key: cord-0998183-ybfkam6c
authors: Shimul, S. N.; Hussain, M.; Faisel, A. J.; Hamid, S. A.
title: COVID-19 Cases Projection in Bangladesh
date: 2020-07-22
journal: nan
DOI: 10.1101/2020.07.20.20158527
sha: 4b1e72eba134b10dbbf95ff784568eeb0c6e9bec
doc_id: 998183
cord_uid: ybfkam6c

In this projection exercise, we analyzed the circumstances of the COVID-19 pandemic in Bangladesh and used multiple methods to characterize the epidemic curve. We merged several publicly available data sets for the purpose. Projections using actual Government data as of June 16, 2020 reveals that the epidemic curve for Bangladesh may be different from that of developed countries and quite similar to such curves in countries in the region. This is true, both in terms of incidence of cases (total number of cases per million population) and length of the epidemic (months to peak or flatten the epidemic curve). We find that while Bangladesh went into lockdown early, efforts to maintain lockdown at a national level was relaxed and new cases accelerated; with significant growth happening since lifting of lockdown on May 31. Our estimates indicate prevalence of COVID-19 may be between 200,000 and 600,000 towards end of the year, may take 9 months (270 days) to flatten the epidemic curve, lifting of the lockdown may have increased total cases by 60 to 100% and may have prolonged the epidemic by additional 2-3 months.

Researchers have determined that COVID-19 is a highly infectious virus that may spread person to person in multiple ways (2) and the rate of transmission often represented by the reproductive number or R 0 was estimated to be higher than known infectious virus (2) . It was also found that 1-6% of those infected may die (3) . As of the writing of this paper there is no cure or vaccination for COVID-19.

Since the early days of the COVID-19 pandemic, researchers have warned that unless mitigated, the dramatic increase in infected cases may overwhelm health systems. Thus, it became critical that estimation exercise of the potential number of cases be undertaken for Bangladesh, to inform timing and levels of potential cases. This paper reviews the current status of COVID-19 pandemic in Bangladesh, discusses different policies undertaken to mitigate the spread, evaluates and estimates total spread of COVID-19 cases in Bangladesh using multiple methods.

Disease dynamics of COVID-19 indicate that transmission may happen from droplets to nose and eyes and may go from infected person to surface or to other persons (2) . Once a person is infected, there is an incubation period where infected person does not show any symptoms.

This incubation period may be between 5 and 14 days. However, the infected person, who does not have symptoms, may be able to transmit the virus (4) . Hence, while symptomatic persons may infect others, it is also possible infected asymptomatic persons may also spread the disease (5) . It is for these reasons that face masks, hand washing hygiene and social distancing has become so important in containing the spread of the disease.

European study estimated initial reproduction number for COVID-19 is 3.8 and NPI has had significant impact on the spread of the disease resulting in reproduction number of less than 1 at 0.86 and final infection rate of about 3.2-4% of the population (12) .

In the context of Bangladesh, the first COVID-19 case was identified on March 9 and by March 23, 2020 total cases reached 33. It is at that point that the government of Bangladesh announced shutdown 1 of public, private offices, schools and colleges from March 26,2020 for 10 days (15). Subsequently all public transport systems were shut down (16). Police, the Army and other security forces were deployed to enforce social distancing. Announcement of the initial lockdown, in the form of a general holiday, for 10 days was followed by multiple extensions.

Since the announcement on March 24, 2020, there have been several occasions where there have been relaxation of lockdown including allowing garments workers to join work(17), allowing stores to open and relaxation of travel during Eid religious holiday(18). Finally, the Government announced that all general holiday will end on May 30 and work and travel can resume on May 31; however, there would be a strict enforcement of NPI such as social distancing, and vulnerable population such as the elderly, pregnant women and those with multiple chronic diseases are discouraged to re-join their workplace. Schools and colleges were to remain closed.

In the mean time the Government has attempted to increase capacity of the health care system, including increased number of COVID-19 testing sites from 2 to more than 50 (19), testing about 15,000 a day; recruitment of 2,000 new doctors and 6,000 nurses (20) and opening To mitigate economic impact of the pandemic and subsequent cyclone, the Government has implemented multiple stimulus packages targeted at workers, small and large business, garments and agricultural sectors, low income/informal workers and those affected by cyclone 1 The government declared a general holiday/ leave instead of calling it 'lockdown' or 'shutdown'. However the objectives were the same-to tame the spread of the virus.

Amphan. The Asian Development Fund (ADB), the International Monetary Fund (IMF) and the World Bank has pledged aid and loan (23) to support Bangladesh in this crisis.

Studies on Bangladesh specific COVID-19 models estimated that by end of May, 2020 about 89 million may be infected and total deaths could be 500,000.(13). Another study estimated that without NPI, daily new cases at peak of the epidemic would be nine Million and deaths over 18 months would be over 200,000 and with NPI, new daily peak would be 36,000 and cumulative deaths would be about 5,000 (14).

Although using simulation, various studies attempted to understand the impact of various interventions, and there have been some attempts to project cases using real data, there is not evidence in the literature where multiple modeling methods are used to predict the cases as well as to understand the impact of policy interventions.

In light of these events, in this paper we attempt to estimate the epidemic curve for Bangladesh under alternative scenarios using multiple methods. Pre-post analysis is done to evaluate impact of the lifting of the lockdown on May 31. The significant contribution of this paper is to apply models from diverse fields: epidemiology, econometrics, and data analytics (machine learning tools) which allowed us to capitalize the strength of various fields. Since there is an insurmountable amount of unknown that follow COVID-19, using various tools can provide a stronger prediction. This paper uses actuals figures as of June 16,2020 obtained from IEDCR, official Government authority to provide COVID-19 data. At the time, about 100 days had passed since the first case and total number of infected cases in Bangladesh was about 90,000.

As noted before researchers employ standard epi models to predict levels of infections in a pandemic. However due to lack of Bangladesh specific parameter data on contact and transmission rates and significant measurement issues with available data, standard method of SIR may not be sufficient for projecting cases in Bangladesh. Second, unknown characteristics of COVID-19 creates substantial uncertainty and finally changing nature of the NPI in Bangladesh since the beginning of the epidemic requires multiple approaches to projecting cases.

One method is the Susceptible-Infected-Recovered (SIR) model developed by Kermack, and McKendrick (24). The model uses parameters of incidence, transmissibility, duration of infections and recovery/deaths to estimate trajectory of incidence of infection and deaths over time. Note recovery here means those who have recovered or died as result of the infection and are removed from those who are susceptible.

The SIR model is used to estimate rates of infections for highly infectious disease. Often termed as compartmental model, the model relates changes in suspected case (S), infected cases (I) and recovered or death (R ) via 3 differential equations as shown below:

where r represent transmission rate and a represent rate of recovery or death among infected. Note R+S+ I= population, which is assumed to be constant and for Bangladesh it is assumed to be 161 Million. Equation (2), can be rearranged to represent R 0 = r S/a, where r/a represent contact ratio or fraction of the population that comes in contact with persons who are infected during the period of the epidemic and R 0 reproduction number. R 0 value of greater than 1 represents a epidemic as number of newly infected will be greater than number of currently infected.

Challenges with the SIR model is that all parameters are subject to local conditions; so, contact rate may be affected by labour market decisions of individuals, recovery rate may be . CC-BY-NC-ND 4.0 International license It 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 July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint affected by availability of local hospital capacity etc. These conditions vary widely country to country.

Variation of the SIR model was developed by Battista (25) and it is that model that is used in this study.

A method to simulate the expected exponential growth of COVID-19 cases in a Sshaped curve can also be estimated using a Gompertz. The simplest form of the Gompertz curve can be expressed as:

CI t is the cumulative cases at time t , a is the projected number of maximum cases, b is the growth rate of infection, t 0 is the start date. The cumulative cases are driven by the three parameters; a, b and t 0 and using regression techniques these may be estimated.

Gompertz curve has been previously used to estimate patterns of human height growth as well as spread of an epidemic (26,27,28). The main advantage of using Gompertz distribution is that it allows having a longer tail which is a likely case for COVID-19. In addition, this distribution is flexible in parameters compared to say logistic and exponential models.

This approach is to fit actual data to a quadratic (parabolic) function that reflects the epidemic curve observed in countries that have experienced the COVID-19 pandemic over a given number of days. The first incidence in Bangladesh was in March 8, full 3 months after the first incidence in China. That means daily data was available for all countries for 90+ days to analyze experience in other countries with the virus.

The approach is to use estimation of new case count based on quadratic function where the dependent variable is a function of square of the dependant variable. Using this approach, number of new infections in a country i at time t after x days of the onset of the infection can be estimated using the following formulae.

. CC-BY-NC-ND 4.0 International license It 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 July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint

In this model, number of daily cases in a country is assumed to depend on the number of days since COVID-19 cases reached 100 (x) and its squared value. It is assumed that x capture epidemies measures such as contact rates and transmissions rates and its square values assumes the epidemic will peak in a country after certain days in the country. It is also important to note maximum level of I when the first differential of equation 4 is set to zero and maximum value of x is reached when

. A useful aspect of this method is that it allows us to change the values of a or b to decrease or increase the days over which the epidemic will reach a peak.

Thus higher values of a or lower values of b will increase the days to peak of the epidemic.

These adjustments to parameters may be informed by evidence of impact of NPI or lack there of.

Using regression methods, parameters a, b and can be estimated and these parameters can then be applied to futures x for specific country to calculate I for the future.

Another method is an open sourced time series forecasting tool developed by Facebook known as Prophet (29). This platform has previously been used in COVID-19 (30). The algorithm automatically selects changes in data points, runs piecewise linear or logistic regressions accounting for variations in data due to various events, say weekdays. Another advantage of this tool is that it can capture recent changes and can provide better forecast taking both old and relatively newer growths. This tool has been extremely popular among the data scientists and machine learning experts, and some famous machine learning sites provide projections using this tools, such as in Katana ML firm (https://katanaml.io/). The prophet model can be run using R or python. We applied python fbprophet library to produce our results.

Although recommended approach to manage a pandemic is to implement NPI strategies until pandemic has reached a peak, many developing countries do not have the economic capacity to sustain strong NPI measures for a lengthy period across the country. As a result, many countries are forced to relax their respective NPI enforcement even when daily cases have . CC-BY-NC-ND 4.0 International license It 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 July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint not peaked. Bangladesh is one such country where NPI strategy was lifted on the 99th day since the first case, even though new case counts were on the rise.

Since each of these models are based on actuals as of a specific date, estimation based on two different times can be used to evaluate impact of NPI policy pre and post lifting of NPI.

Thus impact lifting of NPI is D i for country i can be calculated as

Assume NPI was relaxed on date t and current date is ts, where ts is greater than t, T is the when total cases reaches a peak (new cases approach zero) then impact of NPI is estimated to be the difference in projected total case using actuals as of date t versus actuals as of ts. In such as case, ‫ܫ‬ ప ෩ is the actual or estimated incidence of total cases at time i, ts is the date on which NPI was implemented and t is the date of the most recent actuals and T is the final date of the of the projection period. Note that t is greater than ts.

Country level annual demographic, health care related data and income was obtained . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint Country level COVID-19 related daily data was obtained from three different sources.

Europeans Union's European Centre for Disease Prevention and Control (ECDPC) data provides country level daily cases and deaths for COVID-19 (32). The country specific daily test data was downloaded from the "github" site(33). Country specific daily reduction in cell phone traffic was downloaded from Google mobility (34). Google publishes data on location specific mobility data by type of location for each country and by date. The report shows percentage change in cell phone traffic in a given location in a country or region for a given date compared to a baseline of Jan-3 to Feb-4 of 2020. The location identified include retail, grocery, parks, transit, workplace and residential. Review of the mobility data for Bangladesh showed that data for parks and grocery were varying during the lock down period, workplace data was deemed to not reliable as significant portion of the working population is in the informal sector (thus no specific workplace location) ; finally trends in transit and retail data were the same. Thus for the purpose of this study we use changes in cell phone traffic in transit location in Bangladesh. It is assumed this trend reflects changes in social distancing and other NPI, where lower levels (higher negative levels) reflect stronger social distancing and vice versa.

Country level daily data from these 3 sources and annual figures was then merged to create a single data set for this study.

For comparative purpose, we pursue a cohort type analysis where we compare measures for Bangladesh to similar measures of other selected countries where all countries were in the 99 th day of their respective COVID-19 epidemic.

There about 100 countries selected for comparison. The category created to represent countries per-capita income is INC_RNK_CAT. The data shows Bangladesh is in INC_RNK_CAT =1 among all the countries ranked by Per Capita Income. All measures are reported as median in their respective group. Bangladesh has lower spending on health than its comparators (2.5% vs 5.5%), however in terms of life expectancy, Bangladesh does better (71.8 vs 62.7). Age distribution data shows Bangladesh has more elderly as % of all population compared to its cohort; however this percentage is lower in Bangladesh than other countries with higher per capita income.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint . These aggregate measures appear to show as per capita income increases, countries tend to spend more of their income on health, they appear to have higher life expectancy and as a result percentage of the elderly population increases.

COVID-19 related measures show that transit mobility has decreased by 37% as of June 16 in Bangladesh compared to baseline mobility of January/February,2020. This is within range of the same measure of other countries; however other countries may have different epidemic experience and thus the decline will need to be evaluated in the context of experience of each country. Tests per million population show Bangladesh has higher rates compared to it cohort( 93 vs 85) and lower than rates in higher income countries. COVID-19 total cases per million for Bangladesh appear to be significantly higher than its cohort countries (562 vs 285) ; note this is measured 100 days after the date of the first case for all countries. Thus Bangladesh appear to have much higher rate of infection compared to other countries in same income groups.

For deaths per million, rates for Bangladesh is higher than its cohorts and in terms of case fatality (which equals total deaths divided by total infected), rates for Bangladesh is comparable to its cohort.

Compared to high income countries, Bangladesh appear to have lower cases per million; this may be a reflection of lower levels of testing or other factors. WHO has provided guidance on testing which is based on test positivity rate. The positivity rate is the ratio of the number of people who test positive for COVID-19 to all those who are tested. As testing programs scale up, the number of tests go up and if the virus is contained or on its way to be contained, the number . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint of people testing positive go down, thus the positivity rate is expected to go down. WHO has advised that testing programs should try to test such that positivity rate is between 3 and 12%.

Currently, In Bangladesh, the positivity rate has been about 20%. Thus in Bangladesh testing may need to be at-least doubles to reduce positivity rate. There could be other reasons that results in lower cases in Bangladesh which may include levels of infection spread and different levels of immunity to the virus.

Higher income countries appear to have higher levels of testing, levels of cases and deaths and higher levels of case fatality. High case fatality in high income countries could be explained by the fact that these countries have more elderly population and more of the elderly population are institutionalized. A single outbreak in one of these institutions, which typically houses 100-200 seniors, may result in significant outbreak and a large number may of the seniors may not survive. Thus recent evidence show that most of the elderly deaths (those 65 years or older) were not in the community but in long-term care facilities. About 80% of the deaths in Canada and 40% of the deaths in USA were in long-term care facilities (39). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint elderly as percentage of the population are within range of its neighbors. In terms of COVID-19 measures, Google's Transit data show Bangladesh had higher values than only two of neighbors Vietnam and Pakistan and for two different reasons. By the 99 th day Thailand was registering less than 50 cases a day and Nepal was registering less than 5 cases a day. Vietnam has already reached its peak daily cases and thus was opening up its economy while Pakistan was still experiencing exponential growth in daily cases as is Bangladesh. Total cases data indicate Bangladesh has the highest rate of infections among its neighbors. The rate of deaths in Bangladesh is second highest, highest being that of Pakistan. Table 2 shows Bangladesh specific weekly data for transit, tests, new and total cases.

Noting that Bangladesh announced a general holiday on march 28, 2020 with an attempt to increase levels of NPI, it appears NPI was most effective in Bangladesh in around middle of April when the percentage decline in transit mobility was highest at around 70% but then . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint mobility has been increasing and the most dramatic increase coming in on first week of June with almost 20% increase in mobility. Data also shows significant increases in testing and new cases since mid-April from about 2,000 tests to about 15,000 tests a day and new cases increase from 300 to about 3,000 cases a day. Thus tests per day has increased 7 folds while new cases have increased by 10 folds.

In Bangladesh, national lockdown was announced less than 20 days after the first case was identified. Many developed nations had sub-national lock downs and national lock downs was announced much later than that in Bangladesh. In Canada and United States, for instance, sub-national lock downs were announced and varied by Province/State. In these countries there was never a national lock down. In UK and Italy, national lockdown was announced long after the date of the first case. In Italy for instance first case was detected in January 31 and the national lockdown was announced on March 9, 38 days later (35) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint targeted national lock down was announced about 50 days after the first case (36) and India national lock down was also announced 50 days after the first case (39). As the transit mobility data Graph 1 and Graph2 shows, Bangladesh experienced a significant drop in mobility well before other countries and the reduction was significant to about -70%. That reduction remained in place for about 40 days after which mobility tended to increase signaling decreases in NPI compliance. In most cases, this trend contrasts with other countries where the reduction remained in place for first 100 days.

Trends in new cases per day per million population (Graph 3 and Graph 4) indicate that the cases in Bangladesh is still growing. When compared to regional countries, the levels in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint we use May 30 as the cut-off point for pre-lockdown analysis. It is expected that projected number of cases in the future will be significantly higher if we use actuals cases as of June 16 (post lockdown) versus projection using actuals thru May 30 (pre-lockdown). Further in certain models, the change in mobility between May 30 to June 16 can be used as proxy for reduction of NPI.

Actual new and total cases as of June 16, 2020 were used for this study. As of that date, the epidemic has been in country for about 99 days since the first case and as of June 16, 2020 there were about 90,000 cases with most recent moving 3-day average daily cases of about 3,100

The 'pre' period was defined as the period from date of first case to May 30, 2020 or about 80 days. As of May 30, 2020, there about 43,000 total cases and daily new cases was about 2,500.

Estimation using the SIR methodology, using actual cases as of June 16,2020, indicates that new cases will peak in around the middle of June and total cases will peak at about 194,527 and early phase of peaks in total cases will be reached around August, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint Under the 'quadratic' modelling approach, regression is run using equation 4 above. In this approach country level daily new cases per million is regressed on country specific time measures such as days since 100 th case and its square. The resulting parameters are then adjusted to account for increased transit activity in Bangladesh using Google mobility data. The relationship between Google mobility data and COVID-19 cases was explored by Yilmazkuday Bangladesh is yet to achieve daily maximum;

plus recent relaxation of lockdowns is likely to extend the daily maximum longer. So to adjust for that prospect, the coefficient was adjusted by the transit factor of 0.76.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint

As Graphs 9 and 10 shows, the quadratic method shows that new cases is projected to peak around July-2020 at about 3,600 a day and total cases likely to peak around November-2020 at about 310,000 total cases. If the current case fatality holds to the end, total deaths is estimated to be about 4,000. Given that there is wide-spread community spread, even with the recently announced efforts of Red Zoning, total cases may continue to increase. Projected fatality levels may be under-estimated as current fatality numbers may be under reported and future case fatality rate may increase as a result of health care system will come increased pressure and will struggle to save lives.

The Facebook Prophet approach results in daily cases to peak at about 4,800 around first week of July and total cases to peak around 388,000 in first week of September. Projected daily and total cases tracks very close to actual levels.

Using actuals as of May 30, the projection results in total cases to peak around the same time in September but the number of total cases would be significantly lower at about 247,000.

19 ity se . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint Comparison across all 4 models, as shown in Table 3 , reveals that total cases may peak between September to November and levels of cases may range from 194,000 to 615,000. Using the most recent case fatality rate and assuming same rates remain throughout the epidemic, total deaths may range from 2,600 to 8,200. ng . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint

The outcome of these analytic exercises was shared with key public officials for considerations and they have been using it consistently. So far the researchers have provided to the Health Government officials in the form of Technical Brief. The last one being # 10 and this process is on going. Appendix A provides example of such a report.

It was proposed that such analysis would assist key policy and decision makers in charge of managing the COVID-19 pandemic. After several iterations, it was decided that the results of the SIR model and district level data highlighting areas of improvements and concerns would be created.

All models projects that the current COVID-19 model may peak around the time frame towards end of the year. Total number of cases have a significant range multiple of 3, from about 200,000 to about 600,000; although 3 of the models predicts total cases to be 390,000 or lower.

The high range may reflect the uneven growth of cases since the start of the epidemic. Rate of infection spreading ranges from R 0 of 1.006 to 1.02.

The trajectory of COVID-19 epidemic in Bangladesh also reflects a much longer period of infection, here the daily peak is expected to be in July, 4 months after the first case was detected. NPI impacts supports the idea that lifting of the NPI has had significant impact in increasing potential cases from about 60% to 100% higher. This study has several limitations. The use of multiple models to predict the outcome of an epidemic in a single country may result in outcome that varies significantly from one another, thus increasing the level of uncertainty. Limited data availability of Bangladesh specific transmission and recovery rates forces us to select initial values that may not be appropriate. Time specific influential events such holidays, major climate events and subnational events could not be captured in any of the models. Some of these events may extend the epidemic period. This study also does not take into account recent changes in policy such as assignment of red zone, lock down of neighborhoods, increases in contact tracing and other measures that government has taken to mitigate the spread of the virus. This study does not address the direct and indirect health effects of the COVID-19 epidemic. Future analysis could address these shortcomings.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint There may also be a need to better understand and apply measures not directly captured in epidemiological reporting; we may need to include sociological and anthropological considerations in localities to better address the appropriateness of measures and methods in our modelling approach (37).

Using multiple models, this study estimates the trajectory of the COVID-19 pandemic in Bangladesh. Results show a wide variation in the total number of cases may range from 200,000 to 600,000 towards the end of the year. This paper also shows that NPI polices have significant impact on the intensity and length of the epidemic; suggesting that lifting of NOI may have extend the epidemic by 3 months and increase prevalence between 60 to 100%. Substantial uncertainty exist as the rate of transmission of the virus is unknown, public compliance with NPI efforts has been mixed and governments ability to enforce and sustain NPI policies is challenged due to the negative effects they have on economic life.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 22, 2020. S  I  R  M  o  d  e  l  :  W  e  u  s  e  d  t  h  e  r  e  a  l  d  a  t  a  p  r  o  v  i  d  e  d  b  y  I  E  D  C  R  t  o  e  s  t  i  m  a  t  e  t  h  e  p  a  r  a  m  e  t  e  r  s  r  e  q  u  i  r  e  d  f  o  r  S  I  R  m  o  d  e  l  ,  a  n  d  t  h  e  f  o  l  l  o  w  i  n  g  g  r  a  p  h  s  p  r  o  v  i  d  e  s  t  h  e  r  e  s  u  l  t  s  f  r  o  m  t  h  a  t  m  o  d  e T  h  e  d  e  t  a  i  l  f  i  n  d  i  n  g  s  o  f  t  i  m  e  s  e  r  i  e  s  m  o  d  e  l  a  n  d  S  I  G  M  O  I  D  C  u  r  v  e  (  L  o  g  i  s  t  i  c  a  n  d  G  o  m  p  e  r  t  z  D  i  s  t  r  i  b  u  t  i  o  n  )  a  r  e  n  o  t  s  h  o  w  n  h  e  r  e  .  G  o  m  p  e  r  t  z  d  i  s  t  r  i  b  u  t  i  o  n  i  s  l  e  s  s  r  e  s  t  r  i  c  t  i  v  e  i  n  t  e  r  m  s  o  f  v  a  l  u  e  o  f  p  a  r  a  m  e  t  e  r  s  i  t  c  a  n  a  d  o  p  t  a  n  d  s  o  t  h  i  s  i  s  v  e  r  y  l  i  k  e  l  y  s  i  t  u  a  t  i  o  n  .  S  I  R  m  o  d  e  l  s  e  e  m  s  t  o  d  r  a  w  s  a  m  e  c  o  n  c  l  u  s  i  o  n  a  s  t  h  e  r  e  s  u  l  t  s  f  r  o  m  G  o  m  p  e  r  t  z  d  i  s  t  r  i  b  u  t  i  o  n  -s  o  i  t  i  s  n  o  t  r  e  p  o  r  t  e  d  .  F  a  c  e  b  o  o  k  A  n  a  l  y  t  i  c  T  o  o  l  i  s  m  o  s  t  l  y  l  i  k  e  c  a  s  e  g  i  v  e  n  t  h  a  t  e  v  e  r  y  t  h  i  n  g  i  s  o  p  e  n  n 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 July 22, 2020. . CC-BY-NC-ND 4.0 International license It 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 July 22, 2020. . https://doi.org/10.1101/2020.07.20.20158527 doi: medRxiv preprint

COVID-19 outbreak associated with air conditioning in restaurant

Estimates of the severity of coronavirus disease 2019: A model-based analysis. The Lancet Infectious Diseases

Presymptomatic SARS-CoV-2 infections and transmission in a skilled nursing facility

COVID-19 diagnosis and management: A comprehensive review

Physical distancing, face masks, and eye protection to prevent person-to-person transmission of SARS-CoV-2 and COVID-19: A systematic review and metaanalysis

Escalating infection control response to the rapidly evolving epidemiology of the coronavirus disease 2019 (COVID-19) due to SARS-CoV-2 in Hong Kong

Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in wuhan, china: A modelling study

IHME COVID-19 health service utilization forecasting team. Forecasting COVID-19 impact on hospital beddays, ICU-days, ventilator days and deaths by US state in the next 4 months

Mathematical modelling of COVID-19 transmission and mitigation strategies in the population of ontario, canada

Estimating the effects of non-pharmaceutical interventions on COVID-19 in europe