Disease network analysis to reveal comorbidity patterns in hospitalized patients with COPD using large-scale administrative health data

Abstract

Chronic obstructive pulmonary disease (COPD) is a common respiratory condition with a high comorbidity burden. This study aims to utilize regional administrative health data and employ network analysis to systematically investigate COPD comorbidity patterns across the entire spectrum of chronic diseases. The hospitalization discharge records from all secondary and tertiary hospitals in Sichuan Province, China, from 2015 to 2019 were collected, including 2,004,891 COPD inpatients. We constructed comorbidity networks using the Salton Cosine Index, applied centrality measures to identify central diseases, and the Louvain algorithm to detect clusters. We found that 96.05% of COPD patients had at least one comorbidity, with essential (primary) hypertension (40.30%) being the most prevalent. The comorbidity network identified 11 central diseases including disorders of glycoprotein metabolism as well as gastritis and duodenitis. Sex differences were reflected in the comorbidity relationships of hyperplasia of the prostate in the male network and osteoporosis without pathological fracture in the female network. Urban patients demonstrated higher comorbidity prevalence and exhibited more complex comorbidity relationships compared to rural patients. This study provided a comprehensive understanding of the complex relationships among comorbidities in hospitalized COPD populations, contributing to the advancement of patient-centered care.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-12580-9.

Introduction

Chronic obstructive pulmonary disease (COPD) is a common respiratory condition, which ranks as the fourth leading cause of death globally. The World Health Organization has reported that about 3.5 million deaths worldwide were attributable to COPD in 2021, accounting for 5% of all deaths in that year¹. The number of COPD patients has increased dramatically rising from 8,722,696 in 1990 to 16,214,828 in 2019². In China, COPD has been the third leading cause of death since 1990³, with its incidence and mortality rates expected to continue rising over the next 25 years², thereby making it a significant public health issue.

COPD often co-occurs with other chronic diseases⁴. In a retrospective study in Denmark, 81% of COPD patients had at least one chronic comorbidity⁵. Another retrospective study in Spain found that over 80% of patients with COPD in all groups had at least one comorbid condition⁶. Previous studies have shown that comorbidities significantly affect the health status of COPD patients^7–9. For example, a genetic association study found that COPD patients with asthma experienced more severe symptoms and poor quality of life compared to those without asthma¹⁰. Numerous studies have also demonstrated that comorbid conditions such as diabetes, ischemic heart disease, hypertension, and depression increase the risk of severe deterioration and mortality in COPD patients^11–13. Therefore, investigating the comorbidity patterns of COPD is crucial for effective disease management and targeted prevention and treatment.

To date, most existing studies on COPD comorbidities have primarily focused on a few common diseases^14,15, such as cardiovascular diseases and diabetes, and interpreted them using prevalence statistics¹⁶ and traditional regression methods¹⁷, lacking a comprehensive understanding of the relationship patterns among multiple diseases. In addition, many studies relied on self-reported¹⁸ data or small-scale data collected from a few hospitals¹⁹, potentially introducing systematic biases due to data constraints.

Recent advances in network analysis techniques have provided a new approach for the study of comorbidity patterns^20–22. Disease networks use correlation coefficients such as Pearson’s coefficient, the relative risk (RR), and the Salton Cosine Index (SCI)^23,24 to represent the association between diseases. This approach has been applied to study comorbidity patterns of various diseases, such as cardiovascular diseases^25,26, depression²⁷, hepatocellular carcinoma²⁸, COVID-19²⁹, and migraine³⁰, revealing the complex relationship among multiple comorbidities. Moreover, the rapidly growing administrative health data provides new opportunities for a systematic evaluation of the entire disease spectrum through network analysis^31,32, offering a more in-depth foundation for investigating comorbidity profiles^33,34 and predicting disease risk^35–37.

The network analysis method has been applied to several studies on comorbidities of chronic obstructive pulmonary disease patients in Spain^6,38,39. However, the datasets used in these studies are relatively small. To the best of our knowledge, no study has yet combined large-scale administrative health data with a network analysis approach to systematically investigate the comorbidity patterns of COPD, especially in Chinese populations, whose lifestyle and living environment differ significantly from those in Western populations.

Therefore, the present study aims to utilize regional administrative data and employ network analysis to identify comorbidity patterns among hospitalized patients with COPD in southwest China. We hope that this systematic analysis will provide a more comprehensive understanding of the complex relationships among multiple diseases, enhance our understanding of COPD comorbidities and offer valuable insights for the management of COPD patients.

Methods

Data source and study population

This retrospective study was conducted using longitudinal data from a provincial health database. The database collected anonymized hospital discharge records (HDRs) from all secondary and tertiary hospitals in Sichuan Province, China, since 2015. Each HDR included anonymized identity, age, sex, residential address, admission date, discharge date, one primary discharge diagnosis, and up to 15 secondary discharge diagnoses. All the diagnoses were coded according to ICD-10 (International Classification of Diseases, 10th Revision). A total of 2,004,891 inpatients aged over 18 years old, residing in Sichuan Province, with at least one diagnosis of COPD (ICD: J41-J44) during the study period from January 1, 2015, to December 31, 2019, and without a death record within that period were included in the study.

Identification of comorbidities

We examined three-digit ICD-10 codes for the entire spectrum of conditions and applied the chronic condition indicator to differentiate between acute and chronic ICD-10 codes⁴⁰. Only three-digit ICD-10 codes representing chronic diseases were included for the study. Diagnostic codes from chapters XV to XXII were excluded, as these codes are not diseases or are general symptoms. To avoid rare diseases, we only included chronic diseases with a prevalence significantly ≥ 1%^32,41 (using a Z-test with Bonferroni correction).

Prevalence and enriched diseases

To provide a more detailed and accurate depiction of comorbidity patterns among COPD patients, we stratified the study population based on sex (male and female), geographic region (urban and rural), and age group (18–44, 45–59, 60–69, 70–79, 80+). To assess differences in comorbidities across different sexes and geographic regions, we first calculated the prevalence of comorbidities within each subgroup and applied a Z-test (after Bonferroni correction) to determine statistical significance. For comorbidities with significant differences, we denoted the prevalence in the two subgroups (e.g., male vs. female, urban vs. rural) as prevalence1 and prevalence2, and computed as follows:

1

2

A disease was considered enriched in the subgroup with the higher prevalence if the relative difference was ≥ 0.5.

Comorbidity network construction

In a comorbidity network, each node represents a chronic disease, and the edges between nodes denote the co-occurrence strength of disease pairs. Common metrics for measuring comorbidity strength include the RR, Pearson’s correlation coefficients and the SCI index^23,24. We employed the SCI index because it is not affected by the sample size⁴², providing a more stable measure of disease co-occurrence strength. The Pearson’s correlation coefficient is used to determine the cutoff value for SCI, based on the principle that the number of significantly coexisting disease pairs is equal in networks constructed using the cosine index and Pearson’s correlation⁴³. The process of calculating SCI and determining its cutoff value follows these steps:

1. For any two diseases, compute the SCI (Eq. (3)).

2. Calculate the phi correlation coefficient (Eq. (4)) for each disease pair and select statistically significant correlations at α = 0.01, resulting in a set of significant phi values (Eq. (5)).

3. From the n nodes with computed phi values in Step 2, calculate =/p, where p = .

4. Identify disease pairs satisfying , resulting in q pairs.

5. Rank the disease pairs in descending order of SCI, determine the cutoff based on the top q pairs, and select those exceeding this threshold. The resulting significant disease pairs and their association coefficients are then used to construct an undirected, weighted comorbidity network.

   3

   4

   5

   where represents the number of patients with both disease and disease , and represent the number of patients with only disease or only disease , respectively, and represents the total number of patients.

Comorbidity network analysis

Firstly, we conducted a statistical analysis of the comorbidity network’s structural characteristics, including the number of nodes, the number of edges, network density, and network diameter. To assess the importance of diseases within the comorbidity network, the PageRank algorithm⁴⁴ was employed to calculate the centrality of each node. Higher PageRank scores indicate that a disease is more important within the network. In addition to the PageRank algorithm, degree, weighted degree, betweenness centrality, and eigenvector centrality were also used as supplementary centrality measures. Diseases with a high degree centrality have more direct connections with other diseases in the network. High weighted degree centrality indicates that a disease has a stronger actual influence in the network. High betweenness centrality indicates that a disease has a greater ability to act as a “bridge” connecting other diseases in the network. Eigenvector centrality takes into account both the quality and quantity of a node’s connections in the network, accurately reflecting diseases with strong influence in the network. We identified the top 10% of nodes with the highest centrality as central diseases, which have an important impact on the network.

Besides, a community detection algorithm was used to further identify tightly connected clusters within the network. The Louvain algorithm is a community detection algorithm based on modularity optimization, whose core goal is to divide the network into several “communities”, making the connections within the communities as dense as possible and the connections between communities as sparse as possible⁴⁵. We employed the Louvain algorithm⁴⁶ to perform clustering analysis on the comorbidity network, dividing it into multiple tightly connected communities that may correspond to comorbidities driven by the same risk factors in clinical practice. Within each community, we identified the disease with the highest betweenness centrality as the core disease, indicating that it plays a key role in the development or progression of multiple comorbidities within a cluster.

Furthermore, to more precisely identify comorbidity patterns among COPD patients, the population was stratified into subgroups stratified by sex and age, as well as region and age, and comorbidity networks were constructed and analyzed for each subgroup, enabling the exploration of differences in comorbidity networks across these stratifications.

Statistical testing of network differences

To statistically compare network metrics between stratified groups, we employed a null network model approach⁴⁷. The procedure is as follows: For the two networks to be compared (G1 and G2), we randomly generated 500 networks according to the estimated degree distributions of each of the networks, resulting in 1000 networks in total. We then performed 1000*999/2 = 499,500 comparisons on these 1000 networks to calculate the expected value and standard deviation of the difference for each network metric. Finally, we conducted a one-sample t-test to generate P values, indicating that the differences in metrics between the two networks are statistically significant compared to the expected differences.

Results

Characteristics and comorbidity profiles

The demographic characteristics and comorbidity profiles are shown in Table 1. A total of 2,004,891 hospitalized patients with COPD were included in this study, with an average age of 70.49 years at enrollment. Among them, 1,223,095 were male, accounting for 61.01% of the total population.

Table 1.

Characteristics of study population.

Sex	Group type	Group	N (PCT (%))	Characteristics of comorbidity
				Patients with at least one comorbidity		Number of comorbidities
				PCT (%)	P value	Mean (± SD)	P value
Total	–	–	2,004,891	96.05	–	7.43 (± 5.76)	–
Male	Total	–	1,223,095	95.64	< 0.001	7.38 (± 5.81)	< 0.001
	Region	Urban	482,611 (39.46)	466,013 (96.56)	< 0.001	8.82 (± 6.83)	< 0.001
		Rural	740,484 (60.54)	710,249 (95.92)	< 0.001	7.48 (± 5.99)	< 0.001
	Age	18–44	26,550 (2.17)	19,991 (75.30)	< 0.001	3.63 (± 4.34)	< 0.001
		45–59	162,750 (13.31)	151,078 (92.83)	< 0.001	6.12 (± 5.40)	< 0.001
		60–69	375,653 (30.71)	360,866 (96.06)	< 0.001	7.58 (± 6.04)	< 0.001
		70–79	431,597 (35.29)	421,982 (97.77)	< 0.001	8.85 (± 6.59)	< 0.001
		80+	226,545 (18.52)	222,345 (98.15)	> 0.05	8.98 (± 6.73)	< 0.001
Female	Total	–	781,796	96.88	–	7.64 (± 5.75)	–
	Region	Urban	304,733 (38.98)	298,246 (97.87)	–	9.44 (± 6.96)	–
		Rural	477,063 (61.02)	462,950 (97.04)	–	7.62 (± 5.82)	–
	Age	18–44	12,835 (1.64)	11,267 (87.78)	–	4.36 (± 4.33)	–
		45–59	76,797 (9.82)	72,970 (95.02)	–	6.47 (± 5.40)	–
		60–69	204,446 (26.15)	198,382 (97.03)	–	7.99 (± 6.11)	–
		70–79	293,898 (37.60)	288,315 (98.10)	–	9.06 (± 6.60)	–
		80+	193,820 (24.79)	190,262 (98.16)	–	8.59 (± 6.40)	–

A Z-test was used to assess the significant differences in percentage of patients with at least one comorbidity between male and female groups (P < 0.001); A Mann–Whitney U test was used to assess the significant differences in number of comorbidities between male and female groups (P < 0.001).

The average number of chronic comorbidities was 7.43. 96.05% have at least one comorbidity, and 63.25% have at least two comorbidities. Females exhibited a significantly higher prevalence of comorbidities compared to males (96.88% vs. 95.64%; Z-test, P < 0.001). In both urban and rural areas, the average number of comorbidities in female patients was significantly higher than that in male patients (urban: 8.82 vs. 9.44; rural: 7.48 vs. 7.62; Z-test, P < 0.001). Additionally, the proportion of patients with at least one comorbidity increased progressively with age subgroups, rising from 75.3 to 98.15% in males and from 87.78 to 98.16% in females.

Comorbidity prevalence and enriched diseases

There was a total of 115 comorbidities with a prevalence greater than 1% in patients with COPD (Z-test, P < 0.05/115; Supplementary Table S1). The three most prevalent comorbidities among COPD patients were essential (primary) hypertension (I10, 40.3%), gastritis and duodenitis (K29, 40.0%), and chronic ischemic heart disease (I25, 31.9%). In sex-specific groups, 32 comorbidities were significantly more prevalent in male patients (P < 0.05), among them, 8 comorbidities were enriched in male patients, such as gout (M10, 3.3% in males vs. 1.0% in females). Conversely, 44 comorbidities were significantly elevated in females (P < 0.05), 11 of which were enriched, including osteoporosis without pathological fracture (M81, 16.2% in females vs. 6.6% in males). Across different regions, 80 comorbidities were significantly more prevalent in urban patients (P < 0.05), with 11 enriched diseases, such as hyperplasia of the prostate (N40, 22.0% in urban vs. 6.1% in rural). Rural patients had only 5 significantly prevalent comorbidities, none of which were enriched. Detailed information on enriched diseases is provided in Supplementary Table S2.

Comorbidity network in COPD

The COPD comorbidity network comprised 112 diseases and 1415 comorbidity pairs. Its visualization is shown in Fig. 1a, with the edges representing the top 25% of SCI values filtered for display. On average, each disease in the network exhibited significant associations with 25 comorbidities. The three comorbidity pairs with the highest SCI values were: gastritis and duodenitis (K29) with hyperplasia of the prostate (N40) (SCI: 0.549), essential (primary) hypertension (I10) with cerebral infarction (I63) (SCI: 0.457), and heart failure (I50) with complications and ill-defined descriptions of heart disease (I51) (SCI: 0.430). Figure 1b highlights the key central diseases that have a significant impact on the comorbidity network under the PageRank metric (Central diseases defined by other metrics are included in Supplementary Table S3 and the same presentation method is applied to networks stratified by gender and region). Most of them belong to the circulatory system (essential (primary) hypertension (I10), heart failure (I50), cerebral infarction (I63), other cerebrovascular diseases (I67)), and endocrine/metabolic diseases (disorders of glycoprotein metabolism (E77) and disorders of purine and pyrimidine metabolism (E79).

Fig. 1.

The comorbidity network of patients with COPD. (a) The overall COPD comorbidity network. (b) Central diseases within the comorbidity network. (c) Clustering results of the COPD comorbidity network. The node size is proportional to disease prevalence, edge width is proportional to the SCI value between disease pairs, and the node color indicates the corresponding ICD-10 chapter classification.

Clustering analysis of the network (Fig. 1c) identified six distinct communities with strong internal associations. For instance, in community 1, the correlations were mostly around endocrine and metabolic diseases, cardiovascular and cerebrovascular diseases, and nervous system diseases. In community 5, the diseases mainly consisted of digestive system diseases and genitourinary system diseases.

Sex-specific comorbidity patterns in COPD

Figure 2a illustrates the sex-specific comorbidity networks for males and females. The male comorbidity network exhibited higher complexity, with more comorbidities (117 vs. 106), significant comorbidity pairs (1487 vs. 1249), and a larger average degree (25.42 vs. 23.57) compared to the female network. For a clearer effect, we provide interactive HTML web versions for both male and female network graphs at: . The comorbidity pairs with the highest SCI values in both sexes were essential (primary) hypertension (I10) and cerebral infarction (I63). The second-highest SCI was gastritis and duodenitis (K29) with hyperplasia of the prostate (N40) in males, whereas in females, it was heart failure (I50) and complications and ill-defined descriptions of heart disease (I51). The top 50 SCI-ranked pairs for each sex are listed in Supplementary Table S4.

Fig. 2.

Comorbidity networks, SCI values, central diseases and communities in male and female patients with COPD. (a) The male and female comorbidities networks. (b) The SCI values distribution in sex- and age-specific groups. (c) The central diseases in sex- and age-specific groups. (d) The male and female communities in different age groups. Only communities with diseases more than 4 are shown and the label of each node is the core disease in this community.

Table 2 presents age-stratified comorbidity network metrics across 5 age subgroups. Statistical tests show the differences between the two corresponding subgroups are remarkable. Male comorbidity networks exhibited more pairs and higher average degrees than the female network. Males also had more diseases in four age subgroups, except for the 18–44 age group. As age increases, the number of nodes and edges in the networks of different subgroups ranged from 70 to 121 and 585 to 1559 in males, respectively, from 74 to 104 and 450 to 1225 in females. The average weighted degree of the male comorbidity network increased from 1.81 to 3.07.

Table 2.

Comorbidity network properties grouping by sex and age.

Group	Nodes	Edges	Density	Diameter	Avg. degree	Avg. w-degree	Avg. clo-centrality
Male
18–44	70	585	0.242*	4*	16.7*	1.81*	0.547*
45–59	107	1278	0.225*	3*	23.9*	2.47*	0.557*
60–69	108	1310	0.227*	4*	24.3*	2.70*	0.555*
70–79	121	1559	0.215*	4*	25.8*	3.01*	0.547*
80+	106	1220	0.219*	5*	23.0*	3.07*	0.539*
Female
18–44	74	450	0.167	5	12.2	1.43	0.509
45–59	100	1135	0.229	3	22.7	2.37	0.555
60–69	103	1189	0.226	4	23.1	2.78	0.547
70–79	104	1225	0.229	3	23.6	3.09	0.555
80+	101	1098	0.217	4	21.7	2.91	0.549

Avg. degree average degree of a network, Avg. w-degree average weighted degree of a network, Avg. clo-centrality average closeness of a network.

*The difference of this metric between male subgroup and female subgroup was significant compared to expected (using null network models method, one sample t-test, P < 0.001).

Figure 2b demonstrates the changes in the distribution of SCI values across different age subgroups. For both sexes, median SCI increased progressively in subgroups aged over 45 years. The details on the statistical comparison of sex-specific networks are presented in Supplementary Table S5.

Figure 2c showcases the evolution of central diseases across different age subgroups. Both males and females shared several central diseases in multiple age subgroups, including gastritis and duodenitis (K29), disorders of glycoprotein metabolism (E77), and other anemias (D64). Some diseases exhibited high importance only in males, such as hyperplasia of the prostate (N40) and chronic kidney disease (N28), while sleep disorders (G47) held significant weight only in specific female age subgroups. Chronic kidney disease (N28) stood as a central disease in male subgroups above 60, while sleep disorders (G47) played a central role in female subgroups below 60. Figure 2d displays the clustering results of comorbidity networks for males and females across different age subgroups. Other diseases of the liver (K76) served as the core disease in four male and three female age subgroups, while other anemias (D64) became the core disease in four male and five female age subgroups. Some community core diseases exhibited sex differences; for instance, hyperplasia of the prostate (N40) and chronic kidney disease (N28) were community core diseases only in the 45–59 and 60–69 male age subgroups, respectively, while other disorders of the brain (G93) was exclusively a community core disease in the 45–59 female age subgroup. Detailed clustering data are provided in Supplementary Table S6.

Region-specific comorbidity patterns in COPD

Compared to the rural comorbidity network, the urban comorbidity network was more complex, with more comorbidities (117 vs. 108) and more significantly comorbidity pairs (1496 vs. 1313). Figure 3a is a differential graph that only displays edges with weights greater than 0.05 for a better visualization. Most diseases either had higher prevalence rates in the urban COPD network or were exclusively present in the urban network (71 vs. 7), including atherosclerosis (I70) and other diseases of the liver (K76). In contrast, other pulmonary heart diseases (I27) was one of the few diseases with a higher prevalence in the rural network. The highest SCI value comorbidity pair in urban networks was between essential hypertension (I10) and cerebral infarction (I63), while in rural networks it was between gastritis/duodenitis (K29) and hyperplasia of the prostate (N40). Additionally, the SCI values of more comorbidity pairs in networks from different regions were presented in Supplementary Table S7. Table 3 summarizes the network metrics for different age groups across urban and rural populations. With the exception of the 18–44 age group, the urban network exhibited more diseases, comorbidity pairs, and a higher average degree than the rural network in the other age groups. As age increased, the number of comorbidities across different sex and age subgroups ranged from 69 to 118, the number of comorbidity pairs increased from 427 to 1497, and the average weighted degree rose from 1.48 to 3.43. Figure 3b demonstrates the distribution of SCI values across different age subgroups in urban and rural areas. In both networks, the median SCI value was lowest in the 45–59 age group but increased consistently in subsequent age groups. The details on the statistical comparison of region-specific networks are presented in Supplementary Table S8.

Fig. 3.

Comorbidity networks, SCI values, central diseases and core diseases in urban and rural patients with COPD. (a) The comorbidity networks’ differences between urban and rural patients. (b) The SCI values distribution in region- and age-specific groups. (c) The central diseases in region- and age-specific groups. (d) The core diseases in each community. The node size is proportional to the absolute value of the difference in disease, the edge width is proportional to the absolute value of the difference in SCI, and the node color indicates the corresponding ICD-10 chapter classification.

Table 3.

Comorbidity network properties grouping by region and age.

Group	Nodes	Edges	Density	Diameter	Avg. degree	Avg. w-degree	Avg.clo centrality
Urban
18–44	69	427	0.182*	4*	12.4*	1.48*	0.514*
45–59	102	1206	0.234*	3*	23.6*	2.46*	0.557*
60–69	110	1349	0.225*	4*	24.5*	2.92*	0.550*
70–79	118	1497	0.217*	4*	25.4*	3.36*	0.547*
80+	113	1381	0.218*	4*	24.4*	3.43*	0.549*
Rural
18–44	73	538	0.205	4	14.7	1.58	0.526
45–59	100	1159	0.234	3	23.2	2.31	0.560
60–69	107	1298	0.229	4	24.3	2.68	0.559
70–79	107	1281	0.226	3	23.9	2.88	0.554
80+	100	1076	0.217	4	21.5	2.71	0.554

Avg. degree average degree of a network, Avg. w-degree average weighted degree of a network, Avg. clo-centrality average closeness of a network.

*The difference of this metric between the urban subgroup and the rural subgroup was significant compared to expected (using null network models method, one-sample t-test, P < 0.001).

Figure 3c depicts the central diseases across different age subgroups. Several central diseases, such as other diseases of the liver (K76), other anemias (D64), and disorders of glycoprotein metabolism (E77), were consistently present in both urban and rural networks across multiple age groups. Some diseases, like atherosclerosis (I70) and sleep disorders (G47), were more prominent in the urban comorbidity network, while others, such as hyperplasia of the prostate (N40) and respiratory failure (J96), played a more significant role in rural subgroups. Figure 3d presents the clustering analysis results for comorbidity networks across various age subgroups in urban and rural areas. Notably, other anemias (D64), chronic ischemic heart disease (I25), heart failure (I50), and fibrosis and cirrhosis of the liver (K74) served as core diseases in multiple age-subgroup clusters, while hyperplasia of the prostate (N40) emerged as a core disease predominantly among rural patients. Further details on clustering analysis are provided in Supplementary Table S9.

Discussion

This study utilized network analysis based on large-scale administrative health data to reveal comorbidity patterns among hospitalized patients with COPD in southwest China. We identified high-prevalence diseases within the COPD population, as well as the central diseases in the comorbidity network. Additionally, we uncovered complex community structures within the comorbidity network of hospitalized COPD patients. Furthermore, our study analyzed differences in comorbidity prevalence, central diseases, and clustered communities across subgroups based on sex and region. These central comorbidities, closely interconnected comorbidity clusters, and observed differences should be prioritized in COPD clinical management, providing a basis for optimizing therapeutic strategies and developing targeted preventive measures.

Previous studies demonstrated that most patients with COPD have at least one additional disease^4,48. In our study, 96.05% of COPD inpatients in Sichuan Province had at least one comorbidity. We found that cardiovascular and cerebrovascular diseases (e.g., essential hypertension, chronic ischemic heart disease, heart failure, and other pulmonary heart diseases) and digestive system diseases (e.g., gastritis and duodenitis) were the most prevalent. This finding was consistent with previous studies in which high prevalence rates of cardiovascular and cerebrovascular diseases and digestive system diseases were reported in the general population⁴⁹ and COPD patients³³. Substantial evidence demonstrated a strong association between COPD and cardiovascular diseases^50,51. For instance, elevated levels of inflammatory markers in systemic inflammation may simultaneously affect both COPD and cardiovascular diseases, potentially serving as a mechanism linking the two conditions⁵². Moreover, we found that the chronic kidney disease occurred commonly (17.4% for other disorders of the kidney and ureter) in COPD patients and the prevalence increased with age. Previous studies have found that shared factors such as smoking and air pollution can promote the development of both COPD and chronic kidney disease, providing potential mechanistic support for their co-occurrence^53–55.

In the present study, other anemias, gastritis and duodenitis and other diseases of the liver were identified as central diseases in the COPD comorbidity network based on multiple centrality measures, indicating that their strongest influence on the overall COPD comorbidity network. These findings suggested that interventions targeting these central diseases may more effectively improve comprehensive comorbidity management in COPD patients. In a U.S. network analysis study using degree centrality, the most central disease was systemic hypertension, which was comorbid with 28 other diseases³⁸. In our research, essential (primary) hypertension exhibited high centrality in weighted degree, eigenvector centrality, and PageRank metrics, but ranked relatively low in degree centrality. This indicated that while essential (primary) hypertension is not widely connected to many other diseases, it maintains strong co-occurrence strengths with a few key conditions, and may influence the network through limited but important comorbidity pathways. Although there are differences between the two studies, yet both highlighted the pivotal role of cardiovascular diseases in comorbidity networks. Besides, cardiovascular (e.g., essential hypertension, chronic ischemic heart disease, heart failure) and digestive system diseases (e.g., gastritis and duodenitis) accounted for the highest proportion of comorbidities. Moreover, the highest co-occurrence rates between diseases in these two systems in our study may be driven by shared genetic factors⁵⁶ or similar disease mechanisms⁵⁷, offering new directions for COPD comorbidity prediction and management. Additionally, other anemias and hyperplasia of the prostate were the most frequent “bridge diseases” in the network, which may accelerate comorbidity progression through specific pathological pathways.

There were many differences in comorbidity patterns between males and females. Gout, as well as malignant neoplasm of the bronchus and lung, were more prevalent in males, while other rheumatoid arthritis and other anxiety disorders exhibited higher prevalence rates in females. A community-based study on mild COPD patients in Shanghai found that the prevalence of anxiety disorders in women was 6.41 times higher than in men (12.8% vs. 2.8%)⁵⁸. In our study, the prevalence of other anxiety disorders was 2.07 times higher in women than in men (3.50% vs. 1.69%). These sex differences in anxiety prevalence among COPD patients may be related to the greater activity of neural circuits involved in conflict-related anxiety regulation in women, and the heightened sensitivity of midbrain regions to estrogen⁵⁹. In addition, the prevalence of other anemias was higher in females, and the comorbidity strength between other anemias and osteoporosis without pathological fracture was also stronger in females, which may be related to anemias resulting from osteoporosis⁶⁰. We also found that there are differences in the central diseases of comorbidity networks between different genders. For instance, the chronic kidney disease and hyperplasia of the prostate were central diseases in the male comorbidity networks of multiple subgroups, while in female networks, other intervertebral disc disorders and sleep disorders were central diseases. These diseases may have sex-specific pathophysiological mechanisms, which could contribute to their differing patterns of comorbidity with other diseases.

Such sex-based comorbidity patterns may not only reflect clinical factors but also deeper biological mechanisms. A prior study has shown that genotype-by-sex (GxS) interactions can shape cross-phenotype disease associations⁴⁷. For example, certain SNPs exhibit sex-specific effects on the risk of metabolic or autoimmune diseases, which may in turn contribute to distinct comorbidity structures. In future studies, integrating genetic data may help to further elucidate the biological underpinnings of these sex-specific comorbidity patterns.

Urban–rural differences were also observed in both the prevalence and strength of comorbidity relationships. In our study, urban COPD patients demonstrated higher prevalence rates for most comorbidities compared to the rural patients, including both high-prevalence comorbidities (e.g., hypertension, disorders of glycoprotein metabolism) and lower-prevalence comorbidities (e.g., oesophagitis, Parkinson’s disease). Urogenital diseases, including hyperplasia of the prostate and other disorders of the kidney and ureter, were more prevalent among urban patients and exhibited stronger comorbidity associations. These differences may be attributed to lifestyle differences between urban and rural populations, as well as less timely screening and treatment for reproductive system diseases in rural areas⁶¹.

Our study had some limitations. First, this study only included COPD inpatients. Therefore, it is crucial to interpret our results within the context of hospitalized patients with COPD in China. In future studies, comparing comorbidity patterns across different baseline populations, such as those from other healthcare systems or demographic backgrounds, may help assess the generalizability of current findings. Second, we excluded individuals who died during the study period to ensure a more homogeneous sample, which may underestimate the complexity of certain comorbidities. However, previous studies have indicated that this exclusion criterion did not significantly affect the results^62,63. In addition, the comorbidity network we constructed is undirected, so it does not allow us to determine causal relationships between comorbidities. Using longitudinal data to construct directed networks or multi-layer temporal networks could provide insights into potential causal relationships and predictive risk of diseases. Finally, our study is based on phenotype disease networks and lacks exploration of comorbidity mechanisms. In future studies, other forms of data, such as genomic information or social factors, can be integrated to reveal deeper mechanisms underlying potential comorbidity patterns.

Despite these limitations, this study also has several strengths. First, our study applied network theory and ICD-10 codes extracted from administrative discharge data to investigate COPD comorbidities, systematically exploring a broad spectrum of diseases rather than limited common conditions. This approach enabled us to identify potential comorbid relationships that may have been overlooked in previous studies, providing a scientific basis for early intervention. Moreover, our data was based on large-scale regional data rather than data collected from a few hospitals, which helped avoid bias caused by sampling in previous research results. Furthermore, to the best of our knowledge, this is the first comprehensive study in China revealing the comorbidity patterns in COPD inpatients. This study helps fill the gap in comprehensive research on comorbidities in Chinese COPD patients. Our analysis has generated some new insights by identifying central diseases, detecting disease communities, and comparing subgroups by gender and region, contributing to the development of more precise comorbidity management strategies and providing valuable information for improving the health outcomes of COPD patients.

Conclusion

In conclusion, this study conducted a systematic analysis of comorbidities among hospitalized COPD patients in Sichuan Province, China, providing an overview of comorbid conditions commonly observed in routine inpatient care. We identified comorbidities that significantly influence other diseases and found highly interconnected comorbidity clusters, offering new insights into comorbidity patterns. The comorbidity relationships of COPD patients were more complex among male, urban, and older patients. Research on comorbidities based on network analysis methods and large-scale data enhances the understanding of complex disease relationships, complements gaps in traditional clinical research methods, and supports the development of targeted preventive health strategies.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Y.L. designed the study, performed the experiments, analyzed the data and wrote the first draft of the manuscript. H.Q. conceived the study and revised the manuscript. All authors have read and approved the final manuscript.

Funding

This work was supported in part by the National Science and Technology Major Project of China (2024ZD0523903), the National Natural Science Foundation of China (72342014), and the Key Research Project of Health Information Center of Sichuan Province (2023ZXKY06001).

Data availability

The data that support the findings of this study are available from the Health Information Center of Sichuan Province, but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. The data are however available from corresponding author on reasonable requests and with permission of the Health Information Center of Sichuan Province.

Code availability

The source code for network construction using the Salton Cosine Index is available at: https://anonymous.4open.science/r/Disease-Network-Analysis-CF60.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

This study was performed in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Health Information Center of Sichuan Province (Ethics Approval Number: ZX-EC202300101). All data analyses were conducted within the Health Information Center of Sichuan Province. The requirement to obtain informed consent was waived by the Ethics Committee of Health Information Center of Sichuan Province because of the secondary nature of the de-identified data in the retrospective study design.