An approach to predicting patient experience through machine learning and social network analysis

ObjectiveImproving the patient experience has become an essential component of any healthcare system’s performance metrics portfolio. In this study, we developed a machine learning model to predict a patient’s response to the Hospital Consumer Assessment of Healthcare Providers and Systems survey’s “Doctor Communications” domain questions while simultaneously identifying most impactful providers in a network.Materials and MethodsThis is an observational study of patients admitted to a single tertiary care hospital between 2016 and 2020. Using machine learning algorithms, electronic health record data were used to predict patient responses to Hospital Consumer Assessment of Healthcare Providers and Systems survey questions in the doctor domain, and patients who are at risk for responding negatively were identified. Model performance was assessed by area under receiver-operating characteristic curve. Social network analysis metrics were also used to identify providers most impactful to patient experience.ResultsUsing a random forest algorithm, patients’ responses to the following 3 questions were predicted: “During this hospital stay how often did doctors. 1) treat you with courtesy and respect? 2) explain things in a way that you could understand? 3) listen carefully to you?” with areas under the receiver-operating characteristic curve of 0.876, 0.819, and 0.819, respectively. Social network analysis found that doctors with higher centrality appear to have an outsized influence on patient experience, as measured by rank in the random forest model in the doctor domain.ConclusionsA machine learning algorithm identified patients at risk of a negative experience. Furthermore, a doctor social network framework provides metrics for identifying those providers that are most influential on the patient experience.     

Bias and fairness assessment of a natural language processing opioid misuse classifier: detection and mitigation of electronic health record data disadvantages across racial subgroups

ObjectivesTo assess fairness and bias of a previously validated machine learning opioid misuse classifier.Materials & MethodsTwo experiments were conducted with the classifier’s original (n = 1000) and external validation (n = 53 974) datasets from 2 health systems. Bias was assessed via testing for differences in type II error rates across racial/ethnic subgroups (Black, Hispanic/Latinx, White, Other) using bootstrapped 95% confidence intervals. A local surrogate model was estimated to interpret the classifier’s predictions by race and averaged globally from the datasets. Subgroup analyses and post-hoc recalibrations were conducted to attempt to mitigate biased metrics.ResultsWe identified bias in the false negative rate (FNR = 0.32) of the Black subgroup compared to the FNR (0.17) of the White subgroup. Top features included “heroin” and “substance abuse” across subgroups. Post-hoc recalibrations eliminated bias in FNR with minimal changes in other subgroup error metrics. The Black FNR subgroup had higher risk scores for readmission and mortality than the White FNR subgroup, and a higher mortality risk score than the Black true positive subgroup (P < .05).DiscussionThe Black FNR subgroup had the greatest severity of disease and risk for poor outcomes. Similar features were present between subgroups for predicting opioid misuse, but inequities were present. Post-hoc mitigation techniques mitigated bias in type II error rate without creating substantial type I error rates. From model design through deployment, bias and data disadvantages should be systematically addressed.ConclusionStandardized, transparent bias assessments are needed to improve trustworthiness in clinical machine learning models.     

Predicting outcomes in central venous catheter salvage in pediatric central line–associated bloodstream infection

ObjectiveCentral line–associated bloodstream infections (CLABSIs) are a common, costly, and hazardous healthcare-associated infection in children. In children in whom continued access is critical, salvage of infected central venous catheters (CVCs) with antimicrobial lock therapy is an alternative to removal and replacement of the CVC. However, the success of CVC salvage is uncertain, and when it fails the catheter has to be removed and replaced. We describe a machine learning approach to predict individual outcomes in CVC salvage that can aid the clinician in the decision to attempt salvage.Materials and MethodsOver a 14-year period, 969 pediatric CLABSIs were identified in electronic health records. We used 164 potential predictors to derive 4 types of machine learning models to predict 2 failed salvage outcomes, infection recurrence and CVC removal, at 10 time points between 7 days and 1 year from infection onset.ResultsThe area under the receiver-operating characteristic curve varied from 0.56 to 0.83, and key predictors varied over time. The infection recurrence model performed better than the CVC removal model did.ConclusionsMachine learning–based outcome prediction can inform clinical decision making for children. We developed and evaluated several models to predict clinically relevant outcomes in the context of CVC salvage in pediatric CLABSI and illustrate the variability of predictors over time.     

Development and validation of prediction models for mechanical ventilation,renal replacement therapy,and readmission in COVID-19 patients

ObjectiveCoronavirus disease 2019 (COVID-19) patients are at risk for resource-intensive outcomes including mechanical ventilation (MV), renal replacement therapy (RRT), and readmission. Accurate outcome prognostication could facilitate hospital resource allocation. We develop and validate predictive models for each outcome using retrospective electronic health record data for COVID-19 patients treated between March 2 and May 6, 2020.Materials and MethodsFor each outcome, we trained 3 classes of prediction models using clinical data for a cohort of SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2)–positive patients (n = 2256). Cross-validation was used to select the best-performing models per the areas under the receiver-operating characteristic and precision-recall curves. Models were validated using a held-out cohort (n = 855). We measured each model’s calibration and evaluated feature importances to interpret model output.ResultsThe predictive performance for our selected models on the held-out cohort was as follows: area under the receiver-operating characteristic curve—MV 0.743 (95% CI, 0.682-0.812), RRT 0.847 (95% CI, 0.772-0.936), readmission 0.871 (95% CI, 0.830-0.917); area under the precision-recall curve—MV 0.137 (95% CI, 0.047-0.175), RRT 0.325 (95% CI, 0.117-0.497), readmission 0.504 (95% CI, 0.388-0.604). Predictions were well calibrated, and the most important features within each model were consistent with clinical intuition.DiscussionOur models produce performant, well-calibrated, and interpretable predictions for COVID-19 patients at risk for the target outcomes. They demonstrate the potential to accurately estimate outcome prognosis in resource-constrained care sites managing COVID-19 patients.ConclusionsWe develop and validate prognostic models targeting MV, RRT, and readmission for hospitalized COVID-19 patients which produce accurate, interpretable predictions. Additional external validation studies are needed to further verify the generalizability of our results.     

Using nursing notes to improve clinical outcome prediction in intensive care patients: A retrospective cohort study

ObjectiveElectronic health record documentation by intensive care unit (ICU) clinicians may predict patient outcomes. However, it is unclear whether physician and nursing notes differ in their ability to predict short-term ICU prognosis. We aimed to investigate and compare the ability of physician and nursing notes, written in the first 48 hours of admission, to predict ICU length of stay and mortality using 3 analytical methods.Materials and MethodsThis was a retrospective cohort study with split sampling for model training and testing. We included patients ≥18 years of age admitted to the ICU at Beth Israel Deaconess Medical Center in Boston, Massachusetts, from 2008 to 2012. Physician or nursing notes generated within the first 48 hours of admission were used with standard machine learning methods to predict outcomes.ResultsFor the primary outcome of composite score of ICU length of stay ≥7 days or in-hospital mortality, the gradient boosting model had better performance than the logistic regression and random forest models. Nursing and physician notes achieved area under the curves (AUCs) of 0.826 and 0.796, respectively, with even better predictive power when combined (AUC, 0.839).DiscussionModels using only nursing notes more accurately predicted short-term prognosis than did models using only physician notes, but in combination, the models achieved the greatest accuracy in prediction. ConclusionsOur findings demonstrate that statistical models derived from text analysis in the first 48 hours of ICU admission can predict patient outcomes. Physicians’ and nurses’ notes are both uniquely important in mortality prediction and combining these notes can produce a better predictive model.     

Electronic phenotyping of health outcomes of interest using a linked claims-electronic health record database: Findings from a machine learning pilot project

ObjectiveClaims-based algorithms are used in the Food and Drug Administration Sentinel Active Risk Identification and Analysis System to identify occurrences of health outcomes of interest (HOIs) for medical product safety assessment. This project aimed to apply machine learning classification techniques to demonstrate the feasibility of developing a claims-based algorithm to predict an HOI in structured electronic health record (EHR) data.Materials and MethodsWe used the 2015-2019 IBM MarketScan Explorys Claims-EMR Data Set, linking administrative claims and EHR data at the patient level. We focused on a single HOI, rhabdomyolysis, defined by EHR laboratory test results. Using claims-based predictors, we applied machine learning techniques to predict the HOI: logistic regression, LASSO (least absolute shrinkage and selection operator), random forests, support vector machines, artificial neural nets, and an ensemble method (Super Learner).ResultsThe study cohort included 32 956 patients and 39 499 encounters. Model performance (positive predictive value [PPV], sensitivity, specificity, area under the receiver-operating characteristic curve) varied considerably across techniques. The area under the receiver-operating characteristic curve exceeded 0.80 in most model variations.DiscussionFor the main Food and Drug Administration use case of assessing risk of rhabdomyolysis after drug use, a model with a high PPV is typically preferred. The Super Learner ensemble model without adjustment for class imbalance achieved a PPV of 75.6%, substantially better than a previously used human expert-developed model (PPV = 44.0%).ConclusionsIt is feasible to use machine learning methods to predict an EHR-derived HOI with claims-based predictors. Modeling strategies can be adapted for intended uses, including surveillance, identification of cases for chart review, and outcomes research.     

Developing machine learning models to personalize care levels among emergency room patients for hospital admission

ObjectiveTo develop prediction models for intensive care unit (ICU) vs non-ICU level-of-care need within 24 hours of inpatient admission for emergency department (ED) patients using electronic health record data.Materials and MethodsUsing records of 41 654 ED visits to a tertiary academic center from 2015 to 2019, we tested 4 algorithms—feed-forward neural networks, regularized regression, random forests, and gradient-boosted trees—to predict ICU vs non-ICU level-of-care within 24 hours and at the 24th hour following admission. Simple-feature models included patient demographics, Emergency Severity Index (ESI), and vital sign summary. Complex-feature models added all vital signs, lab results, and counts of diagnosis, imaging, procedures, medications, and lab orders.ResultsThe best-performing model, a gradient-boosted tree using a full feature set, achieved an AUROC of 0.88 (95%CI: 0.87–0.89) and AUPRC of 0.65 (95%CI: 0.63–0.68) for predicting ICU care need within 24 hours of admission. The logistic regression model using ESI achieved an AUROC of 0.67 (95%CI: 0.65–0.70) and AUPRC of 0.37 (95%CI: 0.35–0.40). Using a discrimination threshold, such as 0.6, the positive predictive value, negative predictive value, sensitivity, and specificity were 85%, 89%, 30%, and 99%, respectively. Vital signs were the most important predictors.Discussion and ConclusionsUndertriaging admitted ED patients who subsequently require ICU care is common and associated with poorer outcomes. Machine learning models using readily available electronic health record data predict subsequent need for ICU admission with good discrimination, substantially better than the benchmarking ESI system. The results could be used in a multitiered clinical decision-support system to improve ED triage.     

Improving the delivery of palliative care through predictive modeling and healthcare informatics

ObjectiveAccess to palliative care (PC) is important for many patients with uncontrolled symptom burden from serious or complex illness. However, many patients who could benefit from PC do not receive it early enough or at all. We sought to address this problem by building a predictive model into a comprehensive clinical framework with the aims to (i) identify in-hospital patients likely to benefit from a PC consult, and (ii) intervene on such patients by contacting their care team.Materials and MethodsElectronic health record data for 68 349 inpatient encounters in 2017 at a large hospital were used to train a model to predict the need for PC consult. This model was published as a web service, connected to institutional data pipelines, and consumed by a downstream display application monitored by the PC team. For those patients that the PC team deems appropriate, a team member then contacts the patient’s corresponding care team.ResultsTraining performance AUC based on a 20% holdout validation set was 0.90. The most influential variables were previous palliative care, hospital unit, Albumin, Troponin, and metastatic cancer. The model has been successfully integrated into the clinical workflow making real-time predictions on hundreds of patients per day. The model had an “in-production” AUC of 0.91. A clinical trial is currently underway to assess the effect on clinical outcomes.ConclusionsA machine learning model can effectively predict the need for an inpatient PC consult and has been successfully integrated into practice to refer new patients to PC.     

Machine learning for initial insulin estimation in hospitalized patients

ObjectiveThe study sought to determine whether machine learning can predict initial inpatient total daily dose (TDD) of insulin from electronic health records more accurately than existing guideline-based dosing recommendations.Materials and MethodsUsing electronic health records from a tertiary academic center between 2008 and 2020 of 16,848 inpatients receiving subcutaneous insulin who achieved target blood glucose control of 100-180 mg/dL on a calendar day, we trained an ensemble machine learning algorithm consisting of regularized regression, random forest, and gradient boosted tree models for 2-stage TDD prediction. We evaluated the ability to predict patients requiring more than 6 units TDD and their point-value TDDs to achieve target glucose control.ResultsThe method achieves an area under the receiver-operating characteristic curve of 0.85 (95% confidence interval [CI], 0.84-0.87) and area under the precision-recall curve of 0.65 (95% CI, 0.64-0.67) for classifying patients who require more than 6 units TDD. For patients requiring more than 6 units TDD, the mean absolute percent error in dose prediction based on standard clinical calculators using patient weight is in the range of 136%-329%, while the regression model based on weight improves to 60% (95% CI, 57%-63%), and the full ensemble model further improves to 51% (95% CI, 48%-54%).DiscussionOwingto the narrow therapeutic window and wide individual variability, insulin dosing requires adaptive and predictive approaches that can be supported through data-driven analytic tools.ConclusionsMachine learning approaches based on readily available electronic medical records can discriminate which inpatients will require more than 6 units TDD and estimate individual doses more accurately than standard guidelines and practices.     

Predicting brain function status changes in critically ill patients via Machine learning

ObjectiveIn intensive care units (ICUs), a patient’s brain function status can shift from a state of acute brain dysfunction (ABD) to one that is ABD-free and vice versa, which is challenging to forecast and, in turn, hampers the allocation of hospital resources. We aim to develop a machine learning model to predict next-day brain function status changes.Materials and MethodsUsing multicenter prospective adult cohorts involving medical and surgical ICU patients from 2 civilian and 3 Veteran Affairs hospitals, we trained and externally validated a light gradient boosting machine to predict brain function status changes. We compared the performances of the boosting model against state-of-the-art models—an ABD predictive model and its variants. We applied Shapley additive explanations to identify influential factors to develop a compact model.ResultsThere were 1026 critically ill patients without evidence of prior major dementia, or structural brain diseases, from whom 12 295 daily transitions (ABD: 5847 days; ABD-free: 6448 days) were observed. The boosting model achieved an area under the receiver-operating characteristic curve (AUROC) of 0.824 (95% confidence interval [CI], 0.821-0.827), compared with the state-of-the-art models of 0.697 (95% CI, 0.693-0.701) with P < .001. Using 13 identified top influential factors, the compact model achieved 99.4% of the boosting model on AUROC. The boosting and the compact models demonstrated high generalizability in external validation by achieving an AUROC of 0.812 (95% CI, 0.812-0.813).ConclusionThe inputs of the compact model are based on several simple questions that clinicians can quickly answer in practice, which demonstrates the model has direct prospective deployment potential into clinical practice, aiding in critical hospital resource allocation.     

Development of a “meta-model” to address missing data,predict patient-specific cancer survival and provide a foundation for clinical decision support

ObjectiveLike most real-world data, electronic health record (EHR)–derived data from oncology patients typically exhibits wide interpatient variability in terms of available data elements. This interpatient variability leads to missing data and can present critical challenges in developing and implementing predictive models to underlie clinical decision support for patient-specific oncology care. Here, we sought to develop a novel ensemble approach to addressing missing data that we term the “meta-model” and apply the meta-model to patient-specific cancer prognosis.Materials and MethodsUsing real-world data, we developed a suite of individual random survival forest models to predict survival in patients with advanced lung cancer, colorectal cancer, and breast cancer. Individual models varied by the predictor data used. We combined models for each cancer type into a meta-model that predicted survival for each patient using a weighted mean of the individual models for which the patient had all requisite predictors.ResultsThe meta-model significantly outperformed many of the individual models and performed similarly to the best performing individual models. Comparisons of the meta-model to a more traditional imputation-based method of addressing missing data supported the meta-model’s utility.ConclusionsWe developed a novel machine learning–based strategy to underlie clinical decision support and predict survival in cancer patients, despite missing data. The meta-model may more generally provide a tool for addressing missing data across a variety of clinical prediction problems. Moreover, the meta-model may address other challenges in clinical predictive modeling including model extensibility and integration of predictive algorithms trained across different institutions and datasets.     

STAN: spatio-temporal attention network for pandemic prediction using real-world evidence

ObjectiveWe aim to develop a hybrid model for earlier and more accurate predictions for the number of infected cases in pandemics by (1) using patients’ claims data from different counties and states that capture local disease status and medical resource utilization; (2) utilizing demographic similarity and geographical proximity between locations; and (3) integrating pandemic transmission dynamics into a deep learning model.Materials and MethodsWe proposed a spatio-temporal attention network (STAN) for pandemic prediction. It uses a graph attention network to capture spatio-temporal trends of disease dynamics and to predict the number of cases for a fixed number of days into the future. We also designed a dynamics-based loss term for enhancing long-term predictions. STAN was tested using both real-world patient claims data and COVID-19 statistics over time across US counties.ResultsSTAN outperforms traditional epidemiological models such as susceptible-infectious-recovered (SIR), susceptible-exposed-infectious-recovered (SEIR), and deep learning models on both long-term and short-term predictions, achieving up to 87% reduction in mean squared error compared to the best baseline prediction model.ConclusionsBy combining information from real-world claims data and disease case counts data, STAN can better predict disease status and medical resource utilization.     

Automated model versus treating physician for predicting survival time of patients with metastatic cancer

ObjectiveBeing able to predict a patient’s life expectancy can help doctors and patients prioritize treatments and supportive care. For predicting life expectancy, physicians have been shown to outperform traditional models that use only a few predictor variables. It is possible that a machine learning model that uses many predictor variables and diverse data sources from the electronic medical record can improve on physicians’ performance. For patients with metastatic cancer, we compared accuracy of life expectancy predictions by the treating physician, a machine learning model, and a traditional model.Materials and MethodsA machine learning model was trained using 14 600 metastatic cancer patients’ data to predict each patient’s distribution of survival time. Data sources included note text, laboratory values, and vital signs. From 2015–2016, 899 patients receiving radiotherapy for metastatic cancer were enrolled in a study in which their radiation oncologist estimated life expectancy. Survival predictions were also made by the machine learning model and a traditional model using only performance status. Performance was assessed with area under the curve for 1-year survival and calibration plots.ResultsThe radiotherapy study included 1190 treatment courses in 899 patients. A total of 879 treatment courses in 685 patients were included in this analysis. Median overall survival was 11.7 months. Physicians, machine learning model, and traditional model had area under the curve for 1-year survival of 0.72 (95% CI 0.63–0.81), 0.77 (0.73–0.81), and 0.68 (0.65–0.71), respectively.ConclusionsThe machine learning model’s predictions were more accurate than those of the treating physician or a traditional model.     

Importance-aware personalized learning for early risk prediction using static and dynamic health data

ObjectiveAccurate risk prediction is important for evaluating early medical treatment effects and improving health care quality. Existing methods are usually designed for dynamic medical data, which require long-term observations. Meanwhile, important personalized static information is ignored due to the underlying uncertainty and unquantifiable ambiguity. It is urgent to develop an early risk prediction method that can adaptively integrate both static and dynamic health data.Materials and MethodsData were from 6367 patients with Peptic Ulcer Bleeding between 2007 and 2016. This article develops a novel End-to-end Importance-Aware Personalized Deep Learning Approach (eiPDLA) to achieve accurate early clinical risk prediction. Specifically, eiPDLA introduces a long short-term memory with temporal attention to learn sequential dependencies from time-stamped records and simultaneously incorporating a residual network with correlation attention to capture their influencing relationship with static medical data. Furthermore, a new multi-residual multi-scale network with the importance-aware mechanism is designed to adaptively fuse the learned multisource features, automatically assigning larger weights to important features while weakening the influence of less important features.ResultsExtensive experimental results on a real-world dataset illustrate that our method significantly outperforms the state-of-the-arts for early risk prediction under various settings (eg, achieving an AUC score of 0.944 at 1 year ahead of risk prediction). Case studies indicate that the achieved prediction results are highly interpretable.ConclusionThese results reflect the importance of combining static and dynamic health data, mining their influencing relationship, and incorporating the importance-aware mechanism to automatically identify important features. The achieved accurate early risk prediction results save precious time for doctors to timely design effective treatments and improve clinical outcomes.     

Machine-learning model to predict the cause of death using a stacking ensemble method for observational data

ObjectiveCause of death is used as an important outcome of clinical research; however, access to cause-of-death data is limited. This study aimed to develop and validate a machine-learning model that predicts the cause of death from the patient’s last medical checkup.Materials and MethodsTo classify the mortality status and each individual cause of death, we used a stacking ensemble method. The prediction outcomes were all-cause mortality, 8 leading causes of death in South Korea, and other causes. The clinical data of study populations were extracted from the national claims (n = 174 747) and electronic health records (n = 729 065) and were used for model development and external validation. Moreover, we imputed the cause of death from the data of 3 US claims databases (n = 994 518, 995 372, and 407 604, respectively). All databases were formatted to the Observational Medical Outcomes Partnership Common Data Model.ResultsThe generalized area under the receiver operating characteristic curve (AUROC) of the model predicting the cause of death within 60 days was 0.9511. Moreover, the AUROC of the external validation was 0.8887. Among the causes of death imputed in the Medicare Supplemental database, 11.32% of deaths were due to malignant neoplastic disease.DiscussionThis study showed the potential of machine-learning models as a new alternative to address the lack of access to cause-of-death data. All processes were disclosed to maintain transparency, and the model was easily applicable to other institutions.ConclusionA machine-learning model with competent performance was developed to predict cause of death.     

Bayesian logistic regression for online recalibration and revision of risk prediction models with performance guarantees

ObjectiveAfter deploying a clinical prediction model, subsequently collected data can be used to fine-tune its predictions and adapt to temporal shifts. Because model updating carries risks of over-updating/fitting, we study online methods with performance guarantees. Materials and MethodsWe introduce 2 procedures for continual recalibration or revision of an underlying prediction model: Bayesian logistic regression (BLR) and a Markov variant that explicitly models distribution shifts (MarBLR). We perform empirical evaluation via simulations and a real-world study predicting Chronic Obstructive Pulmonary Disease (COPD) risk. We derive “Type I and II” regret bounds, which guarantee the procedures are noninferior to a static model and competitive with an oracle logistic reviser in terms of the average loss.ResultsBoth procedures consistently outperformed the static model and other online logistic revision methods. In simulations, the average estimated calibration index (aECI) of the original model was 0.828 (95%CI, 0.818–0.938). Online recalibration using BLR and MarBLR improved the aECI towards the ideal value of zero, attaining 0.265 (95%CI, 0.230–0.300) and 0.241 (95%CI, 0.216–0.266), respectively. When performing more extensive logistic model revisions, BLR and MarBLR increased the average area under the receiver-operating characteristic curve (aAUC) from 0.767 (95%CI, 0.765–0.769) to 0.800 (95%CI, 0.798–0.802) and 0.799 (95%CI, 0.797–0.801), respectively, in stationary settings and protected against substantial model decay. In the COPD study, BLR and MarBLR dynamically combined the original model with a continually refitted gradient boosted tree to achieve aAUCs of 0.924 (95%CI, 0.913–0.935) and 0.925 (95%CI, 0.914–0.935), compared to the static model’s aAUC of 0.904 (95%CI, 0.892–0.916).DiscussionDespite its simplicity, BLR is highly competitive with MarBLR. MarBLR outperforms BLR when its prior better reflects the data.ConclusionsBLR and MarBLR can improve the transportability of clinical prediction models and maintain their performance over time.