The variant screening tests were performed by Cerba Laboratory (Saint Ouen L’Aumône, France) on samples that originated from partner centers in mainland France and tested positive for SARS-CoV-2 with a generic qPCR assay. Most of the samples originated from the general population (Tables 1, 2). We did not have access to additional details about patient symptoms; however, according to an earlier study on a similar cohort, nearly all the samples originated from nasopharyngeal swab specimens, and the proportion of symptomatic and asymptomatic individuals were comparable among the positive tests (14). To limit epidemiologic biases, we removed persons >80 years or <5 years of age from the dataset.

We first analyzed 131,478 screening tests performed during September 1–December 18, 2021. The assays used over this first period were ID SARS-CoV-2/VOC Evolution Pentaplex (ID Solutions, https://www.id-solutions.fr) (93,554 tests), VariantDetect (PerkinElmer, https://www.perkinelmer.com) (33,037 tests), and VirSNiP (TIB Molbiol, https://www.tib-molbiol.de) (4,887 tests). These tests targeted 3 mutations in the SARS-CoV-2 spike protein: E484K (mutation A), E484Q (mutation B), and L452R (mutation C). Denoting the absence of a mutation by a 0 and its presence by 1, A0B0C1 mostly corresponds to infections caused by the Delta variant, A0B0C0 to the Alpha or Omicron variant or an ancestral lineage, A0B1C1 to Kappa or Kappa-like variants, A1B0C0 to the Beta or the Gamma variant, and A1B0C1 to a Delta variant with an E484K mutation.

Because of the shift in variant frequencies, new screening assays were implemented in late 2021. We analyzed 193,256 tests performed during December 6, 2021–February 28, 2022, all using the assay ID SARS-CoV-2/VOC Revolution Pentaplex (ID Solutions). This assay still targeted mutations A and mutation C but also targeted S:K417N (mutation D). Denoting nontested mutations with a 9, then A0B9C1D0 most likely indicates infections caused by the Delta variant, A0B9C0D1 by the Omicron variant, A1B9C0D0 by the Gamma variant, A1B9C0D1 by the Beta variant, and A0B9C0D0 by the Alpha variant or the B.1.640 lineage. Finally, A0B9C1D1 can either indicate an infection by Delta with a 417N mutation, Omicron with a 452R mutation, or a Delta–Omicron co-infection.

For the ID Solutions Pentaplex tests, we analyzed 4 Ct values. Three of these values correspond to primers targeting the mutations of interest: S:417N, S:452R, or S:484K, the last to a primer targeting the nucleoprotein gene, which was used as a control.

Next-generation sequencing (NGS) was performed by Cerba Laboratory for 16,973 samples with a Ct <30 using the CovidSeq amplicon-based NGS assay according to supplier recommendations (Illumina, https://www.illumina.com) and after a Janus/Chemagic RNA extraction (Perkin Elmer) from the nasopharyngeal swab. All sequences obtained were submitted to the EMERGEN Consortium Database (Santé Publique France, https://www.santepubliquefrance.fr/dossiers/coronavirus-covid-19/consortium-emergen) and GISAID (https://www.gisaid.org).

We first analyzed variant-specific screening tests collected during September 1–December 18, 2021 (Figure 2, panel A). Most of these tests originated from the general population (96.6%) and showed coverage differences between regions of France (Table 1). The most common assay used (71%) was that from ID Solutions (Evolution). The raw number of tests performed follows the incidence curve of the epidemic (Appendix 1 Figure 1).

Focusing on the tests performed during October 25–December 18, 2021 (i.e., when the epidemic was increasing), we used a multinomial regression model to identify covariates associated with the test outcome (Table 3). A0B0C0 infections (consistent with Omicron) were found in younger persons than were A0B0C1 infections (consistent with Delta); RRR was 0.85 (95% CI 0.83–0.88) per age unit (equal to 56 years in this study). We also detected strong temporal increases in most of the regions of France that had RRR >10 per day. In some regions, we detected a temporal increase of A0B1C1 tests, consistent with the Kappa variant. Finally, in our dataset, the (rare) A1B0C1 tests only showed a slight temporal increase in 2 regions (Bretagne and Hauts-de-France).

We then estimated growth advantages of A0B0C0 over A0B0C1 infections during 21-day time windows. The advantage was adjusted for covariates and assumed to be constant over each window. In September 2021, A0B0C0 infections were spreading less rapidly than A0B0C1 (Figure 2, panel B). This finding is consistent with the rapid increase of the Delta variant at the time (18). The pattern shifted at the end of November, with a 50% growth advantage of A0B0C0 infections, which increased to 105% (95% CI 96.1%–114%) in the last time window. According to this model, A0B0C0 infections became more frequent than A0B0C1 infections during the week of December 20 (Figure 2, panel C), with strong variations across regions (Appendix 1 Figure 1).

The new screening test targeting the K417N mutation enabled us to better document the spread of the Omicron variant (Table 2). In December 2021, the A0B0C0 wave was mainly caused by viruses bearing the K417N mutation (Figure 3, panel A). Furthermore, the proportion of A0B0C0 tests not attributable to Omicron decreased toward the end of the year. Finally, we also noted potential co-infections of Omicron and Delta in December.

We then estimated the growth advantage of A0B9C0D1 over A0B9C1D0 during December 6, 2021–February 28, 2022 (Figure 3, panel B). The resulting estimate (96.5% [95% CI 87.9%–105%]) is very consistent with the results obtained using a less-specific test on the early stages of the wave.

We observed a shift between the Omicron waves in the different regions of France (Figure 4). For instance, in the South-East area, Delta was still dominant during week 51 of 2021. As expected, we also saw that tests consistent with co-infections of Omicron and Delta were more frequent in regions where the 2 variants were cocirculating in substantial frequencies. This shift in different regions can explain the second increase in growth advantage of A0B0C0 tests observed in November (Figure 2, panel C).

Because variant screening tests only target 3 mutations, we analyzed whole-genome sequences of ≈5% of the positive samples (Figure 5, panel A). This analysis revealed that before October 2021, A0B0C0 tests mostly originated from Delta variant infections, whereas in November they originated from rare lineages or from the 20C lineage. A more precise analysis shows that these mostly correspond to the B.1.640 lineage. Beginning near the end of November, half of these tests were associated with the Omicron variant; this percentage increased to >80% during December.

Beginning in the second week of January 2022, some of the screening outcomes consistent with Omicron (A0B9C0D1) were associated with the BA.2 variant (Figure 5, panel B). This proportion increased over the next several weeks. Using the sequencing data, we estimated a growth advantage of BA.2 over the BA.1 Omicron lineage of 48.9% (95% CI 44.2%–53.6%). BA.2 accounted for most variants at the end of February, meaning that the Omicron variant BA.1 lineage only dominated the epidemic in France for <3 months (Figure 5, panel B).

For the tests performed during December 16, 2021–February 28, 2022, we used a linear model to explore differences in Ct values between variants. All the covariates were significant according to ANOVA with a type II error (Appendix 1 Table 1). Ct values tended to decrease with age or to be lower in samples from hospitals (Appendix 1 Table 2), which is consistent with earlier results (14). Furthermore, A0B9C0D1 tests exhibited significantly higher Ct values than A0B9C1D0; fitted median values were 22.1 versus 21.4 (Figure 6, panel A). This result suggests lower amounts of genetic material in the samples.

To further investigate these patterns, we analyzed the Ct values of the mutations targeted by the assay. We found that the Ct for the 417N mutation was higher in single infections (A0B9C0D1) than in co-infections (A0B9C1D1) (Figure 6, panel B). This finding is consistent with the greater ability of Omicron compared with Delta to infect immunized hosts, assuming that such breakthrough infections have a lower virus load (27,28). For the 452R mutation, we found the opposite pattern (Figure 6, panel C).

Finally, we analyzed the Ct values of the control gene as a function of the virus lineage inferred from the NGS data (Figure 6, panel D). BA.1 samples had higher Ct values than did Delta samples. Furthermore, BA.2 samples had lower Ct values than did BA.1 samples.

On December 22, 2021, we incorporated the inferred growth advantage of Omicron/BA.1 over Delta into Covidsim (21) to explore an optimistic and a pessimistic scenario running through mid-March 2022. These scenarios differed in terms of the assumptions made regarding the reduction of Omicron virulence compared with Delta (3-fold vs. 2-fold) and vaccine protection against infection (75% vs. 40%) and severe illness (95% vs. 80%). Even though our assumption that the epidemic was under control at the end of 2021 was too optimistic, both the optimistic and the pessimistic scenarios showed that CCU activity was likely to remain high over January and February 2022, which proved to be accurate (Appendix 1 Figure 4).

Given our estimations of the frequency of the Omicron/BA.2 sublineage in the population and its growth advantage over BA.1, we can predict the temporal increase of the epidemic Rt. We compared this predicted Rt with that calculated for the period March 1–10, 2022, using national hospital admission data, and found that from March 3 the ratio between the 2 was greater than unity (Figure 7, panel A). This result suggests that the epidemic growth cannot solely be explained by variant replacement and involves other drivers (e.g., the end of the holiday periods in some regions starting February 21, 2022).

Finally, on March 17, 2022, by using consolidated estimates of relative virulence (6) and vaccine effectiveness (23) for Omicron variants, we explored 2 prospective scenarios for nationwide COVID-19 CCU activity depending on the intensity of the relaxation of the control over the epidemic: Rt at the peak as 1.1 or 1.6 (Figure 7, panel B). We found that a new hospital peak was possible in the more pessimistic case, but its height remained below half of the peak experienced during the first Omicron wave in January.