A novel scale based on biomarkers associated with COVID-19 severity can predict the need for hospitalization and intensive care, as well as enhanced probabilities for mortality.

Prognostic scales may help to optimize the use of hospital resources, which may be of prime interest in the context of a fast spreading pandemics. Nonetheless, such tools are underdeveloped in the context of COVID-19. In the present article we asked whether accurate prognostic scales could be developed to optimize the use of hospital resources. We retrospectively studied 467 files of hospitalized patients after COVID-19. The odds ratios for 16 different biomarkers were calculated, those that were significantly associated were screened by a Pearson's correlation, and such index was used to establish the mathematical function for each marker. The scales to predict the need for hospitalization, intensive-care requirement and mortality had enhanced sensitivities (0.91 CI 0.87-0.94; 0.96 CI 0.94-0.98; 0.96 CI 0.94-0.98; all with p < 0.0001) and specificities (0.74 CI 0.62-0.83; 0.92 CI 0.87-0.96 and 0.91 CI 0.86-0.94; all with p < 0.0001). Interestingly, when a different population was assayed, these parameters did not change considerably. These results show a novel approach to establish the mathematical function of a marker in the development of highly sensitive prognostic tools, which in this case, may aid in the optimization of hospital resources. An online version of the three algorithms can be found at: http://benepachuca.no-ip.org/covid/index.php.

SARS-CoV-2 airborne infection probability estimated by using indoor carbon dioxide.

Airborne transmission is one of the main routes of SARS-CoV-2 spread. It is important to determine the circumstances under which the risk of airborne transmission is increased as well as the effective strategy to reduce such risk. This study aimed to develop a modified version of the Wells-Riley model with indoor CO2 to estimate the probability of airborne transmission of SARS-CoV-2 Omicron strains with a CO2 monitor and to evaluate the validity of this model in actual clinical practices. We used the model in three suspected cases of airborne transmission presented to our hospital to confirm its validity. Next, we estimated the required indoor CO2 concentration at which R0 does not exceed 1 based on the model. The estimated R0 (R0, basic reproduction number) based on the model in each case were 3.19 in three out of five infected patients in an outpatient room, 2.00 in two out of three infected patients in the ward, and 0.191 in none of the five infected patients in another outpatient room. This indicated that our model can estimate R0 with an acceptable accuracy. In a typical outpatient setting, the required indoor CO2 concentration at which R0 does not exceed 1 is below 620 ppm with no mask, 1000 ppm with a surgical mask and 16000 ppm with an N95 mask. In a typical inpatient setting, on the other hand, the required indoor CO2 concentration is below 540 ppm with no mask, 770 ppm with a surgical mask, and 8200 ppm with an N95 mask. These findings facilitate the establishment of a strategy for preventing airborne transmission in hospitals. This study is unique in that it suggests the development of an airborne transmission model with indoor CO2 and application of the model to actual clinical practice. Organizations and individuals can efficiently recognize the risk of SARS-CoV-2 airborne transmission in a room and thus take preventive measures such as maintaining good ventilation, wearing masks, or shortening the exposure time to an infected individual by simply using a CO2 monitor.

Face mask is an efficient tool to fight the Covid-19 pandemic and some factors increase the probability of its adoption.

This study examines the dynamic impact of face mask use on both infected cases and fatalities at a global scale by using a rich set of panel data econometrics. An increase of 100% of the proportion of people declaring wearing a mask (multiply by two) over the studied period lead to a reduction of around 12 and 13.5% of the number of Covid-19 infected cases (per capita) after 7 and 14 days respectively. The delay of action varies from around 7 days to 28 days concerning infected cases but is more longer concerning fatalities. Our results hold when using the rigorous controlling approach. We also document the increasing adoption of mask use over time and the drivers of mask adoption. In addition, population density and pollution levels are significant determinants of heterogeneity regarding mask adoption across countries, while altruism, trust in government and demographics are not. However, individualism index is negatively correlated with mask adoption. Finally, strict government policies against Covid-19 have a strong significant effect on mask use.

How do psychobehavioural variables shed light on heterogeneity in COVID-19 vaccine acceptance? Evidence from United States general population surveys on a probability panel and social media.

OBJECTIVES: To (1) understand what behaviours, beliefs, demographics and structural factors predict US adults' intention to get a COVID-19 vaccination, (2) identify segments of the population ('personas') who share similar factors predicting vaccination intention, (3) create a 'typing tool' to predict which persona people belong to and (4) track changes in the distribution of personas over time and across the USA. DESIGN: Three surveys: two on a probability-based household panel (NORC's AmeriSpeak) and one on Facebook. SETTING: The first two surveys were conducted in January 2021 and March 2021 when the COVID-19 vaccine had just been made available in the USA. The Facebook survey ran from May 2021 to February 2022. PARTICIPANTS: All participants were aged 18+ and living in the USA. OUTCOME MEASURES: In our predictive model, the outcome variable was self-reported vaccination intention (0-10 scale). In our typing tool model, the outcome variable was the five personas identified by our clustering algorithm. RESULTS: Only 1% of variation in vaccination intention was explained by demographics, with about 70% explained by psychobehavioural factors. We identified five personas with distinct psychobehavioural profiles: COVID Sceptics (believe at least two COVID-19 conspiracy theories), System Distrusters (believe people of their race/ethnicity do not receive fair healthcare treatment), Cost Anxious (concerns about time and finances), Watchful (prefer to wait and see) and Enthusiasts (want to get vaccinated as soon as possible). The distribution of personas varies at the state level. Over time, we saw an increase in the proportion of personas who are less willing to get vaccinated. CONCLUSIONS: Psychobehavioural segmentation allows us to identify why people are unvaccinated, not just who is unvaccinated. It can help practitioners tailor the right intervention to the right person at the right time to optimally influence behaviour.

Spatiotemporal transmission of infectious particles in environment: A case study of Covid-19.

This study explores the dynamic transmission of infectious particles due to COVID-19 in the environment using a spatiotemporal epidemiological approach. We proposed a novel multi-agent model to simulate the spread of COVID-19 by considering several influencing factors. The model divides the population into susceptible and infected and analyzes the impact of different prevention and control measures, such as limiting the number of people and wearing masks on the spread of COVID-19. The findings suggest that reducing population density and wearing masks can significantly reduce the likelihood of virus transmission. Specifically, the research shows that if the population moves within a fixed range, almost everyone will eventually be infected within 1 h. When the population density is 50%, the infection rate is as high as 96%. If everyone does not wear a mask, nearly 72.33% of the people will be infected after 1 h. However, when people wear masks, the infection rate is consistently lower than when they do not wear masks. Even if only 25% of people wear masks, the infection rate with masks is 27.67% lower than without masks, which is strong evidence of the importance of wearing a mask. As people's daily activities are mostly carried out indoors, and many super-spreading events of the new crown epidemic also originated from indoor gatherings, the research on indoor epidemic prevention and control is essential. This study provides decision-making support for epidemic preventions and controls and the proposed methodology can be used in other regions and future epidemics.

[Prediction of protection probability against Omicron BA.1, BA.4 and BA.5 variants in symptomatic infections with prototype strain based on neutralization antibody levels].

Objective: To predict the protection probability of different clinical outcomes after reinfection with Omicron variant in symptomatic and unvaccinated COVID-19 patients who infected with prototype strain. Methods: The data used in this study were derived from a systematic review and meta-analysis which systematically searched PubMed, Embase, Web of Science, and Europe PMC databases, included published and uploaded studies of dynamic changes of neutralizing antibodies in symptomatic COVID-19 patients from 1 January 2020 to 2 October 2022 and extracted the literature information, study design, serological experiment information and antibody results. According to the scatter distribution characteristics of antibody titer data, a generalized additive model based on Gaussian distribution was used to fit the titer value of neutralizing antibody based on logarithmic conversion and the dynamic change pattern of neutralizing antibody in symptomatic and unvaccinated COVID-19 patients infected with prototype strain over time was obtained. In this study, the fitted antibody titers of patients on the 28th, 51st, and 261st day after symptom onset was selected to predict the protection probability. Results: Neutralizing antibodies produced in symptomatic and unvaccinated patients infected with prototype strain could provide protection against Omicron reinfection, and the probability of protection gradually decreased with time. Neutralizing antibody level on day 28 after symptom onset provided protection probability of 30.3% (95%CI: 20.0%-45.5%) against reinfection, 51.5% (95%CI: 33.4%-75.9%) against symptomatic reinfection, and 91.2% (95%CI: 77.1%-97.7%) against severe reinfection caused by Omicron BA.5. The protection probability against Omicron BA.1, BA.4 and BA.5 reinfections decreased significantly 261 days after symptom onset, showing 9.6%-12.9%, 18.4%-23.9% and 63.1%-70.3% against three clinical outcomes, respectively. At the same time point and against the same clinical outcome, the protection probability of BA.1 was the highest, followed by BA.4 and BA.5. Conclusions: Neutralizing antibodies induced in symptomatic and unvaccinated COVID-19 patients previously infected with the prototype strain have limited protection probability against Omicron BA.5 reinfections and symptomatic reinfections. The protection probability against Omicron BA.5 reinfections is 30.3% 28 days after symptom onset and decreases to about 10% after 261 days. However, the protection probability against severe reinfections is considerable, with over 90% 28 days after symptom onset and still exceeding 60% after 261 days.

An age-of-infection model with both symptomatic and asymptomatic infections.

We formulate a general age-of-infection epidemic model with two pathways: the symptomatic infections and the asymptomatic infections. We then calculate the basic reproduction number [Formula: see text] and establish the final size relation. It is shown that the ratio of accumulated counts of symptomatic patients and asymptomatic patients is determined by the symptomatic ratio f which is defined as the probability of eventually becoming symptomatic after being infected. We also formulate and study a general age-of-infection model with disease deaths and with two infection pathways. The final size relation is investigated, and the upper and lower bounds for final epidemic size are given. Several numerical simulations are performed to verify the analytical results.

Temporal dynamics in mental health symptoms and loneliness during the COVID-19 pandemic in a longitudinal probability sample: a network analysis.

Figuring out which symptoms are central for symptom escalation during the COVID-19 pandemic is important for targeting prevention and intervention. Previous studies have contributed to the understanding of the course of psychological distress during the pandemic, but less is known about key symptoms of psychological distress over time. Going beyond a pathogenetic pathway perspective, we applied the network approach to psychopathology to examine how psychological distress unfolds in a period of maximum stress (pre-pandemic to pandemic onset) and a period of repeated stress (pandemic peak to pandemic peak). We conducted secondary data analyses with the Understanding Society data (N = 17,761), a longitudinal probability study in the UK with data before (2019), at the onset of (April 2020), and during the COVID-19 pandemic (November 2020 & January 2021). Using the General Health Questionnaire and one loneliness item, we computed three temporal cross-lagged panel network models to analyze psychological distress over time. Specifically, we computed (1) a pre-COVID to first incidence peak network, (2) a first incidence peak to second incidence peak network, and (3) a second incidence peak to third incidence peak network. All networks were highly consistent over time. Loneliness and thinking of self as worthless displayed a high influence on other symptoms. Feeling depressed and not overcoming difficulties had many incoming connections, thus constituting an end-product of symptom cascades. Our findings highlight the importance of loneliness and self-worth for psychological distress during COVID-19, which may have important implications in therapy and prevention. Prevention and intervention measures are discussed, as single session interventions are available that specifically target loneliness and worthlessness to alleviate mental health problems.

Association between prolonged corticosteroids use in COVID-19 and increased mortality in hospitalized patients: a retrospective study with inverse probability of treatment weighting analysis.

BACKGROUND: Previous studies have demonstrated a beneficial effect of early use of corticosteroids in patients with COVID-19. This study aimed to compare hospitalized patients with COVID-19 who received short-course corticosteroid treatment with those who received prolonged-course corticosteroid treatment to determine whether prolonged use of corticosteroids improves clinical outcomes, including mortality. METHODS: This is a retrospective cohort study including adult patients with positive testing for Sars-CoV-2 hospitalized for more than 10 days. Data were obtained from electronic medical records. Patients were divided into two groups, according to the duration of treatment with corticosteroids: a short-course (10 days) and a prolonged-course (longer than 10 days) group. Inverse probability treatment weighting (IPTW) analysis was used to evaluate whether prolonged use of corticosteroids improved outcomes. The primary outcome was in-hospital mortality. Secondary outcomes were hospital infection and the association of different doses of corticosteroids with hospital mortality. Restricted cubic splines were used to assess the nonlinear association between mortality and dose and duration of corticosteroids use. RESULTS: We enrolled 1,539 patients with COVID-19. Among them, 1127 received corticosteroids for more than 10 days (prolonged-course group). The in-hospital mortality was higher in patients that received prolonged course corticosteroids (39.5% vs. 26%, p < 0.001). The IPTW revealed that prolonged use of corticosteroids significantly increased mortality [relative risk (RR) = 1.52, 95% confidence interval (95% CI): 1.24-1.89]. In comparison to short course treatment, the cubic spline analysis showed an inverted U-shaped curve for mortality, with the highest risk associated with the prolonged use at 30 days (RR = 1.50, 95% CI 1.21-1.78). CONCLUSIONS: Prolonged course of treatment with corticosteroids in hospitalized patients with COVID-19 was associated with higher mortality.

Towards robust and accurate estimates of the incubation time distribution, with focus on upper tail probabilities and SARS-CoV-2 infection.

Quarantine length for individuals who have been at risk for infection with SARS-CoV-2 has been based on estimates of the incubation time distribution. The time of infection is often not known exactly, yielding data with an interval censored time origin. We give a detailed account of the data structure, likelihood formulation and assumptions usually made in the literature: (i) the risk of infection is assumed constant on the exposure window and (ii) the incubation time follows a specific parametric distribution. The impact of these assumptions remains unclear, especially for the right tail of the distribution which informs quarantine policy. We quantified bias in percentiles by means of simulation studies that mimic reality as close as possible. If assumption (i) is not correct, then median and upper percentiles are affected similarly, whereas misspecification of the parametric approach (ii) mainly affects upper percentiles. The latter may yield considerable bias. We suggest a semiparametric method that provides more robust estimates without the need of a parametric choice. Additionally, we used a simulation study to evaluate a method that has been suggested if all infection times are left censored. It assumes that the width of the interval from infection to latest possible exposure follows a uniform distribution. This assumption gave biased results in the exponential phase of an outbreak. Our application to open source data suggests that focus should be on the level of information in the observations, as expressed by the width of exposure windows, rather than the number of observations.

Uncertainty Quantification for Epidemic Risk Management: Case of SARS-CoV-2 in Morocco.

In this paper, we propose a new method for epidemic risk modelling and prediction, based on uncertainty quantification (UQ) approaches. In UQ, we consider the state variables as members of a convenient separable Hilbert space, and we look for their representation in finite dimensional subspaces generated by truncations of a suitable Hilbert basis. The coefficients of the finite expansion can be determined by approaches established in the literature, adapted to the determination of the probability distribution of epidemic risk variables. Here, we consider two approaches: collocation (COL) and moment matching (MM). Both are applied to the case of SARS-CoV-2 in Morocco, as an epidemic risk example. For all the epidemic risk indicators computed in this study (number of detections, number of deaths, number of new cases, predictions and human impact probabilities), the proposed models were able to estimate the values of the state variables with precision, i.e., with very low root mean square errors (RMSE) between predicted values and observed ones. Finally, the proposed approaches are used to generate a decision-making tool for future epidemic risk management, or, more generally, a quantitative disaster management approach in the humanitarian supply chain.

Score to assess the probability of relapse in granulomatosis with polyangiitis and microscopic polyangiitis.

OBJECTIVE: To develop a score assessing the probability of relapse in granulomatosis with polyangiitis (GPA) and microscopic polyangiitis (MPA). METHODS: Long-term follow-up data from GPA and MPA patients included in five consecutive randomised controlled trials were pooled. Patient characteristics at diagnosis were entered into a competing-risks model, with relapse as the event of interest and death the competing event. Univariate and multivariate analyses were computed to identify variables associated with relapse and build a score, which was then validated in an independent cohort of GPA or MPA patients. RESULTS: Data collected from 427 patients (203 GPA, 224 MPA) at diagnosis were included. Mean±SD follow-up was 80.6±51.3 months; 207 (48.5%) patients experienced ≥1 relapse. Relapse risk was associated with proteinase 3 (PR3) positivity (HR=1.81 (95% CI 1.28 to 2.57); p<0.001), age ≤75 years (HR=1.89 (95% CI 1.15 to 3.13); p=0.012) and estimated glomerular filtration rate (eGFR) ≥30 mL/min/1.73 m² (HR=1.67 (95% CI 1.18 to 2.33); p=0.004) at diagnosis. A score, the French Vasculitis Study Group Relapse Score (FRS), from 0 to 3 points was modelised: 1 point each for PR3-antineutrophil cytoplasmic antibody positivity, eGFR ≥30 mL/min/1.73 m² and age ≤75 years. In the validation cohort of 209 patients, the 5-year relapse risk was 8% for a FRS of 0, 30% for 1, 48% for 2 and 76% for 3. CONCLUSION: The FRS can be used at diagnosis to assess the relapse risk in patients with GPA or MPA. Its value for tailoring the duration of maintenance therapy should be evaluated in future prospective trials.

The effect of variation of individual infectiousness on SARS-CoV-2 transmission in households.

Quantifying variation of individual infectiousness is critical to inform disease control. Previous studies reported substantial heterogeneity in transmission of many infectious diseases including SARS-CoV-2. However, those results are difficult to interpret since the number of contacts is rarely considered in such approaches. Here, we analyze data from 17 SARS-CoV-2 household transmission studies conducted in periods dominated by ancestral strains, in which the number of contacts was known. By fitting individual-based household transmission models to these data, accounting for number of contacts and baseline transmission probabilities, the pooled estimate suggests that the 20% most infectious cases have 3.1-fold (95% confidence interval: 2.2- to 4.2-fold) higher infectiousness than average cases, which is consistent with the observed heterogeneity in viral shedding. Household data can inform the estimation of transmission heterogeneity, which is important for epidemic management.

Spatio-Temporal Bayesian Models for Malaria Risk Using Survey and Health Facility Routine Data in Rwanda.

INTRODUCTION: Malaria is a life-threatening disease ocuring mainly in developing countries. Almost half of the world's population was at risk of malaria in 2020. Children under five years age are among the population groups at considerably higher risk of contracting malaria and developing severe disease. Most countries use Demographic and Health Survey (DHS) data for health programs and evaluation. However, malaria elimination strategies require a real-time, locally-tailored response based on malaria risk estimates at the lowest administrative levels. In this paper, we propose a two-step modeling framework using survey and routine data to improve estimates of malaria risk incidence in small areas and enable quantifying malaria trends. METHODS: To improve estimates, we suggest an alternative approach to modeling malaria relative risk by combining information from survey and routine data through Bayesian spatio-temporal models. We model malaria risk using two steps: (1) fitting a binomial model to the survey data, and (2) extracting fitted values and using them in the Poison model as nonlinear effects in the routine data. We modeled malaria relative risk among under-five-year old children in Rwanda. RESULTS: The estimation of malaria prevalence among children who are under five years old using Rwanda demographic and health survey data for the years 2019-2020 alone showed a higher prevalence in the southwest, central, and northeast of Rwanda than the rest of the country. Combining with routine health facility data, we detected clusters that were undetected based on the survey data alone. The proposed approach enabled spatial and temporal trend effect estimation of relative risk in local/small areas in Rwanda. CONCLUSIONS: The findings of this analysis suggest that using DHS combined with routine health services data for active malaria surveillance may provide provide more precise estimates of the malaria burden, which can be used toward malaria elimination targets. We compared findings from geostatistical modeling of malaria prevalence among under-five-year old children using DHS 2019-2020 and findings from malaria relative risk spatio-temporal modeling using both DHS survey 2019-2020 and health facility routine data. The strength of routinely collected data at small scales and high-quality data from the survey contributed to a better understanding of the malaria relative risk at the subnational level in Rwanda.

A comparison of two frameworks for multi-state modelling, applied to outcomes after hospital admissions with COVID-19.

We compare two multi-state modelling frameworks that can be used to represent dates of events following hospital admission for people infected during an epidemic. The methods are applied to data from people admitted to hospital with COVID-19, to estimate the probability of admission to intensive care unit, the probability of death in hospital for patients before and after intensive care unit admission, the lengths of stay in hospital, and how all these vary with age and gender. One modelling framework is based on defining transition-specific hazard functions for competing risks. A less commonly used framework defines partially-latent subpopulations who will experience each subsequent event, and uses a mixture model to estimate the probability that an individual will experience each event, and the distribution of the time to the event given that it occurs. We compare the advantages and disadvantages of these two frameworks, in the context of the COVID-19 example. The issues include the interpretation of the model parameters, the computational efficiency of estimating the quantities of interest, implementation in software and assessing goodness of fit. In the example, we find that some groups appear to be at very low risk of some events, in particular intensive care unit admission, and these are best represented by using 'cure-rate' models to define transition-specific hazards. We provide general-purpose software to implement all the models we describe in the flexsurv R package, which allows arbitrarily flexible distributions to be used to represent the cause-specific hazards or times to events.

Acceptance of COVID-19 vaccine among sub-Saharan Africans (SSA): a comparative study of residents and diasporan dwellers.

BACKGROUND: The COVID-19 vaccines are being rolled out across all the sub-Saharan Africa (SSA) countries, with countries setting targets for achieving full vaccination rates. The aim of this study was to compare the uptake of, resistance and hesitancy to the COVID-19 vaccine between SSA locally residents and in the diasporan dwellers. METHODS: This was a cross-sectional study conducted using a web and paper-based questionnaire to obtain relevant information on COVID-19 vaccine acceptance. The survey items included questions on demography, uptake and planned acceptance or non-acceptance of the COVID-19 vaccines among SSAs. Multinomial logistic regression was used to determine probabilities of outcomes for factors associated with COVID-19 vaccination resistance and hesitancy among SSA respondents residing within and outside Africa. RESULTS: Uptake of COVID-19 vaccines varied among the local (14.2%) and diasporan (25.3%) dwellers. There were more locals (68.1%) who were resistant to COVID-19 vaccine. Participants' sex [adjusted relative risk (ARR) = 0.73, 95% CI: 0.58 - 0.93], education [primary/less: ARR = 0.22, CI:0.12 - 0.40, and bachelor's degree: ARR = 0.58, CI: 0.43 - 0.77]), occupation [ARR = 0.32, CI: 0.25-0.40] and working status [ARR = 1.40, CI: 1.06-1.84] were associated with COVID-19 vaccine resistance among locals. Similar proportion of local and diasporan dwellers (~ 18% each) were hesitant to COVID-19 vaccine, and this was higher among health care workers [ARR = 0.25, CI: 0.10 - 0.62 and ARR = 0.24, CI:0.18-0.32, diaspora and locals respectively]. After adjusting for the potential confounders, local residents aged 29-38 years [ARR = 1.89, CI: 1.26-2.84] and lived in East Africa [ARR = 4.64, CI: 1.84-11.70] were more likely to report vaccine hesitancy. Knowledge of COVID vaccines was associated with hesitancy among local and diasporan dwellers, but perception was associated with vaccine resistance [ARR = 0.86,CI: 0.82 - 0.90] and hesitancy [ARR = 0.85, CI: 0.80 - 0.90], only among the local residents. CONCLUSIONS: Differences exist in the factors that influence COVID-19 vaccine acceptance between local SSA residents and thediasporan dwellers. Knowledge about COVID-19 vaccines affects the uptake, resistance, and hesitancy to the COVID-19 vaccine. Information campaigns focusing on the efficacy and safety of vaccines could lead to improved acceptance of COVID-19 vaccines.

Likelihood-Free Dynamical Survival Analysis applied to the COVID-19 epidemic in Ohio.

The Dynamical Survival Analysis (DSA) is a framework for modeling epidemics based on mean field dynamics applied to individual (agent) level history of infection and recovery. Recently, the Dynamical Survival Analysis (DSA) method has been shown to be an effective tool in analyzing complex non-Markovian epidemic processes that are otherwise difficult to handle using standard methods. One of the advantages of Dynamical Survival Analysis (DSA) is its representation of typical epidemic data in a simple although not explicit form that involves solutions of certain differential equations. In this work we describe how a complex non-Markovian Dynamical Survival Analysis (DSA) model may be applied to a specific data set with the help of appropriate numerical and statistical schemes. The ideas are illustrated with a data example of the COVID-19 epidemic in Ohio.

A dual graph neural network for drug-drug interactions prediction based on molecular structure and interactions.

Expressive molecular representation plays critical roles in researching drug design, while effective methods are beneficial to learning molecular representations and solving related problems in drug discovery, especially for drug-drug interactions (DDIs) prediction. Recently, a lot of work has been put forward using graph neural networks (GNNs) to forecast DDIs and learn molecular representations. However, under the current GNNs structure, the majority of approaches learn drug molecular representation from one-dimensional string or two-dimensional molecular graph structure, while the interaction information between chemical substructure remains rarely explored, and it is neglected to identify key substructures that contribute significantly to the DDIs prediction. Therefore, we proposed a dual graph neural network named DGNN-DDI to learn drug molecular features by using molecular structure and interactions. Specifically, we first designed a directed message passing neural network with substructure attention mechanism (SA-DMPNN) to adaptively extract substructures. Second, in order to improve the final features, we separated the drug-drug interactions into pairwise interactions between each drug's unique substructures. Then, the features are adopted to predict interaction probability of a DDI tuple. We evaluated DGNN-DDI on real-world dataset. Compared to state-of-the-art methods, the model improved DDIs prediction performance. We also conducted case study on existing drugs aiming to predict drug combinations that may be effective for the novel coronavirus disease 2019 (COVID-19). Moreover, the visual interpretation results proved that the DGNN-DDI was sensitive to the structure information of drugs and able to detect the key substructures for DDIs. These advantages demonstrated that the proposed method enhanced the performance and interpretation capability of DDI prediction modeling.