Hello, my name is Nick Watson. Thank you for this opportunity to discuss mortality factors in COVID-19 for the SECM microlearning presentations. I produced the first three versions of this presentation in March and April of 2020 and have created the updated fourth version in September 2021. Over the past 18 months, our prognostication of mortality in COVID-19 has been refined primarily through larger sample sizes and pooled analysis, while simultaneously increasing in complexity due to viral genetic diversity, vaccination availability, and geopolitical differences. Here I will provide a snapshot overview of the major themes in this presentation. Mortality factors in COVID-19 can be simplified into three main divisions. Clinical factors subdivided into pre-existing factors and factors that are acquired during hospitalization or complications that accrue during hospitalization. Laboratory values primarily reflective of immune function, organ failure, and inflammation, as well as modifiers that take into account factors such as viral genetic variation, vaccination status, geopolitical factors, date of infection, and resource availability. Fatality rates are only as good as the data upon which they are based. Here, the information in red font represents pre-vaccination data with large sample sizes of 44,000 patients in China and 1.3 million patients in the United States. This shows that the majority of those presenting with COVID-19 were sent home with a mild form of the disease. Approximately 14% of presentations became hospitalized. 2-5% of all hospitalized patients ended up in an intensive care unit, and some number of those patients ended up on mechanical ventilation. We know the most about these hospitalized patients, and that's where our mortality data primarily comes from. This information in the upper left is from relatively large data sets from early 2020 surges in Detroit City, New York City, and the United Kingdom. While there are ranges demonstrated here, this really shows that the mortality is relatively high percentages for mechanical ventilation, intensive care unit, and hospitalization. Similarly, ECMO data is shown. The case fatality rate is all deaths per all known cases. Once again, data in black font is from the first several months of COVID. The data in red font is more contemporary, yet still pre-vaccination. The Johns Hopkins Coronavirus Resource Center estimates case fatality rate worldwide as 2.05% as of September 24, 2021. Now, the infection fatality rate is all deaths per all infected people, and is estimated rather than measured. It's estimated because of what I have identified here. A number of people are untested or did not present or died at home. However, were infected. This makes the denominator an estimate rather than a known number. Multiple clinical factors are a mortality risk for COVID-19. Older age is a risk for mortality. Chronic comorbidities increase the risk of death in COVID-19, and having multiple chronic comorbidities further increases that risk. Acute organ dysfunction, per se, is a risk for COVID-19. Ethnicity appears to be a risk factor for death from COVID-19, and within this, African-Americans have a higher mortality when compared to Caucasian-Americans. There may be a complex relationship between metabolic disorders, socioeconomic status, access to health care, and race that is not fully elucidated. In addition, the risk of death from COVID-19 appears to be a greater risk for death than does medical staff. And similarly, factors such as pregnancy, secondary infection, and proximal femur fracture, when present in COVID-19 patients, increases the risk of death compared to COVID-19 patients without these factors. Specifically, the risk of death from COVID-19 is higher than the risk of death from COVID-19. Finally, olfactory and gustatory disturbances have become a popular hallmark of the disease. However, they're interestingly associated with reduced mortality. Multiple factors can modify the risk of mortality for COVID-19. Geographic location is a factor because of the risk of death from COVID-19. The risk of death from COVID-19 is higher than the risk of death from COVID-19. Multiple factors can modify the risk of mortality for COVID-19. Geographic location is a factor because it is related, at least in part, to availability of critical care resources and predominance of strain type. Changes in mortality risk over time are likely reflective of a host of factors, including development of more rapid and more accurate testing, experience with providing care to infected patients, and changes in best practice. It appears that viral variants have differing mortality. Within the first several months of the pandemic, different haplotypes were identified with differing virulence. This graphic shows the spread of the L haplotype, which was concerning for a more aggressive clinical course and possibly more virulence. Current focus is on these variants and concern that the virus can adapt beyond the protection of current immunizations. Vaccination status, type, and time interval since vaccination affect mortality in that vaccination clearly reduces mortality among those who became infected, and this vaccine effectiveness varies slightly based on the type of vaccine and wanes with time. The effects of reinfection are not entirely clear in the existing literature. The supply and utilization of resources in the context of demand can have low or high effects on vaccine effectiveness. In general, these laboratory values have been predictive of higher mortality risk. These can be conceptually divided into lab values reflecting immune function, organ system dysfunction, and inflammation. No one lab is highly predictive of mortality, but their use is best exemplified in prognostic models, which I will speak about next. In the first few months of COVID, these were the state-of-the-art prognostic systems in addition to the list of comorbidities and predictive lab values I've shown previously. Prognostic models generally aim to use clinical data to predict outcomes. Now here, I've selected these reviews of prognostic models because they are prior to vaccination availability and thus have at least that homogeneity in the models they discuss. Focusing on reference 59 at the top, this review found that existing prognostic models were not particularly good. This included models that were created specifically for addressing COVID-19 mortality and excluded artificial intelligence, machine learning, and neural network-based models. Looking at reference 67 at the bottom, this review focused primarily on existing severity scoring systems such as MUSE, Apache 2, and so forth. This also demonstrated that there were few good predictors of mortality. Early warning scores are generally a type of prognostic model based on information available in an electronic medical record. As such, they can be based on large amounts of data and even modify in real time as a patient's condition changes. I wanted to show some representative early warning score systems here to demonstrate the EWS from pre-COVID and those developed in response to COVID are being discussed in the literature. To further exemplify how early warning scores can utilize big data, take this example study. The investigators used a tremendous amount of data totaling 2,863 years of observation time to build their EWS model. It is based on a time-varying neural cox model that accounts for risk factors changing over time and potential non-linear interactions between risk factors and COVID-19 related mortality risk, inclusive of 65 predictive factors. Importantly, early warning scores can be used to predict clinical deterioration, severity, and or mortality and are based on a sample population. Therefore, if using an EWS to inform decision making or resource allocation, a clinician needs to pay special attention to how an EWS has been created and evaluated in a study to appropriately apply it in their specific clinical context. In addition to the early warning scores that I just discussed, artificial intelligence represents a subset of the approaches to predicting mortality in COVID-19. Here is a review at the end of 2020 looking at 101 publications of which 17 studies evaluated mortality using artificial intelligence. Most prediction models here were trained on a specific localized dataset. Because of this, the evaluation of artificial intelligence techniques used and the importance of predictors cannot be discerned through meta-analysis. Also, the time frame from evaluation to death was a variable or not reported in these prediction models, so of course the area under the curve varies depending on what temporal values the prediction model holds. Interestingly, using artificial intelligence and specifically machine learning has allowed the integration of computer analysis of radiographic images into some of these prognostic models. Here is an example of the type of data processing that artificial intelligence can produce. In this study, more than 10,000 patient records were analyzed by multiple methods, and in this graphic the results of two of those methods are demonstrated. This shows how a large number of predictive factors for mortality can be weighted when analyzed from a large sample size. I have deliberately focused on discussing mortality prediction rather than severity prediction given the topic of this talk. However, there are obviously overlapping themes in predicting mortality and estimating severity. Scoring systems for severity and mortality have also been integrated into frameworks for ethical allocation of limited resources during the pandemic. When a clinician is using the literature to help assess the likelihood of mortality, several considerations arise. First, mortality rates and therefore mortality estimate tools are prone to error because of changing therapies over time, effective vaccination, virus genetic variation, and local prevalence of particular strains. Moreover, prediction models are based on specific samples with geographic, chronologic, ethnic, and socioeconomic factors. Additionally, allocation and or limitation of resources likely affects mortality in ways that are not easily integrated into prediction models. Finally, changes in goals of care and withdrawal of care based on patient values may affect sample mortality rates independent of the variables used in prediction models. Ultimately, it is the clinician's role to assess and integrate this information for the benefit of the patient. Thank you very much for the opportunity to present this topic on behalf of the SCCM.

mortality factors

COVID-19

clinical factors

laboratory values

modifiers