Impact of multimorbidity patterns in hospital admissions: the case study of asthma

Abstract

Background: Most previous studies assessing multimorbidity in asthma assessed the frequency of individual comorbid diseases. Objective: We aimed to assess the frequency and clinical and economic impact of co-occurring groups of comorbidities (comorbidity patterns using the Charlson Comorbidity Index) on asthma hospitalizations. Methods: We assessed the dataset containing a registration of all Portuguese hospitalizations between 2011–2015. We applied three different approaches (regression models, association rule mining, and decision trees) to assess both the frequency and impact of comorbidities patterns in the length-of-stay, in-hospital mortality and hospital charges. For each approach, separate analyses were performed for episodes with asthma as main and as secondary diagnosis. Separate analyses were performed by participants’ age group. Results: We assessed 198340 hospitalizations in patients >18 years old. Both in hospitalizations with asthma as main or secondary diagnosis, combinations of diseases involving cancer, metastasis, cerebrovascular disease, hemiplegia/paraplegia, and liver disease displayed a relevant clinical and economic burden. In hospitalizations having asthma as a secondary diagnosis, we identified several comorbidity patterns involving asthma and associated with increased length-of-stay (average impact of 1.3 [95%CI=0.6–2.0]–3.2 [95%CI=1.8-4.6] additional days), in-hospital mortality (OR range=1.4 [95%CI=1.0–2.0]–7.9 [95%CI=2.6–23.5]) and hospital charges (average additional charges of 351.0 [95%CI=219.1–482.8] to 1470.8 [95%CI=1004.6–1937.0]) Euro compared with hospitalizations without any registered Charlson comorbidity). Consistent results were observed with association rules mining and decision tree approaches. Conclusions: Our findings highlight the importance not only of a complete assessment of patients with asthma, but also of considering the presence of asthma in patients admitted by other diseases, as it may have a relevant impact on clinical and health services outcomes.

Keywords:

• Economic impact
• asthma multimorbidity patterns
• intelligent data analysis

Introduction

Multimorbidity is defined as the co-occurrence of two or more chronic health conditions within one person having a relevant burden on clinical and economic outcomes.(1–5) Current literature has assessed the burden of multimorbidity patterns on specific diseases such as chronic obstructive pulmonary disease (COPD), motor neuron disease or asthma.(5–8) Regarding the latter, multimorbidity has been studied mainly related to the presence of other allergic diseases.(5) This may be insufficient since the presence of non-allergic comorbidities alongside asthma may also be related to a poorer prognosis or to a worse impact of asthma episodes.(9–11) For example, a recent study emphasized the incremental costs for comorbidities associated with asthma and highlighted that treating associated diseases alongside asthma associated with improved asthma outcomes and overall health.(12) However, that study only defined multimorbidity based on the count of chronic diseases and did not consider their nature or how their patterns or combinations may impact the outcomes. In addition, we do not know the impact of asthma as a comorbidity on hospital admissions due to other conditions.(12) This is particularly relevant not only because of the high prevalence of asthma but also because individuals with asthma are more susceptible to infections and non-communicable chronic comorbidities which are associated with worse health outcomes.(10,11)

Therefore, our study aims to describe the relationship between the co-occurrence of multiple diseases and the clinical and economic outcomes on Portuguese asthma hospitalizations.

Methods

Study design

In this study, we assessed the impact of comorbidities patterns in the length of stay (LOS), in-hospital mortality, and hospital charges in hospitalizations occurring in adult inpatients (above 18 years old) with asthma as main or as secondary diagnosis. We assessed the Portuguese nationwide hospitalizations dataset and applied regression models to assess the impact of patterns of comorbidities from the Charlson Comorbidity Index (CCI) (a validated classification system frequently used for health services research).(13) In addition, we identified association rules between comorbidities, so as to assess whether those co-occurring comorbidities with the highest impact were also those more frequent and/or more strongly associated among themselves. Finally, we assessed the consistency of our results by building decision trees identifying those comorbidity patterns with the highest impact.

Data sources and variables

We extracted data from the Portuguese National Hospital Morbidity Database, which was provided by the Central Administration of the Health System (Administração Central do Sistema de Saúde) and contains data from all hospitalizations occurring in public hospitals from Mainland Portugal. For each hospitalization episode, administrative information (e.g. admission and discharge dates, and discharge outcome), patient demographic variables (e.g. sex, birth date and parish of residence) and information on the main and secondary diagnoses are provided. Diagnoses were coded according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) standards. Charges were calculated taking into consideration the expenditure tables for the Portuguese National Health Service hospital reimbursements and the diagnosis-related groups (DRG)-based budget allocation model.(14)

We retrieved data for both hospital admissions (i) with asthma as primary diagnosis (i.e. main cause of admission) and (ii) with asthma as secondary diagnosis (we defined asthma diagnostic group using the “493.x” ICD-9-CM code).(15) For hospitalizations having a primary diagnosis of asthma, we selected all episodes occurred between 2011–2015. For hospitalizations having asthma as a secondary diagnosis, we selected all 2011-2015 episodes from patients born in 20 randomly-selected birth weeks as correcting for potentially under-coding is computationally demanding.(16) The sample was selected based on a probabilistic gamma distribution which corresponded to the one fitting the years of birth density distribution of all hospitalizations.

Considering that, in the assessed database, the patient identifier number changes every year, we applied a hierarchical clustering algorithm (HCA) + k-means clustering method to identify hospitalization episodes with high probability of being from the same individual patients.(16) This allowed us for correcting for potential under-coding of comorbidities. Comorbidities coded as ICD-9-CM were grouped into the 17 Charlson comorbidity groups which have been previously validated in studies that use administrative health data and widely adopted by the scientific community.(13,17–21)

Ethics

This is a secondary data study that uses an anonymised administrative dataset, conducted in accordance with privacy and data protection principles and regulations and without requirements for an ethical approval.

Data analysis

Data analysis encompassed three different steps – (i) application of regression models to identify the combinations of comorbidities associated with the highest impact on the episodes’ LOS, in-hospital mortality and hospital charges (by comparing hospitalizations with these combinations of comorbidities with hospitalizations without any registered Charlson comorbidity), (ii) association rule mining to assess the relationship and frequency of comorbidities patterns, and (iii) obtention of decision trees to further assess the most impactful comorbidity patterns, and thus assess the consistency of the results obtained with regression models. For each step, separate analyses were performed for episodes with asthma as main and as secondary diagnosis. In addition, within each group of hospitalization episodes, separate analyses were performed by participants’ age group. In particular, we considered the age groups 18-49, 50-64, 65-79, ≥80 years old based on previous research on multimorbidity patterns.(22,23) A detailed description of the used approaches is provided in supplementary section: detailed methods description.

Data analysis was carried out using R software Statistical significance was considered at p < 0.05.

Results

We assessed 74 761 hospitalizations with asthma as main diagnosis, and 123 579 hospitalizations with asthma as secondary diagnosis. A descriptive analysis of both samples is presented in Supplementary Table 1.

Given the large volume and complexity of obtained results, we developed an interactive tool allowing consultation of the complete results of our study (available at https://dianaportelasilva.github.io/asthma-hospitalizations-costs/?key=bzmebtkacv; with the following key: bzmebtkacv). Herein, in the text and tables of the manuscript, only the main results will be presented.

Multivariable regression models

We built multivariable regression models to identify the combinations of comorbidities associated with increased LOS, in-hospital mortality and hospitalization charges. Figure 1 provides the descriptive results for the different outcomes in multimorbidity and non-multimorbidity episodes.

Figure 1 Overall description of the outcomes of the episodes with and without multimorbidity according to age group.

Figure 2 Diagram illustrating the association rules’ selection process.

Tables 1-3 and supplementary tables 2-4 list the top 10 comorbidity patterns with the highest impact on the LOS, in-hospital mortality and hospital charges per age group. Overall, oncologic disease (cancer and metastasis), hepatic disease and hemiplegia/paraplegia were the conditions most frequently present among the most impactful comorbidity patterns. Comorbidity patterns involving these conditions displayed a relevant impact throughout all age groups, even though the preponderance of oncologic disease was particularly higher in older patients. The impact of comorbidity patterns involving cardiovascular diseases was at its highest in middle-aged adults, while for those involving HIV such occurred in young adults.

Table 1 Top 10 association of the combinations of groups of diseases defined in the Charlson comorbidity index groups with length of stay stratified according to age class for hospitalizations with asthma as main diagnosis.

Combinations of groups of diseases defined in the Charlson comorbidity defined groups	beta (95%CI) [p-value]
A. 18-49 years old
Hemiplegia or paraplegia	7.8 (6.9-8.7) [<0.001]
Cerebrovascular disease + Hemiplegia or paraplegia	6.9 (5.5-8.4) [<0.001]
Cerebrovascular disease + CPD	6.2 (4.9-7.5) [<0.001]
HIV	5.9 (4.8-7.0) [<0.001]
Diabetes with chronic complication	5.9 (4.5-7.2) [<0.001]
Diabetes without chronic complication + Renal disease	5.8 (4.2-7.4) [<0.001]
CHF + Renal disease	5.3 (3.9-6.8) [<0.001]
Moderate or severe liver disease	4.5 (3.1-5.9) [<0.001]
CPD + Hemiplegia or paraplegia	4.4 (2.9-6.0) [<0.001]
Cerebrovascular disease	4.4 (3.6-5.2) [<0.001]
B. 50-64 years old
HIV	5.9 (3.1-8.8) [<0.001]
Diabetes without chronic complication + Cancer + Metastasis	5.6 (2.7-8.5) [<0.001]
Diabetes without chronic complication + Metastasis	5.6 (3.0-8.2) [<0.001]
Cerebrovascular disease + Hemiplegia or paraplegia	4.3 (2.4-6.1) [<0.001]
Diabetes without chronic complication + Hemiplegia or paraplegia	4.0 (1.3-6.6) [0.004]
Hemiplegia or paraplegia	3.6 (2.2-5.0) [<0.001]
Cerebrovascular disease + Renal disease	3.6 (1.3-5.9) [0.003]
Hemiplegia or paraplegia + Cancer	3.5 (0.6-6.3) [0.017]
Diabetes without chronic complication + Cancer	3.4 (1.8-5.1.0) [<0.001]
PVD + Renal disease	3.4 (0.8-6) [0.009]
C. 65-79 years old
CHF + CPD + Hemiplegia or paraplegia	7.7 (5.8-9.5) [<0.001]
Hemiplegia or paraplegia + Cancer	6.8 (5.0-8.6) [<0.001]
Mild liver disease + Cancer	6.4 (4.8-8.1) [<0.001]
CHF + Hemiplegia or paraplegia	6.3 (4.9-7.6) [<0.001]
Peptic ulcer disease + Diabetes without chronic complication	5.8 (3.9-7.7) [<0.001]
CHF + Diabetes without chronic complication + Hemiplegia or paraplegia	5.7 (3.9-7.5) [<0.001]
CPD + Hemiplegia or paraplegia	5.5 (4.2-6.9) [<0.001]
CHF + Peptic ulcer disease	5.5 (3.5-7.5) [<0.001]
PVD + Diabetes without chronic complication + Diabetes with chronic complication	5.2 (3.7-6.7) [<0.001]
CHF + Metastasis	5.1 (3.4-6.8) [<0.001]
D. More than 80 years old
Cerebrovascular disease + Metastasis	14.4 (7.9-26.3) [<0.001]
Cerebrovascular disease + Cancer + Metastasis^[†]	13.0 (6.9-24.4) [<0.001]
Metastasis	10.0 (7.3-13.7) [<0.001]
Cancer + Metastasis^[†]	9.8 (7.0-13.6) [<0.001]
Diabetes without chronic complication + Metastasis	8.5 (5.3-13.5) [<0.001]
Diabetes without chronic complication + Cancer + Metastasis	7.8 (4.7-13.0) [<0.001]
Cerebrovascular disease + Cancer	6.8 (4.3-10.8) [<0.001]
MI + Cancer	6.6 (3.4-12.7) [<0.001]
CPD + Metastasis	6.5 (3.8-10.9) [<0.001]
CHF + Metastasis	6.4 (3.6-11.5) [<0.001]

Legend: ^[†] Combinations of diseases included in the Top 10 of the evaluated association rules; MI - Myocardial infarction; CHF - Congestive heart failure; CPD - Chronic pulmonary disease; PVD - Peripheral vascular diseases.

Table 2 Top 10 association of the combinations of groups of diseases defined in the Charlson comorbidity index groups with in-hospital mortalitystratified according to age class for hospitalizations with asthma as main diagnosis.

Combinations of groups of diseases defined in the Charlson comorbidity defined groups	OR (95%CI) [p-value]
A. 18-49 years old
Cancer + Metastasis	49.8 (26.9-92.2) [<0.001]
Metastasis	46.6 (25.4-85.3) [<0.001]
Cancer	20.4 (12-34.7) [<0.001]
Diabetes without chronic complication + Cancer	15.0 (5.7-39.5) [<0.001]
HIV	14.0 (6.7-29.2) [<0.001]
Moderate or severe liver disease	12.7 (5.1-31.4) [<0.001]
Mild liver disease + Moderate or severe liver disease ^[†]	9.3 (3.4-25.7) [<0.001]
CPD + HIV ^[†]	8.6 (3.1-23.9) [<0.001]
Cerebrovascular disease	8.6 (4.5-16.4) [<0.001]
Hemiplegia or paraplegia	8.6 (4.6-16.2) [<0.001]
B. 50-64 years old
Diabetes without chronic complication + Cancer + Metastasis	30.1 (14.8-61.2) [<0.001]
Diabetes without chronic complication + Metastasis	24.1 (12.2-47.7) [<0.001]
Cancer + Metastasis	20.5 (13.3-31.6) [<0.001]
Metastasis	18.8 (12.3-28.6) [<0.001]
Moderate or severe liver disease	16.0 (8.2-31.1) [<0.001]
Mild liver disease + Moderate or severe liver disease	14.7 (7.3-29.7) [<0.001]
Hemiplegia or paraplegia + Cancer	12.6 (6.1-26.1) [<0.001]
Diabetes without chronic complication + Cancer	12.1 (6.6-22.1) [<0.001]
Cancer	11.6 (7.7-17.5) [<0.001]
CPD + Cancer + Metastasis	11.3 (5.6-22.5) [<0.001]
C. 65-79 years old
Cerebrovascular disease + Metastasis	14.4 (7.9-26.3) [<0.001]
Cerebrovascular disease + Cancer + Metastasis	13.0 (6.9-24.4) [<0.001]
Metastasis	10.0 (7.3-13.7) [<0.001]
Cancer + Metastasis	9.8 (7.0-13.6) [<0.001]
Diabetes without chronic complication + Metastasis	8.5 (5.3-13.5) [<0.001]
Diabetes without chronic complication + Cancer + Metastasis	7.8 (4.7-13.0) [<0.001]
Cerebrovascular disease + Cancer	6.8 (4.3-10.8) [<0.001]
MI + Cancer	6.6 (3.4-12.7) [<0.001]
CPD + Metastasis	6.5 (3.8-10.9) [<0.001]
CHF + Metastasis	6.4 (3.6-11.5) [<0.001]
D. More than 80 years old
Metastasis	2.8 (2.0-4.1) [<0.001]
Cancer + Metastasis ^[†]	2.8 (1.9-4.1) [<0.001]
Cancer	1.8 (1.4-2.4) [<0.001]
Diabetes without chronic complication + Metastasis	1.8 (1.1-3.0) [0.026]
Diabetes without chronic complication + Cancer + Metastasis	1.7 (1.0-3.0) [0.049]

Legend: ^[†] Combinations of diseases included in the Top 10 of the evaluated association rules; MI - Myocardial infarction; CHF - Congestive heart failure; CPD - Chronic pulmonary disease; PVD - Peripheral vascular disease.

Table 3 Top 10 association of the combinations of groups of diseases defined in the Charlson comorbidity index groups with hospital charges stratified according to age class for hospitalizations with asthma as main diagnosis.

Combinations of groups of diseases defined in the Charlson comorbidity defined groups	beta (95%CI) [p-value]
A. 18-49 years old
CPD + Cancer + Metastasis	2724.9 (2252.1-3197.6) [<0.001]
Renal disease + Cancer + Metastasis	2649.3 (2199.6-3099.0) [<0.001]
Mild liver disease + Cancer	2644.2 (2227.8-3060.5) [<0.001]
Renal disease + Metastasis	2522.5 (2101.8-2943.2) [<0.001]
CPD + Metastasis	2460.5 (2034.7-2886.2) [<0.001]
Cancer + Metastasis	2321.5 (2137.2-2505.8) [<0.001]
CHF + Cancer + Metastasis	2314.8 (1882.5-2747.2) [<0.001]
Cerebrovascular disease + Metastasis	2189.2 (1742.2-2636.2) [<0.001]
CHF + Metastasis	2173.4 (1793-2553.7) [<0.001]
Diabetes without chronic complication + Cancer + Metastasis	2162.4 (1807.1-2517.6) [<0.001]
B. 50-64 years old
CHF + PVD + Renal disease	2836.4 (2312.8-3360.0) [<0.001]
CHF + PVD	2379.9 (1942.5-2817.3) [<0.001]
PVD + Renal disease	2314.1 (1861.5-2766.7) [<0.001]
Moderate or severe liver disease	2282.6 (1894.4-2670.9) [<0.001]
Mild liver disease + Moderate or severe liver disease	2252.9 (1843.1-2662.7) [<0.001]
Hemiplegia or paraplegia + Cancer	2075.4 (1659.2-2491.6) [<0.001]
Cerebrovascular disease + Cancer	1930.5 (1539.2-2321.8) [<0.001]
CPD + Hemiplegia or paraplegia	1879.9 (1439-2320.8) [<0.001]
Cerebrovascular disease + Hemiplegia or paraplegia	1827.7 (1507.6-2147.9) [<0.001]
PVD + Cerebrovascular disease	1818.8 (1465.2-2172.4) [<0.001]
C. 65-79 years old
CPD + Hemiplegia or paraplegia	2435.2 (2099.5-2771.0) [<0.001]
Hemiplegia or paraplegia + Cancer	2328.8 (1972.5-2685.2) [<0.001]
Diabetes without chronic complication + Hemiplegia or paraplegia + Renal disease	2186.2 (1787.0-2585.4) [<0.001]
Hemiplegia or paraplegia	1970.3 (1737.0-2203.7) [<0.001]
Cerebrovascular disease + Metastasis	1910.9 (1610.9-2210.9) [<0.001]
CHF + Peptic ulcer disease	1864.9 (1500.5-2229.3) [<0.001]
Mild liver disease + Diabetes without chronic complication + Renal disease	1855.0 (1489.1-2220.9) [<0.001]
Cerebrovascular disease + Cancer + Metastasis	1823.0 (1501.1-2144.9) [<0.001]
Hemiplegia or paraplegia + Renal disease	1703.7 (1370.3-2037.2) [<0.001]
CPD + Diabetes without chronic complication + Hemiplegia or paraplegia	1689.7 (1316.5-2063.0) [<0.001]
D. More than 80 years old
Cerebrovascular disease + CPD + Mild liver disease	1871.3 (1387.1-2355.5) [<0.001]
CHF + Cerebrovascular disease + Mild liver disease	1863.6 (1455.4-2271.7) [<0.001]
Cerebrovascular disease + Mild liver disease + Diabetes without chronic complication	1830.0 (1303.8-2356.2) [<0.001]
Cerebrovascular disease + Mild liver disease	1807.1 (1368.3-2246.0) [<0.001]
Mild liver disease + Cancer	1716.5 (1177.4-2255.5) [<0.001]
MI + CPD	1569.7 (1105.3-2034.1) [<0.001]
MI	1436.5 (1072.4-1800.7) [<0.001]
CPD + Renal disease + Cancer	1410.5 (1055.0-1766.0) [<0.001]
Metastasis	1400.8 (1153.4-1648.3) [<0.001]
Cerebrovascular disease + Diabetes without chronic complication + Cancer ^[†]	1379.7 (913.5-1846.0) [<0.001]

Legend: ^[†] Combinations of diseases included in the Top 10 of the evaluated association rules; MI - Myocardial infarction; CHF - Congestive heart failure; CPD - Chronic pulmonary disease; PVD - Peripheral vascular disease.

Top 10 comorbidity patterns associated with increased LOS, in-hospital mortality or hospital charges were not always necessarily the same – we found a probability of 22.0% (episodes with asthma as main diagnosis) or of 40.7% (hospitalizations with asthma as secondary diagnosis) for any comorbidity pattern to be present in the top 10 of at least two of the defined outcomes. Probabilities were as low as 2.2% and 3.7% when considering the presence in the three defined outcomes.

In hospitalizations having asthma as a secondary diagnosis, asthma itself did not appear in any of the top 10 comorbidity patterns with the highest impact on LOS, in-hospital mortality or hospitalizations charges. Nevertheless, we identified several comorbidity patterns involving asthma (isolated or combined with other comorbidities) and associated with increased LOS (average impact of 1.3–3.2 additional days), in-hospital mortality (OR range = 1.4–7.9) and hospital charges (average impact of 351.0–1025.8 Euro) (Tables 4–5 and supplementary Table 5). Except for the comorbidity patterns associated with in-hospital mortality in the elderly, in all remaining comorbidity patterns asthma was always associated with increased LOS, in-hospital mortality or hospitalizations charges. Overall, the top 10 more impactful comorbidity patterns were quite different between hospitalizations with asthma as a main versus as a secondary diagnosis (with only 27% patterns being common to both groups of hospitalizations).

Table 4 Association of the combinations of groups of diseases defined in Charlson comorbidity index groups containing Asthma with in-hospital mortality stratified according to age class for hospitalizations with asthma as secondary diagnosis.

Combinations of groups of diseases defined in the Charlson comorbidity defined groups	OR (95%CI) [p-value]
A. 18-49 years old
Asthma	5.6 (2.1-15.3) [0.001]
CPD + Asthma	5.6 (2.1-15.3) [0.001]
B. 50-64 years old
CPD + Cancer + Asthma	7.9 (2.6-23.5) [<0.001]
Cancer + Asthma	7.9 (2.6-23.0) [<0.001]
Diabetes without chronic complication + Asthma	2.9 (1.0-8.4) [0.048]
CPD + Diabetes without chronic complication + Asthma	2.9 (1.0-8.4) [0.048]
Asthma	2.2 (1.2-4.1) [0.011]
CPD + Asthma	2.2 (1.2-4.1) [0.011]
C. 65-70 years old
CPD + Cancer + Asthma	2.2 (1.1-4.6) [0.030]
Cancer + Asthma	2.2 (1.1-4.6) [0.030]
Asthma	1.4 (1.0-2.0) [0.048]
CPD + Asthma	1.4 (1.0-2.0) [0.048]
D. More than 80 years old
Asthma	0.7 (0.5-1.0) [0.041]
CPD + Asthma	0.7 (0.5-1.0) [0.041]

Legend: MI - Myocardial infarction; CHF - Congestive heart failure; CPD - Chronic pulmonary disease; PVD - Peripheral vascular disease.

Table 5 Association of the combinations of groups of diseases defined in Charlson comorbidity index groups containing Asthma with hospital charges stratified according to age class for hospitalizations with asthma as secondary diagnosis.

Combinations of groups of diseases defined in the Charlson comorbidity defined groups	beta (95%CI) [p-value]
A. 18-49 years old
Asthma	351.0 (219.1-482.8) [<0.001]
CPD + Asthma	351.0 (219.1-482.8) [<0.001]
B. 50-64 years old
CPD + Cancer + Asthma	1025.8 (618.1-1433.5) [<0.001]
Cancer + Asthma	1025.8 (618.1-1433.5) [<0.001]
Asthma	402.0 (240.5-563.6) [<0.001]
CPD + Asthma	402.0 (240.5-563.6) [<0.001]
C. 65-70 years old
Cerebrovascular disease + Asthma	1470.8 (1004.6-1937.0) [<0.001]
Cerebrovascular disease + CPD + Asthma	1470.8 (1004.6-1937.0) [<0.001]
CHF + Asthma	1252.5 (921.9-1583.1) [<0.001]
CHF + CPD + Asthma	1252.5 (921.9-1583.1) [<0.001]
Diabetes without chronic complication + Asthma	1211.7 (904.4-1518.9) [<0.001]
CPD + Diabetes without chronic complication + Asthma	1211.7 (904.4-1518.9) [<0.001]
Renal disease + Asthma	987.9 (574.5-1401.3) [<0.001]
CPD + Renal disease + Asthma	987.9 (574.5-1401.3) [<0.001]
CPD + Cancer + Asthma	856.0 (481.0-1231.0) [<0.001]
Cancer + Asthma	856.0 (481.0-1231.0) [<0.001]
Asthma	624.3 (459.2-789.5) [<0.001]
CPD + Asthma	624.3 (459.2-789.5) [<0.001]
D. More than 80 years old
Diabetes without chronic complication + Asthma	639.9 (272.9-1006.8) [0.001]
CPD + Diabetes without chronic complication + Asthma	639.9 (272.9-1006.8) [0.001]]
CHF + Asthma	634.4 (320.9-947.9) [<0.001]
CHF + CPD + Asthma	634.4 (320.9-947.9) [<0.001]
Asthma	512.4 (298.1-726.7) [<0.001]
CPD + Asthma	512.4 (298.1-726.7) [<0.001]

Legend: MI - Myocardial infarction; CHF - Congestive heart failure; CPD - Chronic pulmonary disease; PVD - Peripheral vascular disease.

Association rules

We assessed association rules as a pattern mining strategy to identify how frequent and strongly associated were the combinations of comorbidities with the highest impact on the assessed outcomes. We obtained a total of 6255 association rules (Figure 2), with an average confidence of 0.46 for patterns on hospitalizations with asthma as main diagnosis and of 0.38 for patterns on hospitalizations with asthma as secondary diagnosis. From the top 10 identified combinations in regression models in each age group, we identified 10 combinations of diseases included on the top 10 of the evaluated association rules (among rules with the highest score of confidence and lift). Overall, the results highlighted the strong association on patterns of combinations that include comorbidities such as cancer, metastasis, cerebrovascular disease, and mild liver disease comorbidities (supplementary web page).

Decision trees

In our regression decision tree algorithm, for both hospitalizations with asthma as main or secondary diagnosis, hemiplegia/paraplegia was the most frequent comorbidity on the first node-attribute to split the data (corresponding to the feature with most discriminative capacity in the decision tree) (supplementary web page). Cerebrovascular disease was most commonly the last node with any decision capacity on decision trees for hospitalization charges. In contrast, LOS leaf nodes were mostly affected by cancer and hemiplegia/paraplegia.

Discussion

In this study, we applied different approaches to assess the impact of different comorbidities patterns in the LOS, in-hospital mortality and hospital charges of hospitalizations occurring in patients with asthma as main or as secondary diagnosis. Our results suggest that in those hospitalizations, specific combinations of diseases may have a relevant clinical and economic burden. In addition, our results point that, in hospitalizations by other diseases, the presence of asthma as a comorbidity associates with worse outcomes.

We observed that the combinations of CCI comorbidities with the strongest associations with the assessed outcomes frequently included diseases such as cancer, metastasis, cerebrovascular disease, hemiplegia/paraplegia, and liver disease. Previous studies have linked multimorbidity to increased risk of mortality, LOS and costs due to the potential synergistic effect resulting from the combination of multiple diseases.(12,24,25) However, it has also been reported that the number of chronic conditions is not always necessarily proportional to the mortality risk, pointing to the importance of the specific comorbidities the patient displays.(24)

Chang et al.(26) and Watson et al.(27) reported that non-allergic comorbidities such as pneumonia or cardiovascular disease were independently associated with an increased risk of death in patients admitted for asthma (particularly for asthma exacerbations). Notwithstanding, those studies have only assessed each comorbidity individually, without considering their combinations. Identifying such combinations is particularly relevant as the systemic inflammation associated with asthma predisposes certain diseases to co-exist in the same patient, leading to a worse prognosis.(12) According to Kankaanranta et al (28), in contrast to early-onset asthma (an predominantly atopic-induced asthma, where specific comorbidities – such as allergic rhinitis or eczema - are well characterized (28,29)), in late-onset asthma the mechanisms of asthma may include several metabolic and inflammatory components that are common to the other diseases such as diabetes mellitus type 2, cardiovascular diseases or psychiatric diseases.(29) For instance, asthma and COPD share many clinical features, which have been characterized as an asthma-COPD overlap syndrome (ACOS).(30) Interestingly, some comorbidities were found to be more frequent in ACOS than in COPD-only, probably due to systemic inflammation mechanisms.(8)

We have further observed that the most impactful combinations of comorbidities were not necessarily the most frequent ones, although there were cases in which that occurred. As with the impact of comorbidities, previous studies have only assessed the frequency of each individual comorbidity in asthma, rather than the frequency of combinations of diseases.(31,32) Moreover, comparisons with studies assessing multimorbidity patterns in patients with other diseases is also difficult, considering the heterogeneity in which the different conditions are defined.(33)

We observed that, when present as a secondary diagnosis, asthma (or comorbidity combinations including asthma) seemed to be associated with higher LOS, in-hospital mortality and hospital charges. This impact of asthma may stem from the occurrence of exacerbations during the hospital admission, which may prompt further medical care and worsen patients’ clinical condition. Such asthma-attributed burden may also partly stem from a combined effect of multiple conditions due to potential common inflammatory mechanisms. A notable exception in the deleterious impact of asthma concerned in-hospital mortality among the elderly: age has been associated with various health problems, and a phenomenon of concurrent events can result in death from other causes, thus explaining this apparent (but misleading) “protective effect of asthma”.(29)

Limitations

This study has some limitations. Firstly, we used secondary data which were originally developed for hospital reimbursement purposes, and which have intrinsic limitations in their use for research. Such constraints may include (i) potential under-coding of comorbidities (which was partly, but probably not fully corrected, since such database relies on an unique patient identification number to the clinical registration of the patient which changes annually even for episodes from the same individual) (16), (ii) possible inaccurate coding (34), and (iii) lack of information on relevant variables, including the need for intensive care unit admission or the actual hospitalization costs (we only had information on prices paid to hospitals for each hospital admission). Secondly, to define comorbidities, we relied on CCI groups; to define comorbidity patterns, we only assessed combinations of diseases with up to three comorbidities. While we opted for these approaches for feasibility purposes (allowing to overcome difficulties in handling individual ICD-9-CM codes or an even higher number of potential combinations of comorbidities), these may have resulted in potential loss of granularity in the definition of comorbidities or of their combinations. Thirdly, for regression models, our comparison group consisted of hospitalizations without any code used for definition of CCI comorbidities. This implies that the impact of combinations of comorbidities was only assessed in comparison with (i) episodes from “the healthiest patients” or (ii) the most under-coded episodes. Lastly, we did not have information on the severity of asthma or on the eventual occurrence of asthma exacerbations during the hospital admission. This may be relevant as costs in comorbidities in asthma are predominantly higher in patients with more severe asthma (12).

Strengths

This study has also several strengths. It has a nationwide scope and encompasses a 5-year period study. Moreover, we applied a combination of different analytical methods, thus allowing us to evaluate the robustness of our results. In addition, our findings provide novel insight regarding how comorbid chronic diseases may affect the clinical and economic burden of asthma and how asthma can impact the outcomes of hospitalizations by other causes (thus providing evidence to support the importance of considering other comorbidities on asthma hospitalizations and of asthma as a comorbidity on hospitalizations by other causes (10,12)).

Our findings have relevant clinical implications. We have observed that specific comorbidity patterns have a relevant impact in asthma admissions, pointing to the importance of considering patients’ comorbidities. Future studies may assess whether these comorbidity patterns associate with worse asthma control or higher risk of exacerbations, whether their proper management can improve asthma control, or whether the main diagnosis (in episodes with asthma as secondary diagnosis) could also potentially impact such outcomes. Our study also points to the deleterious impact of asthma as a comorbidity in admissions by other diseases possibly explained by the occurrence of exacerbations during the hospital admission (pointing to the importance of maintaining an adequate asthma control). This hypothesis should be further confirmed by future studies.

Conclusions

We provide evidence on how specific combinations of comorbidities have a particular relevant impact on the clinical and economic outcomes of asthma hospitalizations. Our findings highlight the importance not only of a complete assessment of patients with asthma, but also of taking into account the presence of asthma in patients admitted by other diseases. Therefore, this study further contributes to raise awareness on asthma impact even when present as a hospitalization’s secondary diagnosis.

Author contributions statement

1. Diana Portela

   • Writing a major portion of the paper;

   • Structuring study design, conducting analysis, interpreting the results.

2. Pedro Pereira Rodrigues

   • Substantial contributions to structuring the study design, organizing and conducting the statistical analysis;

   • Final approval of the published version.

3. Alberto Freitas

   • Revision and critical appraisal of conducted statistical analysis;

   • Revising results and its interpretation;

   • Critically revising the manuscript for important intellectual content.

4. Elísio Costa

   • Critically revising the manuscript for important intellectual content.

5. Jean Bousquet

   • Revision and critical appraisal of conducted statistical analysis;

   • Critically revising the manuscript for important intellectual content.

6. João Almeida Fonseca

   • Critically revising the manuscript for important intellectual content;

   • Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved

7. Bernardo Sousa Pinto

   • Substantial contributions to the conception or design of the work; the acquisition, analysis and interpretation of data for the manuscript;

   • Drafting the work and revising it critically for important intellectual content;

   • Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

All authors reviewed and approved the final version of the manuscript.

Acknowledgements

We acknowledge António Almeida who contributed on building the site presenting supplementary material.

Conflict of interests

The authors declare that they have no competing interest.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.