Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof of Theorem 3.1 We know that, for each database D, ontology Σ and UBCQ q, it holds that D ∪ Σ |= q if and only if chase(D, Σ) |= q (Fagin et al. 2005). Therefore, also Dc ∪ Σc |= qc if and only if chase(Dc, Σc) |= qc. Moreover, by Proposition 3.2, we have that R(chase(Dc, Σc)) = chase(D, Σ) and R(qc) ≡ q. Hence, remain to prove that R(chase(Dc, Σc)) |= R(qc) if and only if chase(Dc, Σc) |= qc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | We prove the “if” part, given that the “only if” part can be obtained retracing the chain of the following implications. Suppose that chase(Dc, Σc) |= qc. Therefore, there is a homomor- phism h from at least one disjunct of qc, say ςj(ψj(Y j))c (where ςj is a canonical substitu- tion), to chase(Dc, Σc), that is h(ςj(ψj(Y j))c) ⊆ chase(Dc, Σc). Therefore, R(h(ςj(ψj(Y j))c)) ⊆ R(chase(Dc, Σc)). Moreover, note that R(h(ςj(ψj(Y j))c)) = h(R(ςj(ψj(Y j))c)). Hence, h(R(ςj(ψj(Y j))c)) ⊆ R(chase(Dc, Σc)). Thus, h is also a homomorphism from a disjunct of R(qc) to R(chase(Dc, Σc)), that is R(chase(Dc, Σc)) |= R(qc). |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof of Proposition 4.1 Let Σ be a shy ontology. Note that, for each rule ρ ∈ Σ, there exists a rule ς(ρ)c ∈ Σc such that ς(X i) = ni for each variable X i occurring in ρ. It is easy to see that a such ς is a safe substitution. We denote by ¯Σc the set of all and anly this kind of rules in Σc. Note that, if Σc is a shy ontology, then ¯Σc ⊆ Σc is also a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | By contradiction, suppose that ¯Σc is not a shy ontology. First, suppose that there exists a rule ς(ρ)c ∈ ¯Σc such that there exists a variable, say X, occurring in more than one body atom and X is not protected in body(ς(ρ)c). Therefore, for each existential variable Y , there exists an atom β ∈ body(ς(ρ)c) and some position pred(β)[i] in which X occurs, and pred(β)[i] is not invaded by Y . Consider the unpacked rule R(ς(ρ)c) = ρ ∈ Σ. Therefore, by construction, for each existential variable Y , there exists α ∈ body(ρ) and some position pred(α)[ j] in which X occurs, and pred(α)[ j] is not invaded by Y . Hence, X occurs in more than one body atom of ρ and X is not protected in body(ρ). So that, ρ is not a shy rule, and, thus, Σ is not a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Then, suppose that there exists a rule ς(ρ)c ∈ ¯Σc such that there are two distinct universal variables, say X and Y , that are not protected in body(ς(ρ)c); occur in head(ς(ρ)c); occur in two different body atoms; and they are attacked by the same variable. Therefore, there exists an existential variable Z such that X and Y occur only in invaded position by Z. Consider again the unpacked rule R(ς(ρ)c) = ρ ∈ Σ. Then, by the unpacking function, X and Y are not protected in body(ρ), and they occur in head(ρ), in two different body atoms, and only in invaded position by Z. Thus, they are attacked by the same variable. Therefore, also in this case, ρ is not a shy rule. Hence, Σ is not a shy ontology. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof of Proposition 4.2 Let M be a finite model of D ∪ Σ. Clearly, if M is a well-supported finite model of D ∪ Σ, we are done. Therefore, suppose that M is not a well-supported finite model of D ∪ Σ. Let Ω1 = (α1, . . . ,αm) be an ordering of the atoms of M. Hence, by assumption, there exists α ∈ M that is not a well-supported atom w.r.t. Ω1. Let αj1 be the first atom in the ordering Ω1 that is not well- supported. And consider a new ordering Ω2 = (α1, . . . ,αj1−1,αj1+1, . . . ,αm,αj1 ), where αj1 is shifted from the position j1 to the position n. As M 6∈ wsfmods(D, Σ), then Ω2 is not a well- supported ordering of M. Moreover, the first j1 − 1 atoms are well-supported w.r.t. Ω2. Therefore, let αj2 be the first atom in the ordering Ω2 that is not well-supported. Again, we consider a new ordering, say Ω3, where αj2 is shifted from position j2 − 1 to the position n. Iteratively, we build a sequence Ω1, Ω2, . . . , Ωm, . . . of orderings that are not well-supported. Note that, as the number of different orderings is finite, there exist at least two orderings in the sequence that are the same. Therefore, let Ωm1 and Ωm2 be the first two orderings of the sequence, with m2 > m1, such that Ωm1 = Ωm2 (i.e., Ωm1 and Ωm2 are the same ordering). Consider the subset A ⊆ M containing the first n − (m2 − m1) elements in Ωm1 , and the set B of the last m2 − m1 atoms in Ωm1 . By construction, A is a well-supported instance. Moreover, each β ∈ B is not well-supported by A, as Ωm2 = Ωm1. That is, there is no rule ρ in Σ and no homomorphism h such that h(body(ρ)) ⊆ A and h(head(ρ)) = {β}. Hence, as M is a model, whenever A |= body(ρ), there exists an atom α in A, such that α |= head(ρ). Therefore, A is a model. |
Processed_Finite_model_reasoning_over_existential_rules.txt | To complete the proof, let M be a finite minimal model of D ∪ Σ. As just proved, there exists a well-supported finite model M′ ⊆ M. By minimality of M, the model M′ must be equal to M. Therefore, M is a well-supported finite model. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Proof of Theorem 4.3 We have to prove that for each M ∈ wsfmods(Dc, Σc a), there exist M′ ∈ wsfmods(Dc, Σc) and a homomorphism h′ such that h′(M′) ⊆ M. Indeed, by hypothesis, there exists a homomorphism h such that h(q) ⊆ M′, and so (h′ ◦ h)(q) ⊆ M. |
Processed_Finite_model_reasoning_over_existential_rules.txt | a), and let (α1, . . . ,αm) be a well-supported ordering of M, and let (hα1i, . . . , hαmi) be a propagation ordering of (α1, . . . ,αm). If there exists a join rule ρ ∈ Σc satisfied by M with a null or a constant t in the join variables, then we consider the set of join atoms in the body of ρ w.r.t. the term t, say A ⊆ M. First, we substitute a term t of some α ∈ A in position l, with the corresponding term ht, j, ki of hαi, that can be considered as a fresh null. This new atom is denoted by α′, so that α′[l] = ht, j, ki. Then, for each αi ∈ M such that hαii[l] = ht, j, ki, for some position l, we set α′ i [l] = αi[l]. In this way, we build an instance M′ = {α′ : α ∈ M} of Σ, and a homomorphism h′ such that h′(ht, j, ki) = t, for each introduced fresh null ht, j, ki to substitute t. By construction, it holds that h′(α′) = α, so that h′(M′) = M. Note that, by construction, M′ is a well-supported finite instance of Dc ∪ Σc. |
Processed_Finite_model_reasoning_over_existential_rules.txt | Therefore, it remains to prove that M′ is a model of Dc ∪ Σc. By contradiction, suppose that M′ is not a model. Hence, there exists a rule ρ ∈ Σc such that M′ |= body(ρ), and M′ 6|= head(ρ). We distinguish two cases. |
Processed_Detecting_Edgeworth_Cycles.txt | responses as a proxy for the general public’s responses to various patterns in gasoline prices. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: The top panels and the bottom panels show examples of daily retail-price series in “cycling” and “non- cycling” station-quarter observations, respectively, for illustration purposes. The vertical axes measure retail gasoline prices in the Australian cent (left, middle) and the Euro cent (right), respectively. The horizontal axes represent calendar days. Note our main analysis uses retail margins (= retail minus wholesale prices) instead, thereby controlling for costs. |
Processed_Detecting_Edgeworth_Cycles.txt | to the correct finding (i.e., positive correlations) in WA, but some of them fail in NSW. |
Processed_Detecting_Edgeworth_Cycles.txt | underscore the relatively high profits that firms can earn when the discount factor is near 1. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: Each “manually labeled” station-quarter observation in the WA data is single-labeled as either “cycling,” “maybe cycling,” or “not cycling,” whereas the NSW and German data are triple-labeled. See main text for details. |
Processed_Detecting_Edgeworth_Cycles.txt | same protocol; a detailed case study of any single region/country is not our main objective. |
Processed_Detecting_Edgeworth_Cycles.txt | would require 24 times more labor: 480 hours $13.50, the total cost would be $155,520. |
Processed_Detecting_Edgeworth_Cycles.txt | classification, as cycles might appear in only one part of the graph but not others. |
Processed_Detecting_Edgeworth_Cycles.txt | follows the other within three hours in 90% of the cases, followed by three other major chains. |
Processed_Detecting_Edgeworth_Cycles.txt | case study of the German fuel markets might want to analyze intra-day patterns as well. |
Processed_Detecting_Edgeworth_Cycles.txt | that we propose, and (iii) the way we optimize the parameter values of each model. |
Processed_Detecting_Edgeworth_Cycles.txt | median price increase to the median price decrease, with 2 as a threshold to define cyclical subsamples. |
Processed_Detecting_Edgeworth_Cycles.txt | where Pi,t (f ) is the periodogram (6) of station-quarter (i, t), and θF T threshold parameter. |
Processed_Detecting_Edgeworth_Cycles.txt | interpolater, which is a spline where each piece is a third-degree polynomial of Hermite form.25 For each (i, t), we fit CS to its demeaned price series, pi,d ≡ meand∈t (pi,d), and count the number of times the fitted function CSi,t (d) crosses the d-axis (i.e., equals 0). |
Processed_Detecting_Edgeworth_Cycles.txt | with many latent states, and uses the terminal state s∗ as a latent score to detect cycles. |
Processed_Detecting_Edgeworth_Cycles.txt | accuracy, the composition of correct and incorrect predictions, and the associated parameter value(s), θ∗, for each method. |
Processed_Detecting_Edgeworth_Cycles.txt | The poor performance of NMC is surprising in three ways. First, it performed well in WA. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: See section 4 for the definition of each method. Appendix B.1 (Table 5) reports additional results for the variants of Methods 5–7. Columns (8)–(10) do not report parameter values because they contain too many parameters to be listed. We randomly split the sample into an 80% training subsample and a 20% testing subsample 101 times. In each split, the former subsample is used for setting parameter values, the medians of which are reported here. The accuracy statistics are also the medians from the 101 testing subsamples. |
Processed_Detecting_Edgeworth_Cycles.txt | mean differences are small relative to their standard deviations (see Tables 6 and 7 in Appendix B.2). |
Processed_Detecting_Edgeworth_Cycles.txt | Note: The exact numbers underlying these plots are reported in Panel (A) of Table 6 in Appendix B.2. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: Columns (1)–(10) use the median-accuracy version of each method in Table 2. The unit of measurement (of means and standard deviations) is the Australian cent in WA and NSW, and the euro cent in Germany, respectively. The p value indicates the probability that the difference in means is zero based on Welch’s t statistic and the approximate degrees of freedom. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: Column (0) is the same as in Table 3, which Column (00) refines by asymmetry based on negative median change. The unit of measurement (of means and standard deviations) is the euro cent. The p value indicates the probability that the difference in means is zero based on Welch’s t statistic and the approximate degrees of freedom. |
Processed_Detecting_Edgeworth_Cycles.txt | because it is the only one (among Methods 1–4) that captures the notion of cyclicality. |
Processed_Detecting_Edgeworth_Cycles.txt | ically require large amounts of training data, but the requirement is minimal in this context. |
Processed_Detecting_Edgeworth_Cycles.txt | is a positive, real-valued function of frequency f , and provides a convenient way to quantify the contribution of each frequency f to the signal g (x). |
Processed_Detecting_Edgeworth_Cycles.txt | tration in oligopolistic industries. A high value of the HHI indicates the market is close to monopoly. |
Processed_Detecting_Edgeworth_Cycles.txt | regressor is a constant term. The idea is that the frequencies with good fit will exhibit high ˜P LS (f ). |
Processed_Detecting_Edgeworth_Cycles.txt | This form ensures the observed values (p0, p1) and their slopes (m0, m1) are fitted exactly. It has become a default specification of CS in SciPy, a set of commonly used Python libraries for scientific computing. |
Processed_Detecting_Edgeworth_Cycles.txt | where ¯pi,d is the demeaned price. The information content of this statistic is similar to the previous one, but its calaculation is simpler. |
Processed_Detecting_Edgeworth_Cycles.txt | Forget Gate” variant due to Cho et al. (2014), which is also referred to as Gated Recurrent Units (GRUs) in the literature. Second, we do not include cl d−1 inside Λ in (12) or inside tanh and Λ in (13). This omission corresponds to their “No Peepholes” variant. Greff et al. |
Processed_Detecting_Edgeworth_Cycles.txt | rate of 0.0005, (iii) the number of epochs is 100, and (iv) the batch size is 30. |
Processed_Detecting_Edgeworth_Cycles.txt | We occasionally encounter cases in which a range of parameter values attain the same (maximum) accuracy. In such cases, we report the median of all θ∗ values that we find in our grid search. These cases typically involve “degenerate” predictions in which [cyclei,t (θ) = 1 or [cyclei,t (θ) = 0 for all (i, t), and hence are mostly irrelevant for the purpose of finding well-performing θs. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: See the text of Appendix sections A.1 and B.1 for the definition of each method. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: The numbers in panel (A) indicate accuracy in the 20% testing subsample. Some or all of the remaining 80% subsample are used as a training subsample to optimize the parameters of each model. We randomly split the sample 101 times, tune the parameters as many times, and report their median performances. The dollar costs in Panels (B) and (C) are based on the total RA hours to manually classify cycles and the hourly wage of $13.50 (see section 3). |
Processed_Detecting_Edgeworth_Cycles.txt | Note: These marginal-cost curves are based on the numbers reported in Table 6 (shown with markers) and a splines-based interpolation (dashed lines). |
Processed_Detecting_Edgeworth_Cycles.txt | Note: These alternative results are based only on “cleaner” subsamples that are either unanimously labeled as “cycling” by all RAs or not labeled as “cycling” by any RA. Other details follow Table 2. |
Processed_Detecting_Edgeworth_Cycles.txt | Note: These alternative results are based only on “cleaner” (i.e., less ambiguous) subsamples that are either unanimously labeled as “cycling” by all RAs or not labeled as “cycling” by any RA. |
Processed_Detecting_Edgeworth_Cycles.txt | on pricing: Evidence from the Collapse of a Gasoline Cartel,” Journal of Industrial Economics, 62 (2): 191–228. |
Processed_Symbolic_Controller_Synthesis_for_Büchi_Specificat.txt | ABSTRACT We consider the policy synthesis problem for continuous-state controlled Markov processes evolving in discrete time, when the specification is given as a Büchi condition (visit a set of states infin- itely often). We decompose computation of the maximal probability of satisfying the Büchi condition into two steps. The first step is to compute the maximal qualitative winning set, from where the Büchi condition can be enforced with probability one. The second step is to find the maximal probability of reaching the already computed qualitative winning set. In contrast with finite-state models, we show that such a computation only gives a lower bound on the maximal probability where the gap can be non-zero. |
Processed_Symbolic_Controller_Synthesis_for_Büchi_Specificat.txt | In this paper we focus on approximating the qualitative winning set, while pointing out that the existing approaches for unbounded reachability computation can solve the second step. We provide an abstraction-based technique to approximate the qualitative winning set by simultaneously using an over- and under-approximation of the probabilistic transition relation. Since we are interested in qual- itative properties, the abstraction is non-probabilistic; instead, the probabilistic transitions are assumed to be under the control of a (fair) adversary. Thus, we reduce the original policy synthesis problem to a Büchi game under a fairness assumption and charac- terize upper and lower bounds on winning sets as nested fixed point expressions in the µ-calculus. This characterization immediately provides a symbolic algorithm scheme. Further, a winning strategy computed on the abstract game can be refined to a policy on the controlled Markov process. |
Processed_Symbolic_Controller_Synthesis_for_Büchi_Specificat.txt | We describe a concrete abstraction procedure and demonstrate our algorithm on two case studies. We show that our techniques are able to provide tight approximations to the qualitative winning set for the Van der Pol oscillator and a 3-d Dubins’ vehicle. |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | ABSTRACT The Australia Telescope Compact Array (ATCA) has been used to image class I methanol masers at 9.9, 25 (a series from J=2 to J=9), 84, 95 and 104 GHz located in the vicinity of IRAS 16547-4247 (G343.12-0.06), a luminous young stellar object known to harbour a radio jet. The detected maser emission consists of a cluster of 6 spots spread over an area of 30′′. Five spots were detected in only the 84- and 95-GHz transitions (for two spots the 84-GHz detection is marginal), while the sixth spot shows activity in all 12 observed transitions. We report the first interferometric observations of the rare 9.9- and 104-GHz masers. It is shown that the spectra contain a very narrow spike (< 0.03 km s−1) and the brightness temperature in these two transitions exceeds 5.3 × 107 and 2.0 × 104 K, respectively. The three most southern maser spots show a clear association with the shocked gas traced by the H2 2.12 µm emission associated with the radio jet and their velocities are close to that of the molecular core within which the jet is embedded. This fact supports the idea that the class I masers reside in the interface regions of outflows. Comparison with OH masers and infrared data reveals a potential discrepancy in the expected evolutionary state. The presence of the OH masers usually means that the source is evolved, but the infrared data suggest otherwise. The lack of any class II methanol maser emission at 6.7 GHz in the source raises an additional question, is this source too young or too old to have a 6.7 GHz maser? We argue that both cases are possible and suggest that the evolutionary stage where the class I masers are active, may last longer and start earlier than when the class II masers are active. However, it is currently not possible to reveal the exact evolutionary status of IRAS 16547-4247. |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | Methanol masers are commonly found in massive star-forming re- gions. They fall into 2 categories first defined by Batrla & Menten (1988) and further considered by Voronkov et al. (2005b): class II masers (of which the 6.7 GHz is the best known and usually strongest) are closely associated with infrared sources and reside in the close environment of exciting (proto-)stars. In contrast, class I masers (e.g. 95 GHz) are usually found offset (by up to a par- sec) from continuum sources (e.g., Kurtz, Hofner & ´Alvarez 2004). Theoretical calculations strongly suggest that class I masers are pumped via collisions with the molecular hydrogen in contrast to class II masers, which have a radiative pump (e.g., Cragg et al. 1992; Voronkov 1999; Voronkov et al. 2005b). |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | masers and outflows in a number of sources (Plambeck & Menten 1990; Menten 1996), although another scenario, which involves cloud-cloud collisions, has also been proposed (Sobolev 1992; Mehringer & Menten 1996; Salii, Sobolev & Kalinina 2002). The common point of these two scenarios is the presence of shocks. There is observational evidence that methanol abundance is signifi- cantly increased in the shock processed regions (e.g., Gibb & Davis 1998) and the gas is expected to be heated and compressed in such regions providing more frequent collisions and, therefore, more ef- ficient pumping. |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | transition series known to exhibit class I maser activity (Voronkov 1999) and covered all known class I maser transitions within the frequency range of existing ATCA receivers. Only the J=5 tran- sition from the J2−J1 E series at 25 GHz has been previously observed towards IRAS 16547-4247 (Voronkov et al. 2005a), al- though with no position measurement. In this paper we present the first detection of 7 other transitions from this series as well as the first detection of the 84-GHz maser towards this source. The other 3 transitions were detected in a number of single dish and ATCA maser line surveys (Slysh et al. 1994; Val’tts et al. 2000, Voronkov et al. unpublished observations). It is worth mentioning that this source is the only example where all these class I maser transitions have been detected together, including the relatively rare 9.9 and 104 GHz masers. |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | IRAS 16547-4257 has recently been the focus of attention as the most luminous young stellar object (YSO) known to harbour a radio jet. At a distance of 2.9 kpc (Bronfman, private communica- tion), the YSO has a bolometric luminosity of 6.2 × 104 L⊙, equiv- alent to that of a single O8 Zero Age Main Sequence (ZAMS) star. The jet was first detected by Garay et al. (2003) using the ATCA at 1.4, 2.5, 4.8 and 8.6 GHz and subsequently observed using the VLA at 8.5 GHz and 14.9 GHz (Rodr´ıguez et al. 2005). This is the first reported case of a radio jet associated with a young O- type star. Located at the centre of a massive dense molecular core (approximately 103 M⊙), the jet is driving a highly energetic colli- mated bipolar outflow that extends over 1.5 pc (Brooks et al. 2003; Garay et al. 2006). |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | Observations were made using the ATCA in 2005 May-August with assorted array configurations and various correlator modes. The details, including the spectral resolutions, velocity coverages, full-width at half maximum (FWHM) of the primary beam, and synthesized beam parameters, are summarized in Table 1. We adopted the rest frequencies determined by M¨uller et al. (2004). The measurement uncertainties are given in Table 1 in parenthe- ses after each corresponding frequency. They are expressed in units of the least significant figure. Corresponding uncertainties in the radial velocity are given in the next column. |
Processed_Class_I_methanol_masers_in_the_outflow_of_IRAS_165.txt | The J=2,3 and 4 transitions of the J2−J1 E series near 25 GHz were observed in a single 8 MHz band centred on the average fre- quency of these transitions. The remaining transitions were ob- served using a separate frequency setup for each individual tran- sition. The correlator setup used for observations of the J=5 and 6 transitions on May 2 allowed us to observe the 24896 MHz con- tinuum emission with a 128 MHz bandwidth simultaneously with the spectral line observations. However, for the subsequent obser- vations we chose a higher spectral resolution rather than a con- tinuum channel. With the exception of the 9.9 GHz observations on June 16 two orthogonal linear polarizations were recorded. For June 16 only one linear polarization was recorded in order to obtain a larger number of spectral channels and achieve a spectral resolu- tion comparable with that of the higher frequency transitions. It is worth noting that Wiesemeyer, Thum & Walmsley (2004) found a number of mm-wavelength class I methanol maser transitions to be polarized. The correlator setup used in our study did not allow us to determine polarization properties. However, during the data reduc- tion both polarizations, if present, were averaged together. Hence, all the flux densities with the exception of that at 9.9 GHz measured on June 16 are not affected by polarization. For the latter we exam- ined the flux density measured from 10 individual scans taken at various hour angles assuming that the source was unresolved and located at the position found by the imaging. No parallactic an- gle dependence was found beyond the noise level of 0.3 Jy (1σ). This fact as well as a comparison with the May observations (see section 3.3) performed in the dual polarization mode leads us to conclude that polarization effects, if present, are less than the cali- bration uncertainty. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | We modelled emergence and spread of the Omicron variant of SARS-CoV-2 in Australia between December 2021 and June 2022. This pandemic stage exhibited a diverse epidemiological profile with emergence of co-circulating sub-lineages of Omicron, further complicated by differences in social distancing behaviour which varied over time. Our study delineated distinct phases of the Omicron-associated pandemic stage, and retrospectively quantified the adoption of social distancing measures, fluctuating over different time periods in response to the observable incidence dynamics. We also modelled the corresponding disease burden, in terms of hospitalisations, intensive care unit occupancy, and mortality. Supported by good agreement between simulated and actual health data, our study revealed that the nonlinear dynamics observed in the daily incidence and disease burden were determined not only by introduction of sub-lineages of Omicron, but also by the fluctuating adoption of social distancing measures. Our high-resolution model can be used in design and evaluation of public health interventions during future crises. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | On 24 November 2021 the B.1.1.529 variant of SARS-CoV-2 was reported to the World Health Organization (WHO) from South Africa, and was soon designated as a variant of concern, subsequently named Omicron 1,2. Around the same time, the Omicron variant reached Australia. The first cases attributed to Omicron were, for example, reported by the New South Wales Health Department during the week ending 4 December, 2021 3. The rapid spread of the Omicron variant dramatically changed the epidemiological situation in the country, starting a new, fourth, pandemic stage in Australia 4. This occurred just at the time when the previous, third, wave caused by the Delta (B.1.617.2) variant, had begun to subside, having been controlled to a large extent by a comprehensive mass-vaccination campaign, as well as strict and long lockdowns across several affected states and territories (e.g., New South Wales, Victoria, Queensland, and Australian Capital Territory) 5,6,7,8. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | A traditional description of a pandemic “wave” assumes that a wave includes both upward and/or downward periods, during which the rising/declining case numbers are substantial 9. The first three pandemic waves in Australia exhibited well-defined upward and downward periods. The first wave in the nation (March–June 2020) peaked at approximately 500 cases per day, i.e., around 20 daily cases per million 10. The second wave was mostly confined to the state of Victoria (June–September 2020), peaking at approximately 700 cases per day, i.e., around 30 daily cases per million 11,12. The third wave (June–November 2021) peaked at approximately 2,750 daily cases, i.e., around 100 daily cases per million. This peak formed by mid-October 2021, and the incidence stabilised in November between 1,200 and 1,600 daily cases, i.e., between 45 and 65 daily cases per million 13,6. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | However, the fourth pandemic stage caused by the Omicron variant (December 2021 – September 2022) has included several different waves. Notably, there were multiple peaks, with the first and largest one observed in mid-January 2022 (around 110,000 daily cases, or 4,250 daily cases per million), while smaller but still substantial peaks occurred in early April 2022 (around 57,000 daily cases, or 2,200 daily cases per million) and mid-May 2022 (around 51,000 daily cases, or 2,000 daily cases per million), see Fig. 1. The fourth pandemic stage was also characterised by the emergence of a number of sub-lineages of the Omicron variant, some of which rapidly spread across Australia 3: sub-lineage BA.1 (first reported in NSW by week 48, 2021), BA.2 (first reported in NSW by week 3, 2022), BA.4 (first reported in NSW by week 16, 2022) and BA.5 (first reported in NSW by week 17, 2022). Starting from March 2022, Omicron BA.1 was mostly replaced by BA.2, and from May 2022, BA.4 and BA.5 (see Fig. 1). The BA.3 sub-variant did not make a notable impact in Australia. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Different transmission rates were observed between sub-variants: BA.2 had a daily growth rate advantage relative to BA.1 of 0.10 (95% CI: 0.10-0.11) 14 and BA.4 and BA.5 have an estimated daily growth advantage of 0.08 (95% CI: 0.08–0.09) and 0.10 (95% CI: 0.09–0.11), respectively, relative to BA.2 15. In addition, BA.4 and BA.5 have resulted in more re-infections than BA.1 and BA.2 16,17,18. Pandemic effects of different Omicron sub-variants have not been traced within an individual-based age-dependent transmission and response model that takes into account demographic heterogeneity of the Australian population. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Age-dependent hospitalisation and fatality rates reportedly differ across all four Omicron sub-variants detected in Australia 19,3,20. In addition, during the Omicron pandemic period, most hospitalisations and deaths occurred among patients aged over 65 years, disabled patients, and those with three or more comorbidities 21. Thus, to model the actual impact of COVID-19 on mortality and ICU admissions, cases where mortality or critical illness (ICU admission) were directly attributable to the COVID-19 must be distinguished from those where this was not the case. This differential characterisation has has not been modelled for the Omicron pandemic stage in Australia. Hence, the impact of indirect disease burden remains an open question, and providing an adequate model, validated by the actual data from the Omicron pandemic period, may elucidate future public health approaches. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | The diverse epidemiological profile of the four co-existing Omicron sub-variants is complicated by fluctuations in the fraction of the Australian population that followed non-pharmaceutical interventions (NPIs), and specifically, social distancing (SD) measures. While during previous pandemic stages the social distancing measures in Australia were mandated by various “stay-in-home-orders”, these restrictions were significantly relaxed during the Omicron pandemic stage. As a result, rather than considering “SD compliance”, we focus on “SD adoption”. Following previous studies 10,22,6, we interpret social distancing measures as comprising several behavioural changes that reduce the intensity of interactions among individuals, including staying at home, mask wearing, physical distancing, etc, and thus, reduce the virus transmission probability (see Appendix A.2). Consequently, the population fraction following one or more of these behaviours, that is, reducing the overall intensity of interactions, is referred to as “SD-adopters”. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Substantial variability in SD-adoption during the Omicron pandemic stage is documented in several reports, including mobility reduction reports, shown in Supplementary Fig. A.12 and Table A.12. The mobility reduction data are categorised in a coarse-grained way (e.g., workplaces, public transport, residential, retail and recreation, etc.) 23. Hence, it is difficult to ascertain the extent of actual SD-adoption, i.e., the actual fraction of SD-adopters, at different times during the fourth pandemic stage, especially when interleaved with summer school holidays. It is, however, notable that during the rapid growth in incidence towards the first peak observed in mid-January 2022, the fraction of people staying at home (not travelling to workplaces) increased. Once the incidence significantly dropped during late February 2022, this fraction was reduced. When the incidence resumed its growth toward the second peak in mid-April, there was a slight but delayed increase in this fraction, albeit of a smaller magnitude. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | These observations of fluctuating SD-adoption indicate a degree of responsiveness within the population. However, such responsiveness has not yet been quantified and comprehensively modelled, with previous studies assuming only static fractions of individuals compliant with (or adopting) social distancing measures. Thus, there is a need to retrospectively and quantitatively estimate the fluctuating fraction of SD-adopters, treating it as a key variable which determined the pandemic response during the Omicron stage, in presence of other NPIs and vaccination interventions. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Overall, we aim to provide and validate a high-resolution pandemic response model on the scale of Australia. This model should be able to trace highly transmissible sub-variants of SARS-CoV-2, reproducing and explaining multiple observed waves, evaluate a range of dynamic public health interventions, and quantify their relative contributions to pandemic response. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | This study aims to delineate distinct phases of the fourth pandemic stage in Australia, induced by the Omicron variant. In doing so, we separate the dynamics induced by the two initial sub-variants (BA.1 and BA.2), as opposed to the subsequent sub-variants (BA.4 and BA.5). We then model the leading part of the fourth pandemic stage, induced by the two initial sub-variants of Omicron (BA.1 and BA.2), that is, the period between early December of 2021 and mid-April of 2022 (see Fig. 1). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Independently, we also aim to identify the periods of fluctuating adoption of social distancing requirements, matching the corresponding simulations with actual pandemic curves. In identifying the impact of the social distancing behaviour on the pandemic spread, we contrast the trajectory modelled for the fluctuating SD-adoption with its static counterparts, set at feasible extremes of SD-adoption (20% and 70%). In addition, we consider transition to the phase dominated by sub-variants BA.4 and BA.5, and possible reasons behind persistence of the Omicron variant of SARS-CoV-2 in Australia. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Finally, we model the disease burden of the fourth pandemic stage in Australia experienced until mid-April 2022, in terms of hospitalisations, intensive care units (ICU) occupancy, and mortality caused by the Omicron variant (BA.1 and BA.2). In doing so, we account for the impacts directly attributable to, or not directly attributable to COVID-19 on these parameters. This analysis considers a balance of the fluctuating extent of SD-adoption, transitions across the sub-variants, and the indirect effects of COVID-19, and explores to what extent this balance shaped the nonlinear dynamics across the three measures of disease burden. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | In order to capture the heterogeneous population structure, we used a high-resolution Agent-based Model (ABM) which comprised an artificial population of over 23.4 million agents generated from the Australian census data. The model was implemented within a large-scale software simulator (AMTraC-19: Agent-based Model of Transmission and Control of the COVID-19 pandemic in Australia 24) and included a range of non-pharmaceutical interventions and vaccination rollout schemes. This model has been previously calibrated and validated to reproduce key variant-specific epidemiological features of the COVID-19, including the ancestral variant 10,22 and the Delta variant 6,25. In this work, we calibrated the model to the Omicron variant (BA.1 and BA.2), using genomic surveillance reports 3; extended the social distancing model to support a fluctuating fraction of SD-adopters; carried out an additional sensitivity analysis; and validated the model using the incidence data reported between December 2021 and April 2022, as described in Supplementary Material. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | The model implements a specific profile for the natural history of the disease, progressing over several states of infection, i.e., Susceptible, Infectious (Asymptomatic or Symptomatic), and Removed (recovered or deceased). This dynamic is governed by several parameters, including the incubation period and recovery period, which are randomised across the agents around mean values. Susceptible hosts or agents which share a mixing context with an infectious agent may become infected and then infectious. Some agents develop asymptomatic infection, with this fraction set differently for adults and children. The asymptomatic infectivity is set lower than the symptomatic one. The model also specifies case detection probabilities which differ for symptomatic and asymptomatic/pre-symptomatic cases, see Supplementary Table A.2. Sensitivity analysis is described in Supplementary Material. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | A pandemic scenario is triggered by infecting a number of agents residing in metropolitan areas around international airports, in proportion to the number of incoming passengers 10,6, as described in Appendix A.1. At each simulation step, the mixing contexts of agents and their trajectory through the natural history of the disease determine their transmission probabilities. Once an agent is exposed to infection, the transmission probability varies: it exponentially grows to a peak of infectivity and then linearly declines during the agent’s recovery. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | In modelling the fourth pandemic stage, we assumed that the population vaccination coverage was high. This assumption is supported by the vaccination levels achieved in Australia by the end of 2021. The proportion of adult Australians (i.e., over the age of 16) who were double vaccinated by 30 November 2021 reached 87.2%, and 92.5% of adults have had by then at least one dose 29. The implemented vaccination scheme follows a pre-emptive mass-vaccination campaign with two vaccines: priority and general (see Supplementary Material). This reflects key features of the actual vaccine rollout in Australia during 2021 22,6. When an agent is vaccinated, their vaccination state is defined by a set of the corresponding vaccine efficacy parameters. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | The model includes several NPIs, including case isolation, home quarantine, school closures, and social distancing behaviour 10,6. Case isolation and home quarantine are activated from the start of simulation. The setup and duration of school closures approximates the period of summer holidays and a gradual resumption of school activities until mid-March of 2022. The extent of social distancing adoption is simulated dynamically, with the SD-adopting population fraction set at any given simulation step according to a predefined assignment profile, see Table A.5. Whenever the SD-adoption fraction changes during the simulation, Bernoulli sampling determines the agents which are selected to follow the social distancing behaviour. Vaccination states, the adoption of social distancing and the compliance with other NPIs modify the transmission probabilities within the agents’ mixing contexts, as described in Supplementary Material. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Estimation of disease burden is carried out in terms of hospitalisation, ICU occupancy and mortality (daily and cumulative deaths). In doing so, we used age-dependent case hospitalisation risks (CHRs) and ICU ratios, scaled to the actual hospitalisation cases in Australia, and age-dependent infection fatality rates (IFRs), while accounting for adjusted vaccine efficacies and a differentiation between sub-variants BA.1 and BA.2, as described in Supplementary Material. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Phase 3: BA.1 dominant, epidemiological week 50 (2021), ending 18 December, 2021, to week 2 (2022). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Phase 5: BA.1 and BA.2 co-exist, epidemiological week 4 to week 9, ending 5 March, 2022. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Phase 7: BA.2, BA.4 and BA.5 co-exist, epidemiological week 16, ending 23 April, 2022, onwards1. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | We note that sub-variant BA.1 has displaced the Delta variant by phase 3, causing a rapid growth in incidence towards the primary incidence peak. The second incidence peak shaped during phase 6 which was dominated by sub-variant BA.2. Phase 7 marked the emergence of BA.4 and BA.5. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | We re-calibrated the ABM to the incidence data observed during the beginning of the fourth pandemic stage in Australia, driven by the Omicron variant (specifically, phase 3 in Fig. 1 shaped by BA.1, but considering also phases 4 to 6 when BA.1 was concurrent with BA.2). For the previous pandemic stage in Australia (June–November 2021), i.e., the third pandemic wave shaped by the Delta variant, the reproductive number was calibrated to be around R0 = 6.20 6. In this study, the calibration resulted in R0 = 19.56 (see Supplementary Material, including Appendix A.6 and Table A.3), which is consistent with other reports estimating R0 of Omicron BA.1 to be approximately 3.1 times as high as the reproductive number of the Delta variant 30. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | In modelling the first six phases during which the two initial sub-variants of Omicron (BA.1 and BA.2) emerged and reached dominance, we focused on a “retrodictive” estimation of the SD-adopting population fraction. A key assumption that we made is that the adoption of social distancing behaviour fluctuates over time and the changes in the extent of SD-adoption are not necessarily synchronous with the pandemic phases. Instead, these changes can be driven by individual perception of the infection risks, life pattern changes associated with summer holidays and return to school/work for many individuals, vaccination availability for children, and other policy setting changes affecting social distancing. Hence the changes in the extent of SD-adoption may precede or follow some salient observed features of the pandemic trajectory. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Selection of the SD-adoption change-points and the corresponding fractions of SD-adopters is a subject of modelling which aims to minimise the difference between the resulting and observed incidence curves. For example, the SD-adoption profile presented in Table A.5 includes six change-points, including the initial assignment, at day 0, of SD = 0.3, that is, 30% of the agents follow social distancing behavior from the pandemic stage onset. This is set to continue until day 54 after which the fraction of SD-adopters sharply rises to 70%, i.e., SD = 0.7 from day 55, and so on. This optimised profile produced the average incidence curve shown in Fig. 2, closely approximating the observed incidence curve, with both primary and secondary peaks well-aligned with the observations. The optimisation process is described in Supplementary Material. The incidence curve produced by the fluctuating (dynamic) SD-adoption is contrasted with two static SD-adoption alternatives, set at 20% and 70%, i.e., SD1 = 0.2 and SD2 = 0.7. This comparative analysis, illustrated by Fig. 2, highlights the nonlinear nature of resultant dynamics: the fluctuating fraction of SD-adopters produces the trajectory which agrees with the actual incidence dynamics, unlike the static SD-adoption profiles which both fail to match the salient pandemic patterns. For example, the lower static SD-adoption, SD1 = 0.2, does not produce the secondary peak at all, and the higher static SD-adoption, SD2 = 0.7, misses the first peak by almost two months and produces the second peak with a larger magnitude than the first one (see Supplementary Table A.13). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 1: Incidence and disease burden of COVID-19 in Australia between 26th November 2021 and 13th June 2022. Shaded areas in grey and blue show the emergence of variants of concern and sub-lineages over time, identified in weekly genomic surveillance reports (NSW Health). Top (a): Black (y-axis, left): a 7-day moving average of reported COVID-19 daily incidence. Orange (y-axis, right): COVID-19 related deaths. Middle (b): Black (y-axis, left): COVID-19 hospitalisations (bed occupancy). Orange (y-axis, right): COVID-19 ICU cases (bed occupancy). Bottom (c): Black (y-axis, left): COVID-19 ICU cases (bed occupancy); Orange (y-axis, right): Daily COVID-19 related deaths (solid), daily COVID-19 deaths (dashed). The data for COVID-19 cases, hospitalisation and ICU occupancy, and mortality are published by the Australian government. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | maintain the infection spread. This can be explained by depletion of the susceptible host population within the simulation: the simulated cumulative incidence at the end of phase 6 reached 23% of the simulated population, with the rest being probabilistically protected by vaccinations (given the partial efficacy of both vaccines). Since the model does not assume re-infections and does not simulate diminishing vaccine efficacy, this divergence strongly suggests that the actual incidence observed during phase 7, associated with the spread of BA.4 and BA.5 (including the third peak, see Fig. 1), was mostly generated by re-infections or new infections in vaccinated individuals with lower vaccine effectiveness (e.g., due to waning vaccine efficacy), rather than infections of the epidemiologically or immunologically “naïve” population (see Discussion). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 2: A comparison of incidence produced by dynamic social distancing (SD) adoption (shown in solid black) and static SD-adoption fractions (SD1 = 0.2, shown in red; SD2 = 0.7, shown in green). Coloured shaded areas around the solid line show standard deviation. Changes in dynamic SD-adoption are marked by vertical dashed black lines. Traces corresponding to each simulated scenario are computed as the average over 20 runs. Simulated SD adoption is combined with other interventions (i.e., school closures, case isolation, and home quarantine). A 7-day moving average of the actual time series (black crosses) is shown for the period between 26th November 2021 and 13th June 2022 (x-axis, bottom). The simulated incidence is scaled up by 10% to reflect the population increase from 23.4 million (2016 census data, model input) to 25.8 million (2021 census data). The simulated incidence is offset by 12 days (x-axis, top) to align with the observed incidence peak. Shaded areas in grey and blue show the emergence of variants of concern and sub-lineages over time, identified in weekly genomic surveillance reports (NSW Health). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 3: A comparison of hospitalisations (bed occupancy) among dynamic social distancing (SD) levels (shown in solid black) and static SD levels (SD1 = 0.2, shown in red; SD2 = 0.7, shown in green). The simulated hospitalisations are offset by 7 days. Coloured shaded areas around the solid line show standard deviation. Changes in dynamic SD-adoption are marked by vertical dashed black lines. Traces corresponding to each simulated scenario are computed as the average over 20 runs. SD adoption is combined with other interventions (i.e., school closures, case isolation, and home quarantine). The actual time series (black crosses), shown from 26th November 2021, aligns with the start of the Omicron outbreak in Australia. Shaded areas in grey and blue show the emergence of variants of concern and sub-lineages over time, identified in weekly genomic surveillance reports (NSW Health). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | deaths) in Australia, as described in Supplementary Material. The key input to the disease burden estimation is provided by the incidence curves produced by the simulation. In other words, a given SD-adoption profile generates a non-linear incidence trajectory which in turn results in hospitalisations and ICU occupancy. The incidence trajectory also leads to estimation of daily and cumulative deaths. Similarly to the analysis of daily incidence, we contrast dynamic and static SD adoption. The difference, however, is that we no longer optimise the SD assignment profile, and use the optimised profile presented in Table A.5. Thus, a comparison between the actual and estimated disease burdens can be used to further validate the model. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 3 and Supplementary Table A.14 show that the hospitalisation curve (occupancy) generated by the dynamic SD-adoption closely matches the actual hospitalisations. This alignment is again in stark contrast with the hospitalisation curves produced by static SD-adoption fractions (SD1 = 0.2 and SD2 = 0.7). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | than the actual ICU occupancy. The ICU occupancy model, driven by ICU admission rates, accounted for a reduction in the disease severity, differentiating between BA.1 and BA.2 (cf. Supplementary Table A.8). Furthermore, we adjusted the ICU trajectory during phase 6 to represent the ICU occupancy directly attributable to COVID-19, rather than the ICU occupancy in patients with COVID-19 (as explained in Appendix A.4). The adjusted ICU occupancy is in a better concordance with the actual data. The static SD-adoption fractions (SD1 = 0.2 and SD2 = 0.7) do not capture the actual trajectory of ICU occupancy. Finally, Fig. 5, Fig. A.7 and Supplementary Table A.14 show that the daily mortality curve produced by the dynamic SD-adoption matches the actual data reasonably well again. In contrast, the mortality estimated by static SD-adoption fractions, SD1 = 0.2 and SD2 = 0.7, consistently overestimate and underestimate the actual observations, respectively. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | These estimations of the disease burden reinforce the argument that the nonlinear dynamics of the observed hospitalisations, ICU occupancy, and mortality result from a nuanced combination of (i) fluctuating fractions of SD-adoption, (ii) a transition from BA.1 to BA.2, and (iii) indirect effects of COVID-19, i.e., the distinction between ICU admissions or deaths arising due to COVID-19 versus those with COVID-19. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Emergence of recurrent epidemic waves of COVID-19 is a systemic phenomenon and its quantitative understanding remains an elusive target 31. Typically, several interacting factors contribute to recurrent dynamics. Firstly, the population structure in the modern world is highly heterogeneous: not only at the level of individual households and local government areas, but also at the regional, national and international level. This heterogeneity forces the infection transmission to follow complex paths, producing distinct urban and rural waves 32,26,33, disproportionate effects on high- and low-density housing 34, as well as in the areas characterised by socioeconomic disadvantage profiles and higher concentrations of essential workers 35,6,36,37. Thus, while infection rates may decline in some areas, they may escalate in others. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Secondly, the population compliance with various stay-at-home orders and adoption of social distancing requirements fluctuates in time: strong initial adherence is typically replaced by sizable fatigue and rarely recovers beyond moderate compliance, thus reducing effectiveness of NPIs 38,6,39. This fatigue cannot be exclusively attributed to “contrarian” individuals and affects broad population groups. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Thirdly, vaccine efficacy wanes over time, requiring booster vaccine doses, which temporarily reestablish protective immunity. Independently, following a typically strong initial vaccination uptake, “the fully up to date” vaccinated population fraction slowly reduces over time, undergoing possible upswings when the risk perception increases 40,41. Similar to the NPI fatigue, this dynamic is driven not only by “anti-vaxxer” groups, but may depend on access to and knowledge about vaccination 42, and involve several drivers of vaccine hesitancy, including confidence, complacency, altered risk calculation, and limited collective responsibility 43,44. In addition, the testing, tracing, isolation and quarantine (TTIQ) capacities themselves may be under stress during rapidly growing outbreaks, diminishing their role in curbing the epidemic. The TTIQ capacities were severely strained in Australia during the rise of the Omicron variant, leading to a decline in their effectiveness 45. Moreover, delays in imposing strict NPIs or initiating vaccination rollouts amplify non-linearly, so that short response lags cause long recovery tails 10,6. All these cyclical spatiotemporal dynamics may combine over time, creating vicious feedback loops and superposition of interacting waves. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | In this study, we sought to increase our understanding of the mechanisms generating multiple pandemic waves. We specifically modelled a pandemic response to the emergence and spread of highly transmissible variant of SARS-CoV-2 (Omicron and its sub-lineages) during the fourth pandemic stage in Australia. This stage produced several local incidence peaks, and we identified the corresponding phases (see Fig. 1), followed by retrospective modelling of dynamic social distancing behaviour. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Importantly, the identification of distinct phases did not reduce the analysis to a description of well-defined waves, but rather attributed the incidence peaks and patterns to nuanced combinations of new sub-variants and fluctuations in social distancing behaviour. For example, the secondary peak in mid-April of 2022 (see phase 6 in Fig. 1) was explained not only by the emergence of BA.2 which started to dominate the preceding BA.1, but also by the reduced adoption of social distancing requirements. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 4: A comparison of ICU (bed occupancy) among (i) dynamic social distancing (SD) levels, adjusted for phase 6 by scaling with 0.77 to capture ICU cases due to COVID-19 (shown in solid black), (ii) non-adjusted dynamic SD levels to represent all ICU cases (dotted black), and (iii) static SD levels (SD1 = 0.2, shown in red; SD2 = 0.7, shown in green). The simulated ICUs are offset by 7 days. Coloured shaded areas around the solid line show standard deviation. Changes in dynamic SD-adoption are marked by vertical dashed black lines. Traces corresponding to each simulated scenario are computed as the average over 20 runs. SD adoption is combined with other interventions (i.e., school closures, case isolation, and home quarantine). The actual time series (black crosses), shown from 26th November 2021, aligns with the start of the Omicron outbreak in Australia. Shaded areas in grey and blue show the emergence of variants of concern and sub-lineages over time, identified in weekly genomic surveillance reports (NSW Health). |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | Figure 5: A comparison of daily deaths between dynamic social distancing (SD) levels (shown in solid black) and static SD levels (SD1 = 0.2, shown in red; SD2 = 0.7, shown in green). The simulated daily deaths are offset by 14 days. Coloured shaded areas around the solid line show standard deviation. Changes in dynamic SD-adoption are marked by vertical dashed black lines. Traces corresponding to each simulated scenario are computed as the average over 20 runs. SD adoption is combined with other interventions (i.e., school closures, case isolation, and home quarantine). Actual daily deaths are derived from reported weekly deaths (shown in orange; solid: COVID-19 related deaths; dashed: COVID-19 deaths). Shaded areas in grey and blue show the emergence of variants of concern and sub-lineages over time, identified in weekly genomic surveillance reports (NSW Health). The timeline is divided into 6 phases as follows: 1) BA.1 detected, 2) Delta and BA.1 co-exist, 3) BA.1 dominant, 4) BA.2 detected, 5) BA.1 and BA.2 co-exist, and 6) BA.2 dominant. |
Processed_Persistence_of_the_Omicron_variant_of_SARS-CoV-2_i.txt | The main result of this modelling is that the fraction of SD-adopters was found to be fluctuating in response to the incidence dynamics. Essentially, these changes appeared in response to the pandemic dynamics observed by the population, sometimes lagging behind and occasionally preempting pandemic turns. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.