Multi-objective Evolutionary Discretization of Gene Expression Profiles: Application to COVID-19 Severity Prediction

Machine learning models can use information from gene expressions in patients to efficiently predict the severity of symptoms for several diseases. Medical experts, however, still need to understand the reasoning behind the predictions before trusting them. In their day-to-day practice, physicians prefer using gene expression profiles, consisting of a discretized subset of all data from gene expressions: in these profiles, genes are typically reported as either over-expressed or under-expressed, using discretization thresholds computed on data from a healthy control group. A discretized profile allows medical experts to quickly categorize patients at a glance. Building on previous works related to the automatic discretization of patient profiles, we present a novel approach that frames the problem as a multi-objective optimization task: on the one hand, after discretization, the medical expert would prefer to have as few different profiles as possible, to be able to classify patients in an intuitive way;on the other hand, the loss of information has to be minimized. Loss of information can be estimated using the performance of a classifier trained on the discretized gene expression levels. We apply one common state-of-the-art evolutionary multi-objective algorithm, NSGA-II, to the discretization of a dataset of COVID-19 patients that developed either mild or severe symptoms. The results show not only that the solutions found by the approach dominate traditional discretization based on statistical analysis and are more generally valid than those obtained through single-objective optimization, but that the candidate Pareto-optimal solutions preserve the sense-making that practitioners find necessary to trust the results. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

An evolutionary approach to the discretization of gene expression profiles to predict the severity of COVID-19

In this work, we propose to use a state-of-the-art evolutionary algorithm to set the discretization thresholds for gene expression profiles, using feedback from a classifier in order to maximize the accuracy of the predictions based on the discretized gene expression levels, while at the same time minimizing the number of different profiles obtained, to ease the understanding of the expert. The methodology is applied to a dataset containing COVID-19 patients that developed either mild or severe symptoms. The results show that the evolutionary approach performs better than a traditional discretization based on statistical analysis, and that it does preserve the sense-making necessary for practitioners to trust the results. © 2022 Owner/Author.

P.0492 Impact of mood and stress during COVID-19 lockdown on perceived immune fitness and the presence and severity of COVID-19 symptoms

Background: To address the rapid spread of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) around the world, in the absence of a vaccine or adequate treatment for the 2019 coronavirus disease (COVID-19), many governments around the world enforced lockdown periods. While lockdowns are beneficial in reducing the spread of the virus, literature shows that lockdowns may also have a significant negative impact on mood, wellbeing, and health. In this context, we hypothesized that negative mood changes increase overall stress levels, which in turn has a negative effect on perceived immune fitness, expressed in a greater presence and severity of COVID-19 symptoms [1]. Objective: The aim of this study was to investigate the impact of mood and stress during COVID-19 lockdown on perceived immune fitness and reported COVID-19 symptoms. Methods: An online retrospective survey was held among Dutch adults, to evaluate the first lockdown in The Netherlands [1]. Questions were answered for the period before the lockdown (15 January–14 March 2020) as well as for the lockdown period (15 March–11 May 2020). Mood was assessed via 1-item scales including “stress”, “anxiety”, “depression”, “fatigue”, “hostile”, “lonely” and “happy” [2]. Perceived immune fitness was assessed on a scale ranging from 0 (very poor) to 10 (excellent) [3]. The COVID-19 Symptoms Scale comprised the items sneezing, running nose, sore throat, cough, and malaise/feeling sick, high temperature (up to 38 Celsius), fever (38 Celsius and higher), shortness of breath, and chest pain [1]. The severity of each of the nine items could be rated as none (0), mild (1), moderate (2), or severe (3). The sum score of items served as COVID-19 symptom severity score, with a possible range from 0 (no complaints) to 27 (severe complaints). In addition, the presence of COVID-19 symptoms was calculated by counting the number of symptoms with a score > 0. Assessments before and during COVID19 lockdown were compared using the Related Samples Wilcoxon Signed Rank Test. For each variable, difference scores (Δ, lockdown – before lockdown) were calculated. Pearson's correlations were calculated between difference scores. Results: Data of 1415 Dutch adults (64% women, age range: 18 to 94 years old) were analyzed. During lockdown, all mood ratings, stress, and perceived immune fitness were significantly poorer compared to the period before lockdown (p<0.0001). The COVID-19 symptom severity score as significantly higher during the lockdown period (p=0.018), but no difference was found for the number of reported COVID-19 symptoms (p=0.256). Significant correlations were found between mood changes and Δ stress (r ranged from 0.334 to 0.557, all p<0.0001). The correlation between Δ stress and Δ perceived immune fitness was also significant (r=-0.310, p<0.0001). Finally, Δ perceived immune fitness correlated significantly with difference scores for the presence (r=-0.223, p<0.0001) and severity (r=-0.245, p<0.0001) of COVID-19 related symptoms. Conclusions: The findings support the hypothesis that lockdown has a negative effect on mood and increases stress. This is reflected in poorer perceived immune fitness, which in turn is associated with a greater presence and severity of COVID-19 symptoms. No conflict of interest

P.0581 The economic costs of perceived reduced immune fitness in the Netherlands

Background: Immune fitness has been defined as a state where an individual's immune system is resilient, having an inbuilt capacity to adapt to challenges by establishing, maintaining, and regulating an appropriate immune response in order to promote health and prevent and resolve disease [1]. Perceived reduced immune fitness can be a possible reason for absenteeism (staying home from work) or presenteeism (attending work despite health problems, possibly with reduced work performance). Objective: The purpose of the current study was to investigate the impact of reduced immune fitness on absenteeism and presenteeism at work, and the estimated associated costs for the Dutch economy. Methods: Dutch adults were recruited via Facebook to complete an online survey [2]. Participants could indicate whether in 2019 (i.e. pre-COVID-19) they were employed (owner or employee). Absenteeism and presenteeism related to perceived reduced immune fitness were assessed. To this end, questions were adapted from a recent study examining the cost of workplace hangovers and intoxication to the UK economy [3]. Questions concerned the number of days in 2019 that participants (a) did not work because they experienced reduced immune fitness and (b) did work although they experienced reduced immune fitness. With regard to presenteeism, they could further indicate, in comparison to a regular working day, how well they performed at work on days when they experienced reduced immune fitness. This was done by rating their performance on a scale ranging from 0% (compared to a regular day I achieved nothing/did not work) to 100% (my work was absolutely not influenced by experiencing reduced immune fitness). Statistics Netherlands provided information on the average Dutch income. Perceived immune fitness was assessed with the immune status questionnaire [1]. In line with Bhattacharya's method [3], the economic costs of absenteeism and presenteeism due to reduced immune fitness were calculated. Results: Data of N=466 participants with a mean (SD) age of 37 (15.2) years old (range: 18-65 years old), 67.4% female, was evaluated. Overall, 4.4 days of absenteeism, 27.1 days of presenteeism, and a 21% reduction of performance level were reported for presenteeism days. Females, higher educated, and older participants reported significantly higher rates of absenteeism and presenteeism, and lower performance levels on days working with reduced immune fitness. Significant correlations (p<0.05) were found between perceived immune fitness and the number of absenteeism (r = -0.350) and presenteeism days (r = -0.339). The estimated economic cost of absenteeism (€6.8± 0.064 billion euro) and presenteeism (€8.8 ± 0.083 billion euro) sum up to a total cost of €15.6± 0.147 billion euro. Conclusions: Perceived reduced immune fitness has a significant negative impact on work performance, expressed in both absenteeism and presenteeism. Based on the present sample, the associated annual costs for the Dutch economy for 2019 were estimated at 15.6 billion Euro. These high costs warrant further investigation. A large nationally representative sample should be conducted to verify these findings and yield a more accurate estimate of the associated economic costs of perceived reduced immune fitness. No conflict of interest

P.0444 The use of prescription analgesics during COVID-19 lockdown in the Netherlands

Background: The 2019 coronavirus disease (COVID-19) poses great demands on hospitals and resulted in delayed care. It can be hypothesized that this delayed professional health care is reflected in a reduction of pharmacy dispensed analgesics. Objective: The purpose of this study was to investigate the possible impact of COVID-19 lockdown on the dispensing rates of analgesics drugs by Dutch pharmacies. Methods: Pharmacy dispensing data was obtained from Stichting Farmaceutische Kentallen on newly dispended analgesics (ATC2 N02), covering 1890 pharmacies and 96% of the Dutch population. Over-the-counter (OTC) painkillers were not considered. Age and sex of the patient were recorded, as well as the week of first-time dispensed drugs. Data were collected for the first half year of 2019 and the first half year of 2020. Patients were allocated to one of the following age groups: children (0-9 years old), adolescents (10-19 years old), adults (20-64 years old), and elderly (65 year and older). Further, data was combined into three time periods: week 1-11 (for 2020 corresponding to the pre-COVID-19 lockdown period), week 12-19 (for 2020 corresponding to the COVID-19 lockdown), and week 20-26 (for 2020 corresponding to the post-COVID-19 lockdown). Using independent t-tests, dispensing data was compared between males and females, age groups, and between 2019 and 2020. Results: An overall reduction was observed in the first half of 2020 for the number of newly dispensed analgesics (379.657 in 2019 and 360.094 in 2020, respectively). In both years, analgesics were significantly more often dispensed to females than to males, and most were prescribed to adults followed by the elderly. In addition, significantly more analgesics were dispensed in week 1-11 compared to the other time periods. In male children, compared to 2019 the reduction was only significant during the COVID-19 lockdown in 2020. For female children, no significant differences were found. For adolescents of both sexes, compared to 2019, a significantly reduced number of dispensed analgesics was found for all three time periods (p<0.001). In adults of both sexes, the reduction in dispensed analgesics in 2020 was only significant during the COVID-19 lockdown. In elderly, a reduction in dispensed analgesics in 2020 was found for pre- and during the COVID-19 lockdown (p<0.001), but not post-COVID-19 lockdown. In elderly males, a significant reduction in 2020 was found for pre-COVID lockdown, but not thereafter. For elderly females, the reduction in 2020 was significant for both week pre- (p<0.001) and during COVID-19 lockdown (p=0.02), but not thereafter. Conclusions: These findings show that the number of dispensed analgesics was significantly lower in 2020 compared to 2019. The reduction in dispensed analgesics was most consistently seen for the 2020 COVID-19 lockdown period, reflecting delayed professional health care across all age groups. Future research should investigate whether, due to delayed care, a substantial number of patients may have suffered from untreated pain complaints during the 2020 COVID-19 lockdown, or whether they have established pain relief via non-prescription painkillers. No conflict of interest

Design of specific primer sets for SARS-CoV-2 variants using evolutionary algorithms

Primer sets are short DNA sequences of 18-22 base pairs, that can be used to verify the presence of a virus, and designed to attach to a specific part of a viral DNA. Designing a primer set requires choosing a region of DNA, avoiding the possibility of hybridization to a similar sequence, as well as considering its GC content and Tm (melting temperature). Coronaviruses, such as SARS-CoV-2, have a considerably large genome (around 30 thousand nucleotides) when compared to other viruses. With the rapid rise and spread of SARS-CoV-2 variants, it has become a priority to breach our lack of specific primers available for diagnosis of this new variants. Here, we propose an evolutionary-based approach to primer design, able to rapidly deliver a high-quality primer set for a target sequence of the virus variant. Starting from viral sequences collected from open repositories, the proposed approach is proven able to uncover a specific primer set for the B.1.1.7 SARS-CoV-2 variant. Only recently identified, B.1.1.7 is already considered potentially dangerous, as it presents a considerably higher transmissibility when compared to other variants. © 2021 ACM.