key: cord-1050122-0z3qiz6p authors: Kivuti-Bitok, L. W.; Momodu, A. S.; Jebet, J. C.; Kimemia, F.; Gichuki, I.; Ngune, I. title: System Dynamics Model of Possible Covid-19 Trajectories Under Various Non-Pharmaceutical Intervention Options in Low Resource Setting. date: 2020-10-08 journal: nan DOI: 10.1101/2020.10.06.20204487 sha: 07baff845ca6eb4e50015ce6ac550c715a5a71bc doc_id: 1050122 cord_uid: 0z3qiz6p We present a population-based System Dynamics Model (SDM) of possible Covid-19 trajectories under various intervention options in the uniqueness of Kenya. We developed a stock and flow based SDM. We parametrized the SDM using published data and where data was not available, expert opinion was sought. Following validation test, the model was simulated to determined possible outcomes of non-pharmaceutical interventions in management of Covid-19. We simulate the possible impact of; social distancing, quarantining, curfew and cross-county travel restriction, lockdown of selected cities in Kenya and quarantining. We varied interventions in terms of start dates, duration of implementation and effectiveness of the interventions. We estimated the outcomes in terms of number of possible infections, recoveries and deaths. With the current state of interventions, we estimated a peak of Covid-19 in September 2020 with an estimated 13.5 Million Covid-19 cases and 33.8 thousand deaths in Kenya. The largest possible reduction in infections and mortality was achievable through increase in the effectiveness of the interventions. The suggested interventions would delay the epidemic peak of Covid-19 to between late Nov 2020 and early December 2020 with an estimated13M cases a 500 thousand reduction in Covid-19 cases and 32.4 deaths( a reduction in 1400 deaths). We conclude that SDM enables an understanding of the complexity and impact of different intervention scenarios of Covid-19 in Kenya. , especially for developing countries like Kenya. System Dynamics modeling approach has been used to demonstrate both qualitative and quantitative varied options to managing such pandemic as COVID-19 [14] . SDM has been advocated as a tool predicting the number of new cases as well as identification of best measures to mitigate SARS-CoV-2 transmission [11] . Developed by Jay Forrester in the late 1950s [15] , [16] SDM ( the origin of current whole systems thinking) is a differential equations-based model that involves a number of steps. The activities are '(1) problem identification and definition, (2) system conceptualization, (3) model formulation, (4) model testing and evaluation, (5) model use, implementation and dissemination, and (6) design of learning strategy/infrastructure' [17] . Mental models of dynamic wicked problems such as Covid-19 are presented using Causal Loop Diagrams (CLD). These are further developed into a comprehensive Computerized model using software such as Stella ® and Vensim ®. The variables identified in these CLD are translated in terms of Stocks (depicting variables that accumulate in number) and flows between the stocks as well as the information that determines the value of the flows (converter variables) [18] [19] . Feedback effects and delays are a key component of SDM. Differential equations are the main drivers of the model. In silico experimentations, which combine findings from literature and computerized mathematical models, allow vast numbers of experiments that may produce more accurate results that gives room for hypothesis generation [20] [21] [22] . Computerized experimentations are cost effective and are less time-consuming alternative to expensive real time laboratory and clinical experimentation. Simulation using computer software enables study of systems behavior over time and supports in silico policy analysis. SDM relies on existing qualitative and quantitative data, and where data is not available, expert opinion is sought [23] The CLDs, stocks and flows provide a common language that can be easily understood by a wide range of stakeholders. SDM has been recommended in analysis and understanding of the impact of different interventions in management of Covid-19 [24] [12] [25] [26] . The effect of quarantine . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020. 10 .06.20204487 doi: medRxiv preprint periods on contacts and deaths in Covid-19 has been modelled [26] . At the National level, SDM has been recommended as a versatile tool in decision making in population-based models [12] in regards to quarantine, social distancing, delivery of testing, hospital capacity, staffing, resource mobilization well as health and wellbeing of the patient. At the global level SDM can be utilized in better understanding of the impact of global quarantine [12] .SDM is generally used for strategic decisions affecting the whole population. Studies applying SDM have been conducted in a variety of settings . Few if any SDM of Covid-19 studies have been done in Sub-Sharan Africa. We adopted Susceptible, Infected and Removed(SIR)structure [27] and hence splits the study population into mutually exclusive groups, subgroups and compartments. In line with other modelling studies on epidemic and pandemics [28] [4] we separated the susceptible to include the exposed and the removed to include the recovered and the dead. Thus we adapted a Susceptible( persons who have not contracted but have potential to contract the virus) , Exposed(Persons who have come into contact with an infected person and may or may not have contracted the virus and are at the same time asymptomatic), Infected(persons who have contracted the disease, they may or may not be symptomatic and may infect others), Recovered (persons who had been infected with the virus and whose infection has cleared and may no longer infect others)and Death ( is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint The general structure of the model presented as stocks and flows diagram is shown in Figure 2 . Stocks represent variables that accumulate and are measured by levels. In this study, the stocks represent the number of people in each state regarding Covid-19 thus one may be susceptible, exposed, infected, recovered or dead. The flows represent the movement from one state to the next at a given time. Transit from one state (stock) to another is guided by a general is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint We used published data to calibrate the model and where data was not available, expert opinion was sought. We simulated the model for Population of 47.2 million people representing the Kenya Population. The basic parameters used to seed are presented in Table 1 and these represent the base case scenario. The information used to verify the model structure was sourced form SEIRD publications [28] [30] , recent case studies on Covid-19, World Health Organizations(WHO), Our World in Data, Kenyan Government press reports and expert judgement. The base case scenario represents the current status of Covid-19 interventions in Kenya. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. The model assumes: 1) A homogenous and static population hence the effect of new births and immigration was excluded. The stocks are therefore conserved. 2) That movement across compartments/stocks is a given time step 3) Incubation period of 5 days 4) That new cases can be detected on testing 5) Untested cases will not be identified; however, they progress through stocks in a similar way as the detected cases while in the community. That the unidentified cases in the community may get 'opportunistic testing' and follow the pathways of detecting cases in corresponding stocks. That there is conferred immunity after recovery. 10) That it would be possible for populace to consciously taken note and reduce the number of person to person contact per day. Since the infection from the virus is reinforcing, and therefore have an exponential growth, the measures taking to limit its transmission are expected to have a counterbalancing effect on its growth. Model validation done through a walk and passed adequacy and extreme conditioning tests. The ability to replicate historical Covid-19 data in Kenya was also demonstrated. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. The ability of the model to reproduce historical data was assessed through comparative runs from raw data reported on our world in data and simulated run at base case scenario The ability to reproduce historical data is shown in Figure 4 demonstrating the comparison between the two curves of daily reported cases and simulation results from our model. The two curves are similar in shape even though the Simulated numbers are higher due to low testing levels of Kenya, hence possibility of missed cases. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint Kenya has a rate of positive cases at 13% by end of July 2020 (Our World in Data, 2020) meaning that country performed many tests relative to the size of the outbreak, thus many cases are likely to be unreported. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint The base case scenario (with current interventions) is presented in Figure 6 . Showing that the peak infections of Covid-19 are likely to occur on day 178 (September 2020) with approximately 13.6 million active cases and 34,000deaths. We compared the Base case scenario through varying effectiveness of selected non pharmaceutical interventions. All activities geared towards physical distancing as well as hand hygiene measures included all behavioral adaptation such as closure of social gatherings and modification of transport systems among others . Holding all other variables as at base case scenario, we varied the levels of effectiveness of physical distancing at the WHO recomemded levels and the effectiveness levels the experts felt were realistic or could be achievable. The impact of various levels of behavioural adaptation were simulated and compared to base case scenario. The results are demonstrated in Figure 7 a and 7b shows that 50%, 65%(realistic level), 80% and 95% levels of effectiveness would push the peak of Covid-19 with 2, 4 and 6 days respectively with a minimal reduction in both active cases and deaths . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint Movement restriction in Kenya was implemented in two ways. A national wide curfew and lockdown of 5 hotspot counties. The lockdown was modified as it involved closure of borders of the hotspot counties as well as some sub-sections of two(Nairobi and Mombasa) of the hotspot counties. A curfew was effected for the first time on the 15 th day since confirmation of first Covid-19 case in Kenya. In the first month, the curfew was effected from 7pm to 5am daily. This was later varied to 9pm to 4am daily. By the time of publication, the 9pm to 4am curfew was ongoing. We did not differentiate this varying of curfew timings in our model. The experts felt that the 7pm to 5am curfew was more is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint effective at approximately 75% while generally people tended to take the 9pm to 4 am curfew less seriously with a suggested effectiveness of 65%. Overall, the base-case effectiveness of curfew was estimated at 65%. We varied the effectiveness of curfew at the 80% and 95% WHO levels and adopted the 80% effectiveness as our realistic level. Since the lockdown of hotspot counties had been lifted by the time of development of this model, we did not vary the variables associated with lock down. The effect of lockdown was however included in the base case scenario. As demonstrated in Fig 8 a and Fig 8 b , curfew may have resulted to shifting the peak of both active cases and deaths from 178 days at base case to 183 days at 95% effectiveness with minimal reduction in the number of active cases as well as the number of deaths. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint The initial approach to quarantining was self-quarantining of any persons who came in to the country after the Covid-19 pandemic was declared. This was found not to be effective and by the of day 21, the government implored compulsory quarantine for new entrants into the country. Special government designated facilities were utilized. Each quarantined person was expected to cater for their own cost of quarantining with the cheapest facility charging $20 per day. This led to an uproar from the general population, and there were reported cases of a few people escaping from some quarantine facilities. The quarantine facilities were viewed and reported by the local print and media as an avenue of perpetuating police brutality. Quarantine in Kenya was therefore a challenge. After about 4 weeks from the reporting of the first Covid-19 case in Kenya, the government resulted to free quarantine at government own centers. We estimated the overall effectiveness of quarantine at 75%. We held all other variables as at base case scenario and varied the effectiveness of quarantining. As demonstrated in Fig 9a and 9b increasing the effectiveness of quarantining from 75% to 80% and 95% would push the peak of active cases and deaths with 3.5 days and 5.5 days respectively. Quarantining effectiveness at 75%(Base Case) Quarantining effectiveness at 80% Quarantining effectivess at 90% . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint Since total lockdown of the country would not be feasible due to possible catastrophic social-economic impact, we assumed a scenario whereby the general population would is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint The realistic scenario was what we felt the Country had potential to achieve in Covid-19 management. As demonstrated in Fig 11, realistic intervention levels of all the selected non pharmaceutical interventions; effectiveness of physical distancing and hand hygiene at 65%, curfew at 80%, quarantining at 80% and person to person contact at maximum of 30) would result to a delay of peak of Covid-19 cases from 178 th day since first confirmed infection to a peak of 246 th day allowing approximately 67 extra days for preparedness of health care system. The new peak would likely be late November to mid-December 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint In this paper, we simulated the effect of non-pharmaceutical measures on the progression of Covid-19 pandemic in Kenya. Our model used locally adapted non-pharmaceutical measures for managing the spread of Covid-19 and flattening the curve. We ran the model using the WHO reported coverage rates of 50%, 80% and 95%, for modelling interventions in a pandemic (We assumed that the general population would make a conscious effort to interact with a specific number of people at most per day) We simulated several scenarios Our simulation suggest that multiple and feasible interventions need to be adopted to limit the spread of the virus and flatted the curve. When we simulated single interventions such as physical distancing and hand hygiene, even with a population uptake of this intervention . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. . https://doi.org/10.1101/2020.10.06.20204487 doi: medRxiv preprint at 95%, the infection and mortality peak rates could only be extended with 3.5 and 5.5 days respectively. Our suggested scenario depicting realistic intervention levels of all the selected non pharmaceutical interventions would delay of peak by 67 days allowing a modest amount of time to prepare the mitigate possible overwhelming of the health care system. The new peak would then be late November to mid-December 2020. While controlling for physical distancing and hand hygiene, the greatest impact was on the extending the curve was seen when person to person contact was varied The ideal scenario would be for Kenya to achieve a population uptake of 100% for all the suggested measures to control the pandemic. However, the social acceptability and feasibility of such level coverage in a resource limited setting like Kenya that has populous cities, overcrowded housing, high usage of public transport is dismal [33] . Majority of the hotspot countries are overcrowded and access to soap and water, and hand sanitizers remains a challenge too with alternative hygiene measures being fronted [34] . We therefore predict that use of realistic model, would allow the government time to organize resources to deal with the mortality and infection peaks. Covid-19 is dynamic and the data may vary drastically. Our model is based on person to person contact and provides suggestions that take into account the current situation in the country. The application of the model may be limited to Kenya because the mixing patterns of individuals may differ in other regions and countries and across cultures. While we acknowledge sufficient data was used to populate the model, we also leave room for incorporating new knowledge to further refine the model. We also did not classify the severity of Covid-19 cases.This model does not attempt to predict the course of Covid-19 in Kenya but rather generates hypothesis as to possible Covid-19 Trajectories from possible non-pharmaceutical interventions. The current non-Pharmaceutical interventions are likely to have pushed the peak date of Covid-19 cases to September 2020.Enhanced intervention would push this peak by Approx. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 8, 2020. A simplified language of the number of person to person contact per day may be a more understandable message. SDM is a useful tool in seeking a deeper understanding of impact of non-pharmaceutical interventions in Covid-19. A scoping review of 2019 Novel Coronavirus during the early outbreak period: Epidemiology, causes, clinical manifestation and diagnosis Review on COVID19 disease so far World Health Organization declares global emergency: A review of the 2019 novel coronavirus (COVID-19) The Global Impact of COVID-19 and Strategies for Mitigation and Suppression. Imp Coll COVID-19 Response Team Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand Spread and dynamics of the COVID-19 epidemic in Italy: Effects of emergency containment measures First case of coronavirus disease confirmed in Kenya Update of Coronavirus in the Country and response measures as at COVID-19 ) in Kenya : Preparedness , response and transmissibility PRESIDENTIAL ADDRESS ON ENHANCED MEASURES IN RESPONSE TO THE COVID-19 PANDEMIC. Nairobi Social network-based distancing strategies to flatten the COVID-19 curve in a post-lockdown world How simulation modelling can help reduce the impact of COVID-19 Using system dynamics to develop policies that matter: global management of poliomyelitis and beyond Understanding COVID-19 spreading through simulation modeling and scenarios comparison: preliminary results. Epub ahead of print 2020 Stock and flow concepts are usually associated with system dynamics modeling System dynamics philosophical background and underpinnings An expert view of the system dynamics modeling process: Concurrences and divergences searching for best practices in system dynamics modeling The potential of system dynamics The greater whole : Towards a synthesis of system dynamics and soft systems methodology On a Resurgence of Management Simulations and Games * Syst Dyn Rev Applying dynamic simulation modeling methods in health care delivery research -The SIMULATE checklist: Report of the ISPOR simulation modeling emerging good practices task force silico clinical trials: Concepts and early adoptions. Briefings in Bioinformatics. Epub ahead of print 2019 Mainstreaming modeling and simulation to accelerate public health innovation An Integrated Participatory Systems Modelling Approach: Application to Construction Innovation Understanding COVID-19 spreading through simulation modeling and scenarios comparison: preliminary results Modeling containing covid-19 infection. A conceptual model Lisa Maile PB. Modelling exit strategies from Covid-19 Lockdown with a Focus on antibody tests A contribution to the mathematical theory of epidemics Monitoring Italian COVID-19 spread by an adaptive SEIRD model System Dynamics Modelling: A Practical Approach Understanding COVID-19 spreading through simulation modeling and scenarios comparison: preliminary results. Epub ahead of print 2020 Cost-Effectiveness Guidelines Data. The daily positive rate, given as a rolling 7-day average Social Distancing and Lockdown in Black Townships in South Africa Godfrey. Epub ahead of print 2020 Dry Taps? A Synthesis of Alternative "Wash" Methods in the Absence of Water and Sanitizers in the Prevention of Coronavirus in Low-Resource Settings We acknowledge the contribution of Robert Eberlein through the Webinar of Modeling the COVID-19 Pandemic: A Primer and Overview, in Isee Systems. We appreciate the invaluable support of Dr Jonathan Moizer and Prof Jonathan Lean of University of Plymouth and Danny Ibarra-Vega of IRCACS(Columbia). The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.