No significant increase in Guillain-Barré syndrome after COVID-19 vaccination in adults: A vaccine adverse event reporting system study

Abstract

Objective: To investigate the association between Guillain-Barré syndrome (GBS) and COVID-19 vaccination.

Background: On July 13, 2021, the US Food and Drug Administration (FDA) released a new warning that Johnson & Johnson COVID-19 vaccine could increase the risk of developing GBS.

Methods: The reporting rate of adult GBS after COVID-19 vaccination, ascertained with Brighton criteria, was compared with the reporting rate after other vaccinations during the same time period, and also compared with the reporting rate during control periods. Statistical methods such as proportion tests, and Pearson's chi-squared test were utilized to identify significant relationships. Self-controlled and case centered analyses were conducted. A machine learning model was utilized to identify the factors associated with a worse outcome defined as emergency room (ER) or doctor visits, hospitalizations, and deaths.

Results: The reporting rate of GBS after COVID-19 vaccination was significantly higher than after influenza and other vaccinations (49.7, 0.19, 0.16 per 10 million, p < 0.0001). However, the reporting rate was within the incidence range of GBS in the general population. Using self-controlled and case centered analyses, there was a significant difference in the reporting rate of GBS after COVID-19 vaccination between the risk period and control period (p < 0.0001). There was an estimated 0.7-1.7 per million excess reports of GBS within 6 weeks of COVID-19 vaccination. Machine learning model demonstrated that female gender and age between 18 and 44 are associated with worse outcome. No association was found between the onset interval of GBS and its prognosis.

Conclusions: Although the reporting rate of GBS after COVID-19 vaccination was not statistically different than that of the general population, the increased reporting of GBS within the first 6 weeks after COVID-19 vaccination, more so than with other vaccinations, suggests that some cases of GBS are temporally associated with COVID-19 vaccination. However, there is a reduction in the reporting rate of GBS after other vaccines, compared to reporting rates pre-COVID-19, highlighting limitations inherent in any passive surveillance system. These findings warrant continuous analysis of GBS after COVID-19 vaccination. Further improvement of the machine learning model is needed for clinical use.

Keywords: COVID-19 vaccination; Guillan Barre syndrome; Machine learning; SARS-CoV-2; Vaccine adverse events.

Fig. 1

Onset interval of GBS and non GBS events after COVID vaccination.

Fig. 2

Panel A: Using the self-controlled analysis, there was a significant difference in the reporting rate of GBS after COVID-19 vaccination between the risk period, which was 3 days–6 weeks and control period, which was 7 weeks to 10 weeks, p < 0.0001. The relative risk demonstrates that cases are 34.16 times more likely to occur 6 weeks after vaccination than the control period which was 11–16 weeks after vaccinations. Panel B: In case centered analysis the relative risk demonstrates vaccination was more likely to occur 9.14 times in the 3 days to 6-week period than the 7-to-12-week period.

Fig. 3

Using this matrix, we classify the results of the predictive model. Panel A represents the ER or Doctor visit model, with a computed accuracy of 76.74%. Panel B is hospitalizations, with a computed accuracy of 70.54%. Panel C is death, with an accuracy of 97.67%. As can be seen in panel A, the model was able to accurately predict which cases would not result in an ER or doctor visit (sensitivity). But lower accuracy in cases that resulted in a visit (specificity). The same result was true for death, panel C, as an endpoint. For hospitalization, the model predicted with a high specificity for cases that would result in hospitalization, but a lower sensitivity, in identifying cases that did not result in hospitalization.

Fig. 4

This graph demonstrates the importance of each variable of the case of GBS and the percentage of the variance of the data that can be explained by each one. A higher value means that the feature has a higher impact on the model predicting a death. This indicates the percentage of importance in the classification model. As shown in this graph, the most important feature of the case, was the patient’s age, which explains 35% of the prediction. Every feature used to train the random forest model is given a value on this scale. The most important variables were age, gender, and presence of an ER or doctor visit to predict death, with the cumulative sum of these explaining approximately 75% of the data.