Diagnostic accuracy of a predictive scoring tool for patients who are eligible for home discharge from a palliative care unit

Introduction

Staying at a preferred place, principally at home, is of great importance for dying patients; moreover, patients who die at home experience higher ‘quality of death’ than those who die at acute hospitals (1-4). Empirical data, however, indicates that a considerable number of patients do not die in places of their own choosing (5-10). In recent years, an increasing number of medical institutions have been proactively providing discharge support. Some studies have suggested that patients and families have a better experience and more survival benefits when patients can remain at home than when they are admitted to palliative care units (PCUs) (11-13).

It is valuable to be able to identify patients who are eligible for home discharge from PCUs so timely care coordination can be provided. To date, several empirical studies have generated data for the purpose of developing predictive tools to identify such patients (14-16). Discharge from PCUs to home was associated with younger age, good performance status, and admission that was not due to intensive symptom control (i.e., absence of dyspnea, unnecessary oxygen therapy, and ascites drainage). One nationwide study developed a clinical tool to predict whether a cancer patient receiving home care dies at home, and this measure can predict the outcome with a sensitivity of 0.72 and a specificity of 0.81 (c-statistic, 0.84), using five parameters: caregivers’ preferences for home death, availability of visiting physicians, 24 h contact between physicians and nurses, caregivers’ experiences of deathwatch at home, and patients’ insights as to their own prognosis (16). These models may be useful, but they still have several limitations, such as coming from retrospective studies with no formal validation process or being developed with populations that were not limited to patients admitted to PCUs.

Therefore, we attempted to create and validate a scoring tool to predict whether a patient admitted to a PCU is likely to die in the hospital or be discharged home using prospectively recorded data from inpatients in a PCU to assist medical personnel in making decisions about the transition to home. We present the following article in accordance with the STARD reporting checklist (available at https://apm.amegroups.com/article/view/10.21037/apm-22-902/rc).

Methods

We conducted a single-center retrospective cohort study analyzing prospectively collected data on patients with cancer who are terminally ill. For each patient, the attending physicians recorded the structured data collecting sheet on the data related to admission and discharge. The data were collected and reported in patients’ electronic medical records as a part of routine practice. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Ethical Review Committee for Clinical Research of Toyama Kouseiren Takaoka Hospital (Approval No. 20190829003; the full protocol is available from the corresponding author). Informed consent was waived due to the national guideline by the Ministry of Health, Labour and Welfare, i.e., Ethical Guidelines for Medical and Biological Research Involving Human Subjects.

In the Japanese healthcare system, 459 certified PCUs provide specialist palliative care for patients with advanced malignancies; this specialist care is fully covered by the national insurance. Each PCU has an average of 20 beds and at least one attending physician, and the PCU usually belongs to a general hospital. The PCU plays a major role in end-of-life care but must also facilitate smooth transitions to home care services if appropriate. Per unit, the average number of patients admitted and the number of patients who died at the PCU was 187 and 153 per year, respectively. In 2016, the average length of stay of 32.2 days.

Subjects

All consecutive cancer patients admitted to the PCU of the Kouseiren Takaoka Hospital, a 533-bed general hospital in Toyama Prefecture in Japan from October 2016 to October 2019 were enrolled from the registry system. The PCU had 16 beds and provided active home discharge services (17,18). The data from the first admission for each patient were used for this study; rehospitalizations (e.g., second or subsequent hospitalizations) were excluded.

Measurement outcomes

The recorded outcomes included whether patients were discharged home. The follow-up period was from the patient’s admission to the PCU until their death, and the location of death was identified.

On the basis of the literature review (5-10,14-16) and discussion among the authors, we measured 22 potential scale items available on the first day of PCU admission, including (I) demographic variables (age, sex, primary sites of cancer, metastasis sites, marriage status, the presence or absence of a spouse, whether there is a daytime caregiver, whether the primary caregiver is the spouse, two or more family members living together, and whether there are multiple generations living together); (II) patient general conditions at admission according to the Eastern Cooperative Oncology Group (ECOG) performance status scale, the Palliative Prognostic Index (PPI) (19), and calories ingested on the first day); (III) vital signs at admission (systolic blood pressure, pulse, SpO₂); (IV) medications (opioid dose, oral morphine equivalent in mg); (V) patient symptoms at admission (fatigue, dyspnea, nausea and vomiting, abdominal distention, disturbance of consciousness, delirium, edema, dysphagia, motor disturbance, itching sensation, bleeding of the tumor, abdominal pain, aching pain, and anorexia), and (VI) preferred place of care and death (the patient’s preferred place of care, the patient’s preferred place for death, the family’s preferred place of care, the family’s preferred place for death).

The primary cancer sites were divided into the following categories: hematology and lymphology, urology, hepatobiliary and pancreatic, breast, gynecology, respiratory, gastroenterology, skin, head and neck, bone and soft tissue, and unknown primary. Cancer metastasis sites were found in the liver, lungs, brain, and bones. The admission period was calculated from the day a patient came to the PCU. The ECOG scale provides a consistent reference of measurement indicating the level of function of patients with cancer in terms of caring for themself, their activities of daily living, and their physical abilities. The items are scored from “0” (fully active, able to carry on all pre-disease performance without restriction) to “4” (completely disabled; cannot carry on any self-care; totally confined to bed or chair) (19). The PPI measures a patient’s general conditions based on performance status and several symptoms, such as dyspnea, delirium, oral intake, and edema (20). Scores range from 0 to 15, with higher scores indicating poorer general condition.

Patients’ caloric intake was calculated based on the designated calories of the meals provided by the hospital and the percentage of the food consumed (e.g., a patient offered a 1,200 kcal meal who ate 20% of the meal was considered to have ingested 240 kcal). The caloric content of meals served in the hospital was routinely reported in the medical record by nutritionists. The percentages of the food consumed were also routinely recorded by ward nurses in 10% increments. Also, food from family members was allowed, and the number of calories eaten was reflected based on the nurse’s report. Patient symptoms were measured using the Japanese version of the Support Team Assessment Schedule (STAS-J) using a scale from “none” [0] to “severe” [4] for each symptom (21,22). The reliability and validity of the Japanese version of the STAS were confirmed in a previous study (21), and we recorded one symptom with the highest score as a “major symptom” that results in admission.

Statistical analyses

We used the training-testing procedure to develop and validate the scoring tool. All eligible patients were dichotomized by an alternating method and assigned to the development model group and the validation group. For example, the patients who were discharged home were defined as the home discharge group, and those who were not discharged (i.e., died in the hospital) were defined as the hospital death group. For the patients who were discharged to other hospitals, we classified them into hospital death group because we followed all of them until death and confirmed that they died in hospitals.

In the development phase, first, to investigate factors associated with home discharge, a univariate logistic regression analysis was performed for each factor between the two groups: the home-discharge group and the hospital-death group. We used ad-hoc cutoff points for all variables because there are no confirmed cutoff points relevant to this study’s aim. No indeterminate or missing cases were identified.

Second, multivariate logistic regression analysis (variable reduction method—likelihood ratio) was performed to examine the association of each factor with items having a P value <0.2 in univariate logistic regression analysis and basic patient characteristics (age and sex). A probability of removal of a p-value of less than 0.1 was set, and factors with P<0.05 were identified as factors linked to home discharge. The following equation was developed using the identified factors (denoted as A1, A2...); their odds ratios were rounded to the nearest whole number (denoted as B1, B2...).

Score=A1×B1+A2×B2+…[1]

A: 1 if applicable to the identified factors, 0 if not applicable; B: rounded odds ratios for items with P values <0.05.

The total score for each patient was calculated using the created model group, and the highest sum of sensitivity and specificity was set as the cutoff value of the score formula by receiver operating characteristic (ROC) analysis. An ROC curve is a statistical method used to analyze the usefulness of a diagnostic method. The curve is formed on the graphical plot by connecting them with a polygonal line. The area under the curve (AUC) is used to quantify the sensitivity and specificity. The AUC value approaches “1” if the diagnostic method has high discriminative power. In our study, if the total score was greater than the cutoff value, the patient was determined to be eligible for discharge home, and this was used as the scoring tool.

In the validation phase, the scoring tool was applied to the validation group for verification. Validation was evaluated in terms of sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). We also evaluated the error rate, which represents the ratio of the number of predictions that differ from the actual observation value to the total number of predictions. Validity was evaluated based on sensitivity, specificity, PPV, NPV of at least 80% each, and an error rate of less than 20%. No formal sample size calculation was performed due to the exploratory nature of this study. IBM SPSS Statistics ver. 27 (IBM Corporation) was used for the analysis.

Results

During the study period, a total of 369 cancer patients were admitted to the PCU. Of them, we excluded 10 cases for which a death location could not be identified; the remaining 180 cases (of which 64 were discharged home) were included in the development of the model, and 179 cases (of which 72 were discharged home) were included in the validation phase. All data used for calculating scores were obtained on the day of admission. Among 136 patients who were discharged home, a total of 67 (49.3%) died at home eventually.

Development phase

The characteristics of the patients are summarized in Table 1. Mean age was 73.7 (range, 36–99) years; 112 (62.2%) patients were male, and 68 (37.8%) were female. The percentage of the patients whose spouse was the primary caregiver was 77% (86/112) among males and 38% (26/68) among females. The mean performance status was 3.1, and the median PCU admission period was 18.5 (range, 1–105) days.

Univariate analyses

The univariate logistic regression analysis results are shown in Table 1. Factors associated with patient outcomes (discharge to home or die in hospitals) included performance status, PPI score, systolic blood pressure, pulse, SpO₂, calorie intake, use of opioids, married and spouse alive, availability of a daytime caregiver, primary caregiver (spouse), primary tumor site, metastasis sites, and symptoms.

Multivariate analyses

The results of the multivariate logistic regression analysis are shown in Table 2. Independent factors associated with the outcome regarding whether a patient was discharged home were sex (female), calorie intake (520 kcal or more), availability of a daytime caregiver, family’s preferred place of care (home), and symptoms that resulted in hospitalization was not fatigue).

Development of scoring model

The most accurate model consisted of four factors: (I) sex (female), 4 points; (II) caloric intake at admission (>520 kcal), 19 points; (III) having a daytime caregiver, 11 points; (IV) family preference for home care, 139 points; and (V) symptoms other than fatigue that resulted in hospitalization, 7 points. From these results, we developed the formula as follows:

Score=(female×4)+(firstdaycalorieintakeof520kcalormore×19)+(presenceofdaytimecaregiver×11)+(familypreferredplaceofcarehomeorfacility×139)+(symptomsthatresultedinhospitalizationwerenotfatigue×7)[2]

Accuracy of the developed model

The value with the highest sensitivity + specificity was used as the cutoff value, and a score of 155 or less was determined to be hospital death, and a score of 155 or more was determined to be home discharge (Table 3). The AUC using this cutoff value was 0.949, with 95% confidence intervals of 0.918 to 0.981 (Figure 1).

Figure 1 ROC curve for the developed model. ROC, receiver operating characteristic.

Validation phase

Patient characteristics were similar in the development phase (data not shown). The scoring tool was applied to the validation group for verification. Table 4 demonstrates a cross-table of censored/non-censored cases and outcomes; and the sensitivity, specificity, NPV, PPV, and error rate were 75.3%, 86.3%, 82.2%, 80.6%, and 18.4% respectively (Table 5). Consequently, the accuracy of the score formula was considered to be valid: a score of more than 155 indicates a high likelihood of returning home.

Discussion

To our best knowledge, this is the first study to develop a clinical tool to identify a cancer patient who would be discharged to their home from an acute PCU. The accuracy of this scoring tool developed in this study is high. The tool is also simple, using only five parameters, and has been validated in oncology palliative care settings. The difference from an existing tool is sample (patients receiving home care vs. patients admitted to PCUs), outcomes (home death vs. discharge to home), and variables (variables obtained from nursing facilities vs. variables from patients on admission) (16). This constructed score could be calculated based on the patient factors and could be used as a useful tool for healthcare providers to easily identify patients necessary for planning their discharge to home.

In the score formula created, five factors were selected: female, calorie intake, presence of a daytime caregiver, family’s preferred place of care is home, and having something other than fatigue as the main symptom that resulted in hospitalization. Meta-analyses have not identified sex as an independent determinant for home death (7,8). A possible reason women were more likely to be discharged home in this cohort is that less than 40% of the female patients reported their spouse as the primary caregiver compared with approximately 80% of the male patients. This marked difference may indicate that the caregivers for the female were younger people capable of caring for patients. For example, children often become the caregivers for their mothers when they are discharged and moved back to their homes.

The results indicate that calorie intake, having a daytime caregiver, and the family preferring home as the place of care are factors associated with being discharged home; these results are consistent with existing studies (7-10). Miura et al. reported that the higher the calorie intake on the first day, the more likely the patient was to transition home, according to the results of a multivariate analysis of the study exploring predictors of possible discharge to home from a palliative care ward (15). Similarly, we found that a “calorie intake of 520 kcal or more on the first day” could be a useful factor for patients to be discharged home. Furthermore, multiple empirical studies have shown that “home” being the family’s preferred place of care is a particularly strong factor facilitating discharge to home (7-10,15,16).

Furthermore, fatigue was known to be a symptom that is associated with a high level of distress (23,24). However, there is conflicting evidence as to whether fatigue, or any other physical symptoms, are associated with home death (7,8). A potential interpretation of the finding in this cohort that fatigue was associated with hospital death (i.e., less likely to discharge to home) is that treatment options to overcome fatigue are limited, and fatigue has a significant impact on daily life that results in a lower possibility of discharge to home (15). The presence of fatigue might also mean that the patient’s general condition has deteriorated; thus, substituting the fatigue item with other items, such as performance status, prognostic scores, or vital signs, might be of value in future studies.

There are several limitations of this study. First, this is a single-center study, and validation in a multicenter study is needed. Second, we did not include the economic status of the patient’s family because, in the Japanese healthcare system, all medical costs are covered by the national insurance program; the upper limits for self-payment are about 1,000 USD per month, even in populations with relatively high incomes. Third, the sample size was relatively small. There was a heavy weight on one item in the equation (i.e., 139 points for family preference for home death). Fourth, although sex was identified as a predictive factor of home discharge, there might be complex interactions among variables such as the patient’s sex, their relationship to the primary caregiver (spouse or not), and the age of the primary caregiver. Fifth, we used ad-hoc cutoff points for all variables. Sixth, there would be cultural differences in interpreting the study results; for example, the culture of looking after one’s elderly family member at home differs among countries. Seventh, we could not explore the effect of the inclusion of admission length before PCU transfer due to lack of data, and future study could be improved if admission length is added as a variable. Eighth, although we adopted “home discharge” as an outcome, “length of stay at home” may be a more important outcome. The finding that the rate of home death was sufficiently high in our patients who had been discharged home (49% compared with the national average of 13%) provides some rationale indicating the value of this outcome. Future studies might explore “length of home stay” or “home death” as outcomes.

Conclusions

In conclusion, whether a patient admitted to a PCU can discharge to their home could be predicted using the simple clinical tool described in this study. However, further validation and outcome studies are warranted.

Acknowledgments

Funding: None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://apm.amegroups.com/article/view/10.21037/apm-22-902/rc

Data Sharing Statement: Available at https://apm.amegroups.com/article/view/10.21037/apm-22-902/dss

Peer Review File: Available at https://apm.amegroups.com/article/view/10.21037/apm-22-902/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://apm.amegroups.com/article/view/10.21037/apm-22-902/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are 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. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Ethical Review Committee for Clinical Research of Toyama Kouseiren Takaoka Hospital. (Approval No. 20190829003; the full protocol is available from the corresponding author). Informed consent was waived due to the national guideline by the Ministry of Health, Labour and Welfare, i.e., Ethical Guidelines for Medical and Biological Research Involving Human Subjects.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Steinhauser KE, Clipp EC, McNeilly M, et al. In search of a good death: observations of patients, families, and providers. Ann Intern Med 2000;132:825-32. [Crossref] [PubMed]
2. Miyashita M, Sanjo M, Morita T, et al. Good death in cancer care: a nationwide quantitative study. Ann Oncol 2007;18:1090-7. [Crossref] [PubMed]
3. Hatano Y, Aoyama M, Morita T, et al. The relationship between cancer patients’ place of death and bereaved caregivers’ mental health status. Psychooncology 2017;26:1959-64. [Crossref] [PubMed]
4. Kinoshita H, Maeda I, Morita T, et al. Place of death and the differences in patient quality of death and dying and caregiver burden. J Clin Oncol 2015;33:357-63. [Crossref] [PubMed]
5. Fereidouni A, Rassouli M, Salesi M, et al. Preferred Place of Death in Adult Cancer Patients: A Systematic Review and Meta-Analysis. Front Psychol 2021;12:704590. [Crossref] [PubMed]
6. Wilson R, Hepgul N, Higginson IJ, et al. End-of-life care and place of death in adults with serious mental illness: A systematic review and narrative synthesis. Palliat Med 2020;34:49-68. [Crossref] [PubMed]
7. Gomes B, Higginson IJ. Factors influencing death at home in terminally ill patients with cancer: systematic review. BMJ 2006;332:515-21. Erratum in: BMJ 2006;332:1012. [Crossref] [PubMed]
8. Costa V, Earle CC, Esplen MJ, et al. The determinants of home and nursing home death: a systematic review and meta-analysis. BMC Palliat Care 2016;15:8. [Crossref] [PubMed]
9. Cai J, Zhang L, Guerriere D, et al. What Variables Contribute to the Achievement of a Preferred Home Death for Cancer Patients in Receipt of Home-Based Palliative Care in Canada? Cancer Nurs 2021;44:214-22. [Crossref] [PubMed]
10. Gamblin V, Prod’homme C, Lecoeuvre A, et al. Home hospitalization for palliative cancer care: factors associated with unplanned hospital admissions and death in hospital. BMC Palliat Care 2021;20:24. [Crossref] [PubMed]
11. Hui D, Elsayem A, Palla S, et al. Discharge outcomes and survival of patients with advanced cancer admitted to an acute palliative care unit at a comprehensive cancer center. J Palliat Med 2010;13:49-57. [Crossref] [PubMed]
12. Sekimoto G, Aso S, Hayashi N, et al. Experience of the temporary discharge from the inpatient palliative care unit: A nationwide post-bereavement survey for end-of-life cancer patients. Asia Pac J Oncol Nurs 2022;9:100058. [Crossref] [PubMed]
13. Hamano J, Yamaguchi T, Maeda I, et al. Multicenter cohort study on the survival time of cancer patients dying at home or in a hospital: Does place matter? Cancer 2016;122:1453-60. [Crossref] [PubMed]
14. Hausner D, Kevork N, Pope A, et al. Factors associated with discharge disposition on an acute palliative care unit. Support Care Cancer 2018;26:3951-8. [Crossref] [PubMed]
15. Miura T, Matsumoto Y, Okizaki A, et al. A retrospective study of prognostic factors for patients who can be discharged home from a palliative care unit. Palliative Care Research 2013;8:107-15. [Crossref]
16. Fukui S, Morita T, Yoshiuchi K. Development of a Clinical Tool to Predict Home Death of a Discharged Cancer Patient in Japan: a Case-Control Study. Int J Behav Med 2017;24:584-92. [Crossref] [PubMed]
17. Tanabe K, Sawada K, Shimada M, et al. Evaluation of A Novel Information-Sharing Instrument for Home-Based Palliative Care: A Feasibility Study. Am J Hosp Palliat Care 2015;32:611-9. [Crossref] [PubMed]
18. Murakami N, Tanabe K, Morita T, et al. Impact of a Six-Year Project to Enhance the Awareness of Community-Based Palliative Care on the Place of Death. J Palliat Med 2018;21:1494-8. [Crossref] [PubMed]
19. Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982;5:649-55. [Crossref] [PubMed]
20. Morita T, Tsunoda J, Inoue S, et al. The Palliative Prognostic Index: a scoring system for survival prediction of terminally ill cancer patients. Support Care Cancer 1999;7:128-33. [Crossref] [PubMed]
21. Miyashita M, Matoba K, Sasahara T, et al. Reliability and validity of the Japanese version of the Support Team Assessment Schedule (STAS-J). Palliat Support Care 2004;2:379-85. [Crossref] [PubMed]
22. Collins ES, Witt J, Bausewein C, et al. A Systematic Review of the Use of the Palliative Care Outcome Scale and the Support Team Assessment Schedule in Palliative Care. J Pain Symptom Manage 2015;50:842-53.e19. [Crossref] [PubMed]
23. Henson LA, Maddocks M, Evans C, et al. Palliative Care and the Management of Common Distressing Symptoms in Advanced Cancer: Pain, Breathlessness, Nausea and Vomiting, and Fatigue. J Clin Oncol 2020;38:905-14. [Crossref] [PubMed]
24. Klasson C, Helde Frankling M, Lundh Hagelin C, et al. Fatigue in Cancer Patients in Palliative Care-A Review on Pharmacological Interventions. Cancers (Basel) 2021;13:985. [Crossref] [PubMed]