Natural Language Processing to Identify Advance Care Planning Documentation in a Multisite Pragmatic Clinical Trial

CONTEXT

Large multisite clinical trials studying decision-making when facing serious illness require an efficient method for abstraction of advance care planning (ACP) documentation from clinical text documents. However, the current gold standard method of manual chart review is time-consuming and unreliable.

OBJECTIVES

To evaluate the ability to use natural language processing (NLP) to identify ACP documention in clinical notes from patients participating in a multisite trial.

METHODS

Patients with advanced cancer followed in three disease-focused oncology clinics at Duke Health, Mayo Clinic, and Northwell Health were identified using administrative data. All outpatient and inpatient notes from patients meeting inclusion criteria were extracted from electronic health records (EHRs) between March 2018 and March 2019. NLP text identification software with semi-automated chart review was applied to identify documentation of four ACP domains: (1) conversations about goals of care, (2) limitation of life-sustaining treatment, (3) involvement of palliative care, and (4) discussion of hospice. The performance of NLP was compared to gold standard manual chart review.

RESULTS

435 unique patients with 79,797 notes were included in the study. In our validation data set, NLP achieved F1 scores ranging from 0.84 to 0.97 across domains compared to gold standard manual chart review. NLP identified ACP documentation in a fraction of the time required by manual chart review of EHRs (1-5 minutes per patient for NLP, vs. 30-120 minutes for manual abstraction).

CONCLUSION

NLP is more efficient and as accurate as manual chart review for identifying ACP documentation in studies with large patient cohorts.

Deep learning to predict long-term mortality in patients requiring 7 days of mechanical ventilation

Background

Among patients with acute respiratory failure requiring prolonged mechanical ventilation, tracheostomies are typically placed after approximately 7 to 10 days. Yet half of patients admitted to the intensive care unit receiving tracheostomy will die within a year, often within three months. Existing mortality prediction models for prolonged mechanical ventilation, such as the ProVent Score, have poor sensitivity and are not applied until after 14 days of mechanical ventilation. We developed a model to predict 3-month mortality in patients requiring more than 7 days of mechanical ventilation using deep learning techniques and compared this to existing mortality models.

Methods

Retrospective cohort study. Setting: The Medical Information Mart for Intensive Care III Database. Patients: All adults requiring ≥ 7 days of mechanical ventilation. Measurements: A neural network model for 3-month mortality was created using process-of-care variables, including demographic, physiologic and clinical data. The area under the receiver operator curve (AUROC) was compared to the ProVent model at predicting 3 and 12-month mortality. Shapley values were used to identify the variables with the greatest contributions to the model.

Results

There were 4,334 encounters divided into a development cohort (n = 3467) and a testing cohort (n = 867). The final deep learning model included 250 variables and had an AUROC of 0.74 for predicting 3-month mortality at day 7 of mechanical ventilation versus 0.59 for the ProVent model. Older age and elevated Simplified Acute Physiology Score II (SAPS II) Score on intensive care unit admission had the largest contribution to predicting mortality.

Discussion

We developed a deep learning prediction model for 3-month mortality among patients requiring ≥ 7 days of mechanical ventilation using a neural network approach utilizing readily available clinical variables. The model outperforms the ProVent model for predicting mortality among patients requiring ≥ 7 days of mechanical ventilation. This model requires external validation.

Long-Term Prognosis of Older Adults Who Survive Emergency Mechanical Ventilation

CONTEXT

Emergent mechanical ventilation represents an important inflection point in seriously ill older adults' illness trajectories. Data are lacking on the long-term prognosis after surviving mechanical ventilation to inform shared decision making in serious illness conversations.

OBJECTIVES

Describe the long-term prognosis of older adults who survive emergency mechanical ventilation to inform shared decision making.

METHODS

This is a retrospective cohort study from a single-center intensive care unit in an academic, urban, and tertiary care medical center. We included adults aged 75 years and older consecutively admitted with mechanical ventilation between 2008 and 2012 in the Multiparameter Intelligent Monitoring of Intensive Care III database. We excluded patients who were electively admitted. Our primary outcome was the long-term prognosis after leaving the hospital stratified by discharge location. Our secondary outcome was the frequency of documented serious illness conversations within 48 hours of hospitalization recommended by the National Quality Forum.

RESULTS

We identified 415 patients (454 hospital admissions) consecutively admitted to the intensive care unit. The median age was 82.6 years, 54.0% were female, 78.2% were white, non-Hispanic, and in-hospital mortality rate was 36.6%. Among the survivors, the median survival after hospital discharge was 163.5 days (interquartile range 37.5-476.8). Only 49.1% of patients had documented serious illness conversations within 48 hours of hospitalization. About 63.4% of patients (59 of 93) who were discharged to long-term acute care hospitals died by six months.

CONCLUSION

This study demonstrated the long-term prognosis of older adults who underwent emergent mechanical ventilation. These data could be used to inform shared decision making in serious illness conversations.

Safety and immunogenicity of a Zika purified inactivated virus vaccine given via standard, accelerated, or shortened schedules: a single-centre, double-blind, sequential-group, randomised, placebo-controlled, phase 1 trial

BACKGROUND

The development of an effective vaccine against Zika virus remains a public health priority. A Zika purified inactivated virus (ZPIV) vaccine candidate has been shown to protect animals against Zika virus challenge and to be well tolerated and immunogenic in humans up to 8 weeks of follow-up. We aimed to assess the safety and immunogenicity of ZPIV in humans up to 52 weeks of follow-up when given via standard or accelerated vaccination schedules.

METHODS

We did a single-centre, double-blind, randomised controlled, phase 1 trial in healthy adults aged 18-50 years with no known history of flavivirus vaccination or infection at Beth Israel Deaconess Medical Center in Boston, MA, USA. Participants were sequentially enrolled into one of three groups: ZPIV given at weeks 0 and 4 (standard regimen), weeks 0 and 2 (accelerated regimen), or week 0 alone (single-dose regimen). Within each group, participants were randomly assigned using a computer-generated randomisation schedule to receive an intramuscular injection of 5 μg ZPIV or saline placebo, in a ratio of 5:1. The sponsor, clinical staff, investigators, participants, and laboratory personnel were masked to treatment assignment. The primary endpoint was safety up to day 364 after final dose administration, and secondary endpoints were proportion of participants with positive humoral immune responses (50% microneutralisation titre [MN50] ≥100) and geometric mean MN50 at observed peak response (ie, the highest neutralising antibody level observed for an individual participant across all timepoints) and week 28. All participants who received at least one dose of ZPIV or placebo were included in the safety population; the analysis of immunogenicity at observed peak included all participants who received at least one dose of ZPIV or placebo and had any adverse events or immunogenicity data after dosing. The week 28 immunogenicity analysis population consisted of all participants who received ZPIV or placebo and had immunogenicity data available at week 28. This trial is registered with ClinicalTrials.gov, NCT02937233.

FINDINGS

Between Dec 8, 2016, and May 17, 2017, 12 participants were enrolled into each group and then randomly assigned to vaccine (n=10) or placebo (n=2). There were no serious or grade 3 treatment-related adverse events. The most common reactions among the 30 participants who received the vaccine were injection-site pain (24 [80%]), fatigue (16 [53%]), and headache (14 [46%]). A positive response at observed peak titre was detected in all participants who received ZPIV via the standard regimen, in eight (80%) of ten participants who received ZPIV via the accelerated regimen, and in none of the ten participants who received ZPIV via the single-dose regimen. The geometric mean of all individual participants' observed peak values was 1153·9 (95% CI 455·2-2925·2) in the standard regimen group, 517·7 (142·9-1875·6) in the accelerated regimen group, and 6·3 (3·7-10·8) in the single-dose regimen group. At week 28, a positive response was observed in one (13%) of eight participants who received ZPIV via the standard regimen and in no participant who received ZPIV via the accelerated (n=7) or single-dose (n=10) regimens. The geomteric mean titre (GMT) at this timepoint was 13·9 (95% CI 3·5-55·1) in the standard regimen group and 6·9 (4·0-11·9) in the accelerated regimen group; antibody titres were undetectable at 28 weeks in participants who received ZPIV via the single-dose regimen. For all vaccine schedules, GMTs peaked 2 weeks after the final vaccination and declined to less than 100 by study week 16. There was no difference in observed peak GMTs between the standard 4-week and the accelerated 2-week boosting regimens (p=0·4494).

INTERPRETATION

ZPIV was safe and well tolerated in humans up to 52 weeks of follow-up. ZPIV immunogenicity required two doses and was not durable. Additional studies of ZPIV to optimise dosing schedules are ongoing.

FUNDING

The Henry M Jackson Foundation for the Advancement of Military Medicine.

Sustained maternal antibody and cellular immune responses in pregnant women infected with Zika virus and mother to infant transfer of Zika-specific antibodies

PROBLEM

Evaluation of Zika virus (ZIKV)-specific humoral and cellular immune response in pregnant women exposed to ZIKV.

METHOD OF STUDY

In this observational, prospective cohort study, we recruited pregnant women presenting for prenatal ultrasound for ZIKV exposure at a single academic teaching hospital in Boston, MA from November 2016 to December 2018. We collected blood, urine, and cervicovaginal swabs antepartum, intrapartum, and postpartum; and cord blood and placenta at delivery. We used experimental assays to calculate quantitative viral loads, ZIKV-specific immunoglobulin titers, and ZIKV-specific T-cell responses.

RESULTS

We enrolled 22 participants, three of which had serologic-confirmed ZIKV infection. No participants demonstrated sustained ZIKV shedding. ZIKV-specific IgG/IgM antibody was sustained throughout pregnancy and postpartum. ZIKV envelope and capsid-specific T-cell responses were also observed, albeit inconsistent. No newborns in this cohort had congenital Zika syndrome. Infant cord blood of infected mothers exhibited ZIKV-specific IgG, but not IgM antibodies.

CONCLUSION

We detected a robust, prolonged maternal humoral immune response to ZIKV during pregnancy and postpartum. We also demonstrated evidence for efficient transplacental antibody transfer from mother to infant at birth, supporting the importance of neonatal passive immunity to ZIKV. Maternal T-cell responses were less consistent among pregnant women infected with ZIKV.

Deep learning algorithms to identify documentation of serious illness conversations during intensive care unit admissions

Background:

Timely documentation of care preferences is an endorsed quality indicator for seriously ill patients admitted to intensive care units. Clinicians document their conversations about these preferences as unstructured free text in clinical notes from electronic health records.

Aim:

To apply deep learning algorithms for automated identification of serious illness conversations documented in physician notes during intensive care unit admissions.

Design:

Using a retrospective dataset of physician notes, clinicians annotated all text documenting patient care preferences (goals of care or code status limitations), communication with family, and full code status. Clinician-coded text was used to train algorithms to identify documentation and to validate algorithms. The validated algorithms were deployed to assess the percentage of intensive care unit admissions of patients aged ⩾75 that had care preferences documented within the first 48 h.

Setting/participants:

Patients admitted to one of five intensive care units.

Results:

Algorithm performance was calculated by comparing machine-identified documentation to clinician-coded documentation. For detecting care preference documentation at the note level, the algorithm had F1-score of 0.92 (95% confidence interval, 0.89 to 0.95), sensitivity of 93.5% (95% confidence interval, 90.0% to 98.0%), and specificity of 91.0% (95% confidence interval, 86.4% to 95.3%). Applied to 1350 admissions of patients aged ⩾75, we found that 64.7% of patient intensive care unit admissions had care preferences documented within the first 48 h.

Conclusion:

Deep learning algorithms identified patient care preference documentation with sensitivity and specificity approaching that of clinicians and computed in a tiny fraction of time. Future research should determine the generalizability of these methods in multiple healthcare systems.

From Pharmacovigilance to Clinical Care Optimization

In order to ensure the continued, safe administration of pharmaceuticals, particularly those agents that have been recently introduced into the market, there is a need for improved surveillance after product release. This is particularly so because drugs are used by a variety of patients whose particular characteristics may not have been fully captured in the original market approval studies. Even well-conducted, randomized controlled trials are likely to have excluded a large proportion of individuals because of any number of issues. The digitization of medical care, which yields rich and accessible drug data amenable to analytic techniques, provides an opportunity to capture the required information via observational studies. We propose the development of an open, accessible database containing properly de-identified data, to provide the substrate for the required improvement in pharmacovigilance. A range of stakeholders could use this to identify delayed and low-frequency adverse events. Moreover, its power as a research tool could extend to the detection of complex interactions, potential novel uses, and subtle subpopulation effects. This far-reaching potential is demonstrated by our experience with the open Multi-parameter Intelligent Monitoring in Intensive Care (MIMIC) intensive care unit database. The new database could also inform the development of objective, robust clinical practice guidelines. Careful systematization and deliberate standardization of a fully digitized pharmacovigilance process is likely to save both time and resources for healthcare in general.

DC-SIGN association with the Th2 environment of lepromatous lesions: cause or effect?

The clinical spectrum of leprosy is related to patients' immune responses. Non-responsiveness towards Mycobacterium leprae (ML) seems to correlate with a Th2 cytokine profile. The reason for such a polarized immune response remains unclear. The C-type lectin, DC-SIGN, expressed by subsets of dendritic cells (DCs) and macrophages, has previously been associated with Th2 responses. Here we show abundant DC-SIGN expression in lepromatous but not borderline tuberculoid leprosy, in both HIV-positive and HIV-negative patients. Moreover, we demonstrate that DC-SIGN can act as an entry receptor for ML, as it does for M. tuberculosis, through the cell wall component lipoarabinomannan. DC-SIGN is expressed on virtually all ML-containing cells, providing further evidence for its role as a receptor. DC-SIGN may therefore be induced on macrophages in lepromatous leprosy and may then contribute to mycobacterial entry into these cells.

[Principles of NASA longitudinal medical studies]

This paper describes approaches to longitudinal studies of the changes in the health status of the US astronauts. The methods include acquisition and analysis of biomedical data accumulated in one and repeated space missions, detection of potential occupational diseases inflight and evaluation of mortality cases associated with them. It is suggested to use pilots and flight controllers as controls. It is indicated that annual physical examinations can be an important source of relevant scientific information.

Concepts for NASA longitudinal health studies

Clinical data collected from a 15-year study of the homogeneous group of pre-Shuttle astronauts have revealed no significant long-term effects from spaceflight. The current hypothesis suggests that repeated exposures to the space environment in the Shuttle era will similarly have no long-term health effects. However, a much more heterogeneous group of astronauts and non-astronaut scientists will fly in Shuttle, and data on this group's adaptation to the space environment and readaptation to Earth are currently sparse. In addition, very little information is available concerning the short- and long-term medical consequences of long duration exposure to space and subsequent readaptation to the Earth environment. In this paper, retrospective clinical information on astronauts is reviewed and concepts for conducting epidemiological studies examining long-term health effects of spaceflight on humans, including associated occupational risks factors, are presented.