How the resource allocation and inpatient behavior affect the expenditures of terminal malignant tumor patients?

Background: The inappropriate use of medical resources and escalating health expenditures of severe diseases patients has been a great concern in China’s health system especially among terminal stage. This study aims to analyze how the resource allocation and hospitalization behavior affect the expenditures of terminal malignant tumor patients, and provide evidences on resource allocation and utilization. Methods: An analysis framework of inuencing factors of medical expenses was built according to Andersen’s Behavioral Model. Hospitalization expenditures of malignant tumor patients who died in medical institutions were tracked in Shanghai in 2016. We use path analysis to analyze the inuencing factors of hospitalization expenditures at terminal stage. Results: Results demonstrated that hospital services and expenditures during the terminal stage were primarily in tertiary hospitals. The top three inuencing factors were length of stay, number of admissions and level of medical institutions. While the inuence of demographic characteristics (age, gender, type of medical insurance, etc.) was relatively low. Conclusions: Data suggest that hospitalization expenditures and patients’ economic burden can be reduced through adjusting allocation of medical resources and service utilization, as well as reducing unnecessary hospitalization days.

tumor patients and their hospital expenditures, thus to provide evidence for a sound allocation of health funding and to improve its utilization e ciency.
Factors including gender, age, type of malignant tumor, time to death, medical institution, number of admissions, length of stay(LOS) and with/without surgery could affect the expenditures of terminal malignant tumor patients. According to Shugarman LR's research, among patients with lung cancer, colon and rectal cancer, women generated higher medical expenditures than men. The gender difference of medical expenditures was more prominent in patients of 68-74 years old due to the interaction of gender and age [iii]. Among patients dying from oral cancer, those who were under 65 years old consumed 819 dollars more than those 65 years old or above in Taiwan Region of China [iv]. Expenditures varied largely among patients of different malignant tumors, from 14,216 Euros for liver cancer to 26,712 Euros for tumors of hematopoietic and lymphoid tissue. Expenditures of brain malignant tumor were about 20 times of that of prostatic cancer during the rst month after diagnosis. Observed by Blakely T, medical expenditures related closely with the disease stage, demonstrating a shape of 'U' from diagnosis to death.
Medical expenditures were high during the rst month after diagnosis, and decreased to a relatively low level in the following stages, and increased again during the last month before death [v]. Reev R found that those who died at home consumed lower medical expenditures than those who died in hospitals [vi].
Researchers also revealed that LOS, hospital level and with/without surgery were main factors that directly in uence hospital expenditures [vii] , [viii] . With/without surgery 6 , type of disease and age[ix]could in uence hospitalization expenditures through utilization of hospital service indirectly. . Among those factors associated with medical expenditures, gender and age belong to the predisposing factors, type of medical insurance is an enabling factor, while type of malignant tumor is classi ed as a need factor. Patients' utilization of health service, level of medical institutions, LOS, number of admissions and with/without surgery directly affect hospitalization expenditures. Given that predisposing factors and need factors can be slightly changed, we focused on analyzing enabling factors such as medical insurance and health service utilization behavior. The analytic framework is shown in

Analysis method
We used SPSS 18.0 to conduct statistical analysis. Chi-squared test was applied for categorical data.
Pearson correlation coe cient was used to demonstrate correlation between variables. Path analysis model was tted by multiple linear regression models. The signi cance level α was set as 0.05.
Path analysis is a derivation from multiple linear regression, explaining relationship quantitatively through covariance structure among variables. In a path analysis model, exogenous variable is affected by external variables, and endogenous variable is affected by internal variables. In this study, hospitalization expenditures and LOS were de ned as endogenous variables. Logarithmic transformation was used, where Y 1 = lg hospitalization expenditures, Y 2 = lg LOS.
The in uence of exogenous variables on Y 1 : Direct path coe cient = the standardized regression coe cient; The indirect in uence of exogenous variables on Y 1 : Indirect path coe cient = direct path coe cient of independent variable Y 2 × (correlation between X and Y 2 ); Total path coe cient= direct path coe cient + indirect path coe cient.
Numeric data were included in the model as continuous variable. Unordered categorical variables were included as dummy variables. Level of medical institutions was classi ed into 7 combinations based on patients' choices. First-level and unrated hospitals and community health service centers were merged into ' rst-level' in the model. Exogenous variables and value assignments can be found in Table 1.

General information
In 2016, there were 13, 835 patients died from malignant tumor in medical institutions of Shanghai, among which male accounted for 62.49%. The average age was 70.42±12.70 years old. The leading types of malignant tumor were lung cancer (24.17%), gastric cancer (9.37%), colon cancer (7.05%), pancreatic cancer (5.98%) and liver cancer (5.83%). The number of admissions in the last 2 years of life amounted to 74, 500. The proportion of admissions to tertiary hospitals, secondary hospitals, rst-level and unrated hospitals, community health service centers were 56.58%, 35.76%, 4.64% and 3.02% respectively. The median LOS were 7 days, 13 days and 15 days in tertiary, secondary, rst-level and unrated hospitals respectively. Chi-squared test demonstrated that there was signi cant difference in constitution of gender and age among different levels of medical institutions (P<0.05). Women and those aged 65 and above were more likely to be hospitalized in lower level institutions.
Hospital services of terminal malignant tumor patients were concentrated in institutions located in central urban areas (Figure 2). Among the top 5 hospitals with the largest number of admissions, the rst one was Pulmonary Hospital (accounted for 5.27% of the total number of admissions), followed by Zhongshan Hospital (4.42%), Chest Hospital (3.63%), the Tenth People's Hospital (3.51%) and Shanghai East Hospital (2.85%). 15 of the top 20 medical institutions were tertiary hospitals, and the other 5 ones were secondary hospitals. Admissions to these 20 hospitals accounted for 53.52% of the total.
Among 13, 835 terminal malignant tumor patients in 2016, 44.87% patients were hospitalized in one single hospital and were not transferred to another hospital in the last 2 years of life, the majority of which chose tertiary hospitals. Around 90% of the patients were hospitalized in no more than three different medical institutions. Less than 1% of patients transferred to more than seven medical institutions.
At the municipal level, a large number of patients were transferred to tertiary hospitals. Hospitals in central urban areas were closely connected with each other, receiving a large number of patients transferred from other hospitals. That was probably due to the central geographic location therefore it was convenient for referral. While at the district level, the gure demonstrated different patterns. In suburbs like Jinshan, Fengxian and Qingpu District, patients were mainly hospitalized within the district, and the majority of patients were transferred into regional tertiary hospital in these suburbs.

Hospitalization expenditures of different levels of institutions
Medical services for malignant tumor patients were mainly provided by tumor hospitals and oncology departments of comprehensive hospitals. Among them, the number of beds in tertiary hospitals accounted for 55.55% of the total number of beds in Shanghai, while the proportion of admissions and hospitalization expenditures accounted for 56.58% and 56.94%. Per-admission hospitalization expenditures ranged from RMB 21, 000 to RMB 235, 000 Yuan among different levels of hospitals, while the number was RMB 7, 900 yuan in community health service centers. Per-day hospitalization expenditures of tertiary hospitals were 2.10 times of that of rst-level and unrated hospitals, 9.96 times of that of community health service institutions (Figure 3).
The out-of-pocket payment in Shanghai includes the self-paid expenditures beyond the scope of medical insurance and copayment within the insurance coverage. The out-of-pocket payment rate of tertiary hospitals was the highest (30.58%), and the rate increased by level of institutions. The patients' copayment rate ranged from 15.98% to 17.22% in the secondary and tertiary hospitals, and that of rstlevel hospitals and community health service centers ranged from 7.94% to 8.01%. However, self-paid rate beyond the scope of medical insurance of tertiary hospitals was 13.36%, which was much higher than that of rst-level (5.52%) and secondary hospitals (8.23%). The rate of community health service centers was only 0.8% (Figure 4).

Differences of hospitalization utilization and expenditures among institutions
Hospitalization utilization and expenditures of malignant tumor decedents mainly occurred in tertiary hospitals in Shanghai. The institutional distribution of hospital service and expenditures basically corresponded with bed allocation. Tertiary hospitals owned the most beds, and terminal patients tended to choose these hospitals. Along with the requirement of curtailing LOS by the government, patients in tertiary hospitals were usually transferred to secondary hospitals in central urban areas after acute stage.
Although patients had a lower economic burden in rst level and below hospitals, only 7.65% of hospitalizations happened in these institutions. On the one hand, it may due to the limited resources of lower level institutions. Tumor beds of secondary and tertiary hospitals accounted for 86.14% of total tumor beds in Shanghai, while beds of rst level and below hospitals only accounted for 13.86%. There were no tumor beds in community health service centers, but hospice care beds were set which mainly provide services for terminal malignant tumor patients. On the other hand, the relatively weak ability and policy limitation on medical equipment has restricted the service delivery of community health service centers. Community hospice care mainly covered supportive treatment such as pain relieving. Medicines for regular nutritional support (e.g. lipid emulsion) were not available, which can hardly meet the needs of terminal patients.
The per-admission expenditures among different levels of medical institutions varied little. However, the per-day hospitalization expenditures were highest in tertiary hospitals, which was double that of rst-level hospitals, 10 times that of community health service centers. The proportion of out-of-pocket was also the highest in tertiary hospitals. The differences can be explained by the severity of the disease, treatment method as well as charging standards of different levels of medical institutions.

In uencing factors of hospitalization expenditures and their interactions
The results of path analysis showed that predisposing factors, namely age, gender, type of medical insurance, and type of malignant tumor have less impact on hospitalization expenditures compared with utilization factors. Correlation analysis also demonstrated that the number of admissions and LOS were positively related with hospitalization expenditures. This study emphasized on the optional way to reduce terminal malignant tumor patients' hospitalization expenditures through adjusting and controlling service utilization.
(1) LOS and number of admissions patients stayed more than 426 days, and 5% of patients stayed more than 208 days. According to the interview of key informers, some patients just need supportive care and considered medical institutions as elderly homes. As to the number of admissions, it not only impacted hospitalization expenditures directly, but also in uenced hospitalization expenditures through LOS.
(2) Level of medical institutions Taking patients hospitalized only in rst-level hospitals as the reference, patients choosing other institutions brought higher expenditures. Patients hospitalizing only in tertiary hospitals had the largest direct path coe cient, that is, when other factors were controlled, their hospitalization expenditures at terminal stage were the highest. However, level of medical institutions was negatively associated with LOS, partly due to the high proportion of severe diseases and short LOS of tertiary hospitals. The median LOS was 7 days in tertiary hospitals and 13 days in secondary hospitals. Expenditures of patients hospitalized in both secondary and tertiary hospitals were the second highest.
(3) With/without surgery Among utilization factors, the total path coe cient of surgery was 0.301, both direct and indirect path coe cients were positive. Hospital expenditures of patients who had surgery were higher than those who didn't. Its in uence on hospital expenditures ranked the forth, mainly caused by high cost of operation, post-operative nursing, drugs and high-value consumables which was consistent with the results of NL Keating, et al 7,15 .
(4) Demographic characteristics Total path coe cient of the in uence of age on hospitalization expenditures was -0.101, meaning that elderly patients generated lower hospitalization expenditures. A plausible explanation was that elderly patients tended to receive less aggressive treatments 5, [vi], [vii] . Total path coe cient of gender is -0.008, that was, men generated higher hospitalization expenditures than women, which was probably due to higher prevalence of unhealthy habits among men. Women's relatively poor economic status and Chinese traditional preference for men also had an impact 8, [viii] . Expenditures of patients hospitalized in tertiary hospitals, and in both secondary and tertiary hospitals were higher; while those hospitalized in rst-level hospitals and community health service centers generated lower expenditures. With/without surgery ranked the 4 th among in uencing factors. Demographic characteristics (age, gender, type of medical insurance, etc.) had less in uence compared with utilization factors.