risk factors. Natural language processing and machine learning can be used to automate the process and enable the screening of large numbers of patients. This study developed and compared four machine learning and statistical models: logistic regression, support vector machine, random forest, and deep neural network, in their performance in classifying colorectal cancer patients. Excellent classification performance is achieved with AUCs over 97%.

Adverse perinatal outcomes in a large United States birth cohort during the COVID-19 pandemic.

BACKGROUND: The impact of coronavirus disease 2019 (COVID-19) on adverse perinatal outcomes remains unclear. OBJECTIVE: This study aimed to investigate whether COVID-19 is associated with adverse perinatal outcomes in a large national dataset and to examine the rates of adverse outcomes during the pandemic compared with the rates of adverse outcomes during the prepandemic period. STUDY DESIGN: This observational cohort study included 683,905 patients, between the ages of 12 and 50, hospitalized for childbirth and abortion between January 1, 2019, and May 31, 2021. During the prepandemic period, 271,444 women were hospitalized for childbirth. During the pandemic, 308,532 women were hospitalized for childbirth, and 2708 women had COVID-19. The associations between COVID-19 and inhospital adverse perinatal outcomes were examined using propensity score-adjusted logistic regression. RESULTS: Women with COVID-19 were more likely to experience both early and late preterm birth (adjusted odds ratios, 1.38 [95% confidence interval, 1.1-1.7] and 1.62 [95% confidence interval, 1.3-1.7], respectively), preeclampsia (adjusted odds ratio, 1.2 [95% confidence interval, 1.0-1.4]), disseminated intravascular coagulopathy (adjusted odds ratio, 1.57 [95% confidence interval, 1.1-2.2]), pulmonary edema (adjusted odds ratio, 2.7 [95% confidence interval, 1.1-6.3]), and need for mechanical ventilation (adjusted odds ratio, 8.1 [95% confidence interval, 3.8-17.3]) than women without COVID-19. There was no significant difference in the prevalence of stillbirth among women with COVID-19 (16/2708) and women without COVID-19 (174/39,562) (P=.257). There was no difference in adverse outcomes among women who delivered during the pandemic vs prepandemic period. Combined inhospital mortality was significantly higher for women with COVID-19 (147 [95% confidence interval, 3.0-292.0] vs 2.5 [95% confidence interval, 0.0-7.5] deaths per 100,000 women). Women diagnosed with COVID-19 within 30 days before hospitalization were more likely to experience early preterm birth, placental abruption, and mechanical ventilation than women diagnosed with COVID-19 >30 days before hospitalization for childbirth (4.0% vs 2.4% for early preterm birth [adjusted odds ratio, 1.7; 95% confidence interval, 1.1-2.7]; 2.2% vs 1.2% for placental abruption [adjusted odds ratio, 1.86; 95% confidence interval, 1.0-3.4]; and 0.9% vs 0.1% for mechanical ventilation [adjusted odds ratio, 13.7; 95% confidence interval, 1.8-107.2]). CONCLUSION: Women with COVID-19 had a higher prevalence of adverse perinatal outcomes and increased in-hospital mortality, with the highest risk occurring when the diagnosis was within 30 days of hospitalization, raising the possibility of a high-risk period.

Association of prehospital antiplatelet therapy with survival in patients hospitalized with COVID-19: A propensity score-matched analysis.

PURPOSE: Coronavirus disease 2019 (COVID-19) is associated with hypercoagulability and increased thrombotic risk. The impact of prehospital antiplatelet therapy on in-hospital mortality is uncertain. METHODS: This was an observational cohort study of 34 675 patients ≥50 years old from 90 health systems in the United States. Patients were hospitalized with laboratory-confirmed COVID-19 between February 2020 and September 2020. For all patients, the propensity to receive prehospital antiplatelet therapy was calculated using demographics and comorbidities. Patients were matched based on propensity scores, and in-hospital mortality was compared between the antiplatelet and non-antiplatelet groups. RESULTS: The propensity score-matched cohort of 17 347 patients comprised of 6781 and 10 566 patients in the antiplatelet and non-antiplatelet therapy groups, respectively. In-hospital mortality was significantly lower in patients receiving prehospital antiplatelet therapy (18.9% vs. 21.5%, p < .001), resulting in a 2.6% absolute reduction in mortality (HR: 0.81, 95% CI: 0.76-0.87, p < .005). On average, 39 patients needed to be treated to prevent one in-hospital death. In the antiplatelet therapy group, there was a significantly lower rate of pulmonary embolism (2.2% vs. 3.0%, p = .002) and higher rate of epistaxis (0.9% vs. 0.4%, p < .001). There was no difference in the rate of other hemorrhagic or thrombotic complications. CONCLUSIONS: In the largest observational study to date of prehospital antiplatelet therapy in patients with COVID-19, there was an association with significantly lower in-hospital mortality. Randomized controlled trials in diverse patient populations with high rates of baseline comorbidities are needed to determine the ultimate utility of antiplatelet therapy in COVID-19.

The Sacred Sites of Houston: A Novel Experiential Course for Undergraduate Medical Education on Religion and Spirituality.

Medical schools are charged to deliver a curriculum on religion and spirituality (R/S), so a novel experiential course, the Sacred Sites of Houston, was developed. Sixty students completed the course consisting of 6 site visits. Post-course, participants described more general knowledge and knowledge of how each faith tradition describes medicine and health (p < 0.05 for all) except for Catholicism (p = 0.564 and p = 0.058). Ten course participants and 6 control non-course participants were interviewed following clinical rotations to assess the impact of the experiential course on R/S in the clinical setting. Themes from qualitative interviews such as R/S, barriers, interactions, and the course impact emerged. The importance of R/S in the patient-provider relationship and end-of-life care was prominent in course participant interviews compared to non-course participant control subjects. Participation in the course resulted in increased chaplain engagement and significant personal impact. These qualitative and quantitative findings indicate that an experiential course may be effective at addressing the deficit in R/S undergraduate medical education and help enhance the spiritually and religiously competent care of patients.

A model-agnostic approach for understanding heart failure risk factors.

OBJECTIVE: Understanding the risk factors for developing heart failure among patients with type 2 diabetes can contribute to preventing deterioration of quality of life for those persons. Electronic health records (EHR) provide an opportunity to use sophisticated machine learning models to understand and compare the effect of different risk factors for developing HF. As the complexity of the model increases, however, the transparency of the model often decreases. To interpret the results, we aimed to develop a model-agnostic approach to shed light on complex models and interpret the effect of features on developing heart failure. Using the HealthFacts EHR database of the Cerner EHR, we extracted the records of 723 patients with at least 6 yeas of follow up of type 2 diabetes, of whom 134 developed heart failure. Using age and comorbidities as features and heart failure as the outcome, we trained logistic regression, random forest, XGBoost, neural network, and then applied our proposed approach to rank the effect of each factor on developing heart failure. RESULTS: Compared to the "importance score" built-in function of XGBoost, our proposed approach was more accurate in ranking the effect of the different risk factors on developing heart failure.

Using Machine Learning Imputed Outcomes to Assess Drug-Dependent Risk of Self-Harm in Patients with Bipolar Disorder: A Comparative Effectiveness Study.

BACKGROUND: Incomplete suicidality coding in administrative claims data is a known obstacle for observational studies. With most of the negative outcomes missing from the data, it is challenging to assess the evidence on treatment strategies for the prevention of self-harm in bipolar disorder (BD), including pharmacotherapy and psychotherapy. There are conflicting data from studies on the drug-dependent risk of self-harm, and there is major uncertainty regarding the preventive effect of monotherapy and drug combinations. OBJECTIVE: The aim of this study was to compare all commonly used BD pharmacotherapies, as well as psychotherapy for the risk of self-harm, in a large population of commercially insured individuals, using self-harm imputation to overcome the known limitations of this outcome being underrecorded within US electronic health care records. METHODS: The IBM MarketScan administrative claims database was used to compare self-harm risk in patients with BD following 65 drug regimens and drug-free periods. Probable but uncoded self-harm events were imputed via machine learning, with different probability thresholds examined in a sensitivity analysis. Comparators included lithium, mood-stabilizing anticonvulsants (MSAs), second-generation antipsychotics (SGAs), first-generation antipsychotics (FGAs), and five classes of antidepressants. Cox regression models with time-varying covariates were built for individual treatment regimens and for any pharmacotherapy with or without psychosocial interventions ("psychotherapy"). RESULTS: Among 529,359 patients, 1.66% (n=8813 events) had imputed and/or coded self-harm following the exposure of interest. A higher self-harm risk was observed during adolescence. After multiple testing adjustment (P≤.012), the following six regimens had higher risk of self-harm than lithium: tri/tetracyclic antidepressants + SGA, FGA + MSA, FGA, serotonin-norepinephrine reuptake inhibitor (SNRI) + SGA, lithium + MSA, and lithium + SGA (hazard ratios [HRs] 1.44-2.29), and the following nine had lower risk: lamotrigine, valproate, risperidone, aripiprazole, SNRI, selective serotonin reuptake inhibitor (SSRI), "no drug," bupropion, and bupropion + SSRI (HRs 0.28-0.74). Psychotherapy alone (without medication) had a lower self-harm risk than no treatment (HR 0.56, 95% CI 0.52-0.60; P=8.76×10-58). The sensitivity analysis showed that the direction of drug-outcome associations did not change as a function of the self-harm probability threshold. CONCLUSIONS: Our data support evidence on the effectiveness of antidepressants, MSAs, and psychotherapy for self-harm prevention in BD. TRIAL REGISTRATION: ClinicalTrials.gov NCT02893371; https://clinicaltrials.gov/ct2/show/NCT02893371.

Understanding Demographic Risk Factors for Adverse Outcomes in COVID-19 Patients: Explanation of a Deep Learning Model.

This study was to understand the impacts of three key demographic variables, age, gender, and race, on the adverse outcome of all-cause hospitalization or all-cause mortality in patients with COVID-19, using a deep neural network (DNN) analysis. We created a cohort of Veterans who were tested positive for COVID-19, extracted data on age, gender, and race, and clinical characteristics from their electronic health records, and trained a DNN model for predicting the adverse outcome. Then, we analyzed the association of the demographic variables with the risks of the adverse outcome using the impact scores and interaction scores for explaining DNN models. The results showed that, on average, older age and African American race were associated with higher risks while female gender was associated with lower risks. However, individual-level impact scores of age showed that age was a more impactful risk factor in younger patients and in older patients with fewer comorbidities. The individual-level impact scores of gender and race variables had a wide span covering both positive and negative values. The interaction scores between the demographic variables showed that the interaction effects were minimal compared to the impact scores associated with them. In conclusion, the DNN model is able to capture the non-linear relationship between the risk factors and the adverse outcome, and the impact scores and interaction scores can help explain the complicated non-linear effects between the demographic variables and the risk of the outcome.

Using multivariate long short-term memory neural network to detect aberrant signals in health data for quality assurance.

BACKGROUND: The data quality of electronic health records (EHR) has been a topic of increasing interest to clinical and health services researchers. One indicator of possible errors in data is a large change in the frequency of observations in chronic illnesses. In this study, we built and demonstrated the utility of a stacked multivariate LSTM model to predict an acceptable range for the frequency of observations. METHODS: We applied the LSTM approach to a large EHR dataset with over 400 million total encounters. We computed sensitivity and specificity for predicting if the frequency of an observation in a given week is an aberrant signal. RESULTS: Compared with the simple frequency monitoring approach, our proposed multivariate LSTM approach increased the sensitivity of finding aberrant signals in 6 randomly selected diagnostic codes from 75 to 88% and the specificity from 68 to 91%. We also experimented with two different LSTM algorithms, namely, direct multi-step and recursive multi-step. Both models were able to detect the aberrant signals while the recursive multi-step algorithm performed better. CONCLUSIONS: Simply monitoring the frequency trend, as is the common practice in systems that do monitor the data quality, would not be able to distinguish between the fluctuations caused by seasonal disease changes, seasonal patient visits, or a change in data sources. Our study demonstrated the ability of stacked multivariate LSTM models to recognize true data quality issues rather than fluctuations that are caused by different reasons, including seasonal changes and outbreaks.

A bottom-up approach to creating an ontology for medication indications.

OBJECTIVES: The study sought to learn if it were possible to develop an ontology that would allow the Food and Drug Administration approved indications to be expressed in a manner computable and comparable to what is expressed in an electronic health record. MATERIALS AND METHODS: A random sample of 1177 of the 3000+ extant, distinct medical products (identified by unique new drug application numbers) was selected for investigation. Close manual examination of the indication portion of the labels for these drugs led to the development of a formal model of indications. RESULTS: The model represents each narrative indication as a disjunct of conjuncts of assertions about an individual. A desirable attribute is that each assertion about an individual should be testable without reference to other contextual information about the situation. The logical primitives are chosen from 2 categories (context and conditions) and are linked to an enumeration of uses, such as prevention. We found that more than 99% of approved label indications for treatment or prevention could be so represented. DISCUSSION: While some indications are straightforward to represent, difficulties stem from the need to represent temporal or sequential references. In addition, there is a mismatch of terminologies between what is present in an electronic health record and in the label narrative. CONCLUSIONS: A workable model for formalizing drug indications is possible. Remaining challenges include designing workflow to model narrative label indications for all approved drug products and incorporation of standard vocabularies.

Healthy Lifestyle and Mood: A Biomedical Informatics Citizen Science Project in a High School Classroom.

Mental health is an increasing concern in adolescents. Mental health disorders can affect academic performance, affect the cultivation of healthy relationships, and even lead to suicide. Healthy lifestyle can improve mental health, though there are gaps in the research, partly resulted from the lack of detailed longitudinal datasets on lifestyle and mental health. To inform and engage students in the research on adolescent lifestyle and mood, the George Washington University and the T.C. Williams High School in Alexandria, Virginia teamed up in a citizen science project. Students generated questions, collected data on themselves, analyzed the data, and produced research reports relating to their mental health and lifestyle. Student feedbacks suggest that the students find the project to be generally interesting and some students (46%) reported that the participation in the project may influence their college and career plans. The anonymized dataset resulted from the project provides another contribution to science.

The effect of age and sex on selected hematologic and serum biochemical analytes in 4,804 elite endurance-trained sled dogs participating in the Iditarod Trail Sled Dog Race pre-race examination program.

Endurance-trained sled dogs provide a unique translational model to characterize changes in hematologic and serum biochemical analytes due to the aging process. The primary objective of this study was to determine the effect of age and sex on specific hematologic and serum biochemical parameters in the endurance trained sled dog. Longitudinal and cross-sectional data were analyzed from 9,746 blood and serum samples from 4,804 dogs collected over 7 years as part of the Iditarod Trail Sled Dog Race pre-race examination program. Mixed models analysis was used for statistical analysis and P < 0.01 was considered significant. Dogs ranged from 1-12 years of age and 39% were female. Serum total calcium and phosphorus concentrations and white blood cell count decreased nonlinearly to asymptotic values by 6.6, 3.1, and 6.9 years of age, respectively, equivalent to estimated physiologic ages in human years of 44, 27, and 46 years. Serum glucose concentrations reached their lowest value at 7.8 years of age, equivalent to an estimated human physiologic age of 50 years, after which time the concentration increased. Serum globulin concentrations increased with age, but nonlinearly for females and linearly for males. Most sex-related differences were <5%; however, females had lower serum urea nitrogen (14.7%) and creatinine (7.3%) concentrations, lower serum alanine aminotransferase activity (16.6%), and higher serum total bilirubin concentration (12.8%) and platelet count (6.0%). The endurance-trained sled dog provides an excellent model to separate the physiologic effects of age from those of a sedentary lifestyle on hematologic and serum biochemical analytes.

LocalControl: An R Package for Comparative Safety and Effectiveness Research.

The LocalControl R package implements novel approaches to address biases and confounding when comparing treatments or exposures in observational studies of outcomes. While designed and appropriate for use in comparative safety and effectiveness research involving medicine and the life sciences, the package can be used in other situations involving outcomes with multiple confounders. LocalControl is an open-source tool for researchers whose aim is to generate high quality evidence using observational data. The package implements a family of methods for non-parametric bias correction when comparing treatments in observational studies, including survival analysis settings, where competing risks and/or censoring may be present. The approach extends to bias-corrected personalized predictions of treatment outcome differences, and analysis of heterogeneity of treatment effect-sizes across patient subgroups.

Imputation and characterization of uncoded self-harm in major mental illness using machine learning.

OBJECTIVE: We aimed to impute uncoded self-harm in administrative claims data of individuals with major mental illness (MMI), characterize self-harm incidence, and identify factors associated with coding bias. MATERIALS AND METHODS: The IBM MarketScan database (2003-2016) was used to analyze visit-level self-harm in 10 120 030 patients with ≥2 MMI codes. Five machine learning (ML) classifiers were tested on a balanced data subset, with XGBoost selected for the full dataset. Classification performance was validated via random data mislabeling and comparison with a clinician-derived "gold standard." The incidence of coded and imputed self-harm was characterized by year, patient age, sex, U.S. state, and MMI diagnosis. RESULTS: Imputation identified 1 592 703 self-harm events vs 83 113 coded events, with areas under the curve >0.99 for the balanced and full datasets, and 83.5% agreement with the gold standard. The overall coded and imputed self-harm incidence were 0.28% and 5.34%, respectively, varied considerably by age and sex, and was highest in individuals with multiple MMI diagnoses. Self-harm undercoding was higher in male than in female individuals and increased with age. Substance abuse, injuries, poisoning, asphyxiation, brain disorders, harmful thoughts, and psychotherapy were the main features used by ML to classify visits. DISCUSSION: Only 1 of 19 self-harm events was coded for individuals with MMI. ML demonstrated excellent performance in recovering self-harm visits. Male individuals and seniors with MMI are particularly vulnerable to self-harm undercoding and may be at risk of not getting appropriate psychiatric care. CONCLUSIONS: ML can effectively recover unrecorded self-harm in claims data and inform psychiatric epidemiological and observational studies.

Diabetes mellitus risk for 102 drugs and drug combinations used in patients with bipolar disorder.

OBJECTIVE: To compare the largest set of bipolar disorder pharmacotherapies to date (102 drugs and drug combinations) for risk of diabetes mellitus (DM). METHODS: The IBM MarketScan® database was used to retrospectively analyze data on 565,253 adults with bipolar disorder without prior glucose metabolism-related diagnoses. The pharmacotherapies compared were lithium, mood-stabilizing anticonvulsants, antipsychotics, and antidepressants (monotherapy and multi-class polypharmacy). Cox regression modeling included fixed pre-treatment covariates and time-varying drug exposure covariates to estimate the hazard ratio (HR) of each treatment versus "No drug". RESULTS: The annual incidence of new-onset diabetes during the exposure period was 3.09 % (22,951 patients). The HR of drug-dependent DM ranged from 0.79 to 2.37. One-third of the studied pharmacotherapies, including most of the antipsychotic-containing regimens, had a significantly higher risk of DM compared to "No drug". A significantly lower DM risk was associated with lithium, lamotrigine, oxcarbazepine and bupropion monotherapies, selective serotonin reuptake inhibitors (SSRI) mono-class therapy and several drug combinations containing bupropion and an SSRI. As additional drugs were combined in more complex polypharmacy, higher HRs were consistently observed. CONCLUSIONS: There is an increased risk of diabetes mellitus associated with antipsychotic and psychotropic polypharmacy use in bipolar disorder. The evidence of a lower-than-baseline risk of DM with lamotrigine, oxcarbazepine, lithium, and bupropion monotherapy should be further investigated.

Will Artificial Intelligence Translate Big Data Into Improved Medical Care or Be a Source of Confusing Intrusion? A Discussion Between a (Cautious) Physician Informatician and an (Optimistic) Medical Informatics Researcher.

Artificial intelligence (AI), the computerized capability of doing tasks, which until recently was thought to be the exclusive domain of human intelligence, has demonstrated great strides in the past decade. The abilities to play games, provide piloting for an automobile, and respond to spoken language are remarkable successes. How are the challenges and opportunities of medicine different from these challenges and how can we best apply these data-driven techniques to patient care and outcomes? A New England Journal of Medicine paper published in 1980 suggested that more well-defined "specialized" tasks of medical care were more amenable to computer assistance, while the breadth of approach required for defining a problem and narrowing down the problem space was less so, and perhaps, unachievable. On the other hand, one can argue that the modern version of AI, which uses data-driven approaches, will be the most useful in tackling tasks such as outcome prediction that are often difficult for clinicians and patients. The ability today to collect large volumes of data about a single individual (eg, through a wearable device) and the accumulation of large datasets about multiple persons receiving medical care has the potential to apply to the care of individuals. As these techniques of analysis, enumeration, aggregation, and presentation are brought to bear in medicine, the question arises as to their utility and applicability in that domain. Early efforts in decision support were found to be helpful; as the systems proliferated, later experiences have shown difficulties such as alert fatigue and physician burnout becoming more prevalent. Will something similar arise from data-driven predictions? Will empowering patients by equipping them with information gained from data analysis help? Patients, providers, technology, and policymakers each have a role to play in the development and utilization of AI in medicine. Some of the challenges, opportunities, and tradeoffs implicit here are presented as a dialog between a clinician (SJN) and an informatician (QZT).

Comparison of 71 bipolar disorder pharmacotherapies for kidney disorder risk: The potential hazards of polypharmacy.

BACKGROUND: This study compared the largest set of bipolar disorder pharmacotherapies to date (71 drugs and drug combinations) for risk of kidney disorders (KDs). METHODS: This retrospective observational study used the IBM MarketScan® database to analyze data on 591,052 adults with bipolar disorder without prior nephropathy, for onset of KDs (of "moderate" or "high" severity) following psychopharmacotherapy (lithium, mood stabilizing anticonvulsants [MSAs], antipsychotics, antidepressants), or "No drug". Cox regression models included fixed pre-treatment covariates and time-varying drug exposure covariates to estimate the hazard ratio (HR) of each treatment versus "No drug". RESULTS: Newly observed KD occurred in 14,713 patients. No regimen had significantly lower risk of KDs than "No drug". The HR estimates ranged 0.86-2.66 for "all" KDs and 0.87-5.30 for "severe" KDs. As additional drugs were combined to compare more complex polypharmacies, higher HRs were consistently observed. Most regimens containing lithium, MSAs, or antipsychotics had a higher risk than "No drug" (p < 0.05). The risk for "all" and "severe" KDs was highest respectively on monoamine oxidase inhibitors (MAOIs) (HR = 2.66, p = 5.73 × 10-5), and a lithium-containing four-class combination (HR = 5.30, p = 2.46 × 10-9). The HR for lithium monotherapy was 1.82 (p = 4.73 × 10-17) for "severe" KDs. LIMITATIONS: The limitations inherent for an observational study were non-randomized assignment of patients to treatment groups, non-standardization of diagnostic decisions, and non-uniform quality of data collection. No correction was made for medication dosage. CONCLUSIONS: The findings support literature concerns about lithium nephrotoxicity and highlight the potential risks of MAOIs, MSAs, antipsychotics and psychotropic polypharmacy.