Association of Unplanned Home Visits, Deaths, Preference for Dying at Home, and Home Deaths With Patient Complexity in a Physician‐Led Home Visit Setting: A Secondary Analysis of a Multicenter Prospective Cohort Study

ABSTRACT

Introduction

The primary objective was to examine the association between unplanned physician‐led home visits and patient complexity, as measured by the Minnesota Complexity Assessment Method (MCAM), and between mortality and patient complexity. The secondary objective was to investigate the relationship between patient complexity and patients' wishes to die at home, families' acceptance of the patients' deaths at home, and the location of death.

Methods

We applied Cox proportional hazards models to estimate cause‐specific hazard ratios (HRs) for patient complexity, with unplanned home visits and deaths as dependent variables. We employed multinomial logistic regression models to evaluate relative risk ratios (RRRs) of patient complexity, with the patient's wish to die at home, the family's acceptance of the patient's death at home, and the location of death as dependent variables.

Results

A total of 712 participants were included in the analysis. The Illness domain of the MCAM was positively associated with unplanned home visits and deaths (adjusted HRs: 1.20 and 1.21, respectively). Furthermore, the Illness domain was also positively associated with patients' wishes to die at home (adjusted RRR [aRRR]: 1.24) and families' acceptance of the patients' deaths at home (aRRR: 1.19), whereas the Social domain was negatively associated with these outcomes (aRRRs: 0.81 and 0.84, respectively).

Conclusion

Patients with higher biomedical and social complexity may require closer clinical attention in physician‐led home visit settings, which could help reduce unplanned visits, improve mortality‐related outcomes, and enable patients and their families to choose the location of death without restrictions imposed by biopsychosocial factors.

1. Introduction

In Japan, the establishment of the community‐based integrated care system is being promoted to address the challenges associated with rapid population aging [1]. At the same time, the importance of physician‐led home visits is gaining increasing attention [2, 3]. With approximately 30% of its population aged 65 and older, Japan has the highest proportion of older adults worldwide [4]. To support older individuals in living independently in familiar environments, a comprehensive system for support and service delivery, known as the community‐based integrated care system, has been developed [1]. In particular, home medical care, including physician‐led home visits, plays a crucial role, with a steadily increasing number of patients receiving such care [5].

As aging leads to biological, psychological, and social challenges, an increasing number of older adults experience the coexistence and interaction of these issues. For example, biological problems include multimorbidity [6, 7] and frailty [8]. Among individuals aged 65 years or older in Japan, the prevalence of multimorbidity exceeds 60% [9], while that of frailty is over 7% [10]. Psychological and social aspects are also becoming increasingly complex [11, 12, 13, 14, 15, 16]. The interplay of these biopsychosocial factors in individual patients is collectively assessed as “patient complexity,” an increasingly important concept for providing effective care and treatment [17].

Managing patients with high levels of complexity is a critical issue in home medical care. To date, various health‐related outcomes have been reported to be associated with patient complexity worldwide [18]. In Japanese primary and secondary healthcare settings, patient complexity has been linked to hospital length of stay [19, 20], burden on physicians and nurses [21], healthcare costs [22], and health outcomes such as alcohol use disorders [23]. However, patient complexity in physician‐led home visit settings remains inadequately explored.

To reduce the number of unplanned physician‐led home visits, improve mortality‐related outcomes, and enable patients and their families to choose the location of death without restrictions imposed by biopsychosocial factors, it is essential to accumulate evidence to identify which patients with high complexity in physician‐led home visit settings should be prioritized for intervention. Furthermore, in physician‐led home visit settings, patient care is provided through interprofessional collaboration involving physicians, nurses, care workers, care managers, and other professionals. Therefore, it is important to assess patient complexity using a shared framework, such as the Minnesota Complexity Assessment Method (MCAM), which is understandable across healthcare, long‐term care, and social welfare professionals and facilitates communication and coordination among them.

The primary objective was to examine the association between unplanned physician‐led home visits and patient complexity, as measured by the MCAM, and between mortality and patient complexity. The secondary objective was to investigate the relationship between patient complexity and patients' wishes to die at home, families' acceptance of the patients' deaths at home, and the location of death.

2. Methods

2.1. Design

This study was a secondary analysis of a multicenter prospective cohort study, the Elderly Mortality Patients Observed Within the Existing Residence (EMPOWER) Japan study, which examined mortality rates and locations of death in physician‐led home visit settings in Japan, as well as the risk factors associated with them [2, 3].

2.2. Participants and Setting

The inclusion criteria for this study were patients from any of the 13 institutions participating in the original EMPOWER Japan study, as detailed below, that agreed to cooperate. The exclusion criterion was patients who declined to participate in this study. We also excluded patients who resided in a nursing home at the beginning of the follow‐up period to align with the study's secondary objective.

The inclusion criteria for the original EMPOWER Japan study were as follows: patients aged ≥ 65 years who began receiving physician‐led home visits from any of the 13 institutions between February 1, 2013, and January 31, 2016. The exclusion criteria for the original EMPOWER Japan study were patients who declined to participate in the study. Participants were followed up until January 31, 2017. More details have been reported in previous studies [2, 3].

The original EMPOWER Japan study was conducted at 12 family medicine clinics and one regional hospital located in the greater Tokyo area, all of which are affiliated with the Centre for Family Medicine Development (CFMD) and are included in its primary care practice‐based research network (CFMD‐PBRN). In Japan, physician‐led home visits are regularly scheduled medical services (typically every 2 weeks) for patients who have difficulty visiting outpatient facilities due to disease or injury. Patients pay between 10% and 30% of their medical costs based on their age and income level, while insurance covers the remaining costs. The majority of the physicians involved in the original EMPOWER Japan study were residents in the Japan Primary Care Association (JPCA) family medicine residency programs, JPCA‐certified family physicians, or JPCA‐certified supervisory physicians.

2.3. Measures

2.3.1. Outcome Measures

The outcome measures included unplanned home visits and deaths, as well as the patient's wish to die at home (yes/no/unknown), the family's acceptance of the patient's death at home (yes/no/unknown), and the location of death (at home/in a hospital or nursing home/surviving). Unplanned home visits were defined as those occurring for the first time after the initiation of home visits. However, unplanned home visits were excluded if the patient's outcome was death, as these visits were most likely for confirmation of the patient's death and were not aligned with the study's primary objectives. The patient's wish to die at home and the family's acceptance of the patient's death at home were recorded based on their initial expression after the initiation of home visits. Discussing the patient's wish to die at home and the family's acceptance of the patient's death at home can be a highly sensitive matter. Therefore, the assessment was based on flexible and individualized communication between physicians and patients (and their families) in clinical practice.

2.3.2. Patient Complexity Assessed Using the MCAM

The MCAM, developed by Peek et al. [17] is a tool for assessing patient complexity across various health aspects. The MCAM categorizes patient complexity into five domains: “Illness,” “Readiness to engage,” “Social,” “Health system,” and “Resources for care.” Each domain consists of two items: “Symptom severity/functional impairment” and “Diagnostic challenge” in the Illness domain; “Distress, distraction, preoccupation with symptoms” and “Readiness for treatment and change” in the Readiness to engage domain; “Current home/residential safety, stability” and “Participation in social network” in the Social domain; “Current organization of care” and “Patient‐clinician (or team) relationships” in the Health system domain; and “Shared language with providers” and “Adequacy/consistency of insurance for care” in the Resources for care domain. Each item has four levels of complexity, scored from 0 to 3, with higher scores indicating greater complexity. The total MCAM score ranges from 0 (minimum) to 30 (maximum), while each MCAM domain score ranges from 0 (minimum) to 6 (maximum). The MCAM was assessed at the initiation of home visits. In this study, we used the original English version of the MCAM [17].

2.3.3. Other Variables

We included biomedical, psychological, and social variables at the initiation of home visits in accordance with previous reports [2, 3]. The biomedical variables included sex, age, cancer, serum albumin, the Barthel Index (The Maryland State Medical Society holds the copyright for the Barthel Index. Mahoney FI, Barthel D. Functional evaluation: the Barthel Index. Maryland State Med Journal 1965;14:56–61. Used with permission. Permission is required to modify the Barthel Index or to use it for commercial purposes), number of medications, opioid use, gastrostomy, respiratory devices, domiciliary oxygen therapy, urinary catheter use, dialysis, and pressure ulcer. The psychological variables included the Cornell Scale for Depression in Dementia and dementia, and the social variables included public assistance, the presence of a full‐time caregiver, and living alone.

2.3.4. Follow‐Up and Survival Status

Participants were observed until death or the end of their observation periods. In cases of deaths occurring outside the home, such as in a hospital or nursing home, the date and location of death were determined through contact with the hospital where the patient had been admitted or via information obtained from the family.

2.4. Study Size

The study size was determined by the number of participants included in the original EMPOWER Japan study.

2.5. Statistical Analyses

We performed descriptive analyses to summarize the following data: participant characteristics, incidence rate of unplanned home visits, mortality rate, the total MCAM scores, the MCAM domain scores, patient's wish to die at home, family's acceptance of the patient's death at home, and location of death. The descriptive data were expressed as means (standard deviation [SD]) or medians (interquartile range [IQR]) for continuous variables and counts (percentage [%]) for categorical variables.

We applied Cox proportional hazards models to estimate cause‐specific hazard ratios (HRs) for patient complexity, with unplanned home visits as the dependent variables while treating competing events (deaths) as censoring. We also utilized the model to investigate the HRs for patient complexity, with deaths as dependent variables. The independent variables included sex (female/male = 1/0) and age [years]. However, we did not include other variables that described participant characteristics—such as cancer, serum albumin, the Barthel Index, number of medications, opioid use, gastrostomy, respiratory devices, domiciliary oxygen therapy, urinary catheter use, dialysis, pressure ulcer treatment, the Cornell Scale for Depression in Dementia, dementia, public assistance, the presence of a full‐time caregiver, and living alone—as independent variables. These were excluded because their effects had already been assessed within the MCAM and were reflected in the MCAM scores.

We employed multinomial logistic regression models to evaluate relative risk ratios (RRRs) of patient complexity, with the patient's wish to die at home (yes/no/unknown), the family's acceptance of the patient's death at home (yes/no/unknown), and the location of death (at home/in a hospital or nursing home/surviving) as dependent variables. Again, we used sex (female/male = 1/0) and age [years] as independent variables and did not include other variables that described participant characteristics.

Multiple imputation using the multivariate imputation by chained equations algorithm was used to deal with missing data. This imputation process used all covariates considered associated with the missing value. The covariates were each outcome measure (time to unplanned home visits, time to deaths, patient's wish to die at home, family's acceptance of the patient's death at home, and location of death), the MCAM domain scores, age, sex, indicator variables of clinics, as well as other participant characteristics with no missing value. The results across 100 imputed datasets were combined.

All statistical analyses were conducted using STATA/SE 18.0 and p‐values of < 0.05 were considered statistically significant [24].

3. Results

Among the 825 participants in the original EMPOWER Japan study, 50 were excluded because one clinic declined to cooperate, and 63 were excluded for residing in a nursing home at the beginning of the follow‐up period. Thus, 712 participants were included in the analysis (Figure 1). However, 69 participants were lost to follow‐up. The total observation period was 332 788 person‐days (911.1 person‐years). The mean (SD) observation period was 467.4 (391.0) days, and the median (IQR) observation period was 395.5 (111–735) days.

FIGURE 1.

Flowchart detailing the inclusion, exclusion, and follow‐up process of participants.

Table 1 presents participant characteristics by follow‐up status (followed up vs. lost to follow‐up).

TABLE 1.

Participant characteristics by follow‐up status.

	Total	Followed up	Lost to follow‐up
	N = 712	N = 643	N = 69
Sex, n (%)
Female	381 (53.5)	340 (52.9)	41 (59.4)
Male	331 (46.5)	303 (47.1)	28 (40.6)
Age [years], mean (SD)	83.3 (7.9)	83.1 (8.0)	84.2 (7.0)
Cancer, n (%)	190 (26.7)	186 (28.9)	4 (5.8)
Serum albumin [g/dL], mean (SD)	3.5 (0.7)	3.5 (0.7)	3.7 (0.5)
Data missing	47	42	5
Barthel Index, mean (SD)	54.4 (32.5)	53.7 (32.7)	60.3 (30.3)
Data missing	6	6	0
Number of medications, mean (SD)	5.5 (3.7)	5.5 (3.7)	5.2 (3.6)
Opioid use, n (%)	26 (3.7)	26 (4.0)	0 (0.0)
Gastrostomy, n (%)	10 (1.4)	10 (1.6)	0 (0.0)
Respiratory devices, n (%)	4 (0.6)	4 (0.6)	0 (0.0)
Domiciliary oxygen therapy, n (%)	48 (6.7)	44 (6.8)	4 (5.8)
Urinary catheter use, n (%)	36 (5.1)	35 (5.4)	1 (1.5)
Dialysis, n (%)	2 (0.3)	2 (0.3)	0 (0.0)
Pressure ulcer treatment, n (%)	20 (2.8)	16 (2.5)	4 (5.8)
Cornell Scale for Depression in Dementia, mean (SD)	2.6 (2.9)	2.7 (2.9)	1.6 (2.2)
Data missing	23	22	1
Dementia, n (%)	396 (56.4)	354 (55.8)	42 (61.8)
Data missing	10	9	1
Public assistance, n (%)	102 (14.3)	83 (12.9)	19 (27.5)
Presence of a full‐time caregiver, n (%)	504 (70.8)	471 (73.3)	33 (47.8)
Living alone, n (%)	179 (25.1)	145 (22.6)	34 (49.3)

Abbreviation: SD, standard deviation.

The mean (SD) total MCAM score was 9.7 (3.4). Additionally, the mean (SD) MCAM domain scores were as follows: Illness, 2.5 (1.3); Readiness to engage, 1.5 (1.3); Social, 2.4 (1.4); Health system, 3.1 (0.8); and Resources for care, 0.2 (0.6). There were 18 missing values for both the total MCAM score and each individual MCAM domain score. The mean (SD) total MCAM scores by follow‐up status were 9.7 (3.4) for participants who were followed up (16 missing values) and 9.3 (3.3) for those lost to follow‐up (2 missing values).

The incidence rate of unplanned home visits was 662.6 per 1000 person‐years (95% confidence interval [CI]: 599.3–732.5 per 1000 person‐years). The mortality rate was 383.0 per 1000 person‐years (95% CI: 344.9–425.4 per 1000 person‐years).

Table 2 presents the results of the Cox proportional hazards model examining the association of patient complexity with unplanned home visits and deaths and patient complexity. The Illness domain was associated with both unplanned home visits and deaths. Using multiple imputation, the adjusted HRs were 1.20 (95% CI: 1.10–1.32) and 1.21 (95% CI: 1.10–1.33), respectively.

TABLE 2.

Cox proportional hazards model examining the association between unplanned home visits/deaths and patient complexity.

	Complete case analysis (N = 694)	p	Multiple imputation (N = 712)	p
	aHR [95% CI]		aHR [95% CI]
Unplanned home visits
Sex	0.98 [0.79–1.21]	0.86	0.98 [0.80–1.21]	0.88
Age	1.00 [0.99–1.01]	0.95	1.00 [0.99–1.01]	0.93
The MCAM Illness domain	1.20 [1.10–1.32]	< 0.01	1.20 [1.10–1.32]	< 0.01
The MCAM Readiness to engage domain	0.97 [0.89–1.06]	0.57	0.97 [0.89–1.07]	0.57
The MCAM Social domain	0.93 [0.87–1.02]	0.11	0.94 [0.87–1.02]	0.11
The MCAM Health system domain	1.02 [0.90–1.15]	0.81	1.02 [0.90–1.15]	0.80
The MCAM Resources for care domain	0.97 [0.80–1.17]	0.73	0.96 [0.79–1.17]	0.68
Deaths
Sex	0.67 [0.54–0.84]	< 0.01	0.66 [0.53–0.83]	< 0.01
Age	1.00 [0.98–1.01]	0.80	1.00 [0.98–1.01]	0.70
The MCAM Illness domain	1.22 [1.10–1.34]	< 0.01	1.21 [1.10–1.33]	< 0.01
The MCAM Readiness to engage domain	0.94 [0.86–1.04]	0.22	0.94 [0.86–1.04]	0.23
The MCAM Social domain	0.93 [0.86–1.02]	0.11	0.93 [0.86–1.01]	0.10
The MCAM Health system domain	1.05 [0.91–1.20]	0.54	1.04 [0.90–1.20]	0.58
The MCAM Resources for care domain	1.03 [0.85–1.25]	0.78	1.03 [0.85–1.25]	0.76

Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; MCAM, Minnesota Complexity Assessment Method.

Among participants, 285 (40.0%) wished to die at home, 126 (17.7%) did not, and 301 (42.3%) had an unknown preference. Additionally, the families of 258 (36.2%) participants accepted the patient's death at home, while 166 (23.3%) did not, and the acceptance status for 288 (40.5%) was unknown. Among the 349 participants who died, 127 (36.4%) died at home, 220 (63.0%) died in a hospital or nursing home, and 2 (0.6%) died in an unknown location. The numbers of deaths, deaths at home, deaths at hospitals or nursing homes, and deaths at unknown locations for each institution are shown in Table S1.

Table 3 presents the results of the multinomial logistic regression model examining the association of patient complexity with the patient's wish to die at home, the family's acceptance of the patient's death at home, and deaths at home. The relative risk ratios represent the likelihood of the patient's wish to die at home compared to no wish, the family's acceptance of the patient's death at home compared to no acceptance, and deaths at home compared to deaths in a hospital or nursing home. The Illness and Social domains were associated with the patient's wish to die at home. After multiple imputation, the adjusted RRRs (aRRRs) were 1.24 (95% CI: 1.02–1.50) and 0.81 (95% CI: 0.69–0.97), respectively. Additionally, these domains were also associated with the family's acceptance of the patient's death at home. After multiple imputation, the aRRRs were 1.19 (95% CI: 1.003–1.42) and 0.84 (95% CI: 0.72–0.99), respectively. However, no significant association was found between patient complexity and deaths at home.

TABLE 3.

Multinomial logistic regression model examining the association of patient complexity with the patient's wish to die at home, the family's acceptance of the patient's death at home, and deaths at home.

	Complete case analysis (N = 694)	p	Multiple imputation (N = 712)	p
	aRRR [95% CI]		aRRR [95% CI]
Patient's wish to die at home
Sex	1.01 [0.65–1.58]	0.96	0.98 [0.63–1.52]	0.93
Age	1.04 [1.01–1.07]	0.01	1.04 [1.01–1.07]	0.01
The MCAM Illness domain	1.24 [1.02–1.50]	0.03	1.24 [1.02–1.50]	0.03
The MCAM Readiness to engage domain	1.14 [0.94–1.39]	0.18	1.14 [0.93–1.38]	0.20
The MCAM Social domain	0.81 [0.68–0.96]	0.02	0.81 [0.69–0.97]	0.02
The MCAM Health system domain	1.03 [0.80–1.34]	0.79	1.04 [0.80–1.34]	0.79
The MCAM Resources for care domain	1.07 [0.65–1.76]	0.79	1.08 [0.65–1.79]	0.78
Family's acceptance of the patient's death at home
Sex	1.32 [0.88–2.00]	0.18	1.27 [0.85–1.91]	0.25
Age	1.00 [0.98–1.03]	0.71	1.01 [0.98–1.03]	0.55
The MCAM Illness domain	1.20 [1.00–1.42]	0.045	1.19 [1.003–1.42]	0.045
The MCAM Readiness to engage domain	1.06 [0.89–1.26]	0.53	1.06 [0.88–1.26]	0.55
The MCAM Social domain	0.84 [0.72–0.98]	0.03	0.84 [0.72–0.99]	0.03
The MCAM Health system domain	1.04 [0.82–1.32]	0.73	1.04 [0.82–1.33]	0.72
The MCAM Resources for care domain	0.91 [0.63–1.32]	0.64	0.91 [0.63–1.32]	0.63

	Complete case analysis (N = 692)	p	Multiple imputation (N = 710)	p
	aRRR [95% CI]		aRRR [95% CI]
Deaths at home
Sex	1.03 [0.65–1.64]	0.88	0.99 [0.62–1.56]	0.95
Age	1.02 [0.99–1.05]	0.30	1.02 [0.99–1.05]	0.25
The MCAM Illness domain	1.13 [0.93–1.38]	0.21	1.13 [0.93–1.38]	0.21
The MCAM Readiness to engage domain	0.89 [0.73–1.10]	0.29	0.89 [0.73–1.09]	0.28
The MCAM Social domain	0.90 [0.75–1.07]	0.23	0.90 [0.75–1.07]	0.24
The MCAM Health system domain	1.08 [0.80–1.45]	0.63	1.08 [0.80–1.46]	0.61
The MCAM Resources for care domain	0.95 [0.62–1.45]	0.80	0.94 [0.61–1.44]	0.78

Note: In the analyses using the location of death as a dependent variable, data from two participants who died at unknown locations were excluded. We employed multinomial logistic regression models to evaluate the relative risk ratios of patient complexity, using the patient's wish to die at home (yes/no/unknown), the family's acceptance of the patient's death at home (yes/no/unknown), and the location of death (at home/in a hospital or nursing home/surviving) as dependent variables. The relative risk ratios represent the likelihood of the patient's wish to die at home compared to no wish, the family's acceptance of the patient's death at home compared to no acceptance, and deaths at home compared to deaths in a hospital or nursing home.

Abbreviations: aRRR, adjusted relative risk ratio; CI, confidence interval; MCAM, Minnesota Complexity Assessment Method.

4. Discussion

The Illness domain of the MCAM was positively associated with unplanned home visits and deaths. Furthermore, the Illness domain was also positively associated with patients' wishes to die at home and families' acceptance of the patients' deaths at home, whereas the Social domain was negatively associated with these outcomes. However, no significant association was observed between patient complexity and deaths at home.

Patients with greater illness‐related complexity may be clinically more vulnerable to sudden changes in condition, which often precipitate unplanned home visits or death. Several illness‐related factors, including cancer, the use of visiting nurse services, high levels of care needs, the use of a urinary catheter, and the use of a central venous catheter, have been reported to be linked to unplanned home visits [25, 26]. Moreover, Kaneko et al., using the same dataset, found that high Charlson Comorbidity Index scores, low Barthel Index scores, and the use of oxygen therapy were associated with increased mortality [2]. These variables were incorporated into the MCAM Illness domain, contributing to the findings of the current study. In contrast, although Kaneko et al. also identified that high Cornell Scale for Depression in Dementia scores and not receiving public assistance [2]—factors relevant to the MCAM Readiness to engage and Resources for care domains—were associated with increased mortality, this study did not find an association between these domains and deaths. This discrepancy is likely attributable to the MCAM's broad evaluative scope. While the MCAM comprehensively assesses patient complexity at the domain and item levels, it does not assess specific individual elements in detail.

Considering that patients with complex biomedical conditions tended to prefer dying at home and that their families often accepted this preference, these findings likely reflect the preferences of terminally ill patients, who comprised a portion of the study participants, and their families [2, 3]. Approximately half of Japanese individuals express a desire to spend the final stage of their lives in a home care setting if diagnosed with a terminal illness [27]. Physician‐led home visits play a vital role in such settings, providing terminally ill patients and their families with end‐of‐life care. Accordingly, in this study, patients with a terminal illness—particularly those with potentially high scores in the Symptom severity/functional impairment item within the MCAM Illness domain—who wished to die at home and whose families accepted this wish were included. Their preferences may have influenced the results of this study.

Although home is considered the most preferred place for end‐of‐life care for both patients and their families [28], complex social conditions have been found in this study to limit their choices. Individual and environmental factors reportedly influence these preferences [28]. For example, being surrounded by family and friends encourages patients and their families to choose home as the place of care. This individual factor corresponds to the Participation in social network item in the MCAM Social domain, which assesses participation with family, work, and friends. Additionally, a familiar and supportive environment is also a key reason for preferring home and represents an environmental factor, aligning with the Current home/residential safety, stability item in the same domain. That is, relatively complex social conditions, such as the absence of family and friends or an unfamiliar and unsupportive environment, can act as barriers to choosing home as the preferred place of care.

This study did not identify an association between the MCAM and deaths at home, which is inconsistent with previous studies. Several factors have been reported to influence the likelihood of dying at home [29, 30, 31]. Notably, Watanabe et al., using the same dataset, found that higher scores on the Barthel Index were associated with a decreased likelihood of dying at home compared to other locations [3], whereas the use of oxygen therapy and having a full‐time available caregiver increased the likelihood [3]. Considering that the Barthel Index and the use of oxygen therapy correspond to the MCAM Illness domain and that having a full‐time available caregiver aligns with the Social domain, these results suggest that the MCAM may not adequately account for these variables. This discrepancy may also be attributed to the MCAM's broad evaluative scope, as described above.

The study has some limitations. First, the generalizability of the findings is limited. This study is a secondary analysis of the EMPOWER Japan study, which was conducted at medical facilities in the greater Tokyo area. Consequently, caution should be exercised when extrapolating the findings to populations outside this region. Second, causal inference cannot be established because this study was a secondary analysis of observational data. The findings should therefore be interpreted as associative rather than causal, and they warrant further investigation in prospective or interventional studies. Third, the follow‐up rate was suboptimal. Participants lost to follow‐up had slightly lower total MCAM scores than those who completed the follow‐up. As a result, total MCAM scores may have been overestimated, potentially introducing bias into the results. Fourth, detailed information on institutional characteristics, such as the numbers of physicians, nurses, and medical social workers, was unavailable. In addition, some outcomes, such as the patient's wish to die at home and the family's acceptance of the patient's death at home, may not have remained stable over time. Therefore, without accounting for variation across institutions and by treating time‐dependent variables as constants, the associations between the MCAM and the outcomes may have been overestimated or underestimated. Fifth, no formal standardization of data collection regarding the patient's wish to die at home and the family's acceptance of the patient's death at home was implemented, which may have affected the quality of the data on these preferences. However, discussing these preferences can be a highly sensitive matter. Therefore, we considered it most appropriate to base the assessment on flexible and individualized communication between physicians and patients (and their families) in clinical practice, and we believe that the data obtained reflect the realities of communication in physician‐led home visit settings.

In conclusion, the Illness domain of the MCAM was positively associated with unplanned home visits and deaths. Furthermore, it was also positively associated with patients' wishes to die at home and families' acceptance of the patients' deaths at home, whereas the Social domain exhibited a negative association. The observed association suggests that patients with higher biomedical and social complexity may require closer clinical attention in physician‐led home visit settings, which could help reduce unplanned visits, improve mortality‐related outcomes, and enable patients and their families to choose the location of death without restrictions imposed by biopsychosocial factors.

Author Contributions

Y.S. was responsible for conceptualizing the study, analyzing and interpreting the data, and drafting and revising the manuscript. R.M., S.H., and M.Y. contributed to the study's design, data interpretation, and manuscript drafting and revision. T.W., T.K., S.Y., K.T., M.M., and M.M. participated in the study's design, data interpretation, and manuscript revision. All authors have read and approved the final manuscript.

Funding

This secondary analysis received no specific grant from any funding agency in the public, commercial, or not‐for‐profit sectors. However, the original research from which this analysis was derived was supported by the JSPS KAKENHI (JP24590819). The funding agency had no role in the study design, data collection, analysis, interpretation, manuscript writing, or the decision to submit the article for publication.

Ethics Statement

This study was conducted in accordance with the Declaration of Helsinki and the Ethical Guidelines for Medical and Biological Research Involving Human Subjects. Ethical approval was obtained from the Ethics Committee of The Jikei University School of Medicine (acceptance number: 34‐434 [11591]) and the Ethical Review Committee of the Tokyo‐Hokuto Health Cooperative Association (application number: 121).

Consent

Information about this study was disclosed to participants, providing them with the opportunity to opt out.

Conflicts of Interest

M.M. wrote a manuscript in a Japanese journal named Chiryo Vol. 106, No. 9, 1018–1021 and received payment for a manuscript. M.M. received speaker's honoraria from The Japanese Society of Clinical Nutrition and The Japanese Clinical Nutrition Association. M.M. received lecture fees and lecture travel fees from the Centre for Family Medicine Development of the Japanese Health and Welfare Co‐operative Federation. M.M. is an adviser for the Centre for Family Medicine Development Practice‐Based Research Network. M.M. is a program director of the Jikei Clinical Research Program for Primary‐care. M.M.'s son‐in‐law has worked at IQVIA Services Japan K.K., which is a contract research organization and a contract sales organization. M.M.'s son‐in‐law works at Syneos Health Clinical K.K., which is a contract research organization and a contract sales organization. The other authors declare that they have no competing interests.

Supporting information

Table S1: Numbers of deaths, deaths at home, deaths at hospitals or nursing homes, and deaths at unknown locations for each institution.

Sugiyama Y., Watanabe T., Mutai R., et al., “Association of Unplanned Home Visits, Deaths, Preference for Dying at Home, and Home Deaths With Patient Complexity in a Physician‐Led Home Visit Setting: A Secondary Analysis of a Multicenter Prospective Cohort Study,” Geriatrics & Gerontology International 26, no. 1 (2026): e70316, 10.1111/ggi.70316.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.