Prevalence of persistent SARS-CoV-2 in a big neighborhood surveillance research – Nature

ONS-CIS

This work incorporates statistical knowledge from ONS, which is Crown Copyright. Using the ONS statistical knowledge on this work doesn’t indicate the endorsement of the ONS in relation to the interpretation or evaluation of the statistical knowledge. This work makes use of analysis datasets that will not precisely reproduce Nationwide Statistics aggregates.

The ONS-CIS is a UK household-based surveillance research during which participant households are approached at random from tackle lists throughout the nation to offer a consultant pattern of the inhabitants²¹. All variations of the research protocol can be found at https://www.ndm.ox.ac.uk/covid-19/covid-19-infection-survey/protocol-and-information-sheets. All people 2 years of age and older from every family who present written knowledgeable consent present swab samples (taken by the participant or mother or father or carer for these underneath 12 years of age), no matter signs, and full a questionnaire at assessments. The survey supplied members the choice of solely having one enrolment evaluation (taken by roughly 1%), or weekly assessments for just one month (taken by roughly 1%; Prolonged Information Fig. 1). All different enrolled members (roughly 98%) have been assessed weekly for the primary month of their enrolment within the survey after which roughly month-to-month (initially for 1 12 months; all such members have been approached for re-consent for ongoing follow-up past 1 12 months). The survey had rolling recruitment to fulfill its goal for taking a sure variety of swabs from the inhabitants every month, however in observe, most recruitments occurred between September and December 2020 (Supplementary Data; additionally see supplementary desk 4 in ref. ⁵⁰). The rolling recruitment enabled the research to attain its total pattern numbers (required to handle its surveillance goals) whereas accounting for members withdrawing from the research. As is commonplace, the protocol additionally allowed a 14-day window across the roughly month-to-month assessments (shifting any following assessments to keep away from swabbing members once more at very quick (and variable) discover); crucially, assessments weren’t missed to fulfill survey targets.

Because the overwhelming majority of recruitment comes from invites despatched to households randomly chosen from tackle lists that we wouldn’t have related demographic info, we aren’t in a position to evaluate traits of these agreeing and never agreeing to take part. From 26 April 2020 to 31 July 2022, assessments have been performed by research employees visiting every family; from 14 July 2022 onwards, assessments have been distant, with swabs taken utilizing kits posted to members and returned by put up or courier, and questionnaires accomplished on-line or by phone. For this evaluation, we included knowledge from 2 November 2020 to fifteen August 2022, spanning a interval from Alpha to Omicron BA.2 sequences inside the ONS-CIS dataset (Prolonged Information Desk 1).

Up to now, of 535,731 members recruited into the ONS-CIS, 109,417 (20%) have both accomplished their participation after a single enrolment go to, visits just for the primary month or just for the primary 12 months (7%) or withdrawn (13%; see Supplementary Data). Transferring home was a significant cause for finishing participation within the survey (as this results in members now not being eligible for follow-up as it’s the authentic tackle that’s sampled), a small variety of members died (0.4%), and in July 2022, the survey moved to a distant knowledge assortment method at which level some members selected to finish their participation. For the time interval of this research, 96.2% of swabs had a destructive consequence and 1.9% had a constructive consequence (1.9% have been void). For these with constructive take a look at outcomes, the imply time for the reason that earlier evaluation was 35.2 days and to the subsequent evaluation was 37.1 days. For these with a destructive take a look at, the related numbers have been 31.8 days and 33.0 days. By definition, 100% of first constructive samples from every persistent an infection had a subsequent evaluation. There was no statistical distinction within the time between sampling for people with persistent an infection in contrast with these testing constructive (Supplementary Data).

Sequencing

From December 2020 onwards, sequencing was tried on all constructive samples with Ct ≤ 30; earlier than this date, sequencing was tried in actual time wherever attainable, with some extra retrospective sequencing of saved samples. The overwhelming majority of samples have been sequenced on Illumina Novaseq, with a small quantity utilizing Oxford Nanopore GridION or MINION. One among two protocols have been used: the ARTIC amplicon protocol⁵¹ with consensus FASTA sequence recordsdata generated utilizing the ARTIC nextflow processing pipeline (v1)⁵², or veSeq, an RNA sequencing protocol based mostly on a quantitative focused enrichment technique^19,53 with consensus sequences produced utilizing shiver (v1.5.8)⁵⁴. Throughout our research interval, we recognized 94,943 people with a single sequence and 5,774 people with two or extra sequences. Right here we solely included sequences with 50% or extra genome protection.

Figuring out candidate persistent infections

We first recognized people with two or extra sequenced samples taken not less than 26 days aside. We selected this cut-off as a result of the vast majority of people with acute an infection shed the virus for lower than 20 days and now not than 30 days within the respiratory tract^24,55. Given the acute heterogeneity within the shedding profiles throughout some acute infections^24,55, we additionally thought-about a extra conservative 56-day cut-off for some analyses. Choice was based mostly on availability of sequences, which have been required for genetic evaluation; it was not attainable to permit for failure to establish any long-term shedding as a consequence of members not having assessments/swabs or assessments failing or subsequent positives having Ct > 30, and subsequently not being despatched for sequencing. Nevertheless, which means some persistent infections are more likely to have been missed and so our estimates ought to be thought-about a decrease certain.

Candidate persistent infections have been outlined in one in every of two methods: (1) pairs of sequenced samples that belonged to the identical main lineage, and (2) pairs of sequenced samples the place one or each had no outlined phylogenetic lineage, however the place the genetic distance between them was decrease than that required to distinguish two main lineages (Prolonged Information Fig. 9). The main lineages that we thought-about have been Alpha (B.1.1.7), Delta (B.1.617.2), Omicron BA.1 and Omicron BA.2, together with their sublineages. We assumed pairs belonging to completely different main lineages have been both co-infections or reinfections with two completely different virus lineages. Solely candidate persistent infections have been thought-about in additional evaluation.

Figuring out persistent infections

We decided whether or not two sequences from the identical particular person are from the identical an infection by whether or not they share a uncommon SNP at two or extra consecutive time factors relative to the population-level consensus. If an intermediate sequence from that particular person had an unknown nucleotide at a website (as a consequence of poor protection), whereas the primary and final sequences shared a uncommon SNP, then the intermediate sequence was additionally assumed to be a part of the identical an infection. Uncommon SNPs have been outlined as people who have been shared by fewer than a threshold variety of sequences, belonging to every main lineage, inside the full ONS-CIS dataset (Prolonged Information Fig. 2). The thresholds have been chosen to maximise the variety of persistent infections recognized whereas minimizing the variety of false positives (see under).

To find out the false-positive price, for every main lineage, we generated a dataset of 1,000 randomly paired sequences from completely different people within the ONS-CIS, every sampled not less than 26 days aside. We decided the proportion of those pairs that might have been incorrectly recognized as persistent infections as a perform of the edge for figuring out whether or not a SNP is uncommon (Prolonged Information Fig. 2). Though the full variety of persistent infections that we recognized (among the many listing of candidate persistent infections) grew as the edge for figuring out whether or not a SNP is uncommon elevated, at very excessive thresholds, the speed of false positives (among the many listing of randomly paired sequences) was additionally excessive. In our research, we selected a threshold of 400 sequences (comparable to all sequences of the identical main lineage inside the full ONS-CIS dataset) for all the main lineages, giving a false-positive price (figuring out an an infection as persistent when it was not) of 0–3%. Utilizing this threshold, roughly 92–98% of all sequences from the 4 main lineages had a uncommon SNP relative to the major-lineage population-level consensus.

Figuring out reinfections with the identical main lineage

Any pair of sequences from the identical particular person, of the identical main lineage and not less than 26 days aside have been thought-about as candidate reinfections. Of those, pairs that had not less than one nucleotide distinction on the consensus stage, and didn’t share any uncommon SNPs, have been classed as reinfections. Pairs that had no similar uncommon SNPs, nor any nucleotide variations on the consensus stage, have been classed as undetermined.

Pattern mix-ups may inflate the true variety of reinfections. Within the ONS-CIS, every pattern has a novel barcode, a small minority of barcodes are constructive, and even fewer nonetheless have a Ct ≤ 30; subsequently, random swapping of barcodes is unlikely to lead to a unsuitable constructive sample with Ct ≤ 30 being despatched for sequencing. For every weekly sampling batch, we additionally checked concordance between lineage from the sequencing laboratory and S gene goal failure from the testing laboratory; concordance between Ct from the testing laboratory and genome protection from the sequencing laboratory (excessive protection is anticipated for low Ct, and low protection for top Ct); and for veSeq, a log-linear relationship between the variety of mapped reads from the sequencing laboratory and Ct from the testing laboratory¹⁹.

Phylogenetic evaluation

For every of the 4 main lineages, we selected 600 consensus sequences with not less than 95% protection from the ONS-CIS dataset utilizing weighted random sampling, with every pattern of main lineage i collected in week j given a weight 1/n_ij, the place n_ij is the variety of sequences of main lineage i collected throughout week j²². These sequences have been added as a background set to the gathering of all consensus sequences for samples from persistent infections and reinfections. Mapping of every sequence to the Wuhan-Hu-1 reference sequence was already carried out by shiver, and thus a full alignment for every of the 4 lineages could possibly be constructed utilizing solely this.

Most probability phylogenetic timber have been constructed utilizing IQ-TREE (v1.6.12)⁵⁶ utilizing the GTR+gamma substitution mannequin and ultrafast bootstrap⁵⁷. Every tree was rooted utilizing the gathering dates of the samples and the heuristic residual imply sq. algorithm in TempEst⁵⁸. Visualization used ggtree⁵⁹.

Measuring the variety of impartial appearances of mutations and their health results

To search out the frequency with which mutations (not together with deletions) that we recognized throughout persistent infections are represented in cross-sectional samples from the inhabitants and their between-host stage health, we used the outcomes from ref. ²⁹ on the estimated variety of appearances of mutations from a consultant world dataset of roughly 6.5 million SARS-CoV-2 sequences (for variety of appearances: https://github.com/jbloomlab/SARS2-mut-fitness/blob/predominant/outcomes/mutation_counts/aggregated.csv; for estimating the health impact of mutations: https://github.com/jbloomlab/SARS2-mut-fitness/blob/predominant/outcomes/aa_fitness/aamut_fitness_by_clade.csv), in addition to a subset of these sequences which might be solely sampled from England (arguably extra related to our sequences from the ONS-CIS). When doing this, we managed for main lineage, which means, for instance, if a mutation occurred in a BA.1 persistent an infection, we solely thought-about the variety of instances it appeared on the BA.1 phylogeny. To map between Pangolin lineages and Nextstrain clades, we assumed B.1.1.7 ≡ 20I, B.1.617.2 ≡ {21A,21I,21J}, BA.1 ≡ 21K and BA.2 ≡ {21L,22C,22D}. We additionally in contrast the frequency and health impact of mutations that appeared in two persistent infections (that’s, recurrent mutations) and people who appeared in just one persistent an infection (that’s, single mutations) as reported in ref. ²⁹.

Estimating the proportion of infections which might be persistent

We recognized 381 and 54 infections that lasted 30 days or longer and 60 days or longer, respectively. Evaluating this with the variety of people that had sequenced samples belonging to Alpha, Delta, BA.1 or BA.2, we recognized roughly 0.49% (381 of 77,561) and 0.07% (54 of 77,561) of infections with not less than one pattern that could possibly be sequenced as persistent for 30 days or longer and 60 days or longer, respectively. Because the ONS-CIS is a consultant pattern of people from the final inhabitants, we are able to estimate the proportion of all SARS-CoV-2 infections that grew to become persistent for 1 month or longer, and which have intermittent excessive viral masses. To do that, we have to decide the likelihood {that a} persistent an infection with one sequenced pattern has not less than yet another sequenced pattern. As most persistent infections most likely final 1–3 months, and with out realizing the true viral kinetics throughout persistent an infection, this may be approximated because the likelihood {that a} persistent an infection has virus that may be sequenced on any given day of sampling.

At one excessive, if a typical persistent an infection has a virus pattern that may be sequenced for less than 4 days per thirty days (assuming viral dynamics much like one acute an infection every month), solely 14% of persistent infections could be detected by roughly month-to-month sampling. Correcting for this, we might estimate the proportion of detected infections which might be persistent within the normal inhabitants for 30 days or longer to be 3.5%, calculated because the ratio of the estimated prevalence of persistent infections (0.49%) to the detection price (14%). Equally, for infections persisting 60 days or longer, the estimated share could be 0.5% (0.07%/0.14). On the different excessive, if we assume typical persistent infections have sequenceable virus for 20 days per month and, subsequently a detection price of 71%, we might estimate the proportion of detected infections which might be persistent infections within the normal inhabitants for 30 days or longer to be 0.7% (0.49%/0.71) and for 60 days or longer to be 0.1% (0.07%/0.71).

Evaluating viral load actions and signs

To quantify the adjustments in viral load actions throughout persistent infections, we in contrast Ct values on the final time level a sequence was obtained to when the primary sequence was collected. Likewise, for reinfections, we in contrast the adjustments in Ct worth between the first an infection and reinfection. We used a paired Student’s t-test to calculate P values in each instances because the distribution of variations in Ct values have been usually distributed for each persistent infections (W = 0.99, P = 0.28) and reinfections (W = 0.99, P = 0.78) as decided by the Shapiro–Wilk take a look at⁶⁰.

We additionally tracked 12 signs constantly solicited from all members at each evaluation. Signs have been fever, weak point/tiredness, diarrhoea, shortness of breath, headache, nausea/vomiting, sore throat, muscle ache, stomach ache, cough, lack of scent and lack of style. At every follow-up evaluation, members have been requested whether or not these 12 signs had been current previously 7 days (obligatory query accomplished in any respect assessments the place a swab was taken). Symptom discontinuation was outlined as the primary prevalence of two successive follow-up visits with out reporting signs. To match symptom counts throughout persistent infections and reinfections, we used the two-sided paired Wilcoxon take a look at because the distribution of symptom variations isn’t usually distributed (Fig. 3e). For calculation of P values and visualization of histograms and field plots, we used Mathematica (v13.1.0.0).

Lengthy COVID evaluation

Attributing persistent signs to a earlier SARS-CoV-2 an infection is tough within the absence of a diagnostic take a look at for lengthy COVID, and lengthy COVID instances are recognized to be under-recorded in digital well being data⁶¹. Lengthy COVID standing was subsequently self-reported by research members, so we can’t exclude some participants’ signs being attributable to a medical situation apart from COVID-19. From February 2021, at each evaluation, members have been requested “would you describe your self as having lengthy COVID, that’s, you’re nonetheless experiencing signs greater than 4 weeks after you first had COVID-19, that aren’t defined by one thing else?”.

When estimating lengthy COVID prevalence on this evaluation, we thought-about the primary evaluation not less than 12 weeks and not less than 26 weeks after an infection. Our comparability group comprised all people with a constructive PCR take a look at and Ct ≤ 30 on the first constructive take a look at, excluding the people with persistent an infection recognized on this research, over the identical time span as persistent infections such that first constructive take a look at was inside the vary of dates of the primary constructive take a look at among the many persistent an infection group. Though the underlying research design for ONS-CIS is a cohort research, this particular evaluation of lengthy COVID focuses on evaluating persistent to non-persistent infections when it comes to the chance of subsequent self-reported lengthy COVID (binary outcomes, not less than 12 weeks and not less than 26 weeks following the primary constructive take a look at). Some lacking knowledge have been inevitable, given the timeframe of the research and participant completion or withdrawal (see above); total, the lengthy COVID query was not accomplished at 368,161 of 6,797,789 (5.4%) of assessments in the course of the research interval from 4 February 2021 when it was launched, with 93% and 86% of members with out persistent an infection however with a constructive take a look at with Ct < 30 having a response to the lengthy COVID query not less than 12 and 26 weeks after an infection, respectively (Prolonged Information Fig. 1). Evaluation used full instances, that’s, excluded those that didn’t have a response to the lengthy COVID query on this timeframe (Prolonged Information Fig. 1). As these are binary outcomes fairly than a time-to-event end result, both an odds ratio or a relative danger could possibly be used to guage the chance of lengthy COVID in people with persistent an infection; right here we used odds ratio. The truth that some persistent infections have been most likely missed as a consequence of sequencing solely being tried in excessive viral load samples and as a consequence of missed assessments signifies that our estimates of the affect of persistent an infection are more likely to be biased in the direction of the null, that’s, the true results of persistent an infection are most likely bigger than we estimate. Observe-up from the beginning of an infection to first lengthy COVID response was related between persistent and non-persistent infections (Desk 3).

In calculating the chances ratio of lengthy COVID in people with persistent an infection relative to the comparability group, we used a binary logistic regression mannequin and accounted for confounding variables comparable to age on the final birthday, intercourse, Ct worth, calendar date, space deprivation quintile group, presence of self-reported long-term well being situations (binary), vaccination standing (unvaccinated or single vaccinated, absolutely vaccinated or booster vaccinated 14–89 days in the past, absolutely vaccinated or booster vaccinated 90–179 days in the past, absolutely vaccinated or booster vaccinated 180 or extra days in the past) and days from first constructive take a look at to lengthy COVID follow-up response. All variables besides the final one have been outlined on the time of the primary constructive take a look at. Steady variables (age, Ct worth, calendar date and days to follow-up response) have been modelled as restricted cubic splines with a single inside knot on the median of the distribution and boundary knots on the fifth and ninety fifth percentiles. Vaccination standing was derived from a mixture of CIS and Nationwide Immunisation Administration System (NIMS) knowledge for members in England, and CIS knowledge alone for members in Wales, Scotland and Northern Eire. Given the variety of potential confounders included, we didn’t take a look at for interplay (impact modification). We didn’t take a look at for goodness of match as a result of the mannequin was solely used to regulate for measured confounders of the connection between persistent positivity and lengthy COVID, which we chosen on substantive, fairly than empirical, grounds (that’s, utilizing a causal inference method).

Though we managed for a lot of confounders that might doubtlessly affect our lengthy COVID evaluation, of be aware, age, intercourse, vaccination standing and former an infection, there should be unknown residual confounders that may affect our outcomes. We have been additionally unable to carry out the lengthy COVID evaluation for the reinfection group as a result of low variety of members on this cohort who reported new-onset lengthy COVID 12 weeks or longer or 26 weeks or longer after infections.

Reporting abstract

Additional info on analysis design is obtainable in the Nature Portfolio Reporting Abstract linked to this text.