Using test positivity and reported case rates to estimate state-level COVID-19 prevalence and seroprevalence in the United States

The public health impact of COVID-19 variants of concern on the effectiveness of contact tracing in Vermont, United States

Case investigation and contact tracing (CICT) are public health measures that aim to break the chain of pathogen transmission. Changes in viral characteristics of COVID-19 variants have likely affected the effectiveness of CICT programs. We estimated and compared the cases averted in Vermont when the original COVID-19 strain circulated (Nov. 25, 2020-Jan. 19, 2021) with two periods when the Delta strain dominated (Aug. 1-Sept. 25, 2021, and Sept. 26-Nov. 20, 2021). When the original strain circulated, we estimated that CICT prevented 7180 cases (55% reduction in disease burden), compared to 1437 (15% reduction) and 9970 cases (40% reduction) when the Delta strain circulated. Despite the Delta variant being more infectious and having a shorter latency period, CICT remained an effective tool to slow spread of COVID-19; while these viral characteristics did diminish CICT effectiveness, non-viral characteristics had a much greater impact on CICT effectiveness.

Estimating actual SARS-CoV-2 infections from secondary data

Eminent in pandemic management is accurate information on infection dynamics to plan for timely installation of control measures and vaccination campaigns. Despite huge efforts in diagnostic testing of individuals, the underestimation of the actual number of SARS-CoV-2 infections remains significant due to the large number of undocumented cases. In this paper we demonstrate and compare three methods to estimate the dynamics of true infections based on secondary data i.e., (a) test positivity, (b) infection fatality and (c) wastewater monitoring. The concept is tested with Austrian data on a national basis for the period of April 2020 to December 2022. Further, we use the results of prevalence studies from the same period to generate (upper and lower bounds of) credible intervals for true infections for four data points. Model parameters are subsequently estimated by applying Approximate Bayesian Computation-rejection sampling and Genetic Algorithms. The method is then validated for the case study Vienna. We find that all three methods yield fairly similar results for estimating the true number of infections, which supports the idea that all three datasets contain similar baseline information. None of them is considered superior, as their advantages and shortcomings depend on the specific case study at hand.

Optimization of Screening Strategies for COVID-19: Scoping Review

BACKGROUND

COVID-19 screening is an effective nonpharmaceutical intervention for identifying infected individuals and interrupting viral transmission. However, questions have been raised regarding its effectiveness in controlling the spread of novel variants and its high socioeconomic costs. Therefore, the optimization of COVID-19 screening strategies has attracted great attention.

OBJECTIVE

This review aims to summarize the evidence and provide a reference basis for the optimization of screening strategies for the prevention and control of COVID-19.

METHODS

We applied a methodological framework for scoping reviews and the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) checklist. We conducted a scoping review of the present publications on the optimization of COVID-19 screening strategies. We searched the PubMed, Web of Science, and Elsevier ScienceDirect databases for publications up to December 31, 2022. English publications related to screening and testing strategies for COVID-19 were included. A data-charting form, jointly developed by 2 reviewers, was used for data extraction according to the optimization directions of the screening strategies.

RESULTS

A total of 2770 unique publications were retrieved from the database search, and 95 abstracts were retained for full-text review. There were 62 studies included in the final review. We summarized the results in 4 major aspects: the screening population (people at various risk conditions such as different regions and occupations; 12/62, 19%), the timing of screening (when the target population is tested before travel or during an outbreak; 12/62, 19%), the frequency of screening (appropriate frequencies for outbreak prevention, outbreak response, or community transmission control; 6/62, 10%), and the screening and detection procedure (the choice of individual or pooled detection and optimization of the pooling approach; 35/62, 56%).

CONCLUSIONS

This review reveals gaps in the optimization of COVID-19 screening strategies and suggests that a number of factors such as prevalence, screening accuracy, effective allocation of resources, and feasibility of strategies should be carefully considered in the development of future screening strategies.

Impact of the COVID-19 Pandemic on Inpatient Utilization for Acute Neurologic Disease

Background and Objective: The initial months of the Corona Virus 2019 (COVID-19) pandemic resulted in decreased hospitalizations. We aimed to describe differences in hospitalizations and related procedures across neurologic disease. Methods: In our retrospective observational study using the California State Inpatient Database and state-wide population-level estimates, we calculated neurologic hospitalization rates for a control period from January 2019 to February 2020 and a COVID-19 pandemic period from March to December 2020. We calculated incident rate ratios (IRR) for neurologic hospitalizations using negative binomial regression and compared relevant procedure rates over time. Results: Population-based neurologic hospitalization rates were 29.1 per 100,000 (95% CI 26.9-31.3) in April 2020 compared to 43.6 per 100,000 (95% CI 40.4-46.7) in January 2020. Overall, the pandemic period had 13% lower incidence of neurologic hospitalizations per month (IRR 0.87, 95% CI 0.86-0.89). The smallest decreases were in neurotrauma (IRR 0.92, 95% CI 0.89-0.95) and neuro-oncologic cases (IRR 0.93, 95% CI 0.87-0.99). Headache admissions experienced the greatest decline (IRR 0.62, 95% CI 0.58-0.66). For ischemic stroke, greater rates of endovascular thrombectomy (5.6% vs 5.0%; P < .001) were observed in the pandemic. Among all neurologic disease, greater rates of gastrostomy (4.0% vs 3.5%; P < .001), intubation/mechanical ventilation (14.3% vs 12.9%, P < .001), and tracheostomy (1.4 vs 1.2%; P < .001) were observed during the pandemic. Conclusions: During the first months of the COVID-19 pandemic there were fewer hospitalizations to varying degrees for all neurologic diagnoses. Rates of procedures indicating severe disease increased. Further study is needed to determine the impact on triage, patient outcomes, and cost consequences.

Accurately summarizing an outbreak using epidemiological models takes time

Recent outbreaks of Mpox and Ebola, and worrying waves of COVID-19, influenza and respiratory syncytial virus, have all led to a sharp increase in the use of epidemiological models to estimate key epidemiological parameters. The feasibility of this estimation task is known as the practical identifiability (PI) problem. Here, we investigate the PI of eight commonly reported statistics of the classic susceptible-infectious-recovered model using a new measure that shows how much a researcher can expect to learn in a model-based Bayesian analysis of prevalence data. Our findings show that the basic reproductive number and final outbreak size are often poorly identified, with learning exceeding that of individual model parameters only in the early stages of an outbreak. The peak intensity, peak timing and initial growth rate are better identified, being in expectation over 20 times more probable having seen the data by the time the underlying outbreak peaks. We then test PI for a variety of true parameter combinations and find that PI is especially problematic in slow-growing or less-severe outbreaks. These results add to the growing body of literature questioning the reliability of inferences from epidemiological models when limited data are available.

The dynamic relationship between COVID-19 cases and SARS-CoV-2 wastewater concentrations across time and space: Considerations for model training data sets

During the COVID-19 pandemic, wastewater-based surveillance has been used alongside diagnostic testing to monitor infection rates. With the decline in cases reported to public health departments due to at-home testing, wastewater data may serve as the primary input for epidemiological models, but training these models is not straightforward. We explored factors affecting noise and bias in the ratio between wastewater and case data collected in 26 sewersheds in California from October 2020 to March 2022. The strength of the relationship between wastewater and case data appeared dependent on sampling frequency and population size, but was not increased by wastewater normalization to flow rate or case count normalization to testing rates. Additionally, the lead and lag times between wastewater and case data varied over time and space, and the ratio of log-transformed individual cases to wastewater concentrations changed over time. This ratio decreased between the Epsilon/Alpha and Delta variant surges of COVID-19 and increased during the Omicron BA.1 variant surge, and was also related to the diagnostic testing rate. Based on this analysis, we present a framework of scenarios describing the dynamics of the case to wastewater ratio to aid in data handling decisions for ongoing modeling efforts.

Dynamic parameterization of a modified SEIRD model to analyze and forecast the dynamics of COVID-19 outbreaks in the United States

The rapid spread of the numerous outbreaks of the coronavirus disease 2019 (COVID-19) pandemic has fueled interest in mathematical models designed to understand and predict infectious disease spread, with the ultimate goal of contributing to the decision making of public health authorities. Here, we propose a computational pipeline that dynamically parameterizes a modified SEIRD (susceptible-exposed-infected-recovered-deceased) model using standard daily series of COVID-19 cases and deaths, along with isolated estimates of population-level seroprevalence. We test our pipeline in five heavily impacted states of the US (New York, California, Florida, Illinois, and Texas) between March and August 2020, considering two scenarios with different calibration time horizons to assess the update in model performance as new epidemiologic data become available. Our results show a median normalized root mean squared error (NRMSE) of 2.38% and 4.28% in calibrating cumulative cases and deaths in the first scenario, and 2.41% and 2.30% when new data are assimilated in the second scenario, respectively. Then, 2-week (4-week) forecasts of the calibrated model resulted in median NRMSE of cumulative cases and deaths of 5.85% and 4.68% (8.60% and 17.94%) in the first scenario, and 1.86% and 1.93% (2.21% and 1.45%) in the second. Additionally, we show that our method provides significantly more accurate predictions of cases and deaths than a constant parameterization in the second scenario (p < 0.05). Thus, we posit that our methodology is a promising approach to analyze the dynamics of infectious disease outbreaks, and that our forecasts could contribute to designing effective pandemic-arresting public health policies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00366-023-01816-9.

Accounting for assay performance when estimating the temporal dynamics in SARS-CoV-2 seroprevalence in the U.S

Reconstructing the incidence of SARS-CoV-2 infection is central to understanding the state of the pandemic. Seroprevalence studies are often used to assess cumulative infections as they can identify asymptomatic infection. Since July 2020, commercial laboratories have conducted nationwide serosurveys for the U.S. CDC. They employed three assays, with different sensitivities and specificities, potentially introducing biases in seroprevalence estimates. Using models, we show that accounting for assays explains some of the observed state-to-state variation in seroprevalence, and when integrating case and death surveillance data, we show that when using the Abbott assay, estimates of proportions infected can differ substantially from seroprevalence estimates. We also found that states with higher proportions infected (before or after vaccination) had lower vaccination coverages, a pattern corroborated using a separate dataset. Finally, to understand vaccination rates relative to the increase in cases, we estimated the proportions of the population that received a vaccine prior to infection.

Vaccine-induced or hybrid immunity and COVID-19-associated mortality during the Omicron wave

BACKGROUND

It is not yet entirely clear to what extent vaccine-induced or hybrid immunity protects individuals in Germany from death during the omicron wave of the COVID-19 pandemic.

METHODS

In this retrospective study, we evaluated 470 159 cases over age 59 in the German federal state of Bavaria who tested positive for SARS-CoV-2 between 1 January and 30 June 2022. Cox models were used to estimate adjusted hazard ratios (aHR) for dying within 60 days of the infection, depending on sex, age, time point of infection, and a range of immunity levels.

RESULTS

Over the period of observation, 3836 COVID-19-associated deaths were registered (case fatality rate 0.82 %). Risk of death was significantly lower in cases with a higher immunity level than in unvaccinated cases (aHR for a full primary immunity level, if reached less than six months before the time of the infection: 0.30, 95 %-confidence interval [0.23; 0.39]; if reached more than six months before: aHR 0.46 [0.35; 0.60]). A boosted immunity level lowered risk of death even further (if reached less than three months before the infection: aHR 0.17 [0.15; 0.20]; if reached more than three months before: aHR 0.25 [0.21; 0.29]).

CONCLUSION

Among elderly persons in Bavaria, a higher immunity level was associated with a substantial degree of protection against death during the Omicron wave; the strength of protection may have diminished somewhat over time.

Estimated SARS-CoV-2 antibody seroprevalence trends and relationship to reported case prevalence from a repeated, cross-sectional study in the 50 states and the District of Columbia, United States-October 25, 2020-February 26, 2022

Background

Sero-surveillance of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) can reveal trends and differences in subgroups and capture undetected or unreported infections that are not included in case-based surveillance systems.

Methods

Cross-sectional, convenience samples of remnant sera from clinical laboratories from 51 U.S. jurisdictions were assayed for infection-induced SARS-CoV-2 antibodies biweekly from October 25, 2020, to July 11, 2021, and monthly from September 6, 2021, to February 26, 2022. Test results were analyzed for trends in infection-induced, nucleocapsid-protein seroprevalence using mixed effects models that adjusted for demographic variables and assay type.

Findings

Analyses of 1,469,792 serum specimens revealed U.S. infection-induced SARS-CoV-2 seroprevalence increased from 8.0% (95% confidence interval (CI): 7.9%-8.1%) in November 2020 to 58.2% (CI: 57.4%-58.9%) in February 2022. The U.S. ratio of the change in estimated seroprevalence to the change in reported case prevalence was 2.8 (CI: 2.8-2.9) during winter 2020-2021, 2.3 (CI: 2.0-2.5) during summer 2021, and 3.1 (CI: 3.0-3.3) during winter 2021-2022. Change in seroprevalence to change in case prevalence ratios ranged from 2.6 (CI: 2.3-2.8) to 3.5 (CI: 3.3-3.7) by region in winter 2021-2022.

Interpretation

Ratios of the change in seroprevalence to the change in case prevalence suggest a high proportion of infections were not detected by case-based surveillance during periods of increased transmission. The largest increases in the seroprevalence to case prevalence ratios coincided with the spread of the B.1.1.529 (Omicron) variant and with increased accessibility of home testing. Ratios varied by region and season with the highest ratios in the midwestern and southern United States during winter 2021-2022. Our results demonstrate that reported case counts did not fully capture differing underlying infection rates and demonstrate the value of sero-surveillance in understanding the full burden of infection. Levels of infection-induced antibody seroprevalence, particularly spikes during periods of increased transmission, are important to contextualize vaccine effectiveness data as the susceptibility to infection of the U.S. population changes.

Funding

This work was supported by the Centers for Disease Control and Prevention, Atlanta, Georgia.

Regional and temporal differences in the relation between SARS-CoV-2 biomarkers in wastewater and estimated infection prevalence - Insights from long-term surveillance

Wastewater-based epidemiology provides a conceptual framework for the evaluation of the prevalence of public health related biomarkers. In the context of the Coronavirus disease-2019, wastewater monitoring emerged as a complementary tool for epidemic management. In this study, we evaluated data from six wastewater treatment plants in the region of Saxony, Germany. The study period lasted from February to December 2021 and covered the third and fourth regional epidemic waves. We collected 1065 daily composite samples and analyzed SARS-CoV-2 RNA concentrations using reverse transcription-quantitative polymerase chain reaction (RT-qPCR). Regression models quantify the relation between RNA concentrations and disease prevalence. We demonstrated that the relation is site and time specific. Median loads per diagnosed case differed by a factor of 3-4 among sites during both waves and were on average 45 % higher during the third wave. In most cases, log-log-transformed data achieved better regression performance than non-transformed data and local calibration outperformed global models for all sites. The inclusion of lag/lead time, discharge and detection probability improved model performance in all cases significantly, but the importance of these components was also site and time specific. In all cases, models with lag/lead time and log-log-transformed data obtained satisfactory goodness-of-fit with adjusted coefficients of determination higher than 0.5. Back-estimation of testing efficiency from wastewater data confirmed state-wide prevalence estimation from individual testing statistics, but revealed pronounced differences throughout the epidemic waves and among the different sites.

Modeling the positive testing rate of COVID-19 in South Africa using a semi-parametric smoother for binomial data

Identification and isolation of COVID-19 infected persons plays a significant role in the control of COVID-19 pandemic. A country's COVID-19 positive testing rate is useful in understanding and monitoring the disease transmission and spread for the planning of intervention policy. Using publicly available data collected between March 5th, 2020 and May 31st, 2021, we proposed to estimate both the positive testing rate and its daily rate of change in South Africa with a flexible semi-parametric smoothing model for discrete data. There was a gradual increase in the positive testing rate up to a first peak rate in July, 2020, then a decrease before another peak around mid-December 2020 to mid-January 2021. The proposed semi-parametric smoothing model provides a data driven estimates for both the positive testing rate and its change. We provide an online R dashboard that can be used to estimate the positive rate in any country of interest based on publicly available data. We believe this is a useful tool for both researchers and policymakers for planning intervention and understanding the COVID-19 spread.

Calibrating COVID-19 community transmission risk levels to reflect infection prevalence

Many organizations, including the US Centers for Disease Control and Prevention, have developed risk indexes to help determine community transmission levels for the ongoing COVID-19 pandemic. These risk indexes are largely based on newly reported cases and percentage of positive SARS-CoV-2 diagnostic nucleic acid amplification tests, which are well-established as biased estimates of COVID-19 transmission. However, transmission risk indexes should accurately and precisely communicate community risks to decision-makers and the public. Therefore, transmission risk indexes would ideally quantify actual, and not just reported, levels of disease prevalence or incidence. Here, we develop a robust data-driven framework for determining and communicating community transmission risk levels using reported cases and test positivity. We use this framework to evaluate the previous CDC community risk level metrics that were proposed as guidelines for determining COVID-19 transmission risk at community level in the US. Using two recently developed data-driven models for COVID-19 transmission in the US to compute community-level prevalence, we show that there is substantial overlap of prevalence between the different community risk levels from the previous CDC guidelines. Using our proposed framework, we redefined the risk levels and their threshold values. We show that these threshold values would have substantially reduced the overlaps of underlying community prevalence between counties/states in different community risk levels between 3/19/2020-9/9/2021. Our study demonstrates how the previous CDC community risk level indexes could have been calibrated to infection prevalence to improve their power to accurately determine levels of COVID-19 transmission in local communities across the US. This method can be used to inform the design of future COVID-19 transmission risk indexes.

Assessment of a mass balance equation for estimating community-level prevalence of COVID-19 using wastewater-based epidemiology in a mid-sized city

Wastewater-based epidemiology (WBE) has emerged as a valuable epidemiologic tool to detect the presence of pathogens and track disease trends within a community. WBE overcomes some limitations of traditional clinical disease surveillance as it uses pooled samples from the entire community, irrespective of health-seeking behaviors and symptomatic status of infected individuals. WBE has the potential to estimate the number of infections within a community by using a mass balance equation, however, it has yet to be assessed for accuracy. We hypothesized that the mass balance equation-based approach using measured SARS-CoV-2 wastewater concentrations can generate accurate prevalence estimates of COVID-19 within a community. This study encompassed wastewater sampling over a 53-week period during the COVID-19 pandemic in Gainesville, Florida, to assess the ability of the mass balance equation to generate accurate COVID-19 prevalence estimates. The SARS-CoV-2 wastewater concentration showed a significant linear association (Parameter estimate = 39.43, P value < 0.0001) with clinically reported COVID-19 cases. Overall, the mass balance equation produced accurate COVID-19 prevalence estimates with a median absolute error of 1.28%, as compared to the clinical reference group. Therefore, the mass balance equation applied to WBE is an effective tool for generating accurate community-level prevalence estimates of COVID-19 to improve community surveillance.

A Statistical Model of COVID-19 Infection Incidence in the Southern Indian State of Tamil Nadu

In this manuscript, we present an analysis of COVID-19 infection incidence in the Indian state of Tamil Nadu. We used seroprevalence survey data along with COVID-19 fatality reports from a six-month period (1 June 2020 to 30 November 2020) to estimate age- and sex-specific COVID-19 infection fatality rates (IFR) for Tamil Nadu. We used these IFRs to estimate new infections occurring daily using the daily COVID-19 fatality reports published by the Government of Tamil Nadu. We found that these infection incidence estimates for the second COVID wave in Tamil Nadu were broadly consistent with the infection estimates from seroprevalence surveys. Further, we propose a composite statistical model that pairs a k-nearest neighbours model with a power-law characterisation for "out-of-range" extrapolation to estimate the COVID-19 infection incidence based on observed cases and test positivity ratio. We found that this model matched closely with the IFR-based infection incidence estimates for the first two COVID-19 waves for both Tamil Nadu as well as the neighbouring state of Karnataka. Finally, we used this statistical model to estimate the infection incidence during the recent "Omicron wave" in Tamil Nadu and Karnataka.

Mobility patterns and COVID growth: Moderating role of country culture

The COVID-19 pandemic has resulted in countries reacting differently to an ongoing crisis situation. Latent to this reaction mechanism is the inherent cultural characteristics of each society resulting in differential responses to epidemic spread. Epidemiological studies have confirmed the positive effect of population mobility on the growth of infection. However, the effect of culture on indigenous mobility patterns during pandemics needs further investigation. This study aims to bridge this gap by exploring the moderating role of country culture on the relationship between population mobility and growth of CoVID-19. Hofstede's cultural factors; power distance, individualism/collectivism, masculinity/femininity, uncertainty avoidance, long-term and short-term orientation are hypothesised to moderate the effect of mobility on the reproduction number (R) of COVID-19. Panel regression model, using mobility data and number of confirmed cases across 95 countries for a period of 170 days has been preferred to test the hypotheses. The results are further substantiated using slope analysis and Johnson-Neyman technique. The findings suggest that as power distance, individualism and long-term orientation scores increase, the impact of mobility on epidemic growth decreases. However, masculinity scores in a society have an opposite moderating impact on epidemic growth rate. These Hofstede factors act as quasi moderators affecting mobility and epidemic growth. Similar conclusions could be not be confirmed for uncertainty avoidance. Cross-cultural impact, as elucidated by this study, forms a crucial element in policy formulation on epidemic control by indigenous Governing bodies.

Social capital's impact on COVID-19 outcomes at local levels

Over the past thirty years, disaster scholars have highlighted that communities with stronger social infrastructure-including social ties that enable trust, mutual aid, and collective action-tend to respond to and recover better from crises. However, comprehensive measurements of social capital across communities have been rare. This study adapts Kyne and Aldrich's (Risk Hazards Crisis Public Policy 11, 61-86, 2020) county-level social capital index to the census-tract level, generating social capital indices from 2011 to 2018 at the census-tract, zipcode, and county subdivision levels. To demonstrate their usefulness to disaster planners, public health experts, and local officials, we paired these with the CDC's Social Vulnerability Index to predict the incidence of COVID-19 in case studies in Massachusetts, Wisconsin, Illinois, and New York City. We found that social capital predicted 41-49% of the variation in COVID-19 outbreaks, and up to 90% with controls in specific cases, highlighting its power as diagnostic and predictive tools for combating the spread of COVID.

Early warning indicators of COVID-19 burden for a prosilient European pandemic response

BACKGROUND

The European Union has been criticized for responding to the COVID-19 pandemic in a reactive, rather than prosilient manner. For the EU bloc to be prosilient, it needs to have the right early warning indicators to allow short-term healthcare system preparedness and agile planning of the public health response.

METHOD

The association of COVID-19 disease burden, as measured by mortality (COVID-19 and all-cause), hospital and ICU occupancy, with incidence rate (IR), total positivity rate (TPR) and adjusted TPR as proposed by Vong and Kakkar, was investigated using Poisson regression analysis. This was carried out using both real-time data and time lags of up to 8 weeks to identify potential for early warning of spikes in disease burden. ECDC weekly figures for these indicators were used, and the analysis was repeated for the subset of data after Week 42  of 2020, when the EU Council introduced minimum COVID-19 testing rates.

RESULTS

TPR and IR were noted to be the most predictive of COVID-19 disease burden whilst adjusted TPR applied on weekly data was not associated. TPR behaved better at predicting all-cause mortality in both analyses. The TPR and IR were both best associated with hospital and ICU occupancy and COVID-19 mortality with a short time lag (2-3 weeks in the case of TPR with hospital occupancy and COVID-19 mortality).

CONCLUSIONS

Monitoring TPR can provide a 2-3-week warning of a spike in hospital occupancy and COVID-19 mortality. This time, if well utilized, could help health systems save countless lives by mobilising resources.

Forecasting the COVID-19 Pandemic: Lessons learned and future directions

I.The Coronavirus Disease 2019 (COVID-19) has demonstrated that accurate forecasts of infection and mortality rates are essential for informing healthcare resource allocation, designing countermeasures, implementing public health policies, and increasing public awareness. However, there exist a multitude of modeling methodologies, and their relative performances in accurately forecasting pandemic dynamics are not currently comprehensively understood. In this paper, we introduce the non-mechanistic MIT-LCP forecasting model, and assess and compare its performance to various mechanistic and non-mechanistic models that have been proposed for forecasting COVID-19 dynamics. We performed a comprehensive experimental evaluation which covered the time period of November 2020 to April 2021, in order to determine the relative performances of MIT-LCP and seven other forecasting models from the United States' Centers for Disease Control and Prevention (CDC) Forecast Hub. Our results show that there exist forecasting scenarios well-suited to both mechanistic and non-mechanistic models, with mechanistic models being particularly performant for forecasts that are further in the future when recent data may not be as informative, and non-mechanistic models being more effective with shorter prediction horizons when recent representative data is available. Improving our understanding of which forecasting approaches are more reliable, and in which forecasting scenarios, can assist effective pandemic preparation and management.