Article Text
PDF
Abstract
Objective To explore differences in end-of-life healthcare utilisation and medication costs between patients with haematological malignancies and patients with solid tumours.
Methods Data on deceased patients with cancer were selected from the sample cohort data of health insurance claims from 2008 to 2015 in South Korea. They were categorised into two groups: patients with haematological malignancies and patients with solid tumours. Longitudinal data comprised the patient-month unit and aggregated healthcare utilisation and medication cost for 1 year before death. Healthcare utilisation included emergency room visits, hospitalisation and blood transfusions. Medication costs were subdivided into anticancer drugs, antibiotics, opioids, sedatives and blood preparation. Generalised linear mixed models were used to evaluate differences between the two groups and time trends.
Results Of the 8719 deceased patients with cancer, 349 died from haematological malignancies. Compared with solid tumours, patients with haematological malignancies were more likely to visit the emergency room (OR=1.36, 95% CI 1.10 to 1.69) and receive blood transfusions (OR=5.44, 95% CI 4.29 to 6.90). The length of hospitalisation of patients was significantly different (difference=2.49 days, 95% CI 1.75 to 3.22). Medication costs, except for anticancer treatment, increased as death approached. The costs of antibiotics and blood preparations were higher in patients with haematological malignancies than in those with solid tumours: 3.24 (95% CI 2.14 to 4.90) and 4.10 (95% CI 2.77 to 6.09) times higher, respectively.
Conclusions Patients with haematological malignancies are at a higher risk for aggressive care and economic burden at the end of life compared with those with solid tumours. Detailed attention is required when developing care plans for end-of-life care of haematological patients.
- cancer
- drug administration
- haematological disease
- supportive care
- terminal care
Statistics from Altmetric.com
Introduction
Quality end-of-life care is essential for patients for the following five reasons: receiving adequate pain and symptom management, avoiding inappropriate prolongation of dying, achieving a sense of control, strengthening the relationships with loved ones and relieving their burden.1 Hence, a comprehensive care plan has to be developed in advance to provide quality end-of-life care, made possible by a better understanding of real clinical care.
Unlike patients with solid tumours, patients with haematological malignancies mainly rely on medications rather than surgery towards the end of their life.2–5 The prognosis of end-of-life in haematological malignancy is unpredictable, as patients rapidly lose their physical functions just before death.6–8 Symptoms of haematological malignancy at the end-of-life are different from those of solid tumours in terms of suffering from delirium and drowsiness.9 10 Haemorrhagic episodes, febrile episodes and infections are common symptoms in patients with haematological malignancies.11 12 However, to the best of our knowledge, the impact of haematological malignancies on the long-term trend of healthcare utilisation and medication costs at the end of life is still unknown. Most studies focused only on the last month before death to demonstrate the aggressiveness of care.13 14 To enable patients with cancer to stay in good physical and mental condition until their end of life, we need to understand real clinical care at the end-of-life.
Therefore, we aimed to explore differences and time trends in end-of-life healthcare utilisation and medication costs between patients with haematological malignancies and patients with solid tumours through longitudinal data analysis.
Materials and methods
Data source
This was a retrospective longitudinal study. South Korea’s National Health Insurance Service-National Sample Cohort (NHIS-NSC) data from 2006 to 2015 were used. The NHIS reports that its database contains claims from almost 80 000 healthcare service providers covering approximately 46 million patients per year, accounting for 98% of South Korea’s total population. The NHIS-NSC is a representative sample cohort of one million randomly selected participants using a stratified sampling method. It contains patient records on diagnosis, hospitalisations, procedures, treatments, surgeries, prescription drugs and patient death details, including the cause, making it a valuable resource for healthcare service research.15
Patient selection and follow-up windows
The study population included patients who had died from cancer—C00-C97 in the International Classification of Diseases 10th Revision (ICD-10)—between 2008 and 2015. Patients whose cause of death was haematological malignancies (ICD-10 codes: C77, C81–C96) were classified as having haematological malignancies. Likewise, patients whose cause of death was solid tumours (ICD-10 codes: C00–C76, C78–C79) were defined as patients with solid tumours. We defined patients’ demise day as the index date (day 0) (figure 1). Patients were required to have at least two cancer-related outpatient visit claims or at least one cancer-related hospitalisation claim within a 24-month preindex period. Patients who had records of both haematological malignancies and solid tumours within the 24-month preindex period were excluded.
Study design to examine end-of-life care in patients who died of cancer.
The NHIS-NSC data only provide the year and month of demise. Hence, the date of demise had to be assumed. We assumed that patients died on the last day of the month and operationally defined the length of the last month as 45 days to avoid overestimation and underestimation (online supplementary material 1A). With this operational definition, the median number of actual claims included is 30 days, given that a patient dies randomly within a month. If the last month’s length is 30 days (online supplementary material 1B), the associated cost of the last month for a patient who died in the middle of the month would have claims for just a median of 15 days, ranging from 0 to 30 days. That is, the cost of the last month would be underestimated. Assuming that the day of demise is the beginning of the month (online supplementary material 1C), the cost would be overestimated. Consequently, the actual length of the last month of each patient’s end-of-life comprised a minimum of 15 days to a maximum of 45 days (online supplementary material 1A). Except for the last month, the other months comprised 30 days each.
Supplemental material
Patient-wise longitudinal records were assessed. The maximum length of the follow-up window was 12 months. Follow-up windows began from the date of patients’ first cancer-related prescription within the 12-month pre-index period and ended at the index date (figure 1). If the date of this first prescription was before the date of 1 year prior to death, a follow-up window started from the first day of 12 months prior to the index date. As this study estimated monthly health utilisation and costs, the length of the follow-up window had to be at least 1 month. Hence, patients whose follow-up windows were less than 1 month were excluded.
End-of-life healthcare utilisation and costs
After defining the follow-up window, all cancer-related administrative records within this window were collected and divided per month by counting backwards from the index date. All outcome variables related to healthcare utilisation and medication costs were collected monthly. Within this window, data claims related to healthcare utilisation and direct healthcare costs for medications were assessed. Outcomes demonstrating healthcare utilisation were as follows: (1) number of emergency room visits, (2) length of stay (LOS), (3) LOS in the intensive care unit (ICU) and (4) number of blood transfusions. Five therapeutic categories of medications used to alleviate symptoms during end-of-life were also assessed and included (1) anticancer drugs, (2) antibiotics, (3) opioids, (4) sedatives and (5) blood preparations.9 16–19 The codes of the drugs are presented in online supplementary material 2. All costs were converted to US dollars using the following exchange rate: US$1=1100 South Korean won.
Statistical analysis
Student’s t-tests were used to evaluate differences between continuous variables. χ2 tests were used to evaluate differences between categorical variables. Trends in health utilisation and medication costs at the end-of-life for haematological malignancies versus solid tumours were reported descriptively. The proportion of estimated medication costs for 1 year was estimated based on the mean of each monthly cost for medications.
Generalised linear mixed models (GLMM) were used to evaluate the differences between the two groups and the time trend in healthcare utilisation and medication costs while controlling for baseline covariates.20 Our study aimed to show the difference and time trends in healthcare utilisation and medication costs of haematological malignancies compared with solid tumours. This was demonstrated in two ways. First, the variable ‘haematological malignancy’ was considered for the GLMM to represent the gap in health utilisation at baseline (12 months before death) between patients with haematological malignancies and patients with solid tumours. The term is similar to the y-intercept in a linear function. The ‘haematological malignancy’ variable was a binary variable which indicated haematological malignancy as ‘1’ and solid tumour as ‘0’ (reference). If the estimate is significant, it can be said that the health utilisation of patients with haematological malignancies is significantly different from that of patients with solid tumours at baseline.
Next, the interaction term ‘time × haematological malignancies’ was considered for GLMM to represent the difference in the increasing/decreasing trend in health utilisation between patients with haematological malignancies and patients with solid tumours after baseline. The term is similar to the difference in slope in a linear function. The ‘time’ variable ranged from 1 (reference) to 12 (the last month before death). The estimates of GLMM were adjusted with the following covariates: age, sex, insurance type, Charlson Comorbidity Index (CCI) and length of follow-up period.21
Depending on the variable’s characteristics, various link functions and distributions were adapted to GLMM. For emergency room visits and blood transfusions, which are relatively rare events, a log link and binomial distribution were used. The exponential form of the coefficient is the odds ratio (OR). For the LOS, we used the identity link and Gaussian distribution. Coefficients for LOS indicate the difference given a one-unit shift in the independent variable while holding other variables. Finally, we used the log link functions and the Gaussian distribution for all medication costs. The exponential form of the coefficients indicates the multiplied value of y, given each one-unit increase in an independent variable. We defined this as the relative ratio.
All statistical analyses were deemed significant at a 5% confidence level using two-sided tests. Statistical analyses were performed using SAS Enterprise V.7.1 software.
All personal identifying information of patients was anonymous; therefore, informed consent was waived by the institutional review board for this study.
Results
Patient characteristics
Among the 43 037 patients with cancer who died between 2008 and 2015, a total of 8719 were included in this study (figure 2). Of the total eligible patients, 349 (4%) were patients with haematological malignancies, while the remaining 8370 (96%) had solid tumours. Table 1 shows the characteristics of the patients. The proportion of younger patients with haematological malignancies was higher than in those with solid tumours (p<0.01). Patients with solid tumours had more comorbidities than patients with haematological malignancies (p<0.01).
Sociodemographic characteristics of patients
Flow chart of patient selection.
Healthcare utilisation and medication costs
The monthly trends of healthcare utilisation and subdivided medication costs are shown in figure 3. During the end-of-life period, overall healthcare utilisation was higher in patients with haematological malignancies than in those with solid tumours. table 2 shows the estimates from the GLMM. The estimated difference between the LOS of patients with haematological malignancies and those with solid tumours was 2.49 (95% CI 1.75 to 3.22) days at baseline. The LOS in the ICU between the two groups was not significantly different (difference=−0.21, 95% CI −0.53 to 0.11). However, considering the interaction term, it increased faster than in patients with solid tumours (difference=0.37, 95% CI 0.32 to 0.42). Patients with haematological malignancies were more likely to receive blood transfusions (OR=5.44, 95% CI 4.29 to 6.90) and were more likely to visit the emergency room (OR=1.36, 95% CI 1.10 to 1.69) than patients with solid tumours. The time trends between the two groups were not significant (OR=1.0, 95% CI 0.97 to 1.04; OR=0.96, 95% CI 0.96 to 1.03).
Risk of healthcare utilisation and cost at the end of life
Monthly trend in (A) healthcare utilisation and (B) medication cost per patient (x-axis as time, month). ICU, intensive care unit; USD, US dollars.
Figure 4 shows the proportion of estimated medication costs for 1 year before death. Notably, the proportions of blood preparation and antibiotic costs were higher in patients with haematological malignancies than in those with solid tumours. The cost of anticancer drugs and opioids was higher for those with solid tumours. Based on the monthly trend, all medication costs except anticancer drugs showed an increasing trend as the end-of-life progressed (figure 3). The proportion of costs spent on anticancer drugs decreased in patients with solid tumours (from 60.8% to 9.4%; data not shown) and in patients with haematological malignancies (from 34.3% to 14.9%; data not shown).
Proportion of estimated total medication costs for 1 year before death.
The blood preparation costs in patients with haematological malignancies were 4.10 (95% CI 2.77 to 6.09) times the costs in patients with solid tumours at baseline (table 2). The costs of antibiotics in patients with haematological malignancies were 3.24 (95% CI 2.14 to 4.90) times the costs in patients with solid tumours at baseline. For patients with solid tumours, the costs of antibiotics were 1.33 (95% CI 1.31 to 1.34) times the costs of the previous month. The costs of antibiotics in patients with haematological malignancies were 1.43 (1.33×1.08) times the costs of the previous month by multiplying the estimate of the interaction term and the estimate of the time term. The costs of sedatives in patients with solid tumours were 1.06 (95% CI 1.05 to 1.07) times the costs in the previous month, the sedative costs in patients with haematological malignancies were 1.14 (1.06×1.08) times the costs in the previous month, and the time trend was significantly different.
Discussion
This study explored differences and time trends in end-of-life healthcare utilisation and medication costs between patients with haematological malignancies and those with solid tumours through longitudinal data analysis. The results confirmed that haematological malignancy was associated with higher levels of healthcare utilisation and higher medication costs. The time trends in healthcare utilisation were notable for ICU hospitalisation. Interestingly, the amount spent on anticancer drugs by patients with solid tumours decreased as the endof-life progressed, while the time trend in patients with haematological malignancies was significantly different. Unlike the decreasing trend in the cost of anticancer drugs, the cost of other medications increased in both groups, but this increasing trend was sharper in patients with haematological malignancies, especially in the case of antibiotics and sedatives.
We established that patients with haematological malignancies received a higher volume of care, including hospitalisation, emergency room visits and transfusions, throughout the year before death. This was consistent with the results of a previous study that examined healthcare utilisation in the year before death. This indicated that LOS, number of outpatient visits, number of emergency room visits and healthcare costs were higher in patients with haematological malignancies than in those with solid tumours.22 In addition to these previous studies, we observed the time trend of haematological malignancies for 1 year, which was significantly different from that of solid tumours. We found that hospitalisation in the ICU notably increased in the last month. According to previous studies, patients with haematological malignancies were more likely to receive intensive care for 1 month before death compared with patients with solid tumours.23–25 This was also observed in our study.
An increasing trend in the cost of antibiotics was observed in both cancer types, and the trend was sharper in haematological malignancy. This implies that infection is a critical issue at the end-of-life in patients with haematological malignancy. Infection was also the main cause of death in patients with multiple myeloma and acute myeloid leukaemia.26 27 To manage the infection, 90% of patients with haematological malignancies were treated with antibiotics as part of intensive palliative care in the week before death.16 Our results show that antibiotics can lead to an economic burden for patients. Nevertheless, several studies28–30 have underscored the benefits of using antibiotics at the end-of-life, as they help to relieve patients’ symptoms of infection28 and improve their quality of life.28–30
Similarly, haematological malignancy was associated with higher costs for blood preparation throughout the year before death, compared with that for solid tumours. This is consistent with a study by Dasch et al,23 who reported that the proportion of patients who received blood transfusions in the last week before death was higher in haematological malignancies than in solid tumours. We confirmed that the higher use of blood preparations was not only evident 1 month before death but also 1 year before death. However, we need to consider the benefits of blood transfusions at the patient’s end of life as have several previous studies, although some studies have challenged these benefits. Although blood transfusion is associated with poor survival in haematological malignancies, it can help mitigate symptoms such as fatigue, breathlessness, generalised weakness, dizziness and so on.19 29 However, for patients with anaemia, blood transfusion is associated with longer survival.31 Smith et al32 argued that blood transfusions should be avoided in medically futile situations, and therefore only a minimum amount of red blood cells should be transfused. We could not investigate the benefits of transfusion because our outcomes were healthcare utilisation and costs. Once the high rate of transfusions is captured, physicians need to consider the benefits of transfusions. If the benefits of transfusion at the patient’s end-of-life through further studies can be established, they may be recommended to improve the quality of patients’ end-of-life care.
This study has several limitations. First, rather than tracking patients from the day of diagnosis, the analysis was based on patients whose data claims were available within 2 years before death. In other words, the analysis did not include patients who had not visited the hospital for 2 years before their death, even though they had a prior diagnosis of cancer. According to Langton et al,33 90% of patients who died of cancer visited hospitals within 6 months prior to their demise. The results suggest that the 2-year window in our study was appropriate for capturing the overall study population, and the analysis of data claims of that population can be representative of the healthcare utilisation of all patients with cancer. Second, the outcomes of our study may not reflect up-to-date oncology care, as the NHIS only provides sample data until 2015. There have been great advances in oncology care since 2015, such as blinatumomab for patients with acute lymphoblastic leukaemia and pembrolizumab for patients with non-small cell lung cancer. Since these medications could impact the total healthcare budget and treatment patterns, our study may not reflect the real costs of cancer care. Hence, based on our findings, further studies using an up-to-date database should be pursued. Third, the study analysed only reimbursed medications, as these analyses were based on data claims. Patients had paid for some expensive anticancer drugs or antibiotics that were not covered by health insurance. In Korea, the rates of non-reimbursement costs among total costs are reported by type of cancer. The rates of non-reimbursement for solid tumours ranged from 10.1% to 26.9%, while the rates for haematological malignancies ranged from 5.8% to 15.1% in 2015.34 As the total cost for haematological malignancies was over threefold of the total cost for solid tumours, the non-reimbursed costs would not significantly affect our results. Lastly, the healthcare systems in Korea and other countries are different, leading to varying medical expenses. Therefore, generalisation of the study’s findings to other countries should be done with caution.
Conclusions
This study estimates the differences in end-of-life care between haematological malignancy and solid tumours in the real world. To the best of our knowledge, this study is the first to identify the differences and time trends in healthcare utilisation and medication costs, specifically controlling for covariates. Since we controlled for potential covariates such as sex, age, insurance type, CCI and length of follow-up period, we can conclude that haematological malignancy is associated with high healthcare utilisation and high medication cost, especially in antibiotics and blood preparations. The economic burden of haematological malignancy increases more rapidly than that of solid tumours.
As patients with haematological malignancies are at risk of aggressive care, planning quality end-of-life care is required and care plans should pay more attention to this need. By recognising the economic burden and features of healthcare utilisation of patients with haematological malignancies, we hope that physicians will consider improving patient care plans and ultimately achieve quality end-of-life care for these patients. It would help patients receive adequate pain and symptom management, avoid inappropriate prolongation of dying, achieve a sense of control and relieve caregivers’ economic burden. The economic burden of haematological malignancy can be considered in the cost-effectiveness analysis for reimbursement, which would enhance accessibility to a new treatment, and therefore the quality of care can also be elevated. In addition, policymakers can efficiently plan their budget based on the volume of healthcare utilisation. Therefore, detailed attention is required when decision-making about end-of-life care plans in haematological patients.
References
Footnotes
Contributors Conception and design: S-YY, E-KL, S-KP, S-HK. Acquisition of data and statistical analysis: S-YY, E-KL. Analysis and interpretation of data: S-YY, S-HK. Drafting of the manuscript: S-YY, S-HK. Critical revision of the manuscript for important intellectual content: H-LK, S-HK. Administrative, technical or material support: S-YY, S-KP, E-KL, H-RK, H-LK. Supervision: S-HK, E-KL.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient consent for publication Not required.
Ethics approval This study was approved by the institutional review board of Sungkyunkwan University (approval no. 2017-06-005).
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data may be obtained from a third party and are not publicly available. The National Health Insurance Service-National Sample Cohort’s (NHIS-NSC) database in South Korea was used.