Early recognition of high-risk and critically ill patients has become a priority in improving treatment and reducing mortality among patients who contracted SARS-CoV-2². In this study, we examined the impact of multiple long-term conditions on the in-hospital mortality in individuals with COVID-19 to determine risk factors for COVID-19-related deaths. Our findings can help improve the effectiveness of management of COVID-19 patients and contribute to further development of policies for prevention and response to COVID-19 and its critical outcomes. Using a large, well-documented, regional cohort of hospitalised COVID-19 patients, we found that pre-existing health conditions, including obesity, liver disease, renal disease, metastatic cancer, solid tumour without metastasis, lymphoma, congestive heart failure, chronic obstructive pulmonary disease, and dementia are clinical risk factors associated with COVID-19 mortality, with chronic obstructive pulmonary disease, renal disease, and dementia being the most prevalent among those that died.

Our results are consistent with several studies^{4,5,7,19,20,21,22}. Chronic obstructive pulmonary disease (COPD) was found to increase the odds of death by nearly threefold in a large meta-analysis of 30 studies that examined the vital status of COVID-19 patients with COPD¹⁹. Substantial mortality rates in COVID-19 patients with COPD were also observed by other studies^20,21. It has been suggested that the association between COPD and risk of poor outcomes in COVID-19 might be related to the fact that the innate and acquired antiviral immune responses in individuals with COPD are impaired, leading to delayed virus clearance¹⁹. Dementia was identified as a major risk factor for death in COVID‐19 cases^22,23. Wang et al.²² showed that the odds of COVID-19-related death in patients with dementia doubled when compared to individuals without dementia, with the highest mortality risk in adults with vascular dementia (OR, 3.17; 95% CI, 2.97–3.37, P < 0.001). Some evidence suggested that elevated risk of neurological complications from COVID-19 in people with dementia might be caused by the pre-existing brain pathology²⁴. For example, the breakdown of the blood–brain barrier, i.e., a defence mechanism against disease-causing pathogens, in patients with Alzheimer’s disease and vascular dementia, can increase the ability of bacterial, fungal, and viral pathogens to access the brain more easily and this in turn may have an effect on the severity of COVID-19 and associated fatal outcomes^25,26. A number of studies investigated the impact of chronic diseases and health conditions on risk of COVID-19-related death through multivariate analyses^{4,5,7,8,9,10,27}; however, some of them were based on a small sample size, included a limited list of underlying medical conditions or focused on the impact of a specific medical condition on COVID-19 mortality adjusted for demographic and/or socioeconomic characteristics^7,9,10. Moreover, previous evidence on the relationship between multi morbidity and COVID-19-related death was limited^4,5. A UK prospective observational study of 20,133 patients who were hospitalised with COVID-19 showed that the risk of death was higher for patients with dementia, chronic pulmonary disease, kidney disease, cardiac disease, liver disease, malignancy, and obesity⁵. Another study based on 10,926 COVID-19-related deaths reported similar results: individuals with COVID-19 and underlying kidney disease, liver disease, cardiovascular disease, chronic respiratory diseases, obesity, and recent history of haematological malignancy or other cancers had a greater risk of dying²⁷. Cancer, possibly due to its ability to cause immunodeficiency inherently or through medication, was identified as a major risk factor for COVID-19-related deaths in several studies^4,5,27. In individuals with heart failure and kidney disease, both the SARS-CoV-2 infection and the immune response to the viral infection could destabilize pre-existing conditions, leading to the development of acute cardiac²⁸ or kidney²⁹ injuries and hence, increase the risk of a fatal outcome associated with COVID-19³⁰. People with obesity have been previously characterized by systemic low-grade inflammation, impaired immune response to infections, and higher susceptibility and mortality associated with infections^17,31,32. These factors may all lead to a greater mortality risk in those who contracted SARS-CoV-2. Several studies emerging from different countries identified obesity as an independent risk factor for hospitalisation and death due to COVID-19^{11,12,18,31,32}, with a BMI ≥ 35 kg/m2 radically increasing the mortality risk³³.

In our study, asthma diagnosis was present among 7.6% of hospitalized patients with COVID-19, which is lower than the 9.8% prevalence of asthma in Northern Ireland³⁴. At the same time, we found asthma to be associated with a lower risk of COVID-19-related death in the full sample, although the results of our sensitivity analysis showed no difference in risk of COVID-19-related death in patients with asthma compared to individuals with no documented comorbidities. This finding supports the mixed evidence on the role of asthma in influencing COVID-19-related outcomes. The Open SAFELY study identified asthma as a significant risk factor of death in patients with COVID-19 and indicated that patients on inhaled corticosteroids have the greatest risk³. In vitro studies have suggested that corticosteroids use can result in impaired antiviral innate immune responses^35,36 and delayed virus clearance³⁷, and this in turn can potentially lead to more severe outcomes in individuals who contracted SARS-CoV-2; however, this hypothesis has to be further tested. Several studies found no statistically significant difference in mortality risk by asthma status^38,39. For example, the prospective case–control study based on the UK Biobank data showed that asthma did not significantly increase the odds of COVID-19 mortality³⁸. More recent work however indicated that people with asthma were in fact less likely to die due to COVID-19⁴⁰. Interestingly, the risk of severe clinical outcomes of COVID-19 was lower in people with allergic asthma⁴¹. Akenroye et al.⁴² suggested that some asthma medications, such as mepolizumab, reslizumab, and benralizumab, may enhance immune responses to viral infections and potentially decrease susceptibility to additional lung injury from diseases such as COVID-19. The association between asthma and COVID-19 mortality could also differ by the degree of asthma severity³. Given that our sensitivity analysis showed no difference in risk of COVID-19-related death in patients with asthma compared to individuals with no documented comorbidities, but our primary results indicated a lower risk of COVID-19-related death in patients with asthma compared to non-asthma controls, further analysis is required to evaluate the impact of differing patterns and extents of multimorbidity in patients with asthma (e.g. cardiometabolic multimorbidity) and the role of different asthma medications when examining COVID-19-related outcomes.

We confirmed that COVID-19-related mortality increased with older age. In particular, patients in age groups ≥ 50 years old had higher odds of COVID-19-related death when compared with those aged 40 to 49 years. Higher COVID-19 mortality among older adults has been known since early in the pandemic and has been described in detail^3,27,43. The analysis based on the US epidemiologic data demonstrated that the overall COVID-19 case-fatality rate among individuals infected with SARS-CoV-19 was highest in those aged ≥ 85 years (range 10–27%), followed by those aged 65–84 years (3–11%), aged 55–64 years (1–3%), and aged < 55 years (< 1%)⁴³. Greater risk of death due to COVID-19 in older adults is likely related to their declining immune defences, however other hypotheses have also been suggested⁴⁴.

Finally, we demonstrated that multimorbidity is an important clinical characteristic to consider in the context of the COVID19 pandemic. Our findings showing the higher COVID-19 in-hospital mortality risk in people with multiple underlying conditions are consistent with other published data⁴⁵. A cross-sectional, multicenter, observational study of Italian COVID-19 population found that increasing multimorbidity, measured by the Charlson Comorbidity Index, was strongly associated with COVID-19-related death⁴⁶. Kim et al.⁴⁷, in their study of 2,491 COVID-19 patients, reported that individuals with 3 or more underlying conditions had a 1.8 times higher risk of in-hospital mortality than patients with no underlying conditions. Multimorbidity was also reported to be a predictor of the risk of COVID-19 infection in a large UK Biobank cohort of 428,199 participants; however, the authors did not report on the relationship between the co-existence of multiple underlying conditions and risk of death⁴⁸.

We also analysed multimorbidity patterns by identifying clusters of conditions in hospitalised COVID-19 patients using an unsupervised machine learning technique (k-mode clustering). Our study revealed recognisable co-occurrences of COVID-19 with different combinations of diseases, with a potentially causal link or underlying mechanism, including cardiovascular diseases, respiratory diseases, mental and neurological disorders, metabolic and endocrine diseases, and renal diseases. For example, Cluster 1 characterized by the group of individuals with cardiovascular diseases had also a high percentage of cases with COPD. Previous studies suggested that the systemic inflammatory response associated with COPD may act as a possible mechanism that links COPD with increased risk for cardiovascular diseases⁴⁹. Furthermore, it was demonstrated that COPD is associated with increased carotid intimal medial thickness (CIMT) and that among those with COPD, CIMT is linked to higher cardiovascular mortality⁵⁰. Evidence also supports the association between renal disease and cognitive impairment (Cluster 2); however, the mechanisms underlying this association are not completely elucidated⁵¹. Although, direct impact of uremic toxins has been proposed as a potential cause of cognitive decline, studies showed that dialysis prescription did not reverse symptoms of cognitive impairment⁵². Co-occurrence of other conditions, such as, (i) metabolic syndrome (defined by the presence of metabolic abnormalities including obesity, glucose intolerance, and elevated blood pressure), cardiovascular disease, and liver disease (Cluster 2, 3 and 5)^53,54; (ii) mental disorders and heart disease (Cluster 6)⁵⁵, and (iii) cancer with obesity and diabetes (Cluster 4)⁵⁶ has also been broadly documented and shown to be associated with high mortality. Therefore, the identification of these multi morbidity patterns among hospitalized individuals with COVID-19 can help identify opportunities to target patient-centred care towards people with high-risk ages and a specific combination of health conditions, leading to improved clinical outcomes. Note that we acknowledge that a presence of a specific chronic disease or a combination of chronic diseases in our analytic sample may have acted as an effect modifier of COVID-19 death but could also be associated with COVID-19 death via the existence of another common cause i.e., other clinical characteristics or socio-economic factors not included in our study.

The strengths of our study include the large, regional cohort of hospitalised COVID-19 patients, providing high statistical power to investigate associations between different risk factors and COVID-19 in-hospital mortality. The use of ICD-10 diagnosis codes assigned by medical professionals working in all hospitals throughout Northern Ireland meant that comprehensive information on a wide range of comorbidities were available. Our results remained robust in a number of sensitivity analyses and reinforced previous findings of a higher risk of COVID-19-related death associated with obesity¹², liver disease⁵⁷, renal disease⁵, metastatic cancer⁵⁸, congestive heart failure⁵⁹, and COPD¹⁹. Furthermore, we investigated the associations between several conditions for which little data exist regarding risk for in-hospital mortality in patients with COVID-19, such as hypothyroidism, solid tumour without metastasis, lymphoma, dementia, and other neurological disorders. We also added to the, so far inconclusive, evidence on the role of asthma in influencing COVID-19-related outcomes. Finally, given a paucity of research on the impact of multi morbidity on the risk of COVID-19 in-hospital mortality, we examined the association between increasing multi morbidity, multi morbidity patterns, and COVID-19-related death. To our knowledge, this is the first study to characterise patterns of multi morbidity in a hospitalised population with COVID 19 using an unsupervised machine learning approach.

The interpretation of our results should be made considering several limitations. First, due to unavailability of ICD-10 diagnosis codes, we were unable to consider the records of 552 patients with COVID-19, admitted to hospital in the period from March 1, 2020, to January 31, 2021. Furthermore, we have only used the comorbidity data collected in the studied period. Therefore, it is possible we missed some comorbidities by not including information reported at prior admissions. Second, the underlying cause of death was allocated based on the hospital records of ICD-10 diagnostic coding and discharge status, not death certificate; this might have led not only to the underestimation of the real magnitude of mortality due to COVID-19 but also potential misclassification of deaths from other causes. To assess the extent to which these inaccuracies may have affected our estimates, a similar analysis should be performed in the future using death certificate–based ICD-10 diagnosis codes. Third, since the ICD-10 diagnosis codes were assigned by medical professionals working in different hospitals, they may not have captured intended disease concepts with complete consistency. Fourth, our analysis is cross-sectional, meaning that both the independent variables and the outcome were collected simultaneously. Although the central element of the cross-sectional design is the lack of the temporal information required to describe the evolution of the underlying dynamics, in case of our study, the temporal link between the outcome and independent variables can be cautiously assumed since the presence of underlying health conditions most likely preceded the outcome studied (i.e., fatal/non-fatal hospitalization due to COVID-19). Fifth, the impact of inpatient treatment or procedures performed during hospitalisation as well as the information on the duration of the long-term conditions on patient outcomes was not considered in this study. In addition, unavailability of data on clinical parameters such as oxygen and ventilation treatment limited the opportunity of a more comprehensive analysis including multiple levels of severity as the outcome. Sixth, the purpose of using the clustering approach was for this method to potentially become part of the pipeline for discovering the various multi morbidity profiles that are associated with mortality due to COVID-19. Extensions to this research may include a more robust look at the effects of the hyper parameters and different analytic samples on the associations between different disease groups and COVID-19-related deaths. Furthermore, the results of the clustering algorithm should be further validated by domain experts to determine its clinical utility. It is important to highlight that this analysis was performed among a cohort of individuals hospitalized with COVID-19 and may not be generalizable to those non-hospitalized. The lack of data on non-hospitalized individuals with SARS-CoV-2 infection limited our ability to examine utilization of our algorithm outside of the hospitalized patients. However, our population represented a diverse cohort with a variety of comorbidities and is representative of high-risk individuals across Northern Ireland. Seventh, as mentioned above, our analytic sample does not represent a true random selection from the population since it is solely based on hospitalised patients, that died or did not die during the admission episode following COVID-19 diagnosis. This non-randomness of sample selection may have had an impact on our study results. Older adults, those obese, and with pre-existing medical conditions are more susceptible to adverse COVID-19 outcomes, while COVID-19 severity likely influences hospitalisation⁶⁰. As such, characteristics related to our sample inclusion, also relate to the hypothesised risk factors and the outcome of interest and hence, investigating these factors within hospitalised patients may have introduced the possibility of collider bias. In future work, the likelihood and extent of collider bias associated with sample selection could be evaluated by comparing means, variances, and distributions of variables in the sample of individuals with COVID-19 that have been hospitalized with those in a representative sample of the NI population⁶¹. This could not be evaluated in the current data set. Finally, our study may be subject to other potential sources of bias. For example, selection bias could have been introduced by not including non-hospitalized individuals living in long-term care or assisted living facilities that suffered from severe COVID-19 symptoms and subsequently died due to COVID-19⁶². Bias due to omission of a confounder from the model (unmeasured confounder) is also a possibility. Several demographic factors (e.g., race/ethnicity, disability status, socio-economic status), as well as structural factors (e.g., literacy) were not included in the analysis^63,64,65,66. Moreover, we did not control for the effect of concurrently prescribed medications such as antipsychotics, proton pump inhibitors, antihistamines, and opioid analgesics that were previously shown to increase the risk of adverse outcomes among patients with COVID-19⁶⁷. The association between certain drug classes, polypharmacy, and the risk of COVID-19 mortality should be addressed in future work. In addition, further analysis on host-specific genetic factors and their relationships with severe manifestations of COVID-19, in particular, in individuals with no underlying health conditions that died as a direct consequence of SARS‐CoV‐2 infection, could allow us to better understand relationships between environmental risk factors and severe outcomes associated with COVID-19 and identify potential targets for therapeutic development.

This study estimated the impact of the broad spectrum of comorbidities on COVID-19 in-hospital mortality using a large, regional cohort of hospitalized patients. We found that individuals with lymphoma, metastatic cancer, solid tumour without metastasis, liver disease, congestive heart failure, chronic obstructive pulmonary disease, obesity, renal disease, and dementia were at significantly increased risk for COVID-19-related death. In addition, we showed that the presence of multiple coexisting health conditions further increased odds of death. Given that effective clinical management of patients with multimorbidity is therefore a critical step towards their survival from COVID-19, multidisciplinary clinical teams should prepare comprehensive care plans that can be streamlined and meet the dynamic needs of such patients. Future work should investigate the associations between different patterns of multimorbidity and COVID-19 mortality to better characterize those individuals who would benefit from enhanced preventive measures.