Six-month sequelae of post-vaccination SARS-CoV-2 infection: a retrospective cohort study of 10,024 breakthrough infections

Introduction

The observation that individuals can be infected with SARS-CoV-2 after being vaccinated against COVID-19 (so-called breakthrough infections) has caused concerns (Nixon and Ndhlovu, 2021). These concerns are mitigated by abundant evidence that the risk of severe COVID-19 illness (as proxied by hospitalisation, admission to intensive care unit, and mortality) is lessened by vaccination (Agrawal et al., 2021; Antonelli et al., 2021; Bahl et al., 2021; Butt et al., 2021; Cabezas et al., 2021; Glatman-Freedman et al., 2021; Haas et al., 2021; Hyams et al., 2021; Mateo-Urdiales et al., 2021; Roest et al., 2021). One case-control study investigated the association between self-reported SARS-CoV-2 infection in 908 pairs of vaccinated and unvaccinated individuals and self-reported symptoms beyond 28 days (Antonelli et al., 2021). It found that compared to unvaccinated individuals, those with breakthrough infection were at a lower risk of symptoms beyond 28 days.

However, how COVID-19 vaccination affects the broad spectrum of sequelae of SARS-CoV-2 infection remains elusive. In particular, it is unknown if vaccinated individuals are at the same risk as unvaccinated individuals of venous thromboembolisms, ischaemic strokes, neuropsychiatric complications, long-COVID presentations, and other post-acute sequelae following infection with SARS-CoV-2. In addition, unvaccinated people might have health behaviours related to vaccination hesitancy (Latkin et al., 2021), and this potential source of bias has not been addressed by previous studies.

This cohort study based on electronic health records compares the 6-months outcomes of SARS-CoV-2 infection among individuals who were (vs. those who were not) vaccinated against COVID-19. We investigated a range of outcomes (both acute and post-acute) with documented associations with COVID-19 across multiple body systems.

Methods

Results

A total of 10,024 individuals with SARS-CoV-2 infections recorded at least 2 weeks after a first dose of vaccine against COVID-19 were identified (mean [SD] age at infection: 57.0 [17.9] years-old, 59.4% female). Among them, 65.1% were vaccinated with BNT162b2 ‘Pfizer/BioNTech’, 9.0% with mRNA-1273 ‘Moderna’, 1.6% with Ad26.COV2.S ‘Janssen’, and 24.4% with unspecified subtype. 9479 of these individuals were matched to 9479 individuals with SARS-CoV-2 infections who did not have a COVID-19 vaccine before SARS-CoV-2 infection. The main demographic features and comorbidities of both cohorts are summarised in table 1 (additional baseline characteristics presented in the appendix pp. 18-19). Adequate propensity-score matching (standardised mean difference < 0.1) was achieved for all comparisons and baseline characteristics and all subgroups (appendix pp. 18-27).

We estimated the HRs for the occurrence within 6 months of infection of a range of health events (combined with death in composite endpoints) previously documented to occur at an increased rate after SARS-CoV-2 infections. As seen in Fig. 1 and Fig. 2 (see Appendix pp. 9-12 for all Kaplan-Meier curves and pp. 28-29 for a summary table), compared to unvaccinated individuals, those who were vaccinated at the time of SARS-CoV-2 infection were at a significantly lower risk of composite outcome of death and respiratory failure (HR 0.70, 95% CI 0.63-0.78, Bonferroni-corrected p<0.0001), intubation/ventilation (HR 0.72, 95% CI 0.61-0.84, Bonferroni-corrected p=0.0024), hypoxaemia (HR 0.72, 95% CI 0.65-0.80, Bonferroni-corrected p<0.0001), seizures (HR 0.73, 95% CI 0.62-0.86, Bonferroni-corrected p=0.0057), ICU admission (HR 0.75, 95% CI 0.65-0.85, Bonferroni-corrected p<0.0001), psychotic disorder (HR 0.75, 95% CI 0.63-0.89, Bonferroni-corrected p=0.036), hair loss (HR 0.75, 95% CI 0.64-0.88, Bonferroni-corrected p=0.024), hypercoagulopathy or venous thromboembolism (HR 0.81, 95% CI 0.72-0.91, Bonferroni-corrected p=0.014), and oxygen requirement (HR 0.83, 95% CI 0.75-0.92, Bonferroni-corrected p=0.011). In contrast, there was no significant difference in the risk of many other outcomes including composite of death and any long-COVID feature (HR 1.01, 95% CI 0.96-1.05, p=0.83, Bonferroni-corrected p=1.0), Type 2 diabetes mellitus (HR 0.97, 95% CI 0.90-1.05, p=0.45, Bonferroni-corrected p=1.0), mood disorder (HR 1.05, 95% CI 0.96-1.15, p=0.27, Bonferroni-corrected p=1.0), and anxiety disorder (HR 1.06, 95% CI 0.97-1.15, p=0.20, Bonferroni-corrected p=1.0), among others. There was no violation of the proportionality assumption for 35 out of 45 outcomes (Appendix pp. 30-31 and p. 13 for the time-varying HR for the other 10 outcomes). For most of the other 10 outcomes (including death), the time-varying HRs were significantly lower than 1 (indicating a stronger association with vaccination status) earlier in the follow-up window.

• Download figure
• Open in new tab

Fig. 1

Hazard ratios for the outcome within 6 months of infection with SARS-CoV-2 between individuals vaccinated vs. unvaccinated against COVID-19. HR lower than 1 indicate outcomes less common among vaccinated individuals. Horizontal bars represent 95% confidence intervals. Each outcome is a composite endpoint with death as a component to address competing risks. The contribution of the outcome of interest to the overall incidence of the composite endpoint is encoded by the colour.

• Download figure
• Open in new tab

Fig. 2

Kaplan-Meier estimates for the incidence of outcomes of SARS-CoV-2 infection between vaccinated and unvaccinated individuals in the whole cohort (top) and for the incidence of death in different subgroups (bottom). 95% confidence intervals are shaded. For Kaplan-Meier curves of all other outcomes in the whole cohort, see appendix pp. 9-12. DVT=Deep vein thrombosis. PE=Pulmonary embolism.

Besides the outcomes shown to be significantly associated with vaccination in the primary analysis, those who had received two vaccine doses at the time of SARS-CoV-2 infection were also at a significantly lower risk of myalgia, myocarditis, cerebral haemorrhage, interstitial lung disease, urticaria, anosmia, and myoneural junction/muscle disease (each as composite endpoints with death), and at a significantly lower risk of death (Fig. 3 and appendix pp. 14, 32-33). In those who had received only one vaccine dose, HRs followed a similar pattern, though they were generally closer to 1 (Fig. 3 and appendix pp. 15, 34-35). For each of the outcome, the HR in those who had received 2 doses was not statistically significantly different from the HR in those who had received 1 vaccine dose (appendix pp. 36-37).

• Download figure
• Open in new tab

Fig. 3

Hazard ratios for the outcomes within 6 months of infection with SARS-CoV-2 between individuals who received one dose of the vaccine (vs. unvaccinated individuals), those who received two doses of the vaccine (vs. unvaccinated individuals), vaccinated vs. unvaccinated individuals under the age of 60, and vaccinated vs. unvaccinated individuals over the age of 60. HR lower than 1 indicate outcomes less common among vaccinated individuals. Horizontal bars represent 95% confidence intervals. Each outcome is a composite endpoint with death as a component to address competing risks. The contribution of the outcome of interest to the overall incidence of the composite endpoint is encoded by the colour. Only a subset of representative outcomes is displayed. The same figures with all outcomes are presented in the appendix pp. 14-17.

We found a substantial effect of age on the results. Many HRs in younger individuals (< 60 years-old) were in general lower (i.e. favouring vaccination even more) for outcomes significantly associated with vaccination (Fig. 3 and appendix pp. 16, 38-39). These lower HRs were observed against the backdrop of lower absolute risks for most outcomes among younger individuals (appendix pp. 38-39). In contrast, in individuals >60 years-old, although most trends were similar to those in the younger group, none of the HRs were statistically significantly different from 1 after Bonferroni correction (Fig. 3 and appendix pp. 17, 40-41). For all but one outcomes associated with vaccination status in the primary analysis, the HR in the older group was significantly higher (i.e. favouring vaccination less) than the HR in the younger group (Appendix pp. 42-43). For instance, the HR for the composite outcome of death or respiratory failure was 0.48 (95% CI 0.39-0.60) in individuals <60 years-old and 0.85 (95% CI 0.76-0.96) in those ≥60 years-old (ratio between the two: 1.77, 95% CI 1.38-2.26, p<0.0001). The only exception was the composite of death and hypercoagulopathy/DVT/PE (HR in the older group 0.93, HR in the younger group 0.72, ratio between the two 1.30, 95% CI 1.00-1.70, p=0.054).

Discussion

It is established that vaccination protects against hospitalisation, ICU admission, and death from COVID-19 (Agrawal et al., 2021; Antonelli et al., 2021; Bahl et al., 2021; Butt et al., 2021; Cabezas et al., 2021; Glatman-Freedman et al., 2021; Haas et al., 2021; Hyams et al., 2021; Mateo-Urdiales et al., 2021; Roest et al., 2021) but little was known about other outcomes of breakthrough SARS-CoV-2 infections. The data presented in this study, from a large-scale electronic health records network, confirm that vaccination protects against death and ICU admission following breakthrough SARS-CoV-2 infection and provide estimates of HR for these outcomes in a general population. Our study also shows that vaccination against COVID-19 is associated with lower risk of additional outcomes that had not been assessed in previous studies, namely respiratory failure, hypoxaemia, oxygen requirement, hypercoagulopathy or venous thromboembolism, seizures, psychotic disorder, and hair loss.

On the other hand, previous vaccination does not appear to be protective against several previously documented outcomes of COVID-19 such as long-COVID features, arrhythmia, joint pain, type 2 diabetes, liver disease, sleep disorders, and mood and anxiety disorders. The narrow confidence intervals (related to the high incidence of these outcomes post-COVID) rules out the possibility that these negative findings are merely a result of lack of statistical power. The inclusion of death in a composite endpoint with these outcomes rules out survivorship bias as an explanation. Instead, these negative findings might indicate that these outcomes arise through different pathophysiological mechanisms than outcomes which are affected by prior vaccination. For example, for anxiety and depression, it might be that antagonistic forces are at play with the protective effect of the vaccine being counteracted by the additional stressor of being infected despite being vaccinated.

The absence of a protective effect against long-COVID features is concerning given the high incidence and burden of these sequelae of COVID-19 (Taquet et al., 2021a). However, the risk of several individual long-COVID features were significantly associated with prior vaccination (but did not survive correction for multiple comparisons): myalgia (HR 0.78, 95% CI 0.67-0.91), fatigue (HR 0.89, 95% CI 0.81-0.97), and pain (HR 0.90, 95% CI 0.81-0.99), with potentially additional protection after a second dose of the vaccine against abnormal breathing (HR 0.89, 95% CI 0.81-0.98) and cognitive symptoms (HR 0.87, 95% CI 0.76-0.99). Relative differences in the incidence of individual long-COVID features might explain why our findings differ from those of an app-based survey suggesting that vaccination is overall protective against long-COVID symptoms (Antonelli et al., 2021). Other reasons might also explain this difference. First, the present study used data routinely collected from a general population rather than self-selected individuals. Second, this study uses a large sample size and matches cohorts on a broad range of recorded comorbidities. To further decrease selection bias, the control cohort of this study was selected among those having received an influenza vaccine thus helping control for the confounding effect of vaccine hesitancy and related health behaviours (Latkin et al., 2021). Third, by using symptoms and diagnoses extracted from health records rather than self-reported, our study might capture the most severe presentation of symptoms and these might be less prone to demand characteristics and other forms of detection bias. Finally, there might be differences in the type of vaccines used between the two study. In particular, no ChAdOx1 nCov-19 (‘Oxford/AstraZeneca’) vaccine was used in our study (as this vaccine is not used in the USA).

The findings that vaccination against SARS-CoV-2 does not protect against some of the post-acute outcomes of COVID-19 should not obscure the fact that vaccination remains an important protective factor against these outcomes at the population level, since the best way to prevent those outcomes is to prevent SARS-CoV-2 infection in the first place. This is also the case in those ≥ 60 years-old (Polack et al., 2020). However, our results highlight that some post-acute outcomes of SARS-CoV-2 (and notably long-COVID presentations) are likely to persist even after successful vaccination of the population, so long as breakthrough infections occur. These findings thus help in determining the necessary service provision. They also underline the urgency to identify other preventive or curative interventions to mitigate the impact of such COVID-19 sequelae.

Death was included in a composite endpoint with each outcome of interest to address competing risks (Manja et al., 2017). This implies that for outcomes much rarer than death (which is the minority of outcomes investigated), HRs might be driven by death rather than the outcome of interest. For instance, the 6-month incidence of psychotic disorders in the unvaccinated cohort was 1.11% whereas the death rate was 4.5% in this cohort. Hence the HR of 0.75 for the composite endpoint of death and psychotic disorder is in part driven by differences in death rate. Observing the HR for the outcome of interest in isolation (i.e. not composed with death) is informative so long as it is lower than 1. A HR lower than 1 suggests a lower rate of the outcome of interest despite survivorship bias (which in the present study increases HRs). If the HR is higher than 1, it is impossible to know whether the outcome is more common among vaccinated individuals or if it is a result of survivorship bias. HRs for all outcomes in isolation are provided in Appendix pp. 44-53 for the primary and secondary analyses. For instance, the HR for psychotic disorder measured in isolation was 0.79, indicating that the incidence of this outcome might indeed be lower in vaccinated than unvaccinated individuals and its significant HR might not be merely driven by differences in death rates.

Importantly, the protective effects of the vaccine against acute severity of infection and some post-acute sequelae appears to affect primarily those <60 years-old. In this group, the effects were large and robust, whereas in the ≥60 year-old group, effects were smaller and not statistically robust. This finding regarding age cannot be explained by differences in statistical power since the younger and older subgroups have similar sample sizes (4633 and 4657 respectively) and outcomes are in general more frequent in the older subgroup. Instead, it might be due to differences in contributions of pathophysiological pathways of breakthrough infections between younger and older individuals. In younger patients, effective B-cell response to vaccination might be followed by infection with variants against which antibodies have less neutralising activity (Wu et al., 2021). In older patients, the B-cell response to vaccination might itself be ineffective (Siegrist and Aspinall, 2009) so that the clinical presentation of breakthrough infection is similar to that of infection in unvaccinated individuals. Regardless of explanation, the seemingly age-limited benefits of vaccination on the severity and sequelae of SARS-CoV-2 infection has clear public health implications.

This study has several limitations beyond those inherent to research using EHRs (Casey et al., 2016; Taquet et al., 2021b) (summarised in the appendix p. 2) such as the unknown completeness of records, no validation of diagnoses, and sparse information on socioeconomic and lifestyle factors. First, we do not know which SARS-CoV-2 variant individual patients were infected with and this might affect the protective effect of vaccines. There is evidence that variants of concerns are overrepresented in breakthrough infections (McEwen et al., 2021). Since such variants tend to be associated with worse outcomes (Nyberg et al., 2021), their enrichment in breakthrough infections means that some HRs presented in this study might be conservative estimates. Second, it might be that vaccination status affects the probability to seek or receive medical attention, particularly for less severe outcomes. Third, this study says nothing about the outcomes in patients infected with SARS-CoV-2 but who did not get tested nor diagnosed with COVID-19. Fourth, this study did not investigate whether the association between vaccination status and outcomes of SARS-CoV-2 infection was moderated by timing of the vaccine with respect to the infection. Fifth, we could not compare the different vaccines against each other since the majority had received the Pfizer/BioNTech vaccine. Finally, as an observational study, causation cannot be inferred (though the specificity of the association with some outcomes and not others and the dose-response relationship support causality as an explanation).

In summary, the present data show that prior vaccination against COVID-19, especially after two doses, is associated with significantly less risk of many but not all outcomes of COVID-19, in younger but not older individuals. These findings may inform service planning, contribute to forecasting public health impacts of vaccination programmes, and highlight the urgent need to identify or develop additional preventive and curative interventions for sequelae of COVID-19.

Role of Funding Source

The funding source had no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.

Contributors

MT and PJH conceptualised the study, acquired funding, defined the methodology, and were involved in project administration and supervision. MT and QD curated the data, did the formal analyses, and validated the findings. MT created visualizations and wrote the original draft. MT, QD, and PJH reviewed and edited the manuscript for style and content. PJH and MT were granted unrestricted access to the TriNetX Analytics network for the purposes of research, and with no constraints on the analyses performed nor the decision to publish. The corresponding author had full access to all the data in the study, conducted the analysis, and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Data sharing

The TriNetX system returned the results of these analyses as csv files which were downloaded and archived. Data presented in this paper and the Appendix can be freely accessed at [URL to be added on publication]. Additionally, TriNetX will grant access to researchers if they have a specific concern (via the third-party agreement option).