Dataset Information

Text-based predictions of COVID-19 diagnosis from self-reported chemosensory descriptions.

ABSTRACT:

Background

There is a prevailing view that humans' capacity to use language to characterize sensations like odors or tastes is poor, providing an unreliable source of information.

Methods

Here, we developed a machine learning method based on Natural Language Processing (NLP) using Large Language Models (LLM) to predict COVID-19 diagnosis solely based on text descriptions of acute changes in chemosensation, i.e., smell, taste and chemesthesis, caused by the disease. The dataset of more than 1500 subjects was obtained from survey responses early in the COVID-19 pandemic, in Spring 2020.

Results

When predicting COVID-19 diagnosis, our NLP model performs comparably (AUC ROC ~ 0.65) to models based on self-reported changes in function collected via quantitative rating scales. Further, our NLP model could attribute importance of words when performing the prediction; sentiment and descriptive words such as "smell", "taste", "sense", had strong contributions to the predictions. In addition, adjectives describing specific tastes or smells such as "salty", "sweet", "spicy", and "sour" also contributed considerably to predictions.

Conclusions

Our results show that the description of perceptual symptoms caused by a viral infection can be used to fine-tune an LLM model to correctly predict and interpret the diagnostic status of a subject. In the future, similar models may have utility for patient verbatims from online health portals or electronic health records.

SUBMITTER: Li H

PROVIDER: S-EPMC10374642 | biostudies-literature | 2023 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Text-based predictions of COVID-19 diagnosis from self-reported chemosensory descriptions.

Li Hongyang H Gerkin Richard C RC Bakke Alyssa A Norel Raquel R Cecchi Guillermo G Laudamiel Christophe C Niv Masha Y MY Ohla Kathrin K Hayes John E JE Parma Valentina V Meyer Pablo P

Communications medicine 20230727 1

<h4>Background</h4>There is a prevailing view that humans' capacity to use language to characterize sensations like odors or tastes is poor, providing an unreliable source of information.<h4>Methods</h4>Here, we developed a machine learning method based on Natural Language Processing (NLP) using Large Language Models (LLM) to predict COVID-19 diagnosis solely based on text descriptions of acute changes in chemosensation, i.e., smell, taste and chemesthesis, caused by the disease. The dataset of ...[more]

PMID: 37500763

Similar Datasets

Project description:ImportanceChanges in everyday functioning are crucial to assessing the long-term impact of COVID-19 infection.ObjectiveTo examine the impact of COVID-19 infection on everyday functioning 18 months after infection among veterans with and without histories of COVID-19 infection.Design, setting, and participantsThis cohort study used data from the US Veterans Affairs (VA) and included 186 veterans who had COVID-19 between October 2020 and April 2021 (ie, COVID-19 cohort) and 186 matched comparators who did not have documented COVID-19 infections (ie, control cohort). This match balanced the risk of COVID-19 based on 39 variables measured in the 24 months before infection or match, using principles of target trial emulation. Data were analyzed from December 2022 to December 2023.ExposureFirst documented COVID-19.Main outcome and measuresThe differences in self-reported everyday functioning 18 months after COVID-19 infection were estimated and compared with their matched comparators. Within-matched pair logistic and linear regressions assessed differences in outcomes and were weighted to account for sampling and nonresponse.ResultsAmong the 186 matched pairs of participants, their weighted mean age was 60.4 (95% CI, 57.5 to 63.2) years among veterans in the COVID-19 cohort (weighted sample, 91 459 of 101 133 [90.4%] male; 30 611 [30.3%] Black or African American veterans; 65 196 [64.4%] White veterans) and 61.1 (95% CI, 57.8 to 64.4) years among their comparators in the control cohort (91 459 [90.4%] male; 24 576 [24.3%] Black or African American veterans; 70 157 [69.4%] White veterans). A high proportion of veterans in the COVID-19 cohort (weighted percentage, 44.9% [95% CI, 34.2% to 56.2%]) reported that they could do less than what they felt they could do at the beginning of 2020 compared with the control cohort (weighted percentage, 35.3%; [95% CI, 25.6% to 46.4%]; within-matched pair adjusted odds ratio [OR], 1.52 [95% CI, 0.79 to 2.91]). There was no association of documented COVID-19 infection with fatigue, substantial pain, limitations in either activities of daily living and instrumental activities of daily living, severely curtailed life-space mobility, employment, or mean health-related quality of life on a utility scale.Conclusions and relevanceIn this cohort study of veterans with and without documented COVID-19, many reported a substantial loss of everyday functioning during the pandemic regardless of whether or not they had a documented infection with COVID-19. Future work with larger samples is needed to validate the estimated associations.

Project description:The pandemic of coronavirus disease 2019 (COVID-19) has affected the Italian community. The widespread use of quarantine had the desired impact of controlling the epidemic, although it caused many psychological consequences. To date, compliance of the Italian public with voluntary home quarantine has been very high, but little is known about the impact of psychological health on sociodemographic categories during the quarantine. The purpose of this study was to assess the prevalence of depressive symptoms in specific sociodemographic categories during the COVID-19 quarantine lockdown and the potential factors that contribute to, or mitigate, these effects. In the very early stage of the nationwide lockdown, 3,672 quarantined Italian adult residents (65% females, ranging from 18 to 85 years) participated in a web-based cross-sectional survey, including measures of depressive symptoms, which were measured by the Teate depression inventory, and state anxiety levels. The overall prevalence was 27.8% for moderate and 9.3% for severe levels of depressive symptoms. A generalized logistic model was used to identify the factors associated with mental health problems. Among these factors, sociodemographic variables (e.g., sex, age, employment status) and adherence to quarantine guidelines were analyzed. Females, younger people, students, singles, residents in northern Italy, people who were reluctant to adhere to quarantine guidelines, and people less worried about being infected with COVID-19 were at high risk of developing depressive symptoms during the COVID-19 epidemic, also after controlling for state anxiety. These findings showed that public levels of depressive symptoms did not increase the greater likelihood of being infected. Our study suggested that the monitoring of psychological outcomes for outbreaks could identify groups at higher risk of psychological morbidities due to the current pandemic in order to target future psychological interventions for implementation.

Dataset Information

Text-based predictions of COVID-19 diagnosis from self-reported chemosensory descriptions.

Background

Methods

Results

Conclusions

Publications

Text-based predictions of COVID-19 diagnosis from self-reported chemosensory descriptions.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets