Complex patterns of multimorbidity associated with severe COVID-19 and long COVID – Nature.com
July 11, 2024
Here, we systematically investigate the risk conferred by the presence and potential causal relevance of 1448 diseases for COVID-19 severity (hospitalisation, severe respiratory failure, and death) and Long COVID (Fig.1), based on medical disorder concepts14,16 defined and collated from >12 million medical records from primary (general practice), secondary care (hospital admissions), and disease registry (cancer registry), death certificates, and patient-reported conditions among 502,460 UKB participants (Fig.1 and Supplementary Data1). Incorporating primary care data more than doubled case numbers for more than half (n=817; 56.4%) of the diseases we considered (Supplementary Data1).
Scheme of the study design and analysis done, illustrating our workflow to define disease mechanisms that may causally contribute to severe COVID-19 or Long COVID. SNPs Single nucleotide polymorphisms; SPA = saddle point approximation; MAF = minor allele frequency; *COVID-19 HGI=COVID-19 Host Genetic Initiative, but excluding contributions from UK Biobank
We identified 1128 significant (p<1.1105) disease COVID-19 outcome associations, including almost half (n=679) of the diseases considered with at least one of the four COVID-19 outcomes derived (Fig.2 and Supplementary Data2). Pre-existing diseases were almost exclusively associated with a higher risk for COVID-19 endpoints (median hazard ratio (HR): 2.39, range: 0.5917.3), only two diseases (benign neoplasm of skin and varicella infection) were associated with a decreased risk. Associated diseases spanned almost all chapters of the ICD-10 (17 out of 18) but were consistently enriched in the chapters respiratory (odds ratio [OR]: 5.96; p-value: 2.7x108), circulatory (OR: 2.95; p-value: 3.5x107), and endocrine/metabolic diseases (OR: 2.76; p-value: 9.1104) when associated with severe COVID-19. In contrast, pre-existing disease-codes classified as symptoms were more than 13-fold enriched among diseases associated with an increased risk for Long COVID (OR: 13.2; p-value: 3.6x108) but also hospitalisation (OR: 5.53; p-value: 9.9x105) and death (OR: 3.06; p-value: 7.3x103).
Each panel contains association statistics, p-values (triangles), from Cox-proportional hazard models (two-sided) testing for an association between the disease on the x-axis and three different COVID-19 outcomes, as well as Long COVID. Disease associations passing the multiple testing correction (dotted line, p<1.1105) are depicted by larger triangles of which facing up ones indicate positive, e.g., increased disease risk, associations and downward facing vice versa. The diseases are ordered by ICD-10 chapters (colours) and the top ten for each endpoint annotated. Underlying sample numbers and statistics can be found in Supplementary Data1 and 2.
For COVID-19 requiring hospitalisation, we replicated and refined known associations with serious pre-existing diseases that have been previously used to identify clinically extremely vulnerable people. This included respiratory diseases like pseudomonal pneumonia (HR: 7.53, 95%-CI: 4.7411.97; p-value<1.2x1017), acute renal failure (HR: 4.02, 95%-CI: 3.744.32, p-value: <10300) or type 2 diabetes with renal complications (HR: 7.44; 95%-CI: 5.679.76; p-value: 1.5x1047), as well as immune deficiencies (e.g., deficiency of humoral immunity HR: 6.02; 95%-CI: 4.368.31; p-value: 1.3x1027) or patients under immune suppression (e.g., liver transplants HR: 7.25 95%-CI: 4.5111.68, p-value: 3.4x1016). However, we further observed strong associations with so far less recognized pre-existing mental health and psychiatric diseases and conditions with effect sizes comparable to those previously considered to identify extremely vulnerable people. This included symptoms of malaise and fatigue (HR: 2.17, 95%-CI: 2.072.27, p-value: 4.4x10222) or suicide attempts (HR 5.33, 95%-CI: 4.456.39, p-value: 3.6x1073). Most diseases (n=641, 95.5%, phetero>103) associated with similar magnitude across all three different definitions of COVID-19 severity, with different forms of dementias (phetero<2.1x1024) being among the few exceptions, associating with hospitalisation (HR: 3.83; 95%-CI: 3.384.34; p-value: 2.3x1097) and death (HR: 10.82; 95%-CI: 9.1512.80; p-value: 1.4x10170), but not severe respiratory failure (HR: 1.15; 95%-CI: 0.512.57; p-value: 0.74) due to COVID-19.
In contrast, pre-existing diseases associated with an increased risk for Long COVID only partially overlapped with those increasing the risk for severe COVID-19. Most notably, we replicated associations with anxiety disorders28 (HR: 2.59; 95%-CI: 2.093.20; p-value:1.8x1018) and other mental health symptoms, but most prominently with symptoms of malaise and fatigue (HR: 2.78; 95%-CI: 2.293.37; p-value:1.5x1025) that are hallmarks of Long COVID and were also strongly associated with severe COVID-19.
Almost all significant associations (99.8%, n=1126) were consistent when considering all-cause death as a competing event (Supplementary Data3), and more than half (63.6%; n=718) remained statistically significant (p<4.4x105) when accounting for a large set of potential confounders in multivariable Cox-models (Supplementary Data3). This suggests that potentially unreported associations, such as the increased risk for severe COVID-19 among patients reporting symptoms of malaise and fatigue (adjusted HR: 1.66, 95%-CI: 1.58 - 1.74, p-value=7.3x1092), are not just a reflection of a general disease burden or other chronic diseases associated with a greater risk for severe COVID-19.
We observed limited evidence for effect modifications by sex (n=7), non-European ancestry (n=1), or age (n=8), but not social deprivation, with 16 disease COVID 19 pairings showing evidence of significant differences (Supplementary Data4; p<3.6x106). All included stronger effects in women compared to men, e.g., gout for hospitalised COVID-19 (women: HR: 2.56, 95%-CI 2.212.96, p-value: 1.3x1036; men: HR: 1.46, 95%-CI: 1.341.58, p-value: 2.1x1019), among Europeans reporting vitamin D deficiencies (Europeans: HR: 2.31, 95%-CI: 2.132.51, p-value: 2.1x1087; non-Europeans: HR: 1.31, 95%-CI: 1.081.60, p-value=5.5x103), or among younger participants, e.g., disorders of magnesium metabolism and death with COVID-19 as a likely result of renal failure (age 65 years: HR: 42.98, 95%-CI: 20.1091.90, p-value: 3.0x1022; age >65 years: HR: 5.35, 95%-CI: 3.518.16, p-value: 5.9x1015).
We next derived a disease-disease network18 (Fig.3a) to understand, whether the large set of diseases associated with an increased risk for severe COVID-19 act independently or rather reflect an increased risk among participants suffering from multiple pre-existing conditions, i.e., multimorbidity. The network contained a total of 1381 diseases connected through 5212 edges based on non-random co-occurrence (Supplementary Data 5a, b). Diseases segregated into 31 communities being more strongly connected to each other compared to the rest of the network (Fig.3b, c).
a Disease disease network based on significant (p<4.8x108) positive partial correlations (two-sided). Nodes (diseases) are coloured by ICD-10 chapters and strength of partial correlation depicted by width of the edges. The underlying data can be found in Supplementary Data 5ac Same network, but only highlighting two disease communities strongly enriched for associations with severe COVID-19. d Hub score for the 30 diseases with highest values and associated association statistics, hazard ratios (rectangle) with 95%-confidence intervals (lines), from Cox-proportional hazard models (two-sided). Significant associations are indicated by filled boxes. Colours according to ICD-10 chapters. All underlying data can be found in Supplementary Data2 and 5b.
Two disease communities were consistently and strongly enriched for diseases associated with severe COVID-19. The first (e.g., OR: 5.20; p-value=2.2x1010; for severe respiratory failure) community was strongly enriched for circulatory (OR: 17.6; p-value: 4.4x1039) and respiratory (OR: 10.3; p-value: 7.8x1016) diseases, closely resembling the cardio-respiratory risk profile already described above (Fig.3b). The second community consisted of diverse endocrine (OR: 6.19; p-value: 1.9x1013) and circulatory disease (OR: 3.75; p-value: 5.4x108), and largely reflected the renal-diabetic risk profile (Fig.3c). Accordingly, for each disease acquired during lifetime within the latter disease community, participants risk increased by 18% and 20% to be hospitalised (HR: 1.18; 95%-CI: 1.171.18; p-value: p<10300) or die with COVID-19 (HR: 1.20; 1.191.20; p-value<10300), respectively.
Diseases increasing the risk for severe COVID-19, but not Long COVID further significantly correlated with hub status (e.g., hospitalisation: r=0.59; p-value: 2.8x10124) in the disease-disease network (Fig.3d), that is, diseases that connect a large cluster of diseases to the rest of the network and might hence be considered as multimorbidity hotspots. For example, acute renal failure, strongly associated with severe COVID-19 (Fig.3d), showed strong partial correlations with 30 other diseases and patients are hence prone to complex multimorbidity. However, the imperfect correlation between hub status and disease-association profiles indicates that certain forms of multimorbidity, such those related to secondary malignancies of lymph nodes, are possibly less related to severe COVID-19.
We next systematically characterised whether diseases identified to be associated with COVID-19 severity or Long COVID shared genetic similarity with host genetic susceptibility to severe COVID-19 to understand potential underlying causal mechanisms. We computed genetic correlation estimates for all 1128 disease COVID-19 outcome pairs and observed 75 pairs (6.6%) that showed evidence for significant (p<4.4x105) and directionally consistent genetic correlations (Fig.4 and Supplementary Data6), indicating a putatively causal link of any of 57 unique diseases on severe COVID-19. We did not observe evidence of convergence for Long COVID, which might likely be explained by the still low statistical power for the respective genome-wide association study13.
The first three panels show association statistics, hazard ratios (rectangle) and 95%-confidence interval (lines), for 57 diseases with evidence of convergence with genetic correlation analysis, that are shown in the last two panels (rectangle genetic correlation; lines 95%-confidence intervals). Disease have been grouped by ICD-10 chapters and coloured accordingly (see Figs.2 or 3 for legend). NOS = not elsewhere specified; All underlying data can be found in Supplementary Data1 (sample numbers), 2 and 6.
The diseases with consistent evidence from survival and genetic analysis included well-described risk-increasing effects of pre-existing endocrine (e.g., type 2 diabetes), respiratory (e.g., respiratory failure), or renal (e.g., chronic kidney disease) diseases, but also digestive (e.g., gastritis and duodenitis), or musculoskeletal (e.g., rheumatoid arthritis) diseases, and further symptoms of malaise and fatigue (rG=0.26; p-value=4.7106) and abdominal pain (rG=0.33; p=2.51011), as well as adverse reactions to drugs (e.g., poisoning by antibiotics: rG=0.38; p-value=2.2x106). Findings that collectively demonstrated the need for a comprehensive assessment of disease-risk beyond few, selected common chronic conditions.
Among the 41 diseases for which we had sufficient genetic instruments to perform more stringent Mendelian randomization (MR) analyses to assess causality, we observed only nominally significant (p<0.05) evidence for gout and hospitalisation (OR: 1.03; 95%-CI: 1.011.05, p-value: 0.03), as well as arthropathy not elsewhere specified (OR: 1.28; 95%-CI: 1.061.55; p-value: 0.02) and unspecified monoarthrtitis (OR: 1.21; 95%-CI: 1.041.41; p-value: 0.02) for severe COVID-19 (Supplementary Data7). While we might have been still underpowered for many diseases, this leaves the possibility that convergence of survival and genetic correlation analysis might, in part, be explained by shared risk factors.
To finally understand possible molecular mechanisms linking the diseasome to COVID-19, we systematically profiled disease associations across 49 independent genomic regions linked to COVID-19 or Long COVID. We observed strong and robust evidence of a genetic signal shared between severe COVID-19 and a total of 33 diseases at nine loci (posterior probability (PP)>80%) (Fig.5a and Supplementary Data8). Apart from known pleiotropic loci, such as ABO and FUT2 coding for blood group types, this included respiratory risk loci, albeit with contradicting effect estimates for three loci (Fig.5b). While COVID-19 risk increasing alleles at LZTFL1 and TRIM4 were consistently associated with a higher risk for viral pneumonia and post-inflammatory pulmonary fibrosis, respectively, risk-increasing alleles at MUC5B, NPNT, and PSMD3 were inversely associated with post-inflammatory pulmonary fibrosis and asthma. An observation that extended even beyond shared loci (Fig.5c) illustrating a general trend of phenotypic divergence of genetic effects on diseases that share pathological features with severe COVID-19.
a Network representation of significant (PP>80%) colocalization results. Loci are depicted as white rectangles and diseases as coloured nodes according to ICD-10 chapters. Edges represent strong evidence for colocalization, and solid lines indicate a risk-increasing effect of the COVID-19 risk increasing allele, whereas dashed lines indicate protective effects. Underlying data can be found in Supplementary Data8. b Forest plot displaying hazard ratios (rectangle) with 95%-confidence intervals (lines) for each variant and different COVID-19 and colocalising disease outcomes. Effect estimates for COVID-19 have been obtained from the COVID-19 Host Genetic Initiative and effect estimates for diseases in the present study. All estimates are derived from logistic regression models. c Heatmap of effect estimates across 49 independent genetic loci associated with increased risk for sever COVID-19 and corresponding effects on six selected traits that showed evidence of colocalization at least one other locus. Black rectangles indicate genome-wide significant effects (p<5108). NOS not elsewhere specified; All underlying data can be found in Supplementary Data1 and 8 or is given in the data availability statement.
A notable observation was the TYK2 locus that has previously been suggested to indicate the efficacy of successfully repurposed drugs for severe COVID-1929. Briefly, TYK2 encodes for tyrosine kinase 2 (TYK2) a protein partially targeted by Janus kinase (JAK) inhibitors like baricitinib, that have been approved for rheumatoid arthritis and successfully repurposed for severe COVID-19, although predating possible evidence from genetic studies30,31,32. Accordingly, we observed that the same genetic variant, rs34536443 (PP=99.8%), associated with the risk for severe COVID-19 was also associated with, amongst others, the risk of rheumatoid arthritis, but in opposing effect directions (Fig.5b). Rs34536443 is a loss-of-function missense variant (p.Pro1104Ala) for TYK2 and the functionally impairing minor C allele was associated with a 50% increased risk for severe COVID-19 (odds ratio: 1.50; 95%-CI: 1.40 1.62, p-value= 4.3x1029) but a 23% reduced risk for rheumatoid arthritis (odds ratio: 0.77; 95%-CI: 0.720.83; p-value=2.4x1012) as well as other autoimmune diseases, in particular psoriasis (Supplementary Data8). While the discrepancy between the success of the drug and genetic inference might be explained by the rather weak affinity of baricitinib for TYK233, patients undergoing trials with TYK2-inhibitors for psoriasis34 might be at an elevated risk for severe COVID-19. This observation seemingly aligns with studies on Tyk2-/- mouse models reporting an impaired immune response to viral infections35.
See the article here:
Complex patterns of multimorbidity associated with severe COVID-19 and long COVID - Nature.com