SARS-CoV-2 rapid antibody test results and subsequent risk of hospitalisation and death in 361,801 people

The study population included 361,801 participants, of whom 143,774 (39.7%) tested positive on self-administered LFIAs in rounds 5 and 6 (January–February 2021 and May 2021, respectively) of the REACT-2 study. Table 1 shows the characteristics of the study population overall and by “high” or “low” COVID-19 risk based on the UK Joint Committee on Vaccination and Immunisation (JCVI) definition of being clinically extremely vulnerable for COVID-19 (see Supplementary Methods, S.1). The mean age was 55.5 years; 31.4% were aged 50 years or under; and 56.0% were women. A single vaccination was reported by 28.9% of participants and two vaccinations by 24.3%. Over 1 in 6 (15.8%) reported prior COVID-19, although only 24.9% of these cases had been confirmed (e.g. by polymerase chain reaction (PCR) test) as most occurred in the first COVID-19 wave when community testing was limited.

Among the 192,637 people who had received one or more vaccinations at least 14 days prior to the LFIA test, 128,282 (66.6%) were LFIA positive; 88,026 people had received two vaccinations, of whom 74,534 (84.7%) tested positive. Of the 14,271 with prior COVID-19 confirmed with a test, 12,065 (84.5%) were LFIA positive; of the 57,317 who reported confirmed or suspected prior COVID-19, 29,700 (51.8%) were LFIA positive.

We obtained follow-up data on hospitalisations to September 30, 2021, with mean follow-up time of 184 days and a total of 182,659 person years of follow-up. For mortality we obtained data to December 1, 2021, with a mean follow-up time of 245 days and a total of 243,125 person years of follow-up.

For hospitalisations, 16,802 (4.6%) people were admitted for any cause (excluding pregnancy and accident/injury related, see “Methods”), 5330 of whom were unplanned (emergency) admissions, 91 had COVID-19 as the primary diagnosis and 151 had COVID-19 listed anywhere on their hospital record. There were 389 deaths, including 19 with COVID-19 as the underlying condition and 21 with COVID-19 mentioned anywhere on the death certificate (Table 1).

In N = 25,824 high-risk individuals with one or more vaccinations, those testing LFIA negative had a risk of all-cause mortality of 3.6, compared to 1.7 per 1000 in those testing LFIA positive, an absolute difference of 1.9 per 1000 (Supplementary Table 1).

Figure 1 shows Kaplan–Meier plots of the cumulative hazard functions for hospitalisations and deaths in individuals who had received one or more vaccinations (N = 192,604). Those testing negative on the LFIA antibody test had higher cumulative hazards for all outcomes at almost all time points during follow-up. Findings were similar when stratified by risk group, by age and by vaccination status (Fig. 1 and Supplementary Fig. 1).

95% pointwise confidence intervals are shown in shaded colour.

In mutually adjusted Cox regression models accounting for age, sex, risk group, prior COVID-19 and vaccination count, those with a negative LFIA after one or more vaccinations had an increased mortality from all causes (HR: 1.59 [1.07, 2.37], and mortality with COVID-19 as underlying cause (HR: 20.6 [1.83, 232]), as well as increased risk of hospitalisation with COVID-19 as the primary diagnosis (2.73 [1.15, 6.48]).

After two or more vaccinations, those with a negative LFIA had increased all-cause mortality (HR: 1.87 [1.11, 3.13]), mortality with COVID-19 as underlying cause (HR: 12.46 [1.13, 138]), hospitalisation with COVID-19 as the primary diagnosis (2.79 [1.01, 7.72]) and hospitalisation with COVID-19 mentioned (2.54 [1.01, 6.41]) compared to people who tested positive (Table 2 and Supplementary Tables 2 and 3).

Stability selection

We used 100x subsampled Least Absolute Shrinkage and Selection Operator (LASSO)-penalised Cox models with demographic, biological and health variables together with LFIA result to identify which were preferentially selected as predictive of outcomes (see “Methods”). In these multivariable models, LFIA result was not selected as a predictor of subsequent hospitalisation, emergency hospitalisation, hospitalisation with COVID-19 or death in the population with one or more vaccines, nor when stratified by risk group (Fig. 2). The main positive predictors selected were risk group, age (continuous), anaemia, smoking and body mass index (BMI, continuous).

Results shown for five main outcomes (l-r: hospitalisation, emergency hospitalisation, hospitalisation with COVID-19, hospitalisation with COVID-19 mentioned, death) and for vaccinated (one or more) cohort (top, N = 192,604), high-risk vaccinated cohort (middle, N = 35,812) and low-risk vaccinated cohort (bottom, N = 156,792).

Boosted tree models

We used boosted tree models trained on different subsets of predictors, with and without LFIA, cross-validated on unseen data, to assess the predictive value of LFIA result over and above the other variables (age, sex, risk group, prior COVID-19 history and vaccination count, see “Methods”) in the population who had received one or more vaccines. Adding LFIA result to the boosted tree model did not substantially improve the prediction of hospitalisation or death (Fig. 3); however predictive accuracy was improved by including additional survey data (BMI, smoking status, deprivation, education status, comorbidities) to the model. The findings were consistent when analysis was repeated in the full population (Supplementary Fig. 3).

Models built using boosted tree models (Catboost) with 5-fold cross-validation. Outcome is a binary event within follow-up y/n variable. 95% confidence intervals, derived from 1000x bootstrap resamples, are shown in shaded colour around the curves.