[MUSIC] Hi and welcome to the lecture, my name is Emily Gurley and I'm an infectious disease epidemiologist at the Bloomberg School of Public Health at Johns Hopkins University. In this lecture, we're going to talk about using infectious disease transmission Modeling output to make public health Policy Decisions. By the end of this lecture, you should be able to give examples of public health policy decisions that can be informed by infectious disease transmission models. Including an illustrative example of using models to consider rubella vaccine policy. You will also be able to describe how policy can be incorporated into models to aid the decision making process. And you'll be able to list some important limitations to consider when using modeled output for making policy decisions. Let's start by talking about how policy decisions can be informed by models. Public health policy seeks to take some action. And one important role that models can play in considering policy decisions is by setting expectations for what could happen if nothing is done? If no action is taken, if there's no policy. This kind of baseline expectation could be provided by a forecasting model or a scenario model. The first question to ask from a policy perspective is, is any action required? Modeled predictions of the likely outbreak trajectory could lead you to decide that no policy or intervention is needed. However, if you decide that there should be some kind of policy to reduce transmission or burden of disease, then this baseline estimate can serve as a way to evaluate the impact of your policy. After all, if output from credible models show that there is likely to be a large outbreak and you take action to prevent the outbreak, then there will be no case counts to observe. If an outbreak is prevented, how will you know? Baseline modeled estimates of burden of disease can serve as one way to estimate the impact of a policy. Another possible scenario, is that you have no good policy options to offer to reduce transmission or disease severity. In this circumstance, the baseline estimates can be used to allocate healthcare resources. In this slide, you can see output from a scenario planning model of daily hospital beds needed for COVID-19 patients from early in the pandemic. Under the assumption that no interventions would be used to reduce transmission. You can see that there are multiple lines indicating uncertainty about the timing, duration. And peak values of the hospital beds that would be required to care for patients based on the model assumptions. However, all of the possible curves far surpassed the black line, which indicates the total number of hospital beds available in the model population. In this scenario, the modeled output has significant uncertainty but if the assumptions of the model hold, it is clear that there will be insufficient beds for COVID-19 patients. This is the kind of scenario where policymakers might want to take action to reduce transmission. But how would a policy be included in a model to estimate the possible effects of the policy on disease burden estimates? The first step, is to clearly define what the policy would be and how it would change transmission. What would the intervention link to the policy be? For example, is the intervention reducing transmission by reducing contact with infectious people? Or by reducing the duration of infectiousness or the level of infectiousness? Or is the intervention increasing immunity in the population? Next, you'll need to consider how efficacious the intervention is? What evidence do you have that the intervention will produce the desired effect? Even if the intervention was shown to reduce transmission however, in research or public health studies, how confident are you that the policy that you will make will influence the use of that intervention? So there's always a gap between how well an intervention can work and how well it does work based on how many people use the intervention. You need to ask yourself how well will the intervention be adopted in the population of interest next? Next, you must also consider how long it will take for the policy to take effect? And how long you believe the intervention will take to make meaningful changes to transmission? For example, often interventions implemented early in an outbreak will be far more effective at reducing transmission than those taken later. Next, you must consider how long it will take for the policy to take effect? And how long you believe the intervention will take to make meaningful changes to transmission. You must also consider when the intervention will be implemented. For example, often interventions implemented early in an outbreak will be far more effective at reducing transmission than those implemented later. When considering how well the intervention will work, you must also take into account when it's going to be applied. Finally, depending on how you think the policy will work and how well it will work. You can adjust the model parameters or structure of the model to estimate what the effect of the policy might be. This slide shows the remaining modeled output from the scenario model shown on the last slide. This is the output of a stochastic scenario model on the potential effect of various social distancing policies, on the number of hospital beds required for COVID-19 patients from early in the pandemic. You'll recognize the first panel as the scenario where transmission is uncontrolled and there are no policies to reduce transmission. The following four panels represent output from various models with social distancing policies incorporated into them. At the top of each panel, you can see how the color varies from dark blue to light blue. Dark blue indicates times with little to no social distancing intervention. While the light blue indicates strong intervention implementation. In the second panel, social distancing intervention starts early in the time period and is consistently implemented throughout the time period. This panel is named Social Distancing Fixed because the effect of the social distancing intervention is fixed over time. In the third panel, social distancing interventions are implemented at a similar time point and level as in the second panel. But fatigue for social distancing sets in and the intervention adoption decreases over time. Since this intervention does not change the underlying susceptibility of the population when the intervention is stopped, transmission begins again. And you can see the result in the outbreak curves that are predicted. In the fourth panel, the model assumes that social distancing policy has a pulsed timing. Where social distancing measures are implemented strictly for a short time, then lifted, then implemented again, then lifted again and again. In the fifth panel, social distancing is incorporated somewhat, but only a few of the areas in this hypothetical population actually adopt social distancing measures. These scenarios can help policymakers consider how much they might be able to rely on social distancing policies to stop transmission. Models can also be used to make decisions about how to optimally allocate resources given constraints. This is a common problem for making policy because almost always there are constraints on the resources available. Often, one of the greatest constraints is how quickly a policy or intervention can be implemented to change transmission. A good example of this kind of model is shown here. The question the model is trying to answer is, if we want to use cholera vaccines to stop an outbreak, how should we deploy them to maximize health impact? Cholera vaccine efficacy is high for a two dose vaccine schedule. But there's good evidence that one dose of vaccine will induce good immunity at least in the short term. So there are questions about whether an optimal outbreak response should focus on giving two doses to fewer numbers of people or just one dose, but covering more of the population. Cholera vaccine doses are received at these outbreaks from an international stockpile. So there are always constraints around access to the vaccine. Here, you can see the modeled projection of a cholera outbreak epidemiologic curve from Zimbabwe in grey. In the green line, you can see the projected trajectory of the outbreak if the one dose vaccine option is used. Where more people are vaccinated quickly versus the blue line where fewer people are vaccinated. But each person vaccinated is better protected from infection. In this case, the one dose schedule seems to perform better, though they are similar. The main message, however, is that both the one dose and two dose options are much better than doing nothing. Now let's look at another example with all the details of incorporating policy into a model. To make decisions about optimal vaccine rollout strategies given constraints. Early in the COVID-19 pandemic, models were used to determine how to best deploy vaccines to save lives given that supply of vaccines would be severely limited. One model considered five possible age groups to target for vaccination. Including those aged under 20 years, those aged 20 -49. Targeting groups, age 20 +, 60 + or a random allocation across all age groups. Although it was clear that older age groups were at higher risk for severe disease. It was unclear how vaccinating other age groups would change transmission patterns to also reduce risk in older age groups, since younger age groups often drive a lot of transmission. At the time, we did not have any results on what the possible efficacy of a COVID-19 vaccine would be to prevent infection. So this model assumed that 90% of people vaccinated would be fully protected and that the remaining 10% of people vaccinated would have no protection. Furthermore, the model assumed that 70% of people in the targeted age group would actually accept the vaccine and be vaccinated due to vaccine hesitancy or other concerns. They also had to make assumptions about how and when the vaccine would be rolled out. So they estimated that 0.2% of the population could be vaccinated per day. And then based on assumptions of how many vaccines were available to be administered, the rollout would take between 50-150 days. Here, you can see some of the output from this model. Let's consider each panel. Starting on the left, you can see the allocation of vaccines by age group across the five different policy options. The color of each of these options carries through to the subsequent panels. So you can see how the predictions varied by each age group strategy. There are two scenarios here. And in scenario 1, the reproductive number is assumed to be lower than in scenario 2. And this was done because of the uncertainty of transmission leading up to the vaccine rollout. However, I wouldn't pay too much attention to this because the results did not vary by scenario. Next, however, I want you to focus on the top four panels here. The first two show the proportion of people infected under a scenario of 10% and 30% of vaccine supply. The dashed line show projected infections without any vaccine as a comparison. The main takeaway from these results is that the policy option to reduce the most infections is indicated in the green line, where adults age 20- 49 are targeted for vaccination. If you look at the next two panels on the top row, you can see that again, the greatest reductions and infections are achieved if the 20 -49 year age group is targeted for vaccination. And again, that's represented by the green line. Now, let's look at the bottom four panels which is showing differences by policy in the cumulative mortality and reductions in deaths. Here, the output clearly shows that cumulative mortality is lowest and reductions and deaths are highest for the purple line. Which represents the policy of vaccinating those over the age of 60 before other age groups. This kind of modeled output was used to determine vaccine rollout strategies in the United States and many other countries. [MUSIC]
