RESUMO
A causal effect can be defined as a comparison of outcomes that result from two or more alternative actions, with only one of the action-outcome pairs actually being observed. In healthcare, the gold standard for causal effect measurements is randomized controlled trials (RCTs), in which a target population is explicitly defined and each study sample is randomly assigned to either the treatment or control cohorts. The great potential to derive actionable insights from causal relationships has led to a growing body of machine-learning research applying causal effect estimators to observational data in the fields of healthcare, education, and economics. The primary difference between causal effect studies utilizing observational data and RCTs is that for observational data, the study occurs after the treatment, and therefore we do not have control over the treatment assignment mechanism. This can lead to massive differences in covariate distributions between control and treatment samples, making a comparison of causal effects confounded and unreliable. Classical approaches have sought to solve this problem piecemeal, first by predicting treatment assignment and then treatment effect separately. Recent work extended part of these approaches to a new family of representation-learning algorithms, showing that the upper bound of the expected treatment effect estimation error is determined by two factors: the outcome generalization-error of the representation and the distance between the treated and control distributions induced by the representation. To achieve minimal dissimilarity in learning such distributions, in this work we propose a specific auto-balancing, self-supervised objective. Experiments on real and benchmark datasets revealed that our approach consistently produced less biased estimates than previously published state-of-the-art methods. We demonstrate that the reduction in error can be directly attributed to the ability to learn representations that explicitly reduce such dissimilarity; further, in case of violations of the positivity assumption (frequent in observational data), we show our approach performs significantly better than the previous state of the art. Thus, by learning representations that induce similar distributions of the treated and control cohorts, we present evidence to support the error bound dissimilarity hypothesis as well as providing a new state-of-the-art model for causal effect estimation.
Assuntos
Algoritmos , Aprendizado de Máquina , Humanos , Causalidade , Ensaios Clínicos Controlados Aleatórios como AssuntoRESUMO
BACKGROUND: Observational studies are increasingly being used to provide supplementary evidence in addition to Randomized Control Trials (RCTs) because they provide a scale and diversity of participants and outcomes that would be infeasible in an RCT. Additionally, they more closely reflect the settings in which the studied interventions will be applied in the future. Well-established propensity-score-based methods exist to overcome the challenges of working with observational data to estimate causal effects. These methods also provide quality assurance diagnostics to evaluate the degree to which bias has been removed and the estimates can be trusted. In large medical datasets it is common to find the same underlying health condition being treated with a variety of distinct drugs or drug combinations. Conventional methods require a manual iterative workflow, making them scale poorly to studies with many intervention arms. In such situations, automated causal inference methods that are compatible with traditional propensity-score-based workflows are highly desirable. METHODS: We introduce an automated causal inference method BCAUS, that features a deep-neural-network-based propensity model that is trained with a loss which penalizes both the incorrect prediction of the assigned treatment as well as the degree of imbalance between the inverse probability weighted covariates. The network is trained end-to-end by dynamically adjusting the loss term for each training batch such that the relative contributions from the two loss components are held fixed. Trained BCAUS models can be used in conjunction with traditional propensity-score-based methods to estimate causal treatment effects. RESULTS: We tested BCAUS on the semi-synthetic Infant Health & Development Program dataset with a single intervention arm, and a real-world observational study of diabetes interventions with over 100,000 individuals spread across more than a hundred intervention arms. When compared against other recently proposed automated causal inference methods, BCAUS had competitive accuracy for estimating synthetic treatment effects and provided highly concordant estimates on the real-world dataset but was an order-of-magnitude faster. CONCLUSIONS: BCAUS is directly compatible with trusted protocols to estimate treatment effects and diagnose the quality of those estimates, while making the established approaches automatically scalable to an arbitrary number of simultaneous intervention arms without any need for manual iteration.
Assuntos
Viés , Causalidade , Humanos , Pontuação de PropensãoRESUMO
BACKGROUND: Multimorbidity clinical risk scores allow clinicians to quickly assess their patients' health for decision making, often for recommendation to care management programs. However, these scores are limited by several issues: existing multimorbidity scores (1) are generally limited to one data group (eg, diagnoses, labs) and may be missing vital information, (2) are usually limited to specific demographic groups (eg, age), and (3) do not formally provide any granularity in the form of more nuanced multimorbidity risk scores to direct clinician attention. OBJECTIVE: Using diagnosis, lab, prescription, procedure, and demographic data from electronic health records (EHRs), we developed a physiologically diverse and generalizable set of multimorbidity risk scores. METHODS: Using EHR data from a nationwide cohort of patients, we developed the total health profile, a set of six integrated risk scores reflecting five distinct organ systems and overall health. We selected the occurrence of an inpatient hospital visitation over a 2-year follow-up window, attributable to specific organ systems, as our risk endpoint. Using a physician-curated set of features, we trained six machine learning models on 794,294 patients to predict the calibrated probability of the aforementioned endpoint, producing risk scores for heart, lung, neuro, kidney, and digestive functions and a sixth score for combined risk. We evaluated the scores using a held-out test cohort of 198,574 patients. RESULTS: Study patients closely matched national census averages, with a median age of 41 years, a median income of $66,829, and racial averages by zip code of 73.8% White, 5.9% Asian, and 11.9% African American. All models were well calibrated and demonstrated strong performance with areas under the receiver operating curve (AUROCs) of 0.83 for the total health score (THS), 0.89 for heart, 0.86 for lung, 0.84 for neuro, 0.90 for kidney, and 0.83 for digestive functions. There was consistent performance of this scoring system across sexes, diverse patient ages, and zip code income levels. Each model learned to generate predictions by focusing on appropriate clinically relevant patient features, such as heart-related hospitalizations and chronic hypertension diagnosis for the heart model. The THS outperformed the other commonly used multimorbidity scoring systems, specifically the Charlson Comorbidity Index (CCI) and the Elixhauser Comorbidity Index (ECI) overall (AUROCs: THS=0.823, CCI=0.735, ECI=0.649) as well as for every age, sex, and income bracket. Performance improvements were most pronounced for middle-aged and lower-income subgroups. Ablation tests using only diagnosis, prescription, social determinants of health, and lab feature groups, while retaining procedure-related features, showed that the combination of feature groups has the best predictive performance, though only marginally better than the diagnosis-only model on at-risk groups. CONCLUSIONS: Massive retrospective EHR data sets have made it possible to use machine learning to build practical multimorbidity risk scores that are highly predictive, personalizable, intuitive to explain, and generalizable across diverse patient populations.
Assuntos
Aprendizado de Máquina , Multimorbidade , Adulto , Estudos de Coortes , Registros Eletrônicos de Saúde , Humanos , Pessoa de Meia-Idade , Estudos Retrospectivos , Fatores de RiscoRESUMO
Applying state-of-the-art machine learning techniques to medical images requires a thorough selection and normalization of input data. One of such steps in digital mammography screening for breast cancer is the labeling and removal of special diagnostic views, in which diagnostic tools or magnification are applied to assist in assessment of suspicious initial findings. As a common task in medical informatics is prediction of disease and its stage, these special diagnostic views, which are only enriched among the cohort of diseased cases, will bias machine learning disease predictions. In order to automate this process, here, we develop a machine learning pipeline that utilizes both DICOM headers and images to predict such views in an automatic manner, allowing for their removal and the generation of unbiased datasets. We achieve AUC of 99.72% in predicting special mammogram views when combining both types of models. Finally, we apply these models to clean up a dataset of about 772,000 images with expected sensitivity of 99.0%. The pipeline presented in this paper can be applied to other datasets to obtain high-quality image sets suitable to train algorithms for disease detection.
Assuntos
Neoplasias da Mama/diagnóstico por imagem , Aprendizado de Máquina , Mamografia/classificação , Mamografia/métodos , Automação , Conjuntos de Dados como Assunto , Feminino , Humanos , Sistemas de Informação em Radiologia , Sensibilidade e EspecificidadeRESUMO
With the increasing availability of rich, longitudinal, real-world clinical data recorded in electronic health records (EHRs) for millions of patients, there is a growing interest in leveraging these records to improve the understanding of human health and disease and translate these insights into clinical applications. However, there is also a need to consider the limitations of these data due to various biases and to understand the impact of missing information. Recognizing and addressing these limitations can inform the design and interpretation of EHR-based informatics studies that avoid confusing or incorrect conclusions, particularly when applied to population or precision medicine. Here we discuss key considerations in the design, implementation and interpretation of EHR-based informatics studies, drawing from examples in the literature across hypothesis generation, hypothesis testing and machine learning applications. We outline the growing opportunities for EHR-based informatics studies, including association studies and predictive modeling, enabled by evolving AI capabilities-while addressing limitations and potential pitfalls to avoid.
Assuntos
Registros Eletrônicos de Saúde , Humanos , Aprendizado de Máquina , Pesquisa Biomédica , Medicina de Precisão/métodos , Informática Médica/métodosRESUMO
Trafficking and exploitation for sex or labor affects millions of persons worldwide. To improve healthcare for these patients, in late 2018 new ICD-10 medical diagnosis codes were implemented in the US. These 13 codes include diagnosis of adult and child sexual exploitation, adult and child labor exploitation, and history of exploitation. Here we report on a database search of a large US health insurer that contained approximately 47.1 million patients and 0.9 million provider organizations, not limited to large medical systems. We reported on any diagnosis with the new codes between 2018-09-01 and 2022-09-01. The dataset was found to contain 5,262 instances of the ICD-10 codes. Regression analysis of the codes found a 5.8% increase in the uptake of these codes per year, representing a decline relative to 6.7% annual increase in the data. The codes were used by 1,810 different providers (0.19% of total) for 2,793 patients. Of the patients, 1,248 were recently trafficked, while the remainder had a personal history of exploitation. Of the recent cases, 86% experienced sexual exploitation, 14% labor exploitation and 0.8% both types. These patients were predominantly female (83%) with a median age of 20 (interquartile range: 15-35). The patients were characterized by persistently high prevalence of mental health conditions (including anxiety: 21%, post-traumatic stress disorder: 20%, major depression: 18%), sexually-transmitted infections, and high utilization of the emergency department (ED). The patients' first report of trafficking occurred most often outside of a hospital or emergency setting (55%), primarily during office and psychiatric visits.
Assuntos
Tráfico de Pessoas , Adulto , Feminino , Humanos , Masculino , Ansiedade , Atenção à Saúde , Tráfico de Pessoas/psicologia , Classificação Internacional de Doenças , Estudos Retrospectivos , Adolescente , Adulto JovemRESUMO
Opioid dependence is a national crisis, with 30 million patients annually at risk of becoming persistent opioid users after receiving opioids for post-surgical pain management. Translational Pain Services (TPS) demonstrate effectiveness for behavioral health improvements but its effectiveness in preventing persistent opioid use is less established, especially amongst opioid exposed patients. Prohibitive costs and accessibility challenges have hindered TPS program adoption. To address these limitations, we designed and implemented a remote telehealth TPS protocol focusing on preventing continued opioid use while improving behavioral health. Licensed therapists trained in the opioid-tapering CBT protocol delivered sessions reimbursed through standard payer reimbursement. Our prospective study evaluated the protocol's effectiveness on preventing persistent opioid use and behavioral health outcomes amongst both opioid naïve and exposed patients. In an opioid-naive patient cohort (n=67), 100% completely tapered off opioids, while in an opioid-exposed cohort (n =19) 52% completely tapered off opioids, demonstrating promising results. In both cohorts, we observed significant improvements in behavioral health scores, including pain. This opioid-tapering digital TPS is effective, adoptable, and incurs no out-of-pocket cost for healthcare systems. We provide the opioid-tapering CBT protocol in the supplement to facilitate adoption. Trial Registration Impact of Daily, Digital and Behavioral Tele-health Tapering Program for Perioperative Surgical Patients Exposed to Opioids and Benzodiazepines registered at clinicaltrials.gov, NCT04787692. https://clinicaltrials.gov/ct2/show/NCT04787692?term=NCT04787692&draw=2&rank=1.
RESUMO
Causal inference is a broad field that seeks to build and apply models that learn the effect of interventions on outcomes using many data types. While the field has existed for decades, its potential to impact healthcare outcomes has increased dramatically recently due to both advancements in machine learning and the unprecedented amounts of observational data resulting from electronic capture of patient claims data by medical insurance companies and widespread adoption of electronic health records (EHR) worldwide. However, there are many different schools of learning causality coming from different fields of statistics, some of them strongly conflicting. While the recent advances in machine learning greatly enhanced causal inference from a modeling perspective, it further exacerbated the fractured state in this field. This fractured state has limited research at the intersection of causal inference, modern machine learning, and EHRs that could potentially transform healthcare. In this paper we unify the classical causal inference approaches with new machine learning developments into a straightforward framework based on whether the researcher is most interested in finding the best intervention for an individual, a group of similar people, or an entire population. Through this lens, we then provide a timely review of the applications of causal inference in healthcare from the literature. As expected, we found that applications of causal inference in medicine were mostly limited to just a few technique types and lag behind other domains. In light of this gap, we offer a helpful schematic to guide data scientists and healthcare stakeholders in selecting appropriate causal methods and reviewing the findings generated by them.
RESUMO
In the past decade, there has been exponentially growing interest in the use of observational data collected as a part of routine healthcare practice to determine the effect of a treatment with causal inference models. Validation of these models, however, has been a challenge because the ground truth is unknown: only one treatment-outcome pair for each person can be observed. There have been multiple efforts to fill this void using synthetic data where the ground truth can be generated. However, to date, these datasets have been severely limited in their utility either by being modeled after small non-representative patient populations, being dissimilar to real target populations, or only providing known effects for two cohorts (treated vs. control). In this work, we produced a large-scale and realistic synthetic dataset that provides ground truth effects for over 10 hypertension treatments on blood pressure outcomes. The synthetic dataset was created by modeling a nationwide cohort of more than 580, 000 hypertension patient data including each person's multi-year history of diagnoses, medications, and laboratory values. We designed a data generation process by combining an adapted ADS-GAN model for fictitious patient information generation and a neural network for treatment outcome generation. Wasserstein distance of 0.35 demonstrates that our synthetic data follows a nearly identical joint distribution to the patient cohort used to generate the data. Patient privacy was a primary concern for this study; the ϵ-identifiability metric, which estimates the probability of actual patients being identified, is 0.008%, ensuring that our synthetic data cannot be used to identify any actual patients. To demonstrate its usage, we tested the bias in causal effect estimation of four well-established models using this dataset. The approach we used can be readily extended to other types of diseases in the clinical domain, and to datasets in other domains as well.
RESUMO
Type-2 diabetes is associated with severe health outcomes, the effects of which are responsible for approximately 1/4th of the total healthcare spending in the United States (US). Current treatment guidelines endorse a massive number of potential anti-hyperglycemic treatment options in various combinations. Strategies for optimizing treatment selection are lacking. Real-world data from a nationwide population of over one million high-risk diabetic patients (HbA1c ≥ 9%) in the US is analyzed to evaluate the comparative effectiveness for HbA1c reduction in this population of more than 80 different treatment strategies ranging from monotherapy up to combinations of five concomitant classes of drugs across each of 10 clinical cohorts defined by age, insulin dependence, and a number of other chronic conditions. A causal deep learning approach developed on such data allows for more personalized evaluation of treatment selection. An average confounder-adjusted reduction in HbA1c of 0.69% [-0.75, -0.65] is observed between patients receiving high vs low ranked treatments across cohorts for which the difference was significant. This method can be extended to explore treatment optimization for other chronic conditions.
Assuntos
Aprendizado Profundo , Diabetes Mellitus Tipo 2 , Humanos , Estados Unidos , Hipoglicemiantes/uso terapêutico , Hemoglobinas Glicadas/análise , Diabetes Mellitus Tipo 2/tratamento farmacológico , Diabetes Mellitus Tipo 2/epidemiologia , Doença CrônicaRESUMO
We conduct exploratory analysis of a novel algorithm called Model Agnostic Effect Coefficients (MAgEC) for extracting clinical features of importance when assessing an individual patient's healthcare risks, alongside predicting the risk itself. Our approach uses a non-homogeneous consensus-based algorithm to assign importance to features, which differs from similar approaches, which are homogeneous (typically purely based on random forests). Using the MIMIC-III dataset, we apply our method on predicting drivers/causers of unexpected mechanical ventilation in a large cohort patient population. We validate the MAgEC method using two primary metrics: its accuracy in predicting mechanical ventilation and the similarity of the proposed feature importances to a competing algorithm (SHAP). We also more closely discuss MAgEC itself by examining the stability of our proposed feature importances under different perturbations and whether the non-homogeneity of the approach actually leads to feature importance diversity. The code to implement MAgEC is open-sourced on GitHub (https://github.com/gstef80/MAgEC).
Assuntos
Algoritmos , Respiração Artificial , Estudos de Coortes , Consenso , HumanosRESUMO
There is a great and growing need to ascertain what exactly is the state of a patient, in terms of disease progression, actual care practices, pathology, adverse events, and much more, beyond the paucity of data available in structured medical record data. Ascertaining these harder-to-reach data elements is now critical for the accurate phenotyping of complex traits, detection of adverse outcomes, efficacy of off-label drug use, and longitudinal patient surveillance. Clinical notes often contain the most detailed and relevant digital information about individual patients, the nuances of their diseases, the treatment strategies selected by physicians, and the resulting outcomes. However, notes remain largely unused for research because they contain Protected Health Information (PHI), which is synonymous with individually identifying data. Previous clinical note de-identification approaches have been rigid and still too inaccurate to see any substantial real-world use, primarily because they have been trained with too small medical text corpora. To build a new de-identification tool, we created the largest manually annotated clinical note corpus for PHI and develop a customizable open-source de-identification software called Philter ("Protected Health Information filter"). Here we describe the design and evaluation of Philter, and show how it offers substantial real-world improvements over prior methods.
RESUMO
Importance: Knowing the future condition of a patient would enable a physician to customize current therapeutic options to prevent disease worsening, but predicting that future condition requires sophisticated modeling and information. If artificial intelligence models were capable of forecasting future patient outcomes, they could be used to aid practitioners and patients in prognosticating outcomes or simulating potential outcomes under different treatment scenarios. Objective: To assess the ability of an artificial intelligence system to prognosticate the state of disease activity of patients with rheumatoid arthritis (RA) at their next clinical visit. Design, Setting, and Participants: This prognostic study included 820 patients with RA from rheumatology clinics at 2 distinct health care systems with different electronic health record platforms: a university hospital (UH) and a public safety-net hospital (SNH). The UH and SNH had substantially different patient populations and treatment patterns. The UH has records on approximately 1 million total patients starting in January 2012. The UH data for this study were accessed on July 1, 2017. The SNH has records on 65â¯000 unique individuals starting in January 2013. The SNH data for the study were collected on February 27, 2018. Exposures: Structured data were extracted from the electronic health record, including exposures (medications), patient demographics, laboratories, and prior measures of disease activity. A longitudinal deep learning model was used to predict disease activity for patients with RA at their next rheumatology clinic visit and to evaluate interhospital performance and model interoperability strategies. Main Outcomes and Measures: Model performance was quantified using the area under the receiver operating characteristic curve (AUROC). Disease activity in RA was measured using a composite index score. Results: A total of 578 UH patients (mean [SD] age, 57 [15] years; 477 [82.5%] female; 296 [51.2%] white) and 242 SNH patients (mean [SD] age, 60 [15] years; 195 [80.6%] female; 30 [12.4%] white) were included in the study. Patients at the UH compared with those at the SNH were seen more frequently (median time between visits, 100 vs 180 days) and were more frequently prescribed higher-class medications (biologics) (364 [63.0%] vs 70 [28.9%]). At the UH, the model reached an AUROC of 0.91 (95% CI, 0.86-0.96) in a test cohort of 116 patients. The UH-trained model had an AUROC of 0.74 (95% CI, 0.65-0.83) in the SNH test cohort (n = 117) despite marked differences in the patient populations. In both settings, baseline prediction using each patients' most recent disease activity score had statistically random performance. Conclusions and Relevance: The findings suggest that building accurate models to forecast complex disease outcomes using electronic health record data is possible and these models can be shared across hospitals with diverse patient populations.