The best COVID-19 predictor is recent smell loss: a cross-sectional study

Richard C. Gerkin, Kathrin Ohla, Maria G. Veldhuizen, Paule V. Joseph, Christine E Kelly, Alyssa J. Bakke, Kimberley E. Steele, Robert Pellegrino, Marta Yanina Pepino, Cédric Bouysset, Graciela M Soler, Veronica Pereda‐Loth, Michele Dibattista, Keiland W Cooper, Ilja Croijmans, Antonella Di Pizio, Mehmet Hakan Özdener, Anna D’Errico, Florian Ph. S. Fischmeister, María Adelaida Bock, Paloma Paloma Domínguez, Hüseyin Yanık, Sanne Boesveldt, Jasper H. B. de Groot, Caterina Dinnella, Jessica Freiherr, Tatiana K. Laktionova, Sajidxa Mariño, Erminio Monteleone, Alexia Nunez-Parra, Olagunju Abdulrahman, Marina Ritchie, Thierry Thomas‐Danguin, Julie Walsh‐Messinger, Rashid Al Abri, Rafieh Alizadeh, Emmanuelle Bignon, Elena Cantone, Maria Paola Cecchini, Jingguo Chen, María Dolors Guárdia, Kara C. Hoover, Noam Karni, Marta Navarro, Alissa A Nolden, Patricia Portillo Mazal, Nicholas R. Rowan, Atiye Sarabi‐Jamab, Nicholas Archer, Ben Chen, Elizabeth Di Valerio, Emma L. Feeney, Johannes Frasnelli, Mackenzie E. Hannum, Claire Hopkins, Hadar Klein, Cyril Mignot, Carla Mucignat‐Caretta, Yuping Ning, Elif Öztürk, Mei Peng, Özlem Saatçi, Elizabeth Sell, Carol H. Yan, Raul Alfaro, Cinzia Cecchetto, Gérard Coureaud, Riley Herriman, Jeb M. Justice, Pavan Kumar Kaushik, Sachiko Koyama, Jonathan B. Overdevest, Nicola PirastumedRxiv 2020 · DOI

Abstract

As smell loss is the best predictor of COVID-19, we developed the ODoR-19 tool, a 0-10 scale to screen for recent olfactory loss. Numeric ratings ≤2 indicate high odds of symptomatic COVID-19 (10<OR<4), especially when viral lab tests are impractical or unavailable.

Introduction

While fever and cough are traditionally considered the "cardinal" symptoms of COVID-19, this study demonstrates that recent smell loss is actually the single best predictor of the disease. Early in the pandemic, screening relied heavily on symptoms like fever and cough, which lack diagnostic specificity. This research sought to quantify the predictive power of chemosensory loss using continuous scales rather than binary observations, moving beyond simple "yes/no" questions to determine if the degree of loss could provide a more reliable diagnostic signal, especially in resource-limited settings.

Methods and Datasets

The researchers conducted a preregistered, cross-sectional study using a crowdsourced questionnaire deployed by the Global Consortium for Chemosensory Research (GCCR) in 23 languages. The primary dataset consisted of 15,747 participants, categorized by their COVID-19 status (lab-confirmed or clinical diagnosis).

Participant flow diagram showing the selection and categorization of the study sample.

The study utilized Visual Analog Scales (VAS) from 0–100 to measure smell, taste, and chemesthesis before and during illness. To identify the most parsimonious predictors, L1-regularized logistic regression and information entropy analysis were employed. The team also addressed potential collider bias (the risk that the sample was over-represented by people interested in smell loss) by benchmarking against a representative dataset from the YouGov COVID-19 Public Monitor (n=8,674).

Key Insights

  • Scales Beat Binary: Continuous ratings (0–10 or 0–100) are significantly more informative than binary "Yes/No" questions. The resolution of a scale captures diagnostic nuances that a simple checkbox misses.
  • Smell vs. Stuffy Nose: COVID-19-related smell loss is "orthogonal" to nasal obstruction. C19+ patients lose their sense of smell even without a "blocked nose," distinguishing it from the common cold or flu.
  • Parsimony: A model using only "Smell During Illness" and "Days Since Onset" performed as well (AUC = 0.72) as a complex model using 70 features, enabling the creation of a simple, accurate tool.

Results

Smell loss during illness was the best predictor of COVID-19 (ROC AUC = 0.71), significantly outperforming the combination of fever, cough, and shortness of breath (AUC = 0.55). Based on these findings, the authors proposed the ODoR-19 tool, a 0–10 scale for current smelling ability:

  • Scores 0–2: High odds of COVID-19 (Odds Ratio between 4 and 10).
  • Score 3: Borderline risk.
  • Scores >3: Lower probability, though COVID-19 cannot be ruled out.

Additionally, approximately 50% of C19+ participants reported smell recovery within 40 days. Recovery was best predicted by the time elapsed since the onset of illness, rather than the severity of the initial loss.

Limitations and Future Work

The study is limited by self-selection bias and the use of self-reported data rather than objective clinical tests like "Sniffin' Sticks." Future work involves validating the ODoR-19 tool in clinical settings and tracking long-term "Persistent Smell Loss" clusters to understand the pathophysiology of "Long COVID" as it relates to the olfactory system.

Summary updated

Abstract

As smell loss is the best predictor of COVID-19, we developed the ODoR-19 tool, a 0-10 scale to screen for recent olfactory loss. Numeric ratings ≤2 indicate high odds of symptomatic COVID-19 (10<OR<4), especially when viral lab tests are impractical or unavailable.
BibTeX
@misc{gerkin2020best,
  author = {Richard C. Gerkin and Kathrin Ohla and Maria G. Veldhuizen and Paule V. Joseph and Christine E Kelly and Alyssa J. Bakke and Kimberley E. Steele and Robert Pellegrino and Marta Yanina Pepino and Cédric Bouysset and Graciela M Soler and Veronica Pereda‐Loth and Michele Dibattista and Keiland W Cooper and Ilja Croijmans and Antonella Di Pizio and Mehmet Hakan Özdener and Anna D’Errico and Florian Ph. S. Fischmeister and María Adelaida Bock and Paloma Paloma Domínguez and Hüseyin Yanık and Sanne Boesveldt and Jasper H. B. de Groot and Caterina Dinnella and Jessica Freiherr and Tatiana K. Laktionova and Sajidxa Mariño and Erminio Monteleone and Alexia Nunez-Parra and Olagunju Abdulrahman and Marina Ritchie and Thierry Thomas‐Danguin and Julie Walsh‐Messinger and Rashid Al Abri and Rafieh Alizadeh and Emmanuelle Bignon and Elena Cantone and Maria Paola Cecchini and Jingguo Chen and María Dolors Guárdia and Kara C. Hoover and Noam Karni and Marta Navarro and Alissa A Nolden and Patricia Portillo Mazal and Nicholas R. Rowan and Atiye Sarabi‐Jamab and Nicholas Archer and Ben Chen and Elizabeth Di Valerio and Emma L. Feeney and Johannes Frasnelli and Mackenzie E. Hannum and Claire Hopkins and Hadar Klein and Cyril Mignot and Carla Mucignat‐Caretta and Yuping Ning and Elif Öztürk and Mei Peng and Özlem Saatçi and Elizabeth Sell and Carol H. Yan and Raul Alfaro and Cinzia Cecchetto and Gérard Coureaud and Riley Herriman and Jeb M. Justice and Pavan Kumar Kaushik and Sachiko Koyama and Jonathan B. Overdevest and Nicola Pirastu},
  title = {{The best COVID-19 predictor is recent smell loss: a cross-sectional study}},
  year = {2020},
  doi = {10.1101/2020.07.22.20157263},
  howpublished = {medRxiv},
}
Figure