Vaccine effectiveness against emerging COVID-19 variants using digital health data | Communications Medicine – Nature.com

To estimate VE, we adapted case-control methods1 for prevalent COVID-like illness (CLI) as a proxy for confirmed COVID-19 cases. Therefore, our estimates of VE measure a vaccines ability to prevent suspected symptomatic infections defined by CLI. To allow for changes in variant-specific symptomatology, we iterate across all possible CLI defined by 66 pair-wise combinations of 12 self-reported symptoms (fever, cough, difficulty breathing, fatigue, stuffy or runny nose, aches or muscle pain, sore throat, chest pain, nausea, loss of smell or taste, headache, chills). We then cluster the vaccine effectiveness estimates according to a single symptom of interest and evaluate the median vaccine effectiveness across all CLI definitions in the cluster. As an example, using a COVID-19-specific symptom (loss of smell or taste) as an anchor symptom, we evaluate VE estimates for all CLI definitions inclusive of this symptom during Delta and Omicron waves of infections, resulting in VE estimates for 11 pairwise combinations of symptoms. Consistent with previous estimates of VE that used PCR test data as the outcome2, our analyses reveal a median VEDelta of 0.77, IQR[0.76, 0.80] (Fig.1a, triangle). In comparison, analyzing the data from the Omicron period reveals a median VEOmicron of 0.47, IQR[0.41, 0.53] (Fig.1a, circle). Further expanding the approach to all CLI definitions reveals a median VEDelta of 0.71, IQR[0.65, 0.75] (Fig.1b). In contrast, the VEOmicron estimate is even lower (median 0.29, IQR[0.20, 0.38]). Notably, our findings align with those from a recent meta-analysis study focused on real-world vaccine effectiveness for fully vaccinated individuals. This study reported a VE of 70.9% (95% CI, 68.972.7) against Delta infections and a VE of 23.5% (95% CI, 17.029.5) against Omicron variant infections12. To understand how VE estimates for each CLI definition vary by wave, we take the difference between the two VE period estimates (VEOmicronVEDelta) for each CLI definition. Doing so reveals a median within-CLI definition change of 0.40, IQR[0.45, 0.35] (Fig.2a), suggesting lower VEOmicron regardless of the CLI definition that is used. Additionally, we find that the pattern of change in VE across CLI definitions is similar when evaluating individual country estimates (see Supplementary Fig.1).

a VE estimates for symptoms paired with the loss of smell or taste for the Delta (triangle) and Omicron (circle) periods. 95% confidence intervals are calculated for each VE estimate, with Delta and Omicron period estimates derived from 64,283 and 79,697 survey responses, respectively. b Box and whisker plot of VE estimates across all 66 possible CLI defined by pairwise combinations of symptoms for Delta and Omicron periods. The box represents the interquartile range (IQR) of estimates, with the horizontal line inside the box indicating the median. The whiskers extend to the largest/smallest values up to 1.5 times the IQR. Outlier values are represented as points. The sample size for each VE estimate is consistent with the sample sizes described in panel (a).

a Distribution of within-CLI change (VEOmicronVEDelta) across all CLI definitions. b Distributions of VEOmicronVEDelta among CLI definitions within each anchor symptom. Each box-plot contains estimates for an anchor symptom paired with the 11 other symptoms. Box-plots are ordered according to the magnitude of the median change, with the median across all VE indicated by the gray dashed line. Each box represents the interquartile range (IQR) of estimates, with the horizontal line inside the box indicating the median. The whiskers extend to the largest/smallest values up to 1.5 times the IQR. Outlier values are represented as points. Each VE estimate from the Delta and Omicron periods is derived from 64,283 and 79,697 survey responses, respectively.

To identify potential alterations in COVID-19 symptomatology, we evaluate the change in VE estimates for CLI definitions with a single anchor symptom, like loss of smell and taste. We reason that if symptoms are similar across variants, the within-anchor median change in VE will be similar across anchor symptoms. Our analyses provide evidence for a potential change in COVID-19 symptomatology from the Delta period to the Omicron period, as we note that some symptoms have more or less decline in VE (Fig.2b). Specifically, we find that CLI definitions that include loss of smell or taste have the smallest median change in VE (median: 0.31, IQR[0.34, 0.28]), while definitions with the largest median change include a cough, or sore throat (cough median: 0.49, IQR[0.52, 0.45]; sore throat median: 0.47, IQR[0.49, 0.45]). The observed pattern of change in VE across anchor symptoms is similar when evaluating VE estimates from individual countries (see Supplementary Fig.2), however, with increased uncertainty in estimates as measured by the span of anchor symptom distributions (seeSupplementary Results). Similarly, a survey-based study that used PCR testing data as the outcome demonstrated a shift away from symptomatology that includes loss of smell or taste and towards upper-respiratory type symptoms (i.e., sore throat) during the Omicron period13. Furthermore, a study conducted in Jalisco, Mexico, analyzed reported symptoms for confirmed infections with wild-type SARS-CoV-2, Delta, and Omicron variants, revealing that Omicron infections were linked to a higher incidence of runny nose and sore throat, aligning with the findings of our country-level analysis for Mexico (see Supplementary Fig.3)14. These results corroborate our overall findings, which also identified increased reporting of sore throat during a wave of COVID-19 infections dominated by the Omicron variant. Collectively, these findings suggest a shift in symptomatology associated with the Omicron variant towards more upper respiratory-type symptoms.

In addition to providing insights into changes in COVID-19 symptomatology, the VE estimates also include information about a vaccines ability to protect against COVID-19 illness presenting at different levels of severity as defined by pairwise combinations of symptoms. Importantly, we do not have information about the true severity of each respondents reported illness, and we instead infer severity based on the presence and absence of key symptoms. For instance, all CLI definitions that include at least a fever, cough, aches or muscle pain, sore throat, nausea, loss of smell or taste, or a headache in the absence of difficulty breathing or chest pain are considered mild syndromes. However, according to the NIH, CLI definitions that include difficulty breathing or chest pain are considered more severe forms of illness15. To understand potential changes in VE against mild and severe COVID-19 syndromes, we partition our CLI-informed VE estimates according to the above classifications. As a result, we end up with 42 mild and 21 severe definitions of CLI. We find that severe definitions of illness were more protected than mild definitions during the Delta period (median severe VE: 0.74, IQR[0.70, 0.79], median mild VE: 0.54, IQR[0.45, 0.64]) (Fig.3). However, protection against mild and severe illness was similar during Omicron (median severe VE: 0.30, IQR[0.25, 0.38], median mild VE: 0.22, IQR[0.16, 0.33]). Importantly, VE against severe illness may appear higher, as vaccines are producing milder illness when an individual is infected with COVID-1916, making it seem as if VE against mild illness is less effective. During the Delta wave of infections, we observed a total of 13,220 reports of mild illness and 5316 reports of severe illness. In contrast, during the Omicron wave of infections, there were 24,408 reports of mild illness and 10,234 reports of severe illness.

VE estimates for pairwise combinations of symptoms that include a fever, cough, aches or muscle pain, sore throat, nausea, loss of smell or taste, or a headache in the absence of difficulty breathing or chest pain (mild illness), and pairwise combinations of symptoms that include difficulty breathing or chest pain (severe illness). Each box represents the interquartile range (IQR) of estimates, with the horizontal line inside the box indicating the median. The whiskers extend to the largest/smallest values up to 1.5 times the IQR. Outlier values are represented as points. Each VE estimate from the Delta and Omicron periods is derived from 64,283 and 79,697 survey responses, respectively.

More:

Vaccine effectiveness against emerging COVID-19 variants using digital health data | Communications Medicine - Nature.com