Introduction

The global burden of influenza involves seasonal epidemics and occasional pandemics.1 2 As poultry and swine farms have expanded and thus increased dynamic changes in influenza viruses via transmission among different host species, the future threat of novel influenza viruses leading to pandemics may be inevitable.3 However, accurately diagnosing influenza virus in suspected patients is a common challenge faced by all clinicians.2 Influenza virus infection may not be identified in many instances because influenza virus is only detectable for a short period of time and/or many people do not seek medical care until after the first few days of acute illness. Influenza-like illness (ILI) became a surrogate indicator for practical use in the timely surveillance and measurement of the disease burden of influenza.2 4 The aim of this study is to assess the prognosis of hospitalised patients with ILI, which includes common presentations of influenza: fever, sore throat, headache, musculoskeletal and gastrointestinal symptoms.5–7 The outcomes from ILI cases are mostly mild or self-limited, but a small percentage of them can develop into severe morbidity or mortality.2 The severe cases of ILI typically occur in the very young, the elderly and patients with chronic illness.8–14 It is essential to identify, at an early stage, the most vulnerable ILI patients with higher risk to death15 16 and provide clinical recommendations and public health prevention strategies to minimise fatality.

Elderly and young children, as well as patients with chronic underlying medical conditions, are typically associated with complications of respiratory infections.8 11 13 15 17 Clinicians and public health decision-makers are expecting better evidence that requires careful and thorough examining of the roles that comorbidities play in patients of different age groups. The better evidence aims to assist in the prognosis at the beginning of the first medical visit for patients presenting ILI.18–20 To date, global guidelines on the prevention and management of ILI-related complications are mostly consensus-based; thus, evidence-based guidelines about the risks of comorbidities have been under investigation.8 21 22

Free influenza vaccination programme in Taiwan started in 1998, for population aged over 65 years. The target vaccination population expanded year after year with the consideration of age, comorbidity and occupational exposure. In 2018, all adults over 50 years and paediatric population from 6 months to 18 years were recommended for vaccination. In addition, high-risk occupations in healthcare, livestock workers, live-poultry market workers and residents in long-term care facilities or prisons and jails were all covered. Furthermore, patients with obesity and chronic underlying diseases and pregnant women are recommended to receive vaccination after physician evaluation.23 In this study, using a nationwide cohort data, we aimed to conduct a comprehensive analysis on the age-specific comorbidities associated with fatality on hospitalised patients with ILI.

We believe that the identification of patients with high-risk comorbidities can lead to the support of clinical recommendations and guideline directions for public health policies, including the timely management of antibiotics/antiviral treatments20 and other preventions, such as influenza and pneumococcal immunisation3 21 and non-pharmaceutical preventive measures through health education.24 25

Methods

We conducted a database analysis using the National Health Insurance Research Database (NHIRD) containing records of approximately 1 million patients randomly selected from the 24 million beneficiaries of the National Health Insurance in Taiwan. The database collects encrypted claim records of emergency department (ED) patient, outpatient and inpatient data for billing purposes. The nationwide cohort study was conducted using the diagnostic codes of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).26 Hospitalised ILI patients from January 2005 to December 2010 were included for data analyses. The definition of ILI was based on Taiwan Emergency Department-Based Syndromic Surveillance System.4 We added six codes (382.9, 461.9, 465.8, 466.0, 487.1, 780.6) after comparing the code-based syndromic surveillance for ILI from the USA.27 The codes are presented in online supplementary table 1 . In the instances where the patient had been re-admitted to the hospital with ILI within 14 days from a previous discharge of ILI-related hospitalisation, the two admissions were combined as the same episode. The flow diagram of the selection process of the study subjects is shown in figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow diagram of the selection process of the study population. ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ILI, influenza-like illness; NHIRD, National Health Insurance Research Database in Taiwan.

Comorbidities included in the study analyses were selected after a literature review and a group discussion among physicians specialising in infection and emergency medicine, paediatrics, occupational medicine, and public health professionals. The comorbidities related to malignant diseases were defined, based on the records from the Taiwan Catastrophic Illness Card (TCIC) codes for malignant diseases both 12 months before and 6 months after the first ILI admission date. We included pregnant women when the first date of hospitalisation with ILI occurred up to 3 months before delivery/abortion. We included postpartum women when delivery/abortion occurred up to 3 months before the first date of ILI hospitalisation. Data on other type of comorbidities were collected through diagnoses from reimbursement records 12 months before the first date of ILI admission. A comorbidity was confirmed with at least one diagnosis of hospitalisation claim or through two or more diagnoses from outpatient and/or ED visits. Grouping of comorbidity with ICD-9-CM codes was modified from the Charlson et al,28 Deyo et al29 and Elixhauser et al30 measurements using the most frequently applied codes. For autoimmune disease, organ transplantation, congenital immunodeficiency and malignant, hepatic and mental diseases, we also confirmed the disease status through the TCIC information to improve specificity. Definition of each comorbidity is shown in online supplementary table 2. The ILI-associated fatalities were the main outcome of interest. Fatality case was defined by discharge information recorded in the NHIRD as ‘hospital mortality’ (discharge transfer code ‘4’). Additionally, we also included cases that discharged under critical conditions (discharge transfer code ‘A’), who also withdrew from the National Health Insurance and had not received any healthcare services in the following 12 months after hospital discharge. All analyses were performed in the overall study population, as well as stratified by three age groups: paediatric patients (0–17 years), adult patients (18–64 years) and elderly patients (≧65 years).

In total, 25 clinically important comorbidities were selected after panel discussion and logistic regression analyses were used to evaluate the significance of individual comorbidity. Furthermore, the outcome measures were stratified by influenza season months (December to April) and non-influenza season months31 to compare and evaluate the fatalities by different comorbidities in hospitalised ILI patients. Finally, we performed multivariable analyses to measure the independent predictability of death from these comorbidities among the three age groups. We used stepwise logistic regression analyses with forward selection method at the entry criteria of p<0.15 to achieve a good model fit.32 33 Stepwise logistic regression analyses selected 16 comorbidities in the ‘all patients group’, 9 in the ‘paediatric patient group’, 13 in the ‘adult patient group’ and 17 in the ‘elderly patient group’. We performed generalised estimating equation (GEE) models to adjust for the hospital clustering effect and repeated hospitalisations as patients had two or more ILI-related admissions within an ILI episode.32 The variables adjusted in the GEE model included age, sex and different study periods (from previous year’s July to next year’s June), patient-admitted hospital levels (local hospital, regional hospital or medical centre), and prior frequencies of hospitalisation in the past 12 months (0, 1–2 or >3 admissions). The statistical analyses were performed using SAS V.9.4 (SAS Institute, Cary, North Carolina, USA). A two-tailed p<0.05 was considered statistically significant.

Results

In total, 83 227 events of ILI-related hospitalisation from 1 January 2005 to 31 December 2010 were included for analysis. The demographic data are shown in table 1. There were significantly more male than female ILI cases with a ratio of 1.49 (p<0.001). The majority of ILI-related admissions were elderly patients (44.93%), followed by adult patients (31.23%), then paediatric patients (23.85%). We identified 5282 deaths (6.35%) in our study population, where the case fatality rate of these hospitalised ILI cases was significantly higher among elderly patients (11.40%) than among adult (3.79%) and paediatric patients (0.19%) (p<0.001). Among the overall comorbidities, chronic obstructive pulmonary disease (COPD)/asthma (39.23%) was the top prevalent comorbidity, followed by hypertension (36.84%), heart disease (26.16%), cerebrovascular accident (CVA) (20.35%) and diabetes (19.72%).

Logistic regression analyses were used to evaluate the significance of the 25 comorbidities relating to ILI-associated fatality in patients of different age groups. The case fatality rate and univariate OR of each comorbidity are summarised in table 2. Ten comorbidities showed small difference in proportions between influenza and non-influenza seasons among hospitalised ILI cases with statistical differences in table 3. Among them, three comorbidities (cardiovascular diseases, dementia/epilepsy and tuberculosis) demonstrated slightly higher case fatality rates during influenza seasons than those in non-influenza seasons with statistical significance.

The results of the multivariable analysis by GEE models are shown in figure 2. In all patients, the highest adjusted OR (aOR) of ILI-associated fatality was in those patients with AIDS (aOR 6.53, 95% CI: 2.89 to 14.76), followed by those with metastatic cancer (aOR 3.49, 95% CI: 3.16 to 3.86), haematological malignancy (aOR 3.02, 95% CI: 2.43 to 3.74) and solid organ cancer without metastasis (aOR 1.72, 95% CI: 1.54 to 1.91). Other high-risk underlying medical conditions with statistical significance included chronic kidney disease (CKD) (aOR 1.57, 95% CI: 1.44 to 1.71), tuberculosis (aOR 1.37, 95% CI: 1.22 to 1.54), hypothyroidism (aOR 1.37, 95% CI: 1.04 to 1.79), CVA (aOR 1.24, 95% CI: 1.15 to 1.33), peripheral vascular disease (PVD) (aOR 1.22, 95% CI: 1.05 to 1.42), diabetes (aOR 1.22, 95% CI: 1.13 to 1.30), heart diseases (aOR 1.19, 95% CI: 1.11 to 1.27) and dementia/epilepsy (aOR 1.12, 95% CI: 1.03 to 1.22). However, hypertension (aOR 0.75, 95% CI: 0.70 to 0.81), hyperlipidaemia (aOR 0.69, 95% CI: 0.62 to 0.76) and allergic rhinitis (aOR 0.61, 95% CI: 0.54 to 0.70) showed decreased risk of ILI-related deaths.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Adjusted OR (aOR) of death in admitted patients with influenza-like illness among the three age groups and all patients in Taiwan, 1 January 2005 to 31 December 2010. CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; CVA, cerebrovascular accident; PVD, peripheral vascular disease.

For the risk of age-specific comorbidity, we further stratified our study population into three age groups. In paediatric patients, hypertension (aOR 23.82, 95% CI: 6.84 to 82.97) and dementia/epilepsy (aOR 4.87, 95% CI: 1.83 to 13.01) were distinctively associated with increased risk of ILI-associated deaths (p<0.05) (figure 2). In adult patients, AIDS (aOR 9.34, 95% CI: 4.18 to 20.91), CKD (aOR 1.85, 95% CI: 1.44 to 2.36), tuberculosis (aOR 1.89, 95% CI: 1.41 to 2.55) and diabetes (aOR 1.50, 95% CI: 1.26 to 1.80) displayed significantly increased risk of ILI-associated deaths (p<0.001), whereas hyperlipidaemia showed decreased risk (figure 2). In elderly patients, hypothyroidism (aOR 1.48, 95% CI: 1.11 to 1.98) and severe liver disease (aOR 1.94, 95% CI: 1.13 to 3.30) showed significantly increased risk of ILI-associated deaths (p<0.05) (figure 2).

Discussion

This study, using a large national cohort dataset with simultaneously adjusted clinically important comorbidities and several epidemiological covariates, is the first to demonstrate the independent effect of certain age-specific comorbidities on fatality risk among hospitalised ILI patients. Identifying high-risk patients with ILI or influenza to support evidence-based preventive strategies has been advocated in many high-income countries.8 13 Although meta-analyses have highlighted the important effect of comorbidities on influenza, most of these observational studies have either lacked power or inadequately adjusted for important covariates and thus provided limited evidence.8 11 13 Moreover, most of the published guidelines regarding ILI preventive strategies have targeted elderly populations or those over 50 years of age, thus limiting the applicability of these guidelines to other age groups.34–36 Furthermore, previous studies found that several comorbidities were associated with the increased fatality of ILI patients8 13; however, their analyses did not fully control for important covariates (e.g., age, sex and relevant comorbidities). Our results showed that the contribution of a comorbidity to fatality risk among hospitalised ILI patients was age-specific.

Host defence responses, targeting viral entry, transcription, translation, replication and extracellular release, can limit the disease severity and spread of influenza virus infection.37 We found that malignant diseases were associated with increased fatality rate among hospitalised ILI patients across all three age groups. The trends of aORs were also consistent in all three groups, with the highest trend being cancer with metastasis, followed by haematological malignancy and cancer without metastasis. Because cancer patients tend to have more depressed immunity directly from malignancy or indirectly from anticancer therapy-related immunodeficiency, our study again highlights the important role of malignant diseases, particularly cancer with metastasis, in contributing to increased fatality risk among hospitalised patients with ILI. Furthermore, heart disease and CVA were another two risky comorbidities across the three age groups. Patients with cardiovascular diseases were associated with increased risk of mortality during influenza epidemics, possibly due to the exacerbation of limited cardiovascular reserve from silent or fulminant viral myocarditis or the induction of acute myocardial infarction in patients with pre-existing coronary artery disease.38 39 The three comorbidities of cancer, heart disease and CVA with higher prevalence rates in all age groups provide evidence-based recommendations for clinical management as well as public health planning.

However, age-specific differences were present in several comorbidities associated with ILI-related fatality. AIDS was an independent predictor for death only in hospitalised adult ILI patients. Other chronic diseases, such as tuberculosis, diabetes and CKD, predicted hospital fatality in both adult and elderly patients and showed similar trends in increasing ILI-associated fatality. In fact, ILI patients with co-infection of HIV are associated with a higher risk of superimposed bacterial infections and increased severity of influenza.40–42 Patients with tuberculosis have also proven to be associated with increased influenza deaths in both the 1918 pandemic43 and seasonal influenza.44 Lost immune homeostasis of patients with CKD during influenza infection may result in further renal cell damage,45 eventually leading to kidney failure and subsequent fatality. In hospitalised ILI adult patients, metabolic diseases have more impact on deaths. A previous study found that diabetes tripled the risk of hospitalisation and quadrupled the risk of intensive care unit admission during the 2009 pandemic influenza.46 All these data illustrate the importance of considering preventive measures for age-specific immunocompromised patients to prevent subsequent ILI fatality.

Infants and young children with pre-existing conditions, particularly neurological disorders, are over-represented among influenza-associated paediatric deaths.47 48 However, previous studies were mostly conducted with univariable analysis without using multivariable analysis to adjust for other comorbidities; hence the level of evidence was low.8 11 Our study showed that epilepsy was a specific risk for paediatric patients. However, epilepsy was not documented as a risk in a previous meta-analysis.8 In addition, hypertension was another significant comorbidity risk only in paediatric ILI cases. The identification of 2 deaths among the 28 paediatric hypertension cases led to a high fatality rate. However, these two fatal cases also showed other comorbidities—one had leukaemia, while the other had moyamoya disease. Certainly, there was not enough evidence to derive any conclusions on hypertension in the paediatric group. Conversely, epilepsy/dementia and hypothyroidism were the specific high-risk comorbidities for elders in this study. A previous meta-analysis showed increased risk of epilepsy/dementia in seasonal and pandemic influenza patients and also indicated that endocrinological disease documented increased risk of death in seasonal influenza patients.8 However, this meta-analysis did not specify which endocrinological disease. By contrast, we addressed hypothyroidism to be the most probable disease predisposing fatality among the endocrinological diseases.

On the other hand, several comorbidities were associated with a reduced case fatality rate of ILI either for all age groups (allergic rhinitis) or in a specific age group (e.g., COPD/asthma in the elderly group). ILI patients with allergic rhinitis and COPD/asthma may be more likely to be hospitalised when they presented more prominent respiratory manifestations.49 50 Conversely, the non-infectious airway symptoms caused by allergic rhinitis and asthma/COPD could also mimic ILI, resulting in increased ILI-related hospitalisation. Our study showed a lower mortality risk for COPD/asthma in the study population of hospitalised ILI patients in Taiwan. We should carefully interpret the results since that risk was different from outpatient setting. The hospitalised patients in our study was 39.23% with COPD/asthma (table 1). However, the prevalence of COPD was about 6%,51 and the prevalence of asthma was about 11%52 among general population in Taiwan. Certainly, it is important to consider probable mitigation efforts, including vaccination and antiviral medication to alleviate disease burden.23 53 Adult and elderly ILI patients with hyperlipidaemia associated with lower case fatality rates could be explained by the use of statin to treat hyperlipidaemia. Pre-admission statin use has been shown to reduce 30-day mortality of less severe sepsis patients.54 Therefore, the use of statin in hyperlipidaemia patients may benefit the outcome of ILI. As hypertension is a common comorbidity in adult and elderly populations, our results adjusted for covariates and other comorbidities, thereby demonstrating that hypertension had no influence on adult ILI case fatality and decreased elderly ILI-related fatality. These findings are different from those of a previous meta-analysis showing that hypertension increased all-cause mortality in pandemic influenza patients.8 Certainly, adjusting for other comorbidities should be important in generating final conclusions for patients with multiple comorbidities.

This study had several limitations. First, our study population involved ILI cases (rather than laboratory-confirmed influenza cases) that have more variations in terms of virulence, vaccination and antibacterial/antifungal/antiviral treatment to influenza and other respiratory pathogens. The fatality risk assessment of ILI in our study was not equal to influenza. It is difficult to obtain laboratory confirmation for every case of influenza in clinical settings due to virological identifications for influenza are time-consuming, expensive and impractical to cover all suspected cases with mild outcomes. Therefore, studying patients with ILI provided more clinically relevant scenarios for physicians to identify vulnerable patients. Second, the severity of comorbidity, compliance of medical treatment and possible mitigation efforts of vaccination, early admission and antiviral treatment could not be evaluated in this study. The coding-associated diagnostic error55 may also have limited the accuracy of our disease/comorbidity diagnosis. Third, the comorbidities that we selected here according to the literature review and subsequent group discussion may still have had limitations, such as being unstudied or lower prevalent comorbidities. Obesity is associated with increased risk of mortality following influenza infections.56–58 However, obesity, immunisation and tobacco use were seldom coded in the National Health Insurance system. Fourth, the aetiology of ILI with/without co-infections of other microbial agents and environmental factors may also have influenced the ILI prognoses.59–61 Lastly, infants aged under 1 year have distinctive physiology and immunology responses to infections. The prognosis in this age group may be different from other young children, but subgroup analysis could not be done for limited cases.

The prevalence of comorbidities and environmental factors varied across the globe, with variations on national guidelines.22 34–36 62 The current vaccination guideline in Taiwan suggested patients with chronic underlying diseases, obesity and pregnant women to be evaluated for influenza vaccination, but the level of evidence for high-risk comorbidities was low.8 Our study analysed a national dataset in Taiwan with good representativeness and well-adjusted covariates. The results could support a reference of vulnerable groups to target, which may reduce the burden and prevent the ILI fatalities caused by influenza and other respiratory infectious disease agents.63 64 Since early antiviral treatment may be beneficial for patients with severe influenza,65–67 our findings may assist policy makers in amending guidelines to promptly prescribe antiviral medications for high-risk individuals during the influenza season or pandemic outbreak. Therefore, we offer the following recommendations: first, higher priority should be given to patients with malignancy, heart disease, CVA and AIDS. Second, adult and elderly patients with tuberculosis, diabetes and CKD should also receive more attention. Third, paediatric patients with epilepsy and elderly patients with dementia and hypothyroidism should also be treated with more attention.

In conclusion, our study confirmed that age-specific chronic medical conditions were associated with increased fatalities in hospitalised ILI patients. In the era of personalised medicine and an increasingly ageing population involving various comorbidities, we suggest future studies to calculate the combined risks of comorbidities in an individual patient for better prediction and case management.