Scoping review and meta-analysis of COVID-19 epidemiological parameters for modeling from early Asian studies

21 Retrospective epidemiological models are powerful tools to understand its transmission 22 dynamics and to assess the efficacy of different control measures. This study 23 summarises key epidemiological parameters of COVID-19 for retrospective 24 mathematical and clinical modeling. A review of scientific papers and preprints 25 published in English between 1 January and 15 April 2020 in PubMed, MedRxiv and 26 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. 2 BioRxiv was performed to obtain epidemiological parameters of the initial stage of 27 COVID-19 pandemic in Asia. After excluding articles with unacceptable risks of bias 28 and those that remained as preprints as of 15 November 2021, meta-analyses were 29 performed to derive summary effect estimates from the data collected using the 30 statistical software R. Out of 4,893 articles identified, 88 provided data for 22 31 parameters for the overall population and 7 specifically for children. Meta-analyses 32 were conducted considering time period as a categorical moderator when it was 33 statistically significant. The results obtained are essential for building more reliable 34 models to help clinicians and policymakers improve their knowledge on COVID-19 and 35 apply it in future decisions. 36 37

38 Without any doubt the COVID-19, caused by the novel coronavirus SARS-CoV-2, is 39 the most significant global public health threat in recent decades. It was declared a 40 “Public Health Emergency of International Concern” and “pandemic” by the World 41 Health Organization (WHO) in 30 January and 11 March 2020, respectively. Since it 42 was first detected in December 2019 in Wuhan, China, it has spread over 215 countries 43 and territories, with almost 600 million confirmed cases and more than 6,4 million 44 deaths worldwide as of 21 August 2022 1, although these figures are likely 45 underestimated 2–4. 46 47 Currently, preventive therapeutic strategies are being implemented worldwide, to a 48 greater or lesser extent depending on the country 5. Although vaccines are effective in 49 reducing infection and contributing to community protection by reducing the likelihood 50 of virus transmission, their effectiveness varies among the different variants of SARS-51 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 3 COV-2 that have emerged 6,7. In this context, where the appearance of the Omicron 52 variant has once again placed restrictions and non-pharmaceutical actions on the agenda 53 of many governments 8, retrospective epidemiological modeling may allow us to gain 54 understanding on the chronology of the epidemic progression, the effectiveness of 55 different control measures, etc. 4,9,10 which can be very useful in trying to contain 56 ongoing and future outbreaks. 57 58 Since the beginning of the pandemic, thousands of preprints and papers have been 59 published about COVID-19. Given the high volume of existing literature and the 60 heterogeneity of the studies on COVID-19, it has been very difficult for researchers to 61 agree on the most accurate parameters. Additionally, most studies have examined only 62 few variables, making data mining particularly onerous and tedious. A broader 63 perspective on generalised values for the key parameters would be extremely useful for 64 improving retrospective epidemiological models and the knowledge of past decisions 65 and events, which could help future decisions. A meta-analysis is the best tool to 66 systematically assess the results of previous research to derive conclusions. Thus, the 67 aim of this study was to identify and summarise key epidemiological parameters of the 68 initial stages of COVID-19 from the vast existing literature through meta-analysis in 69 order to make that scattered information useful for retrospective mathematical modeling 70 and also, when data are available, for comparative analysis with other pandemic stages, 71 where new variants predominate over the initial one. 72 73

74 During the study period, 6,969 scientific articles related to COVID-19 were published 75 or posted in the consulted libraries. Those were filtered using the keywords and terms 76 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 4 described in the method section and the titles and abstracts screened, leaving the full 77 text of 361 papers to review. Of those, 131 contained extractable data (see 78 Supplementary Table S1), but only 88 were considered in our meta-analysis for one or 79 more parameters (see Figure 1, the PRISMA flow diagram, for details in exclusion 80 criteria). Unacceptable risk of bias was detected in 5 studies, mainly due to unclear 81 selection approach, and insufficient follow-up period (see the complete evaluation in the 82 Supplementary Table S2). As of 15 November 2021, 13 papers remained unpublished, 83 and they were also excluded. 84 85 Pooled means or percentages and their associated 95% CI were determined for 22 86 epidemiological parameters for the general population (disaggregated into survivor/non-87 survivor when possible) and 7 for children. Tables 1 and 2 show the results provided by 88 the meta-analysis for those parameters for which, having performed the meta-regression 89 using the period of admission of the patients in the hospital as a categorical moderator, 90 the results were not significant and those that did, respectively. 91 For the general population parameters, the number of studies considered ranged from 3 92 to 27 (median 8.5, Q1 6 and Q3 15.75) and the number of patients involved ranged from 93 187 to 5771 (mean 1692), whereas for children studies ranged from 3 to 4 (median 3, 94 Q1 3 and Q3 4) and patients from 25 to 945 (mean 341). The estimated I2, τ 2, and Q 95 (values summarised in the Supplementary Meta-analysis Report for each parameter) 96 indicated that high heterogeneity was present for most parameters (except for onset of 97 symptoms to hospital ICU admission) which reduced their usefulness. The last two 98 columns of both tables also show the results of the parameters for the general 99 population that allowed removal of outliers without compromising their utility, i.e. 100 yielding a lower heterogeneity (values summarised in the Supplementary Meta-analysis 101 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 5 Report) and maintaining a minimum of 3 included studies. Given these constraints, the 102 number of studies considered ranges from 3 to 12 (median 7, Q1 5 and Q3 9.75), with 103 the parameters with 9 or more studies included: onset of symptoms to hospital admission 104 (all), hospital stay length (survivor, period B), asymptomatic patients, percentage of 105 ICU admissions, percentage of deaths, percentage of discharged and percentage in 106 hospital at the end of the study. Parameters with fewer than 5 studies comprised onset of 107 symptoms to hospital admission (survivor), hospital stay length (all patients, both 108 periods), percentage of deaths from ICU and percentage of discharged from ICU. For 109 children, all parameters were initially estimated from fewer than 5 studies and neither 110 the meta-regression with period as moderator nor outlier diagnosis were carried out. 111 112 Forest plots for each parameter (considering all studies or group by period if significant 113 differences between the two subsets were detected) are provided in the Supplementary 114

(see Supplementary Figures S1-31). In all cases, a summary of the results is 115 provided when all the studies are included in the meta-analysis, and with 116 outliers/influencers removed if any. 117 118

119 We present a review and meta-analysis of essential epidemiological parameters (22 for 120 the general population, 7 for children) of COVID-19 based on Asian studies during the 121 initial stage of the pandemic. Our main goal was to provide a summary of key 122 parameters useful for modeling, and therefore an exhaustive clinical explanation of our 123

is beyond the scope of this study. Discussion of the most relevant findings 124 follows. 125 126 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 6 In order to see the influence of the changes in the way of dealing with the virus resulting 127 from the COVID-19 Forum convened by WHO on the key epidemiological parameters 128 we register the time period of the different studies. We classified and, therefore 129 compared, studies with cases occurring prior February 12, 2020 (period A) and studies 130 with patients enrolled before and after (period B). Significant results in the meta-131 regression considering period as a categorical moderator were obtained for hospital stay 132 length (all patients), hospital stay length (survivor), ICU stay length (non-survivor) and 133 onset of symptoms to hospital admission (survivor)(table 2). Note that since period A is 134 included in period B, the differences are more remarkable. The length of hospital stay 135 (both for all patients and for the survivors) were longer in period B than in A, probably 136 because, but not limited to, the shortage of medical resources during period A. On the 137 other hand, results for ICU stay length (non-survivor) and onset of symptoms to hospital 138 admission (survivor), are not conclusive because only one and two studies are available 139 in period B and period A, respectively. 140 141 From the parameters analysed in both groups (overall population versus children), 142

point that children had longer incubation periods, higher percentages of 143 asymptomatic individuals, fewer ICU admissions and shorter hospital stays. This 144 suggests, as found previously 11, less effect of the disease on children than in the overall 145 population. 146 147 For the overall population, the mean serial interval obtained (4.97 days, 95% CI: 3.88-148 6.36) when all studies were considered was shorter than the incubation period (5.61 149 days, 95% CI: 4.74-6.63), supporting the hypothesis of substantial pre-symptomatic 150 transmission of the virus 12. Our own results showed that 29% of disease transmission 151 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 7 would be pre-symptomatic. Nevertheless, the results of meta-analysis are sensitive to 152 outliers, and when outliers are removed for this parameter, this theory is not supported. 153 This is due to the fact that Du et al. 13 was identified as an outlier. Outliers always must 154 be taken with caution, for example, Du et al. conducted a study with a large population 155 (468 out of a total of 662 patients from all the 7 studies) and reported a mean serial 156 interval (3.96 days) lower than the other studies. Thus, when outliers were removed, the 157 number of studies was reduced to 5 with a total of 187 patients considered and the mean 158 serial interval increased to 5.80 days (95% CI: 5.69-5.92). In a meta-analysis any 159

that are sensitive to outliers must be stated with caution, while those that 160 are robust to outliers are more reliable. 161 162 Our study has some limitations. First, only papers published in English were evaluated. 163 Although many papers have been published in Chinese and a few in other languages 164 during the period analysed, papers in English provided a large enough sample. Second, 165 we performed the search strategy only with three libraries until 15 April 2020. Our 166 work might not include all the published data with the selected parameters, but given 167 the large amount of papers that was posted daily, and considering that the initial phase 168 of the pandemic lasted until April 2020 in China and other Asian countries 14–16 we took 169 the pragmatic decision to limit the search in this way. Third, data was restricted to Asian 170 countries. We detected a few useful papers from the USA, Italy and Spain, but we 171 decided not include them in the analysis in order to not introduce additional 172 heterogeneity due to cultural (illness is shaped by cultural factors governing perception, 173 labelling, explanation and valuation of the discomforting experience 17) and care 174 pathways differences between Eastern and Western countries. Finally, we presented 175 some results based on fewer than 5 studies and results that showed high heterogeneity, 176 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 8 all of which also should be interpreted with caution. Outlier analysis were not 177 appropriate for small groups of studies. We only perform a moderated analysis by time 178 period, but considering other characteristics of the studies would also help to identify 179 causes of heterogeneity. Nevertheless, several of the primary studies had incomplete or 180 ambiguous information for some characteristics that could serve as potential moderators 181 of the disease. Thus, we were unable to identify moderators without drastically reducing 182 the number of studies in the meta-analysis. Therefore, we performed our analysis 183 carefully and considering all these difficulties. 184 185 In this paper we summarised more epidemiological parameters than other available 186 multi-parameter reviews on COVID-19 18–24. However, most of them use more studies 187 for most of the parameters than us, mainly due to our language and geographic 188 restrictions, as well as the application of risk of bias analysis and preprints exclusion. 189 Since a mathematical model is as good as the data it uses 25, we believe that the 190 considerations we have applied are essential to obtain the most reliable and accurate 191

available for modeling purposes. 192 193 In summary, we reviewed and gathered the most relevant epidemiological parameters 194 obtained during early phase of COVID-19 pandemic to date in a single article. 195 Improvement of our understanding of the dynamics of this infectious disease and 196 evaluation of the effectiveness of the adopted intervention measures at initial stages of 197 pandemic are crucial to successfully cope with COVID-19 until new vaccines (or 198 boosters) or antiviral treatments are implemented. These data may be of interest both to 199 modelers, clinicians, managers, and national and regional policymakers when creating 200 and interpreting the retrospective epidemiological models. Similar studies for different 201 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 9 stages of the pandemic in several regions would be highly desirable upon data 202 availability, not only to provide a complete picture of the pandemic, but also to be able 203 to identify differences between healthcare systems and their efficiency on dealing with 204 the COVID-19 pandemic, which could also be useful for future pandemics. 205 206

207 Following the drafted protocol (freely available in the Open Science Framework 208 repository, https://osf.io/26mvb), we searched PubMed, MedRxiv and BioRxiv for 209 articles and preprints published in English between 1 January and 15 April 2020 with 210 the keywords “COVID-19”, “SARS-CoV-2”, “severe acute respiratory syndrome 211 coronavirus 2”, “2019-nCoV” or “novel coronavirus”, as well as key terms to obtain 212 data about the following parameters (see Supplementary Text S1 for our search strategy 213 and Supplementary Text S2 for definitions and reclassifications): 214 • pre-symptomatic transmission period, 215 • serial interval, 216 • incubation period, 217 • onset of symptoms/illness onset to first medical visit, 218 • from first medical visit to diagnosis 219 • from first medical visit hospital admission 220 • onset of symptoms/illness onset to diagnosis, 221 • onset of symptoms/illness onset to hospital admission, 222 • onset of symptoms/illness onset to ICU (Intensive Care Unit) admission, 223 • hospital stay length, 224 • ICU stay length, 225 • hospital admission to death, 226 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 10 • hospital admission to discharge, 227 • onset of symptoms/illness onset to death, 228 • onset of symptoms/illness onset to discharge/recovery, 229 • percentage of pre-symptomatic transmissions, 230 • percentage of asymptomatic patients, 231 • percentage of ICU admissions, 232 • percentage of deaths, 233 • percentage of deaths from ICU, 234 • percentage of discharged, 235 • percentage of discharged from ICU, 236 • percentage in hospital, 237 • percentage in hospital from ICU, 238 • percentage transferred from ICU to general hospital wards. 239 240 Preliminary screening by title and abstract identified potential relevant studies. 241 Subsequently, the full texts of those studies were evaluated and their reference lists 242 examined for additional records (Figure 1). Data related to these epidemiological 243 parameters, such as sample size, mean, standard deviation (SD), confidence intervals, 244 median, interquartile range (IQR) and the fitted distribution used in its estimation, when 245 applicable, as well as sociodemographic information (i.e. patient’s age, gender and 246 location) and the period time of the studies, were extracted in Supplementary Table S1. 247 Reviews, non-original research papers and articles not based on Asian studies were 248 excluded. In case of data overlap, the article with the largest sample size was chosen. 249 Finally, papers that remain as preprints as of 15 November 2021 were also removed 250 from the analysis (the corresponding authors were contacted by e-mail for 251 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 11 confirmation). In addition, because COVID-19 does not seem to affect children and 252 teens in the same way as adults 11, we analysed the data for pediatric patients separately. 253 Initial screening was performed by the author ESF, followed by an ultimate extraction 254 of data assessed independently by the authors ESF, DGG and MIV. When discrepancies 255 were detected, MIV and ESF, experienced researchers, made the final decision whether 256 or not to include the study. 257 258 Critical appraisal of selected studies was conducted following the assessment tool 259 developed by Murad et al. 26, which was based on four domains: selection, 260 ascertainment, causality, and reporting. The results from this tool provided a quality 261 assessment of the studies for quantitative synthesis. Articles with unacceptable risk of 262 bias (unclear or high risk of bias in ≥ 1 domain) were excluded from the analysis. Risk 263 of bias was summarised in Supplementary Table S2. 264 265 For each parameter, in order to integrate all the reviewed information from studies with 266 a low/acceptable risk of bias in one overall estimate that could be introduced in 267 mathematical models, we performed different analyses. First of all, taking into account 268 that in Asia the way to deal with the virus changed over time as they gained experience 269 and that on February 12, 2020, WHO convened a Global Research and Innovation 270 Forum on COVID-19 where 300 experts from 48 countries attended 27, we classified the 271 studies considered in this review between those which cases were reported before 272 February 12, 2020 (period A) and, given the impossibility of finding studies only with 273 cases recorded after that date, those who include cases throughout the whole study 274 period (period B). Subsequently, we performed a meta-regression for all the parameters 275 with five or more studies considering the period as a categorical moderator, allowing the 276 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 12 amount of residual heterogeneity to be different in each subset. Afterwards, we carried 277 out the meta-analysis for all included studies and for the two subsets separately when 278 they were confirmed to significantly affect the results, as following: for those 279 parameters including mean and SD, or median and IQR, we conducted a meta-analysis 280 of single means (when mean and SD were not reported, they were estimated from 281 median and IQR 28,29), and when parameters were presented as percentages, a meta-282 analysis of proportions was performed. To account for the variability between and 283 within studies, random-effects models were used. We considered model fitting by both 284 the method of moments 30 and the Restricted Maximum Likelihood Method (REML)31. 285 The natural logarithm transformation was applied to meet the normality assumption 286 underlying the meta-analysis. The null hypothesis of no variance among studies (τ 2=0) 287 was tested using the Q-statistic and the degree of heterogeneity was quantified by the I2 288 index 32. Given the high heterogeneity among studies for most parameters considered, 289 outliers and influencers diagnoses were also performed 33. Thus, we avoided biasing the 290

from the meta-analysis by a few (potentially unusual) studies. Data were 291 analysed using the statistical software R version 3.6.2 34 and the “meta” 35, “metafor” 36 292 and “dmetar” 37 packages. 293 294 The PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-295 analyses extension for Scoping Reviews) checklist was used to endorse good reporting 296 in this article (see Supplementary PRISMA ScR Checklist) 38. 297 298 Data availability 299 The data used to support the findings of this study are included in Supplementary table 300 S1. 301 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 13 302 Code availability 303 The meta-analysis has been conducted in R version 3.6.2 34. Core packages for 304 replicating the results are publicly available and include and include “meta” 35, 305 “metafor” 36 and “dmetar” 37. 306 . 307 308

309 1. WHO. Weekly epidemiological update on COVID-19 - 24 August 2022. 310 https://www.who.int/publications/m/item/weekly-epidemiological-update-on-311 covid-19---24-august-2022. (2022). 312 2. Kung, S. et al. Underestimation of COVID-19 mortality during the pandemic. 313 ERJ Open Res. 7, 00766–02020 (2021). 314 3. Modi, C., Böhm, V., Ferraro, S., Stein, G. & Seljak, U. Estimating COVID-19 315 mortality in Italy early in the COVID-19 pandemic. Nat. Commun. 12, (2021). 316 4. García-García, D. et al. Retrospective methodology to estimate daily infections 317 from deaths (REMEDID) in COVID-19: the Spain case study. Sci. Rep. 11, 318 (2021). 319 5. WHO. Dedicated COVID-19 vaccination dashboard. 320 https://www.who.int/emergencies/diseases/novel-coronavirus-2019/covid-19-321 vaccines (Accessed 23rd December 2021). (2021). 322 6. Lopez Bernal, J. et al. Effectiveness of Covid-19 Vaccines against the B.1.617.2 323 (Delta) Variant. N. Engl. J. Med. 385, 585–594 (2021). 324 7. Burki, T. Omicron variant and booster COVID-19 vaccines. Lancet Respir Med 1 325 (2021). 326 8. Wamsley, L. As omicron surges, nation announce rules on gatherings and travel 327 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 14 ahead of holidays. NPR 1 (2021). 328 9. Fochesato, A. et al. A retrospective analysis of the COVID-19 pandemic 329 evolution in Italy. Biology (Basel). 10, 311 (2021). 330 10. Marzianoa, V. et al. Retrospective analysis of the Italian exit strategy from 331 COVID-19 lockdown. Proc. Natl. Acad. Sci. U. S. A. 118, (2021). 332 11. Du, W. et al. Clinical characteristics of COVID-19 in children compared with 333 adults in Shandong Province, China. Infection Apr 16, 1–8 (2020). 334 12. He, X. et al. Temporal dynamics in viral shedding and transmissibility of 335 COVID-19. Nat. Med. 26, 672–675 (2020). 336 13. Du, Z. et al. Serial Interval of COVID-19 among Publicly Reported Confirmed 337 Cases. Emerg. Infect. Dis. 26, (2020). 338 14. Xu, J. & Cao, B. Letter from China: Managing the second-wave COVID-19 339 outbreak in Beijing. Respirology 26, 275–276 (2021). 340 15. Xu, J. W. et al. Effective actions against the second wave of COVID-19: The 341 front-line experience from China. European Review for Medical and 342 Pharmacological Sciences 24, 11995–11997 (2020). 343 16. Lee, H., Noh, E., Jeon, H. & Nam, E. W. Association between traffic inflow and 344 COVID-19 prevalence at the provincial level in South Korea. Int. J. Infect. Dis. 345 108, 435–442 (2021). 346 17. Kleinman, A., Eisenberg, L.., & G ood, B. Culture, Illness, and Care. Ann. Intern. 347 Med. 88, 251–258 (1978). 348 18. Park, M., Cook, A. R., Lim, J. T., Sun, Y. & Dickens, B. L. A Systematic Review 349 of COVID-19 Epidemiology Based on Current Evidence. J. Clin. Med. 9, 967 350 (2020). 351 19. Khalili, M. et al. Epidemiological Characteristics of COVID-19: A Systematic 352 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 15 Review and Meta-Analysis. Epidemiol. Infect. 1–39 (2020). 353 doi:10.1017/S0950268820001430 354 20. Koh, J., Shah, S. U., Chua, P. E. Y., Gui, H. & Pang, J. Epidemiological and 355 Clinical Characteristics of Cases During the Early Phase of COVID-19 356 Pandemic: A Systematic Review and Meta-Analysis. Front. Med. 7, 295 (2020). 357 21. Izadi, N. et al. The epidemiologic parameters for COVID-19: A Systematic 358 Review and Meta-Analysis. medRxiv 2020.05.02.20088385 (2020). 359 doi:10.1101/2020.05.02.20088385 360 22. Zhang, P., Wang, T. & Xie, S. X. Meta-analysis of several epidemic 361 characteristics of COVID-19. J. Data Sci. 18, 536–549 (2020). 362 23. Gallo, L. G. et al. Ten Epidemiological Parameters of COVID-19: Use of Rapid 363 Literature Review to Inform Predictive Models During the Pandemic. Front. 364 Public Heal. 8, (2020). 365 24. Allen, K., Parry, A. E. & Glass, K. Early reports of epidemiological parameters 366 of the COVID-19 pandemic. Western Pacific surveillance and response 367 journal/i1: WPSAR 12, 65–81 (2021). 368 25. Panovska-Griffiths, J. Can mathematical modelling solve the current Covid-19 369 crisis? BMC Public Health 20, (2020). 370 26. Murad, M. H., Sultan, S., Haffar, S. & Bazerbachi, F. Methodological quality and 371 synthesis of case series and case reports. Evid. Based. Med. 23, 60–63 (2018). 372 27. WHO. COVID-19 Public Health Emergency of International Concern (PHEIC). 373 Global research and innovation forum: towards a research roadmap. (2020). 374 28. Luo, D., Wan, X., Liu, J. & Tong, T. Optimally estimating the sample mean from 375 the sample size, median, mid-range, and/or mid-quartile range. Stat. Methods 376 Med. Res. 27, 1785–1805 (2018). 377 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 16 29. Wan, X., Wang, W., Liu, J. & Tong, T. Estimating the sample mean and standard 378 deviation from the sample size, median, range and/or interquartile range. BMC 379 Med. Res. Methodol. 14, (2014). 380 30. DerSimonian, R. & Laird, N. Meta-analysis in clinical trials. Control. Clin. Trials 381 7, 177–188 (1986). 382 31. Raudenbush, S. W. & Bryk, A. S. Empirical Bayes Meta-Analysis. J. Educ. Stat. 383 10, 75–98 (1985). 384 32. Higgins, J. P. T. & Thompson, S. G. Quantifying heterogeneity in a meta-385 analysis. Stat. Med. 21, 1539–1558 (2002). 386 33. Viechtbauer, W. & Cheung, M. W.-L. Outlier and influence diagnostics for meta-387 analysis. Res. Synth. Methods 1, 112–125 (2010). 388 34. R Core Team. R: A Language and Environment for Statistical Computing. R 389 Foundation for Statistical Computing: Vienna, Austria (2014). 390 35. Balduzzi, S., Rücker, G. & Schwarzer, G. How to perform a meta-analysis with 391 R: A practical tutorial. Evid. Based. Ment. Health 22, 153–160 (2019). 392 36. Viechtbauer, W. Conducting meta-analisys in R with metafor package. J. Stat. 393 Softw. 36, 1–48 (2010). 394 37. Harrer, M., Cuijpers, P., Furukawa, T. A. & Ebert, D. D. Doing Meta-Analysis in 395 R: A Hands-on Guide. Available: 396 https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R. (Accessed 397 18th May 2020). (2019). 398 38. Tricco, A. C. et al. PRISMA extension for scoping reviews (PRISMA-ScR): 399 Checklist and explanation. Ann. Intern. Med. 169, 467–473 (2018). 400 401

402 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 17 We acknowledge Parques Nacionales (Ministerio para la Transición Ecológica y Reto 403 Demográfico, Spain), and Generalitat Valenciana (Regional Government of Valencia, 404 Spain) for the support of the Montgó-Dénia Research Station. We also thank John Y. 405 Dobson and Elena Fonfría for their valuable comments during the editing process. 406 407 Author Contributions 408 CB, ZH and MN: conceptualization. ESF and MIV: Study design, data analysis and 409

interpretation. ESF, MIV and DGG: literature review and data curation. ESF and 410 MIV: writing-original draft preparation. CB: funding acquisition and project 411 administration. All authors contributed to the manuscript writing-review and editing, 412 and approved the submitted version. 413 414 Funding 415 This work was supported by the University of Alicante [grant number COVID-19 2020-416 41.30.6P.0016] and the Conselleria de Agricultura, Desarrollo Rural, Emergencia 417 Climática y Transición Ecológica de Generalitat Valenciana, Ajuntament de Dénia and 418 University of Alicante through the Montgó-Dénia-UA Research Station Agreement 419 [grant number 20202-41.30.6O.00.01]. 420 421 Competing interests 422 The authors declare no competing interests. 423 424 Figures 425 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 18 426 Fig. 1. PRISMA flow diagram. Abbreviations: SD, Standard deviation; CI, Confidence 427 Intervals; IQR, Interquartile range. 428 429 Tables 430 Table 1. Meta-analysis results for parameters where period is not a significant 431 moderator. 432 Parameter All studies included Outliers removed Mean 95% CI Studies included N (patients) I2 (%) Mean I2 (%) General Population Serial interval (days) 4.97 3.88-6.36 7 662 91 5.80 18 Incubation period (days) 5.61 4.74-6.63 16 2437 96 5.30 52 OS to hospital admission (all)(days) 6.29 5.32-7.44 27 4323 98 6.51 12 OS to hospital admission (non-survivor) (days) 10.15 9.61-10.71 7 574 56 10.18 47 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 19 OS to hospital ICU admission (all) (days) 9.74 9.20-10.31 6 187 0 ND ND Hospital stay length (non-survivor) (days) 8.63 7.23-10.31 12 785 93 8.12 0 ICU stay length (survivor)(days) 9.08† 6.47-12.76 4 212 92 NA NA OS to death (days) 17.79 16.06-19.71 10 516 86 16.66 57 OS to discharge (days) 21.76† 18.13-26.12 3 362 98 NA NA Pre-symptomatic transmission (%) 29.33 11.29-57.50 5 870 97 ND ND Asymptomatic patients (%) 8.54 4.04-17.18 14 2430 90 5.44 18 ICU admissions vs all hospital admissions (%) 13.09 9.14-18.41 19 4994 94 16.75 40 Deaths vs all hospitalized patients (%) 7.62 4.59-12.41 20 5771 94 13.23 41 Deaths from ICU (%) 35.03 19.42-54.67 5 453 79 38.30† 0 Discharged vs all hospitalized patients (%) 49.34 36.01-62.76 25 4840 97 34.93 0 Discharged from ICU (%) 44.96 25.73-65.82 6 487 85 54.61† 2 In hospital (%) 59.79 47.01-71.36 21 3902 97 63.13 21 In hospital (ICU) (%) 16.63 8.53-29.91 7 537 86 16.30 57 Children (<18years) Incubation period (days) 6.69† 5.49-8.15 4 25 0 NA NA OS to diagnosis (days) 3.27† 1.35-7.92 3 744 92 NA NA Hospital stay length (days) 12.77† 8.51-19.17 3 52 90 NA NA Asymptomatic patients (%) 16.66† 11.52-23.48 4 945 56 NA NA ICU admissions vs all hospital admissions (%) 5.06† 1.64-14.60 4 211 42 NA NA Discharged vs all hospitalized patients (%) 19.39† 3.39-62.23 3 205 86 NA NA In hospital (%) 80.27† 37.87-96.45 3 205 86 NA NA Abbreviations: CI, Confidence Intervals; ICU, Intensive Care Unit; NA, Not Applicable; ND, Not 433 Detected; OS, Onset of symptoms. 434 † Results are based on fewer than 5 studies. 435 436 Table 2. Meta-analysis results for parameters where period is a significant moderator. 437 Parameter Period All studies included Outliers removed Mean 95% CI Studies included N (patients) I2 (%) Mean I2 (%) General Population OS to hospital admission (survivor)(days) A 9.81† 8.32-11.57 2 298 91 NA NA B 6.14† 4.38-8.59 4 250 97 NA NA . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint 20 Hospital stay length (all) (days) A 12.70 11.58-13.94 7 1959 96 13.13† 0 B 17.14 13.42-21.88 6 777 97 19.70† 0 Hospital stay length (survivor) (days) A 11.06† 10.26-11.91 4 392 59 NA NA B 16.38 14.34-18.70 11 835 99 14.88 72 ICU stay length (non-survivor) (days) A 8.36† 7.19-9.71 4 270 41 NA NA B 11.80† 9.61-14.48 1 51 NA NA NA Abbreviations: CI, Confidence Intervals; ICU, Intensive Care Unit; NA, Not Applicable. Periods: A, 438 includes cases before Feb 12, 2020; B, includes cases before and after Feb 12, 2020. 439 † Results are based on fewer than 5 studies 440 441 442 443 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 25, 2022. ; https://doi.org/10.1101/2022.10.23.22281408doi: medRxiv preprint