Spatial variation in the basic reproduction number of COVID-19: A systematic review
Renate Thiede, Nada Abdelatif, Inger Fabris-Rotelli, Raeesa Manjoo-Docrat, Jennifer Holloway, Charl Janse van Rensburg, Pravesh Debba, Nontembeko Dudeni-Tlhone, Zaid Kimmie, Alize le Roux
SSpatial variation in the basic reproduction number of COVID-19: Asystematic review
RN Thiede , N Abdelatif , IN Fabris-Rotelli , R Manjoo-Docrat , J Holloway , CJanse van Rensburg , P Debba , N Dudeni-Tlhone , Z Kimmie , and A le Roux
Department of Statistics, University of Pretoria Biostatistics, South African Medical Research Council Department of Statistics and Actuarial Science, University of the Witwatersrand Council for Scientific and Industrial Research, Pretoria, South Africa Foundation for Human Rights, South Africa * Corresponding author: [email protected] a ORCiD ID: 0000-0003-0934-5374 b ORCiD ID: 0000-0002-3185-1284 c ORCiD ID: 0000-0002-2192-4873 d ORCiD ID: 0000-0003-2039-0440 e ORCiD ID: 0000-0003-0948-4796 f ORCiD ID: 0000-0002-6539-7833 g ORCiD ID: 0000-0003-4870-988X h ORCiD ID: 0000-0002-8853-3121 i ORCiD ID: 0000-0002-6392-0619 j ORCiD ID: 0000-0002-9214-5076November 2020
Funding: This work is supported by the NRF-SASA (National Research Foundation-South African StatisticalAssociation) Crisis in Statistics Grant. 1 a r X i v : . [ s t a t . O T ] D ec bstract OBJECTIVES: Estimates of the basic reproduction number ( R ) of COVID-19 vary across countries.This paper aims to characterise the spatial variability in R across the first six months of the global COVID-19 outbreak, and to explore social factors that impact R estimates at national and regional level.METHODS: We searched PubMed, LitCOVID and the WHO COVID-19 database from January to June2020. Peer-reviewed English-language papers were included that provided R estimates. For each study, thevalue of the estimate, country under study and publication month were extracted. The median R valuewas calculated per country, and the median and variance were calculated per region. For each country, theHuman Development Index (HDI), Sustainable Mobility Index (SMI), median age, population density anddevelopment status were obtained from external sources.RESULTS: A total of 81 studies were included in the analysis. These studies provided at least oneestimate of R , along with sufficient methodology to explain how the value was calculated. Values of R ranged between 0.48 and 14.8, and between 0.48 and 6.7 when excluding outliers.CONCLUSIONS: This systematic review provides a comprehensive overview of the estimates of the basicreproduction number of COVID-19 globally and highlights the spatial heterogeneity in R . The value wasgenerally higher in more developed countries, and countries with an older population or more sustainablemobility. Countries with higher population density had lower R estimates. For most regions, variability in R spiked initially before reducing and stabilising as more estimates became available. Keywords:
Coronavirus, Basic Reproduction Number, Data Visualization, Uncertainty, Spatial Analy-sis
In 2020, the world observed the spread of the COVID-19 pandemic, which was of an unprecedented nature. TheWorld Health Organisation released a statement on 9 January 2020 reporting that a novel respiratory diseasehad been identified by Chinese authorities, originating in Wuhan. This disease was identified as 2019-nCovin early January, and spread rapidly outside of Wuhan and into the rest of the world. By 17 January, Chinarecorded 62 cases, and 3 travelers had exported the disease, with two diagnosed in Thailand and one in Japan[1]. At the time of writing, 220 countries have been affected by the disease, with the total death toll estimatedat over 1 500 000 . This highlights the importance of understanding all aspects of the disease in order to combatit effectively.The basic reproduction number ( R ) is of particular importance in epidemiological modelling [2]. This isthe average number of susceptible individuals that are infected by a single infected individual [3]. It is bestmeasured in the early phase of the outbreak, before control measures have had time to take effect, and whenmost of the population is susceptible. Estimation of R has proved complex [4]. R estimation depends on theextent to which the disease characteristics are understood, which is challenging in the case of a novel disease.Assumptions must also be made about the ways in which people come into contact with one another. Theseassumptions and uncertainties limit the precision and accuracy of models used to determine R . Previousreviews have determined that values of the estimate range between 0.6 and 14.8 [5, 6]. The reviews of [7] and[8] determined R to be between 1 and 7, while [9, 10, 11] found R estimates to be between 2 and 4. Thereview of [12] states that estimates appear to be stabilising between 2 and 3. It should be noted that these R estimates were all determined at the beginning of the pandemic, and do not reflect changes in the disease overtime.Reasons for the great variability in R estimates remain an open question. The review of [8] indicated thatmathematical models and stochastic models gave different values for R , however, [10] stated that heterogeneityin the estimates was not influenced by the estimation methods chosen. The data used to estimate R introducesanother source of variability. This is supported by the review of [11], which states that R varies from place toplace, but does not describe the variation. While comprehensive, none of the above reviews investigated howthe estimates of R differ across countries or regions.This systematic review aims to provide a comprehensive overview of the estimates of the basic reproductionnumber of COVID-19 globally and to highlight the spatial heterogeneity in the estimates over the first half of2020. As R is measured at the beginning of the pandemic, changing R values do not reflect the progress ofthe disease, but instead illustrate a growing understanding of the disease characteristics over time. We furtherinvestigate possible factors that influence this spatial variability, namely their economic development statusas classified by the World Bank, their human development index (HDI) as an indicator of social development,social mobility index (SMI) as a measure of mobility, median age and population density. Finally, an illustrativeexample considering a model specific to South Africa is given to guide the discussion on spatial heterogeneityin R . Materials and Methods
Inclusion criteria
Only English peer-reviewed research papers were included. These had to provide an estimate of R , and hadto provide enough information in the methodology to determine how R was calculated. Reviews, conferencepapers and other sources were not considered directly in our review, but were consulted as secondary sources. Search strategy
The search included papers published between 31 December 2019 and 30 June 2020, allowing for the initial stagesof the disease spreading internationally. The WHO COVID-19 database, PubMed and LitCovid were searchedfor English-language papers including the terms “Coronavirus”, “COVID-19”, “SARS-CoV-2”, or “2019-nCoV”in combination with the phrases “reproduction number” or “reproductive number”. Two reviewers (RT and NA)independently evaluated the eligibility of the studies obtained from the literature searches. All articles yieldedby the database searches were screened by their titles and abstracts to obtain studies that met the inclusioncriteria. After this, full-text screening was performed. In cases of uncertainties, agreement was reached througha third reviewer (IFR).
Data extraction
Data extracted included the country for which R was estimated, whether actual data or simulated data wasused, the value of R as well as its 95% confidence interval and whether a spatial component was consideredin the estimation of R . Data was extracted independently by two reviewers (RT and NA) and in the case ofno consensus, IFR was consulted. Risk of bias was not considered in this study because no meta-analysis wasdone. Ethics statement
This study was a systematic review which did not employ any human or animal subjects.
Description of included articles
Figure 1 shows the review process. A total of 592 documents were found by searching PubMed, LitCOVID andthe WHO COVID-19 database up to 30 June. In total, 425 were screened after removing duplicates, pre-printsand non-English articles. Of these, 81 were included in the study (Supplementary File 1). These containedsufficient methodology to calculate the basic reproduction number, as well as a numerical estimate for R . Foreach paper, the estimated values of R were included, along with the country (countries) under study, theareas within countries where applicable, and the publication month. The studies covered 65 countries across 5continents, as shown in Figure 2.Estimates for the basic reproduction number were given both at a national level and at local levels. Theestimates for local areas were calculated independently and did not consider spatial auto-correlation betweenthe regions. China had the most overall estimates at 100, followed by Italy with 91 and the USA with 61,thanks largely to the paper of [14], which provided R estimates for each state of the USA. The paper of [15]contributed to the large number of R estimates for India, namely 27 estimates. Spain recorded the fifth largestnumber of estimates, namely 12. These estimates came from a greater variety of papers. The countries whichrecorded the highest number of national R estimates were Italy with 17 estimates, China with 15, Japan with11 and Spain with 10 national R estimates. A total of 36 countries only recorded one national R estimateeach.Since the focus herein is on spatial variation in R , we do not consider the estimation techniques. However,we note the most popular approaches observed in the study. SEIR-like compartmental models were used intwo thirds (54) of the included papers, 15 papers employed exponential or logistic growth models, while themethods of [16] and [17] were used in 9 and 6 papers respectively. We refer to the review of [8] for a thoroughinvestigation of the effects of estimation techniques on the R estimate.3 .. u !E .. C ., � ... C 'i: ., � u "' .,, ., .,, ::, u .E Records identified through database Additional records identified searching through other sources (n = = O) Records after duplicates removed (n = • Records screened Records excluded (n = = ,, reasons Full-text articles assessed (n =
97) for eligibility - Reasons for exclusion: (n = - Ro value not given/not estimated SARS-Co V parameters not COVID-19 - Simulated data used rather than actual data Confidence intervals around
Ro not given Ro values given from other sources
Studies included in qualitative synthesis (n = Figure 1: Article screening and data extraction process for the systematic review [13]Figure 2: Countries included in the study.4igure 3: Distribution of R values by publication month. Variability in R observed in the reviews Monthly distribution of R value Figure 3 shows the shift in R estimates, both national and local, by publication month. For each month, themedian was calculated based on all the papers that had been published up to and including that month. Thisdoes not indicate a change in the R value over time, as most of the R estimates were calculated based ondata from the beginning of the pandemic. Rather, this reflects a change in the understanding of the diseaseover time. This may be due to various factors, such as more sophisticated estimation techniques or improvingepidemiological knowledge. The estimates published in January and February were all obtained for China, andfollow a similar pattern, being between 2 and 4. The estimates in January were all local, in particular calculatedfor Wuhan. For February, it should be noted that an estimate of size 14.8 for the Diamond Princess cruise shipis not visualised here, as it was removed from the analysis as an outlier. In March, estimates from the rest of theworld became available, including Italy and Japan. A large amount of variability was experienced for this month.In April, 20 countries had estimates. Both local and national estimates evidenced a bi-modal distribution, withthe greater mass concentrated between 1 and 3. The smaller concentration of mass, between 4 and 6, is largelydue to the local estimates for the United States from the paper of [14]. In May, the pattern shifted, with smaller R estimates generally. Very few estimates were obtained above 4. By June, most estimates were between 1and 2.5. The higher estimates were all obtained at a national level, with the top five recorded for Brazil, India,China, Spain, the UK and France. The histogram of all R values is shown in Figure S1, exhibiting a bi-modaldistribution similar to the patterns in Figure 3, with the greatest concentration of values between 1.5 and 3, andthe second smaller mass between 4 and 6, mostly influenced by the local estimates published by [14] in April. National distribution of R value Figure 4 shows the median R per country. As of June, most of the estimates were between 1 and 4, with onlythe USA (median R =4.78) and Spain (median R =4.25) being over 4. Only two were below 1, namely HongKong (median R =0.61) and Singapore (median R =0.7). Variance could not be measured at a national leveldue to the small number of estimates for many countries. Of the 65 countries, 36 had only one estimate availableper country. The fact that many of these countries were included at all is largely due to the contribution of[18], who provided R estimates for 53 countries. 5igure 4: National R medians as of June 2020.Figure 5: Regional R medians as of June 2020. Regional distribution of R value To investigate variability, the countries were aggregated, as variance can not be reliably calculated for the 36countries with only one estimate. The countries were grouped into regions as defined by the World Bank . Thegroupings are shown in Table S1.The regional median estimates are displayed on the map in Figure 5, while Figure 6 shows the change inregional medians by publication month of the estimates. The medians and standard deviations are cumulative,meaning they are based on all the estimates up to the relevant month. There is an initial increase for the EastAsia & Pacific region, from 2.65 in February (based on 2 estimates) to 3.19 in March. From March it decreasessteadily to 1.77 in June. Similarly, the median decreases for all regions as shown in Figure 6, except for NorthAmerica and South Asia. For these regions, the median R increased. It is worth noting that neither of theseregions had many estimates in earlier months. As of June, the region with the highest median estimate is NorthAmerica at R =4.28, and the lowest median is obtained for South Asia at R =1.47. This places the R estimatefirmly above 1.The bands in Figure 6 show 10% of the standard deviation, providing a visual representation of the variabilityper region. The 10% is chosen for visual clarity, as showing the full standard deviation would cause the bandsto overlap to the point of illegibility. The standard deviation increased in the East Asia & Pacific region from https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups Social factors that affect the variability of R From the previous section, it is clear that the value of R differs across countries and regions. Since the R value was estimated at the beginning of the pandemic, this is not a result of varying laws or non-pharmaceuticalinterventions in the face of the pandemic, but rather some factors related to the ordinary state of affairs in acountry prior to lockdown. A brief analysis is conducted to explore how R varies based on social or mobilityfactors.The United Nations Human Development Index (HDI) is considered to measure social development, alongwith the Sum4All Sustainable Mobility Index (SMI) as a transport index. The HDI measures the quality of lifeof a country’s population. It considers the length and health of citizens’ lives, their education and knowledge,and standard of living. It is easily interpretable and data is collected yearly, therefore the 2019 HDI was usedherein. The SMI is a multi-faceted metric which measures how well a country is performing in terms of achievingsustainable transport for the future. It takes into account rural access, urban access, equality in the treatmentof male and female users of the transport system, efficiency, safety, as well as air pollution and GHG emissions.The data for nearly all countries globally is publicly available on the Sum4All website. Countries were groupedinto developing economies or transitioning/developed economies as classified by the UN [20].Based on the R values in the study, R is not strongly correlated with HDI ( ρ = 0 . ρ = 0 . ρ = 0 . ρ = − . R values wereobtained for countries with low HDI, SMI or median age, or a high population density. Based on the availabledata, variation in R increases as HDI, SMI and age increase, and decreases as population density increases,in perhaps a non-linear way. The boxplots show that the median R is similar for developing versus developedcountries, however the spread of values is negatively skewed for developing countries and positively skewed fordeveloped countries. Additionally, the variability is higher for developed countries. When considering medianage, a roughly symmetrical distribution was experienced for countries with a younger population, while the http://hdr.undp.org/en/content/human-development-index-hdi Accessed on 16 October 2020. https://sum4all.org/ Accessed on 15 October 2020. R values was negatively skewed for countries with an older population. The latter category alsoexperienced more variability than the former.a) b)c) d)Figure 7: Relationship of R to social factors. a) Median R by HDI. Points are grouped according to developed-or-transitioning or developing economies. b) Median R by median age. Points are grouped according tocountries with a median age of 30 and over, or below 30. c) Median R by SMI. d) Median R by populationdensity. For the purpose of the visualisation, countries with population density above 12 000 were excluded,namely Bangladesh, Singapore and Hong Kong. Illustrative example for South Africa
South Africa’s first case of COVID-19 was announced on 5 March 2020. The government took a swift risk averseapproach placing the country in full lockdown due to high levels of co-morbidity in the population and concernsregarding hospital preparedness. We illustrate spatial variability of R in South Africa as a special case of thevariability observed in the systematic review to further guide the discussion of this paper. Values of R in South Africa There are three main approaches for estimating R [21], namely, using a mathematical expression of the pa-rameters for contact amongst the population; using data of cases over time; or using expressions based on thevalues estimated from the endemic equilibrium of infection. An R package R [22] is also available to estimatethe parameter for epidemic outbreaks, depending on data one has available.In South Africa, the National Institute For Communicable Diseases (NICD) has estimated R (as well asthe effective reproduction number) [23]. They used the maximum likelihood method of [24] and the first 15 –19 days of the epidemic in South Africa to estimate R . The main effect was due to international travellersentering South Africa at the start of the epidemic. Table 1 provides the estimates of [23] for the most affectedregions. The variability of the values indicates a non-homogeneous spreading of COVID-19 within South Africa.Accounting for spatial heterogeneity is therefore critical when modelling COVID-19 transmission, and R inparticular. 8able 1: Estimates of R in South Africa [23].Region South Africa Western Cape Gauteng Eastern Cape KwaZulu-Natal R estimate 2.07 1.76 2.19 1.84 1.7395% CI (1.69, 2.5) (1.11, 2.62) (1.95, 2.39) (1.1, 2.84) (1.15, 2.47)Figure 8: Graph of the correlations between R and the proportion of infected individuals, at South Africanward level as per the 2016 census. Sensitivity analysis of spatial R In order to demonstrate the importance of the spatial component, we conduct a local sensitivity analysis using aSEIR model for the COVID-19 cases in South Africa. We consider the sensitivity of the proportion of individualsthat had been infected by the end of 201 days to the spatial R . R was sampled from a Uniform(1,5) distribution over 960 simulations , and the correlation between the R and proportion of infected individualswas calculated. Figure 8 illustrates the results of this spatial sensitivity analysis visualised as correlations.The figure illustrates that many correlations are between 0.8 and 0.9, indicating a strong relationship betweenthe spatial R and the proportion of infected individuals. The figure further demonstrates that the strength ofthis relationship varies across geographical space. This necessitates the use of a spatially varying R to modelCOVID-19. Future work aims to provide such a spatial SEIR model for South Africa. This study provided a comprehensive review of initial R estimates of COVID-19, as modelled at the beginningof the pandemic in each country. The changes in the published estimates over the first half of 2020 displayeda growing understanding of the R value in the modelling community. R values were visualised at a nationaland regional level to investigate the spatial heterogeneity evident in the estimates. The relationship between R and various social factors were visualised and described, and it was found that a higher R value was relatedto higher values of HDI, SMI, and median age, and a lower population density. An illustrative example ofCOVID-19 modelling in South Africa highlighted the importance and the challenges of accounting for spatialheterogeneity when estimating R .The study included 81 papers and covered 65 countries. The countries with the highest number of estimates,local as well as national, were China and Italy. The highest national R medians were recorded in the USA(median R =4.78) and Spain (median R =4.25). The USA is well connected to other countries, includingChina, along direct flight routes and other channels of transport. This may have led to a higher influx ofinfected individuals before travel restrictions were put in place. A study by [25] suggests that while the initialSpanish case was imported from Germany in late January, numerous cases were received from Italy throughFebruary before local infections began to rise by the end of February. Italy also recorded high estimates (median R =3.14). As it was one of the first countries to experience the pandemic, recording the second largest outbreakin March [26], the country had little time to make preparations. This includes unofficial decisions by individualsto take preventative measures such as self-isolation and social distancing. While the values in this review were Here we conduct the spatial analysis at a ward level as per 2016 census data. Based on the ranges observed in this systematic review. Chosen due to the number of cores available on the high performance computer made use of. R than Italy were Germany (median R =3.37) and South Africa (median R =3.26).The highest individual value of R was 14.8 for the Diamond Princess cruise ship. This number may not berepresentative of COVID-19 transmission, as the disease spread rapidly in the confined conditions of the cruiseship. The influence of super-spreaders would also have been exacerbated by these conditions. Recent studiessuggest that super-spreaders are responsible for the majority of infections [27, 28].The lowest median R values were recorded for Hong Kong (median R =0.61) and Singapore (median R =0.7), the most densely-populated countries in the study. Both countries have a history of managing pan-demics, so that unofficial non-pharmaceutical control measures, such as mask-wearing, may have been in placebefore official COVID-19 regulations were implemented. The lowest median R values above 1 were recordedfor Uruguay (median R =1.03), Romania (median R =1.04) and Saudi Arabia (median R =1.09). Note thata low R does not necessarily mean fewer infections, since a low estimated R might also be due to a lack oftesting or unreliable case data. R is usually estimated from the number of confirmed cases, which can be adirect result of a country’s testing strategy.At a regional level, the highest regional median was recorded for North America (median R =4.28), whileSouth Asia obtained the lowest median (median R =1.47). The median shows an increasing trend, with higherestimates for June than for May. This does not necessarily imply that the estimate will continue to rise. Mostregions experienced a sharp decline in median R from the first to the second month of estimation, with aslight increase into the third month. This indicates an initial overestimation by modellers, followed by an over-correction, finally stabilising to a slightly higher value that is nonetheless much lower than initial estimates.The exceptions to this trend were East Asia & Pacific, which experienced a sharp peak before a decline, andNorth America, which increased steadily to a regional median of 4.28. It is worth noting that neither of thesetwo regions had a large number of estimates. It is therefore possible that the median R may decrease as moreestimates become available, following the pattern of the East Asia & Pacific Region. All other regional mediansstabilised, showing a growing consensus among researchers that the R value of COVID-19 is between 1 and 3.Patterns in the variance differed by region. East Asia & Pacific had a low initial variance, calculated ontwo estimates for China in February. This increased slightly into March, spiked to a value of nearly 8 in April,then decreased sharply in May and more gradually in June. For the Middle East & North Africa, Europe &Central Asia, and North America, the variance in the published estimates showed a spike in uncertainty fromApril to May followed by stabilisation in June. Sub-Saharan Africa followed a similar trend, experiencing agreater decrease from May to June. The above mentioned regions all follow a similar pattern to East Asia &Pacific, lagged by two months. This may be due to later outbreak dates. Latin America & Caribbean shows asharp decrease from the initially high variance (close to 7) in April, to a variance below 5 in May, and a slowerdecrease in June. This follows precisely the trend in East Asia & Pacific, not lagged. The variance for SouthAsia increased sharply from May to June. Based on the trends of the other regions, we may expect this varianceto decrease in July.The impact of various social and demographic factors on R was studied. In particular, the HDI anddevelopment level were used as indicators of socio-economic development, while the SMI indicated the level ofdevelopment of the transport network of each country. The median age and population density were consideredas demographic factors. While no linear relationships were found between R and any of these factors, clearpatterns were visible. In general, there were no high median R values for countries with low measures of socialdevelopment and infrastructure, i.e. a developing economy, low HDI or low SMI. It should be noted that mostestimates were obtained using the data of reported cases. In the case of developing countries, a substantialportion of the population may not have access to testing facilities due to lack of infrastructure, leading to fewercases being detected. No countries with younger populations had high median R values. In contrast, no lowmedian R values were recorded for countries with a high population density. Additionally, the data showedthat the variability in R depended on the value of these factors. Variability and hence uncertainty in R increases with age, HDI and SMI, and decreases with population density.The South African example illustrates the necessity of accounting for spatial heterogeneity when modelling R , while highlighting the difficulty of modelling R in real-time. The R values estimated by the NICD illustratethat the spread of COVID-19 varies by geographical region within South Africa. This supports the findings ofthe literature review that R varies spatially. The sensitivity analysis further demonstrated a spatially varyingrelationship between R and the proportion of infected individuals, with differing correlations observed perward. Accounting for spatial variation in R is crucial for modelling the spread of COVID-19. This musthowever be done with caution, as adding a spatial component to a model will increase complexity and mayincrease uncertainty. A thorough understanding of the ways in which R varies spatially is therefore required.This study provides a critical stepping stone towards such an understanding.As more data becomes available, future work should incorporate R literature of later months, including the10est of 2020 and beyond. In addition, this would allow further modelling to better quantify the relationship of R to the social and demographic factors. The date of the initial outbreak per country could also be included,as well as unofficial social prevention measures put in place by citizens prior to official lockdown, as these mayaffect the value of R . Estimation techniques could also be considered, as in the review of [8]. Since R isheavily reliant on the testing strategy of a country, testing statistics per country may also be included, subjectto data availability. Lastly, the methodology of this study could be extended to investigate spatial variation ofthe time-varying reproduction number. This systematic review presented a summary of the R estimates of COVID-19 across the first six months of thepandemic, indicating a changing understanding of the disease characteristics. Spatial variation in the estimateswas presented at a national and regional level. In addition, the study explored the relationships between R anddevelopment and SMI. Countries with a lower level of development, a younger population, a lower sustainablemobility score or a higher population density did not record high R values. The example for South Africademonstrated the necessity of introducing a spatial component into transmission modelling. This study providesa vital step towards characterising spatial variation in R . A deeper understanding of the factors that influencethe spread of COVID-19 is critical for combating this disease which has disrupted so many lives across theglobe. Conflict of interest
The authors have no conflict of interest to declare for this study.
Acknowledgements
This work is supported by the NRF-SASA (National Research Foundation-South African Statistical Associa-tion) Crisis in Statistics Grant. We acknowledge the use of the support of the Center of High PerformanceComputing. The contributions of and discussions with Sally Archibald, University of the Witwatersrand, arealso acknowledged.
Author contributions
Conceptualization: IFR, RT, NA, RMD, JH, CJVR, PD, NDT, ZK, ALR. Data curation: RT, NA, IFR. Formalanalysis: RT, NA. Funding acquisition: - Methodology: NA, RT, RMD, JH. Project administration: IFR, PD.Visualization: RT, RMD. Writing – original draft: RT, NA, IFR, RMD, ZK. Writing – review & editing: IFR,RT, NA, RMD, JH, CJVR, PD, NDT, ZK, ALR.