Adler Perotte
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Adler Perotte has a total of 207 co-authors. Here are the top co-authors by number of shared publications:
George Hripcsak
Steele Institute for Health Innovation, Geisinger, Danville, Pennsylvania and Department of Biomedical Informatics, Columbia University, New York, New York
6 shared publications- Chunhua Weng
Department of Biomedical Informatics, Columbia University, New York, USA
4 shared publications - George Hripcsak
Department of Biomedical Informatics, Columbia University Medical Center, New York, USA
4 shared publications Noémie Elhadad
Department of Biomedical Informatics, Columbia University, New York, New York, USA
4 shared publications- Bowen Zhou
Tsinghua University
2 shared publications Carol Friedman
Department of Biomedical Informatics, Columbia University, New York, New York, USA
2 shared publications
Latest Publications
- Open Access
- 29 January 2025
- Published by SAGE Publications in Journal of Diabetes Science and Technology
- Vol. 20 (3) , 922-929
A Prospective Pilot Study Demonstrating Noninvasive Calibration-Free Glucose Measurement
Background: Glucose is an essential molecule in energy metabolism. Dysregulated glucose metabolism, the defining feature of diabetes, requires active monitoring and treatment to prevent significant morbidity and mortality. Current technologies for intermittent and continuous glucose measurement are invasive. Noninvasive glucose measurement would eliminate this barrier toward making glucose monitoring more accessible, extending the benefits from people living with diabetes to prediabetes and the healthy. Methods: A novel spectroscopy-based system for measuring glucose noninvasively was used in an exploratory, prospective, single-center clinical study (NCT06272136) to develop and test a machine learning-based computational model for continuous glucose monitoring without per-subject calibration. The study design blinded the development investigators to the validation analyses. Results: Twenty subjects were enrolled. Fifteen were used for the development set, and five in the validation set. All study participants were adults with insulin-treated diabetes and median glycated hemoglobin (HbA 1c ) of 7.3% (interquartile range [IQR] = 6.7-7.7). The computational model resulted in a mean absolute relative difference (MARD) of 14.5% and 96.5% of the paired glucose data points in the A plus B zones of the Diabetes Technology Society (DTS) error grid. The correlation between the average model sensitivity by wavelength and the spectrum of glucose was 0.45 ( P < .001). Conclusions: Our findings suggest that Raman spectroscopy coupled with advanced computational methods can enable continuous, noninvasive glucose measurement without per-subject invasive calibration.
- 29 January 2025
- Published by SAGE Publications in Journal of Diabetes Science and Technology
- Vol. 20 (3) , 922-929
Open AccessA Prospective Pilot Study Demonstrating Noninvasive Calibration-Free Glucose Measurement
Background: Glucose is an essential molecule in energy metabolism. Dysregulated glucose metabolism, the defining feature of diabetes, requires active monitoring and treatment to prevent significant morbidity and mortality. Current technologies for intermittent and continuous glucose measurement are invasive. Noninvasive glucose measurement would eliminate this barrier toward making glucose monitoring more accessible, extending the benefits from people living with diabetes to prediabetes and the healthy. Methods: A novel spectroscopy-based system for measuring glucose noninvasively was used in an exploratory, prospective, single-center clinical study (NCT06272136) to develop and test a machine learning-based computational model for continuous glucose monitoring without per-subject calibration. The study design blinded the development investigators to the validation analyses. Results: Twenty subjects were enrolled. Fifteen were used for the development set, and five in the validation set. All study participants were adults with insulin-treated diabetes and median glycated hemoglobin (HbA 1c ) of 7.3% (interquartile range [IQR] = 6.7-7.7). The computational model resulted in a mean absolute relative difference (MARD) of 14.5% and 96.5% of the paired glucose data points in the A plus B zones of the Diabetes Technology Society (DTS) error grid. The correlation between the average model sensitivity by wavelength and the spectrum of glucose was 0.45 ( P < .001). Conclusions: Our findings suggest that Raman spectroscopy coupled with advanced computational methods can enable continuous, noninvasive glucose measurement without per-subject invasive calibration.
- Open Access
Maximum Likelihood Estimation of Flexible Survival Densities with Importance Sampling
Survival analysis is a widely-used technique for analyzing time-to-event data in the presence of censoring. In recent years, numerous survival analysis methods have emerged which scale to large datasets and relax traditional assumptions such as proportional hazards. These models, while being performant, are very sensitive to model hyperparameters including: (1) number of bins and bin size for discrete models and (2) number of cluster assignments for mixture-based models. Each of these choices requires extensive tuning by practitioners to achieve optimal performance. In addition, we demonstrate in empirical studies that: (1) optimal bin size may drastically differ based on the metric of interest (e.g., concordance vs brier score), and (2) mixture models may suffer from mode collapse and numerical instability. We propose a survival analysis approach which eliminates the need to tune hyperparameters such as mixture assignments and bin sizes, reducing the burden on practitioners. We show that the proposed approach matches or outperforms baselines on several real-world datasets.
Open AccessMaximum Likelihood Estimation of Flexible Survival Densities with Importance Sampling
Survival analysis is a widely-used technique for analyzing time-to-event data in the presence of censoring. In recent years, numerous survival analysis methods have emerged which scale to large datasets and relax traditional assumptions such as proportional hazards. These models, while being performant, are very sensitive to model hyperparameters including: (1) number of bins and bin size for discrete models and (2) number of cluster assignments for mixture-based models. Each of these choices requires extensive tuning by practitioners to achieve optimal performance. In addition, we demonstrate in empirical studies that: (1) optimal bin size may drastically differ based on the metric of interest (e.g., concordance vs brier score), and (2) mixture models may suffer from mode collapse and numerical instability. We propose a survival analysis approach which eliminates the need to tune hyperparameters such as mixture assignments and bin sizes, reducing the burden on practitioners. We show that the proposed approach matches or outperforms baselines on several real-world datasets.
- 15 March 2023
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 30 (6) , 1022-1031
EvidenceMap: a three-level knowledge representation for medical evidence computation and comprehension
To develop a computable representation for medical evidence and to contribute a gold standard dataset of annotated randomized controlled trial (RCT) abstracts, along with a natural language processing (NLP) pipeline for transforming free-text RCT evidence in PubMed into the structured representation. Our representation, EvidenceMap, consists of 3 levels of abstraction: Medical Evidence Entity, Proposition and Map, to represent the hierarchical structure of medical evidence composition. Randomly selected RCT abstracts were annotated following EvidenceMap based on the consensus of 2 independent annotators to train an NLP pipeline. Via a user study, we measured how the EvidenceMap improved evidence comprehension and analyzed its representative capacity by comparing the evidence annotation with EvidenceMap representation and without following any specific guidelines. Two corpora including 229 disease-agnostic and 80 COVID-19 RCT abstracts were annotated, yielding 12 725 entities and 1602 propositions. EvidenceMap saves users 51.9% of the time compared to reading raw-text abstracts. Most evidence elements identified during the freeform annotation were successfully represented by EvidenceMap, and users gave the enrollment, study design, and study Results sections mean 5-scale Likert ratings of 4.85, 4.70, and 4.20, respectively. The end-to-end evaluations of the pipeline show that the evidence proposition formulation achieves F1 scores of 0.84 and 0.86 in the adjusted random index score. EvidenceMap extends the participant, intervention, comparator, and outcome framework into 3 levels of abstraction for transforming free-text evidence from the clinical literature into a computable structure. It can be used as an interoperable format for better evidence retrieval and synthesis and an interpretable representation to efficiently comprehend RCT findings.
- 15 March 2023
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 30 (6) , 1022-1031
EvidenceMap: a three-level knowledge representation for medical evidence computation and comprehension
To develop a computable representation for medical evidence and to contribute a gold standard dataset of annotated randomized controlled trial (RCT) abstracts, along with a natural language processing (NLP) pipeline for transforming free-text RCT evidence in PubMed into the structured representation. Our representation, EvidenceMap, consists of 3 levels of abstraction: Medical Evidence Entity, Proposition and Map, to represent the hierarchical structure of medical evidence composition. Randomly selected RCT abstracts were annotated following EvidenceMap based on the consensus of 2 independent annotators to train an NLP pipeline. Via a user study, we measured how the EvidenceMap improved evidence comprehension and analyzed its representative capacity by comparing the evidence annotation with EvidenceMap representation and without following any specific guidelines. Two corpora including 229 disease-agnostic and 80 COVID-19 RCT abstracts were annotated, yielding 12 725 entities and 1602 propositions. EvidenceMap saves users 51.9% of the time compared to reading raw-text abstracts. Most evidence elements identified during the freeform annotation were successfully represented by EvidenceMap, and users gave the enrollment, study design, and study Results sections mean 5-scale Likert ratings of 4.85, 4.70, and 4.20, respectively. The end-to-end evaluations of the pipeline show that the evidence proposition formulation achieves F1 scores of 0.84 and 0.86 in the adjusted random index score. EvidenceMap extends the participant, intervention, comparator, and outcome framework into 3 levels of abstraction for transforming free-text evidence from the clinical literature into a computable structure. It can be used as an interoperable format for better evidence retrieval and synthesis and an interpretable representation to efficiently comprehend RCT findings.
- Open Access
- 19 January 2022
- Published by Elsevier in American Journal of Transplantation
- Vol. 22 (5) , 1372-1381
Deceased donor kidneys allocated out of sequence by organ procurement organizations
- 19 January 2022
- Published by Elsevier in American Journal of Transplantation
- Vol. 22 (5) , 1372-1381
Open AccessDeceased donor kidneys allocated out of sequence by organ procurement organizations
- Open Access
CEHR-BERT: Incorporating temporal information from structured EHR data to improve prediction tasks
Embedding algorithms are increasingly used to represent clinical concepts in healthcare for improving machine learning tasks such as clinical phenotyping and disease prediction. Recent studies have adapted state-of-the-art bidirectional encoder representations from transformers (BERT) architecture to structured electronic health records (EHR) data for the generation of contextualized concept embeddings, yet do not fully incorporate temporal data across multiple clinical domains. Therefore we developed a new BERT adaptation, CEHR-BERT, to incorporate temporal information using a hybrid approach by augmenting the input to BERT using artificial time tokens, incorporating time, age, and concept embeddings, and introducing a new second learning objective for visit type. CEHR-BERT was trained on a subset of Columbia University Irving Medical Center-York Presbyterian Hospital's clinical data, which includes 2.4M patients, spanning over three decades, and tested using 4-fold cross-validation on the following prediction tasks: hospitalization, death, new heart failure (HF) diagnosis, and HF readmission. Our experiments show that CEHR-BERT outperformed existing state-of-the-art clinical BERT adaptations and baseline models across all 4 prediction tasks in both ROC-AUC and PR-AUC. CEHR-BERT also demonstrated strong transfer learning capability, as our model trained on only 5% of data outperformed comparison models trained on the entire data set. Ablation studies to better understand the contribution of each time component showed incremental gains with every element, suggesting that CEHR-BERT's incorporation of artificial time tokens, time and age embeddings with concept embeddings, and the addition of the second learning objective represents a promising approach for future BERT-based clinical embeddings.
Open AccessCEHR-BERT: Incorporating temporal information from structured EHR data to improve prediction tasks
Embedding algorithms are increasingly used to represent clinical concepts in healthcare for improving machine learning tasks such as clinical phenotyping and disease prediction. Recent studies have adapted state-of-the-art bidirectional encoder representations from transformers (BERT) architecture to structured electronic health records (EHR) data for the generation of contextualized concept embeddings, yet do not fully incorporate temporal data across multiple clinical domains. Therefore we developed a new BERT adaptation, CEHR-BERT, to incorporate temporal information using a hybrid approach by augmenting the input to BERT using artificial time tokens, incorporating time, age, and concept embeddings, and introducing a new second learning objective for visit type. CEHR-BERT was trained on a subset of Columbia University Irving Medical Center-York Presbyterian Hospital's clinical data, which includes 2.4M patients, spanning over three decades, and tested using 4-fold cross-validation on the following prediction tasks: hospitalization, death, new heart failure (HF) diagnosis, and HF readmission. Our experiments show that CEHR-BERT outperformed existing state-of-the-art clinical BERT adaptations and baseline models across all 4 prediction tasks in both ROC-AUC and PR-AUC. CEHR-BERT also demonstrated strong transfer learning capability, as our model trained on only 5% of data outperformed comparison models trained on the entire data set. Ablation studies to better understand the contribution of each time component showed incremental gains with every element, suggesting that CEHR-BERT's incorporation of artificial time tokens, time and age embeddings with concept embeddings, and the addition of the second learning objective represents a promising approach for future BERT-based clinical embeddings.
- 21 June 2021
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 28 (9) , 1970-1976
Clinically relevant pretraining is all you need
Clinical notes present a wealth of information for applications in the clinical domain, but heterogeneity across clinical institutions and settings presents challenges for their processing. The clinical natural language processing field has made strides in overcoming domain heterogeneity, while pretrained deep learning models present opportunities to transfer knowledge from one task to another. Pretrained models have performed well when transferred to new tasks; however, it is not well understood if these models generalize across differences in institutions and settings within the clinical domain. We explore if institution or setting specific pretraining is necessary for pretrained models to perform well when transferred to new tasks. We find no significant performance difference between models pretrained across institutions and settings, indicating that clinically pretrained models transfer well across such boundaries. Given a clinically pretrained model, clinical natural language processing researchers may forgo the time-consuming pretraining step without a significant performance drop.
- 21 June 2021
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 28 (9) , 1970-1976
Clinically relevant pretraining is all you need
Clinical notes present a wealth of information for applications in the clinical domain, but heterogeneity across clinical institutions and settings presents challenges for their processing. The clinical natural language processing field has made strides in overcoming domain heterogeneity, while pretrained deep learning models present opportunities to transfer knowledge from one task to another. Pretrained models have performed well when transferred to new tasks; however, it is not well understood if these models generalize across differences in institutions and settings within the clinical domain. We explore if institution or setting specific pretraining is necessary for pretrained models to perform well when transferred to new tasks. We find no significant performance difference between models pretrained across institutions and settings, indicating that clinically pretrained models transfer well across such boundaries. Given a clinically pretrained model, clinical natural language processing researchers may forgo the time-consuming pretraining step without a significant performance drop.
- 6 May 2021
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 28 (8) , 1703-1711
A neuro-symbolic method for understanding free-text medical evidence
We introduce Medical evidence Dependency (MD)–informed attention, a novel neuro-symbolic model for understanding free-text clinical trial publications with generalizability and interpretability. We trained one head in the multi-head self-attention model to attend to the Medical evidence Ddependency (MD) and to pass linguistic and domain knowledge on to later layers (MD informed). This MD-informed attention model was integrated into BioBERT and tested on 2 public machine reading comprehension benchmarks for clinical trial publications: Evidence Inference 2.0 and PubMedQA. We also curated a small set of recently published articles reporting randomized controlled trials on COVID-19 (coronavirus disease 2019) following the Evidence Inference 2.0 guidelines to evaluate the model’s robustness to unseen data. The integration of MD-informed attention head improves BioBERT substantially in both benchmark tasks—as large as an increase of +30% in the F1 score—and achieves the new state-of-the-art performance on the Evidence Inference 2.0. It achieves 84% and 82% in overall accuracy and F1 score, respectively, on the unseen COVID-19 data. MD-informed attention empowers neural reading comprehension models with interpretability and generalizability via reusable domain knowledge. Its compositionality can benefit any transformer-based architecture for machine reading comprehension of free-text medical evidence.
- 6 May 2021
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 28 (8) , 1703-1711
A neuro-symbolic method for understanding free-text medical evidence
We introduce Medical evidence Dependency (MD)–informed attention, a novel neuro-symbolic model for understanding free-text clinical trial publications with generalizability and interpretability. We trained one head in the multi-head self-attention model to attend to the Medical evidence Ddependency (MD) and to pass linguistic and domain knowledge on to later layers (MD informed). This MD-informed attention model was integrated into BioBERT and tested on 2 public machine reading comprehension benchmarks for clinical trial publications: Evidence Inference 2.0 and PubMedQA. We also curated a small set of recently published articles reporting randomized controlled trials on COVID-19 (coronavirus disease 2019) following the Evidence Inference 2.0 guidelines to evaluate the model’s robustness to unseen data. The integration of MD-informed attention head improves BioBERT substantially in both benchmark tasks—as large as an increase of +30% in the F1 score—and achieves the new state-of-the-art performance on the Evidence Inference 2.0. It achieves 84% and 82% in overall accuracy and F1 score, respectively, on the unseen COVID-19 data. MD-informed attention empowers neural reading comprehension models with interpretability and generalizability via reusable domain knowledge. Its compositionality can benefit any transformer-based architecture for machine reading comprehension of free-text medical evidence.
- Open Access
- 5 January 2021
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 10 (1) , e018476
Admission Cardiac Diagnostic Testing with Electrocardiography and Troponin Measurement Prognosticates Increased 30‐Day Mortality in COVID‐19
Background: Cardiovascular involvement in coronavirus disease 2019 (COVID‐19) is common and leads to worsened mortality. Diagnostic cardiovascular studies may be helpful for resource appropriation and identifying patients at increased risk for death. Methods and Results: We analyzed 887 patients (aged 64±17 years) admitted with COVID‐19 from March 1 to April 3, 2020 in New York City with 12 lead electrocardiography within 2 days of diagnosis. Demographics, comorbidities, and laboratory testing, including high sensitivity cardiac troponin T (hs‐cTnT), were abstracted. At 30 days follow‐up, 556 patients (63%) were living without requiring mechanical ventilation, 123 (14%) were living and required mechanical ventilation, and 203 (23%) had expired. Electrocardiography findings included atrial fibrillation or atrial flutter (AF/AFL) in 46 (5%) and ST‐T wave changes in 306 (38%). 27 (59%) patients with AF/AFL expired as compared to 181 (21%) of 841 with other non‐life‐threatening rhythms ( P <0.001). Multivariable analysis incorporating age, comorbidities, AF/AFL, QRS abnormalities, and ST‐T wave changes, and initial hs‐cTnT ≥20 ng/L showed that increased age (HR 1.04/year), elevated hs‐cTnT (HR 4.57), AF/AFL (HR 2.07), and a history of coronary artery disease (HR 1.56) and active cancer (HR 1.87) were associated with increased mortality. Conclusions: Myocardial injury with hs‐cTnT ≥20 ng/L, in addition to cardiac conduction perturbations, especially AF/AFL, upon hospital admission for COVID‐19 infection is associated with markedly increased risk for mortality than either diagnostic abnormality alone.
- 5 January 2021
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 10 (1) , e018476
Open AccessAdmission Cardiac Diagnostic Testing with Electrocardiography and Troponin Measurement Prognosticates Increased 30‐Day Mortality in COVID‐19
Background: Cardiovascular involvement in coronavirus disease 2019 (COVID‐19) is common and leads to worsened mortality. Diagnostic cardiovascular studies may be helpful for resource appropriation and identifying patients at increased risk for death. Methods and Results: We analyzed 887 patients (aged 64±17 years) admitted with COVID‐19 from March 1 to April 3, 2020 in New York City with 12 lead electrocardiography within 2 days of diagnosis. Demographics, comorbidities, and laboratory testing, including high sensitivity cardiac troponin T (hs‐cTnT), were abstracted. At 30 days follow‐up, 556 patients (63%) were living without requiring mechanical ventilation, 123 (14%) were living and required mechanical ventilation, and 203 (23%) had expired. Electrocardiography findings included atrial fibrillation or atrial flutter (AF/AFL) in 46 (5%) and ST‐T wave changes in 306 (38%). 27 (59%) patients with AF/AFL expired as compared to 181 (21%) of 841 with other non‐life‐threatening rhythms ( P <0.001). Multivariable analysis incorporating age, comorbidities, AF/AFL, QRS abnormalities, and ST‐T wave changes, and initial hs‐cTnT ≥20 ng/L showed that increased age (HR 1.04/year), elevated hs‐cTnT (HR 4.57), AF/AFL (HR 2.07), and a history of coronary artery disease (HR 1.56) and active cancer (HR 1.87) were associated with increased mortality. Conclusions: Myocardial injury with hs‐cTnT ≥20 ng/L, in addition to cardiac conduction perturbations, especially AF/AFL, upon hospital admission for COVID‐19 infection is associated with markedly increased risk for mortality than either diagnostic abnormality alone.
- Open Access
- 11 May 2020
- Published by Springer Nature in npj Digital Medicine
- Vol. 3 (1) , 1-10
Translating evidence into practice: eligibility criteria fail to eliminate clinically significant differences between real-world and study populations
Randomized controlled trials (RCTs) are regarded as the most reputable source of evidence. In some studies, factors beyond the intervention itself may contribute to the measured effect, an occurrence known as heterogeneity of treatment effect (HTE). If the RCT population differs from the real-world population on factors that induce HTE, the trials effect will not replicate. The RCTs eligibility criteria should identify the sub-population in which its evidence will replicate. However, the extent to which the eligibility criteria identify the appropriate population is unknown, which raises concerns for generalizability. We compared reported data from RCTs with real-world data from the electronic health records of a large, academic medical center that was curated according to RCT eligibility criteria. Our results show fundamental differences between the RCT population and our observational cohorts, which suggests that eligibility criteria may be insufficient for identifying the applicable real-world population in which RCT evidence will replicate.
- 11 May 2020
- Published by Springer Nature in npj Digital Medicine
- Vol. 3 (1) , 1-10
Open AccessTranslating evidence into practice: eligibility criteria fail to eliminate clinically significant differences between real-world and study populations
Randomized controlled trials (RCTs) are regarded as the most reputable source of evidence. In some studies, factors beyond the intervention itself may contribute to the measured effect, an occurrence known as heterogeneity of treatment effect (HTE). If the RCT population differs from the real-world population on factors that induce HTE, the trials effect will not replicate. The RCTs eligibility criteria should identify the sub-population in which its evidence will replicate. However, the extent to which the eligibility criteria identify the appropriate population is unknown, which raises concerns for generalizability. We compared reported data from RCTs with real-world data from the electronic health records of a large, academic medical center that was curated according to RCT eligibility criteria. Our results show fundamental differences between the RCT population and our observational cohorts, which suggests that eligibility criteria may be insufficient for identifying the applicable real-world population in which RCT evidence will replicate.