Impact of patient characteristics and treatment procedures on hospitalization cost and length of stay in Japanese patients with influenza: A structural equation modelling approach

Objectives Little is known about the economic burden of influenza‐related hospitalizations in Japan. This study sought to identify the factors that contribute to the total healthcare costs (THCs) associated with hospitalizations due to influenza in the Japanese population. Study design A retrospective cross‐sectional database analysis study. Methods A structural equation modelling approach was used to analyse a nationwide Japanese hospital claims data. This study included inpatients with at least 1 confirmed diagnosis of influenza and with a hospital stay of at least 2 days, who were admitted between April 2014 and March 2015. Results A total of 5261 Japanese inpatients with a diagnosis of influenza were included in the final analysis. The elderly (≥65 years) and the young (≤15 years) comprised more than 85% of patients. The average length of stay (LOS) was 12.5 days, and the mean THC was 5402 US dollars (US$) per hospitalization. One additional hospital day increased the THC by 314 US$. Intensive care unit hospitalizations were linked to higher costs (+4957 US$) compared to regular hospitalizations. The biggest procedure‐related cost drivers, which were also impacted by LOS, were blood transfusions (+6477 US$), tube feedings (+3501 US$) and dialysis (+2992 US$). Conclusions In Japan, the economic burden due to influenza‐related hospitalizations for both children and the elderly is considerable and is further impacted by associated comorbidities, diagnostic tests and procedures that prolong the LOS.


| INTRODUCTION
Influenza is a contagious respiratory illness caused by a highly infectious viral pathogen. The illness ranges from mild to severe and can lead to numerous complications such as superimposed infections, exacerbation of cardiovascular conditions and asthma. Most of the fatal cases occur in the elderly over 65 years old 1 and in high-risk populations including children younger than 2 years old, 2 pregnant women, 3 healthcare workers 4 and patients with associated comorbidities such as asthma, chronic lung disease, kidney disorders and blood disorders. 5,6 Annual epidemics of influenza result in approximately 250 000 to 500 000 deaths worldwide. 7 Cases of influenza also have a substantial socioeconomic impact in terms of medical care, healthcare utilization (eg increase in consultations, hospitalizations and length of stay [LOS]) and work absenteeism. 8 In Europe, influenza is responsible for approximately 10% of sickness-related workplace absence. 9 As influenza is an epidemic disease, it may disturb the healthcare services by acute overloading during the epidemic.
Elderly patients comprise the group with the highest burden of influenza-related complications. Patients aged ≥85 years are 6 times more likely to be hospitalized and 16 times more likely to die compared with patients aged 65-69 years. 10,11 The costs associated with influenza and its complications can be substantial. In the United States, a study based on the 2003 population estimated that the annual burden of influenza was 3.1 million hospital days and 31.4 million outpatient visits. From a societal perspective, the total economic burden of influenza (direct costs and indirect costs, including loss of earnings and loss of life) has been estimated to be 87.1 billion US$ annually, with direct costs accounting for more than 10 billion US$, of which 40% is spent on the treatment of patients older than 65 years of age. 12 In the United States, the mean total cost of hospitalization for influenza-related illness for children was 13 159 US$ (39 792 US$ for patients admitted to an intensive care unit (ICU) and 7030 US$ for patients cared for exclusively on the wards). High-risk patients had a higher mean total cost than low-risk patients (15 269 vs 9107 US$, respectively). 13 In Japan, it is estimated that 5%-10% of the population develops influenza annually, resulting in approximately 1000 to 2000 deaths from influenza alone and an additional 5000 deaths due to complications such as pneumonia. 14 Approximately 20% to 25% of elderly Japanese patients with influenza develop pneumonia, 5% of whom die. 14 From 1988From to 1991.0% of all admissions to paediatric hospitals during the winter season in Japan were due to influenza viral infections, while respiratory syncytial virus accounted for 17.5% of admissions. 15 Despite these statistics, there is limited information available about the extent of the disease burden due to influenza-related illness in Japan. Therefore, the aim of this study was to identify factors that impact hospitalization costs for patients with influenza in Japan utilizing a Japanese administrative database.

| Patient selection
We utilized a commercially available hospital claims databank from Medical Data Vision Co., Ltd (MDV, Tokyo, Japan). This is an administrative database including approximately 4 400 000 patients, which represents approximately 3% of the total Japanese population. 16 The MDV database has been used to investigate a wide range of conditions in Japan such as rheumatoid arthritis, 17,18 schizophrenia, 19 infectious diseases, 20 multiple sclerosis 21 and hypertension. 22 We considered the inpatient claims from patients who were admitted between 1 April (defined by at least 1 night was spent in the hospital).

| Hospitalization cost calculation
Total healthcare costs (THCs) comprised all costs of healthcare services incurred during each hospitalization. These included basic management fees, examination, procedures and medication. Both Diagnosis Procedure Combination cost (DPC cost, which is a case-mix reimbursement cost) and total actual heath care cost were calculated.
All costs were converted from Japanese yen to US$ based on the average exchange rate during April 2014-March 2015 (Financial Market Department, Bank of Japan; 1 US$ = 109.33 yen). 23

| Statistical analysis
Descriptive analyses were performed on baseline characteristics as well as resource use, LOS and THC. As LOS is usually an important driver of the total hospitalization costs, 24 26 The SEM framework allows evaluating relationships among variables by combining the strengths of factor analysis and multiple regression in a single model that can be tested statistically. 27 More specifically, in this study, a path analysis was conducted, which is a special case of the SEM framework that allows an exploration of the causal links (direct and indirect effects) between exogenous variables and 1 or more endogenous variables. In this framework, the total effects of a covariate on the main dependent variable can be decomposed into 2 categories of effects: (i) the indirect effects, consisting of the effect of the covariate on 1 or more intermediary endogenous variables, which in turn translates into an effect on the main variable; and (ii) the direct effect, which is the remaining effect of the covariate on the main variable while controlling for their indirect effects. 28 In our case, the main endogenous variable of interest in the analysis was the total hospitalization cost expressed in Japanese yen, while we assumed that independent variables would have both a direct effect on total hospitalization costs and indirect effects through the LOS. Figure 1 depicts the underlying path diagram showing the relationship between each variable. We also conducted subgroup analyses of the children (≤15 years), the elderly (≥65 years) populations and the infants and toddlers (children ≤2 years old), 3 groups that are particularly susceptible to influenza complications and hospitalization.
Statistical analyses were performed using Stata 15.0. 29

| RESULTS
A total of 5261 Japanese inpatients with influenza were included in the final analysis. We excluded 15 rehospitalized admissions due to the limited number of patients ( Figure 2). home or welfare facilities are less costly than those who were hospitalized from home.
The majority of additional procedures were significantly associated with higher THC both directly and due to an increase in the LOS. Among surgeries and interventions, the largest cost drivers were blood transfusions (+6477 US$), tube feedings (+3501 US$) and dialysis (+2992 US$). Bronchoscopy and echocardiography were the imaging procedures that increased the THC most significantly (+3482 and +1511 US$, respectively). Overall, the effects on DPC costs compared with total costs were similar (Data S1).
Subgroup analyses of children (≤15 years) (Table 3), the elderly (≥65 years) ( Table 4) and the infants and toddlers (≤2 years old) ( Table 5) showed similar results, although the magnitude of the effect was higher in children for most of the surgeries and interventions.

| DISCUSSION
Using an administrative database of hospitalized Japanese patients with influenza, we found that influenza-related hospitalizations mostly consisted of elderly and young patients, confirming that these 2 age groups are at high risk of influenza complications.

| Impact of comorbidities on THC
It is not surprising that healthcare costs significantly increase when influenza strikes in association with other medical disorders. Our data revealed that Parkinson's disease had the highest impact on cost although it represented only 1.6% of the population, followed by cancer and ischaemic heart disease, which were 9.6% and 7.7% of the cases, respectively. The increased healthcare cost is most likely a reflection of the high incidence of influenza-related complications that occur with these comorbidities.
Despite the low incidence of neurologic disorders associated with influenza viral infection, patients have a high risk of developing complications. 30,31 In addition, Parkinson's disease has been reported to be a clinical manifestation of influenza, 32 and parkinsonian-like symptoms such as tremors have also been described in severe influenza cases. 33 Of note, influenza A is one of several viruses that have been implicated in the pathogenesis of Parkinson's disease. 34   influenza is also linked to both increases in AMI 43 and AMI-related deaths. 44,45 Influenza infection has also been associated with damage to the heart muscle leading to cardiomyopathy and myocarditis. 46 Taken together, these observations are consistent with our findings that patients with heart disease comprised a significant share of influenza-related hospitalizations, and heart disease was an important driver of the increase in THC.

| Role of patient origin
It was found that patients who were referred from nursing home or welfare facilities incurred less cost than those who were hospitalized from home. One possible interpretation of this interesting finding is that institutions such as nursing homes or welfare facilities do monitor their clients well and send them to the hospital even in case of a

| Impact of procedures and ER and ICU admissions on THC
The most significant cost drivers among procedures were blood transfusions and tube feedings, which increased the THC by 380 000 Japanese yen (approximately 3450 US$).
Our findings of a high cost burden associated with ICU or ER admissions when compared with routine hospitalizations are consistent with other reports. In European countries, for instance, the daily cost of ICU admissions ranged from €1168 to €2025 (1240 to 2150 US$), 47 while in the United States the estimated additional cost was 2190 US$ per day. 48 These statistics underscore the importance of avoiding ICU or ER admissions whenever possible.

| A role of vaccinations and antiviral treatment
The potential policy implication of our findings is that vaccination programmes should be promoted to avoid influenza-related hospitalizations. From 1977 to 1987, there was already a vaccination programme for Japanese schoolchildren that achieved between 50% and 85% annual coverage in children aged 3-15 years. It was shown that this vaccination programme was associated with a decrease in the overall number of influenza-related excess deaths and that excess deaths increased once the programme was discontinued. 49 Furthermore, because the vaccination of schoolchildren can reduce influenza-related morbidity and mortality among non-immunized contacts as well as the elderly, it was estimated that the vaccination programme could also save 1000 elderly lives per year. 50 For those patients still requiring hospitalization, medical treatment may be an option to reduce hospital LOS and healthcare costs. A recent study in the United States that included 1 557 437 cases of influenza from 4 influenza seasons found an overall 11% reduction in the risk of complications in oseltamivir-treated patients (an 81% reduction in those treated <2 days after the diagnosis). 31 Antiviral treatment also decreased the risk of hospitalizations and emergency room visits by 29% and 24%, respectively. A recent cost-effectiveness analysis in the Japanese healthcare context, for instance, demonstrated that treatment with oseltamivir was highly cost-effective with an incremental cost-effectiveness ratio (ICER) of 398 571 Japanese yen (3645 US$) per quality-adjusted life year from a health insurance perspective.
With the inclusion of productivity costs, the ICER for oseltamivir turned negative, meaning that medical treatment with oseltamivir was both cost-saving and more effective. 51

| Limitations
There are several limitations to our study. First, this analysis is based on a 1-year database. Thus, we could not capture the potential changes due to prescribing behaviour changes and the change of treatment guideline over time. Second, due to the limitations of the database, potentially useful information that might explain costs was lacking. For instance, we could not retrieve hospital ID numbers, which could have been used to identify heterogeneity between hospitals as well as patient characteristics such as region, social and professional status and clinical severity of their disease. Nevertheless, our analysis examined all available patient characteristics (such as age, gender and relevant comorbidities) that could be retrieved from the database. Third, bias may have resulted from the current DPC system that allows hospitals to choose the diagnosis that is incurring the most medical resource utilization as the main diagnosis. In general, patients with comorbidities will receive a higher reimbursement if hospitals choose comorbidities as the primary diagnosis. Finally, a major limitation of this study is that influenza-related hospitalizations can be difficult to identify because influenza is not always detected as the primary cause of the hospitalization, especially in severe cases. 10 As a result, this study may underestimate the true burden of influenza as well as the cost of influenza-related hospitalizations because of the coding incentive.
Only hospitalizations with influenza diagnosis, which are less costly, were included. 10