Person:

Reis, Ben

Background: Physician notes routinely recorded during patient care represent a vast and underutilized resource for human disease studies on a population scale. Their use in research is primarily limited by the need to separate confidential patient information from clinical annotations, a process that is resource-intensive when performed manually. This study seeks to create an automated method for de-identifying physician notes that does not require large amounts of private information: in addition to training a model to recognize Protected Health Information (PHI) within private physician notes, we reverse the problem and train a model to recognize non-PHI words and phrases that appear in public medical texts. Methods: Public and private medical text sources were analyzed to distinguish common medical words and phrases from Protected Health Information. Patient identifiers are generally nouns and numbers that appear infrequently in medical literature. To quantify this relationship, term frequencies and part of speech tags were compared between journal publications and physician notes. Standard medical concepts and phrases were then examined across ten medical dictionaries. Lists and rules were included from the US census database and previously published studies. In total, 28 features were used to train decision tree classifiers. Results: The model successfully recalled 98% of PHI tokens from 220 discharge summaries. Cost sensitive classification was used to weight recall over precision (98% F10 score, 76% F1 score). More than half of the false negatives were the word “of” appearing in a hospital name. All patient names, phone numbers, and home addresses were at least partially redacted. Medical concepts such as “elevated white blood cell count” were informative for de-identification. The results exceed the previously approved criteria established by four Institutional Review Boards. Conclusions: The results indicate that distributional differences between private and public medical text can be used to accurately classify PHI. The data and algorithms reported here are made freely available for evaluation and improvement.

Objective: To determine whether longitudinal data in patients’ historical records, commonly available in electronic health record systems, can be used to predict a patient’s future risk of receiving a diagnosis of domestic abuse. Design: Bayesian models, known as intelligent histories, used to predict a patient’s risk of receiving a future diagnosis of abuse, based on the patient’s diagnostic history. Retrospective evaluation of the model’s predictions using an independent testing set. Setting: A state-wide claims database covering six years of inpatient admissions to hospital, admissions for observation, and encounters in emergency departments. Population: All patients aged over 18 who had at least four years between their earliest and latest visits recorded in the database (561 216 patients). Main outcome measures: Timeliness of detection, sensitivity, specificity, positive predictive values, and area under the ROC curve. Results: 1.04% (5829) of the patients met the narrow case definition for abuse, while 3.44% (19 303) met the broader case definition for abuse. The model achieved sensitive, specific (area under the ROC curve of 0.88), and early (10-30 months in advance, on average) prediction of patients’ future risk of receiving a diagnosis of abuse. Analysis of model parameters showed important differences between sexes in the risks associated with certain diagnoses. Conclusions: Commonly available longitudinal diagnostic data can be useful for predicting a patient’s future risk of receiving a diagnosis of abuse. This modelling approach could serve as the basis for an early warning system to help doctors identify high risk patients for further screening.

To evaluate the effectiveness of the BNT162b2 mRNA vaccine among pregnant women, we conducted an observational cohort study of pregnant women 16 years or older, with no history of SARS-CoV-2, who were vaccinated between December 20, 2020 and June 3, 2021. 10,861 vaccinated pregnant women were matched to 10,861 unvaccinated control women using demographic and clinical characteristics. Study outcomes included documented infection with SARS-CoV-2, symptomatic COVID-19, COVID-19-related hospitalization, severe illness and death. Estimated vaccine effectiveness from 7 through 28 days after the second dose was 97% (95% CI 91%-100%) for any documented infection, 96% (86-100%) for infections with documented symptoms, and 85% (32%-100%) for COVID-19-related hospitalization. Only one event of severe illness was observed in the unvaccinated group, and no deaths were observed in either group. In summary, the BNT162b2 mRNA vaccine was estimated to have high vaccine effectiveness among pregnant women, similar to the effectiveness estimated in the general population.