Background & Aims

It is well known that the SARS-CoV-2 virus responsible for the COVID-19 pandemic can cause long term symptoms including pain. The reasons for why some develop long-COVID pain, whereas other remain symptom free, are currently not fully understood. Some of the predictors are known from smaller cohort studies but larger epidemiological studies utilizing advanced machine learning approaches have not been done. The aim of this explorative, large, national, cohort study was to use a machine learning AI model approach to identify predictors separating those suffering from long-COVID pain from those who experienced no symptoms. The hypothesis is that understanding such prognostic factors could be applied to other clinical conditions with the same underlying pathophysiology causing pain in some but not others.

Methods

Two questionnaires were distributed via the secured national digital mail system to 593,741 residents infected by the SARS-CoV-2 in the period March 2020 to December 2021. A total of 137,260 responded to the questionnaire resulting in approximately 13 million responses related to demographics, pre-existing medical comorbidities, previous pain-related symptoms, pain medication, pain intensity (4-point scale), and development of de novo widespread pain. A total of 19,201 reported long-COVID pain. An eXtreme Gradient Boosting (XGBoost) machine learning model approach was used to construct a classification model with the importance of predictive features identified via SHapley Additive exPlanations (SHAP) explanation values to evaluate the importance of each data variable for the ability to predict the outcome expressed via global parameter importance- and bee swarm plots.

Results

The SHAP values showed that the ratio between the ranked highest mean SHAP values (Age: +0.11, Female sex: +0.09, Height: +0.09, Hospital admission: +0.07, Pain change: +0.07, Weight: +0.06, and Physical activity: +0.04) was high with only a few other factors (Prior diagnosis with stress, anxiety, or nothing, and hospital admission because of COVID-19) contributing with minor effects and the sum of the remaining other 36 factors was equal to zero.

The bee swarm plot showed the change in predictive values when the individual question’s value changed. Predicted value changes revealed that age, height, hospital admission, female sex, prior diagnosis of stress, anxiety, and stay in intensive care at the hospital had the highest positive impact on the model output (development of pain). Younger age, lower weight, and higher height had a negative impact on the model output.

Conclusions

Using an AI machine learning prediction model approach the present large cohort, national survey identified a set of questionnaire responses which contributes to the prediction of long-COVID pain. A core set of identified predictors for development of long-COVID pain are shared with other areas of pain development (e.g., chronic pain after surgery), and hence substantiate the insight into important prognostics to understand why the same underlying disease or intervention (e.g., surgery) may cause pain in some but not in others. This is likewise important when counselling patients for their vulnerability to develop a given pain problem.

References

Ebbesen BD, Giordano R, Valera-Calero JA, Hedegaard JN, Fernández-de-Las-Peñas C, Arendt-Nielsen L. Prevalence and Risk Factors of De Novo Widespread Post-COVID Pain in Non-hospitalized COVID-19 Survivors: A Nationwide Exploratory Population-Based Survey. J Pain. 2024 Jan;25(1):1-11
Fernández-de-Las-Peñas C, Nijs J, Neblett R, Polli A, Moens M, Goudman L, Shekhar Patil M, Knaggs RD, Pickering G, Arendt-Nielsen L Phenotyping Post-COVID Pain as a Nociceptive, Neuropathic, or Nociplastic Pain Condition. Biomedicines. 2022 Oct 13;10(10):2562.
Fernández-de-Las-Peñas C, Raveendran AV, Giordano R, Arendt-Nielsen L. Long COVID or Post-COVID-19 Condition: Past, Present and Future Research Directions. Microorganisms. 2023 Dec 11;11(12):2959.

Presenting Author

Lars Arendt-Nielsen

Poster Authors

Brian Ebbesen

MSc

Aalborg University

Lead Author

E-mail

Lars Arendt-Nielsen

PhD

Aalborg University

Lead Author

E-mail

Rasmus A Nielsen

MSc

Brandheroes, Aarhus, Denmark

Lead Author

E-mail

Jakob Nebeling Hedegaard

MSc

Aalborg University, Aalborg, Denmark

Lead Author

E-mail

Rocco Giordano

Center for Neuroplasticity and Pain, HST, Faculty of Medicine, Aalborg University, Aalborg, DK

Lead Author

E-mail

César Fernández-de-las-Peñas

dr med sci

Universidad Rey Juan Carlos (URJC), Madrid, Spain

Lead Author

E-mail

Topics

  • Assessment and Diagnosis